The structure and mechanism of serine acetyltransferase from Escherichia coli.

Serine acetyltransferase (SAT) catalyzes the first step of cysteine synthesis in microorganisms and higher plants. Here we present the 2.2 A crystal structure of SAT from Escherichia coli, which is a dimer of trimers, in complex with cysteine. The SAT monomer consists of an amino-terminal alpha-helical domain and a carboxyl-terminal left-handed beta-helix. We identify His(158) and Asp(143) as essential residues that form a catalytic triad with the substrate for acetyl transfer. This structure shows the mechanism by which cysteine inhibits SAT activity and thus controls its own synthesis. Cysteine is found to bind at the serine substrate site and not the acetyl-CoA site that had been reported previously. On the basis of the geometry around the cysteine binding site, we are able to suggest a mechanism for the O-acetylation of serine by SAT. We also compare the structure of SAT with other left-handed beta-helical structures.

Serine acetyltransferase (SAT) 1 participates in the dual-step process of sulfur assimilation of microorganisms (1-3) and higher plants (4,5). First, SAT catalyzes the production of O-acetyl-L-serine from acetyl-CoA and L-serine, and then Oacetyl serine (thiol) lyase (OAS-TL) converts O-acetyl-L-serine into L-cysteine in the presence of sulfide. Kredich and co-workers (2,6) reported that SAT (from Salmonella typhimurium) represents the rate-limiting component and is reversibly associated with ϳ5% of the total cellular OAS-TL to form the multi-enzyme complex referred to as "cysteine synthase." Cysteine constitutes the almost exclusive metabolic entrance for reduced sulfur into cell metabolism, where it is required for biosynthesis of essential compounds including methionine, several vitamins, and metal clusters (7). The production of cysteine is therefore of biotechnological interest for pharmacological processes and as a nutritional supplement for food and feed.
SAT is known to be a member of the bacterial O-acetyltransferases subfamily of O-acyltransferases (8,9), where amino acid sequence, tertiary structures, and mechanisms are known. The folding pattern that dominates the O-acetyltransferase family is the left-handed ␤-helix, which is recognized by a hexapeptide repeating signature in which residue i is aliphatic, i ϩ 1 is usually glycine, and i ϩ 4 is a small residue, thus [LIV]-[GAED]-X 2 -[STAV]-X. Structures are triangular in crosssection and are formed by parallel ␤-strands folding into a helix with three strands per turn. The first of such proteins to be studied by x-ray crystallography was the lpxA gene product that was shown to be a trimeric protein with this left-handed ␤-helical fold (10). A folding pattern arises from a hexapeptide repeat, which occurs, albeit to varying degrees, in other members of the O-acyltransferase family. SAT has such hexapeptide repeats at its carboxyl-terminal region. The three clefts between the three subunits form the catalytic centers in which a histidine residue is essential for transfer of the acetyl or succinyl moiety from CoA to the second substrate. From the sequence alignment in Fig. 1, it can be seen that the aminoterminal region of SAT is more varied in length and sequence than the carboxyl-terminal region. The carboxyl-terminal region of SATs is highly conserved and is of special interest because it is responsible for the hetero-oligomerization with OAS-TL (11). Left-handed ␤-helices pack together in a trimeric structure with the 3-fold axis parallel to the helical axis; the main interactions between monomers are hydrophobic ones.
Preliminary crystallographic analysis (12) revealed that SAT is likely to associate in a hexameric form with 3⅐2 symmetry; this was supported by chemical cross-linking and gel filtration experiments. Because only preliminary crystallographic analysis (12) and the sequential mechanism as proposed by Leu and Cook (13) were available, Wirtz et al. (14) proposed a structural model for SAT and aimed to validate it by site-directed mutagenesis. Virtually nothing was known about the tertiary structure of the SAT hexapeptide-repeat domain or its oligomerization with OAS-TL. Wirtz et al. (14) proposed that the carboxyl-terminal domain was likely to comprise a ␤-helix and was involved not only in SAT-OAS-TL interactions but also in homotrimeric interactions of SAT. The amino terminus, proposed to be an ␣-helical domain (11), was suggested to be involved in SAT-SAT interactions, thus forming a homodimer of homotrimers. Here we present the three-dimensional structure of SAT from Escherichia coli, solved by x-ray crystallography at 2.2 Å resolution, and confirm the quaternary arrange- http://www.jbc.org/ Downloaded from ment of this enzyme. In light of this high-resolution structure, we are also able to discuss the mechanism of SAT from a structural viewpoint.
