pentapeptide PG intermediate terminating in

Research Article 463 Enzymes of vancomycin resistance: the structure of D-alanine Dlactate ligase of naturally resistant Leuconostoc mesenteroides Alexandre P Kuzin 1, Tao Sun 1, Jodi Jorczak-Baillass 1, Vicki L Healy 2, Christopher T Walsh 2 and James R Knox 1 * Background: The bacterial cell wall and the enzymes that synthesize it are targets of glycopeptide antibiotics (vancomycins and teicoplanins) and β-lactams (penicillins and cephalosporins). Biosynthesis of cell wall peptidoglycan requires a crosslinking of peptidyl moieties on adjacent glycan strands. The D-alanine D-alanine transpeptidase, which catalyzes this crosslinking, is the target of β-lactam antibiotics. Glycopeptides, in contrast, do not inhibit an enzyme, but bind directly to D-alanine D-alanine and prevent subsequent crosslinking by the transpeptidase. Clinical resistance to vancomycin in enterococcal pathogens has been traced to altered ligases producing D-alanine D-lactate rather than D-alanine D-alanine. Results: The structure of a D-alanine D-lactate ligase has been determined by multiple anomalous dispersion (MAD) phasing to 2.4 Å resolution. Co-crystallization of the Leuconostoc mesenteroides LmDdl2 ligase with ATP and a di-d-methylphosphinate produced ADP and a phosphinophosphate analog of the reaction intermediate of cell wall peptidoglycan biosynthesis. Comparison of this D-alanine D-lactate ligase with the known structure of DdlB D-alanine D-alanine ligase, a wild-type enzyme that does not provide vancomycin resistance, reveals alterations in the size and hydrophobicity of the site for D-lactate binding (subsite 2). A decrease was noted in the ability of the ligase to hydrogen bond a substrate molecule entering subsite 2. Addresses: 1 Department of Molecular and Cell Biology, The University of Connecticut, Storrs, CT 06269-3125, USA and 2 Department of Biological Chemistry and Molecular Pharmacology, Harvard Medical School, Boston, MA 02115, USA. *Corresponding author. E-mail: knox@uconnvm.uconn.edu Key words: antibiotic resistance, ATP-binding, drug design, enzyme mechanism, MAD, X-ray crystallography Received: 11 January 2000 Revisions requested: 16 February 2000 Revisions received: 1 March 2000 Accepted: 6 March 2000 Published: 18 April 2000 2000, 8:463 470 0969-2126/00/$ see front matter 2000 Elsevier Science Ltd. All rights reserved. Conclusions: Structural differences at subsite 2 of the D-alanine D-lactate ligase help explain a substrate specificity shift (D-alanine to D-lactate) leading to remodeled cell wall peptidoglycan and vancomycin resistance in Gram-positive pathogens. Introduction Bacterial cell wall biosynthesis has been an attractive target for antibacterial drugs, including the vancomycin family of glycopeptide antibiotics and both the penicillin and cephalosporin families of β-lactam antibiotics. Both of these classes of antibiotics interdict the processing of peptidoglycan (PG) intermediates bearing D-alanine D-alanine termini. The covalent crosslinking of adjacent peptide strands in the PG layer provides mechanical strength to the cell wall. The transpeptidase activity of bifunctional transglycosylase/transpeptidase enzymes makes the crosslinks with displacement of the terminal D-alanine. β-lactams inhibit the activity of the transpeptidases [1,2], whereas vancomycins bind to the D-alanine D-alanine PG termini to block transpeptidase action [3]. Both drugs lead to decreased crosslinking as the molecular basis of their antibacterial activity. The D-alanine D-alanine moiety