The Molecular Basis of Glyphosate Resistance by an Optimized Microbial Acetyltransferase*

GAT is an N-acetyltransferase from Bacillus licheniformis that was optimized by gene shuffling for acetylation of the broad spectrum herbicide, glyphosate, forming the basis of a novel mechanism of glyphosate tolerance in transgenic plants (Castle, L. A., Siehl, D. L., Gorton, R., Patten, P. A., Chen, Y. H., Bertain, S., Cho, H. J., Duck, N., Wong, J., Liu, D., and Lassner, M. W. (2004) Science 304, 1151-1154). The 1.6-Å resolution crystal structure of an optimized GAT variant in ternary complex with acetyl coenzyme A and a competitive inhibitor, 3-phosphoglyerate, defines GAT as a member of the GCN5-related family of N-acetyltransferases. Four active site residues (Arg-21, Arg-73, Arg-111, and His-138) contribute to a positively charged substrate-binding site that is conserved throughout the GAT subfamily. Structural and kinetic data suggest that His-138 functions as a catalytic base via substrate-assisted deprotonation of the glyphosate secondary amine, whereas another active site residue, Tyr-118, functions as a general acid. Although the physiological substrate is unknown, native GAT acetylates d-2-amino-3-phosphonopropionic acid with a kcat/Km of 1500 min-1 mm-1. Kinetic data show preferential binding of short analogs to native GAT and progressively better binding of longer analogs to optimized variants. Despite a 200-fold increase in kcat and a 5.4-fold decrease in Km for glyphosate, only 4 of the 21 substitutions present in R7 GAT lie in the active site. Single-site revertants constructed at these positions suggest that glyphosate binding is optimized through substitutions that increase the size of the substrate-binding site. The large improvement in kcat is likely because of the cooperative effects of additional substitutions located distal to the active site.

Glyphosate (N-phosphonomethylglycine) is a widely used herbicide that acts nonselectively through inhibition of 5-enolpyruvylshikimate-3-phosphate (EPSP) 4 synthase in the aro-matic biosynthesis pathway (2). Its efficacy against all plant species, low cost, low mammalian toxicity, and benign environmental impact favor its use in crops that have a transgenic tolerance mechanism (3). The predominant transgene currently in use encodes a bacterial EPSP synthase that is nearly insensitive to inhibition by glyphosate (2). We reported an alternative strategy for glyphosate resistance involving enzymatic conversion of glyphosate to N-acetylglyphosate (NAG) (1). Because the acetylated form of glyphosate has low affinity for the active site of EPSP synthase, it is nonherbicidal. Three variants of an enzyme that catalyzes N-acetylation of the secondary amine of glyphosate were discovered from Bacillus licheniformis. These GAT 5 enzymes are 17 kDa in size, are most active at pH 6.8, and have a K m for acetyl coenzyme A (AcCoA) of 1-2 M. The native enzymes are very poor catalysts for acetylation of glyphosate, with a k cat of 5.3 min Ϫ1 and K m,GPJ of 1.3 mM (for ST401 GAT, referred to hereafter as native GAT). 6 The B. licheniformis enzymes belong to a diverse family of largely uncharacterized bacterial N-acetyltransferases sharing between 30 and 64% amino acid identity to native GAT (Fig. 1). Despite extensive screening of biological amines, including amino acids, nucleotides and antibiotics, the physiological substrates for the native enzymes are unknown (4).
To develop gat as a transgene for glyphosate resistance in crops, we subjected the three B. licheniformis enzymes to 11 rounds of gene shuffling and obtained optimized variants with up to a 4,500-fold increase in catalytic efficiency (k cat /K m ) relative to the native enzyme. This was achieved through a combination of improvements in k cat (190-fold) and K m,GPJ (24-fold). When introduced into plants, optimized gat genes confer robust tolerance to glyphosate (1).
To gain insight into the molecular mechanism of glyphosate N-acetylation by GAT, we determined the structure of a 7th round GAT variant (termed R7 GAT) in ternary complex with AcCoA and the competitive inhibitor 3-phosphoglycerate (3PG) and in binary complex with oxidized CoA. We also car-* 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 atomic coordinates and structure factors ( ried out a detailed kinetic analysis of the activity of native, optimized, and site-directed mutants of GAT against a panel of substrate analogs. Together, these data allow us to propose a detailed mechanism for glyphosate N-acetylation by GAT. Furthermore, comparison of the native and optimized enzymes provides insight into the structural basis of the activity enhancements achieved through gene shuffling.
