Structure of Acyl Carrier Protein Bound to FabI, the FASII Enoyl Reductase from Escherichia coli*

Acyl carrier proteins play a central role in metabolism by transporting substrates in a wide variety of pathways including the biosynthesis of fatty acids and polyketides. However, despite their importance, there is a paucity of direct structural information concerning the interaction of ACPs with enzymes in these pathways. Here we report the structure of an acyl-ACP substrate bound to the Escherichia coli fatty acid biosynthesis enoyl reductase enzyme (FabI), based on a combination of x-ray crystallography and molecular dynamics simulation. The structural data are in agreement with kinetic studies on wild-type and mutant FabIs, and reveal that the complex is primarily stabilized by interactions between acidic residues in the ACP helix α2 and a patch of basic residues adjacent to the FabI substrate-binding loop. Unexpectedly, the acyl-pantetheine thioester carbonyl is not hydrogen-bonded to Tyr156, a conserved component of the short chain alcohol dehydrogenase/reductase superfamily active site triad. FabI is a proven target for drug discovery and the present structure provides insight into the molecular determinants that regulate the interaction of ACPs with target proteins.

structure of ACP bound to AcpS (15), these studies confirmed the importance of the ACP recognition helix (helix ␣2) in FASII protein binding and identified a patch of basic residues on the E. coli ␤-ketoacyl-ACP synthase (FabH) and ␤-ketoacyl-ACP reductase (FabG) enzymes responsible for interaction with the ACP helix ␣2 (16,17).
The efforts in our laboratory are focused on the FASII enoyl reductase enzyme FabI, a target for antibacterial diazaborine compounds (18) and triclosan (19 -23), whereas the anti-TB drug isoniazid inhibits InhA, the FabI homolog from M. tuberculosis (24 -26) (Scheme 1). Modeling studies, based on the information provided by the interaction of ACP with FabH and FabG (16,17), suggest that ACP should interact with a cluster of basic residues adjacent to the FabI substrate binding loop. This loop is disordered in binary FabI-cofactor complexes, but becomes ordered in the ternary FabI⅐NAD ϩ ⅐triclosan complex. The ordered loop provides two entries into the active site, termed the major and minor portals. Sacchettini and co-workers (27) have determined the structure of a C16 substrate bound to InhA and have discussed the role of the major and minor portals in substrate recognition by the FabI enzymes. The orientation of the C16 substrate in InhA suggests that substrates enter the FabI active site through the major portal.
In the current study we have determined the structure of ACP bound to the E. coli FabI enzyme. X-ray crystallographic data obtained from a complex between FabI and dodecenoyl-ACP revealed most of the main chain electron density for both FabI and ACP. However, the observed relative orientation of ACP and FabI leaves the ACP Ser 36 residue too far from the active site to deliver the substrate through the major portal.
Because some aspects of the structure were not resolved by the crystallographic data, we employed computational methods to model the missing details and importantly to ascertain whether the ACP could deliver substrate into the FabI active site in the observed complex. Through molecular dynamics simulations, we generated a model for a productive complex between ACP and FabI. The details of the interaction between FabI and ACP in the resulting model are supported by mutagenesis studies, and provide the first detailed description of ACP recognition by a FASII enzyme. Intriguingly, the structural data indicate that the substrate enters the FabI active site through the minor portal and furthermore, suggest that the substrate thioester carbonyl group does not form a hydrogen bond with Tyr 156 , a conserved active site residue.