is preassembled by the enzyme D-alanine D-alanine ligase (Ddl) and then added by the murf ligase to a tripeptide cosubstrate to yield a pentapeptide PG intermediate terminating in D-alanine D-alanine [4]. The Ddl group of enzymes are of special interest for two reasons. First, they are themselves targets for antibiotics, although only D-cycloserine has been developed for human use. Second, variants of the Ddls are the agents of peptidoglycan remodeling in the three major clinical phenotypes of vancomycin-resistant enterococci (VRE). In the VanA (resistant to vancomycin and teicoplanin) and VanB (resistant to vancomycin but susceptible to teicoplanin) phenotypes the PG termini are D-alanine D-lactate, which has an ester instead of an amide linkage, whereas in VanC the replacement is to a D-alanine D-serine [5]. The VanC-type D-alanine D-serine terminus has a tenfold reduced affinity for vancomycin, the VanA-type and VanB-type depsipeptide terminus has a much higher 1000-fold reduction of affinity. In a real sense, therefore, vancomycin resistance in VRE is caused by a switch of D-alanine D-alanine to D-alanine D-X ligases (where X is any amino acid) and the ability of the VanA and VanB ligases to make depsipeptide is a true

464 2000, Vol 8 No 5 Figure 1 ATP ADP P 3 2 D-Ala + H 3 N N H Reactions catalyzed by ATP-requiring D-alanine D-X ligases. Common to both is the phosphorylated D-alanyl intermediate. The binding of vancomycin to the depsipeptide is reduced 1000-fold relative to the dipeptide. + H 3 N D-Ala Ligase + H 3 N D-Lac + H 3 N gain of function, which involves the selective activation of the 2-H of D-lactate as a nucleophile towards D-alanine P3 2 (Figure 1). When Ddls are compared with VanA, B and C by homology searches, five subfamilies can be distinguished [6,7]. There are two Ddl subfamilies (DdlA and DdlB), a VanCtype D-alanine D-serine ligase subfamily and the two clusters of D-alanine D-lactate ligases. ne D-alanine D-lactate subfamily is represented by the VanA and VanB ligases, as well as the depsipeptide ligases of the vancomycin producers [8]. The second D-alanine D-lactate ligase subfamily is found in the lactic acid bacteria, soil organisms with natural resistance to vancomycin producers. This group includes Leuconostoc mesenteroides [9], from which we have previously purified the LmDdl2 ligase [10] and proved its depsipeptide activity. Deciphering the molecular basis of alanine alanine versus alanine lactate production by the ligase subfamilies will require three-dimensional structures of complexes with substrates or inhibitors. The first picture of a D-alanine D-alanine ligase emerged with the X-ray structure for the Escherichia coli DdlB in complex with ADP and a tetrahedral analog of the reaction intermediate [11]. Among the findings from that work were the identification of an omega loop closed over the intermediates and a hydrogen-bonding network involving Tyr216 on that loop. The Tyr216 Phe mutant was observed to have a small but detectable gain of D-alanine D-lactate depsipeptide Table 1 D-Alanine D-lactate production by ligases. Ligase Depsipeptide/dipeptide Increase E. coli DdlB < 0.001 1 E. coli DdlB Y216F 0.04 40 L. mesenteroides LmDdl2 3 3000 S. toyoceansis DdlM 4* 4000 Enterococcal VanA 30 30,000 *From [40]. The structure of this ligase is not known. ligase activity, approximately 40-fold greater in catalytic efficiency than the wild-type DdlB, perhaps reflecting an initial step along the path to an efficient depsipeptide ligase [12]. Relative to DdlB, both of the alanine lactate