A New Family of Cyanobacterial Penicillin-binding Proteins

It is largely accepted that serine β-lactamases evolved from some ancestral DD-peptidases involved in the biosynthesis and maintenance of the bacterial peptidoglycan. DD-peptidases are also called penicillin-binding proteins (PBPs), since they form stable acyl-enzymes with β-lactam antibiotics, such as penicillins. On the other hand, β-lactamases react similarly with these antibiotics, but the acyl-enzymes are unstable and rapidly hydrolyzed. Besides, all known PBPs and β-lactamases share very low sequence similarities, thus rendering it difficult to understand how a PBP could evolve into a β-lactamase. In this study, we identified a new family of cyanobacterial PBPs featuring the highest sequence similarity with the most widespread class A β-lactamases. Interestingly, the Ω-loop, which, in the β-lactamases, carries an essential glutamate involved in the deacylation process, is six amino acids shorter and does not contain any glutamate residue. From this new family of proteins, we characterized PBP-A from Thermosynechococcus elongatus and discovered hydrolytic activity with synthetic thiolesters that are usually good substrates of DD-peptidases. Penicillin degradation pathways as well as acylation and deacylation rates are characteristic of PBPs. In a first attempt to generate β-lactamase activity, a 90-fold increase in deacylation rate was obtained by introducing a glutamate in the shorter Ω-loop.

It is largely accepted that serine ␤-lactamases evolved from some ancestral DD-peptidases involved in the biosynthesis and maintenance of the bacterial peptidoglycan. DD-peptidases are also called penicillin-binding proteins (PBPs), since they form stable acyl-enzymes with ␤-lactam antibiotics, such as penicillins. On the other hand, ␤-lactamases react similarly with these antibiotics, but the acyl-enzymes are unstable and rapidly hydrolyzed. Besides, all known PBPs and ␤-lactamases share very low sequence similarities, thus rendering it difficult to understand how a PBP could evolve into a ␤-lactamase. In this study, we identified a new family of cyanobacterial PBPs featuring the highest sequence similarity with the most widespread class A ␤-lactamases. Interestingly, the ⍀-loop, which, in the ␤-lactamases, carries an essential glutamate involved in the deacylation process, is six amino acids shorter and does not contain any glutamate residue. From this new family of proteins, we characterized PBP-A from Thermosynechococcus elongatus and discovered hydrolytic activity with synthetic thiolesters that are usually good substrates of DD-peptidases. Penicillin degradation pathways as well as acylation and deacylation rates are characteristic of PBPs. In a first attempt to generate ␤-lactamase activity, a 90-fold increase in deacylation rate was obtained by introducing a glutamate in the shorter ⍀-loop.
D-Alanyl-D-alanine peptidases are enzymes involved in the synthesis of the peptidoglycan, the bacterial cell wall constituent that is responsible for the cell shape and resistance to osmotic pressure (1). These enzymes catalyze transpeptidation and carboxypeptidation reactions, thus controlling the peptidoglycan synthesis and cross-linking. As a result of the acylation of their catalytic serine, these enzymes form stable acylenzymes with ␤-lactam antibiotics, such as penicillins, the stability of these complexes being at the origin of the antibiotic effect. Hence, DD-peptidases are also called penicillin-binding proteins or PBPs 2 (2,3). PBPs are divided into two main groups: the low M r PBPs are monofunctional catalytic entities (4), and the high M r PBPs comprise an additional N-terminal domain (5). To counteract ␤-lactam antibiotics, some bacteria produce ␤-lactamases (6 -8), which are enzymes able to hydrolyze penicillins up to 10 8 times faster than PBPs. According to their primary structure, serine ␤-lactamases can be divided into three classes: A, C, and D (class B constituting the group of metalloenzymes). The penicillin-binding module of DD-peptidases and ␤-lactamases share a similar three-dimensional structure composed of two domains, an all-␣-domain and an ␣/␤-domain, the catalytic site being at the interface (7). Three common essential motifs line up the active site, following Ambler numbering (9), S 70 XXK containing the active nucleophile serine, (S/Y) 130 XN on a loop in the all-␣-domain and (K/R) 234 (S/T)G on a ␤-sheet forming the opposite wall of the catalytic cavity. Regarding ␤-lactamases, the rapid hydrolysis of the penicilloyl-enzyme is performed by a water molecule activated by an additional enzymatic element. In class A ␤-lactamases, this role is carried out by a strictly conserved Glu 166 (10) situated on a 16-residue loop (Arg 164 -Asp 179 ) usually referred as the ⍀-loop (11,12). For class C and D ␤-lactamases, a tyrosine (13,14) and an unusual carbamylated lysine (15), respectively, are involved in the deacylation mechanism.
On the basis of sequence similarities, it was proposed that each class of PBP is subdivided into three subclasses, some of which are related to specific classes of ␤-lactamases (2). Sequence alignments and comparative analysis of the structures and the reaction mechanisms indicate that these enzymes are phylogenetically linked (3,16,17) and that each class of ␤-lactamase evolved most probably from an ancestral PBP, bringing to the active site an efficient catalysis mechanism for the deacylation reaction. The low M r class C PBPs, including PBP4 from Escherichia coli and Actinomadura R39 are believed to be the phylogenetically closest PBPs to class A ␤-lactamases, although sequence similarities are almost undetectable except for the presence of the three essential motifs (16). Hence, despite a similar organization of the catalytic sites, the evolutionary pathway has disappeared, and the precise structural and chemical elements that are required in the different reactivities toward ␤-lactams are still not understood.
