Resistance to AFN-1252 Arises from Missense Mutations in Staphylococcus aureus Enoyl-acyl Carrier Protein Reductase (FabI)*

Background: AFN-1252 is a FabI inhibitor developed to treat Staphylococcus aureus. Results: AFN-1252 resistance arises from a single missense mutation that produces a FabI(M99T) protein. Conclusion: The interaction between Met-99 and AFN-1252 accounts for the staphylococcal selectivity of the drug. Significance: The AFN-1252-resistant strains remain susceptible to drug concentrations typically achieved in antibiotic therapy. AFN-1252 is a potent antibiotic against Staphylococcus aureus that targets the enoyl-acyl carrier protein reductase (FabI). A thorough screen for AFN-1252-resistant strains was undertaken to identify the spectrum of mechanisms for acquired resistance. A missense mutation in fabI predicted to encode FabI(M99T) was isolated 49 times, and a single isolate was predicted to encode FabI(Y147H). AFN-1252 only bound to the NADPH form of FabI, and the close interactions between the drug and Met-99 and Tyr-147 explained how the mutations would result in resistant enzymes. The clone expressing FabI(Y147H) had a pronounced growth defect that was rescued by exogenous fatty acid supplementation, and the purified protein had less than 5% of the enzymatic activity of FabI. FabI(Y147F) was also catalytically defective but retained its sensitivity to AFN-1252, illustrating the importance of the conserved Tyr-147 hydroxyl group in FabI function. The strains expressing FabI(M99T) exhibited normal growth, and the biochemical properties of the purified protein were indistinguishable from those of FabI. The AFN-1252 Kiapp increased from 4 nm in FabI to 69 nm in FabI(M99T), accounting for the increased resistance of the corresponding mutant strain. The low activity of FabI(Y147H) precluded an accurate Ki measurement. The strain expressing FabI(Y147H) was also resistant to triclosan; however, the strain expressing FabI(M99T) was more susceptible. Strains with higher levels of AFN-1252 resistance were not obtained. The AFN-1252-resistant strains remained sensitive to submicromolar concentrations of AFN-1252, which blocked growth through inhibition of fatty acid biosynthesis at the FabI step.

acquired resistance, examines the biochemical properties of the resistant enzymes, and evaluates the effect of their expression on bacterial physiology. Although strains expressing FabI mutants had increased resistance to AFN-1252 inhibition, they still had submicromolar MICs for AFN-1252 due to on-target inhibition of FabI.

EXPERIMENTAL PROCEDURES
Bacteria and Molecular Biology-All strains used in this work were derivatives of S. aureus strain RN4220 (27). The MICs were determined by growing strains to an optical density at 600 nm (A 600 ) of 1.0 and diluted 30,000-fold in Luria-Bertani (LB) medium. A 10-l aliquot of diluted cells was added to each well of a U-bottom 96-well plate containing 100 l of LB medium with the appropriate concentration of compound. The plate was incubated at 37°C for 20 h, and cell growth was determined using a Fusion plate reader at 600 nm. Cells grown in medium containing dimethyl sulfoxide (DMSO; 1%) were used as the control for 100% growth. LB medium containing 0.1% Brij-58 at 37°C with or without supplementation with 133 M anteiso-15:0, 66 M anteiso-17:0, and 10 M lipoate was used to test the influence of exogenous fatty acids on growth.
The fabI gene from S. aureus was cloned into pET15b (Novagen), overexpressed, and purified as described previously (6). The plasmids expressing the FabI(M99T) and FabI(Y147H) mutants were generated by mutagenizing the pET15b-fabI plasmid using the QuikChange mutagenesis kit from Agilent Technology, following the manufacturer's protocol. The mutagenesis primers for the M99T mutant were 5Ј-tgtatatcattcaatcgcatttgctaatacggaagacttacgcgg and 5Ј-ccgcgtaagtcttccgtattagcaaatgcgattgaatgatataca. The primers for the Y147H mutant were 5Ј-tagcattgttgcaacaacacatttaggtggcgaattcgc and 5Ј-gcgaattcgccacctaaatgtgttgttgcaacaatgcta. Expression and purification of the mutant proteins were the same as those used for the wild-type protein. The FabI(M99T,Y147H) double mutant was generated by sequential mutagenesis of the FabI(M99T) plasmid with the FabI(Y147H) primers. The wildtype and mutant fabI genes were cloned into the plasmid pCL15, an S. aureus expression vector, and the plasmids were transformed into strain RN4220 as described previously (26).