MATERIALS AND METHODS
Preparation of Substrates and Enzymes-Plasmids for wildtype FabI and the Y156F mutant were available from a previous study (23). The Y146F, K201A, K201E, R204A, R204E, K205A, and K205E FabI mutations were introduced using the QuikChange mutagenesis kit (Stratagene). Wild-type and mutant FabI proteins were overexpressed and purified as described previously (23). Trans-2-dodecenoyl-CoA (DD-CoA) was synthesized from trans-2-dodecenoic acid using the mixed anhydride method (28). Apo-ACP was a gift from Dr. J. Shanklin and Dr. S. Booker. Apo-ACP was also expressed and purified from plasmid pET23b as previously described (29,30) with the following modifications (Dr. S. Booker). After growing cultures at 37°C until an A 600 of 0.6 was reached, casamino acids were added to a final concentration of 2 g/liter and expression was induced by the addition of 500 M isopropyl 1-thio-␤-D-galactopyranoside. Following an additional 4 h of growth at 30°C, centrifugation yielded 8 g/liter of wet cell paste that was subsequently frozen. Frozen cell pellets (50 g) were then resuspended in 50 ml of 25 mM MES, pH 6.1, containing 200 mM NaCl (buffer A) and sonicated for 6 min using 30-s pulses at 4°C. Cellular debris was removed by centrifugation at 33,000 rpm for 1 h at 4°C and the supernatant was loaded onto a Q-Sepharose column (8 ml) preequilibrated with 25 ml of buffer A. The column was washed with 50 ml of buffer A and ACP was eluted using a linear gradient (50 ml) of NaCl from 200 to 850 mM NaCl in buffer A. Fractions were analyzed by 18% SDS-PAGE, pooled, concentrated using an Amicon membrane (YM3, NMWL 3,000) and stored at Ϫ80°C. Conformationally sensitive SDS-PAGE indicated that the ACP was predominantly in the apo form and ESI mass spectrometry revealed that the apo-ACP sample was comprised of two forms in which the N-terminal Met was present or had been cleaved. The apo-ACP was used without further purification for the synthesis of trans-2-dodecenoyl-ACP (DD-ACP).
DD-ACP was a gift from Dr. M. Schaeffer and was also synthesized from DD-CoA and apo-ACP using E. coli AcpS (31), which was overexpressed and purified from E. coli as described previously (32). Briefly, 0.9 mg of apo-ACP was incubated with 50 M DD-CoA (1.4-fold excess) and 50 g of AcpS in 1.4 ml of 50 mM Tris-HCl, 25 mM MgCl 2 , 1 mM dithiothreitol, pH 7.5, buffer for 1 h at 30°C, followed by quenching the reaction by placing it into a dry ice/ethanol bath for 5 min. Subsequently, an SCHEME 1. The type 2 fatty acid biosynthesis pathway and FabI inhibitors.

Structure of ACP Bound to FabI
equal volume of isopropyl alcohol was added and the reaction mixture was incubated for 2 h at 10°C. AcpS was removed by centrifugation at 6000 rpm for 15 min and the supernatant was applied to a 1-ml Q-Sepharose column, equilibrated with 20 mM BisTris, 1 mM dithiothreitol, pH 6.5 (Buffer B), containing 50% isopropyl alcohol. The column was washed three times with Buffer B containing 50% isopropyl alcohol, and then five times with Buffer B alone. DD-ACP was eluted with 5 column volumes of Buffer B containing 600 mM NaCl. The fractions containing DD-ACP were identified by SDS-PAGE, pooled, concentrated, dialyzed into 20 mM Tris⅐HCl, pH 7.0, and stored at Ϫ80°C. The concentration of protein was determined by a BCA assay.
Kinetic Assays-All kinetic assays were performed on a Cary 100 Bio (Varian) spectrophotometer at 25°C in 30 mM PIPES, 150 mM NaCl, pH 8.0. Kinetic parameters were determined by following the oxidation of NADH to NAD ϩ at 340 nm (⑀ ϭ 6.3 mM Ϫ1 cm Ϫ1 ). k cat and k cat /K m for DD-CoA and DD-ACP were determined at a fixed, saturating concentration of NADH (250 M) and by varying the concentration of DD-CoA (5-85 M) or DD-ACP (1-200 M), respectively. Kinetic parameters were obtained by fitting the initial velocity data to the Michaelis-Menten equation using Grafit 3.09b (Erithacus Software Ltd.).
