The structure of (3R)-hydroxyacyl-acyl carrier protein dehydratase (FabZ) from Pseudomonas aeruginosa.

Type II fatty acid biosynthesis systems are essential for membrane formation in bacteria, making the constituent proteins of this pathway attractive targets for antibacterial drug discovery. The third step in the elongation cycle of the type II fatty acid biosynthesis is catalyzed by beta-hydroxyacyl-(acyl carrier protein) (ACP) dehydratase. There are two isoforms. FabZ, which catalyzes the dehydration of (3R)-hydroxyacyl-ACP to trans-2-acyl-ACP, is a universally expressed component of the bacterial type II system. FabA, the second isoform, as has more limited distribution in nature and, in addition to dehydration, also carries out the isomerization of trans-2- to cis-3-decenoyl-ACP as an essential step in unsaturated fatty acid biosynthesis. We report the structure of FabZ from the important human pathogen Pseudomonas aeruginosa at 2.5 A of resolution. PaFabZ is a hexamer (trimer of dimers) with the His/Glu catalytic dyad located within a deep, narrow tunnel formed at the dimer interface. Site-directed mutagenesis experiments showed that the obvious differences in the active site residues that distinguish the FabA and FabZ subfamilies of dehydratases do not account for the unique ability of FabA to catalyze isomerization. Because the catalytic machinery of the two enzymes is practically indistinguishable, the structural differences observed in the shape of the substrate binding channels of FabA and FabZ lead us to hypothesize that the different shapes of the tunnels control the conformation and positioning of the bound substrate, allowing FabA, but not FabZ, to catalyze the isomerization reaction.

Type II fatty acid biosynthesis systems are essential for membrane formation in bacteria, making the constituent proteins of this pathway attractive targets for antibacterial drug discovery. The third step in the elongation cycle of the type II fatty acid biosynthesis is catalyzed by ␤-hydroxyacyl-(acyl carrier protein) (ACP) dehydratase. There are two isoforms. FabZ, which catalyzes the dehydration of (3R)-hydroxyacyl-ACP to trans-2-acyl-ACP, is a universally expressed component of the bacterial type II system. FabA, the second isoform, as has more limited distribution in nature and, in addition to dehydration, also carries out the isomerization of trans-2-to cis-3-decenoyl-ACP as an essential step in unsaturated fatty acid biosynthesis. We report the structure of FabZ from the important human pathogen Pseudomonas aeruginosa at 2.5 Å of resolution. PaFabZ is a hexamer (trimer of dimers) with the His/Glu catalytic dyad located within a deep, narrow tunnel formed at the dimer interface. Site-directed mutagenesis experiments showed that the obvious differences in the active site residues that distinguish the FabA and FabZ subfamilies of dehydratases do not account for the unique ability of FabA to catalyze isomerization. Because the catalytic machinery of the two enzymes is practically indistinguishable, the structural differences observed in the shape of the substrate binding channels of FabA and FabZ lead us to hypothesize that the different shapes of the tunnels control the conformation and positioning of the bound substrate, allowing FabA, but not FabZ, to catalyze the isomerization reaction.
In eubacteria and their endosymbiotic descendants (the plastids of plants and apicomplexan parasites) fatty acids are produced by what is known as the type II fatty acid biosynthetic pathway (1)(2)(3). The steps in this pathway are catalyzed by a universal set of enzymes, each encoded by a separate gene, that have been most closely studied in the model organism Escherichia coli (4,5). The growing acyl chain is shuttled between the pathway enzymes attached to the 4Ј-phosphopantetheine prosthetic group of a dedicated carrier protein, ACP. 1 This system contrasts with the type I fatty acid biosynthesis system that exists in metazoans, where multifunctional polypeptide chains (6) encode all activities in chain initiation and elongation. The intermediates in the type I system are shuffled from one catalytic site to another without being released from the complex. In light of the profound differences in these two systems, enzymes of the type II pathway have emerged as attractive targets for the development of novel antimicrobial or antiparasitic agents (2,7,8). The core feature of the type II pathway is the fatty acid elongation cycle, which extends the fatty acid chain by two carbons in each round. There are four steps in the cycle, and the proteins involved in E. coli are 1) condensation of malonyl-ACP with acyl-ACP, catalyzed by the FabB-and FabF-condensing enzymes, 2) reduction of the ␤-keto moiety by the NADPH-dependent FabG reductase, 3) dehydration of the ␤-hydroxyacyl-ACP by the FabA or FabZ dehydratases and 4) reduction of the trans-2-acyl-ACP by the NAD(P)H-dependent FabI or FabK reductases. The resulting acyl-ACP is two carbons longer and either re-enters the cycle for another round of elongation or is transferred to glycerol phosphate to produce phospholipids when the chain lengths reach 16 -18 carbons (4, 5). The crystal structures of FabB (9), FabF (10,11), FabG (12)(13)(14), FabA (15), and FabI (16) are known.
