Crystal Structure of β-Ketoacyl-Acyl Carrier Protein Synthase III

β-Ketoacyl-acyl carrier protein synthase III (FabH), the most divergent member of the family of condensing enzymes, is a key catalyst in bacterial fatty acid biosynthesis and a promising target for novel antibiotics. We report here the crystal structures of FabH determined in the presence and absence of acetyl-CoA. These structures display a fold that is common for condensing enzymes. The observed acetylation of Cys112 proves its catalytic role and clearly defines the primer binding pocket. Modeling based on a bound CoA molecule suggests catalytic roles for His244 and Asn274. The structures provide the molecular basis for FabH substrate specificity and reaction mechanism and are important for structure-based design of novel antibiotics.

Condensing enzymes catalyze carbon-carbon bond formation by condensing an acyl primer with an elongating carbon source often attached to a holo-acyl carrier protein (ACP). 1 These enzymes act either as single polypeptides or domains of multienzyme complexes. They are responsible for the biosynthesis of a diverse set of natural products, including fatty acids (1,2) and polyketides (3), and have recently been used to catalyze the synthesis of "unnatural" compounds (4,5). Three condensing enzymes are involved in bacterial fatty acid biosynthesis: FabB, FabF, and FabH (␤-ketoacyl-ACP synthase I, II, and III, respectively). Each catalyzes a distinct biochemical reaction (1,2). FabH plays a central role in fatty acid synthesis because it catalyzes the initiating condensation reaction and is responsible for the feedback regulation of the pathway via product (palmitoyl-ACP) inhibition (6 -9). FabH catalyzes the condensation of acetyl-CoA and malonyl-ACP to yield acetoacetyl-ACP. The reaction is believed to occur in three steps: the acetyl transfer from acetyl-CoA to a catalytic cysteine, the decarboxylation of malonyl-ACP to form a carbanion, and the condensation of the acetyl group and carbanion (1,2). FabH most likely acts as a homodimer of ϳ70 kDa (10,11). In a similar manner to other components of the dissociated (type II) fatty acid synthase system, FabH exists as an isolated enzyme. In vertebrates, however, this enzymatic activity is encoded in the N terminus of a large multi-functional Type I fatty acid syn-thase, which displays no overall sequence homology to the bacterial enzyme. Emerging resistance to currently used antibacterial agents has generated an urgent need for antibiotics acting via novel mechanisms (12,13). FabH exists ubiquitously in bacteria and is essential for viability, 2 suggesting it is a potential broad-spectrum antibacterial target. Moreover, thiolactomycin has been shown to be an effective inhibitor of bacterial condensing enzymes; it also possesses antibacterial activity (1,2). The significant differences between bacterial and human fatty acid synthase systems suggest great potential for specific and selective inhibition. Most condensing enzymes contain a consensus sequence of 45 amino acids, but FabH is the most divergent member of the family and matches few of the consensus residues (14). The antibiotic cerulenin inhibits almost all condensing enzymes but not FabH (15). Thus, FabH could have a fold and evolutionary origin different from other condensing enzymes. Here we report the crystal structure of Escherichia coli FabH in the presence and absence of ligand and show that the enzyme contains a quasi-2-fold symmetry, most likely resulting from gene duplication. The structures suggest a conserved fold for all condensing enzymes, provide the structural basis for the catalytic mechanism, and will serve as molecular templates for structure-based drug design.

EXPERIMENTAL PROCEDURES
Protein Purification and Crystallization-Native E. coli FabH protein was overexpressed in E. coli DH10B cells using the pET29 (Novagen) vector and purified to homogeneity in three chromatographic steps (Q-Sepharose, MonoQ, and hydroxyapatite) at 4°C. The selenomethionine-substituted protein was expressed in E. coli BL21 cells and purified in a similar way. Apo FabH crystals were obtained with the native protein using 20% polyethylene glycol 8000 at neutral pH, whereas the acetyl-CoA complex crystals were obtained with the selenomethionine protein using saturating acetyl-CoA and 14% polyethylene glycol 6000 at neutral pH. Crystals were ϳ0.3 mm in size and were frozen before data collection.