ligase subfamilies have substantially higher ratios of alanine lactate/alanine alanine activity (Table 1). The LmDdl2 ligase shows a ratio of 3000:1, whereas the VanA ligase is tenfold better with a ratio of 30,000:1. In an effort to understand these variations in function, we report here the structure of a ligase from the second depsipeptide ligase subfamily, the LmDdl2 ligase from the vancomycin-resistant organism L. mesenteroides. Now available are crystal structures of three D-alanine D-X ligases, and a fourth is under investigation [13]. Each ligase varies greatly in its production of depsipeptide (Table 1), and thereby provides very different levels of resistance to antibiotics of the glycopeptide family, ranging from no resistance in wild-type organisms with the DdlB-type ligase, through mid-level resistance in organisms such as L. mesenteroides with the LmDdl2 ligase, to high-level resistance in enterococcal pathogens using the VanA ligase. In the crystal structures [11,14], all ligases are complexed with similar dipeptidyl analogs of known properties and kinetics [10,15,16], so that comparison of binding sites might explain the subtle differences in ligand specificities. Here we focus on a comparison of the LmDdl2 and DdlB ligases, with a brief inclusion of the recently established VanA structure (D Roper and G Dodson, unpublished observations). Results and discussion Crystal structure of the LmDdl2 ligase LmDdl2 ligase is dimeric in the crystal (Figure 2a) as well as in solution [10], with a noncrystallographic diad near residues Ile85, Leu114, Leu127, Val132 and Leu142, most of which are on α helices forming a hydrophobic interface. LmDdl2 comprises three α+β domains (Figure 2b), a folding motif first seen in glutathione synthetase [17,18]. The amide bond-forming site lies generally at the intersection of all three domains. The ATP-binding site is between two antiparallel β sheets of the central and C-terminal

Research Article D-Alanine D-lactate ligase Kuzin et al. 465 Figure 2 Folding of the LmDdl2 ligase. (a) The LmDdl2 dimer viewed perpendicular to the noncrystallographic diad axis (vertical). The C-terminal helix (center) of each monomer extends to the other monomer. ADP and phosphorylated analog 2 are observed only in monomer 1 (shown in red). Its omega loop is seen on the right-hand side. (b) verlay of LmDdl2 (red) and DdlB (yellow) based on minimization of Cα atoms in 16 core residues listed in the caption to Figure 3. The N terminus is at the right of the figure. This figure was created using RIBBNS [41]. domains and has the so-called palmate [17] or grasp [19] motif now seen in other ATP-dependent enzymes [20,21]. Comparison of ligase foldings Global similarity of the LmDdl2 and DdlB D-alanine D-X ligases is apparent in Figure 2b. LmDdl2 (and VanA) contain a large 36-residue insertion in the N-terminal domain that is not present in the smaller E. coli DdlB enzyme (Figure 3). This insertion is unlikely to have an influence on ligand specificity, however, as it is located on the surface about 20 Å distant from the binding site for the second ligand (subsite 2). The LmDdl2 ligase, the largest of the three ligases, has a five-residue insertion in the omega loop at position 247. LmDdl2 also has a much longer C-terminal peptide that extends in the direction of the other monomer in the dimer (Figure 2a). About half of this 20-residue C-terminal extension is seen in the electron-density map, and the invisible half is possibly near the omega loop of the other monomer. C-terminal disorder might be required