Dehydratases, which catalyze the third step in the type II elongation cycle, are the focus of this study. FabA was the first dehydratase discovered. This enzyme, in addition to performing the dehydration step, has the unique property of isomerizing trans-2-to cis-3-decenoyl-ACP as an essential step in the formation of unsaturated fatty acids in Escherichia coli (17)(18)(19)(20). However, the distribution of FabA is limited to the type II systems found in Gram-negative bacteria that produce unsaturated fatty acids and is always found with its partner, FabB, a condensing enzyme that is also essential for unsaturated fatty acid formation (21). Thus, FabA catalyzes the essential isomerization reaction necessary to introduce the double bond into the 10-carbon intermediate of the growing acyl chain, and FabB is postulated to be required to elongate one or more of the critical cis unsaturated intermediates (5,22). Accordingly, mutational inactivation of either the fabA or fabB gene results in an unsaturated fatty acid auxotroph phenotype (20,21). The fabA and fabB mutants retain the ability to produce saturated fatty acids, leading to the conclusion that an additional dehydratase and elongation-condensing enzyme exists in E. coli. The second elongation-condensing enzyme is called FabF. It is the most widely distributed condensing enzyme in bacteria and is often the sole elongation-condensing enzyme in bacterial type II systems. However, in E. coli FabF plays a dispensable role in the temperature control of cis-vaccenate production (23,24). The second dehydratase FabZ was discovered as a suppressor of temperature-sensitive mutants in lipid A biosynthesis (25). This isozyme of ␤-hydroxylacyl-ACP dehydratase is ubiquitously expressed in type II systems, cannot carry out the isomerase reaction, and is the only dehydratase that exists in most bacteria. Although FabA and FabZ have many primary sequence characteristics in common, bioinformatics analysis clearly divide the two subtypes (Fig. 1). Specifically, there are distinct differences in the active site residues, an Asp in FabA and a Glu in FabZ (Fig. 1), that were generally thought likely to underlie the fact that FabA is capable of isomerization, and FabZ is not. This does not mean that FabZ plays no role in unsaturated fatty acid synthesis, since a recent characterization of the biochemical properties of E. coli FabA and FabZ suggests that FabZ functions in the metabolism of both saturated and unsaturated long chain acyl-ACP, whereas FabA most efficiently processes chain lengths shorter than 10 carbons (26). The essentiality and broad distribution of FabZ make it an potential target for drug discovery. Recently, inhibitors that target the Plasmodium falciparum FabZ were described (27).
We report the x-ray structure of the FabZ dehydratase of Pseudomonas aeruginosa, which completes the structural characterization of representative members of each isozyme family in the type II fatty acid elongation cycle as found in the model organism E. coli. Like E. coli, P. aeruginosa has a fabA-fabB system for unsaturated fatty acid synthesis and a fabZ gene that cannot perform the isomerase reaction (28). Surprisingly, the differences in the active site catalytic residues of FabA and FabZ do not explain their different catalytic properties, but rather, their different biochemistry is explained by the structural features of the active site tunnels.

EXPERIMENTAL PROCEDURES
Cloning and Purification of PaFabZ-PaFabZ was amplified from P. aeruginosa 633 (ATCC 17933D) genomic DNA using the primers 5Ј-GCGGCGGCCCATATGATGGACATCAACGAGATTC and 5Ј-GCG-CGGATCCTAGTTTGCGTTCCGCACAG, cloned into the BamHI and NdeI sites of a modified pET-15b C-terminal His 6 -tag expression vector; this vector was transformed into BL21(DE3)-Gold cells. The authenticity of the resulting clone was checked by DNA sequencing. A single transformant was cultured in 10 ml of Tryptone broth with antibiotic (50 g/ml ampicillin and 50 g/ml kanamycin) at 37°C until the A 600 reached 0.8, then transferred into 1 liter of minimal media supplemented with 50 g/ml ampicillin, 50 g/ml kanamycin, and 40 mg L-selenomethionine along with a mix of amino acids calculated to suppress methionine biosynthesis. When an A 600 of 0.6 was reached, isopropyl-1-thio-␤-D-galactopyranoside was added to a final concentration of 0.5 mM, and incubation was continued overnight at 15°C. Cells were then collected by centrifugation (3500 rpm, 4°C, 20 min), cell pellets were lysed by sonication, and cell debris was removed by centrifugation for 60 min at 14,000 rpm. The supernatant, containing the soluble protein, was applied to a Ni 2ϩ -agarose column and washed with 30 mM imidazole, 20 mM Tris-HCl, pH 7.4 and 0.5 M NaCl. The His-tagged protein was eluted with 0.5 M imidazole, 20 mM Tris-HCl, pH 7.4, and 0.5 M NaCl. Protein was quantitated by the Bradford method (29). The concentrated (40 mg/ml), and purified protein was stored at Ϫ80°C.