Selection of AFN-1252-resistant Mutants-AFN-1252-resistant mutants were selected on LB-agar plates containing 40 ng/ml AFN-1252. An aliquot of 200 l of A 600 ϭ 1 (1 ϫ 10 8 cells) S. aureus strain RN4220 was spread on a 10-cm LB-agar plate containing AFN-1252. The plate was incubated at 37°C for 48 h, and on average, 5-8 resistant colonies were found per plate. Colonies that grew on the plate were restreaked onto another AFN-1252-containing LB agar plate to purify individual clones. The sequences of the fabI genes of the confirmed mutants were amplified via PCR and sequenced. Strains RN4220, MWF32 (FabI(M99T)), and MWF33 (FabI(Y147H)) were subjected to additional rounds of selection on LB-agar plates containing 0.2-1 g/ml AFN-1252 as described above. No resistant colonies were isolated at this AFN-1252 concentration.
Synthesis of Crotonyl-ACP-The S. aureus acpP gene was amplified via Taq polymerase (Invitrogen) from S. aureus genomic DNA and cloned into the pET-15b plasmid (Novagen) at the NdeI and BamHI restriction sites. The resulting plasmid was transformed into BL21 DE3 (Novagen) cells. Following a 3-h induction of log phase cells, the cells were lysed, and the amino-terminal His 6 -tagged ACP was purified using nickel-nitrilotriacetic acid affinity chromatography. A yield of 40 mg of pure apoACP was obtained from each liter of culture. The protein was dialyzed against 20 mM Tris, pH 8.0, to remove the imidazole from purification. The His 6 tag was cleaved from apoACP by treatment with thrombin (5 units/mg purified protein) overnight at 4°C. Thrombin and the cleaved histidine tag were separated from the ACP via size exclusion chromatography. The resulting apoACP had a Gly-Ser-His addition at the amino terminus of the protein. Crotonyl-ACP was biosynthesized by incubating 250 M crotonyl-CoA with 200 M apo-SaACP, 10 mM MgCl 2 , and 1 M purified Streptococcus pneumoniae ACP synthase in 50 mM Tris-HCl, pH 7.0, at 30°C for 2 h (28). Completion of the reaction was determined by analysis using a 15% polyacrylamide gel containing 0.5 M urea gel to separate apoACP from crotonyl-ACP. The His-tagged ACP synthase was removed using a 0.2-ml nickel-nitrilotriacetic acid column. The flow-through containing the crotonyl-ACP was concentrated with an Amicon Ultra centrifugal filter (3 kDa cut-off). The MgCl 2 and CoA were then removed using Zeba-Spin desalting columns.
Real-time PCR for Analysis of Gene Expression-The strains were grown to an A 600 of 1.0. Cells were washed with 1 ml of RNAlater solution from Ambion and treated with Lysostaphin for 15 min at room temperature. Total RNA was isolated from the bacterial cells using the RNAqueous kit (Ambion) per the manufacturer's instructions, including the LiCl precipitation. The pelleted RNA was resuspended in nuclease-free water, and a 10-g aliquot was treated with Ambion's Turbo DNA-free kit to remove genomic DNA. First-strand cDNA was generated from 500 ng of total RNA using random primers and Super-Script II reverse transcriptase (Invitrogen) according to the manufacturer's protocol. Quantitative real-time PCR was performed using the ABI Prism 7700 sequence detection system. Experimental samples were run in triplicate; negative controls (distilled water) and RNA samples without the reverse transcription step were run in duplicate. PCRs contained SYBR Green PCR Master Mix (Applied Biosystems), a 150 nM concentration of each primer, and cDNA synthesized from 10 ng of total RNA. The expression level of various housekeeping genes (proC, gyrB, gmk, glyA, rpoD, rho, recF, and pyk) was checked, and gyrB was determined to be the calibrator least changed by AFN-1252 treatment and subsequently was used as the control to normalize mRNA levels between samples. Template curves of seven points ranging from 50 ng to 50 pg of total RNA input were run for each primer set and analyzed for linearity and relative efficiency of the PCR as compared with the control. Dissociation curves were generated after each real-time PCR run to check for the presence of nonspecific amplification.
Analysis of [ 14 G200 column equilibrated with 20 mM Tris HCl, 150 mM NaCl,  pH 7.8, at 4°C. Then the sample (100 l) was applied to the column, the column was eluted with 20 mM Tris-HCl, 150 mM NaCl, pH 7.8, at 1 ml/min, and 200-l fractions were collected. The fractions were added to 3 ml of ScintSafe scintillation solution and counted in a Beckman LS6500 scintillation counter.