for the movement of the omega loop, which must open and close for ligand entry and exit for each cycle of catalysis [22]. Consistent with this idea is the finding that the more disordered C terminus of monomer 1 lies near the disordered omega loop of ligand-less monomer 2. This disorder might be the cause of the high mosaicity that is observed in crystals. With the LmDdl2 and DdlB sequences aligned according to the crystallographic alignment in Figure 3, their similarity is 44%. For the calculation of the root mean square deviation (rmsd) of the two structures, 16 common residues surrounding the ATP and ligand-binding sites were chosen as a core set. The Cα atoms of the core residues overlay rather closely (0.73 Å). When adding the Figure 3 Amino acid sequences of the D-alanine Dalanine ligase from the ddlb gene of E. coli [42] (top) and the D-alanine D-lactate ligase from the lmddl2 gene of L. mesenteroides [43]. Boxes contain segments clearly matched by overlay of crystallographic structures. Important spatially equivalent core residues in DdlB/LmDdl2 are: Glu15/Glu16, His63/His102, Lys97/Lys136, Lys144/Lys180, Ser150/Ser186, Ser151/Ser187, Glu180/Glu216, Lys181C/Glu217C, Glu187/Glu224, Arg255/Arg301, Asp257/Asp303, Glu270/Glu316, Asn272/Asn318, Gly276/Gly322, Ser281/Ser327 and Leu282/Leu328. The α and β secondary structures are indicated by boxes and arrows, respectively. DdlB MTDKIAVLLGGTSAEREVSLNSGAAVLAGLREGG IDAY 38 PVDPKEV 45 LmDdl2 MTKKRVALIFGGNSSEHDVSKRSAQNFYNAIEATGKYEII 40 VFAIAQNGFFLDTESSKKILALEDEQPIVDAFMKTVDASD 80 DdlB DVTQLKSM GFQKVFIALHGRGGEDGTLQGMLELMGL 81 PYTGSGVMASALSMDKLRSKLLWQGAGLPVAPWVALTRAE 121 LmDdl2 PLARIHALKSAGDFDIFFPVVHGNLGEDGTLQGLFKLLDK 120 PYVGAPLRGHAVSFDKALTKELLTVNGIRNTKYIVVDPES 160 DdlB FEKGLSDKQLAEIS ALGLPVIVKPSREGSSVGMSKVVAE 160 NALQDALRLAFQHDEEVLIEKWLSG PEFTVAILG EE 196 LmDdl2 ANNW SWDKIVAELGNIVFVKAANQGSSVGISRVTNA 196 EEYTEALSDSFQYDYKVLIEEAVNGARELEVGVIGNDQPL 236 DdlB ILPSIRIQPSG TFYDYEAKYLS DETQYFCPAGLE 230 ASQEANLQALVLKAWTTLGCKGWGRIDVMLDSDGQFYLLE 270 LmDdl2 VSEIGAHTVPNQGSGDGWYDYNNKFVDNSAVHFQIPAQLS 276 PEVTKEVKQMALDAYKVLNLRGEARMDFLLDENNVPYLGE 316 DdlB ANTSPGMTSHSLVPMAARQAGMSFSQLVVRILELAD 306 306 LmDdl2 PNTLPGFTNMSLFKRLWDYSDINNAKLVDMLIDYGFEDFA 356 QNKKLSYSFVSLGEEKIGKFN 377

466 2000, Vol 8 No 5 Figure 4 P 2 P ATP ADP 3 P + H 3 N Ligase + H 3 N 1 2 Co-crystallization of phosphinate 1 with the LmDdl2 ligase and ATP produced ADP and the transition-state analogue 2. Cα atoms of all aligned residues (as defined in Figure 3) to the core Cα atoms the rmsd increases to 1.4 Å. Transition-state analog Co-crystallization of the LmDdl2 ligase with ATP and 1(S)-aminoethyl (2-carboxy-2(R)-methyl-1-ethyl) phosphinic acid 1 resulted in transfer of the γ-phosphate group and production of a phosphorylated phosphinate 2 (Figures 4,5). Because dipeptide or depsipeptide formation is initiated by the attack of the first D-alanine (D-alanine 1 ) on the γ-phosphate of ATP to yield the acylphosphate, the phosphorylated phosphinate is an excellent analog of the presumed transition-state for the ligation of the activated D-alanine with the second substrate, either D-alanine or D-lactate. The phosphinophosphate analog 2 is known [23,24] to inhibit wild-type Ddls with a very long half-life (in the order of days). Its inhibitory effect on the VanA depsipeptide ligase is much less striking [16]. For the LmDdl2 ligase, the inhibition level is unmeasured but thought to be poor, and capture of 2 in a crystal structure was not assured. In fact, we find that only one monomer of the LmDdl2 dimer contains the analog, and there