Crystallization and Structure Determination-PaFabZ was crystallized by vapor diffusion using 1.5 l of 15 mg/ml protein plus 1.5 l of well solution (comprising 2.4 M ammonium sulfate plus 0.1 M sodium citrate, pH 5.5) in a hanging drop configuration. Crystals, which grew up to 200 m, were flash-frozen in a liquid nitrogen stream after a brief soak in the above well solution diluted with 20% v/v glycerol. Data were collected at the Advanced Photon Source, beamline ComCAT. A single pass of data was collected at the selenium anomalous peak wavelength for the purposes of phasing. Selenomethionyl crystals of PaFabZ diffracted to a resolution of 2.5 Å and were found to belong to the orthorhombic space group P2 1 2 1 2 1 with cell dimensions a ϭ 93.031 Å, b ϭ 100.897 Å, c ϭ 177.252 Å. The data were integrated using the program Denzo and scaled using Scalepack (30).
The intensity data were converted to structure factors, and a random subset of 10% of the reflections were selected for cross-validation purposes. Anomalous scatterer searching and refinement, experimental phasing, and density modification were performed using standard procedures in CNX (31). Automated heavy atom searching found 21 selenium sites; after refinement, this yielded useful phases with an overall 0.33 figure of merit and phasing power of 1.8. Solvent flipping yielded interpretable maps with a figure of merit of 0.94, which was then used for auto-tracing in MAID (32). This yielded a partial model that was subsequently built using the program Turbo-Frodo. Data collection, processing, and phasing statistics are given in Table I.
Refinement was conducted in CNX using a maximum likelihood function that includes the experimental phases as part of the target (33). Successive rounds of model rebuilding and adding solvent molecules followed by positional and restrained individual B refinements produced the current model with R work of 0.232 (R free ϭ 0.258). Structure refinement statistics are given in Table II. The final model contains six polypeptide chains, five sulfate ions, and 227 water molecules and is missing the following residues: chain A, 82-84 and 146; chain C, 81-82; chain D, 82-83; chain E, 83. Chain C also has the His-tag residues GS visible at the C terminus. All structure figures were prepared using PyMol (pymol.sourceforge.net) except Fig. 5A, which was prepared using Spock (51).
Expression and Purification of EcFabA and EcFabZ-The pET-15b vectors of the wild type and mutant proteins were transformed to E. coli Rosetta competent cells for protein expression. The selected transformant was cultured in LB medium with antibiotic (50 g/ml carbenicillin and 34 g/ml chloramphenicol) at 37°C until the A 600 reached 0.6. Isopropyl-1-thio-␤-D-galactopyranoside was added to a final concentration of 1 mM, and incubation was continued for a further 3 h at 37°C. Cells were collected by centrifugation (6000 rpm, 4°C, 15 min), and cell pellets were lysed with a French press. Soluble proteins were applied to a Ni 2ϩ -agarose column and washed with 40 mM imidazole-containing metal chelation affinity chromatography buffer (20 mM Tris-HCl, pH7.4, 0.5 M NaCl). His-tagged proteins were eluted with 200 mM imidazole (for FabA) or 500 mM imidazole (for EcFabZ) in the same buffer. Proteins were quantitated by the Bradford method (29). The purified proteins were stored at Ϫ20°C.
Purified EcFabA or EcFabZ proteins were applied to a Superdex TM 200 HR 16/60 column (Amersham Biosciences) and eluted with 20 mM Tris-HCl, pH 7.4, 1 mM EDTA, 100 mM NaCl, 1 mM dithiothreitol. The apparent molecular weights of FabA and FabZ were estimated using globular protein standards.
Site-directed Mutagenesis-The construction of pET-15b His 6 -tagged expression vectors for the E. coli fabA and fabZ genes were described previously (26). Site-directed mutagenesis was performed using QuikChange TM site-directed mutagenesis kit (Stratagene) according to the manufacturer's instructions. All the mutants were sequenced to establish that only the desired mutation was present.
Enzymatic Assays-The ability of the individual His 6 -tag FabA and FabZ mutant proteins to carry out the dehydratase and isomerase activities was measured using a reconstituted system essentially as described previously (26,34,35). The reaction mixtures contained 100 M ACP, 1 mM ␤-mercaptoethanol, 0.1 M sodium phosphate buffer, pH The assay mixtures were incubated at 37°C for 30 min and analyzed by conformationally sensitive gel electrophoresis in 13% polyacrylamide gels containing 2.5 M urea. Electrophoresis was performed at 25°C and 32 mA/gel. The gels were dried, and the radiolabeled bands were quantitated using a phosphorimaging screen. Specific activities were calculated from the slopes of the plot of product formation versus protein concentration in the assay. [2-14 C]Malonyl-CoA (specific activity, 52.0 mCi/mmol) was purchased from Amersham Biosciences; octanoyl-CoA was purchased from Sigma. The Mycobacterium tuberculosis MtFabH, E. coli EcFabD, Ec-FabG, and EcFabA, and Streptococcus pneumoniae SpFabF proteins were purified as described previously (36 -39).
