Crystal Structure of Carbapenam Synthetase (CarA)*

Carbapenam synthetase (CarA) is an ATP/Mg2+-dependent enzyme that catalyzes formation of the β-lactam ring in (5R)-carbapenem-3-carboxylic acid biosynthesis. CarA is homologous to β-lactam synthetase (β-LS), which is involved in clavulanic acid biosynthesis. The catalytic cycles of CarA and β-LS mediate substrate adenylation followed by β-lactamization via a tetrahedral intermediate or transition state. Another member of this family of ATP/Mg2+-dependent enzymes, asparagine synthetase (AS-B), catalyzes intermolecular, rather than intramolecular, amide bond formation in asparagine biosynthesis. The crystal structures of apo-CarA and CarA complexed with the substrate (2S,5S)-5-carboxymethylproline (CMPr), ATP analog α,β-methyleneadenosine 5′-triphosphate (AMP-CPP), and a single Mg2+ ion have been determined. CarA forms a tetramer. Each monomer resembles β-LS and AS-B in overall fold, but key differences are observed. The N-terminal domain lacks the glutaminase active site found in AS-B, and an extended loop region not observed in β-LS or AS-B is present. Comparison of the C-terminal synthetase active site to that in β-LS reveals that the ATP binding site is highly conserved. By contrast, variations in the substrate binding pocket reflect the different substrates of the two enzymes. The Mg2+ coordination is also different. Several key residues in the active site are conserved between CarA and β-LS, supporting proposed roles in β-lactam formation. These data provide further insight into the structures of this class of enzymes and suggest that CarA might be a versatile target for protein engineering experiments aimed at developing improved production methods and new carbapenem antibiotics.

The carbapenem class of ␤-lactam antibiotics exhibits a broad spectrum of activity and is resistant to inactivation by ␤-lactamases (1). The core structure of carbapenems consists of a four-membered ␤-lactam ring fused to a five-membered ring containing only carbon atoms (2). Chemical modification of naturally occurring carbapenems, which include olivanic acids from Streptomyces olivaceus and thienamycin from Streptomy-ces cattleya (2,3), has led to the development of successful drugs such as imipenem or meropenem, used to treat hospital acquired infections (1,4). Commercial fermentation of carbapenems is limited by low yields and chemical instability, however (2,4). Synthetic processes have been developed but are expensive and not as efficient as the semisynthetic procedures used to generate penicillins and cephalosporins (4,5). A detailed understanding of carbapenem biosynthesis is thus crucial to developing improved production procedures as well as new antibiotics.
Both CarC and CarA are related to enzymes in the clavulanic acid biosynthetic pathway (2). CarC is weakly homologous to clavaminate synthase, a nonheme iron ␣-ketoglutaratedependent enzyme that catalyzes hydroxylation, cyclization, and desaturation reactions in clavulanic acid biosynthesis (13,14). The crystal structure of CarC reveals a jellyroll fold characteristic of this enzyme family and details of Fe(II), ␣-ketoglutarate, and substrate binding (5). CarA is homologous to ␤-lactam synthetase (␤-LS), an ATP/Mg 2ϩ -dependent enzyme that catalyzes formation of the ␤-lactam ring in clavulanic acid ( Fig. 1) (15,16). Based on kinetic (17) and structural (18,19) data, the ␤-LS mechanism is proposed to involve substrate adenylation followed by ␤-lactamization via a tetrahedral intermediate or transition state. The chemistry performed by CarA is similar. Its ability to catalyze ␤-lactam formation in the presence of ATP and Mg 2ϩ has been demonstrated in vitro, and activity assays with alternative substrates are consistent with formation of an adenylated intermediate (12).