In this work, we identified a new family of PBPs in cyanobacterial genomes featuring the highest sequence similarity to class A ␤-lactamases so far reported. The protein of the thermophilic Thermosynechococcus elongatus was chosen for biochemical characterization. After cloning, expression, and purification, the protein was examined for hydrolytic and aminolytic activity toward various substrates and for ␤-lactam reactivity. Based on the sequence alignment with class A ␤-lactamases, a few mutations were introduced with the aim of increasing the ␤-lactamase activity. The single mutation L158E in the shorter ⍀-loop afforded a 90-fold improved deacylation of the penicilloyl-enzyme.

EXPERIMENTAL PROCEDURES
Chemicals and Antibiotics-All ␤-lactam antibiotics and D-Ala-D-Ala, Ac 2 -Lys-D-Ala-D-Ala, and Ala-D-␥-Glu-Lys-D-Ala-D-Ala substrates were purchased from Sigma. D-Ala paranitroanilide was from Bachem. Thiolester substrates were a kind gift from M. Galleni (Université de Liège, Belgium). Oligonucleotides were from Eurogentec, with "fp" and "rp" standing, respectively, for forward and reverse primer: pbp-a-fp, 5Ј-TCG CCA TGG CCG CTC CTG AGG CAC C-3Ј; pbp-a-rp, 5Ј Bacterial Strains-T. elongatus BP-1 was used as a source of DNA for PCR amplification of the pbp-a gene (locus: tll2115), and a total genome extract of the cyanobacteria was kindly provided by M. Sugiura (Department of Applied Biological Chemistry, Osaka Prefecture University). E. coli Top 10 (Invitrogen) was used for standard subcloning procedures as well as for protein expression.
Cloning of pbp-a Gene and Mutants-The pbp-a gene was identified in the complete genome of T. elongatus (18,19), available in CyanoBase, the genome data base for cyanobacteria (available on the World Wide Web). Since probable signal peptide and/or transmembrane segments were predicted by various programs available from the ExPASy Proteomics server, between amino acids 1 and 92 (see Fig. 3), the gene encoding the PBP domain of PBP-A protein (i.e. truncated from its 92 first N-terminal residues) was recovered by PCR using primers pbpa-fp and pbp-a-rp and the purified T. elongatus chromosome as a template. In this way, NcoI and XbaI restriction sites were added at the 5Ј and 3Ј termini of the pbp-a gene. The amplified 853-bp fragment was then cloned in the pGEM-T Easy cloning vector (Promega). After sequencing confirmation, the pbp-a gene was cloned in the pBAD/Myc-HisB-Tet vector in fusion with a Myc tag and a His 6 tag. This expression vector is a derivative of pBAD/Myc-HisB (Invitrogen), but the ampicillin resistance marker has been replaced by a tetracycline resistance gene. In the final construct, a methionine was added to the N-terminal end of the truncated protein, giving the numbering Met 1 -Pro 2 . . . , which is used throughout the report. The expression vector for the ⍀-loop deletion mutant PBP-A-⌬ was constructed similarly but the initial PCR fragment was generated by running an overlap extension of two PCR fragments obtained by amplification with the pbp-a-⌬1/pbp-a-rp and pbp-a-⌬2/pbp-a-fp pairs of primers. D160A, L158D, L158E, and S61A mutations were generated with the QuikChange site-directed mutagenesis kit (Stratagene) using the respective pairs of fp and rp primers and the wild type expression vector as a template.
Culture and PBP-A Production-Transformed E. coli Top 10 were grown in 1 liter of Luria-Bertani broth medium supplemented with tetracycline (7.5 g/ml) at 37°C until A 600 reached ϳ0,6. The temperature was lowered to 30°C, and protein production was induced for 3-4 h by the addition of L-arabinose at a final concentration of 0.2% (w/v). The culture was centrifuged at 4400 ϫ g for 10 min, and the harvested cells were stored overnight at Ϫ20°C. The pellet was resuspended in 8 ml of 50 mM Tris-HCl, pH 8, 1 mM EDTA, and 100 mM NaCl supplemented with 100 l of egg white lysozyme (10 mg/ml). After a 20-min incubation at room temperature followed by a 45-min incubation at 37°C, 200 l of DNase (1 mg/ml) was added, and the lysate was incubated for 15 min at room temperature. A cytoplasmic extract was obtained by centrifugation at 21,000 ϫ g for 15 min at 4°C.