Previous kinetic studies of SAT (13) suggested a ping-pong mechanism involving an acetyl-enzyme intermediate; recently, Hindson and Shaw (15) have strongly argued that SAT of E. coli (like each of the microbial O-acetyltransferases studied thus far) is likely to have a steady-state random-order mechanism one that involves a productive ternary complex of substrates and enzyme without a covalent enzyme-substrate in- termediate. Furthermore, Hindson (16) has shown that cysteine competes with serine and not with acetyl-CoA as previously proposed (1,2) and also suggested that binding of cysteine to the serine binding site of SAT may give rise to a reduction in affinity for acetyl-CoA, thus explaining the apparent phenomenon of competitive inhibition with respect to acetyl-CoA observed by steady-state kinetics. The structure of SAT presented here is in complex with cysteine, permitting us to discuss the competitive nature of cysteine with serine and acetyl-CoA binding and appraise the manner by which cysteine negatively regulates the first step in its own synthesis.

EXPERIMENTAL PROCEDURES
Expression, Purification, and Crystallization-The E. coli CysE gene (17) was cloned between the NdeI and HindIII restriction sites in expression vector pET28a (Novagen), yielding plasmid pCOCE3. This construct engineers a polyhistidinyl tag at the amino terminus of the recombinant protein. Seleno-L-methionine substituted (SeMet) SAT was overexpressed and purified from E. coli strain B834(DE3) harboring plasmid pCOCE3 using a method adapted from Budisa et al. (18). Cells from a 40-ml overnight culture in LB were pelleted by centrifugation and resuspended in 2 ml of minimal media, and this was used to inoculate 1.2 liters of minimal media containing trace compounds (CuCl 2 ⅐ 2H 2 O, MnCl 2 ⅐ 4H 2 O, ZnCl 2 , and Na 2 MoO 4 ⅐ 2H 2 O), thiamine (10 mg liter Ϫ1 ), biotin (10 mg liter Ϫ1 ), CaCl 2 (10 mg liter Ϫ1 ), glucose (3 g liter Ϫ1 ), kanamycin (50 mg liter Ϫ1 ), and seleno-L-methionine (50 mg liter Ϫ1 ). Suppression of methionine biosynthesis was achieved by adding lysine, phenylalanine, and threonine at 80 mg liter Ϫ1 and isoleucine, leucine, and valine at 40 mg liter Ϫ1 . After 15 h of growth at 37°C (A 600 ϭ 0.7), the cultures were induced with 1 mM isopropyl-D-thiogalactopyranoside, and additional thiamine (10 mg liter Ϫ1 ), biotin (10 mg liter Ϫ1 ), and glucose (3 g liter Ϫ1 ) were also added at this stage. Cells were harvested, after an additional 12 h, by centrifugation at 4°C and stored frozen (Ϫ20°C) overnight. Cell pellets were defrosted and resuspended in 25 ml of an ice-cold solution comprising 10 mM Tris/HCl buffer, pH 7.5, 2 mM dithiothreitol, and 15 mM phenylmethylsulfonyl fluoride; lysed with lysozyme (20 g ml Ϫ1 ) and DNase I (10 g ml Ϫ1 ); and sonicated. Cellular debris was removed by centrifugation, solid ammonium sulfate was stirred slowly into the ice-cold supernatant to give ϳ45% saturation, and the solution was left on ice to equilibrate for 20 min. Precipitated protein was pelleted by centrifugation, resuspended in 25 ml of wash buffer (10 mM imidazole, 0.5 M NaCl, and 10 mM Tris/HCl, pH 7.5), and dialyzed overnight against wash buffer. The sample was loaded onto a HiTrap Chelating column (Sigma) precharged with Ni 2ϩ and equilibrated with wash buffer. SAT was eluted from the column with an imidazole gradient as per the manufacturer's guidelines. Leading fractions were assessed for purity by SDS-PAGE, fractions containing SAT were pooled, and buffer was changed to 10 mM Tris/HCl, 50 mM