ACP Immunoblotting-Cultures (50 ml) of strains RN4220, MWF32, and MWF33 were grown in LB medium to an A 600 of 0.5 at 37°C. AFN-1252 (final concentration of 100 ng/ml for strain RN4220 and 2 g/ml for the mutant strains) was added to the culture. Control samples had only DMSO (1%) used to dissolve the drug. The treated cultures were incubated for 30 min; the cells were harvested and lysed; and the cell extracts were fractionated on a 15% polyacrylamide, 0.5 M urea gel and immunoblotted using anti-ACP antibodies (29) to determine the ACP species (26). Protein from the same number of cells was loaded into each lane by normalizing the extract volume to the A 600 of the culture.
Acetate Labeling-[ 14 C]Acetate labeling experiments were conducted to measure the effect of AFN-1252 on lipid biosynthesis in strains RN4220, MWF32, and MWF33 when treated with AFN-1252. Starter cultures were grown overnight, and the starter culture was back-diluted to an initial A 600 of 0.1 and grown until A 600 reached 0.5 at 37°C and 225 rpm shaking. The culture was split into 10-ml aliquots, and each aliquot was incubated with the appropriate concentration of AFN-1252 and 10 Ci of [ 14 C]acetate for a single doubling time (30 min for strains RN4220 and MWF32, 120 min for strain MWF33). The cells were collected by centrifugation and washed twice with phosphate-buffered saline. Lipids were harvested from pelleted cells using the method of Bligh and Dyer (30). [ 14 C]Acetate incorporation was measured by counting the lipid extracts on an LS6500 multipurpose scintillation counter. Measurements were made in duplicate, and the averages with S.E. values were reported.
Analysis of FabI Mutants by Heterologous Expression in E. coli-The Met-99 residues of the S. aureus fabI gene in the Bluescript vector (6) was mutagenized to other amino acids using the QuikChange site-directed mutagenesis kit (Agilent) and primers generated by the QuikChange Primer Design program (Agilent). The changes were verified by DNA sequencing. The Bluescript vectors harboring the mutant fabI genes were transformed into E. coli strain JP1111 (fabI(Ts)) via electroporation. Strain JP1111 harboring pBluescript expression vector containing the mutant fabI gene were streaked on LB-carbenicillin plates at 42°C to determine if the S. aureus FabI mutants supported cell growth. These same vectors were also transformed into E. coli strain ANS1 (tolC) to rank the AFN-1252 resistance phenotype of the mutants by determining their ability to shift the MIC for AFN-1252 as described previously (26). Strain ANS1 cells harboring the empty vector without the S. aureus fabI gene were included as control.
FabI Enzymatic Assays-The FabI enzymatic progress was determined by measuring the conversion of NADPH to NADP ϩ at 340 nm. The enzyme reactions were performed in a 100-l volume in Costar UV half-area 96-well plates with a SpectraMax 340 instrument taking 340-nm readings at 10-s intervals at 30°C. For velocity measurements, FabI enzyme buffered in 100 mM glutamate, pH 7.8, was added to either 50 M crotonyl-ACP or 200 M crotonyl-CoA and 250 M NADPH buffered in 20 mM Tris-HCl, pH 7.8. Assays for E. coli FabI were the same except that NADH was substituted for NADPH instead. For determination of K m of NADPH for the wild type and mutant FabI enzymes, 50 nM enzyme was added to 50 M crotonyl-ACP and 15, 25, 50, 75, 100, 150, or 250 M NADPH. The reaction was mixed for 10 s by the plate reader, and data were acquired at 10-s intervals for 10 min. Initial velocities were calculated from the initial linear phase of the progress curve. All kinetic experiments were run in duplicates, and parameters were fit to the duplicate data sets. Determination of K m of crotonyl-ACP was similar, except with 200 M NADPH and 2.5, 5, 7.5, 10, 15, 25, and 50 M crotonyl-ACP.
The affinity of AFN-1252 for FabI was determined by measuring the initial velocity with 50 nM enzyme, 200 M NADPH, 50 M crotonyl-ACP, and 15.6, 31.2, 62.5, 125, 250, 500, or 1000 nM AFN-1252. The affinity of triclosan for FabI was determined similarly except that FabI was incubated with triclosan and 200 M NADP ϩ for 10 min before starting the assay to account for the slow binding nature of triclosan (3). Under standard steady state conditions, inhibitors are kept at concentrations 10-fold or more above the concentration of the enzyme, so that that formation of the enzyme⅐inhibitor complex does not alter the concentration of free inhibitors. However, AFN-1252 had affinity values in the nanomolar range, whereas nanomolar concentrations of FabI are necessary to generate a detectable signal in our kinetic assay. Thus, the data were fit to the Morrison quadratic equation for fitting tight binding inhibitors (Equation 1), where v 0 is the velocity with no inhibitor, v i is the velocity at the given concentration of inhibitor, [E] T is the total concentration of FabI in the reaction, [I] T is the total concentration of inhibitor in the reaction, and K i app is the apparent dissociation constant of the inhibitor (31).