PBP-A Purification-The cytoplasmic extract was dialyzed against 20 mM phosphate buffer, pH 7.4, 0.5 M NaCl, 10 mM imidazole (buffer A), using a Spectra/Por membrane (10,000 M r cutoff) overnight, at 4°C. Following passage through a 0.22-m filter, the protein mixture was applied to a Ni 2ϩ -chelating column (HiTrap chelating; Amersham Biosciences), washed with 10 column volumes of buffer A, and finally eluted with a 10 -500 mM imidazole gradient in buffer A. Fractions were analyzed by SDS-PAGE, and those containing PBP-A were pooled and submitted to steric exclusion chromatography on a HiLoad 26/60 Superdex 200 preparation grade column (Amersham Biosciences), using 20 mM Hepes, 1 mM EDTA, pH 7.4, as elution buffer. Protein fractions were analyzed by 10% polyacrylamide gel electrophoresis and either stained with Coomassie Blue R250 or submitted to Western blot using an anti-Myc antibody (Invitrogen). The protein concentration in pure fractions was determined by measuring the absorbance at 280 nm using an extinction coefficient of 13,370 M Ϫ1 cm Ϫ1 .
Enzymatic Assays-The DD-peptidase activity was tested by incubating PBP-A with Ac 2 -L-Lys-D-Ala-D-Ala (1 mM) or D-Ala-D-Ala (1 mM) in 20 mM Tris-HCl buffer, pH 8, in presence or not of a glycine acceptor (1 mM), at 25°C. The release of D-Ala was estimated continuously using the D-amino acid oxidase method (20). Incubation with Ac 2 -L-Lys-D-Ala-D-Ala was also performed at 55°C, aliquots were taken at different times, reaction was stopped by the addition of 1 mM penicillin G, and the release of D-Ala was estimated by the same D-amino acid oxidase method. D-Amino peptidase activity was also assayed on the chromogenic substrate D-Ala paranitroanilide (21) at 25 and 55°C, with concentrations varying from 5 to 18 mM in 20 mM Hepes, 1 mM EDTA, pH 7.4, and following the absorbance of p-nitroaniline at 405 nm (⑀ ϭ 10,200 M Ϫ1 cm Ϫ1 ).
Hydrolysis of 0.3 mM S2a, racemic S2c and S2d, and diastereoisomeric PhacATl thiolester substrates was performed at 37°C and monitored spectrophotometrically at 250 nm (22)(23)(24). Kinetic parameters were extracted by numerical simulation of the progression curves using the following equations, where ⑀ S and ⑀ P are the extinction coefficients of the substrate and the product, and the subscripts 1 and 2 stand for the two enantiomers or diastereoisomers. k 0 is the constant for spontaneous thiolester hydrolysis and was measured independently. ⑀ S , ⑀ P , k cat1 , K m1 , k cat2 , and K m2 were optimized together during the simulations. This is possible because couples of parameters are affecting different regions of the curve.
Penicillin Binding Assay and Rate Measurements-The kinetic scheme for the interaction of a PBP with a ␤-lactam antibiotic is as follows, where E⅐S is the noncovalent Michaelis complex, E-S is the covalent acyl-enzyme complex, and P is the released product. The deacylation rate constant, k 3 , was derived from the first order decay rates of the [ 14 C]penicilloyl-PBP complex. PBP-A protein (between 3 and 10 M) was incubated for 2 min with 10 M [ 14 C]penicillin G in Hepes 20 mM, EDTA 1 mM, pH 7.4. 1 l of penicillinase from Bacillus cereus (Sigma) was added at a concentration such that the remaining free penicillin is hydrolyzed in a few seconds. At various times, 15-l aliquots were removed, and the deacylation was stopped by the addition of 0.5 volumes of SDS-PAGE denaturing buffer (0.5 M Tris-HCl, pH 6.8, 200 mM dithiothreitol, 4% SDS, 0.2% bromphenol blue, and 20% glycerol) and immediate freezing at Ϫ80°C. Reaction mixtures were subjected to SDS-PAGE, gels were dried and exposed for 4 days to the CS screen of a PhosphorImager (GS 525 Molecular Imager System; Bio-Rad), and the quantity of radioactive protein was measured with Molecular Analyst software, version 1.4 (Bio-Rad). The optimized k 3 constants were determined by fitting an exponential to the experimental points. Acylation reactions were performed by mixing 2 M [ 14 C]penicillin G with 3 M protein in 20 mM Hepes, 1 mM EDTA, pH 7.4. Every 10 s, 15-l aliquots were removed and treated using the same protocol described for deacylation. Since (k Ϫ1 ϩ k 2 )/k 1 (or K S ) is most probably much higher than the experimental concentrations of enzyme and substrate, the acylation proceeds via a second order reaction with a rate constant k ac ϭ k 1 k 2 /(k Ϫ1 ϩ k 2 ). The optimized constant was extracted from the data by numerical simulation using the following differential equation, where k 3 is known from the previous deacylation experiment.