is disorder around the empty binding site of the second monomer. In the ordered monomer, the omega loop and a smaller serine serine loop (185 188) close over the cavity holding ADP and 2. A hydrogen bond from Ser187 in the serine loop to the backbone C of Tyr255 in the omega loop is maintained in all ligases (Table 2 and Figure 6). Because a second hydrogen bond linking the two loops in DdlB is absent in this LmDdl2 enzyme, where Phe261 replaces Tyr216, one would expect that these loops might be more floppy than in the DdlB enzyme. But this LmDdl2 structure and the structure of the Tyr216 Phe mutant of DdlB [14] show that there is insignificant increase in loop movement and indicate that the H group is not required for loop closure. Binding sites for ADP and D-alanyl phosphate The geometry for ATP and D-alanine 1 substrates in LmDdl2 is similar to that in DdlB [11]. The adenine group of the nucleotide is in a hydrophobic area with its amine group hydrogen bonded to a sidechain of a conserved glutamic acid (Glu216) and a backbone carbonyl group (Glu217). The hydroxyl groups of the ribose hydrogen bond with Glu224, and conserved lysines Lys136 and Lys180 interact with the α- and β-phosphate groups of ADP. The ammonium group of a lysine on the omega loop (Lys260) joins two magnesium cations in bridging the β- and γ-phosphate groups. In VanA, however, this omega-loop lysine residue is absent and Figure 5 The 2F o F c electron-density map of ADP, phosphorylated phosphinate 2 and bridging magnesium ions in monomer 1 at a contour level of 0.9σ. This figure was created using CHAIN [36]. N N

Research Article D-Alanine D-lactate ligase Kuzin et al. 467 Table 2 Distances in binding sites*. Distance LmDdl2 DdlB Y216F DdlB A1 2.9 3.0 2.8 A2 2.8 2.8 2.6 B1 3.1 3.0 3.2 B2 3.0 3.0 3.1 C1 2.6 2.6 2.3 C2 4.0 3.1 2.8 C3 2.8 2.5 2.4 C4 3.3 D1 2.7 3.0 2.8 D2 3.1 2.9 3.4 E 2.9 2.8 2.5 Figure 6 Glu16 A1 NH + 3 W Ser186 CH A2 3 Ser187 H B1 HN Gly322 P P ADP H B2 H 2 N Arg301 E D2 D1 + CH 3 C3 W CH 3 C4 C2 C1 H Tyr255 S H N Lys260 Ser327 Leu328 Phe261 Met326 His243 Glu226 C *See Figure 6. replaced by two water molecules (D G Dodson, unpublished observations). Roper and The α-ammonium group of the phosphorylated D-alanine 1 is hydrogen bonded to the carboxylate group of the strictly conserved Glu16 (Table 2 and Figure 6). Polarization of the terminal P bond in the intermediate is accomplished by hydrogen bonds from the sidechain amide of conserved Arg301 and the backbone amide of Gly322 in an oxyanion hole. Placement of the D-methyl group near Val19 and the face of His102 helps define substrate chirality. Specificity for the second ligand The D-alanine D-X ligases must have the architecture necessary to discriminate between zwitterionic D-alanine 2 [7,25] and anionic D-lactate competing for subsite 2. Can we explain why, in the LmDdl2 ligase, the binding affinity of D-lactate is tenfold higher than that of D-alanine 2 [10], and that D-lactate readily ligates with the activated D-alanine 1 even though D-lactate is a weaker nucleophile than D-alanine 2? We first address the greater affinity of subsite 2 for D-lactate over D-alanine. External to the site is a positive sidechain, Arg22 (Lys22 in VanA), which would be better than the neutral Asn21 in DdlB in attracting the anionic lactate. Internal groups surrounding subsite 2 include the conserved residues Val19, Thr324 and Leu328. But two changes in LmDdl2 will alter the hydrophobicity and size of subsite 2 relative to DdlB ligase. Firstly, a buried Met326 is found in van der Waals contact with