Crystallization, Data Collection, and Refinement-The FabI⅐NAD ϩ ⅐DD-ACP complex was crystallized by the hanging drop vapor diffusion method. Crystals were grown at 22°C from a protein solution containing FabI, NAD ϩ , and DD-ACP in a 1:10:1 ratio mixed with a precipitant solution containing 18 -22% PEG 4000 and 100 mM HEPES, pH 7.0. Crystals were flash-frozen in liquid nitrogen with 30% (w/v) glycerol as cryoprotectant and data were collected at beam line X26C at the National Synchrotron Light Source at Brookhaven National Laboratory. Data were indexed, integrated, and scaled using the HKL software package (33). Crystals belonged to the hexagonal space group P6 5 22, with a ϭ b ϭ 127.7 Å, and c ϭ 206.7 Å and contained one FabI dimer and one ACP molecule in the asymmetric unit. The structure was solved by molecular replacement using the program MOLREP (34) and the FabI dimer as a search model (Protein Data Bank entry 1C14 (35)). Two unambiguous solutions with R-factors of 0.418 and 0.426 and correlation coefficients 0.569 and 0.552, respectively, were obtained. Structure refinement was performed with REFMAC (36). To locate the ACP molecule, several cycles of rigid-body refinement followed by solvent flattening and NCS averaging were carried out using DM (37). Additional electron density was observed after DM, which to a large extent was a good fit for the main chain of butyryl-ACP (12). All residues of the ACP were changed to Ala, because no density for the side chains was observed and the loop region 195-199 in molecule B of FabI was disordered as well. The model was improved with restrained refinement using REFMAC (36) and then manually improved using the program O (38). The structure was refined to an R free of 0.263 and R of 0.226. An omit map calculated with CNS (39) for the final model showed no additional electron density. The stereochemistry of the model was good, with r.m.s. deviations of 0.010 Å and 1.235°in bond lengths and bond angles, respectively. The structure was analyzed with SFCHECK (40) and PROCHECK (41) and 87.1% of all residues were in the core region, 11.7% allowed, 1.2% generously allowed, and 0% in the disallowed regions of the Ramachandran diagram. Crystallographic statistics are given in Table 1. The coordinates and structure factors have been submitted to the RCSB (PDB code 2FHS).
Molecular Dynamics Simulations-Computational modeling and molecular dynamics simulations were performed with the Amber suite of programs (42). The missing ACP atoms were added using the butyryl-ACP crystal structure (PDB code 1L0I) (12) by overlapping the two structures, and replacing the partial coordinate set with the coordinates from 1L0I. Similarly, the structure of triclosan bound to FabI in the presence of NAD ϩ (PDB code 1QSG) (21) was used to place the NADH cofactor. Missing FabI side chains and all hydrogens were added using Xleap. The coordinates for the phosphopantetheine moiety were taken from the crystal structure of the holo-acyl carrier protein-synthase in complex with holo-acyl carrier protein (PDB code 1F80) (15). Maestro Molecular Modeling software was used to build the acyl chain for the phosphopantetheine moiety. Force field parameters were the ff99 set for proteins (43) and published parameters (44,45) for NADH. The Amber antechamber module and GAFF force field (46) with am1bcc charges (47) were used to generate the parameters for phosphopantetheine and the attached acyl chain. To present the correct face of the crotonyl double bond to the NADH, the substrate must be bound in an s-trans conformation such that the si face of C␤ is oriented toward the NADH pro4(S) proton so that hydride transfer will generate the expected 3(S)-enoyl product (48). Because spontaneous isomerization of this double bond is unlikely to occur during relatively short molecular dynamics (MD) simulations, we modeled the crotonyl in the s-trans conformation.
where I i is the ith measurement and ͗I͘ is the weighted mean of all measurements of I. ͗I/I͘ indicates the average of the intensity divided by its average standard deviation. Numbers in parentheses refer to the respective highest resolution data shell in each data set.
where F o and F c are the observed and calculated structure factor amplitudes. R free same as R cryst for 5% of the data randomly omitted from the refinement. Ramachandran statistics indicate the fraction of residues in the most favored, additionally allowed, generously allowed, and disallowed regions of the Ramachandran diagram, as defined by the program PROCHECK (41).