Molecular Dynamics Simulations-Molecular dynamics simulations were performed with the program AMBER (40) for the two isoforms, EcFabA and PaFabZ, complexed with (3R)-hydroxydecanoyl-NAC substrates in explicit solvent. (3R)-Hydroxydecanoyl-NAC substrate structure was modeled after the suicide inhibitor in the EcFabA 3-decynoyl-NAC complex (PDB code 1MKA) reported by Leesong et al. (15). A superposition of this EcFabA complex onto the PaFabZ structure was used to initially position the substrate within the PaFabZ active site tunnel. Then, to simulate the positioning effect that the ACP has on the substrate, a tethering restraint of 2 kcal was imposed on the substrate terminal amine nitrogen atom, preserving its hydrogen bond with the backbone carbonyl of Ala-105 in the EcFabA structure and Phe-85 for PaFabZ. A solvation cap, centered at the alkyl-hydroxy oxygen of the substrate, was generated and expanded to 40 Å. The protein was allowed to be flexible within 20 Å of any atom of the substrate.
Both the EcFabA and the PaFabZ systems were then subjected to the same computational protocol. First, the water molecules were allowed to re-orient themselves and maximize the hydrogen bond network of the solvated phase. Next, the water positions were fully minimized while keeping the enzyme and ligand fixed. Then the protein side chains, the hydrogen atoms, and the substrate were allowed to minimize while maintaining positional restraints on the enzyme backbone atoms. This was followed by a minimization of the entire system. Upon convergence (⌬r.m.s. Ͻ10 Ϫ6 Å) the water molecules were subjected to equilibration as the bath was gradually heated to 298 K over 50 ps. After an additional minimization the complete system was brought to equilibrium with a gradual heating over a 200-ps period, after which the temperature was kept constant at 298 K for 50 ps to ensure equilibration in energy, density, and pressure. At this stage production runs were undertaken for the duration of 250 ps, saving trajectory snapshots every 1 ps of the last 200 frames. Statistical estimates of the binding energy were performed on the atomic coordinates saved from the various snapshots using an internally developed molecular mechanics Poisson-Boltzmann statistical analysis protocol (41). Free energies of binding were estimated at Ϫ8.44 (.02) and Ϫ8.13 (.02) kcal/mol for the EcFabA and PaFabZ complexes, respectively. The quantities in parenthesis depict the 95% confidence interval obtained over the last 200 frames of the production runs. The enthalpic contributions to the free energy of binding were decoupled on a per-residue basis. No net repulsion energies were measured.

RESULTS AND DISCUSSION
Monomer Structure of PaFabZ-As was anticipated from the homology to FabA at the amino acid sequence level (Fig. 1), the PaFabZ monomer adopts a ␤ ϩ ␣ "hot dog"-fold ( Fig. 2A) that is very similar to that of EcFabA (15). Two other structures shar-ing this same fold have recently been noted; they are 4-hydroxybenzoyl-CoA thioesterase from Pseudomonas sp. strain Cbs-3 (PDB code 1BVQ) (42) and acyl-CoA thioesterase II from E. coli (PDB code 1C8U) (43). In PaFabZ, each monomer is composed of a six-stranded anti-parallel ␤-sheet with topology 1/2/4/5/6/3 that wraps around a long central six-turn ␣-helix, ␣3, located between ␤2 and ␤3. In addition, a short two-turn ␣-helix is located at the N terminus, and another short, slightly irregular helix is inserted between ␤2 and ␣3. Most of the loops joining the ␤-strand elements are short, whereas significantly longer loops connect the ␣-helices to their adjoining secondary structure elements; these loops localize to the inside of the hot dog "bun" and contribute key residues to the active site and also play a role in the higher order oligomerization of the FabZ dimers (see below). Similar to FabA, the FabZ monomers dimerize with the ␤3 strands, interacting with the last two turns of the central ␣-helix by packing against each other in an antiparallel fashion. The active site is formed along the dimer interface with the critical His-49 and Glu-63 active site residues contributed by different monomers. This dimeric interface buries a surface area of 1420 Å 2 per monomer, or about 18% of the total monomer surface.
Hexamer Structure-Unlike the structure of EcFabA and other hot dog-fold proteins, the PaFabZ monomer forms a higher order hexameric structure (Fig. 2, B and C). The hexamer displays a classic "trimer of dimer" arrangement, except that contacts between dimers are mediated by a single molecule only, giving the entire arrangement a ring-like contact topology (where A contacts B contacts C contacts D contacts E contacts F contacts A). The hexamer is organized with the highly curved ␤-strands to the outside of the complex, whereas all of the long loops connecting the main secondary structural elements localize to the inside of the complex, so that the structure viewed down the 3-fold axis forms a trefoil shape (Fig. 2B). An open channel, ranging from 4 to 6 Å wide, runs down the center of the hexamer; this channel seems to have no functional role.