Briefly, the Morrison equation allows the determination of affinity in terms of free and bound concentrations of enzyme and inhibitor, accounting for the impact of enzyme⅐inhibitor binding on the free concentration of inhibitor. For determination of the slow binding mechanism of triclosan to FabI(M99T), 50 nM enzyme was added to a solution of 400 M NADPH, 120 M crotonyl-ACP, and 0.25, 0.5, 0.75, 1, 1.25, 1.5, or 2 M triclosan. The reaction progress was measured over 5 min. Data were analyzed via the method of determining the mechanism of slow binding inhibitors described by Copeland (32). Briefly, the progress curve was fit to Equation 2, where [P] is the concentration of product at time t, v i is the initial velocity, v s is the steady state velocity, and k obs is the rate constant for the change from initial velocity to steady state velocity. DECEMBER 20, 2013 • VOLUME 288 • NUMBER 51

JOURNAL OF BIOLOGICAL CHEMISTRY 36263
Then k obs was plotted against the concentration of inhibitor to determine the mechanism of slow binding. If the relationship is linear, then the inhibitor is a simple slow binding inhibitor, where the on rate for the binding of the inhibitor is slow. If the relationship is hyperbolic, then the inhibitor is a slow tight binding inhibitor, where the EI complex isomerizes into an EI* complex at a slow rate.

RESULTS
Selection of AFN-1252-resistant Strains-AFN-1252 was a potent antibiotic against S. aureus strain RN4220 with a MIC of 3.9 ng/ml in LB broth (Fig. 1B). Individual clones of S. aureus strain RN4220 resistant to AFN-1252 inhibition were selected on LB-agar plates containing 40 ng/ml AFN-1252. Mutant colonies were purified, and the sequence of the fabI genes was determined in 50 of the resistant clones. Forty-nine resistant clones had the same single missense mutation in fabI (T296C) predicted to encode a FabI(M99T) protein, and one isolate had a missense mutation (T439C) predicted to encode FabI(Y147H). Strain MWF32 (FabI(M99T)) was typical of all 49 mutant strains with an AFN-1252 MIC of 250 ng/ml (Fig. 1B). Strain MWF33 (FabI(Y147H)) was consistently 2-fold more resistant to AFN-1252 than strain MWF32 with a MIC of 500 ng/ml (Fig. 1B). Strain PS01 (⌬accD) did not require a functional FabI for growth and was completely refractory to AFN-1252 inhibition. This strain cannot synthesize fatty acids due to the accD deletion, and the fact that it was not affected by AFN-1252 illustrated that the drug did not have an off-target effect in S. aureus.
The wild-type and mutant fabI genes were cloned into the S. aureus expression vector pCL15, and strain RN4220 was transformed with these expression constructs. Plasmid-driven expression of the wild-type FabI increased the MIC for AFN-1252 from 3.9 to 15.6 ng/ml, consistent with FabI as the cellular target for the drug (Fig. 1C). Expression of either FabI(M99T) or FabI(Y147H) increased the MIC to 500 or 1000 ng/ml, respectively (Fig. 1C). This analysis confirmed that FabI(Y147H) was 2-fold more resistant to AFN-1252 than FabI(M99T) in vivo. Together, the selection and expression experiments show that the expression of FabI(M99T) or FabI(Y147H) mutants were necessary and sufficient to confer increased resistance to AFN-1252. The cross-resistance of strains MWF32 and MWF33 against other FASII-targeted drugs was tested (Fig. 1D). Strain MWF33 exhibited increased resistance to the prototypical FabI inhibitor triclosan compared with the wild-type strain (from 62.5 to 500 ng/ml). This result was consistent with the isolation of a mutant fabI allele encoding FabI(Y147H) in a study of triclosan resistance (33). In contrast, strain MWF32 had increased sensitivity to triclosan (from 62.5 to 3.9 ng/ml; Fig. 1D). AFN-1252 selection did not give rise to commonly found triclosan-resistant fabI alleles predicted to encode FabI(A95V), FabI(I192S), or FabI(F204S) (3,9). The fact that triclosan had a MIC of 1 g/ml against strain PS01 (⌬accD) showed that triclosan also had a non-FASII target in S. aureus. The wild-type and AFN-1252resistant strains MWF32 and MWF33 had the same MIC for FASII drugs that targeted the condensing enzymes of FASII. These included cerulenin (31.3 g/ml), thiolactomysin (25 g/ml), platensimycin (0.8 g/ml), and platencin (0.8 g/ml). Thus, the selected fabI mutations conferred resistance only to AFN-1252 and not to drugs that targeted another pathway enzyme.