Data Collection and Multiple-wavelength Anomalous Dispersion Phasing-Diffraction data from the apo crystal were measured to 2.0 Å resolution at the Industrial Macromolecular Crystallography Association 17-ID beam line at the Advanced Photon Source (Table I). The crystal was of the orthorhombic P2 1 2 1 2 1 space group with cell dimensions a ϭ 63.1, b ϭ 65.1, and c ϭ 166.5 Å, with one FabH dimer per asymmetric unit. The selenomethionine FabH⅐acetyl-CoA complex crystal belongs to the tetragonal space group P4 1 2 1 2, with a ϭ b ϭ 72.4 and c ϭ 102.8 Å, and contains a monomer per asymmetric unit. The data (Table I), obtained at the X12C beam line of the Brookhaven National Laboratory, were measured at three wavelengths optimal for Se multiple-wavelength anomalous dispersion phasing experiments. Although the crystal mosaicity was as high as 1.4 o (compared with that of 0.7 o for the apo crystal) and the data were only 80% complete, the data quality was improved by its high redundancy. The acetyl-CoA complex structure was determined to 1.9-Å resolution using the multiple-wavelength anomalous dispersion phasing technique with the program SOLVE (16). There are 8 methionines in E. coli FabH; all were correctly located by SOLVE and refined to reasonable occupancies (0.62-1.00) and tem-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The perature factors (23 -41 Å 2 ). The overall figure of merit was 0.6 from 30 to 1.9 Å resolutions, and the overall Z score was 148. The resulting electron density map was of very high quality (Fig. 1).
Model Building and Refinement-The whole FabH protein (317 residues), the bound acetyl and CoA, as well as 98 solvent molecules were built in the initial model. The data set collected at the inflection point wavelength (0.9785) (I Ͼ 2[I]) were used for structural refinement with XPLOR (17) because its completion is the highest among the three. The apo FabH structure (P2 1 2 1 2 1 ) was solved using the molecular replacement program AMORE (18) with the protein atoms from the acetyl-CoA complex structure as the search model. The two monomers were identified, and the solution gave an R-factor of 33% (8.0 -3.5 Å). The model was initially built using the 2-fold averaged electron density map and refined to the final model using XPLOR. Both models (Table I) have good geometry with no Ramachandran outliers. The apo structure is clearly of good quality, as indicated by the low R-factors; the acetyl-CoA complex structure is also valid for the discussions in this report. This is because the high-quality electron density map ( Fig. 1) is calculated from experimental phases; the complex structure has been used successfully to solve the apo structure; and the two structures are practically the same at ␣-carbon level (r.m.s., 0.3 Å). We have not included the Se atoms and their anomalous signal in the current stage of refinement, resulting in the slightly higher R-factors of the complex structure. The figures were created with the computer programs XTAL-VIEW (19), MOLSCRIPT (20), and GRASP (21).

RESULTS
Overall Structure: E. coli-FabH displays a five-layered core structure, ␣-␤-␣-␤-␣, where each ␣ comprises two ␣-helices and each ␤ is made of a five-stranded, mixed ␤-sheet ( Fig. 2A). Despite the lack of overall sequence homology, the FabH core structure is similar to that of FabF, the other condensing enzyme of known structure (22,23). This fold was first identified in thiolase I, which catalyzes the degradation of ketoacyl-CoA structure (24,25). The superposition of the FabH and FabF cores results in a r.m.s. difference of 1.7 Å for 160 pairs of ␣-carbon atoms. In the comparison, Na2, Ca2, Cb3, and part of Ca3 were excluded because of large differences. The structures beyond the cores fold into a domain for substrate recognition ( Fig. 2A) and intermolecular interactions, and this domain is completely unrelated to that of FabF or thiolase. The overall sequence identity is only 14% between FabH and FabF, but the 160 pairs of structurally matched residues have a higher sequence identity of 21%, indicating that FabH is evolutionally related to FabF. Because FabF fits well with the consensus sequence (14), our results suggest a common ancestor for all condensing enzymes. Their sequences and peripheral structures have diverged significantly throughout evolution to accommodate the differences in substrates. The FabH monomer appears to have been generated through gene duplication. The two halves, residues 1-170 and 171-317, have only 11% overall sequence identity. However, their structures are similar except for the insertion and loop regions (Fig. 2B). The best matched structural elements of the two FabH halves are strands b1, b2, b4, and b5. Within these strands, the 33 pairs of ␣-carbons can be superimposed with a r.m.s. difference of only 1.0 Å, and the sequence identity between them is as high as 24%. A search of sequence data bases revealed no homologous proteins half the size of FabH, except for a few hypothetical proteins deduced from DNA sequences.
