Pseudomonas aeruginosa LD-Carboxypeptidase, a Serine Peptidase with a Ser-His-Glu Triad and a Nucleophilic Elbow*

ld-Carboxypeptidases (EC 3.4.17.13) are named for their ability to cleave amide bonds between l- and d-amino acids, which occur naturally in bacterial peptidoglycan. They are specific for the link between meso-diaminopimelic acid and d-alanine and therefore degrade GlcNAc-MurNAc tetrapeptides to the corresponding tripeptides. As only the tripeptides can be reused as peptidoglycan building blocks, ld-carboxypeptidases are thought to play a role in peptidoglycan recycling. Despite the pharmaceutical interest in peptidoglycan biosynthesis, the fold and catalytic type of ld-carboxypeptidases are unknown. Here, we show that a previously uncharacterized open reading frame in Pseudomonas aeruginosa has ld-carboxypeptidase activity and present the crystal structure of this enzyme. The structure shows that the enzyme consists of an N-terminal β-sheet and a C-terminal β-barrel domain. At the interface of the two domains, Ser115 adopts a highly strained conformation in the context of a strand-turn-helix motif that is similar to the “nucleophilic elbow” in αβ-hydrolases. Ser115 is hydrogen-bonded to a histidine residue, which is oriented by a glutamate residue. All three residues, which occur in the order Ser-Glu-His in the amino acid sequence, are strictly conserved in naturally occurring ld-carboxypeptidases and cannot be mutated to alanines without loss of activity. We conclude that ld-carboxypeptidases are serine peptidases with Ser-His-Glu catalytic triads.

LD-Carboxypeptidase activity can be detected in many Gram-negative and Gram-positive bacteria, but is absent in Caulobacter crescentus (5). It is still not clear whether all LD-carboxypeptidase activity is due to a single enzyme. Metz et al. (6,7) reported that they could distinguish two LD-carboxypeptidase activities in Escherichia coli (termed I and II) by their sensitivity to D-amino acids and to the ␤-lactam antibiotic thienamycin. Activity I was purified and attributed to a 12-kDa norcardicin A-sensitive enzyme (7). There are no reports on purification of activity II. In later independent work, Ursinus et al. (2) reported the purification of a norcardicin A-sensitive and thienamycin-insensitive activity, which was originally ascribed to a 43-kDa dimer-forming protein in the periplasm. When this activity was cloned, the gene was found to code for a 33.6-kDa protein without leader sequence, and a reassessment of its localization suggested that the enzyme is cytosolic (1).
Deletion of the cloned LD-carboxypeptidase gene has little effect on the phenotype during logarithmic growth, but makes the strain prone to lysis in the stationary phase, unless a compensating mutation in AmpD, another peptidoglycan amidase, is present. In contrast to the results of Metz et al. (6,7), Templin et al. (1) found a complete loss of all soluble LD-carboxypeptidase activity in the deletion strain. In this study, we therefore use the term LD-carboxypeptidase for the cloned enzyme, even though unrelated LD-carboxypeptidases may remain to be discovered.
The substrate specificity of the E. coli LD-carboxypeptidase has been well characterized (8). The enzyme is specific for tetrapeptides and discriminates against pentapeptides, but is permissive with regard to the attached sugar moieties: free peptides and peptides linked to MurNAc, GlcNAc-MurNAc, and UDP-MurNAc are substrates. High molecular mass murein sacculi or cross-linked muropeptides are not cleaved (see Fig. 1) (1,8).
The inhibitor sensitivity of LD-carboxypeptidases is less clear, in part because many studies have been done with poorly characterized enzyme preparations that may not have been homogeneous (6, 7, 9 -11). LD-Carboxypeptidase is sensitive to norcardicin A, but biochemical results suggest that the interaction is noncovalent and does not involve an opening of the lactam ring of the antibiotic (2). The sensitivity of LD-carboxypeptidase to other lactam antibiotics varies widely between antibiotics of the same class, but correlates with the chirality of the amino acid substituent of the antibiotic, again suggesting that lactam ring opening is not essential (2,6). EDTA, which was used for standard periplasm extractions when the enzyme was thought to reside in the periplasm, does not inactivate the enzyme (1,12). Extractions with other stronger metal chelators have not been reported.
Therefore, the catalytic class of LD-carboxypeptidase remains unknown, although the pH optimum of ϳ8.4 suggests that the enzyme is probably not an aspartic protease (2). As peptidases of unknown fold and catalytic class, LD-carboxypeptidases are currently classified as family U61 (U for "unknown") in the MEROPS Database (13). Here, we (a) show that the U61 enzyme from Pseudomonas aeruginosa has LD-carboxypeptidase activity, (b) present the crystal structure of the enzyme at 1.5-Å resolution, (c) demonstrate that the enzyme is a serine peptidase with a Ser-His-Glu catalytic triad, and (d) highlight an unexpected similarity of the region around the active-site serine residue to the "nucleophilic elbow" in ␣␤-hydrolases.