Initial minimization of the FabI⅐NADH⅐ACP ternary complex was performed in a stepwise fashion by restraining the backbone C␣ atoms and allowing the side chains to move, with each step consisting of 1000 cycles. The restraints were gradually removed in each step (force constants from 10, 7, 4, 1, and 0 kcal mol Ϫ1 Å Ϫ2 ). Equilibration dynamics was performed on the minimized structure, with a constant temperature of 300 K maintained by coupling to a thermostat using the Langevin algorithm with the collision frequency set to 1 ps Ϫ1 . This reduced viscosity has been shown to facilitate rapid structural rearrangement (49). During dynamics, restraints on the backbone atoms were gradually released in a stepwise fashion in 25-ps increments (force constants 10, 7, 4 to 1 kcal/mol/Å 2 ). Further equilibration with no restraints was performed for 500 ps at 300 K. The time step was 1 fs. During all simulations, all possible nonbonded interactions were evaluated at each time step (i.e. no cutoff was employed). Solvation effects were incorporated using the Generalized Born model as implemented in Amber (50 -52). r.m.s. deviation values were calculated using the initial model structure (prior to equilibration) as a reference.
The production phase consisted of 1-ns simulations at 300 K with parameters as described above. Simulations were fully unrestrained with the exception of a distance restraint between the C-3 atom of the substrate and C-4 of the NADH nicotinamide ring, with a force constant of 10 kcal mol Ϫ1 Å Ϫ2 . The initial separation was ϳ18 Å, and the final distance was 4.6 Å. We modified Amber to apply the restraint force only to the substrate C-3 to avoid pulling NADH out of the binding pocket. After the substrate was drawn into the active site, a 100-ps simulation was performed to equilibrate the system with the substrate in the active site. Finally, this restraint was released and an additional 100 ps of fully unrestrained simulation was performed.
The sensitivity of the results to the force field parameters was studied by repeating the procedure using the ff99SB protein force field (53) and a newer variant of the GB solvation model (54,55). We recently showed that this combination of force field and solvent model was able to accurately reproduce experimental data for large conformational changes in human immunodeficiency virus type 1 protease that occur upon addition or removal of an inhibitor (56,57). After the substrate was drawn into the FabI active site, the system was simulated for an additional 2500 ps.

RESULTS
X-ray Crystallography-FabI forms a 110.9-kDa tetramer with 222 symmetry in which each monomer is formed by a central ␤-sheet that contains seven ␤-strands sandwiched by eight helices. Two FabI monomers (A and C) within the tetramer form a complex with ACP, resulting in a stoichiometry of 2:1 for the FabI-ACP interaction (Fig. 1). The ACP-bound FabI monomers and the uncomplexed FabI monomers are oriented in such a way that they form a nearly continuous ␤-sheet across the dimer interface.
Superposition of the ACP-bound FabI monomers with FabI from the triclosan-bound structure (PDB code 1QSG) (21) results in an r.m.s. deviation of 0.58 Å for 252 C␣ atoms for residues 2-194 and 200 -259. Thus, the overall structure of FabI does not change upon binding to ACP. However, the sub-strate binding loop (residues 191-200) undergoes a major conformational change upon complexation with ACP, whereas FabI helix ␣8 (residues 201-213) is shifted toward helix ␣2 of ACP (21) (Fig. 2). Although the density is missing for most of the side chains in ACP, the distance between the main chain atoms of these two helices indicates that helix ␣8 of FabI interacts with helix ␣2 of ACP with a main chain separation of about 8.5 Å, over a helix-helix interface length of roughly 11 Å. The substrate binding loop in the triclosan-bound structure, which is connected to helix ␣8, forms a lid on top of the triclosan moiety and the nicotinamide ring, thereby shielding them from the solvent. In contrast, this loop adopts an open lid conformation in the FabI-ACP structure, likely due to interaction with ACP,  although the details of this interaction cannot be elucidated from the crystallographic data due to missing side chain density. Finally, the observation that only two ACP molecules are observed in the complex with the FabI tetramer could be due to crystal packing. We attempted to dock ACP molecules to the corresponding positions on the other monomers and obtained steric conflicts between the existing ACP molecules, suggesting that the lack of 1:1 stoichiometry is biologically relevant.
Model Building and Molecular Dynamics-Because the side chains for both molecules in the region of the FabI-ACP interface were not resolved by the crystallographic data, we began our computational study by building a model that included these coordinates (see "Materials and Methods" for details), followed by a series of energy minimization and MD simulations to relax the resulting structure of the complex. The proteins were stable, with the r.m.s. deviation of FabI and ACP individually remaining below 3 Å. The relatively small contact interface between the proteins permitted a somewhat larger degree of relative motion of the two proteins, with the overall r.m.s. deviation of the system reaching a plateau of ϳ3.5 Å.