Mutagenesis, Protein Expression, and Purification-Native and variant GAT enzymes were prepared as described (1). Briefly, gat genes were expressed in Escherichia coli from a pQE80 vector (Qiagen), modified to remove the His 6 tag. GAT was purified from cell lysates by affinity chromatography on coenzyme A-agarose (Sigma) and gel filtration chromatography on Superose 75 (Amersham Biosciences).
Site-directed mutagenesis of the R7 gat gene in pQE80 was performed using a QuikChange site-directed mutagenesis kit (Stratagene). Oligonucleotide pairs containing the substituted nucleotides were purchased from Operon Biotechnologies, Inc. All resulting plasmids were sequenced to verify that only the intended mutation was present in the coding region. Double mutants were obtained by successive application of the sitedirected mutagenesis protocol.
Protein concentrations were calculated from measurements of absorbance at 205 nm (E mg/ml ϭ 30.5) after buffer exchange (NAP-5 column, Amersham Biosciences) into 50 mM Na 2 SO 4 . The extinction coefficient was determined by the procedure of Scopes (5), in which extinction, cm Ϫ1 (mg/ml) Ϫ1 ϭ 27 ϩ 120(A 280 /A 205 ). Kinetic Analysis-Apparent K m values for glyphosate and the compounds in Table 3 were determined as described previously (4) in a continuous spectrophotometric assay in which cleavage of the thioester bond of AcCoA is monitored at 235 nm in a Spectramax Plus 384 plate reader (Molecular Devices, Sunnyvale, CA). Reaction mixtures contained 25 mM Hepes, pH 6.8, 10% ethylene glycol, varied concentrations of acceptor substrate and saturating AcCoA (167 M). The Spectramax software was programmed to convert change in absorbance to reaction velocity, M/min, and to return parameters of the Lineweaver-Burk transformation of the Michaelis-Menten equation. The k cat value was calculated from V max and the concentration of GAT. The apparent K m value for AcCoA was determined from initial rate measurements at saturating glyphosate, using mass spectrometric detection of NAG, as described (4), and the Lineweaver-Burk method.
The pH versus k cat profiles were fit to the equation, k cat,app ϭ k cat /(1 ϩ 10 (pK1ϪpH) ϩ 10 (pHϪpK2) ), where pK 1 is the dissociation constant for the ionizable group that promotes the reaction in its deprotonated form, and pK 2 is the dissociation constant of the ionizable group that promotes the reaction in its protonated form.
For R7 and R11 GAT, K i values for the competitive inhibitors NAG, CoA, 3PG, and phosphoenolpyruvate were determined from slope re-plots of double-reciprocal plots in which AcCoA was fixed at 167 M; the glyphosate was varied, and inhibitor was supplied at a selection of fixed concentrations. For native GAT, the K i value for CoA was also determined as above. For all other enzyme and inhibitor combinations, K i values were determined from a single concentration of inhibitor using the equation K m,app ϭ K m ϩ K m /K i [I]. Inhibitor concentrations were as follows: 2 mM NAG, 50 M 3PG, 10 M phosphoenolpyruvate, 15 M phosphonopyruvate with native GAT; 20 M phosphonopyruvate with R7 and R11 GAT; and 5 M phosphonoacetate and 2 mM phosphonobutyrate with all enzymes.
Crystallization and Data Collection-Purified R7 GAT was buffer-exchanged and concentrated to 10 mg/ml in 5 mM Hepes, pH 7.5. For crystallization the protein was incubated in the presence of 2 mM AcCoA alone or with 2 mM AcCoA and 20 mM 3PG. Crystals were obtained overnight by hanging drop vapor diffusion against a crystallization solution containing 100 mM NaOAc, pH 4.6, 150 -300 mM ammonium sulfate, and 20 -25% PEG4000. The crystals were mounted in nylon loops, after first transferring them to cryoprotectant comprising 100 mM NaOAc, pH 4.6, 25% PEG4000, and 20% glycerol supplemented with 2 mM AcCoA and 250 mM ammonium sulfate (binary complex) or 2 mM AcCoA, 50 mM 3PG, and 50 mM ammonium sulfate (ternary complex). Data were measured at 100 K at beamline 5.0.2 of the Advanced Light Source (ALS, Berkeley, CA) using an ADSC Q210 CCD detector. Although the crystals diffract to high resolution, as evidenced by the low R sym and high ͗I/I͘ values in the highest resolution shell, the detector geometry limited the data collection to 1.6 Å. All data were processed using MOSFLM (6) and programs from the CCP4 suite (7). Data collection and refinement statistics are summarized in Table 1.