The sequence homology and mechanistic parallels between CarA and ␤-LS extend to class B asparagine synthetase (AS-B), which converts aspartic acid to asparagine (20). Whereas CarA and ␤-LS catalyze intramolecular amide bond formation, the AS-B reaction involves intermolecular amide bond formation between adenylated aspartic acid and ammonia. AS-B contains two active sites, an N-terminal glutaminase site where ammonia is generated by hydrolysis of glutamine and a C-terminal synthetase site where aspartic acid is adenylated and then reacts with ammonia to form asparagine. The two active sites are connected by an extended tunnel (21). Although the AS-B N-terminal domain is conserved in ␤-LS, glutaminase activity is not observed (15), consistent with the absence of crucial catalytic residues and the tunnel between domains (18). The differences between the ␤-LS and AS-B N-terminal domains together with variations in the C-terminal active sites reveal how the two enzymes accommodate ␤-lactam formation versus asparagine synthesis (18). To further investigate the evolutionary connections among this family of ATP/Mg 2ϩ -dependent amide synthesizing enzymes, we have determined the x-ray structure of CarA in the absence of exogenous ligands (apo) and in the presence of substrate, CMPr, ATP analog, ␣,␤-methyleneadenosine 5Ј-triphosphate (AMP-CPP), and a single Mg 2ϩ ion.

EXPERIMENTAL PROCEDURES
Protein Overexpression and Purification-CarA for initial crystallization trials and soaking experiments was expressed as described previously (12). To generate selenomethionine-substituted protein, the expression vector pET24a/carA (10) was transformed into Escherichia coli strain B834(DE3) (Novagen). Colonies from this transformation were used to inoculate two 5-ml tubes of LB medium containing 50 g/ml kanamycin. After 16 h at 37°C, 5 ml were used to inoculate a 100-ml culture in LB. This culture was incubated for 6 h, and then the cells were split into two 500-ml cultures of M9 minimal medium supplemented with amino acids, vitamins, 40 mg/liter selenomethionine, and 50 mg/liter kanamycin (22). The cultures were grown in 2-liter baffled Erlenmeyer flasks for 16 h and then transferred to a fermenter containing 10 liters of the same medium supplemented with 200 ml 20% (w/v) glucose. At an A 600 of 0.9, 1 mM isopropyl-␤-D-thiogalactopyranoside, 200 ml 20% (w/v) glucose, and 2 ml antifoam (Sigma) were added. During the induction period, the impellor speed was increased to 750 rpm, the airflow rate was maximized, and a pH of 7 was maintained by manual addition of ammonium hydroxide. Three hours after induction, at an A 600 of 2.0, the cells were harvested by centrifugation at 6000 ϫ g for 15 min. The cell paste was frozen in liquid nitrogen and stored at Ϫ80°C.
The purification protocols for both native and selenomethioninesubstituted CarA were based on described procedures (12). Thawed cell paste (ϳ50 g) was resuspended in 200 ml of 100 mM Tris, pH 8.0, 1.8 mM EDTA, and 1 mM DTT. While stirring on ice, 0.75 g/ml lysozyme (Research Organics) was added to the solution followed 10 min later by 0.1 mg/ml DNase I (Research Organics) and MgCl 2 to a final concentration of 1 mM. The solution was sonicated at 4°C using a Sonic Dismembrator (VWR Scientific) with a 0.5 inch probe for 5 min of 1 s pulses at power level 7. The sonicated suspension was then centrifuged for 20 min at 10,000 ϫ g at 4°C. After 35 and 65% ammonium sulfate fractionation, the pellet from the 65% cut was resuspended in 50 mM HEPES, pH 7.5, 1 mM DTT. This solution was transferred to a 10,000 molecular weight cutoff SnakeSkin dialysis membrane (Pierce) and dialyzed against 1 liter of 50 mM HEPES, pH 7.5, 1 mM DTT at 4°C for 12 h with three changes of buffer. The dialyzed sample was then loaded onto a 53 ml High Load 16/10 Q-Sepharose fast flow column (Amersham Biosciences). After washing with 4 column volumes of 50 mM HEPES, pH 7.5, 5% (v/v) glycerol, and 1 mM DTT, CarA was eluted with a 0 -1 M NaCl 10 column volume gradient. CarA-containing fractions were identified by SDS-PAGE, pooled, applied in 5-ml aliquots to a 175-ml Superdex 200 column (Amersham Biosciences), and eluted at 1 ml/min with 50 mM HEPES, pH 7.5, 500 mM NaCl, 5% (v/v) glycerol, and 1 mM DTT. Purified CarA was concentrated to 13 mg/ml using an Amicon stirred cell with a YM-10 membrane followed by a Centriprep YM-10 (Millipore) and stored at 4°C. The protein concentration was determined using the Bradford assay calibrated with a bovine serum albumin standard. The native molecular mass of purified CarA was estimated by gel filtration chromatography using a Sephacryl S300HR column calibrated with the following protein standards (Amersham Biosciences): ferritin (440 kDa), catalase (232 kDa), aldolase (158 kDa), albumin (67 kDa), ovalbumin (43 kDa), and chymotrypsinogen A (25 kDa). Each protein was eluted with 50 mM Tris, pH 7.5, 100 mM NaCl at 0.4 ml/min at 4°C.