NaCl, and 10 mM EDTA, pH7.5, and concentrated to ϳ10 mg ml Ϫ1 in centricon modules (M r 10,000 cutoff; Amicon). Protein concentration was assessed by serial dilutions on SDS-PAGE; total yield of soluble SAT was 30 mg/liter culture. SAT was stored in the above-mentioned buffer, without additives, at Ϫ80°C. Crystals grew from 0.1 M MES, pH 6.6, 10% (Ϯ)-2-methyl-2,4-pentanediol, 0.5 M sodium thiocyanate, 5.5 mM cysteine, and SeMet SAT (10 mg ml Ϫ1 ) sitting drops, 20°C, within 7 days. Crystals grew with approximate dimensions of 1.0 ϫ 0.8 ϫ 0.4 mm. Crystals were transferred to a solution containing 0.1 M MES, pH 6.6, 40% (Ϯ)-2-methyl-2,4-pentanediol, and 5.5 mM cysteine and flash-frozen.
Data Collection-A three-wavelength MAD data set was collected from a single SeMet SAT crystal on station ID29, European Synchrotron Radiation Facility (Grenoble, France). X-ray fluorescence was used to select the three optimal wavelengths around the K absorption edge of selenium. Data were collected to 2.2 Å at 100 K using a Quantum4 charge-doupled device detector, 90°of data were collected using a 0.75°o scillation per image, and MOSFLM (19) was used to predict a data collection strategy to ensure complete data sets. Diffraction intensities from the data sets were processed using DENZO and SCALEPACK (20). Subsequently, only the peak wavelength data set was required to solve the structure; data collection statistics are shown in Table I. The crystals belong to space group P3 2 21 with unit cell dimensions a ϭ b ϭ 122.0 Å c ϭ 127.5 Å.
Structure Determination and Refinement-Phases were determined using SOLVE (21) with 30 -2.2 Å data from the peak wavelength data set. The mean figure of merit from SOLVE was only 0.26, with a Z-score of 94.0 for one solution containing 27 sites, indicating one trimer per asymmetric unit. After density modification using the CCP4 package (22), the figure of merit increased to 0.75, and a readily interpretable map was calculated for the P3 2 21 enantiomer. The 27 sites and phases found by SOLVE were put into ARP_WARP (23), mode warpNtrace, option H, to automatically build the protein chain, and this resulted in 39 chains, 242 residues, and a connectivity index of 0.59. The output from this was used as input for further auto-building, mode warpNtrace, option R; this resulted in 16 chains, 716 residues, and a connectivity index of 0.95, providing a solid base from which to start manual building. Initial free-R and R values were 36.5% and 34.0%, respectively. One trimer is present per asymmetric unit. Manual rebuilding and refinement were performed using packages CNS (24) and XtalView (25). Eleven residues from each of the 273-residue monomers are not visible at the carboxyl termini and the polyhistidine tag is not visible at the amino termini in the electron density maps (maps calculated with A-weighted Fourier coefficients). The active site regions of the SAT trimer are located between monomers, and cysteine was manually built into the positive lFo-Fcl density that was present at each site. Water molecules were added gradually during further rounds of side chain adjustment coupled with positional and B-factor refinement. The refinement statistics for the final model are given in Table I

RESULTS