The hexameric interface in the PaFabZ structure buries 1110 Å 2 per monomer, or about 14% of each monomer's molecular surface; the extensive nature of this interface coupled with the highly conserved nature of the participating residues ( Fig. 1), indicates that a hexameric packing is unlikely to be a crystallization artifact. Our studies on His-tagged EcFabZ also indicate a hexameric state in solution; the elution behavior from a gel filtration column corresponds to a Stokes radius appropriate for a 112-kDa globular assembly (Fig. 3). The dimeric His-tagged FabA molecule, in contrast, elutes as a 40-kDa protein (Fig. 3). Previously reported non-denaturing gel electrophoresis results for EcFabZ are also consistent with a hexamer (25). In contrast, the FabZ orthologue in the P. falciparum type II fatty acid synthase (PfFabZ) is reported to be dimeric by both dynamic light scattering and gel filtration criteria (27), with a reported crystal form that is incompatible with a hexameric structure (44). This is puzzling because many of the residues that mediate hexameric contacts in PaFabZ are conserved, even in PfFabZ (Fig. 1). Although the residues that directly interact with the acyl chain of the substrate are confined to the dimer, the hexameric interface in PaFabZ may play an important role in stabilizing the active site. Long narrow tunnels in proteins, particularly ones lined with predominantly hydrophobic residues, are intrinsically unstable, since a more compact repacking of the component residues will be energetically favorable. In PaFabZ the loops lining the active site tunnel are stabilized and fixed by the additional packing interactions afforded by the hexameric interface. As a consequence, these loops are far more tolerant of amino acid substitutions in a R-factor is defined as ⌺ hkl ʈF obs ͉ Ϫ ͉F calc ʈ/⌺ hkl ͉F obs ͉. b R free is a cross-validation residual calculated using 10% of the data, which were chosen randomly.
FabZ than in FabA, as is reflected in the far greater proportion of conserved amino acids in FabA than in FabZ when comparing sequences derived from a similarly divergent set of organisms (Fig. 1).
The Substrate Binding Tunnel-Like EcFabA, PaFabZ possesses a long, narrow, hydrophobic tunnel that spans both monomers. The substrate tunnel begins at the surface with a narrow opening that can be completely occluded by Tyr-88 (see the discussion below), snakes under the N terminus of the long, central The tunnel entrance region near the ACP binding site is obscured by Tyr-88. This residue is highly mobile, as judged by its high B-factors, and adopts two distinct conformations. In five of the monomers in the asymmetric unit Tyr-88 points toward Pro-100Ј and is held in place by van der Waals interactions with Pro-100Ј and Val-90. This apparently preferred conformation completely blocks access to the substrate binding tunnel. In a second conformation, observed only in a single monomer, the residue rotates ϳ120°around the C␣-C␤ bond to pack against the side chains of Leu-86 and Arg-144 and forms a hydrogen bond with the C-terminal carboxylate. This opens a narrow channel roughly 4.5 Å wide between Tyr-88, Phe-89, Val-90, Phe-97, Pro-100, and Ile-55Ј that permits access to the catalytic tunnel from the bulk solvent. This parallels an analogous pattern in EcFabA where Phe-171 blocks access to the pocket in the apoenzyme structure and is moved aside to allow access to the active site in the complex with the inhibitor, 3-decynoyl-NAC (15). However, in EcFabA Phe-171 is located at the C terminus of the polypeptide chain, and thus, it is a topologically distinct residue that performs a similar structural role. Possibly, this "trap door" motif plays some functional role in prohibiting the binding of small hydrophobic molecules from solution to the unliganded enzyme.
Comparison of FabZ to FabA-The overall structure of the PaFabZ dimer strongly resembles that of EcFabA (Fig. 4); 466 atoms of the EcFabA dimer can be superimposed onto the corresponding atoms of PaFabZ with a root mean square deviation of 1.40 Å using the "magic fit" function in the program Swiss-Pdb viewer (45). This structural similarity is most pronounced for residues of the ␤-sheet and the N terminus of the central ␣3 helix. The C terminus of the ␣3 helix and the loops connecting the secondary structure elements superpose more loosely. This general impression of similarity extends to the active site of the enzymes. Both proteins have an active site histidine with the N␦ atom hydrogen-bonded to a backbone carbonyl oxygen. Both proteins have a catalytic water molecule that is held in place by hydrogen bonds to a backbone amide at the N terminus of the central helix, and to an acidic residue, Asp-84, in FabA and Glu-63 in FabZ. Although the acidic residues are different, they hold the water molecule in essentially the same spatial position.