Repeated attempts at generating mutants resistant to higher concentrations of AFN-1252 by selection on plates containing 1 g/ml AFN-1252, using either the wild-type strain RN4220 or strains MWF32 or MWF33, yielded no colonies. Thus, we were unable to directly select for more resistant clones or to evolve a more resistant clone starting with either of the two resistant clones we obtained in the first round of selection. These data indicated that these two mutants were the only missense mutations in fabI that led to a significantly increased MIC for AFN-1252. We investigated this hypothesis by generating all possible missense mutations at position 99 and used heterologous expression in E. coli to rank the resistance phenotypes of each mutant based on the MIC (Table 1). FabI(M99T) was the most resistant of the single missense mutations that could occur at position 99. Also, amino acid changes at position 99 that would not be expected from the selection method because they involved either two or three base pair changes also were not more resistant than FabI(M99T) ( Table 1).
Phenotypes of the AFN-1252-resistant Strains-The growth rates of the two fabI mutant strains were compared with wild type (Fig. 2A). Strain MWF32 expressing the FabI(M99T) protein had a doubling time in LB medium that was indistinguishable from the parental strain RN4220 (27-30-min doubling time). In contrast, strain MWF33 (FabI(Y147H)) had a significantly impaired growth rate (120-min doubling time) (Fig. 2A). If the slow growth rate of strain MWF33 was due to the FabI(Y147H) protein being catalytically defective, then supplementation with exogenous anteiso-15:0/17:0 fatty acids and lipoate should cure the deficiency in FASII activity and accelerate the growth rate. Supplementation with exogenous fatty acids did restore the doubling time of strain MWF33 to a wildtype rate (35 min) ( Fig. 2A). Supplementing strains RN4220 or MWF32 with exogenous fatty acids did not increase their growth rate. The transcription of the fab genes in S. aureus is controlled by the FapR repressor (34). This repressor is released from its DNA binding sites by malonyl-CoA, which accumulates when FASII activity is reduced (35). Thus, a hallmark of FASII inhibition in S. aureus is an increase in the transcription of the fab genes (26). The mRNA levels of plsX in the fapR-plsX-fabD-fabG operon, fabH in the fabH-fabF operon, and fabI were measured by quantitative RT-PCR to determine if the mutations compromised FabI function in vivo, which would be reflected by a compensatory up-regulation of gene expression in the mutant strains. All three genes exhibited increased expression in both mutant strains, indicating that both fabI mutations resulted in catalytically compromised enzymes and triggered the up-regulation of the FapR regulon (Fig. 2B). These data suggested that deficiencies in the activity of the mutant fabI alleles led to the compensatory up-regulation of pathway enzymes to maintain the rate of FASII. In the case of strain MWF33, these compensatory changes were not sufficient to restore the normal growth of the strain (Fig. 2A). The elevated fab gene expression in strain MWF32 suggested that FabI(M99T) was also catalytically deficient in vivo but that the compensatory up-regulation of the fab genes (including fabI) was sufficient to prevent a deficit in fatty acid production resulting in a normal growth phenotype. Methionine is the only residue found at position 99 in S. aureus FabIs, and the gene expression data indicated that FabI(M99T) was catalytically compromised in vivo. These data provided insight into why we were not able to select for a FabI(M99T,Y147H) double mutant using higher concentrations of AFN-1252 starting with either strain MWF32 or MWF33. The elimination of these two clearly important drug-protein interactions should give rise to a more resistant enzyme, and indeed when we synthesized the FabI(M99T,Y147H) double mutant and tested it in the E. coli system, the MIC for AFN-1252 was shifted more than either of the single mutants alone (Table 1). Thus, the FabI(M99T,Y147H) was more resistant to AFN-1252 than either FabI(M99T) or FabI(Y147H) but was also predicted to be more catalytically impaired in vivo than the single mutants. We tested this idea by

Complementation and AFN-1252 sensitivity for missense mutations affecting positions 99 and 147 of S. aureus FabI
Complementation was scored as growth (Y) or no growth (N) at 42°C when plasmid expression of the indicated mutants was transformed into E. coli strain JP1111 (fabI(Ts)). The sensitivity to AFN-1252 was determined after transforming the plasmids into E. coli strain ANS1, and the MICs were determined as described under "Experimental Procedures." DECEMBER (26); therefore, we determined if FASII was still inhibited in the mutant strains (Fig. 2C). AFN-1252 most potently inhibited [ 14 C]acetate incorporation in strain RN4220 and was less effective in strain MWF32, and strain MWF33 was the most resistant. The ACP pools of AFN-1252-treated wild-type and mutant strains were compared with the ACP pool of the untreated cells (Fig. 2D). In untreated cells, non-esterifed ACP was the only clearly discernible protein form. In AFN-1252treated cells, a series of short-chain acyl-ACP accumulated. Treatment of the two mutant strains with higher concentrations of AFN-1252 resulted in the same pattern of acyl-ACP accumulation, as noted in the wild-type strain. Thus, strains MWF32 and MWF33 both exhibited increased resistance to AFN-1252, but the MIC for AFN-1252 in these strains was still attributed to on-target inhibition of the mutant FabI enzymes.