EXPERIMENTAL PROCEDURES
Cloning-The gene for LD-carboxypeptidase from P. aeruginosa strain PA01 was amplified by standard PCR methods from genomic DNA, adding EcoRI and BamHI sites for cloning. The fragment was digested with EcoRI and BamHI, and the digestion product was ligated with T4 ligase into a derivative of pET15b (Novagen) that lacks the original EcoRI site of the vector and carries six histidine residues and an EcoRI site downstream of the original NcoI site. Thus, the N-terminal sequence of our construct was MGHHHHHHEFMTS . . . , where MTS comprises the first residues of the native LD-carboxypeptidase sequence. Restriction analysis and automated DNA sequencing confirmed that the expression construct harbored the full LD-carboxypeptidase gene. The conceptually translated sequence differed from the Swiss-Prot sequence (accession number Q9HTZ1) at two positions. Asp 30 and Gln 104 were both mutated to glutamate. As the recombinant protein is active (see below), it is likely that the point mutations are due to strain differences, but this was not rigorously checked. Site-directed mutations were introduced into the expression construct by the QuikChange method (Stratagene) essentially according to the manufacturer's protocol, but with Pfu polymerase (EURx Ltd.) instead of PfuTurbo polymerase.
Protein Purification-Cells were harvested; resuspended in buffer A (50 mM Tris-HCl (pH 7.5) and 200 mM NaCl); and treated with lysozyme, DNase I, and phenylmethylsulfonyl fluoride. After sonication and centrifugation at 145,000 ϫ g, the supernatant was applied to a nickel-nitrilotriacetic acid-agarose column (Qiagen Inc.) equilibrated with buffer A. The column was washed first with buffer A and then with buffer A containing 100 mM imidazole and developed with 300 mM imidazole in buffer A. The eluate was dialyzed for 4 h against 200 volumes of 10 mM Tris-HCl (pH 7.5) and concentrated to 6 mg/ml by ultrafiltration (Centricon YM-30 filters). All purification and concentration steps were done at 7°C.
Substrate Preparation-Pseudomonas putida peptidoglycan was isolated essentially as described by Glauner (15). Briefly, bacteria were grown in 1 liter of LB medium to A 600 ϳ 2.0, harvested, resuspended in 10 ml of cold water, and added dropwise with vigorous mixing to 10 ml of boiling 8% SDS. Samples were boiled for an additional 30 min and incubated overnight at room temperature. Peptidoglycan was recovered by centrifugation at 145,000 ϫ g for 1 h at room temperature and washed several times to remove traces of SDS. The polymer was resuspended in 1.2 ml of 10 mM Tris (pH 7.0) and treated first with 120 g of ␣-amylase (Fluka) for 2 h at 37°C to degrade high molecular mass glycogen and then incubated with 240 g of Pronase (Fluka) for 1.5 h at 60°C to release covalently bound lipoprotein. After another wash with boiling 8% SDS, the pellet was resuspended in 1.2 ml of 10 mM Tris (pH 7) and treated overnight with 200 -300 g of lysozyme at 37°C. Material that remained insoluble after the lysozyme treatment was pelleted and discarded. The supernatant was filtered through a 5-kDa cutoff filter (Vivascience) and subjected to time-of-flight mass spectrometric analysis. For the most prominent peak, a molecular mass of 940.4 Ϯ 0.1 Da was found, consistent with the molecular mass of the GlcNAc-MurNAc disaccharide with a tetrapeptide side chain (see Fig. 1). A priori, the glycosidic linkage in the disaccharide could connect either C-1 of Glc-NAc and C-4 of MurNAc or, alternatively, C-1 of MurNAc and C-4 of GlcNAc. The assignment in Fig. 1 is based on the known specificity of lysozyme (16) and was not checked experimentally. To estimate the amount of GlcNAc-MurNAc tetrapeptide, several methods were tried. Derivatization of the free amino group in meso-diaminopimelic acid with Sanger reagent (fluorodinitrobenzene) failed; thus, reduction of the muramic acid to muramitol was used to titrate the amount of disaccharide. The calculations assumed that 4 mol of disaccharide can be reduced per mol of NaBH 4 with 100% reaction efficiency and are therefore likely to overestimate the true amount of GlcNAc-MurNAc tetrapeptide.