Enzymatic Properties of EcFabA and EcFabZ Mutants-Sitedirected mutagenesis was used to generate EcFabA and EcFabZ mutants to directly test the role of the key differences in the amino acid sequences of these two proteins for isomerase activity. The mutation sites were selected based on the alignment of EcFabA and EcFabZ sequences to test whether the obvious differences in the primary sequence and active site residues that distinguish these two proteins account for the distinct differences in their catalytic activity (Fig. 1) Table III, active site residues of FabA are important to both isomerization and dehydratase activities. Enzyme activity was decreased by the mutation of either His-70 or Asp-84 to Ala. Surprisingly, EcFabA[D84E] exhibited isomerization and dehydratase activities, whereas EcFabZ[E68D] retained dehydratase activity but was inactive as an isomerase. These data clearly show that the most distinct difference in the active site residues of FabA and FabZ is not a determining factor for isomerase activity. Additionally, exchanging or substituting several conserved residues that distinguish EcFabA and EcFabZ in the active site channel did not provide insight into the requirements for isomerase activity. These mutants included EcFabZ[H19P], EcFabZ [F23M], and EcFabZ [V64C], which all retained dehydratase activity and did not exhibit isomerase activity. Also, EcFabA [P29H] and EcFabA[C80V] mutants retained both dehydratase and isomerase activities. The complementation of E. coli fabA(Ts) mutant validated the in vitro biochemical results that none of the above mutants evaluated in this work exhibited altered dehydratase/isomerase properties (Table III). Thus, mutation of the key active site histidine and acidic amino acids in EcFabA and EcFabZ showed that these two residues were key players in catalysis; the residue-swapping experiments showed that the differences in the active site did not alter the ability of the proteins to perform the isomerization reaction.
Putative ACP Binding Site-Recent work indicates that the enzymes of type II fatty acid synthesis possess a common electropositive/hydrophobic surface feature adjacent to the active site entrance that serves as the docking site for ACP (14,46,47). Close to the entrance of the PaFabZ active site tunnel, the twist of a subsection of the ␤-sheet comprising the Cterminal end of ␤3 of one monomer and ␤3, ␤6, ␤5, and ␤4 from the second monomer defines a wide groove (Fig. 5). The floor of this groove is lined with hydrophobic residues (Tyr-88, Val-90, Gly-91, Ile-120, Ile-140, Ala-142), whereas the edge is lined by conserved basic residues (Arg-96Ј, Arg-98Ј, Lys-117) (Fig. 5). A hydrophobic groove flanked by a basic ridge is a conserved surface feature in all ACP binding enzymes and interacts with the acidic/hydrophobic surface of ACP helix ␣2 (46 -48). Of these residues, Arg-96Ј is most likely a key residue for ACP docking since the arginine residues that dominate the ionic interactions with ACP in other Fab enzymes are highly conserved in these protein classes (14,46,47), and Arg-96Ј is completely conserved in all FabZs (Fig. 1). Arg-98Ј and Lys-117 are the other two residues most likely to contribute to ACP binding, and positively charged residues are found at these two surface locations in the FabZs (Fig. 1). In FabZ, like FabA and other Fab enzymes, the prosthetic group of the docked ACP extends into the narrow substrate tunnel to correctly orient the acyl chain with respect to the active site residues.
Molecular Dynamics Simulations-In the absence of an experimental structure detailing the binding of substrate to PaFabZ, molecular dynamics was used as a tool to gain insights into the conformation and position in which the substrate binds, the free energy of binding, and to identify the site residues that play an active role in stabilizing the substrates. A (3R)-hydroxydecanoyl-NAC substrate analog was modeled into the two active sites, minimized, and equilibrated at 298 K, and 250-ps trajectories were computed as described under "Experimental Procedures." Statistical estimates of the substrate binding energy were calculated from these trajectories yielding estimated free energies of binding of Ϫ8.44 and Ϫ8.13 kcal/mol for the EcFabA and PaFabZ substrate analog complexes, re- A stereo view of the overlay of the FabA and FabZ dimers is shown, looking approximately down the dimeric 2-fold axis. The PaFabZ structure is drawn as cyan, the EcFabA structure is in yellow. In general, the ␤-strands overlay very well, whereas the central helix diverges toward its C terminus, and the other helices and loops show differences on the order of 1-2 Å. spectively. These values indicate similarly favorable binding energies for this substrate analog for both enzymes. Enthalpic contributions to the free energy of binding were analyzed on a per-residue basis which revealed that the binding enthalpy is dominated by the hydrogen bonds between the carboxyl of Asp-84/Glu-63 and the OH of the substrate and two hydrogen bonds to the thioester carbonyl and the amide, but significant van der Waals contributions were also derived from Tyr-87/ Gly-103 and Tyr-88/Arg-104 interacting with the substrate acyl chain.
The average structure derived from modeling the (3R)-hydroxydecanoyl-NAC model substrate in molecular dynamics simulations provides a useful yardstick for discussing the possible roles of the features of the two active sites. In these simulations, the three hydrogen bonds afforded by the 3-hydroxy group, the carbonyl of the thioester, and the terminal amide provide a strong constraint on this portion of the molecule and act to tether it during simulations. The strongly favorable overall enthalpy of binding, the lack of significant clashes between the substrate and individual residues, and the similarity to the structure of the bound decynoyl-NAC in the FabA structure (15) strongly support the argument that the atoms of the acyl chains are at the approximately correct depth in the deeper part of the tunnel. Thus, we have a fairly accurate idea of which residues of the substrate are in contact with specific residues that line the tunnel, allowing us to identify the differences between the pockets that are likely to affect enzymatic activity.