Structure of the Dimer-The crystal structure of FabH revealed the presence of a tight dimer (Fig. 3), consistent with our biochemical data 3 and similar to the structures of FabF (22,23) and thiolase I (24,25). The dimer interface mainly involves Nb3, Na3, and Na2 and buries ϳ2670 Å 2 of accessible surface area. Strand Nb3 of one monomer interacts with that of the other monomer to form a continuous 10-stranded ␤-sheet in the dimer. Helix Na3, with the catalytic Cys 112 near its N terminus (Gly 114 ), uses its C-terminal half to interact with its counterpart from the other monomer. This suggests that the dimer 3 A. Konstandinidis, unpublished data. interface is important for positioning the catalytic apparatus of FabH. The two active sites, facing opposite sides in the dimer, are formed mostly by residues within each monomer. However, Phe87 (in a loop preceding helix Na2) of one monomer projects into the active site of the other monomer. These structural features suggest that the observed dimer is the biologically relevant structure, and that its formation is essential for the proper functioning of the FabH enzyme. Catalytic Residues-Each FabH active site, located at the center of the monomer, is formed at the interface of the N-and C-terminal domains ( Fig. 2A). The catalytic residues in thiolase I were proposed to be Cys 125 , His 375 , and Cys 403 (24,25), whereas those in FabF were predicted to be Cys 163 , His 303 , and His 340 (22,23). The corresponding residues in FabH were found by structural alignment to be Cys 112 , His 244 , and Asn 274 . Although FabH Cys 112 aligns well with Cys 163 of FabF and Cys 125 of thiolase I, the structural superposition revealed significant differences among other catalytic residues (Fig. 4). The lack of a Cys 403 -like residue in FabH implies a role for that residue that is relevant only to the thiolase reaction. Asn 274 superimposes well with the histidines in FabF (His 340 ) and thiolase I (His 375 ), but His 244 is far from its counterpart (His 303 ) in FabF, suggesting different substrate specificity. A previously unsuspected thiolase residue, Asn 343 , is found near His 244 of FabH. If Asn 343 plays a role in catalysis, it is possible that the His-Asn pairs of FabH and thiolase are inverted structurally to match the reversed reactions they catalyze.
Substrate or Ligand Binding-Both FabF and thiolase I structures were determined in the absence of a substrate; thus, the structural basis for catalysis and substrate specificity has not been established. To further investigate these important questions, we co-crystallized FabH with its first substrate, acetyl-CoA. The overall structure of FabH in the acetyl-CoA complex agrees well with the apo structure, with a r.m.s. difference of 0.3 Å for all 317 ␣-carbons. The electron density reveals a covalent attachment of the acetyl moiety to Cys 112 (bond distance, 1.8 Å), indicating that the substrate was turned over by the enzyme to yield an initial product complex. This is direct evidence that Cys 112 is the catalytic nucleophile. The acetyl oxygen makes hydrogen bonds to the backbone nitrogens of Gly 306 (2.8 Å) and Cys 112 (3.0 Å). These backbone nitrogens are ideal for stabilizing the oxyanion formed during the transition state of acetyl transfer. The same role can be suggested for the backbone nitrogens of Phe 400 and Cys 163 of FabF by structural analogy. FabH Ala 113 has a slightly unfavorable torsion angle, which may be required to make Cys 112 N available for the interaction. The acetyl methyl group occupies a hydrophobic pocket in the active site, formed by the side chains of Leu 142 , Phe 157 , Leu 189 , Leu 205 , and Phe 87 Ј, which belongs to the other monomer (Fig. 5A). The van der Waals interactions between the acetyl methyl and Phe 87 Ј suggest that the dimer interface plays a critical role in determining the primer specificity of FabH.