␤-Lactam Antibiotics Hydrolysis-All of the spectrophotometric measurements were performed with the help of a thermostated CARY 3 Bio (VARIAN analytical instruments). The time courses of the hydrolysis of the ␤-lactam compounds were performed in 20 mM Hepes, 1 mM EDTA, pH 7.4, with an enzyme concentration around 3 M and penicillin antibiotics at 0.5 mM or cehalosporins at 0.25 mM, following the absorbance at the appropriate wavelength. The rate of the reaction was calculated by dividing the rate of absorbance decrease by the difference of extinction coefficients between the product and the substrate (penicillin G,

RESULTS
Primary Sequence Analysis of PBP-A Family-By searching bacterial genomes for genes encoding penicillin recognizing proteins, we identified a whole family of cyanobacterial proteins annotated as putative class A ␤-lactamases but with a 6-amino acids deletion in the region of the ⍀-loop and missing the essential Glu 166 . A multiple alignment with some class A ␤-lactamases of known structure is shown in Fig. 1. Although the three other motifs SXXK, SDN, and K(S/T)G are conserved, the absence of Glu 166 suggested that these proteins are rather PBPs than ␤-lactamases. Although substitutions in the ⍀-loop can occur naturally, giving rise to extended spectrum lactamases (25), insertions or deletions were mainly genetically engineered in this region (26,27). Only one example of a natural duplication in the ⍀-loop of a class A ␤-lactamase has been reported (28).
Focusing on the shorter ⍀-loop of this new family of proteins reveals a consensus motif X 2 LP(D/E)X 2 GTN. A leucine is aligned with the essential Glu 166 and is followed by a conserved proline in a position where a cis-proline is frequently observed in ␤-lactamases. The presence of a conserved glycine and the absence of the salt bridge usually found across the -loop of ␤-lactamases (R152-D167) must be important for the conformation of this short loop.
For better understanding the evolutionary situation of these cyanobacterial proteins, we searched for homologous sequences in the Protein Data Bank at the NCBI Web site. In this way, we identified several other putative lactamases, most of them from recently sequenced genomes, with variations in the ⍀-loop length, but all of them having the three other motifs SXXK, SXN, and K(S/T)G conserved. A multiple alignment with these proteins of unknown functions, together with some well known LMW class A and C PBPs and class A ␤-lactamases was generated, and a phylogenetic tree was elaborated (Fig. 2). For this purpose, raw protein sequences were manipulated. The putative ⍀-loop of each protein was removed in order to prevent alignment upon the ⍀-loop size that could introduce a bias. Putative sequence signal, transmembrane segments, or insertions at the extremities of the proteins, possibly introducing noise in the alignment, were also removed. Finally, large insertions from LMW class C were removed on the basis of structure comparisons with class A ␤-lactamases. It is important to note that LMW class A and C PBPs could only be aligned correctly when all these manipulations were done, whereas all other proteins aligned well with class A ␤-lactamases using the raw sequences. The alignment with LMW class A and C PBPs, on the other hand, is essentially based on the presence of the three conserved motifs of the active site. Without these elements, the homologies between these two families of proteins and all the others in the tree become unrecognizable. For this reason, the uncertainty of the branchings of both families to the rest of the tree is highlighted by the dotted lines in Fig. 2. As shown in the tree, the family of cyanobacterial proteins appears as the phylogenetically closest family to class A ␤-lactamases, which makes them particularly interesting. We called it the PBP-A family.
Since no member of that family has been characterized so far, we decided to undertake a full biochemical study of one member. We focused on the PBP-A from T. elongatus BP-1 (NCBI accession number 22295842). Compared with ␤-lactamases, it FIGURE 1. Sequence alignment of the ⍀-6 family proteins and some ␤-lactamases of known structure. The alignment was generated using the T-Coffee program (67). Residues conserved over a threshold of 90% are colored. Three essential motifs of the active sites of ␤-lactamases are indicated by red triangles. The ⍀-loop and the essential Glu 166 are indicated by a blue bar and triangle, respectively.
is the closest member of the PBP-A family, and its thermostability makes it the most attractive protein for future protein engineering. The polypeptide deduced from the nucleotide sequence of PBP-A consists of 368 amino acids. Pairwise alignment using BLAST revealed no significant similarity with any known PBP but 29% identity (over 194 residues) with TEM-1 ␤-lactamase from Escherichia coli. Following Ambler numbering, the aligned ␤-lactamase sequence extends from Gly 45 down to the K(S/T)G motif with uniformly distributed identi-ties. With regard to the tridimensional structure, it means that the ␣-domain and the external strands of the ␤-sheet are the main elements that bring PBP-A so close to class A ␤-lactamases.
Cloning and Expression of T. elongatus pbp-a and Mutants-Submission of the protein primary sequence to several Expasy predicting programs revealed different potential start codons, signal peptides and transmembrane segment between amino acids 1 and 92 (Fig.  3). Depending on whether the option Gramϩ or GramϪ was chosen, a signal sequence with cleavage site between Ala 66 2Ala 67 or Ala 92 2Pro 93 , respectively, was predicted with a low score by SignalP. Proteomics studies of Synechocystis sp. PCC 6803 showed that in some cases, start codons for periplasmic proteins were badly assigned by CyanoBase (29). There are several methionines in the N-terminal sequence of PBP-A, suggesting that it may be the case for this protein.
We tried to express two putative full proteins, starting at the first and third methionine Met 28 , but in both cases protein expression was undetectable. A transmembrane segment from residue 58 to 78 was also predicted with high confidence, suggesting that the protein could be membrane-anchored, as is the case for many DD-peptidases. Hence, we decided to remove the first 92 amino acids and cloned the coding sequence from Pro 93 to the C terminus in fusion with a Myc tag and a His 6 tag. Soluble expression of the protein in the cytoplasm was detected, and the 33-kDa PBP-A was purified as described under "Experimental Procedures" with a final yield of 0.22 mg of protein/liter of culture. All mutants were expressed under the same conditions and were obtained as pure proteins.