The Structure of SAT-The crystal structure of SAT from E. coli was solved by SAD phasing from SeMet SAT to 2.2 Å. The SAT monomer (Fig. 2a) is composed of two domains: residues 1-140 form an ␣-helical domain, and residues 141-262 form a left-handed ␤-helical domain. The ␣-helical domain is comprised of eight ␣-helices (␣1-␣8). From ␣1 (Cys 3 -Cys 23 ) there is a short turn, leading to ␣2 (Pro 25 -Thr 34 ), which runs anti-parallel to ␣1; the amino-terminal coil of ␣2 has 3 10 helical geometry. ␣3 (Leu 41 -Leu 53 ) deviates from the ␣1-␣2 hairpin at an angle of ϳ80 o , following ␣3 there are two ␤-turns (type II and type VIII), which then run into ␣4 (Ala 60 -Ala 73 ); ␣4 and ␣3 run in an anti-parallel fashion. At the carboxyl-terminal end of ␣4 there is another change in direction, at an angle of ϳ60 o , into ␣5 (Glu 76 -Arg 91 ); ␣5 is in close proximity to ␣1. There is a ␤-turn meander, again type II and type VIII, to ␣6 (Ser 99 -Leu 103 ). From ␣6 runs ␣7 (Lys 106 -Gln 123 ), which is almost perpendicular in direction, an inverse ␥-turn leads to the final helix, ␣8 (Arg 126 -Phe 140 ), which in turn leads into the ␤-helical domain. The ␤-helical domain is a typical left-handed ␤-helix comprising of fourteen ␤-strands forming five coils of the helix. The apex of the ␤-helix is at residue Pro 240 . Residues Ala 241 -Gly 262 form a meandering loop that covers one side of the ␤-helix prism, reaching almost as far as the ␣-helical domain before it meanders back toward the apex of the ␤-helix, terminating in a short ␣-helix. This carboxyl-terminal loop has a helical propensity and forms some H-bonding interactions with itself, but very few with the rest of the monomer. Apart from the final carboxyl-terminal loop there is only one break from the ␤-helix, this is a loop from residue Gly 184 to His 193 that has random coil topology. ␣5 and ␣8 of the ␣-helical domain form H-bonding interactions with the first coil of the ␤-helix.
Quaternary Structure of SAT-The asymmetric unit of the SAT crystals is a trimer with 3-fold symmetry, independent of the crystallographic 3-fold symmetry, forming a three-sided pyramid shape with approximate dimensions of 65 Å (aminoterminal base), 40 Å (carboxyl-terminal apex), and 50 Å in height (Fig. 2b). The monomer-monomer interactions within the trimer are ϳ70% hydrophobic and involve a vast network of hydrogen bonds, many via water molecules. The main interacting regions between monomers are as follows: ␣1-␣2 loop with ␣3-␣4 loop; ␤-turns between ␣5 and ␣6 with loop Thr 185 -Lys 187 from the ␤-helix; ␣6-␣7 loop with ␣8; ␤-helix corner including residues Asp 157 -Thr 160 , Gln 178 , Ala 204 , and Ala 222 with the side of the adjacent ␤-helix comprising residues Asp 143 , Val 163 , Thr 181 , Lys 207 , and Leu 227 ; residue Gly 238 with Val 239 at the top of the ␤-helices; and the carboxyl-terminal loop Val 255 -Gln 258 with the loop Thr 185 -Lys 187 from the ␤-helix. The interacting surface between two monomers in the trimer is 2720 Å 2 . Observing the crystal packing, it seems evident that the SAT trimer interacts with another SAT trimer at the amino-terminal ends (Fig. 2c). The total buried surface between the two trimers is 4450 Å 2 , consistent with an oligomer interaction rather than a crystallization artifact, and has a similar hydrophobic nature to the monomer-monomer interactions (ϳ65%). The dimer of trimers also concurs well with previous preliminary crystallographic analysis, in which SAT crystallized in space group P2 1 2 1 2 1 and exhibited 3⅐2 symmetry; in addition, chemical crosslinking and gel filtration studies are consistent with the same quaternary arrangement (12). The Cysteine Binding Site-Between each monomer in the trimeric interaction, a cysteine molecule is observed (Fig. 2b). The cysteine binding site is situated in a small cleft between adjacent