Activity Assay-2 g of GlcNAc-MurNAc tetrapeptide (ϳ2 nmol) was incubated with 20 ng of LD-carboxypeptidase (0.57 pmol of protomer) or control for 20 min at 37°C in either 10 mM Tris (pH 7.0) or crystallization buffer (50 mM citric acid/citrate (pH 4.5)). The wild-type enzyme was active in both buffers. The activity of mutants with altered triad residues was undetectable under these conditions. To establish the limits of their activity, 2 g of mutant enzyme (57 pmol of protomer) was incubated with 2 g of substrate (ϳ2 nmol) for 20 min or overnight at 37°C. The reaction was monitored by HPLC 2 (Waters) using a C18 column (CC250/3 Nucleosil 100-5-C18, Macherey Nagel). The column was equilibrated with buffer B (50 mM phosphoric acid/sodium dihydrogen phosphate (pH 4.3)), washed for 5 min with buffer B after injection, and ramped to buffer C (75 mM sodium dihydrogen phosphate/ disodium hydrogen phosphate (pH 4.95) and 15% methanol) for 40 min. The flow rate was 0.5 ml/min throughout. Retention times are relative to the appearance of the injection peak.
Crystallization-The wild-type protein could be concentrated only to 6 mg/ml. Needle-shaped crystals were grown at room temperature (21°C) by vapor diffusion of 4 l of sitting drops with 0.5 ml of reservoir buffer from a 6 mg/ml protein solution mixed in 1:1 ratio with reservoir buffer containing 0.02 M calcium chloride dihydrate, 0.1 sodium acetate trihydrate (pH 4.6), and 30% (v/v) 2-methyl-2,4-pentanediol. Crystals could be flash-cryocooled from mother liquor and diffracted in-house to ϳ2.9 Å and showed clear signs of high anisotropic mosaicity and disorder.
Much better, plate-like crystals were obtained by vapor diffusion at room temperature (21°C) when a 6 mg/ml protein solution was mixed in 1:1 ratio with reservoir buffer containing 50 mM citric acid (pH 4.5). Unlike the wild-type protein, the selenomethionine variant of LD-carboxypeptidase could be concentrated to 27 mg/ml. Crystallization of the selenomethionine variant was successful with this protein concentration and the same crystallization buffer. For cryoprotection, crystals were soaked in a mixture containing 75% reservoir buffer and 25% (2R,3R)-(Ϫ)-2,3-butanediol. Crystals belonged to space group P2 1 with cell constants a ϭ 52.2, b ϭ 78.3, and c ϭ 71.3 Å and ␤ ϭ 104.3°. The presence of a strong signal for a local 2-fold axis in the self-rotation function and the solvent content indicated that the crystals contained two subunits of LD-carboxypeptidase in the asymmetric unit.
Structure Determination-A native data set to 1.5-Å resolution and a three-wavelength selenomethionine MAD data set to 2.4-Å resolution were collected at beamline BW6 (Deutsches Elektronen Synchrotron, Hamburg, Germany). As each protomer could contain up to four selenomethionine residues (the initiator methionine of the N-terminal His tag, the native initiator methionine, and two methionine residues in the sequence), up to eight selenomethionine sites were expected, but only four sites were identified by SHELXD (17). Phasing and solvent flattening suggested a clear preference for one hand over its enantiomeric alternative, but the resulting map was clearly of insufficient quality for model building. Nevertheless, it was good enough for the program GETAX (18) to define the location of the dimer axis, which turned out to be consistent with the positions of four selenium atoms. To further improve phases, we screened heavy atom soaks for the possible generation of derivatives in-house. A tungstate soak showed a clear peak in the y ϭ 0.5 Harker section. Its location, originally identified with the Patterson search program RSPS (19), was consistent (except for the floating y coordinate) with the location derived from the difference Fourier maps that were phased with the selenomethionine phases. The reverse cross-phasing step, starting from the phases of the tungstate soak and interpreting anomalous selenomethionine difference Fourier maps, was successful as well. Joint MLPHARE phasing (20) of the single isomorphous replacement and MAD data sets followed by automasked 2-fold averaging and solvent flattening with the program DM at 2.5-Å resolution yielded improved phases (TABLE ONE). The map that was calculated with these phases and the amplitudes of the native data set could in part be interpreted manually. Sequence assignment was greatly aided by the information on the locations of the methionine seleniums in the selenomethionine variant. A manually built model of the two protomers that covered ϳ50% of the sequence was sufficient for ARP/wARP to deliver a nearly complete and almost fully refined model (21). The LD-carboxypeptidase vari-ants were analyzed by mapping the model of the wild-type enzyme into the nearly identical unit cells of the mutants, followed by rigid body refinement and omit map calculations. The model was manually corrected in the vicinity of the active site, but otherwise left untouched. Further refinement lowered the R-factor, but introduced undesirable distortions and was therefore not used.

RESULTS
Choice of Enzyme-The best characterized LD-carboxypeptidase is the enzyme from E. coli, which is very unstable (2). Therefore, we decided to concentrate on homologs from other bacterial species. The conserved hypothetical protein from P. aeruginosa (Swiss-Prot accession number Q9HTZ1) could be expressed in E. coli, was stable enough for purification, and could be crystallized. As the open reading frame had never been characterized, a substrate was required to test its activity.