Molecular Mechanism of AFN-1252 Resistance
Enzymology of FabI Mutant Proteins-The wild-type and the mutant FabI enzymes were expressed and purified. FabI(Y147H) was not stable to short term storage and precipitated from the storage buffers, making this enzyme difficult to characterize. Freshly isolated FabI(Y147H) exhibited a marked catalytic defect (Fig. 3A). The calculated FabI(Y147H) k cat was 2.3 Ϯ 0.8 min Ϫ1 compared with the FabI k cat of 30.8 Ϯ 5.5 min Ϫ1 . This catalytic deficiency of FabI(Y147H) was consistent with the fatty acid-dependent growth phenotype of strain MWF33 (Fig.  2A). The roles played by Tyr-157 and Lys-164 located in the conserved YX 6 K motif in NADP(H) binding and hydride transfer to carbon-3 of the substrate are an established characteristic of this protein family (36). Whether the Tyr-147 hydroxyl was important for activity was tested by constructing the FabI(Y147F) mutant. The E. coli MIC analysis system confirmed that the FabI(Y147F)-expressing cells did not have increased resistance to AFN-1252 compared with FabI ( Table  1), suggesting that FabI(Y147F) bound AFN-1252 with approximately the same affinity as FabI. Like FabI(Y147H), purified FabI(Y147F) was not stable and had a similar low k cat (3.9 Ϯ 0.3 min Ϫ1 ). The conserved Tyr-147 may form a hydrogen bond with the thioester carbonyl of enoyl-ACP to align the substrate in the active site and promote the reaction. This interpretation for the role for Tyr-147 in catalysis was also arrived at based on molecular dynamics simulations of enoyl-ACP binding to E. coli FabI (37).
Whether AFN-1252 bound to the free enzyme or to one of the nucleotide-bound forms was determined by incubating [ 14 C]AFN-1252 with FabI, FabI plus NADP ϩ , or FabI plus NADPH. The samples were then fractionated with a size exclusion column, the location of FabI was determined by UV detection, and [ 14 C]AFN-1252 was determined by scintillation counting. [ 14 C]AFN-1252 co-eluted with the protein only in the presence of NADPH, illustrating that AFN-1252 only bound to the FabI⅐NADPH complex (Fig. 3B).
The high affinity binding of AFN-1252 presented a challenge in measuring its binding to FabI or its mutant derivatives. The usual FabI assay uses enoyl-N-acetylcysteamines or enoyl-CoAs as substrate analogs (11,38). However, the low affinity of FabI for enoyl-CoA meant that this assay required low micromolar FabI concentrations to observe detectable rates. These high protein concentrations cannot be used to kinetically determine the affinity of nanomolar inhibitors with any accuracy. Therefore, our assays used crotonyl-ACP, which allowed us to measure initial rates using 20 -50 nM FabI. Nonetheless, AFN-1252 binding was still high enough that the formation of the FabI⅐NADPH⅐AFN-1252 complex affected the free concentration of the inhibitor, necessitating the use of the Morrison quadratic equation (31) to take this effect into account in calculating binding constants (see "Experimental Procedures"). The affinity of AFN-1252 for FabI and FabI(M99T) was measured kinetically with saturating concentrations of NADPH and crotonyl-ACP. As expected, the FabI(M99T) mutant enzyme was more resistant against AFN-1252, with a K i app of 69 nM versus 4 nM for FabI under saturating substrate concentrations, accounting for the increased resistance against AFN-1252 at the cellular level (Fig. 3C). The low activity of FabI(Y147H) (Fig. 3A) meant that micromolar concentrations of enzyme were needed in the assays, precluding an accurate measurement of AFN-1252 binding to this mutant enzyme. However, all of the biological data indicated that FabI(Y147H) was  Table 1).