Crystallization and Data Collection-Initial crystallization trials for both native and selenomethionine-substituted CarA were conducted using the Hampton Research Crystal Screens HT TM and Index HT TM in 96-well sitting drop trays. Crystallization drops containing 1 l of protein and 1 l of well solution were equilibrated versus 50 l of each screen solution. The optimized precipitant solution for selenomethionine-substituted protein consisted of 30% (w/v) PEG 3350, 15% (v/v) glycerol, and 100 mM sodium citrate, pH 5.3. For data collection, these crystals were transferred to a cryosolution composed of 30% (w/v) PEG 3350, 25 mM HEPES, pH 7.5, 50 mM sodium citrate, pH 5.3, 15% (v/v) glycerol, and 250 mM NaCl. After a 1-5-min soak, the crystals were flash-cooled in liquid nitrogen. Crystals of native CarA were obtained by the hanging drop method using 23-26% (w/v) PEG 4000, 100 mM sodium citrate, pH 5.6, 210 -250 mM ammonium acetate, and 100 mM cesium chloride. Microseeding was required to produce sufficiently large crystals. Soaking was performed in a solution containing 26% (w/v) PEG 4000, 100 mM sodium citrate, pH 5.6, 50 mM ammonium acetate, 20% (v/v) glycerol, 100 mM cesium chloride, 5 mM AMP-CPP, 5 mM MgCl 2 , and 5 mM CMPr. CMPr was synthesized as described pre- viously (12). After 6 h, crystals were flash-cooled in liquid nitrogen. All data were collected at Ϫ160°C by using an ADSC Quantum 315 CCD detector at Stanford Synchrotron Radiation Laboratory beamline 9 -1. Data for multiwavelength anomalous dispersion phasing were collected using the inverse beam technique. Both the selenomethionine-substituted and native CarA crystals belong to the space group P2 1 with unit cell dimensions of a ϭ 61.1 Å, b ϭ 180.4 Å, c ϭ 96.6 Å, ␤ ϭ 97.5°and a ϭ 103.0 Å, b ϭ 61.9 Å, c ϭ 175.9 Å, ␤ ϭ 97.0°, respectively. Data sets were processed with MOSFLM (23) and SCALA (24) (Table I).
Structure Determination-Of 24 possible selenium positions, 20 were located using the program SOLVE, and an initial model was built with RESOLVE (25). The asymmetric unit contains four monomers labeled A, B, C, and D. The initial model from RESOLVE included 1453 of a possible 2012 residues and served as a starting point for manual model building with XtalView (26). Initially, only monomer A was built and refined with iterative cycles of simulated annealing and individual B-factor refinement using CNS (27). Once monomer A was complete, it was used to replace the partially modeled B, C, and D monomers using the program LSQMAN (28), and iterative refinement and model adjustment were continued. The final model consists of residues A2-A441, A451-A501, B2-B441, B451-B501, C2-C441, C451-C501, and D2-D441, D468-D500, and 312 water molecules. The missing residues were not visible in the electron density map. Ramachandran plots generated with PROCHECK (29) indicate that the model exhibits good geometry with 90.7% of the residues in the most favored region and all others in the additionally allowed regions.