Structure Determination-The structures were determined by difference Fourier analysis using the isomorphous crystal structure of selenomethionine-substituted GAT (PDB entry 2BSW) (8) as the starting model, after removing ligands and water molecules, and applying a random positional displacement to all atoms. Inspection of initial F o Ϫ F c difference maps revealed clear electron density for cofactor and 3PG in the ternary complex and cofactor and an active site sulfate ion in the binary complex. The electron density for the acetyl group of AcCoA in the ternary complex is weak, suggesting rotational disorder. The acetyl group was modeled (and subsequently refined) into the density in the chemically reasonable orientation in which its carbonyl oxygen points toward the backbone amides of Gly-74 and Met-75. There is also evidence for a minor (ϳ30% occupancy) active site configuration in which 3PG is replaced by a sulfate ion and an additional water molecule. We did not observe convincing electron density for the acetyl group of AcCoA in the binary complex structure. Based on careful inspection of F o Ϫ F c difference maps, we modeled the cofactor as oxidized CoA, although we cannot exclude the possibility that the bound species is AcCoA with a rotationally disordered acetyl group. No evidence for transfer of the acetyl group to the thiol of Cys-108 was observed, either kinetically or in the electron density maps of either structure. Refinement was carried out in REFMAC5 (9), and COOT (10) was used for model building. The use of anisotropic B-factors in the final stage of refinement was supported by a Ͼ1% drop in R free and R cryst for both data sets. All amino acids except for the N-terminal methionine, which is disordered, are included in the final models. PyMOL was used to generate all structure figures (11). Atomic coordinates and structure factors have been deposited in the PDB with accession numbers 2JDC (binary complex) and 2JDD (ternary complex).

Structural Overview of GAT
General Fold-We determined the 1.6-Å crystal structure of an optimized GAT (R7) in ternary complex with AcCoA and a competitive inhibitor, 3PG, and in binary complex with oxidized CoA (Table 1). Attempts to obtain a ternary complex with glyphosate were unsuccessful, presumably because of the low affinity of glyphosate for R7 GAT in the presence of ammonium sulfate, which was required for crystallization. GAT is a compact enzyme ( Fig. 2A) with approximate dimensions of 43 ϫ 40 ϫ 38 Å. The crystal packing suggests that GAT functions as a monomer, in agreement with gel filtration results. Despite low primary sequence homology, the overall fold identifies GAT as a member of the GCN5-related N-acetyltransferase (GNAT) superfamily (12). GAT shows the closest structural similarity to YJCF (PDB code 1Q2Y; 19% sequence identity; 1.9 Å r.m.s.d.), a distantly related protein of unknown function from Bacillus subtilis, and more modest similarity to the well characterized aminoglycoside 6Ј-N-acetyltransferase Aac(6Ј)-Ii, (PDB code 1B87; 16% sequence identity; 2.7 Å r.m.s.d.) (13) and serotonin N-acetyltransferase (PDB code 1CJW; 14% sequence identity; 3.0 Å r.m.s.d.) (14).
Cofactor-binding Site-The interactions between cofactor and GAT are similar to those observed throughout the GNAT superfamily (15,16). The adenosine group of AcCoA, which makes few direct interactions with the protein, is largely solvent-exposed, whereas the pyrophosphate and pantetheine moieties form a series of direct and water-mediated hydrogen bonds to the enzyme.
3PG-binding Site-The competitive inhibitor, 3PG, binds in a solvent-exposed cleft adjacent to the cofactor and the V-shaped wedge formed by ␤4 and ␤5. The binding site is flanked on one side by the region extending from the C-terminal portion of ␣1 into ␣2b and on the other side by the ␤6 -␤7 loop ( Fig. 2A). These flanking regions are among the least conserved structural features within the GNAT superfamily, a reflection of the fact that these enzymes bind to a diverse set of substrates. The binding site is positively charged and surrounded by an otherwise electronegative surface (Fig. 2B). A total of 8 amino acids, 1 water molecule, and the acetyl group of AcCoA interact directly (Ͻ4 Å) with the inhibitor (Fig. 2C). The hydrophobic side chains of Leu-20, Phe-31, and Val-135 all contact 3PG, as does the main chain nitrogen of Gly-74, which forms a hydrogen bond with the carboxylate of 3PG. The majority of contacts, however, are made between charged groups, including side chain interactions with the phosphate end (Arg-21, Arg-111, and His-138) and with the carboxylate end (Arg-21 and Arg-73) of 3PG. Of particular note is a short, 2.46-Å hydrogen bond between the N-⑀ of His-138 and a phosphate oxygen of 3PG.