The FabI(M99T) enzyme had turnover rates as fast as the wild type enzyme (Fig. 3A). The kinetic parameters of the FabI(M99T) mutant were compared with wild type enzyme, and no significant differences were observed. The FabI(M99T) mutant had a K i app of 42.1 Ϯ 8.6 M for NADPH and 5.5 Ϯ 0.8 M for crotonyl-ACP compared with 55 Ϯ 7.5 M for NADPH and 14.4 Ϯ 5.5 M for crotonyl-ACP for FabI (Fig. 3D). FabI exhibited substrate inhibition by crotonyl-ACP (K i app ϭ 20.1 Ϯ 8.4 M), whereas the M99T mutant did not, although the observed substrate inhibition was probably not physiologically relevant because enoyl-ACPs were not normally detected in vivo (Fig. 2D), and thus their concentrations were probably below the FabI K m .
Triclosan Binding in the M99T Mutant-The increased susceptibility of strain MWF32 to triclosan was explored by examining the time-dependent nature of triclosan binding to FabI. Triclosan binding to E. coli FabI exhibited slow binding kinetics, with the full effect of the drug occurring 50 -80 s after the reaction was initiated (Fig. 4A). Triclosan binding to S. aureus FabI was significantly faster, reaching equilibrium within 20 s (Fig. 4B). However, triclosan binding to FabI(M99T) was a slower process, taking close to 100 s from the onset for the full effect of triclosan inhibition to be manifested (Fig. 4C). To compare affinities for triclosan to the FabI and FabI(M99T) enzymes, both enzymes were preincubated with triclosan and NADP ϩ for 10 min to allow inhibitor to bind before the start of the enzymatic reaction. These experiments showed that FabI(M99T) had a higher affinity for triclosan with a K i app ϭ 7.9 nM compared with K i app ϭ 13.1 nM for FabI (Fig. 4D). The FabI(M99T) enzyme was added to assay mixtures containing different concentrations of triclosan, and the progress curves were measured (Fig. 4C). The k obs term, which measures the rate of the onset of inhibition, was determined from the progress curve for each concentration of triclosan and plotted against the concentration of triclosan (Fig. 4C, inset). The k obs versus [triclosan] plot fit best to a straight line indicating reversible, slow binding of triclosan to FabI(M99T), with an apparent on rate (k 3 ) of 1.7 ϫ 10 4 M Ϫ1 s Ϫ1 . This "slow binding" behavior of triclosan is characteristic of its interactions with many FabIs (7,12,39), and the importance of slow off rates to the in vivo drug efficacy has recently become clear (12,39). However, the basis for the lower MIC in our closed system was attributed to the higher affinity of triclosan for the FabI(M99T)⅐NAD ϩ complex.

DISCUSSION
This work identifies the M99T mutation as the most common mechanism for the acquisition of increased resistance to AFN-1252 in S. aureus. Met-99 resides on a conformationally flexible loop that covers the FabI active site following the binding of NADPH and substrate (11). The helical segment containing Met-99 in the FabI⅐NADPH⅐AFN-1252 ternary complex is not a well conserved structural element in bacterial FabIs. Residues 95-117 are disordered in the FabI crystal structure to allow the entry of substrates into the active site and form a structured lid over the active site in the FabI⅐NADP(H), FabI⅐NADP(H)⅐triclosan, and FabI⅐NADP(H)⅐AFN-1252 complexes (11,40). Met-99 forms a close interaction with the oxotetrahydronaphthyridine portion of AFN-1252 that contributes a significant hydrophobic interaction that is clearly required for high affinity AFN-1252 binding (Fig. 5). An analysis of FabI protein sequences deposited in the NCBI Microbial Genomes database shows that staphylococcal species uniformly have Met-99, whereas other FabIs have more hydrophilic and smaller residues in this position. This fact suggests that the interaction between Met-99 and AFN-1252 accounts for the selective anti-staphylococcal activity of the drug (22). Although FabI(M99T) did not have a pronounced catalytic defect using crotonyl-ACP as substrate, the actual in vivo substrates are an elongating series of branched-chain enoyl-ACPs. This blind spot in our analysis may be relevant because the up-regulation of the lipid biosynthetic genes in strain MWF32 compared with strain RN4220 illustrates that the FapR genetic regulatory switch is increasing protein expression to compensate for a deficiency in the pathway. Although the nature of this inferred defect is not clear, the cells expressing the mutant protein are able to compensate and do not have an observable growth phenotype in the laboratory. Further experiments will be required to determine if strains with the FabI(M99T) defect are attenuated for growth in animal models.