The structure of selenomethionine-substituted CarA was used as a starting model for refinement of the structure of native CarA in the presence of AMP-CPP, Mg 2ϩ , and CMPr. All four monomers in the asymmetric unit were located with molecular replacement using MOL-REP (30). After rigid body refinement with CNS (27), F o Ϫ F c difference maps at the active sites revealed positive electron density that was modeled as AMP-CPP, Mg 2ϩ , and CMPr. The structure was refined by multiple rounds of simulated annealing and individual B-factor refinement with CNS (27) and model rebuilding with XtalView (26). The final model includes residues 2-501 for each of the four monomers and 399 water molecules. All four active sites contain one molecule each of AMP-CPP, Mg 2ϩ , and CMPr. The model exhibits good geometry according to Ramachandran plots generated using PROCHECK (29) with 88.4% of the residues in the most favored regions and all others in the additionally allowed regions. Buried surface area calculations were performed with CNS (27), and figures were generated with MOL-SCRIPT (31), PyMOL (32), and Raster3D (33).

RESULTS AND DISCUSSION
Overall Fold-CarA comprises two domains (Fig. 2a). The N-terminal domain includes residues 2-207, and the C-terminal domain consists of residues 208 -501. The N-terminal domain is composed of two antiparallel six-stranded ␤-sheets that form a sandwich, flanked on each side by two ␣-helices. This fold is quite similar to that of the ␤-LS (18) and AS-B (21) N-terminal domains (Fig. 3). In ␤-LS, one of the sheets is five-stranded and is covered by disordered loop regions. These regions correspond to an additional strand and two helices in CarA. In AS-B, the N-terminal domain houses the glutaminase active site, which includes the catalytically important N-terminal cysteine residue and other residues involved in glutamine binding. The cysteine is replaced with a phenylalanine in ␤-LS, and nine additional N-terminal residues fill the space corresponding to the AS-B glutaminase site. The CarA N-terminal residue, a serine, occupies approximately the same position as the AS-B cysteine, but several key structural elements that constitute the AS-B glutaminase site are missing. First, an extended loop comprising AS-B residues 49 -59 is not conserved in CarA (Fig. 3). Second, residues 224 -228 from the AS-B C-terminal domain extend toward the glutamine binding site, whereas the equivalent residues in CarA are shifted back toward the second domain. Finally, specific residues that interact with glutamine in AS-B are not conserved in CarA. As a result, there is no glutamine binding pocket in CarA. The N-terminal domain of CarA is further distinguished from those of the other two enzymes by the presence of an extended loop region encompassing residues 148 -173 (Fig. 2). These residues interact with the C-terminal domain and are completely absent in ␤-LS and AS-B (Fig. 3). Part of the equivalent region in AS-B is occupied by the C-terminal residues of the second domain.
An extended linker region connects the N-terminal and Cterminal domains (Fig. 2a). The interface between the two domains is large with ϳ3800 Å 2 buried surface area. The Cterminal domain consists of a five-stranded parallel ␤-sheet   (Fig. 2b). There are extensive contacts between the A and B subunits, also observed for the D and C subunits, and between the A and D subunits, repeated between B and C (Table II). The buried surface area at these two types of interfaces exceeds 2000 Å 2 and is within the range expected for a stable complex (34). Although native gel analysis indicated that CarA is a dimer in solution (12), the molecular mass measured by calibrated gel filtration chromatography is 235 kDa. This value is consistent with the calculated molecular mass for a homotetramer of 224 kDa and the tetramer observed in the crystal. Both ␤-LS (18) and AS-B (21) form dimers but not tetramers or higher order species.
The interactions between the A and B monomers primarily involve residues from one side of the N-terminal domain (Fig.  2b). Several key hydrogen bonds are formed between residues from the last strand in the N-terminal domain and the linker region. For example, the side chain of Asp-A191 interacts with the side chains of Lys-B181 and Tyr-B373 and vice versa. In addition, the side chain of Glu-A199 is hydrogen-bonded to the backbone nitrogen of Thr-B202. Other important interactions derive from the loop encompassing residues 148 -173, including a salt bridge between Glu-A156 and Lys-B175. The involvement of the domain linker and the head-to-tail arrangement of the two monomers are similar to the ␤-LS dimer, but the N-terminal part of the linker forms the interactions in CarA, whereas the C-terminal portion is important in ␤-LS (18). As a result, the monomers are shifted with respect to one another in the two enzymes.