In contrast with the conformation observed in the binary complex, which contains a sulfate ion bound in the substratebinding site, residues in the ␤6 -␤7 loop are displaced by up to 1.5 Å away from the active site in the ternary complex. This loop possesses the highest average B-factors in the ternary complex structure. In addition, side chains of residues lining the opposite face of the binding site, including Arg-21 and Arg-73, are displaced by up to 1.3 Å away from the active site to accommodate 3PG binding.
Certain features of the GAT structure appear to link the substrate-and cofactor-binding sites. Three well ordered water molecules are buried at the interface where ␣1, ␣2a, and ␣2b pack against the central ␤-sheet ( Fig. 2A). The most deeply buried of these waters makes a series of hydrogen bonds to residues from three distinct regions of the enzyme: Arg-16, located at the C-terminal end of ␣1; Ser-55, located in the center of ␤3; and the backbone carbonyl of Gly-74, located in the ␤-bulge region of ␤4. Ser-55 is conserved in the GAT-like family of N-acetyltransferases, as is Arg-16, which forms a buried salt bridge with another conserved residue Asp-34 (on ␣2b). The proximity of these residues to the cofactor-(Gly-74) and substrate-binding sites (Leu-20, Arg-21, Phe-31, Arg-73, and Gly-74) suggests that the network of hydrogen bonds formed between ␤3, ␤4, ␣2a, and ␣2b is required for optimal binding of both ligands. Indeed, substitution of Asp34 with alanine in an 8th round GAT variant decreased K m values for both AcCoA and glyphosate by 44-and 13-fold, respectively (data not shown).

Cofactor and Substrate Specificity for Native GAT and Optimized Variants
The steady-state kinetic parameters of native and optimized GAT enzymes used in this study are summarized in Table 2.
Cofactor Specificity-To characterize cofactor specificity, we compared the ability of GAT to utilize different acyl-CoAs for glyphosate N-acetylation. As reported previously, no activity was observed when 1 mM of the malonyl, crotonyl, succinyl, or 3-hydroxy-3-methylglutaryl derivatives of coenzyme A were incubated with 5 mM glyphosate and either native or R7 GAT (4). In contrast, activity was detected with propionyl-CoA and R7, and the following kinetic constants were determined: k cat ϭ 350 min Ϫ1 , K m,PrCoA ϭ 100 M, k cat /K m ϭ 3.5 min Ϫ1 M Ϫ1 . The corresponding values for AcCoA are 1040 min Ϫ1 , 1.5 M, and 690 min Ϫ1 M Ϫ1 . Thus, k cat /K m values with propionyl-CoA is only 0.5% that with AcCoA. This strict specificity is consistent with a lack of space in the active site for CoA derivatives with acyl groups appreciably larger than acetyl.
Acceptor Substrate Specificity-To define the range of sizes and functional groups that are preferred in substrates utilized by native, R7, and R11 GAT, we tested a series of compounds related to glyphosate (Table 3). Aminomethylphosphonate (AMPA), a glyphosate analog lacking a carboxyl group, was acetylated with nearly the same k cat value as glyphosate by the native, R7, and R11 enzymes, but its K m values are much higher than those of glyphosate. Relative to glyphosate, the K m value for AMPA is 3.6-fold greater with native GAT, 17-fold greater with R7, and 133-fold greater with R11. Interestingly, N-methyl-AMPA had K m values similar to those of AMPA, but k cat values were 27-to 50-fold lower, suggesting steric hindrance from the methyl group. Glycine and N-methylglycine (sarcosine) were tested to probe the effect of removing the phosphonate group from glyphosate. Even when supplied at 100 mM, essentially no activity was observed with any of the three GAT variants. The contrasting effects obtained from removal of the phosphonate group compared with removal of the carboxyl group show that the former provides more binding energy for catalysis with glyphosate.