subunits and is formed by ␤-strand (Asp 157 -Thr 160 ) and the ␤-turn meander from ␣5-␣6 of one subunit and the extended loop (Gly 184 -His 193 ) from the left-handed ␤-helix of the neighboring subunit. The Gly 184 -His 193 loop is rich in conserved small residues and thus provides ample space for substrate/inhibitor binding. The cysteine ligand is flanked by two histidine residues (His 158 and His 193 ), one from each subunit, which are able to form hydrogen bonds with the cysteine ligand sulfur (His 158 N⑀2-Cys S␥ 3.2 Å, His 193 N⑀2-Cys S␥ 3.4 Å). Pro 93 flanks the carboxyl group of the cysteine. Three other charged residues direct their side chains toward the cysteine: two aspartic acid residues (Asp 92 and Asp 157 ), one from each subunit, and arginine (Arg 192 ) from the adjacent subunit. Asp 92 and Asp 157 are attracted into place by the amide of the cysteine ligand. The close proximity of these negative side chains is neutralized by local main chain amides (Asp 92 with Pro 93 and Ala 94 ; Asp 157 with His 158 ); nevertheless, the closeness of these aspartic acids is reminiscent of aspartic proteases (28,29). These residues, important in cysteine coordination, are conserved throughout species (Fig. 1). All other residues around the binding pocket are small and uncharged. Many ordered water molecules are observed in the binding pocket and help to form a network of hydrogen bonds (Fig. 3a). The carboxyl terminus of the cysteine ligand forms direct hydrogen bonds with the side chain of Arg 192 (N1 and N⑀) and with the main chain Glu 166 nitrogen via a water molecule. The amino terminus of the cysteine forms direct hydrogen bonds with the carboxylate side chains of Asp 157 and Asp 92 and to the same side chains via two water molecules.
Ternary Intermediate Modeling-Based on the cysteine binding site and structure of SAT presented here, we have manually modeled a ternary intermediate for this reaction. The carboxylate and amine of the serine moiety would be able to form the same hydrogen bonding pattern as the cysteine ligand, with the side chains Arg 192 and with Asp 92 and Asp 157 , respectively. The O␥ would form a hydrogen bond with the imidazole of His 158 . This allows us to suggest a mechanism for acyl transfer similar to that described for chloramphenicol acetyltransferase (30) shown in Fig. 3c.   FIG. 3. The cysteine binding site and proposed mechanism. a, balland-stick representation of the cysteine (pink) binding pocket and key residues (one subunit, orange; adjacent subunit, blue). The main chain is solid-colored, and side chains are transparent; water molecules are represented as cyan spheres, and hydrogen bonds are represented as red dotted lines. b, stereoview of cysteine and water omit map (red) and 2Fo-Fc map of key residues, contoured at 1.2 . c, schematic representation of the proposed mechanism for O-acetylation of serine.

DISCUSSION
Previous preliminary quaternary structure analysis of SAT by chemical cross-linking, gel filtration, and analysis of crystal packing (12) had indicated that SAT exists as a dimer of trimers, which was in contrast to the tetrameric arrangement proposed originally (31). The structure presented here concludes that the quaternary arrangement of SAT is a dimer of trimers. Each monomer is comprised of an amino-terminal ␣-helical domain and a carboxyl-terminal left-handed ␤-helical domain. The trimer formation is approximately parallel with the axis of the ␤-helical domain, and the dimer of trimers interface is at the amino-terminal ␣-helical domain.