Structural analysis of the resulting model indicates that the complex is predominantly stabilized through hydrogen bonding interactions between basic residues of FabI in helix ␣8 and acidic residues of ACP in helix ␣2, consistent with the structural and modeling studies on the interaction of ACP with AcpS, FabH, and FabG (15)(16)(17). However, whereas the x-ray crystallographic data and computational modeling studies provide information on the nature of the interaction between FabI and ACP, the absence of electron density for the ACP pantetheine hindered our ability to predict how the ACP delivers the substrate into the FabI active site. Consequently, we set out to build a model of a productive FabI⅐ACP complex using the computationally refined crystallographic model as a starting point and attaching a crotonyl-phosphopantetheine group to the side chain of Ser 36 in ACP. Importantly, these simulations also help determine whether the relative positions of FabI and ACP observed in the crystal structure permit delivery of substrate to the active site.
During MD simulations of this structure, the substrate remained outside the FabI active site. This is expected because diffusion of the substrate into the protein is likely an inaccessible event during standard MD simulations on the nanosecond time scale. We thus obtained a model for the productive complex by placing a distance restraint between the crotonyl C-3 carbon and the NADH C-4 carbon, reducing the target value during a 1-ns MD simulation to draw the substrate into the active site. No other restraints were employed to enforce any particular binding mode. Subsequently, the restraint between the substrate and FabI active site was released and a further completely unrestrained simulation was performed for 100 ps.
Similar to the behavior without substrate, the individual proteins remained relatively stable during these simulations, with backbone r.m.s. deviation less than or equal to 3.0 Å as compared with the initial model. We observed a greater movement of the ACP molecule with respect to FabI, which allowed Ser 36 to be oriented toward the active site cavity. To gain more specific insight into the structural changes that occurred during substrate entry, we performed a superposition of the initial and final structures of the entire protein and then separately calculated the r.m.s. deviation values for each residue without refitting (Fig. 3). These show that the r.m.s. deviation for the majority of FabI was Ͻ2.0 Å, whereas the active site loop residues (95-115 and 190 -210) showed a much larger deviation of Ͼ5 Å. The larger fluctuations for these same loop regions even in the absence of substrate suggests that their flexibility may accommodate entry of substrate into the active site. The r.m.s. deviation values for ACP are distributed more evenly across the sequence (3.5 Å); this is not unexpected because the initial model was based on the crystal structure of ACP in the absence of FabI with the butyryl moiety accommodated in an internal cavity of the ACP.
Interactions in the Productive FabI⅐ACP Complex-The structure of the final FabI⅐ACP structure is shown in Fig. 4. ACP interacts with FabI helix ␣8 and delivers the substrate to the active site between helix ␣8 and a loop comprised of FabI residues 152-156. Analysis of this structure reveals several important interactions at the FabI⅐ACP interface and also between the phosphopantetheine and the FabI protein (Fig. 5). Residues Lys 201 , Arg 204 , and Lys 205 from helix ␣8 of FabI contact residues Asp 35 , Asp 38 , Glu 41 , and Glu 48 of ACP helix ␣2, whereas FabI Lys 201 also interacts with Gln 14 in ACP. The side chain amino group of FabI Lys 205 , which interacts with Asp 35 in ACP helix ␣2, is hydrogen bonded to the phosphopantetheine phosphate (O-7). In addition, the backbone carbonyl of FabI Lys 205 forms a hydrogen bond to the pantetheine hydroxyl group (O-10), whereas H⑀-2 of His 209 is hydrogen bonded to the pantetheine 4Љ amide carbonyl oxygen. Finally, the pantetheine 2Љ amide nitrogen forms a hydrogen bond to the backbone carbonyl of Asp 202 (Fig. 5). These interactions between the ACP pantetheine and FabI undoubtedly play an important role in stabilizing the FabI⅐ACP complex and in positioning the substrate within the active site. In the final structure, the distance between the  DECEMBER 22, 2006 • VOLUME 281 • NUMBER 51 crotonyl C-3 and the NADH pro4(S) proton is 3 Å. Analysis of the structure also reveals that the crotonyl thioester carbonyl is located 4 Å from the Tyr 146 hydroxyl group, suggesting that Tyr 146 may form a hydrogen bond to the thioester during substrate reduction (Fig. 5).