The only compound we tested that was acetylated appreciably by native GAT was D-2-amino-3-phosphonopropionate (D-AP3), exhibiting a k cat /K m of 1500 min Ϫ1 mM Ϫ1 . Activity with D-AP3 was less with R7 and R11 (Table 3). Activity with the L-isomer was very poor with the native enzyme and almost undetectable with the optimized variants. The DL form of the next longer compound, DL-2-amino-4-phosphonobutyrate (DL-AP4), showed only slight activity at 100 mM with all enzymes, and the longest, DL-2-amino-5-phosphonopentanoate (DL-AP5), was devoid of reactivity (see below).
We also tested a series of compounds for their ability to inhibit glyphosate N-acetylation (Table 3). NAG inhibited each enzyme competitively with glyphosate. K i values decreased in parallel with K m,GPJ and were 1.3 mM (native), 0.35 mM (R7), and 0.06 mM (R11). Despite being a poor substrate, L-AP3, was a potent inhibitor of both R7 and R11, with K i values of 5 and 10 M, respectively ( Table 3). The next longer compound, DL-AP4, was a poor inhibitor of native GAT but bound 53-fold more tightly to R11 than to native GAT. The longest compound, DL-AP5, did not inhibit native GAT and only weakly inhibited R7 and R11. A similar trend was observed with the competitive inhibitors 3PG and phosphoenolpyruvate (PEP). The shorter of the two compounds, PEP, showed lower K i values than 3PG for native, R7, and R11, but the fold difference decreased throughout the optimization as follows: 12-fold tighter with the native enzyme, 5.3-fold with R7, and only 2.4-fold with R11. Similar results were obtained with the PEP analog, phosphonopyruvate. The shortest compound tested, 2-phosphonoacetate was a potent inhibitor of the native, R7, and R11 enzymes.

Kinetic Analysis of Site-directed Mutants
Glyphosate Binding-In the crystal structure of R7 GAT bound to AcCoA and 3PG, there are 8 residues located within 4 Å of the inhibitor. To examine the role of these residues in binding to glyphosate, we constructed a series of point mutants in R7 GAT and determined their kinetic parameters ( Table 4). Mutation of Arg-111, Arg-73, or Arg-21 to alanine resulted in a 170 -250-fold increase in K m,GPJ . Likewise, a significant increase in glyphosate K m was observed for H138A (44-fold) and the more conservative L20I (32-fold) mutant.
Enzyme Mechanism-To probe the mechanism of glyphosate N-acetylation by GAT, we characterized the kinetic parameters for the native enzyme and optimized variants, as well as for site-directed mutants constructed in R7 GAT. The structure reveals a conserved cysteine in the active site (Cys-108) that assumes alternate conformations within the crystal. In the conformation pointing toward the active site, the cysteine sulf- hydryl is located 3.8 Å from the carbonyl carbon of AcCoA (Fig.  2C). To clarify the role of this cysteine in catalysis, we substituted it with alanine in R7 GAT. The resulting enzyme was fully active with glyphosate (Table 4). Reaction with the sulfhydryl reagent 5,5Ј-dithiobis-2-nitrobenzoate inactivated any variant with Cys-29 present, as well as C108A, but had no effect on C29A variants. Evidently, Cys-108 is inaccessible to the bulky 5,5Ј-dithiobis-2-nitrobenzoate molecule, and inactivation was because of reaction with Cys-29. The smaller alkylating agent iodoacetamide had no effect on any GAT variant.
To identify ionizable groups that are relevant to the catalytic mechanism, we measured the pH dependence of k cat for R7 GAT. As shown in Fig. 3A, the pH profile indicates that k cat depends on the ionization of two groups. The first must be deprotonated for activity, and likely corresponds to a catalytic base with a pK value of 5.7. The second must be protonated for activity, and likely corresponds to a catalytic acid with a pK value of 7.5. Given its sequence conservation (Fig. 1) and its location 5.5 Å from the presumptive position of the nitrogen atom of glyphosate (pK a of 10.3 (3)) (Fig. 2C), His-138 is the most likely amino acid to function as a general base. Replacing His-138 with ala -TABLE 3 Substrate specificity of native, R7 and R11 GAT, performed as described under "Experimental Procedures" "None" means that activity at a compound concentration of 10 mM was below the detection limit (LOD 90% ϭ Blank AVE ϩ 3(Blank SD); typically ϳ1 M/min). "Trace" means that activity was above the LOD but insufficient for kinetic analysis. nine in R7 GAT caused a 110-fold decrease in k cat relative to the R7 GAT enzyme (Table 4). Substituting alanine for any of the conserved active site arginines, Arg-111, Arg-21 and Arg-73, reduced k cat more modestly, by 26-, 4.3-, and 1.3-fold, respectively.