Structural Basis for FabZ Acyl Chain Length Selectivity-One of the paradoxes presented by the FabZ structure is that the binding tunnel, about 11 Å from the catalytic water site to the deepest accessible recesses, is long enough to accommodate a 10-carbon chain but does not seem adequate to accommodate longer substrates. How then does the enzyme bind 12-, 14-, and 16-carbon substrates that it has been experimentally demonstrated to dehydrate? In PaFabZ, the carbonyl oxygen of Tyr-10, the ring of Pro-12, the ring of Phe-74, and the carbonyl oxygen of Thr-85 approach one another to form a constriction close to the exterior surface. The transient nature of the constriction is emphasized by the observation that calculating a solvent-accessible surface using a probe radius of 1.3 Å (instead of the more standard 1.4 Å) results in a tunnel in two of the six monomers that connects to the external surface at two ends. Residues 73-86 (comprising the last two turns of ␣3, the con-necting loop, and the first couple of residues of ␤3) are clearly the most mobile feature of the structure. These residues have both the highest inter-chain r.m.s. distances and the highest B factors of any region of the structure, and the central part of the loop is so poorly constrained that it can only be reliably traced in one monomer. In the presence of a long acyl chain, small coordinated outward motions of the intrinsically mobile residues Thr-85 and Phe-74 should open adequate space to allow the chain to escape to the external surface. The physical coupling of the two entrances to the active site tunnel through the ␤3 residues Tyr-88 through Thr-85 makes for the intriguing conjecture that the opening of the two ends may be energetically coupled to each other and perhaps also to ACP binding.
Structural Basis for Isomerase Activity in FabA and Not FabZ-Why is FabA, but not FabZ, able to function as a trans-2to cis-3-decenoyl-ACP isomerase? Comparison of the two structures reveals that the structure of PaFabZ possesses the same key active site structural features previously observed in EcFabA (15) (Fig. 6). In EcFabA, His-70 is sandwiched between the rings of Pro-78 and Phe-71, with an unusual non-prolyl cis-peptide bond between residues 70 and 71 making this arrangement possible. This histidine residue is locked in place with a hydrogen bond between the carbonyl oxygen of Val-76 and His-70 N␦, and NMR experiments on the FabA enzyme have shown that the N⑀ is unprotonated, even at pH 5 (49). His-70, which forms a covalent bond with the suicide inhibitor 3-decynoyl-NAC in FabA, has been hypothesized by Leesong et al. (15) to abstract a proton from C2 during dehydration and to donate a proton to the same atom during isomerization. Both the close van der Waals contact between this residue and the C2 atom in the dynamics simulations and the substantial reduction in both the isomerase and dehydratase activities of the EcFabA[H70A] mutant also support this hypothesis. In PaFabZ, this histidine along with the details of its structural environment is closely mirrored by His-49, Pro-57, Phe-50, and Ile-55 (Fig. 6B). In principle, His-49 of PaFabZ could similarly support both a dehydratase and an isomerase reaction. Leesong et al. (15) also suggest that the FabA Asp-84 supports dehydration by acting as a proton donor to the hydroxyl group of (3R)-hydroxy substrates and that the water molecule bound between this residue and the amide of Cys-80 is the product water from the dehydration reaction (15). The widening of the catalytic tunnel at the point occupied by this water molecule makes this the only site available that would accommodate a water molecule while still allowing the tunnel to bind a trans-2-acyl chain in the aftermath of the dehydration reaction. The same authors also contend that Asp-84 supports isomerization by abstracting a proton from C4. Dynamics simulations on FabA suggest that the (3R)-hydroxy substrate is stably bound when placed with its hydroxyl group in the same position as the catalytic water molecule and that C4 remains close to the carboxylate group. Also, the EcFabA[D84A] mutant showed a 2 order of magnitude impairment in both reactions, lending strong experimental support to the idea of dual roles for this residue. Superposing the structures reveals that a shift in the orientation of the ␣3 helix results in the carboxylates of the acidic residues and the respective catalytic water being almost identically positioned with respect to the substrate and the catalytic histidine on the opposite side of the tunnel (Fig. 6). Also, the backbone carboxylate group of PaFabZ Asp-84/Ec-FabA Glu-63 is being held in place by a hydrogen bond to a glutamine residue (PaFabZ Gln-67/EcFabA Gln-88) that is conserved in both proteins. Therefore, the structures demonstrate that the active site environments are essentially the same.