Structure of ACP Bound to FabI
To investigate the influence of the computational protocol on the results, we repeated the process using a different protein force field and solvent model (see "Materials and Methods"). We recently employed this combination to successfully simulate ligand-induced conformational changes in human immunodeficiency virus type 1 protease (56,57). Importantly, we also demonstrated that this particular implicit solvent model accurately reproduces the stability of salt bridges between protein side chains as compared with simulations in explicit solvent (58). Simulations with these parameters gave very similar results to those described above. The crotonyl thioester carbonyl is located even closer (2.7 Å) to the Tyr 146 hydroxyl group, providing further evidence that Tyr 146 may form a hydrogen bond to the thioester during substrate reduction.
Kinetic Analysis of Wild-type and Mutant FabI Enzymes-The structural studies described above identify several key interactions in the FabI⅐ACP complex that have been evaluated by replacing the basic FabI residues with acidic groups and Ala. The data in Table 2 demonstrates that replacement of Lys 201 , Arg 204 , and Lys 205 with Ala has little or no effect on the kinetic parameters for reduction of DD-CoA, whereas k cat /K m for reduction of DD-ACP is reduced 5 (Lys 201 and Lys 205 ) to 50 (Arg 204 )-fold. In addition, replacement of Arg 204 and Lys 205 with Glu causes a further reduction in k cat /K m for reduction of DD-ACP without affecting k cat /K m for the DD-CoA substrate. Similar to the Ala mutants, substitution of Glu for Arg 204 has a larger impact on substrate reduction (250-fold) compared with Lys 205 (14-fold). Finally, replacement of Lys 201 with Glu resulted in an enzyme with little activity toward either substrate and we were unable to determine accurate kinetic parameters for this mutant.
We have also examined the importance of Tyr 156 and Tyr 146 , the two active site Tyr residues, in substrate reduction. Replacement of Tyr 156 with Phe has no effect on substrate reduction in agreement with previous studies on both FabI (23) and InhA (22,28), which questioned the importance of Tyr 156 (Tyr 158 in InhA) in catalysis (59). In contrast, mutagenesis of Tyr 146 has a larger impact on catalysis, with k cat and k cat /K m for DD-ACP decreasing by around 50-fold compared with wild-type FabI. A similar decrease in kinetic parameters for Y146F was also observed for the DD-CoA substrate.

DISCUSSION
ACPs are small, acidic proteins that fulfill an essential role in metabolism through their interactions with a diverse array of  Interactions between FabI and ACP. A, interactions between ACP (cyan) and FabI (green) at the helix ␣2 (ACP)-helix ␣8 (FabI) interface. B, interactions between crotonyl-pantetheine and FabI. The pantetheine (cyan) is hydrogen bonded to residues in FabI helix ␣8 (green). FabI residues in the conserved active site triad (Tyr 146 , Tyr 156 , and Lys 163 ) are colored yellow. The crotonyl group of the substrate (cyan) is bound in the s-trans conformation and the crotonyl carbonyl group is oriented toward Tyr 146 (yellow). The C-3 carbon of the crotonyl group is 3 Å from the NADH pro4(S) proton (white). In addition, the NADH ribose (cyan) is hydrogen bonded to Tyr 156 and Lys 163 . The figure was made with pymol (64). target enzymes. However, despite their central role, detailed structural information on the acyl group-specific recognition of ACPs by their target proteins has remained elusive, presumably partly as a result of the conformational flexibility of the ACP molecule. Here we report the first structural data for the direct interaction of an acyl-ACP substrate with a target enzyme, the FASII FabI enzyme, based on a combination of x-ray crystallography and computational modeling. FabI is the enoyl reductase in the bacterial FASII pathway and a target for antibacterial drug discovery. Thus, not only does the structure of the FabI⅐ACP complex provide general insight into how target proteins recognize and bind to acyl-ACPs, the present structure also provides a foundation for the development of novel FabI inhibitors that antagonize the interaction of FabI with its natural substrate.