FabI(Y147H) was the only other AFN-1252 mutant found in our study (1 in 50). In contrast to Met-99, Tyr-147 is a conserved residue in eubacterial FabIs. The 3-methylbenzofuran rings of AFN-1252 bind in the catalytic pocket, making strong hydrophobic interactions with Tyr-147 (Fig. 5). The absence of this interaction in FabI(Y147H) accounts for the lower affinity of AFN-1252. The low catalytic activity of FabI(Y147H) prevented the calculation of a binding constant from kinetic experiments. However, the collective in vivo data suggest that the AFN-1252 K i for FabI(Y147H) is 2-fold higher than for FabI(M99T). In bacterial FabIs, this invariant tyrosine residue is flanked by a serine/threonine and a large hydrophobic residue, such as leucine or phenylalanine. Structurally, the tyrosine residue is at a spatially conserved position forming a part of the active site of the FabI enzyme in the crystal structures of FabI from S. aureus (11,40), E. coli (41), Helicobacter pylori (42), Bacillus subtilis (43), and Bacillus cereus (43). The analysis of these structures, our site-directed mutagenesis, and molecular dynamics simulations (37) all point to Tyr-147 promoting the reaction by hydrogen-bonding to the substrate thioester carbonyl. This hypothesis explains the compromised catalytic activity of FabI(Y147H) and the marked growth defect of strain MWF33.
Understanding the number and nature of AFN-1252 resistance mutations in FabI has important implications to the potential therapeutic use of AFN-1252. The low number of AFN-1252-resistant mutations is attributed to many of the key interactions that occur between AFN-1252 and the backbone amide of Ala-97 and the hydrogen bond network and stacking interactions that bridge the drug, protein, and NADPH cofactor (25). It is difficult to envision missense mutations that would selectively break these key interactions and still preserve cofactor binding and catalysis. Both mutant strains are more resistant to AFN-1252, but because AFN-1252 is such a high affinity inhibitor, they remained sensitive to 0.25-0.5 g/ml of the drug. This suggests that animals infected with the resistant strains could be treated by administering higher concentrations of AFN-1252 because the MIC for AFN-1252 in the resistant strains is still lower than most antibiotics, such as linezolid (1-4 g/ml), a front-line drug used to treat S. aureus (22). Animal models testing the efficacy of AFN-1252 deliver doses of the drug that generate serum levels over 1.5 g/ml (44). The idea that AFN-1252-resistant mutants could be successfully treated by simply delivering the same or higher doses of drug will be important to test in animal infection models. It is also significant that we were unable to obtain S. aureus mutants that were resistant to higher AFN-1252 concentrations than FabI(M99T) and FabI(Y147H) even when starting with strains MWF32 and MWF33. We reasoned that the FabI(M99T,Y147H) double mutant would be more resistant to AFN-1252 due to the elimination of two key interactions and confirmed this hypothesis by constructing and analyzing the double mutant enzyme. However, the in vivo gene expression data indicate that both FabI(M99T) and FabI(Y147H) are functionally compromised, leading us to conclude that the FabI(M99T,Y147H) double mutant is even more deficient. This idea was corroborated by the finding that a strain possessing a plasmid expressing FabI(M99T,Y147H) grew very poorly on low concentrations of AFN-1252 that knocked out the endogenous wild-type FabI. These data suggest that the double mutant does not arise in our selection schemes because a single copy of this defective enzyme cannot support the growth of S. aureus.
Our study also provides insight into cross-resistance against drugs targeting the same enzyme. Whereas AFN-1252 binds to the FabI⅐NADPH complex (Fig. 5) and triclosan binds to the FabI⅐NADP ϩ complex (3), the two inhibitors occupy the same substrate/product binding pocket. Of the two AFN-1252-resistant mutants, FabI(Y147H) has increased resistance against triclosan, whereas FabI(M99T) is actually more sensitive to triclosan. Future infection models will be required to determine if the poor growth of the strain MWF33 expressing the FabI(Y147H) protein compromises the virulence of S. aureus or whether the strain can obtain sufficient fatty acids from the host to support colonization. CG400462, a FabI inhibitor developed by CrystalGenomics, gave rise to the same two resistant mutants as we found with AFN-1252 (14), suggesting that it binds in a similar manner to the FabI⅐NADPH complex. Other FabI-targeted compounds, including CG400549 (13), MUT056399 from Mutabilis (15), and triclosan analogs from Xu et al. (9), give rise to resistant mutants (F204C/L/S/F and A95V) more typically found for triclosan (45,46) that were not detected in our study. Thus, the concern that spread of triclosan resistance would invalidate the use of FabI-targeted drugs (47) is limited to the FabI(Y147H) mutant in the case of AFN-1252.