We also investigated the ability of the conserved active site residue, Tyr-118, to function as a general acid. Tyr-118 makes a direct hydrogen bond (3.3 Å) to the sulfur atom of AcCoA in the crystal structure (Fig. 2C), where it is poised to protonate the thiolate anion of the CoA leaving group. Substituting Tyr-118 with phenylalanine in R7 GAT caused a significant (17-fold) decrease in k cat at pH 6.8, and eliminated the basic limb of the k cat versus pH profile (Fig. 3A). At pH 8.5, where tyrosine is largely in the phenolate form, k cat for R7-Y118F is nearly equivalent to that of R7.
Shuffling Changes-There are a total of 21 amino acid substitutions in R7 GAT resulting in a 200-fold increase in k cat and a 5.4fold decrease in K m,GPJ relative to the native enzyme (Table 2). This sequence variation is distributed across the entire structure (Fig.  2D), with only 4 of the 21 substitutions located within 5 Å of either AcCoA or 3PG (Figs. 1 and 2C). To examine the effect of individual shuffling changes on catalysis, we generated four site-directed mutants in which these active site residues (Phe-31, Ala-114, Thr-132, and Val-135) in the R7 enzyme were changed back to the native sequence. Surprisingly, the individual revertants had little impact on k cat , with 3 of the 4 revertants actually showing slight improvement (Table 4). In contrast, A114V (13-fold) and V135I (6.3-fold) showed a significant increase in K m,GPJ relative to R7 GAT. The other two revertants, T132I and F31Y, show more modest increases in K m,GPJ (Table 4). Most significant is that in all cases, the variant residue possesses a smaller side chain than the native residue.

Mechanism of Glyphosate N-Acetylation by GAT
The kinetic and structural data presented here allow us to propose a detailed mechanism for glyphosate binding and N-acetylation by GAT. The lack of an essential cysteine in the GAT active site, as evidenced by the fully functional C108A mutant, as well as the observed ternary complex structure, suggests that the reaction proceeds via direct acetyl transfer from AcCoA to the secondary amine of glyphosate (Fig. 3B), rather than by a ping-pong mechanism involving an acetyl-enzyme intermediate. This general mechanism is used by nearly all known GNAT enzymes (16).
Based on the stereochemical similarity of the compounds, we infer that glyphosate binds in the same location and orientation as 3PG, with its carboxyl group in contact with Arg-73 and Arg-21, and its phosphonyl group in contact with  His-138, Arg-21, and Arg-111. Site-directed alanine mutants implicate each of these residues in glyphosate binding. In particular, alanine substitutions at any of the active site arginines dramatically increase K m,GPJ by 170 -250-fold (Table 4). With a pK a of 10.3 (3), the glyphosate amine is largely protonated at physiological pH and must be deprotonated prior to acetyl transfer. In GAT, the same basic residues implicated in glyphosate binding also contribute to a positively charged active site (Fig. 2B). As proposed in the case of the yeast histone acetyltransferase HpaII (17), this net positive electrostatic potential is likely to lower the pK a value of the substrate amine upon substrate binding. However, individual alanine substitutions constructed at each of the basic GAT residues reveal significant and differential effects on k cat . Although alanine mutations at any of the active site arginines reduce k cat between 1.3-and 26-fold, mutation of His-138 to alanine results in a 110-fold reduction in k cat (Table 4). A decline of this magnitude is consistent with that expected for loss of a catalytic base (18 -20), and it suggests that His-138 plays a specific catalytic role during the reaction.
All GAT variants that we tested, including the three native B. licheniformis GAT enzymes (1), contain histidine at position 138 and exhibit an optimal k cat at pH ϳ 6.8, as shown for R7 in Fig. 3A. The pH-rate analysis of R7 GAT with glyphosate ( Fig.  3A) suggests the presence of one or more ionizable groups in the enzyme-substrate complex with a pK 1 of 5.7 that must be deprotonated for maximal activity, and we attribute this to His-138 (solution pK a ϳ 6.0) and to the phosphonate group of glyphosate (solution pK a ϳ 5.7).