Substrate Preparation-If the P. aeruginosa hypothetical protein (Swiss-Prot accession number Q9HTZ1) were an LD-carboxypeptidase, it should be able to cleave the GlcNAc-MurNAc tetrapeptide, an abundant fragment in lysozyme digests of many Gram-negative bacteria. As P. aeruginosa is a human pathogen and subject to handling restrictions, we purified and digested the peptidoglycan from the closely related P. putida instead. Mass spectrometry showed that the monoisotopic mass of the dominant fragment in the lysozyme digests matched the expected mass of the disaccharide tetrapeptide fragment. Most preparations were significantly contaminated with a second species, which was identified as the disaccharide tripeptide according to its mass. Based on prior data on the murein structure in Gram-negative bacteria and on the specificity of lysozyme, which splits between C-1 of MurNAc and C-4 of GlcNAc (16), the structures in Fig. 1 were assigned to the disaccharide tri-and tetrapeptides in the peptidoglycan digests.
When murein lysozyme digests were chromatographed on a C18 reverse phase column, four peaks instead of the expected two peaks were found ( Fig. 2A). Such peak splitting was reported previously for lysozyme digests of E. coli peptidoglycan and was attributed to the equilibrium between the anomeric forms of muramic acid with the C-1 OH in the ␣and ␤-positions (15). The behavior of P. putida peptidoglycan digests is consistent with this interpretation. When material from a single chromatography peak was re-injected, a peak doublet was again observed (data not shown). When the mixture was pretreated with the reducing agent sodium borohydride, only two instead of the usual four peaks were observed. The altered positions of the new peaks indicated that reduction of muramic acid to muramitol had indeed taken place (Fig. 2D). The P. aeruginosa Hypothetical Protein Has LD-Carboxypeptidase Activity-The recombinant P. aeruginosa hypothetical protein (Swiss-Prot Accession Number Q9HTZ1) converted the GlcNAc-MurNAc tetrapeptide efficiently to the tripeptide. As both substrate peaks disappeared, we concluded that both anomeric forms of the GlcNAc-Mur-NAc tetrapeptide are substrates. The GlcNAc-MurNAc tetrapeptide with muramitol instead of muramic acid, which resulted from the sodium borohydride treatment, was still a substrate (Fig. 2, D and E), suggesting that the region around the anomeric carbon of the GlcNAc-MurNAc tetrapeptide is not important for substrate recognition. As the LD-carboxypeptidase substrate preparations were contaminated with the product, the reaction was best monitored by subtracting the HPLC trace prior to the reaction from the trace after the reaction. In this mode of presentation, the destruction of substrates gives rise to negative ("downward") peaks, and the emergence of products manifests itself as positive ("upward") peaks (Fig. 2, C and F).
Structure Determination-To identify candidate active-site residues of LD-carboxypeptidase, the crystallographic approach was taken. P2 1 crystals of LD-carboxypeptidase from P. aeruginosa with two protomers in the asymmetric unit could be grown and diffracted to a resolution of 1.5 Å on beamline BW6 (Deutsches Elektronen Synchrotron). The selenomethionine variant of the protein was produced and crystallized, but because it contained only two internal well ordered selenomethionine residues/protomer, the MAD phases were not sufficient for model building and had to be combined with the single isomorphous replacement phases for a tungstate derivative. The combined phases were improved and extended by 2-fold averaging and solvent flattening as described under "Experimental Procedures." A partial, manually built model served as the starting point for automatic density interpretation by the ARP/wARP procedure (21). The final refinement statistics appeared satisfactory (TABLE TWO). The C-␣ trace of the LD-carboxypeptidase dimer is presented in Fig. 3.
A Dimer in the Crystal and Likely in Solution-The P2 1 crystals contain two protomers in the asymmetric unit, which are related by a pure 2-fold rotation without a screw component (Fig. 3). The interface buries a total of 2800 Å 2 and extends over 1400 Å 2 and is therefore sufficiently large to be potentially relevant in solution. We therefore scored this contact region with the DCOMPLEX server, which applies statistical potentials to distinguish biologically relevant from purely crystallographic protein-protein contacts (22). The server suggested a value of ϳ25 kcal/mol for the affinity between the protomers and classified the interface as a "true interface." To confirm this conclusion experimentally, we performed size chromatography runs in two different salt concentrations. P. aeruginosa LD-carboxypeptidase migrated with apparent molecular masses of 56 and 51 kDa in low salt (10 mM Tris (pH 7.5) and 50 mM NaCl) and high salt (10 mM Tris (pH 7.5) and 200 mM NaCl) buffers, respectively. These values are in between the calculated protomer mass of 34.6 kDa and the calculated dimer mass of 69.2 kDa (Fig.  4, A and B) and favor the dimer, at least for the high protein concentrations that were used for size chromatography. As the E. coli LD-carboxypeptidase is known to dimerize as well, we suspected that dimerization may be a general feature of LD-carboxypeptidases. To assess the conservation of the dimerization interface, we used the ConSurf server to map amino acid conservation scores to the protein surface (Fig. 5) (23). Fig. 5 (compare A, B, and C) shows that the interface is only slightly more conserved than other regions of the protein surface. Therefore,

Data collection and refinement statistics for wild-type and mutant LD-carboxypeptidases
We note that each LD-carboxypeptidase protomer contains two residues (Ser 115 and Arg 57 ) at position iϩ1 of a type IIЈ ␤-turn, which is normally reserved for glycines. Arg 57 is far away from the active site and therefore not discussed. r.m.s.d., root mean square deviation.