the second ligand, replacing a solvent-exposed His280 in DdlB (Figure 7). The -CH2 S CH3 portion of Met326 will fill 55 60 Å 3 [26] of the binding cavity and could restrict the binding of the zwitterionic amino acid [25] in favor of the slightly smaller lactate anion. (The NH 3+ and H groups are estimated to differ in volume by 10 12 Å 3 ). And secondly, Tyr216 in DdlB is replaced by the more hydrophobic Phe261 in LmDdl2. Together, these alterations in cavity size and Schematic showing the distances listed in Table 2. Residues Tyr255, Lys260 and Phe261 are on the omega loop. Gly322NH and Arg301 form an oxyanion hole. hydrophobicity might contribute to the tenfold greater binding for lactate over alanine 2 in LmDdl2. pposing this trend, however, is the fact that the hydrophobic sidechain of Leu328, although it is common to LmDdl2 and DdlB, has rotated approximately 3 Å away from the C-terminal carboxylate, having been replaced by a water molecule. In searching for enzymic influences on the nucleophilicity of the second ligand, we see that the local binding interactions of the C terminus of the analog indeed differ from those observed in DdlB (Table 2, Figures 6,7). Subsite 2 lies at the N terminus of the 327 335 helix, the dipole of which might assist alignment of the ligand [11]. Unlike the situation in the DdlB complex, however, in LmDdl2 the secondary structure of this helix is somewhat irregular. The carboxylic acid group of 2 is not as strongly hydrogen bonded to the hydroxyl group of conserved Ser327, and, significantly, no carboxylate hydrogen bond exists to the backbone amide of Leu328. Structurally, the longer distances in LmDdl2 result from an expansion of the ligation site along the dipeptide axis. Thus, distances from Glu16 or Ser186 (residues near the N-terminal alanine 1 ) to Ser327 or Leu328 (residues near the C-terminal ligand) are longer in LmDdl2 by 0.5 1.4 Å. Given the prepositioning of the N-terminal D-alanine 1, the C-terminal carboxylic acid group of the second ligand falls short of one of the two potential hydrogen-bonding groups on the enzyme. Instead, a water molecule bridges between a carboxylate oxygen atom and the backbone amide of Leu328. For an incoming lactate molecule, therefore, weaker hydrogen bonding by its carboxylate group would increase the nucleophilicity of the α-hydroxyl group, and hence facilitate its attack on the activated carbon of alanine 1 in the first site. Catalytic residues? A plausible catalytic base was never identified in the DdlB ligase [11,15]. Based on the crystal structure, the

468 2000, Vol 8 No 5 Figure 7 Glu16 Ser186 NH3 Ser187 Glu16 Ser186 NH3 Ser187 verlay of phosphorylated phosphinate 2 in the binding site in LmDdl2 (red) and DdlB (orange). LmDdl2 residues are numbered. Distances are given in Table 2. This figure was drawn using MLSCRIPT [44]. Phe261 Phe261 Arg301 Arg301 Glu226 Glu226 His243 His243 Leu328 Met326 Ser327 Leu328 Met326 Ser327 Tyr216 Phe mutant of the DdlB enzyme was made and found to exhibit a small amount of new depsipeptide activity [12] (Table 1). The markedly more active LmDdl2 ligase in fact contains phenylalanine (Phe261) at the corresponding spatial position (Figure 7), but other changes exist as well, the importance of which can be evaluated by site-directed mutagenesis. For example, a buried glutamic acid Glu226 in LmDdl2 has replaced a threonine in DdlB. Glu226 reaches toward the analog 2, forming a weak water-bridged hydrogen bond with the acyl phosphate group and another with the C-terminal carboxylate group of 2. Within 2.7 Å of