Between adjacent subunits in each trimer, a small cleft accommodates the cysteine ligand. Upon examination of the structure, a solvent-filled channel is observed extending away from the cysteine ligand binding site. Inspection of the cysteine binding site suggests a mechanism of O-acetylation of serine. When serine is bound in place of cysteine, its O␥ would form a classical "catalytic triad" with the imidazole of His 158 from one subunit of the trimer and the carboxylic acid of Asp 143 of the adjacent subunit. This provides a mechanism for His 158 to act as a base to activate the serine hydroxyl for nucleophilic attack on the carbonyl carbon of the acetyl group of acetyl-CoA. This would form an oxyanion tetrahedral intermediate. The imidazole of His 193 is in a prime position to stabilize the negative charge, especially if protonated, although nearby main chain amides would also be candidates for this role. The collapse of the tetrahedral intermediate forms CoA and O-acetylserine (Fig. 3c). The competitive binding of cysteine and serine for the catalytic site provides an explanation for the competitive inhibition (and therefore control) by cysteine described by Hindson (16) but does not explain the apparent lack of S-acetylation of cysteine by acetyl-CoA. Although sulfur is not as electronegative as oxygen, it would still be expected to make an excellent nucleophile for attack on the acetyl carbonyl carbon. The slightly longer sulfur-carbon bond pushes the sulfur atom into a position such that it can make potential hydrogen bonds with both His 158 and His 193 that are significantly longer than the corresponding oxygen hydrogen bonds (32 His 158 and the O␥ of serine would prevent the formation of a hydrogen bond to His 193 at the same time, thus the binding of cysteine is expected to be significantly tighter than that of serine, and this may increase the activation energy sufficiently to prevent catalysis. This difference in binding may account for some of the 4.4 kcal⅐mol Ϫ1 apparent difference in binding energies of serine and cysteine calculated by Hindson (16). Maier (33) showed that mutating Thr 167 to alanine significantly reduced the inhibition by cysteine; the O␥ of Thr 167 is in a position to hydrogen bond to the N␦ of the imidazole of His 193 , holding it in position for the N⑀ to interact with the cysteine S␥. Removal of this interaction could significantly reduce cysteine binding energy. We would argue, based on the structure presented here, the modeling of the tetrahedral intermediate, and on previous biochemical assays (15,16), that it is the serine binding pocket and not the acetyl-CoA binding channel that is occupied by cysteine to inhibit the O-acetylation of serine reaction performed by SAT. The tight binding of cysteine, particularly the interactions of the two histidine side chains from adjacent monomers, may involve a slight distortion of the acetyl-CoA binding site, such that the affinity for the coenzyme is reduced. This may explain why binding of cysteine appears to inhibit acetyl-CoA binding (16).
The left-handed ␤-helical fold observed in the SAT structure, which arises from the hexapeptide repeat, is common in the O-acyltransferase family. The most similar structure to SAT is found to be xenobiotic acetyltransferase (34), which superimposes with a 2.1 Å r.m.s.d. over 105 aa (C␣). Other similar structures are galactoside O-acetyltransferase (35), which superimposes with 2.7 Å r.m.s.d. over 119 aa; carbonic anhydrase (36), which superimposes with 2.9 Å r.m.s.d. over 106 aa; tetrahydrodipicilinate N-succinyltransferase (37), which superimposes with 3.0 Å r.m.s.d. over 115 aa; and, finally, UDP-Nacetylglucosamine acyltransferase (10,38), which superimposes with a 3.3 Å r.m.s.d. over 114 aa (Fig. 4). The overall sequence identities of these structures are quite low, but the important hexapeptide repeat is present. It is observed that the main conserved region in these structures is the left-handed ␤-helical domain, which does vary in length from ϳ5 to ϳ8 coils. Each of these structures oligomerizes to form similar trimeric structures with the axis of the left-handed ␤-helix running approximately parallel to the 3-fold axis. In each case, a loop extending from the left-handed ␤-helix interacts with the neighboring subunit and covers the active site. These enzymes are presumed to have similar mechanisms, with a histidine acting as a general base catalyst for acetyl transfer (34). All of these structures have varying amounts of ␣-helical content, which is at either the amino-or carboxyl-terminal end of the left-handed ␤-helix.
The structure of SAT also allows one to visualize possible regions of the structure important in the interaction with OAS-TL. The carboxyl-terminal loop is in close proximity to the active site; this is intriguing because it has been shown previously (14) that it is the very carboxyl-terminal end of SAT that is crucial for the interaction with OAS-TL. No stoichiometric details of the SAT-OAS-TL interactions are known. It can be envisaged that OAS-TL binds to SAT on the side of the trimer, on the exterior of the interface between ␤-helical domains, and is in a prime position to receive the O-acetylserine product form of SAT to complete the cysteine synthesis. Consequently, six OAS-TL molecules could bind on to the SAT dimer of trimers, resulting in a very efficient cysteine synthesis machinery requiring tight control by negative feedback inhibition. During the process of review of this article, Olsen et al. (39) published the x-ray structure of the homologous enzyme from Haemophilus influenzae in complexes with CoA and cysteine. The results are in agreement with those presented here.