C-terminal Domain-The C-terminal domain is built around a seven-stranded ␤-barrel with a strand order of 1-7-6-2-3-4-5 and sheer number 10, which is intermediate between the minimal sheer 8 and the maximal sheer 12, which have been found for seven-stranded ␤-barrels (25,26). The barrel is made from a three-stranded parallel sheet and a four-stranded antiparallel sheet that have been joined together on the edges. The preference for right-handed ␤-␣-␤ motifs in the antiparallel part of the barrel places all connecting helices on the outside of the barrel (Figs. 3 and 6 and supplemental Fig. 1, C and D).
Active Site-The insensitivity of E. coli LD-carboxypeptidase to EDTA (1, 12) and our failure to locate zinc ions in a crystal of LD-carboxypeptidase that had been soaked with 1 mM Zn 2ϩ prior to the collection of a two-wavelength MAD data set suggested that the enzyme is probably not a metallopeptidase. We therefore inspected all serine and cysteine residues of LD-carboxypeptidase for environments that could render them more nucleophilic. The only strong candidate for an enzyme nucleophile from this search was Ser 115 . As shown in Fig. 7, the O-␥ atom of this serine residue is located in the plane of the imidazole ring of His 285 and only 3.0 Å away from the N-⑀ atom. We presume that His 285 is uncharged and remains in the tautomeric state with the proton on the N-␦ atom. In this scenario, the serine O-␥ atom could donate a hydrogen bond to the N-⑀ atom of the imidazole ring either directly or via a bridging water molecule, and the N-␦ atom of the imidazole ring could donate a hydrogen bond to an oxygen atom of the terminal carboxylate of Glu 217 . We point out that such a hydrogen-bonding arrangement would not only orient the plane of the imidazole ring, but would also be expected to make His 285 N-⑀ more basic and thus better suited as a general base in catalysis.
Structural Features in Support of the Proposed Triad-Many features of the crystal structure support the interpretation that LD-carboxypeptidase is a serine protease and that Ser 115 , His 285 , and Glu 217 form a functional catalytic triad. (a) It is known that active sites are often located in regions with rare backbone conformations (27,28). Ser 115 is located in such a region. For both subunits, our model places this residue at position 2 (also called "iϩ1") of a perfect type IIЈ turn, which is normally reserved for glycines (29). The serine residue at this position in the LD-carboxypeptidase structure has (, ) ϳ (60°, Ϫ130°) and falls in either the generously allowed or disallowed region of the widely used versions of the Ramachandran plot for non-glycine residues (30 -33). (b) Ser 115 is strategically located at the N terminus of a helix, where the helix dipole moment would be expected to favor deprotonation and thus catalysis. For cysteines in this location, it is known that deprotonation can occur prior to catalysis (34); for serine residues, which are more basic, deprotonation is more likely to happen during catalysis. (c) The proposed active site is strategically located at the interface between the sheet and barrel domains. The proposed nucleophile (Ser 115 ) is anchored on the N-terminal sheet domain, whereas the other two triad residues (His 285 and Glu 217 ) are placed on the C-terminal barrel domain, allowing a putative substrate to bind to the cleft at the interface between the two domains. Moreover, such a location would place a substrate essentially at the "top" of the seven-stranded barrel in the C-terminal domain, thought to be the preferred binding place for substrates in both ␤and ␣␤-barrels. (d) A large number of ordered water molecules are found in the vicinity of the active site, some of which may be bound to otherwise unfilled substrate-binding subsites. (e) All three proposed triad residues are strictly conserved among LD-carboxypeptidases and their homologs. Moreover, ConSurf mapping of alignment scores to the protein surface showed that the active-site region is among the most conserved regions of the protein (Fig. 5A).