Glu226 is a buried histidine residue, His243. Although the catalytic involvement of this histidine is unknown, it is interesting that the histidine glutamate pair is only found in the ligase with moderate depsipeptide activity. The inactive DdlB ligase, its weakly acitve Tyr216 Phe mutant, and the highly active VanA ligase all have unreactive residues at these two positions. The recent crystal structure of VanA (D Roper and G Dodson, unpublished observations) shows that a much better-positioned histidine His244 replaces Phe216 of DdlB and Phe261 of LmDdl2, an alignment unforeseen in earlier sequence matches [7,11]. Biological implications Bacteria are rapidly changing in response to heavy antibiotic use, either by means of microbial selection, mutations, or transfer of genetic material from drug-resistant strains to susceptible strains [27]. The steady spread of resistance to the penicillins is well documented [28]. Vancomycin, the backup antibiotic for Gram-positive infections that do not respond to penicillin therapy, is increasingly used in the treatment of methicillin-resistant Staphylococcus aureus. With the discovery of vancomycin resistance in Enterococcus faecium in 1988, there is growing concern that the plasmid-borne resistance will be transferred to Staphylococcus aureus [29,30]. The wild-type D-alanine D-alanine ligases have long been good antibiotic targets [31], in that they are present in all bacteria, are necessary for growth, and are absent from humans. The newer D-alanine D-lactate ligases such as VanA and VanB, found in highly vancomycin-resistant gram-positive pathogens, might have arisen from the chromosomal D-alanine D-lactate ligases present in producing organisms such as Streptomyces toyocaensis or in the intrinsically resistant lactic acid organisms such as Leuconostoc mesenteroides. A comparison of three-dimensional structures of D-amino acid ligases might help establish ancestry and reveal enzymic changes leading to modification of cell wall peptidoglycan and glycopeptide resistance. Materials and methods Strains and chemicals E. coli strains BL2(DE3) and B834(DE3) were obtained from Novagen (Madison, WI). Culture media was obtained from Difco Laboratories (Detroit, MI). Amino acids, nucleotides, dithiothreitol (DTT), ampicillin, and buffers were from Sigma Chemical Co. (St. Louis, M). Isopropyl-thio-β-D-galactoside (IPTG) was purchased from Bachem Biosciences (Torrance, CA) and polyethyleneglycols (PEGs) were from Hampton Research (LaGuna Niguel, CA). Phosphinate 1 was a generous gift from PA Bartlett and BA Ellsworth (University of California, Berkeley). verexpression of wild-type and selenomethionine LmDdl2 LmDdl2 ligase was overexpressed as described previously [10]. For overexpression of the selenomethionine-substituted LmDdl2 ligase, plmddl2 was transformed into the E. coli met auxotroph expression strain B834(DE3). The corresponding strain was grown overnight in 21 ml of medium, which was 5% (v:v) Luria broth and 95% M9 medium supplemented with 1 mg l 1 thiamine, 1 g l 1 glucose, 489 mg l 1 MgS 4.7H 2, 14.7 mg l 1 CaCl 2, 7.5 mg l 1 FeS 4, 100 µgl 1 ampicillin, 50 mg l 1 each of 18 amino acids (excluding methionine and cysteine), and 50 mg l 1 selenomethionine [32]. The cells were spun down, resuspended in M9 media (supplemented as above), and used to inoculate 2 l of supplemented M9 media. The culture media was incubated at 37 C until A 595 = 0.8. IPTG at a final concentration of 0.5 mm was added to induce lmddl2 gene expression, and the culture was further incubated for 5 h at 30 C.