In the ternary complex His-138 is located 5.5 Å from C-3 of 3PG, the atom corresponding to the nitrogen of glyphosate (Fig.  2C). This is too far for His-138 to function directly as a catalytic base. In the true substrate complex subtle differences in the conformation of glyphosate (relative to 3PG) or in the position of His-138 might bring the substrate amine closer to His-138. Alternatively, the specific effect of His-138 on k cat could be mediated by water molecules in the active site, as observed in other members of the GNAT family (15,16). However, none of the active site waters observed in the crystal structure are located within hydrogen bonding distance of the His-138 side chain.
We favor an alternative mechanism by which His-138 catalyzes substrate deprotonation, shown in Fig. 3B. In the crystal structure, determined at pH 4.6, His-138 forms a short 2.46-Å hydrogen bond with one of the phosphate oxygens of 3PG. Similar short hydrogen bonds are observed in the active sites of a number of enzymes, and they are proposed to play a variety of significant roles in catalysis (21)(22)(23). At physiological pH, the His-138 side chain and the glyphosate phosphonate group are likely in a deprotonated state. Upon substrate binding we propose that a proton from the secondary amino group of glyphosate is stabilized on a phosphonate oxygen atom, resulting in formation of a strong hydrogen bond between His-138 and glyphosate (Fig. 3B) and activation of the substrate amine. This substrate-assisted proton transfer mechanism is consistent with the observed pH dependence of k cat (Fig. 3A) and explains the dual role of His-138 in substrate binding and as a catalytic base.
Polarization of the AcCoA carbonyl thioester and stabilization of the tetrahedral intermediate formed after attack by the substrate amine lone pair is typically performed in GNAT enzymes by a pair of main chain amide hydrogens (15,16). In R7 GAT, the main chain amides of residues Gly-74 and Met-75, part of the structurally conserved ␤-bulge in ␤4, are positioned to fulfill this role (Fig. 2C).
In some members of the GNAT superfamily an active site residue functions as a general acid to protonate the thiolate anion that forms on CoA upon collapse of the tetrahedral intermediate (15,16). In GAT, the hydroxyl group of Tyr-118 is located within 3.3 Å of the sulfur atom of CoA, where it is perfectly positioned to perform this function. Mutating Tyr-118 to phenylalanine resulted in an 17-fold drop in k cat at pH 6.8 (Table 4) and eliminated the basic limb of the pH versus k cat plot (Fig. 3A), suggesting that Tyr-118 should be in the neutral form for optimal GAT activity. Indeed, k cat for R7 at pH 8.5, where Tyr-118 would largely be de-protonated, is nearly the same as for R7-Y118F. The rate acceleration we attribute to Tyr-118 (17-fold) is typical for acid-base catalysts (18) and is similar to that seen in other members of the GNAT superfamily (14,19).

Substrate Specificity and Optimization for Glyphosate
Substrate Specificity-The physiological role and natural substrates of the bacterial GAT-like enzymes (Fig. 1) are unknown. The most active compound we tested with native GAT is the D-isomer of AP3, a 4-atom (main chain) molecule that is acetylated with a k cat /K m value of 1500 min Ϫ1 mM Ϫ1 . Although it is significantly more active than any of the aminocontaining compounds we tested against native GAT, D-AP3 is a relatively poor substrate compared with the substrates of other N-acetyltransferases (4, 24 -26). Furthermore, although AP3 is a naturally occurring phosphonate, found in Tetrahymena and the soft coral Zoanthus (27)(28)(29), there is no defined physiological role for the D-isomer of AP3.
Analysis of the structure-activity relationship for the series of inhibitors tested against the native and optimized GAT enzymes reveals two general trends as follows: 1) native GAT binds shorter ligands (chiefly 3 and 4 atoms in the main chain) more tightly than longer ones, and 2) progressive optimization for glyphosate activity is accompanied by improved binding to longer ligands (up to ϳ5 atoms in the main chain) and retained binding to shorter ligands (Table 3). For example, the 3-and 4-atom main chain compounds (e.g. 2-phosphonoacetate, phosphonopyruvate, and PEP) are all significantly better inhibitors (K i ϳ 10 M) of native GAT than are the 5-and 6-atom main chain compounds (K i values ranging from 140 M to no measurable inhibition). After optimization for the 5-atom main chain substrate glyphosate, the 5-and 6-atom main chain compounds (e.g. NAG, DL-AP4, and DL-AP5) bind more tightly (up to 50-fold) to R11 than to native GAT, whereas the shorter compounds remain potent inhibitors of the native and optimized enzymes.