In agreement with previous predictions, several acidic residues in and close to the ACP helix ␣2 (Asp 35 , Asp 38 , and Glu 41 ) form stable electrostatic interactions with three basic amino acids (Lys 201 , Arg 204 , and Lys 205 ) located adjacent to the FabI substrate binding loop. Importantly, replacement of Lys 201 , Arg 204 , or Lys 205 by Ala or Glu results in significant decreases in k cat /K m for reduction of DD-ACP, caused both by a decrease in k cat and an increase in K m , without affecting the kinetic parameters for reduction of the corresponding CoA substrate. The effect on DD-ACP reduction was larger for the Glu substitutions, with Arg 204 demonstrating the most sensitivity to mutation. Thus, these residues are involved in specific interactions with the protein portion of DD-ACP, as was shown in similar experiments with FabH and FabG (16,17). The FabI-ACP interactions position Ser 36 , the ACP residue that carries the phosphopantetheine, above an opening into the active site formed by the substrate binding loop, helix ␣8 comprised of residues 192-206 and a mobile loop comprised of residues 152-156. These two loops move apart to allow the phosphopantetheine to deliver the substrate to the active site through the minor portal (27). Whereas the C-3 carbon of the enoyl substrate is at a distance of ϳ3 Å from the pro4(S) NADH proton, the positioning of the substrate into the active site is unexpected given previous structural data on inhibitors bound to FabI and the hexadecenoyl-N-acetylcysteamine (C16-NAC) substrate (27) bound to InhA. Below we discuss the orientation of the substrate with respect to the catalytic triad in the FabI active site.
FabI is a member of the short chain alcohol dehydrogenase/ reductase family. This superfamily is characterized by a conserved triad of active site residues. In FabI the triad is comprised of Tyr 146 , Tyr 156 , and Lys 163 , whereas in InhA, the M. tuberculosis enoyl reductase, the triad is Phe 149 , Tyr 158 , and Lys 165 . Mechanistic information on the role of these residues in catalysis has been provided by site-directed mutagenesis coupled with structural data primarily arising from enzyme-inhibitor rather than enzyme-substrate complexes. Inhibitors that have been structurally characterized in complex with these enzymes include compounds such as the diazaborines (18) and isoniazid (26), which modify the NAD(H) cofactor, and those, such as triclosan, that bind noncovalently to the enzyme-cofactor complex (20,21).
Triclosan, which binds with picomolar affinity to the E. coli FabI (23,60,61), has been proposed to bind to the enzyme as a substrate analog (62). This hypothesis has gained support from the structure of C16-NAC bound in a stable ternary complex to InhA in the presence of NAD ϩ (27). Tyr 156 of FabI (Tyr 158 in InhA) forms a hydrogen bond to the phenol of triclosan and the carbonyl oxygen of the C16 substrate. Mutagenesis clearly supports the importance of Tyr 156 (Tyr 158 ) in triclosan binding, and replacement of this residue by Phe increases the K i for triclosan inhibition by 400 (160)-fold (22,23). However, the impact of mutating Tyr 156 (Tyr 158 ) with respect to its role in catalysis is much less pronounced. In FabI, the kinetic parameters are unaffected by replacement of Tyr 156 with Phe, whereas the corresponding mutation in InhA (Y158F) has a slightly stronger effect on catalysis, reducing k cat 24-fold compared with wild-type. Intriguingly, however, the Y158S InhA mutant displays wild-type activity (28). Thus, Tyr 156 does not play a significant role in substrate reduction and, in agreement with these data, the substrate carbonyl group in the present FabI-ACP structure is pointing away from Tyr 156 (Fig. 5).