There are no other residues in the vicinity of the C2 and C4 carbon atoms with the potential to influence catalysis. PaFabZ His-13 (Pro in FabA) is not conserved and is a little too distant from C4 to likely play a role in quenching the isomerization reaction, the EcFabA[P29H] mutant retains both catalytic activities, and the EcFabZ [H19] remains inert as an isomerase. Similarly, the nonessential nature of the Cys-80/Val-64 dichotomy is demonstrated again by the fact that the EcFabA[C80V] mutant retains both activities, whereas the EcFabZ [V64C] mutant remains without isomerase activity. This leads to the somewhat surprising conclusion that the active site machinery required for the dehydration and isomerization chemistry in FabA is also present in the FabZ active site. Why then does FabZ lack isomerase activity? Our structural analysis suggests that, unlike FabA, FabZ is incapable of binding the trans-2decenoyl substrate in a conformation that allows the substrate to adopt a cis conformation. The acyl chain only interacts with the enzyme through van der Waals interactions; therefore, the conformations that the acyl chain can adopt are solely dictated by the shape of the active site tunnel.
Detailed examination of the van der Waals surface of the pocket in contact with the C2-C5 atoms of the substrate shows that are complementary changes in sequence results in the overall shape of the tunnel to be essentially identical in the two enzymes. For example, in EcFabA the ␣3 residue Trp-87 is an important contributor to the tunnel surface. In PaFabZ the corresponding residue is the smaller Ala-64. However, a complementary substitution on ␤3, EcFabA Ala-105 to PaFabZ Phe-89, results in an aromatic ring in roughly the same place in the two structures, and this combined with shifts in the underlying secondary structural elements results in a surface with similar overall shape and properties. In contrast, the distal portion of the tunnel (i.e. the area predicted to bind C6 through C10) is built from secondary structural elements where the backbone atoms show the most significant differences between the two structures. Thus, the residues comprising the ␣1-␤1 loop, the ␣2-␣3 loop, ␣3, and ␤3 (add 1 to these PaFabZ ␤-strand numbers to get those of EcFabA) all show substantial shifts that result in two tunnels whose centers of masses are displaced roughly 2 Å relative to one another. The effect of all of these changes is clearly seen in the averaged structures for the (3R)-hydroxydecanoyl-NAC substrate derived from the dynamics simulations, where the different shapes of the pockets cause the two acyl chains to occupy quite different regions of space for the atoms C6-C10. We propose that the differences in the shapes of the active site tunnels explain why the two enzymes differ with respect to their isomerization behavior (Fig. 7). Our hypothesis is that the transition from a covalent bond between C3 and the hydroxyl in the (3R)-hydroxy substrate to a van der Waals contact between a tightly bound, polar water molecule and the trans-2 acyl chain requires that the substrate shift away from the water binding position. The trans-2 decenoyl substrate is therefore positioned more deeply in the substrate pocket than the (3R)-hydroxydecanoyl substrate, forcing the C5 atoms of the trans-2 substrate into a portion of the tunnel where the van der Waals surfaces presented by FabA and FabZ are quite differently shaped. Differing positions of the C5 atom then fix the C3-C4 dihedral at a different angle, which is significantly more cislike for FabA than for FabZ. The ability of FabA, but not FabZ, to place the trans-2 substrate in an appropriate conformation to allow isomerization is then a consequence of the shifts of the protein backbone in the distal part of the substrate binding tunnel, including elements such as the ␣3 helix and ␤3 strand. In summary, the FabA pocket surface is uniquely able to preorganize C2 through C5 of the acyl chain in a configuration appropriate to exploit the dehydration machinery to also catalyze isomerization.
Conclusions-The differences between EcFabA and PaFabZ structures that account for the differing catalytic properties of these proteins are subtle. It is clear that the catalytic machinery in both proteins is equivalent and is capable of carrying out both the dehydratase and isomerase activities. Differences in the orientation of helix ␣3 and other secondary structural elements imposed by differences in the overall structure of the enzyme lead to differently shaped active site tunnels that in turn impose a conformation and position on the substrate that in FabA, but not in FabZ, is appropriate for the isomerization reaction. This three-dimensional difference in the shape of the active site tunnel required to transform a dehydratase into an isomerase is a property that cannot be reconstituted with simple site-directed mutagenesis work. Indeed, the line dividing FabAs and FabZs into functional classes is blurred by the recent finding that a FabZ homolog in Enterococcus faecalis also functions as an isomerase (50). However, this instance appears to be a unique circumstance since most bacteria that have only an FabZ dehydratase do not make unsaturated fatty FIG. 7. Schematic diagram showing the active sites of FabA and FabZ with a bound decenoyl substrate. FabA is depicted with a cis-3 acyl chain bound and FabZ with a trans-2 acyl chain. In FabA the shape of the tunnel in the region of atoms C2 through C5 allows (or perhaps, even forces) the adoption of a cis-3 conformation; in FabZ the tunnel is shaped differently in the region of C5 and prevents the adoption of a cis-3 configuration. Consequently it is the shapes of their respective tunnels, ultimately underpinned by shifts in the underlying backbone, that determines the ability of EcFabA, but not PaFabZ, to perform the isomerization reaction. acids or in one case utilize another type of isomerase, the FabM found in S. pneumoniae (35). At the present level of understanding, biochemical and genetic characterization will be required to ascertain the function(s) of FabZs in bacterial type II fatty acid synthases since bioinformatics analysis alone is not adequate.