Mutagenesis Data Support the Proposed Triad-The role of the proposed triad residues was checked experimentally by mutating them individually to alanine. In a direct comparison of the activities of the wild-type enzyme and the S115A, H285A, and E217A mutant proteins,  Fig. 3. B, view of one protomer after removal of the other protomer and a 90°rotation about the horizontal axis to display the dimerization interface. The green line marks the surface above residues that have at least one atom in contact (cutoff distance of 3.9 Å) with the other protomer. The yellow line surrounds the surface above residues that have at least one side chain atom in contact with the other protomer. C, view of the same protomer in B after a further 180°rotation about the horizontal axis. Relative to the orientation in A, this view represents a view of the top of the dimer. The dimer axis is represented by a dotted line in A-C. Arrows mark the locations of the active-site serine residues. This figure was made with the PyMOL program (available at www.pymol.org).  DECEMBER 9, 2005 • VOLUME 280 • NUMBER 49 all mutant proteins behaved indistinguishably from the control without enzyme (data not shown). To make the comparison more quantitative, we incubated 2 nmol of substrate in a 24-l sample volume with either 20 ng of wild-type enzyme or a 100-fold excess of the mutant proteins. Total substrate turnover after 20 min and after an overnight incubation was quantified by HPLC (Fig. 8). We concluded that any residual activity The numbering and secondary structure information are for the P. aeruginosa sequence. The boundaries for secondary structure elements were determined with the DSSP program (55). ␣-Helices and ␤-strands are shown in black, and 3 10 -helices are shown in gray. The information on the 3 10 -register (gray) at the N terminus of the helix after the active-site serine residue was added manually. Only ␤-strands within the ␤-sheet in the N-terminal domain (N␤) and the ␤-barrel in the C-terminal domain (C␤) are numbered. The catalytic triad residues Ser 115 , Glu 217 , and His 285 are marked by arrowheads. Poorly ordered or disordered amino acids at the interface between the N-and C-terminal domains are boxed. Background shading in the alignment indicates the residue conservation scores for all amino acids in the P. aeruginosa sequence. The darker the background, the higher the sequence conservation score. Conservation scores include sequence information from all LD-carboxypeptidase homologs in the NCBI Database. Iterative searches with the BLAST program (56) were used to find LD-carboxypeptidase homologs. Representative sequences (threshold of 0.9) were identified with the CD-HIT program (57) and were aligned with the ClustalX program (58), which was also used to calculate the sequence conservation scores. This figure was done with the ALSCRIPT program (59).

LD-Carboxypeptidase
of the mutants was within experimental error and at least 500-fold lower than the activity of the wild-type enzyme.
Crystallization and Structure Determination of Active-site Mutants-The mutagenesis results showed that S115A, H285A, and E217A are required for enzyme activity, but did not necessarily prove that these residues are directly involved in catalysis because the defects of the mutant enzymes could also be due to impaired substrate binding or an overall folding problem. To test these possibilities, we assayed the ability of the mutants to protect substrates from the activity of the wild-type enzyme. The results were inconclusive and poorly reproducible, suggesting that the mutants may not be very stable (data not shown). The CD spectra of the wild-type and mutant enzymes suggested that all proteins were folded, but it remained unclear whether significant differences in the spectra could be attributed to different protein contaminations alone (data not shown). Therefore, we decided to verify the structures of the mutant proteins crystallographically. The S115A and H285A mutants could be crystallized under the same conditions as the wild-type enzyme and formed crystals in the same space group as the wild-type enzyme with very similar cell constants. The S115A mutant was essentially indistinguishable from the wild-type enzyme, except that there was no density in the place of the Ser 115 O-␥ atom, which confirmed the mutation. The H285A mutant enzyme was also indistinguishable outside the active-site region, but Glu 217 , anchored by hydrogen bonds from His 285 and Asn 192 in the wild-type enzyme, flips into a new position in the H285A mutant enzyme (Fig. 7B). Altogether, the crystallographic analysis of the mutants showed that, for the S115A and H285A mutant LD-carboxypeptidases, the lack of activity cannot be attributed to a folding defect.

DISCUSSION
A Novel Peptidase-Dali quantitative structure comparisons (35) between LD-carboxypeptidase and the proteins in the Protein Data Bank indicated that LD-carboxypeptidase is structurally most similar to uroporphyrinogen III synthase (code 1wd7) (36), the receiver domain of a plant ethylene receptor (code 1dcf) (37), and a signaling protein (code 1tjy) (38). The Dali Z-scores, which measure the similarity in S.D. above average, were 7.5, 6.2, and 6.0, respectively. The high values probably result from the canonical architecture of LD-carboxypeptidase, which is built around a ␤-sheet and a ␤-barrel, and provide no insight about function or mechanism. Altogether, the Dali search algorithm identified Ͼ60 proteins with Dali Z-scores Ͼ4, with no clear cutoff between "true" and "false" positives. We therefore searched this list for hydrolases. S-Adenosylhomocysteinase (Protein Data Bank code 1b3r), a thioether hydrolase that uses reversible redox chemistry to facilitate the hydrolysis reaction (39), scored highest and is likely a purely structural match. Interestingly, the second highest scoring hydrolase, with a Dali Z-score of 4.7, was dienelactone hydrolase (Protein Data Bank code 1din), an ␣␤-hydrolase with a Cys-His-Asp catalytic triad (40). The good structural match between this enzyme and LD-carboxypeptidase is the result of the nearly identical ␤-sheet topologies of the two enzymes. The optimal structural superposition brings the nucleophiles of the active sites in roughly the same region, but places them in nonequivalent loops of the two proteins (data not shown).