The interactions between the A and D monomers include residues 63-78 from two helices in the N-terminal domain, residues from the domain linker region, and a few residues from the C-terminal domain (Fig. 2b). In particular, the carbonyl oxygen of Gly-A70 is linked to the side chain of Ser-D404, the side chains of Glu-A73 and Glu-A75 interact with Ser-D208, and the carbonyl oxygen of Arg-A91 is hydrogen-bonded to the side chain of Asn-D95. Two tyrosine residues, Tyr-A77 and Tyr-D226, are also within hydrogen bonding distance. Contacts between the A and C and between the B and D monomers are minimal and only involve Glu-49 and Arg-50 from all four monomers. A third residue, Asn-A112, points toward the interface between A and C, but away from that between B and D, accounting for the size difference. The interface between B and D is smaller in the substrate structure because Arg-D50 is directed toward the hole in tetramer center rather than toward the interacting monomer (Table II).
The Active Site-Soaking crystals of apo CarA in solutions containing CMPr, AMP-CPP, and MgCl 2 resulted in an active site with a full complement of ligands (CarA/CMPr/AMP-CPP structure) (Fig. 4). The substrate is located in a hydrophobic pocket lined by residues Leu-319, Ile-323, Leu-349, and Ile-354 (Fig. 5a). The presence of a hydrophobic site was predicted from kinetic studies with alternate substrates (12). The CMPr mol-ecule was initially positioned with its carboxymethyl group directed toward the AMP-CPP ␣-phosphate by analogy to the ␤-LS structures (18,19). Subsequent refinement and careful inspection of simulated annealing omit maps indicated that it is best modeled in the opposite orientation, however, with the carboxymethyl group, which is not as well ordered as the rest of the CMPr molecule, pointing away from the AMP-CPP. One of the carboxylate oxygen atoms is within hydrogen bonding distance of the side chain of Gln-371 in three of the four monomers (Fig. 5a). In monomers A and C, the side chain of Arg-374 is also within 4 Å. The ␤-LS counterpart to Arg-374 is Phe-377, probably because an arginine in this position would be too close to the CEA guanidinium group. One ␣-carboxylate oxygen atom is hydrogen bonded to a water molecule (Fig. 5a), present in all four monomers, which is linked to the carbonyl oxygen of Gly-344, the backbone nitrogen of Asp-348, and one of the AMP-CPP ␣-phosphate oxygen atoms. This orientation of CMPr is clearly not a productive conformation for ␤-lactam formation. Notably, the second ␣-carboxylate oxygen atom of CMPr is within 3.5 Å of the AMP-CPP ␣ phosphate oxygen atom in all four monomers, suggesting that the CMPr ␣-carboxylate may be protonated in the pH 5.6 soaking solution. Such a pH effect could account for the presence of this aborted substrate complex in the crystal. Additional stabilization of the unproductive complex might derive from the interactions of Gln-371 and Arg-374 with the carboxymethyl group.