Enzymatic Activity of PBP-A-The DD-carboxypeptidase activity was assessed by incubating PBP-A with D-Ala-D-Ala and Ac2-Lys-D-Ala-D-Ala substrates. No hydrolysis of these compounds was observed at 25°C or at 55°C, which is the optimal growth temperature of T. elongatus. A very weak activity (v/e 0 ϭ 3.7 10 Ϫ3 s Ϫ1 ) was detected with Ala-D-␥-Glu-Lys-D-Ala-D-Ala (0.8 mM). PBP-A was also inactive on D-Ala-paranitroanilide, a substrate used for the D-aminopeptidase activity of Ochrobacterium anthropi, a member of the serine penicillinrecognizing enzyme (PRE) family (30,31).
Since the peptidoglycan composition is not known for cyanobacteria, substrates containing meso-diaminopimelic acid  . Some well known LMW class A and C PBPs are also represented, but since their homologies with the other proteins are almost unrecognizable, the corresponding evolutionary distances are much less accurate. The distance scale represents the evolutionary distance, expressed in the number of substitutions per amino acid. instead of lysine were also tested. Indeed, this is the most common variation among peptidoglycans of prokaryotes (32). We performed qualitative assays by incubating PBP-A with Lactobacillus plantarum extracts containing cytoplasmic precursors of the peptidoglycan (i.e. UDP-N-acetylmuramyl-pentapeptides) and monitoring cleavage of the C-terminal residue by HPLC (33,34). A very weak cleavage activity was observed with UDP-N-acetylmuramyl-L-Ala-␥-D-Glumeso-diaminopimelic acid-D-Ala-D-Ala, whereas the enzyme was significantly more active on a precursor ending with an ester-linked C-terminal D-lactate instead of the amide-linked D-alanine (data not shown).
Hydrolysis of thiolester substrates at 37°C was observed for both thioglycolates and thiolactates, with a preference for alanine-derived substrates compared with glycine (Table 1). Pha-cATl is a diastereoisomer mixture of phenylacetyl-D-alanylthio-L-lactate and phenylacetyl-D-alanyl-thio-D-lactate. Results show that PBP-A is specific for the D-alanyl group and has a marked preference for one thiolactic stereoisomer. PBP-A was also found to be very active on one enantiomer of the racemic S2d substrate, most probably the D-alanyl-thioglycolate, given the previous result on PhacATl. No activity was detected on the L-alanyl-thioglycolate isomer, indicating a strict D-stereospeci-ficity on that carbon. A lower specificity observed with the racemic glycyl-thio-DL-lactate S2c substrate indicates a more tolerant recognition on the leaving group side. The hydrolysis efficiency of thiolester substrates is very variable among PBPs; the hydrolysis rates and k cat /K m were lower for Streptomyces K15 DD-peptidase, which is homologous to class A ␤-lactamases; and the catalytic activity of PBP-A was in the same range as Streptomyces R61, but the K m values were smaller (22,35). Compared with TEM-1 ␤-lactamase, PBP-A features similar activity on glycyl substrates (S2c and S2a) but is 350-fold more active on the S2d D-alanyl substrate, indicating a key difference in the specificity for the penultimate D-Ala.
Hence, the kinetic values for PBP-A are typical of DD-peptidases, suggesting that this protein is a DD-carboxypeptidase. No increase in the rate of deacylation in the presence of D-alanine as an acceptor could be observed; hence, transpeptidase activity could not be evidenced. As suggested by the high K m observed with all substrates, the limiting step for PBP-A with thiolester seemed to be acylation and may be correlated with the absence of PBP-A reactivity with peptidic D-Ala-D-Ala substrates in the range of concentrations used.
Reactivity toward Penicillins-If a protein is a PBP, it forms a covalent complex with ␤-lactam antibiotics. PBPs incubated

TABLE 1 Kinetic parameters for hydrolysis of thiolester substrates by PBP-A and PBP-A-L158E
Experiments were run at 37°C. X and Y are the two possible stereoisomeric forms of the substrates. Most values have been measured several times with a 10% error in reproducibility. with a radioactive ␤-lactam can be visualized following SDS-PAGE and autoradiography (36). Penicilloylated PBP-A could be visualized as a large band by autoradiography, indicating that the protein is indeed a PBP. PBP-A was also able to hydrolyze penicillin G at a slow rate: k cat ϭ 2.7 ϫ 10 Ϫ4 s Ϫ1 at 20°C and 3.8 ϫ 10 Ϫ3 s Ϫ1 at 55°C. The kinetic parameters of [ 14 C]penicillin G acylation and deacylation by PBP-A were also determined and compared with kinetic constants of TEM-1 and some low M r PBPs ( Table 2). The key element in PBP inhibition by ␤-lactam antibiotics is the high acylation rate followed by a slow deacylation rate, making the acyl-enzyme stable and allowing fast inactivation of the PBP. Large variations in the kinetic constants can be observed among PBPs; the highest k 2 /K s value of 3 ϫ 10 5 M Ϫ1 ⅐s Ϫ1 is observed in Actinomadura R39. In the same way, k 3 varies between 10 Ϫ3 and 10 Ϫ6 s Ϫ1 . By comparison, one of the best ␤-lactam catalysts, TEM-1 from E. coli, has an acylation rate 10 2 to 6 ϫ 10 4 higher than PBPs, and the deacylation rate is 10 5 to 10 8 times faster. The kinetic values of PBP-A are far from being in that range, and with such a molecular weight, this protein can be clustered in the low M r PBP family.