The function of the second aromatic residue in the enoyl reductase triad (Tyr 146 /Phe 149 ) is also not clear, and Rozwarski et al. (27) have argued that Phe 149 in InhA is involved in directing the NADH to deliver a hydride to the correct position on the substrate and/or in modulating the interaction of the bound substrate with a channel of water molecules that leads away from the active site. In the simulation model, the FabI Tyr 146 hydroxyl is located within hydrogen bonding distance from the substrate carbonyl, suggesting that Tyr 146 rather than Tyr 156 provides electrophilic assistance during substrate reduction. To probe the role of this residue in catalysis, the Y146F mutant was characterized and shown to catalyze substrate reduction with a k cat /K m value 14 (DD-CoA) to 50 (DD-ACP)-fold lower than wild-type FabI. These data suggest that the Tyr 146  DECEMBER 22, 2006 • VOLUME 281 • NUMBER 51

Structure of ACP Bound to FabI
hydroxyl is directly involved in catalysis, consistent with the hydrogen bonding interaction revealed by the MD simulations. We note, however, that the homologue of Tyr 146 in InhA is a Phe (Phe 149 ), and so is unable to provide a hydrogen bond to the substrate. Indeed, Raman studies on deuterated NADD cofactor bound to InhA suggest that Phe 149 most likely is involved in correctly positioning the cofactor for hydride transfer. 8 Thus, one possibility is that Tyr 146 (FabI) and Phe 149 (InhA) play different roles in catalysis.
Returning to previous structural studies, Sacchettini and coworkers (27) have discussed two entry points for substrates into the active sites of enoyl reductases termed the major and minor portals. Based on the structures of inhibitors bound to FabI and InhA and that of the InhA⅐C16-NAC complex, the fatty acid substrate would have been expected to enter the FabI active site through the major portal. Instead, the present structure indicates that binding of ACP to FabI delivers the acyl-pantetheine to the active site between loops comprised of residues 192-206 and 152-156 such that the fatty acid enters the active site through, or adjacent to, the minor portal. This structure is consistent with our mutagenesis data and also the ACP-FabI binding interface proposed by Rock and co-workers (16,17). Indeed, interaction of ACP with the basic patch of residues on FabI leaves the ACP Ser 36 residue too far from the active site to deliver the substrate through the major portal, thus predicating an alternative entry point for the substrate. Interestingly, in the recently determined structure of 5-octyl-2-phenoxyphenol bound to InhA, the analogous channel in InhA is occupied by the alkyl chain of the inhibitor that binds to the enzyme with a K i value of 1 nM (PDB code 2B37) (63). Taken together, these data indicate that substrate binding occurs in the opposite orientation to that expected from studies using inhibitors or the truncated C16-NAC substrate. In the case of InhA, we do not as yet have a structure of the relevant complex with ACP and it is possible that ACP delivers the substrate to this enzyme through the major portal, as suggested by the C16-NAC structure. Alternatively, without the ACP to locate it, the C16-NAC molecule could bind to InhA in a mode that more closely approximates an inhibitor rather than a substrate. The two resulting possibilities then are that the C16-NAC could be reduced by the enzyme in this alternative binding mode or that the C16-NAC is bound nonproductively to the enzyme in the complex used for the structural studies. Whereas we are unable to differentiate between these possibilities at the present time, we note that the C16-NAC molecule is bound to the enzyme in the presence of NAD ϩ rather than NADH, as observed for the diphenyl ether inhibitors. Future studies on the interaction of ACP with InhA will shed light on these possibilities. In addition, we are also extending our efforts to probe the interaction of ACPs with other components of the FASII pathway and designing compounds to antagonize the interaction of the natural substrate with FabI.
In summary, we have used a combination of x-ray crystallography and molecular dynamics to provide the first structural insight into the detailed interaction of ACP with a target enzyme. The structural data are substantiated by mutagenesis, and reveal that interactions between ACP and FabI are largely electrostatic in nature. Of key importance are interactions between acidic residues in ACP helix ␣2 and basic residues in FabI helix ␣8. The ACP phosphopantetheine delivers the enoyl substrate to the active site between helix ␣8 and a mobile loop comprised of residues 152-156. In agreement with site-directed mutagenesis, the conserved active site Tyr (Tyr 156 ) is not directly involved in interactions with the substrate. Whereas some of the specific interactions between FabI and ACP differ somewhat between simulations performed using different protein force fields and solvent models, they provide the consistent view that ACP is able to successfully deliver the substrate into the FabI active site through the minor portal and from the position observed in the crystal structure of the complex. Knowledge of the structural determinants that control the interaction of ACP with target enzymes is of critical importance for the design of inhibitors against these enzymes and for fully understanding the multiple roles of ACP in metabolic pathways.