The AMP-CPP molecule is well ordered in all four monomers. The adenosine hydroxyl groups interact with the carbonyl oxygen of Pro-244 and the amide nitrogens of Gly-344 and Tyr-345. The adenine nitrogens are hydrogen-bonded to the backbone oxygen and nitrogen atoms of Ile-270. The corresponding residues in ␤-LS, Val-247, Gly-347, Tyr-348, and Met-273 (Fig.  3), participate in a similar set of interactions. A single Mg 2ϩ ion is coordinated by terminal oxygen atoms from the ␣and ␥-phosphates, the oxygen atom bridging the ␤and ␥-phosphates, two water molecules, and the carbonyl oxygen atom of Ile-444 (Fig. 5a). The water molecules are not consistently present in all four monomers. The position of this Mg 2ϩ ion is different from that observed in the structure of ␤-LS with CEA and AMP-CPP (␤-LS/CEA/AMP-CPP) (18). In ␤-LS, the Mg 2ϩ ion is located on the opposite side of the AMP-CPP phosphorus atoms and is coordinated by side chain oxygens from two aspartic acid residues as well as a water molecule and phosphate oxygen atoms (Fig. 5b). Notably, the water molecule, which forms a hydrogen bond with the ␤-carboxylate of CEA, is also observed in CarA, interacting with the substrate despite the absence of the Mg 2ϩ ion. Although the equivalent aspartic acid residues are present in CarA, Asp-250 and Asp-348, they do not coordinate a Mg 2ϩ ion (Fig. 5b). By contrast, both possible Mg 2ϩ binding sites are occupied in structures of ␤-LS with ATP alone, with N 2 -(carboxylmethyl)-L-arginine (CMA)-adenylate plus PP i , and with AMP plus the product deoxyguanidinoproclavaminic acid plus PP i (19). In these three structures, one Mg 2ϩ ion, designated Mg2, is coordinated like in the CarA/ CMPr/AMP-CPP structure and the other, designated Mg1, is coordinated as in the ␤-LS/CEA/AMP-CPP structure.
Since all the structural features involved in Mg 2ϩ coordination are conserved between CarA and ␤-LS, it is not obvious why distinct sites are occupied in the two structures with substrate and AMP-CPP. One difference is the pH at which crystallization and subsequent ligand soaking was conducted. Whereas all the ␤-LS crystals were grown and handled at pH 8.0, the CarA crystals were maintained at pH 5.6. The lower pH could affect the protonation state of the ␥-phosphate oxygens, one of which has a pK a of ϳ6 (35). Both Mg 2ϩ sites involve coordination by a terminal ␥-phosphate oxygen atom (Fig. 5b).  If the ␥-phosphate oxygen atom near the two aspartic acid residues is protonated in CarA, Mg 2ϩ binding on the opposite side might be more favorable. It is likely that neither ␥-phosphate oxygen atom is protonated in the pH 8.0 ␤-LS/CEA/AMP-CPP structure, leaving both sides of the ␥-phosphate group available for coordination. In this situation, the site with the two aspartic acid residues might have higher affinity for the Mg 2ϩ ion. Consistent with this hypothesis, two lysine residues, Lys-423 and Lys-443, are hydrogen-bonded to the ␥-phosphate oxygen atoms in ␤-LS (18,19), whereas Lys-443 has shifted away from the AMP-CPP in CarA (Fig. 5b) leaving just one lysine, Lys-421, within hydrogen bonding distance. This altered conformation of Lys-443 could reflect a decrease in negative charge due to protonation of one of the ␥-phosphate oxygen atoms. The presence of one rather than two Mg 2ϩ ions in the ␤-LS/CEA/AMP-CPP structure was attributed to the substitution of a carbon atom for the oxygen bridging the ␣and ␤-phosphate groups (19), and the same explanation may apply to CarA. A second difference between the CarA/CMPr/AMP-CPP and ␤-LS/CEA/AMP-CPP structures is the ordering of a loop comprising residues 441-450 in CarA. The corresponding region remains disordered in the ␤-LS/CEA/AMP-CPP structure (18), although it is visible in the structures with CMA-adenylate and product (19). Since this loop provides a ligand to the Mg 2ϩ ion, the carbonyl oxygen of Ile-444, its ordering could explain why this site is preferentially occupied in the CarA structure. Conversely and perhaps more likely, Mg 2ϩ binding at this site could organize these residues. The loop is further linked to the active site by a salt bridge between Arg-441 and Glu-277, which in turn interacts with one of the solvent ligands to the Mg 2ϩ ion. The corresponding residues in ␤-LS, Arg-441 and Glu-280, only interact in one monomer of the structure with CMAadenylate. In all the other ␤-LS structures, regardless of whether the following residues are disordered, Arg-441 adopts a different conformation.