A Strained Active-site Serine Residue-The most remarkable feature of the suggested active site in LD-carboxypeptidase is the very unusual main chain conformation of the active-site serine residue, which has Ramachandran angles in the lower right quadrant of the Ramachandran plot. We believe that the unusual serine conformation is a feature of the model and not a crystallographic error: the structure was solved at 1.5-Å resolution, which is sufficient to locate carbonyl oxygen atoms with confidence. Weighted 2F o Ϫ F c and omit maps show robust density for this region, and B-factor values for all non-hydrogen atoms are below 15 Å 2 (except for the Ser 115 O-␥ atom, which has B-factors slightly below 20 Å 2 ) in the two subunits. The modeled conformation implies hydrogen bonds from Phe 114 O to Ile 117 NH and from Ser 115 O to Ser 118 NH  and O-␥H, which together could compensate for the repulsion between Ser 115 C-␤ and Phe 114 O. In our model (which was refined with standard van der Waals terms), the distance between these two atoms is 2.9 Å and is thus significantly shorter than the 3.3-Å van der Waals equilibrium distance (41), but still longer than the 2.8-Å "normally allowed" distance that was used in the original work of Ramachandran et al. (30). Thus, the clash between Ser 115 C-␤ and Phe 114 O cannot be very severe.
We note that the conformation of the active-site serine residue in LD-carboxypeptidase is not unprecedented and that similar conformations have been described in two recent surveys of Ramachandran outlier residues. Gunasekaran et al. (42) found such conformations in the second position (iϩ1) of perfect type IIЈ ␤-turns. In a follow-up on this work, Pal and Chakrabarti (43) reported that such residues (in their nomenclature, "region II" residues) are "common in the first position of 3 10 -helices, which in the majority of cases lead into ␣-helices." As type IIЈ ␤-turns are essentially short 3 10 -helices, the results of Gunasekaran et al. and Pal and Chakrabarti are perfectly consistent. The active-site serine residue in LD-carboxypeptidase fits their description: it is the iϩ1 residue of a type IIЈ ␤-turn, which itself is part of a 3 10 -helix, which leads into a regular ␣-helix.
A Nucleophilic Elbow in LD-Carboxypeptidase and ␣␤-Hydrolases-The location of a serine residue at a sharp kink between a ␤-strand and an ␣-helix reminded us of the nucleophilic elbow in ␣␤-hydrolases. ␣␤-Hydrolases form a structurally defined, very large group of hydrolases that use serine or cysteine as the nucleophile, histidine as the general base, and aspartate or glutamate as the third triad residue (44,45). In the lipase from Geotrichum candidum, which was analyzed at 1.8-Å resolution, the triad is Ser-His-Glu, as in LD-carboxypeptidase (46,47). Therefore, we chose this enzyme as a representative of the ␣␤-hydrolases for a detailed comparison (Fig. 9). In both structures, the joint in the nucleophilic elbow is a type IIЈ ␤-turn with the serine in the iϩ1 position, even though the turn in ␣␤-hydrolases was originally and inaccurately described as "␥-like" (44). In both structures, the serine residue has the same unusual main chain conformation, which in ␣␤-hydrolases is thought to make the serine easily approachable by substrate and the hydrolytic water molecule (45). ␣␤-Hydrolases and LD-carboxypeptidase superimpose perfectly well in the immediate vicinity of the active-site serine residue, but they differ downstream of the type IIЈ ␤-turn: in LD-carboxypeptidase, it is part of a 3 10 -helix, whereas in the lipase from G. candidum, it is immediately followed by a standard ␣-helix. This difference results in a significantly different orientation of the helix relative to the serine residue. Remarkably, there is an almost equivalent difference in the orientation of the ␤-strand upstream of the active-site residue, so the overall shape of the nucleophilic elbow is very similar in both proteins. For ␣␤-hydrolases, it has been noted that the sharp turn at the nucleophilic elbow, which packs the helix very tightly against the preceding strand, constrains the sequence around the active-site serine residue to GXSXG (44,45). In LD-carboxypeptidases, which have a slightly different backbone, the motif is GXSDX, without the requirement for the second glycine residue. The conservation of the aspartate residue immediately downstream of the active-site serine residue is understandable: its side chain carboxylate oxygen atoms accept hydrogen bonds from main chain amides and thus help to fix the sharp turn in the nucleophilic elbow (Fig. 9).