Surprisingly, we found a CoA molecule bound in the acetyl-CoA complex, although the electron density is weak for parts of the molecule (Fig. 5B). Observing CoA in the crystal suggests that the reaction product has a notable FabH affinity. The bound CoA allows us to define the active site cavity, which is long (ϳ15 Å) and narrow, ideal for fitting the pantetheine group of substrates in a more or less extended conformation. The adenine ring of CoA stacks between the side chains of Trp 32 and Arg 151 (Fig. 5C), a classical adenine binding mode. The Arg 151 side chain also interacts with the ribose hydroxyl (3.4 Å) and phosphate (3.2 Å), whereas the N-cap dipole of helix Cb1 and Arg 36 are involved in stabilizing the diphosphate. The pantetheine group interacts with the hydrophobic wall of the active site cavity, comprised of residues such as Ile 156 , Met 207 , Val 212 , Phe 213 , and Ile 250 . The pantothenate also forms a couple of hydrogen bonds (O9 to Asn 247 ND2, 2.9 Å; N8 to Gly 209 O, 3.1 Å) with the enzyme, but the ␤-mercaptoethylamine tail of CoA is flexible, as indicated by the weak electron density and the lack of specific protein interactions. Cys 112 is at the bottom of this cavity, with the acetyl carbon 3.9 Å away from the CoA sulfur atom (Fig. 5A).

DISCUSSION
It has been shown that FabH can accept primers such as acetyl-, propionyl-, and butyryl-CoA, but not long chain acyl-CoA (7,8). This is supported by our FabH structure, in which the primer binding pocket is indeed quite small (Fig. 5A). The acetyl methyl group takes up most of the space in this pocket and is within reasonable van der Waals distances to Leu 142 , Phe 157 , and Phe 87 Ј. It is possible to model an additional carbon pointing between the side chains of Leu 189 and Phe 157 , suggesting a propionyl group can also fit. Adding yet another carbon appears to require side chain rearrangements, consistent with a much reduced efficiency for butyryl-CoA. This is in contrast to the FabF primer binding pocket that is much more extended and is the locus of binding of cerulenin (23). A helical segment of FabF (Phe 133 -Ile 138 ) is found by structural analogy to play a role similar to FabH Phe 87 Ј, suggesting the use of dimer interface as a conserved feature in condensing enzymes to modulate primer specificity. It is clear from structural inspection that cerulenin or a long acyl chain would not bind the FabH enzyme because the pocket is too small.
Although our acetyl-CoA structure turns out to be a product complex, the observed CoA provides a basis for modeling either of the FabH substrates. As indicated by the nature of the interactions, the mode of CoA binding should approximate that of the acetyl-CoA from adenine to most of the pantothenate. The observed flexible CoA tail is probably not an accurate representation of acetyl-CoA, but the substrate binding mode can be modeled with only a few free bond rotations. In such a model, this tail is probably in a more extended conformation. The hydrogen bonds between the acetyl carbonyl oxygen and the backbone nitrogens of Gly 306 and Cys 112 may guide the acetyl group into the active site, positioning it for nucleophilic attack by Cys 112 (Fig. 6A). His 244 , 3.4 Å from Cys 112 , is in position to act as a general base and extract a proton from halves have a similar fold. The core secondary structural elements are labeled as b1-b5 and a1-a3, respectively and referred in the text with a prefix of N or C to indicate the domain they belong to. These ␤-sheet and ␣-helices are drawn in magenta (N, b1-b5), cyan (N, a1-a3), red (C, b1-b5), and blue (C, a1-a3). N is the N terminus, which precedes N, b1. C is the C terminus, which comes off C, b5. Secondary elements of insertion regions are drawn in yellow and orange for N-and C-terminal domains, respectively. The catalytic Cys 112 is shown in a red ball-andstick drawing, and the CoA molecule (green) was taken from the acetyl-CoA complex structure to orient the view. B, stereo view of the superposition between the N-terminal (red) and C-terminal (blue) domains of FabH. Core ␤-strands are labeled; the overlay was generated by matching the 33 ␣-carbon pairs from four of the strands. Cys 112 . Asn 274 is too far (4.2 Å from Cys 112 ) to assume this role. The side chain of Ser 276 is 3.8 Å from Cys 112 , but its nature prohibits it from accepting a proton. The reaction likely goes through a tetrahedral intermediate, with the carbonyl group forming an oxyanion. The backbone nitrogens of Gly 306 and Cys 112 can stabilize this oxyanion. After the acetyl transfer to Cys 112 , His 244 could replenish the CoA sulfur with a proton to produce a neutral CoA (Fig. 6A). In well-known examples such as serine proteases, catalytic residues can be superimposed regardless of large differences in the overall protein folding (26). However, His 244 and its analogous residue in FabF (His 303 ) are in different positions with respect to the catalytic cysteines (Fig. 4). This is most likely attributable to differences in the substrates, suggesting that the histidine may interact directly with substrates or intermediates in addition to its role as the catalytic base. The positional differences in the catalytic apparatus suggest that condensing enzymes are under stringent evolutionary pressure to recognize and process different substrates, another reason for their sequence divergence.