To determine the profile of PBP-A for ␤-lactam reactivity, we spectrophotometrically followed the hydrolysis of different substrates at 55°C at the appropriate wavelength ( Table 3). The protein shows a stronger hydrolytic activity for penicillins than for cephalosporins, which is the same type of profile as the classic TEM-1 class A ␤-lactamase (37).
The penicillin degradation products released by PBP-A were also examined. It has been shown that benzylpenicillin fragmentation in DD-carboxypeptidase-transpeptidases of Streptomyces R61 and R39, Staphylococcus aureus, Bacillus subtilis, Bacillus stearothermophilus, and E. coli can occur at the C7-N4 or C5-C6 bonds, releasing penicilloic acid or phenylacetylglycine and N-formyl-D-penicillamine, respectively, whereas the only product from ␤-lactamases is penicilloic acid (38 -42). In the experiments carried out with Streptomyces R61 DD-peptidase, phenylacetylglycine may be a direct breakdown product of the acyl-enzyme, and N-formyl-D-penicillamine may be a secondary product from hydrolysis of D-5,5-dimethyl-⌬ 2 -thiazoline-4-carboxylic acid (Fig. 4) (38, 42). Release of D-5,5-dimethyl-⌬ 2 -thiazoline-4-carboxylic acid can be followed at 257 nm, and the appearance of free thiol groups from N-formyl-Dpenicillamine can be kinetically followed by using Ellman's reagent (43). For PBP-A, monitoring at 257 nm during incubation with benzylpenicillin showed a first phase with an increase of the absorbance and a second phase where the absorbance slowly decreased. These observations suggest the formation of a product during the first phase followed by its degradation, similar to the transient accumulation of D-5,5-dimethyl-⌬ 2 -thiazoline-4-carboxylic acid observed for some PBPs. Free thiols

. Reaction of benzylpenicillin 1 with DD-peptidases leads to formation of a covalent intermediate 2 and release of a complex mix of products, among which is the normal hydrolysis product, penicilloic acid 4, but also phenylacetyl-glycine 7 and N-formyl-D-penicillamine 8.
Deacylation may occur by two mechanisms: direct hydrolysis or an intramolecular attack generating an oxazolone, 3, which could hydrolyze into penicilloic acid or fragment into D-5,5-dimethyl-⌬2-thiazoline-4-carboxylic acid 6 and oxazolinone 5. Late hydrolysis of these fragments may release products 7 and 8 (38,42).  were also detected using Ellman's reagent. A time-dependent first order increase was preceded by a short lag phase, suggesting the release of N-formyl-D-penicillamine. When extrapolated to the time required for complete penicillin degradation, the free thiols represented 10% of the starting benzylpenicillin concentration. Thus, penicillin degradation in PBP-A appears to occur via the two pathways, hydrolysis of the C7-N4 bond being favored.

PBP-A-D160A, PBP-A-S61A, and PBP-A-⌬ Mutants-An
aspartic acid in position 160 is the only well conserved acidic residue in the short ⍀-loop of the PBP-A family. Since PBP-A showed weak penicillin hydrolytic activity and no Glu 166 is present in the ⍀-loop, we wanted to assess whether Asp 160 is involved in the protein activity. For this purpose, we expressed the mutant D160A and spectrophotometrically measured the activity. There was no loss of activity when Asp 160 was mutated, suggesting that it is not involved in the catalytic mechanism (Table 4).
We assessed in the same way the role of Ser 61 . Since PBP-A-S 61 A could not bind [ 14 C]penicillin G and no activity could be detected spectrophotometrically with benzylpenicillin as a substrate, we deduced that the mutant lost the penicillin binding property of PBP-A (Table 4). In all enzymes of the active site serine penicillin-recognizing enzyme (ASPRE) family (31), the serine of the SXXK motif is the catalytic residue forming an acyl-enzyme with the substrate. PBP-A-S61A lost the capacity of forming a penicilloyl-enzyme, confirming the alignment evidence that Ser 61 is the catalytic serine involved in the ester bond with the substrate.