A Ser-His-Glu Triad-A Ser-His-Glu triad is not only present in the lipase from G. candidum (47), but was also found in several other ␣␤-hydrolases such as acetylcholinesterase (48) and butyrylcholinesterase (49). Interestingly, all these enzymes cleave phosphate esters, not amide bonds. In peptidases with a serine nucleophile and histidine general base, aspartate and not glutamate is the norm for the third triad residue. The rule applies to trypsin-like (MEROPS clan PA), subtilisinlike (MEROPS clan SB), and prolyl oligopeptidase-like (MEROPS clan SC) peptidases, which are so different in their overall folds that they cannot have arisen from a single ancestor. The preference for aspartate over glutamate as the third triad residue in serine peptidases is not understood and may be fortuitous; nevertheless, to the best of our knowledge, aspartyl dipeptidase with its Ser-His-Glu triad is so far the only exception to this rule (50). Our present work on LD-carboxypeptidase adds another peptidase to the still very short list of hydrolases with a Ser-His-Glu triad that cleave peptide bonds and not phosphate esters.
Bacterial Versus Eukaryotic LD-Carboxypeptidases-Early biochemical data indicated that several different LD-carboxypeptidase activities may be present in bacteria (6), but only one such activity has been traced back to a cloned and characterized protein (1). Surprisingly, weak LDcarboxypeptidase activity has also been reported for a eukaryotic protein, the Drosophila peptidoglycan recognition protein-SA, which is involved in the sensing of bacterial infection and in the activation of the Toll pathway. Peptidoglycan recognition protein-SA shares the fold with T7 lysozyme, but has lost the metal-binding site and has a Ser-His dyad instead (51). Based on the crystal structure, a threonine residue might act as the third triad residue, but this hypothesis was not checked biochemically (51). Either way, it is clear that the LD-carboxypeptidases from Drosophila and P. aeruginosa are different.
LD-Carboxypeptidase Homologs with a Different Activity-Bacterial LD-carboxypeptidases are genome-encoded proteins. E. coli strains that produce the antibiotic microcin C7 additionally contain a plasmid-encoded LD-carboxypeptidase homolog, which mediates resistance against exogenously added microcin C7 and is therefore also known as microcin C7 self-immunity protein MccF (52). Sequence alignment of LD-carboxypeptidases and MccF showed that all three triad residues of LD-carboxypeptidase are conserved in MccF (Fig. 6), strongly suggesting that MccF shares not only the fold, but also the hydrolytic activity with LD-carboxypeptidases. Substrates of MccF are not known, but microcin C7 precursors are obvious candidates. However, microcin C7 maturation requires only one hydrolysis reaction, for which no enzyme has yet been found: the conversion of asparagine to aspartate (53). As this reaction could also occur spontaneously and because the link of this reaction to self-immunity is unclear, the question of the physiological activity of MccF remains unresolved.
More LD-Carboxypeptidases-At present, proteins with homology to LD-carboxypeptidase are variously annotated as conserved hypothetical proteins, (putative) muramoyl-tetrapeptide carboxypeptidases, or (putative) microcin C7 self-immunity proteins in the sequence data bases. According to the annotations, there are about as many self-immunity proteins as LD-carboxypeptidases. This classification is suspicious because (a) most bacteria are not expected to produce microcin; (b) nearly all proteins are chromosomally encoded; (c) the presence of a microcin C7 self-immunity protein usually correlates with the lack of an LD-carboxypeptidase; and (d) the LD-carboxypeptidases and microcin C7 self-immunity proteins do not segregate into two separate branches in phylogenetic trees. Therefore, it is likely that many (but certainly not all) proteins that are currently annotated as (putative) microcin C7 selfimmunity proteins are actually LD-carboxypeptidases.
Other LD-Carboxypeptidases?-Our crystallographic and biochemical results have unexpected bearing on the still unanswered question of whether bacteria (and E. coli in particular) contain enzymes with LD-carboxypeptidase activity that are not homologous to the LD-carboxypeptidase described in this work. Previous work had traced some LD-carboxypeptidase activity to a 12-kDa protein that behaved as a monomer (7). When it was later found that the knockout of the only known LD-carboxypeptidase gene abolished all soluble LD-carboxypeptidase activity (1), it became likely that the 12-kDa protein was its proteolytically truncated version. Based on the results of this work, we can now rule out this possibility: Ser 115 and His 285 of the P. aeruginosa LD-carboxypeptidase triad align with Ser 106 and His 270 of E. coli LD-carboxypeptidase. As these residues are separated by 163 amino acids, any active degradation product with peptidase activity cannot be Ͻ18 kDa. To strengthen this conclusion, we looked for the minimal spacer between the active-site serine and histidine residues in LD-carboxypeptidase homologs and found that the spacer is always larger than the 12-kDa mass of the unknown LD-carboxypeptidase activity. Therefore, if 12 kDa is indeed the correct molecular mass of the unknown active enzyme, this enzyme cannot be a homolog of the LD-carboxypeptidase described in this work.