Mechanistic Implications-By analogy to ␤-LS and AS-B and in accordance with kinetic data, the CarA mechanism has been proposed to involve substrate adenylation followed by ␤-lactam formation (12). Similar to ␤-LS (17) and AS-B (36), ATP binds first, followed by CMPr, and the last product released is PP i (12). The arrangement of ligands in CarA is consistent with this scheme. The AMP-CPP is situated deep in the active site cleft covered by CMPr and residues 441-450. The interactions between AMP-CPP and CarA are very similar to those observed in the ␤-LS structures, suggesting that the ATP binding site in this family of enzymes is highly conserved. In ␤-LS, 5b) and does not change conformation upon CMPr and AMP-CPP binding. The CarA equivalent to ␤-LS Tyr-348, Tyr-345, also remains in the same position. The similarity in position between the apo and substrate structures combined with the substitution of Ile for Tyr suggest that these two residues may not function to secure the substrate in the active site, as suggested for ␤-LS (19). It is also possible, however, that these residues or others do change conformation upon substrate binding in the correct position.
Although CMPr is present in a nonproductive orientation, the structure provides insight into the details of substrate binding. Whereas the carboxylate of the carboxymethyl group interacts with two amino acid residues, Gln-371 and Arg-374, the ␣-carboxylate group interacts only with a water molecule. Therefore, in the proper orientation, the carboxymethyl group is not expected to form specific contacts with the enzyme. Similarly, the carboxyethyl chain of CEA does not interact with residues in the ␤-LS active site (18,19). The lack of specific interactions is consistent with the ability of CarA to catalyze cyclization of different diastereomers of CMPr and of various diacids (12). This apparent promiscuity taken together with the nonproductive substrate conformation observed in the crystal structure raises the question of how CarA places substrates correctly in the active site. Whereas the arginine side chain provides a means for positioning CEA in the ␤-LS active site via specific interactions, CMPr is more symmetrical, with the two sides of the proline ring differing only in one methylene group.
Once substrate and ATP are assembled in the active site, the next step is attack of ATP by the CMPr carboxymethyl group to yield an adenylated intermediate. The formation of such a species is supported by hydroxylamine-dependent AMP formation in the presence of diacid substrates lacking an amino group (12). Intramolecular amide bond formation then yields a tetrahedral intermediate or transition state, followed by collapse to (3S,5S)-carbapenam-3-carboxylic acid, AMP, and PP i . Deprotonation of the ammonium ion is required to initiate ␤-lactam formation and by analogy to ␤-LS (19) may be facilitated by a catalytic dyad consisting of residues Glu-380 and Tyr-345 (Fig. 5b). These two residues form a strong hydrogen bond in all four monomers in both structures. Another residue proposed to be critical in ␤-LS (19), Lys-443, is also conserved in CarA (Fig. 5b). Although this residue points away from the AMP-CPP in the current structure, a change in side chain conformation would enable it to stabilize the oxyanion intermediate or transition state. The sequential and structural conservation of these three residues in the two enzymes underscores their potential mechanistic importance. Finally, the adenylation and ␤-lactamization steps are assisted by Mg 2ϩ , which both activates the ␣,␤-bond in ATP and counterbalances the negative charge on the PP i and nascent AMP, as proposed previously (18,19). Although only one Mg 2ϩ ion is present in the CarA/CMPr/AMP-CPP structure, all the ligands used by ␤-LS to bind two Mg 2ϩ ions are present, suggesting that CarA can also bind two Mg 2ϩ ions.
In sum, the structure of CarA reveals the overall fold, details of cofactor and substrate binding, and key residues in the active site. Comparison of the N-terminal domain to the corresponding domains in AS-B and ␤-LS demonstrates another manner in which the AS-B glutaminase site can evolve away.
The function of this domain in CarA and ␤-LS remains unclear. Comparison of the CarA C-terminal active site to those in AS-B and ␤-LS indicates that the ATP binding site is highly conserved but that the substrate binding pocket changes to accommodate different molecules. The positions of the Glu-380/Tyr-345 catalytic dyad and of Lys-443 are consistent with previously proposed functional roles for these residues. A major outstanding question is how CarA properly positions CMPr and other substrates in the active site. Further biochemical and structural studies are required to resolve this issue, but the current data suggest that CarA might prove to be an even more versatile target than ␤-LS for protein engineering experiments.