Deletion of 10 residues in the short ⍀-loop of PBP-A affords a mutant (PBP-A-⌬) that is essentially inactive on thiolesters but still capable of forming and slowly hydrolyzing the penicilloyl-enzyme. In class A ␤-lactamases, the ⍀-loop is an essential structure for catalytic activity and protein specificity against ␤-lactam antibiotics. Banerjee et al. (44) studied the effects of the ⍀-loop deletion in S. aureus PC1 ␤-lactamase. This ␤-lactamase mutant shows a decreased activity against nitrocefin and is still able to bind a cephalosporin type substrate but completely loses the ability to bind penicillins. The deletion of amino acids 163-178 had a severe impact on the protein structure, such as the exposure of internal residues to the solvent and the enlargement of the active site. By implying steric restrictions, the ⍀-loop is also involved in the formation of the oxyanion hole, an essential structural element formed in the active site of serine enzymes for catalysis. The loop deletion thus induced the loss of the oxyanion hole. For PBP-A, the acylation rate for benzylpenicillin is decreased by a factor of 4, but deacylation is increased by a factor 3, and activity toward nitrocefin is conserved. Thus, the ⍀-loop influences somewhat the protein reactivity with penicillin but does not seem to be directly involved in catalysis.

PBP-A-L158D and PBP-A-L158E
Mutants-Glu 166 plays a critical role in both acylation and deacylation of ␤-lactamase, as shown by studies of TEM-1 mutants severely impaired in enzymatic activity. It was first shown by Adachi and al. (10) that mutating Glu 166 into Gln, Asn, or Ala abolished penicillin hydrolysis, but these mutants were still acylated by the substrates. A kinetic study of TEM-1 E166N mutant by Guillaume et al. (45) showed that deacylation is decreased by 8 orders of magnitude for benzylpenicillin, whereas acylation is decreased by only 2 orders of magnitude.
Leu 158 , which is aligned with Glu 166 , was mutated in either Asp or Glu. PBP-A-L158D and PBP-A-L158E were still acylated by benzylpenicillin. Preliminary results at 55°C showed no significant difference of activity of PBP-A-L158D mutant toward penicillin G, whereas the k cat of PBP-A-L158E mutant was increased by 24 times at 55°C and 50 times at 20°C. Acylation and deacylation constants were then determined for PBP-A-L158E at 20°C (Table 4). The deacylation rate is increased by a factor of 90, whereas acylation was 10 times slower. Moreover, no thiols were detected during incubation of PBP-A-L158E with benzylpenicillin; nor was an increase in absorbance at 257 nm detected, indicating that only the ␤-lactamase-like hydrolysis of the C7-N4 bond is taking place.
The increase in penicillinase activity of PBP-A-L158E confirms the positioning of the ⍀-loop deletion in the alignment between PBP-As and ␤-lactamases (Fig. 1), since it shows that Leu 158 , which is aligned with Glu 166 in ␤-lactamases, is most probably occupying a very similar spatial position.
We wanted to assess whether the L158E mutation was concomitant with a change in substrate specificity. As had been done for the wild type protein, hydrolysis of different lactams at 55°C was followed spectrophotometrically ( Table 3). The profile was exactly the same, with a preference in favor of penicillins. Determination of k cat at various temperatures allowed us to calculate the activation energy from Arrhenius plots: 68.8 kJ⅐mol Ϫ1 was obtained for the wild type and 49 kJ⅐mol Ϫ1 for the L158E mutant. This decrease of the activation energy is consistent with the increase in the reaction catalysis.

DISCUSSION
Physiological Role of PBP-A-We have shown that PBP-A is a penicillin-binding protein that hydrolyzes ␤-lactam substrates more than 6 orders of magnitude more slowly than TEM-1 ␤-lactamase. Moreover, T. elongatus is sensitive to concentrations of ampicillin of 5 and 25 g/ml, 3 and no lactamase is clearly annotated in its genome. Hence, the annotation of PBP-A as a putative ␤-lactamase should certainly be revised as well as those for the whole family. These observations also rule out a function as penicillin sensor PBP as found in Bacillus licheniformis and S. aureus for the induction of class A ␤-lactamases (46,47).
In relation with its activity toward ␤-lactam compounds and its molecular weight, PBP-A is suggested to be a low M r PBP. However, low M r PBPs are usually membrane-anchored from their C-terminal part, and high M r PBPs are linked to the membrane by their N-terminal part (5,48). Although the subcellular location is not known, the strongest prediction is that the protein is anchored through its N-terminal end. Subsequent analysis on all the family of proteins gave the same results. Hence, we propose that we identified a new family of low M r PBPs. High M r PBPs catalyze transglycosylation and transpeptidation reactions whereas low M r PBPs do not synthesize peptidoglycan and are involved in its modification by removing the terminal D-Ala from pentapeptide side chains (carboxypeptidase activity) or by cleaving the peptide cross-links that hold the glycan chains together. Considering the molecular weight, its probable exportation in the periplasm and its activity toward ␤-lactams and thiolester analogues of D-Ala-D-Ala substrates, PBP-A can be considered as a low M r DD-peptidase. The absence or very weak activity on D-Ala-D-Ala-derived substrates only reflects that they may not be appropriate or that the protein may be active on peptidic substrates only under certain conditions. This is often the case for DD-peptidases; for example, the soluble forms of PBP5 and PBP6 from E. coli are also inactive on Ac 2 -L-Lys-D-Ala-D-Ala substrate (49,50). Moreover, it was shown that some PBPs feature strong substrate preferences, the most striking being PBP3 from Streptococcus pneumoniae that has a much longer ⍀-loop-like structure and recognizes only peptides with a diaminopimelic acid instead of a lysine (51). Finally, although cyanobacteria cell wall shows an overall Gram-negative structure, the peptidoglycan is thicker, the degree of cross-linking between the peptidoglycan chains is higher (more similar to the Gram-positives), teichoic acids characteristic of Gram-positive walls are absent, and carotenoids are found, unlike Gram-negative membranes (52). Thus, the cell wall structure seems to be different from that of Gram-positive and Gram-negative bacteria. Little is known about its composition, and with the latter observations, this can lead us to the supposition that PBP-A should act on a specific cyanobacterial substrate.