The active site of GAT dictates not only the acceptable size of a substrate but also its stereochemical configuration. This is best illustrated by AP3. Although the D-isomer is a substrate for each enzyme, the L-isomer is a potent inhibitor of R7 and R11.
Taken together with previous results (4), the structure-activity data indicate a narrow substrate range for native GAT and variants optimized for acetylation of glyphosate. Appreciable activity requires an amine-containing compound with a phosphonyl or phosphoryl and a carboxyl group with a main chain length of 5 or fewer atoms. Of the numerous compounds we tested (Table 3) (4), those that lie outside these parameters all fail to exhibit appreciable activity with native GAT or any variant.
It is interesting to note that the activity for D-AP3 drops by less than 5-fold in R11 relative to the native enzyme, whereas glyphosate activity increases by 4,500-fold. Thus, the original GAT activity against D-AP3 is robust to the large number of changes accumulated in R11 GAT during the course of the optimization for glyphosate. This is consistent with other experiments in which optimization of a weak starting activity is driven by mutations that have little effect on the original (strong) activity but a dramatic effect on the optimized activity (30).
Glyphosate Optimization-The 1100-fold increase in glyphosate N-acetylation by R7 GAT is achieved through a combination of improvements in k cat (200-fold) and in K m,GPJ (5.4-fold). There are 21 amino acids differences (out of 146) between R7 GAT and the native enzyme (Fig. 2D, yellow  spheres). Strikingly, none of these changes occur at sites comprising the catalytic machinery. Indeed, only four of the changes are located within 5 Å of the cofactor or 3PG (Fig. 2D,  blue spheres). In almost all of these cases, Y31F, V114A, I132T, and I135V, the optimized enzyme contains a smaller side chain than that present in the native enzyme, consistent with the kinetic data suggesting that GAT is optimized to accommodate a larger substrate, glyphosate.
Single-site revertants constructed at each of the four active site variable positions reveal three that significantly increase K m,GPJ (Table 4). In particular, a 6.2-fold increase in K m,GPJ is observed for the R7-V135I revertant. This substitution alone accounts for the entire 5.4-fold improvement in K m,GPJ observed in the R7 enzyme. The inability of glyphosate to inhibit the glyphosate-insensitive G96A mutant of EPSP synthase from E. coli arises from a van der Waals clash between the methyl group of Ala-96 and the phosphonate group of glyphosate (31). In the crystal structure of GAT, Val-135 is in an analogous position, just 3.5 Å from the glyphosate phosphonate group. It appears that gain-of-function substitutions in the optimized R7 GAT serve, in part, to relieve similar steric clashes in the substrate-binding pocket of native GAT.
The present results offer little insight into how the 200-fold increase in k cat was achieved in the optimized R7 enzyme. Single-site revertants constructed at each of the four active site positions reveal minimal effects on k cat . Substitutions located distal to the active site (Fig. 2D, yellow spheres) may remodel the active site through subtle rearrangements that influence the enzyme structure and flexibility (32)(33)(34)(35) to speed up the ratelimiting step of the overall reaction.
The outcome of our optimization experiments with GAT is an enzyme with activity against glyphosate that is comparable with that of other N-acetyltransferases for their natural substrates. This includes another herbicide resistance enzyme, phosphinothricin acetyltransferase, which acetylates phosphi-nothricin (PPT) with a k cat /K m,AcCoA of 6 ϫ 10 4 M Ϫ1 s Ϫ1 and a k cat /K m,PPT of 3 ϫ 10 5 M Ϫ1 s Ϫ1 (4). The k cat /K m of the R11 GAT enzyme for AcCoA and glyphosate is 2 ϫ 10 7 and ϳ5 ϫ 10 5 M Ϫ1 s Ϫ1 , respectively. In contrast, the diffusion-limited maximal value for k cat /K m is 10 9 M Ϫ1 s Ϫ1 (18). Thus, despite the dramatic improvement in activity with glyphosate, GAT is far from the limits of optimization.