The observed active site cavity appears to be the only access to Cys 112 and, therefore, the possible binding cavity for the second substrate malonyl-ACP and the feedback inhibitor palmitoyl-ACP. Malonyl-ACP is a common substrate for many condensing enzymes. The phosphopantetheine group is linked to a serine of the ACP, which is an acidic protein of 78 residues (27). A region above the FabH cavity is positively charged and seems to be suitable for ACP binding (Fig. 6B). This region mainly includes two helices (Ca1 and Ca2) and contains several positively charged residues (Arg 36 , Arg 40 , Lys 214 , His 222 , Arg 235 , Arg 249 , Lys 256 , and Lys 257 ). Binding of the prosthetic group of malonyl-ACP may be modeled based on the bound CoA. This is reasonable given that the acetylated FabH is known to react with ACP to produce acetyl-ACP (6), which is the reverse of the acetyl transfer reaction and a good indication that CoA and ACP can bind FabH similarly in the pantothenate region.
The acetyl transfer reaction is clearly a distinct step in the FabH reaction. A stepwise mechanism has also been proposed for decarboxylation and condensation (14, 28). FabH can catalyze the production of acetyl-ACP from butyryl-CoA and malonyl-ACP (8), further suggesting that decarboxylation need not occur in concert with condensation. The active sites are very similar between the apo and acetylated structures; only His 244 shifts marginally (0.5 Å). However, it is known that malonyl-ACP only binds acetylated FabH but not the apo form of the enzyme (7,8). This suggests that either the acetyl group interacts directly with the malonyl moiety, or the change of Cys 112 electrostatic potential upon acetylation is important for the binding of this substrate. The former is unlikely because iodoacetamide can also induce the binding of the malonyl group (29). The latter is probably true because His 244 is counterbalanced by Cys 112 in apo FabH (Fig. 6A)  acetylated FabH for a charged interaction with the malonyl carboxylic acid (Fig. 6C). His 244 is the only potentially positively charged residue in the active site, which provides another reason for the proposed His 244 -carboxylic acid interaction during decarboxylation. In addition, His 244 may stabilize the carbanion for the nucleophilic attack and donate a proton to Cys 112 after condensation. Modeling clearly suggests that Asn 274 forms a hydrogen bond with the malonyl carbonyl oxygen. Therefore, Asn 274 may promote decarboxylation by gen-eral acid catalysis via stabilizing an enol intermediate (Fig. 6C) and hold the carbanion for nucleophilic attack in condensation. Other condensing enzymes use a histidine in that position, and the histidine is suitable to play these same roles of Asn 274 . Based on our structures, mutagenesis and biochemical experiments can be devised to test these models and their application in other condensing enzymes. The structure will also facilitate the structure-based design of novel antibiotic drugs against this important target. FIG. 6. Proposed mechanism of the FabH reactions. A, schematic diagram of the acetylation reaction mechanism. Dotted lines indicate hydrogen bonds; thick arrows indicate attack or electron transfer directions. His 244 extracts a proton from Cys 112 , which may then attack the carbonyl of acetyl-CoA. The tetrahedral intermediate is stabilized by backbone nitrogens of Gly 306 and Cys 112 . His 244 may donate a proton at the end of acetylation to form a neutral CoA. B, FabH surface charge distribution viewed toward the active site cavity. The figure was calculated with the acetyl-CoA complex structure after removing the bound CoA. Only the monomer was used, because the dimer interface is not in the area of interest. Red indicates negative potential, and blue indicates positive potential. The CoA molecule is shown to occupy the active site, with white for carbon atoms, blue for nitrogens, red for oxygens, yellow for phosphors, and green for the sulfur. C, proposed structural basis for the decarboxylation and condensation mechanisms. According to the mechanism (14), the carboxylic acid moiety probably binds His 244 , whereas the carbonyl group interacts with Asn 274 . The decarboxylation produces a carbanion, which resonates with an enol intermediate that is stabilized by Asn 274 . The nucleophilic attack from the carbanion to the enzyme-bound acetyl group leads to a tetrahedral intermediate and finally yields the product acetoacetyl-ACP.