However, sequence similarity of PBP-A with low M r PBPs is nearly undetectable, and five PBPs homologous to PBPs 1, 2, 3, 4, and 5, respectively, have been identified in T. elongatus and may be sufficient for bacteria survival (53). For these reasons, PBP-A may be not directly involved in cell wall synthesis. Among the 40 sequenced cyanobacterial genomes, PBP-A is found in 34 and absent in six Prochlorococcus marinus strains. In T. elongatus, no mobility elements are found in the vicinity of the gene. Overall, this suggests that PBP-A is probably involved in a function of ancient origin. An operon composed of a methyltransferase and pbp-a is predicted at the Comprehensive Microbial Resource Web site and is followed by a gene encoding a protein homologous to high affinity ABC transporter. ABC transporters are a large family of proteins permitting import and export of peptides, polysaccharides, drugs, and many other molecules across the membrane and utilizing ATP as a source of energy (54). Since this gene is only 6 nucleotides away from pbp-a, this ABC transporter and PBP-A may work in synergy. It is known from the studies of Goodell (55) that bacteria are able to recycle their muropeptides. Moreover, in S. aureus, the pbp4 gene is also flanked by a gene encoding an ABC-transporter that could participate in muropeptide recycling (56) and has been shown to be involved in cell autolysis (57). Thus, PBP-A may have a similar function.
Place of PBP-A in the Evolution of PBPs and ␤-Lactamases-According to Massova and Mobasherry (16), evolution is not a linear process, but several PBPs would have evolved separately from a primordial PBP into different classes of ␤-lactamases in parallel and independent processes. In a first step, PBPs would have acquired the ability to support deacylation of the penicilloyl-enzyme. Then the acylation and deacylation processes would have been refined by substitutions, insertions, or deletions to approach a full catalytic activity. This two-step evolution is supported by the asymmetry in ␤-lactamases catalytic machinery; acylation and deacylation involve different catalytic residues, illustrating that they evolved in response to different selection pressures at different evolution time points. The last step in this evolution process is the loss of PBP original activity to gain efficiency in the new function by incorporating structural features that disfavor interaction with peptidoglycan. However, ␤-lactamase and DD-peptidase activities appear mutually exclusive in nature. If coexistence of both activities is impossible, then loss of the peptidoglycan interaction (e.g. by loss of the penultimate D-Ala binding pocket) may have to accompany or even precede the acquisition of significant ␤-lactam-hydrolyzing ability (59).
PBP5 from E. coli contains all the features of a PBP but also displays an ⍀-loop-like structure. Although primary sequence similarities are almost undetectable between PBP-5 and class A ␤-lactamases, it was suggested from the protein crystal structure that class A ␤-lactamases evolved directly from a carboxypeptidase PBP5-like by incorporating two amino acids in the ⍀-loop and removing the specific residues of carboxypeptidase activity (60). Based on the sequence alignment, PBP-A contains all of the specific features of class A ␤-lactamases except for 6 amino acids in the ⍀-loop and the essential Glu 166 . All of the proteins from this family are expressed in the early branched bacterial species of cyanobacteria, and they share significant sequence similarities with class A ␤-lactamases and none with known PBPs. Therefore, it is the known PBP family that is phylogenetically the closest to class A ␤-lactamases. This is further supported by the 90-fold increase in the deacylation rate obtained by introducing a glutamate at the position aligned with Glu 166 . This effect is similar to what was observed by Chesnel et al. (61) for the F450D mutant of PBP2x, although the mutation was designed based on the tertiary structure.
In summary, we identified a new family of PBPs with significant sequence similarity to class A ␤-lactamases and undetectable homology to known DD-peptidases. Compared with class A ␤-lactamases, proteins from this family have a 6-residue shorter ⍀-loop with no equivalent of the essential glutamate. The ␤-lactamase activity of PBP-A from T. elongatus is more than 6 orders of magnitude slower than TEM-1, but a single mutation introducing a glutamate in the predicted equivalent position in the shorter ⍀-loop affords a 90-fold improvement of the deacylation rate. Further attempts to increase the penicillinase activity by substitution mutagenesis or ⍀-loop reinstitution are under progress. Compared with DD-peptidases, acylation and deacylation with penicillins occur with similar rates, and degradation of the penicilloyl enzyme occurs via the same two-pathway mechanism that has been observed for many of these enzymes. Although PBP-A is unable to hydrolyze classical D-alanyl-D-alanine-derived substrates, its stereoselective activity on thiolester-mimicking substrates indicates that PBP-A is most probably a D-Ala-D-Ala-carboxypeptidase. Direct studies on T. elongatus cyanobacterium may help to shed some light on its physiological role.