Research Article D-Alanine D-lactate ligase Kuzin et al. 469 Protein purification Purification of wild-type LmDdl2 ligase was performed as previously described [10]. Selenomethionine LmDdl2 was purified similarly except that degassed 50 mm HEPES at ph 7.2, 5 mm MgCl 2, 1 mm EDTA, 200 mm KCl, and 1 mm DTT was used for cell lysis and subsequent purification steps, yielding 60 mg of purified protein. Mass spectrometry analysis Wild-type and selenomethionine LmDdl2 were analyzed by electrospray mass spectrometry at the Mass Spectrometry Facility of Harvard University (Cambridge, MA). For wild-type LmDdl2, which contains six methionine residues in a total of 377, the measured mass is 41,686 Da. The measured mass for selenomethionine LmDdl2 is 41,947 Da. The 261 mass difference is between that expected for five and six selenium substitutions per molecule. Crystallization Both wild-type and selenomethionine-substituted LmDdl2 protein could be crystallized under similar conditions. Crystals were grown at room temperature by the vapor equilibrium method. The drop contained protein at 3.5 mg ml 1 in 15% PEG 8000 (or mpeg 5000) for the selenoprotein), 0.1 M ammonium sulfate, 50 mm NaCl (0 mm for selenoprotein), 2.4 mm DTT for selenoprotein, 1 mm MgCl 2, 3 mm ATP (AMP PNP for the selenoprotein), 3 mm phosphinate, in 50 mm MES buffer at ph 6.5. The reservoir contained 30% PEG or mpeg and 0.2 M ammonium sulfate in 0.1 M MES buffer. Diamond-shaped plates appeared after one to two weeks. Separation of glycol and salt phases occurred in most drops after one month. Two monoclinic forms grew under these conditions. Form 1, the subject of this paper, had unit-cell dimensions 118.2 90.8 82.1 Å, β = 92.1 (100 K) in space group C2 with two monomers in the asymmetric unit (V M = 2.6 Å 3 Da 1 ). Form 2, also with two monomers in the asymmetric unit, had cell dimensions 167.2 64.7 112.3 Å, β = 112.3 in C2 (V M =3.5Å 3 Da 1 ). The two monoclinic forms often grew in the same drop, with form 1 appearing two to three weeks after form 2. Both forms diffracted to approximately the same resolution, 2.2 2.4 Å. X-ray data from the native LmDdl2 crystal A form 1 crystal was immersed in the 30% PEG holding solution containing 25% MPD for 4 min prior to being loop-mounted and flash cooled to 100 K in a liquid nitrogen gas stream. Single-wavelength (0.908 Å) data collection was performed at MacCHESS station A1 at the Cornell High Energy Synchrotron Source, Ithaca, NY. A CCD area detector (Area Detector Systems Corp.) was used in 1k 1k mode at a distance of 80 mm from the crystal. A total of 204 of data, made from 1 phi oscillations of 20 sec each, was processed using the HKL programs [33]. Calculated mosaicity was high (1.9 ) and overall completeness was only 82% to 2.38 Å resolution. The R merge (on intensities) was 7.2% (Table 3). X-ray data from the selenomethionine LmDdl2 crystal A form 1 seleno-crystal was immersed in the PEG holding solution containing 22% MPD for 45 sec and flash cooled to 100 K in a liquid nitrogen gas stream. MAD data were collected by the reverse-beam method at three wavelengths (0.9500 Å remote, 0.9786 Å peak and 0.9789 Å edge) at station 12c at NSLS at Brookhaven National Laboratory. A CCD area detector (Brandeis B4) 344 mm from the crystal recorded 2k 2k images for a 1 phi oscillation. Count times were 50 sec per image. Data were collected at the edge wavelength over a 36 wedge of phi, then recollected at the peak and remote wavelengths until 107 of data were recorded. This process was repeated at phi plus 180. Intensities were indexed and scaled by HKL [33]. Poor crystal mosaicity (0.8 ) reduced the completeness of the three data sets to 55 63% at 2.4 Å resolution. R merge values were 4.8%, 5.8% and 5.2% at the three wavelengths (Table 3). solution A search for a molecular replacement solution using DdlB structures (2DLN, 1IV or 1IW) as models was not successful. In spite of poor data completeness for the selenomethionine crystal, the structure was Table 3 X-ray data for native and selenomethionine crystals*. Data set Native Se remote Se peak Se edge λ(å) 0.9080 0.9500 0.9786 0.9789 f 2.6 8.3 8.2 f 3.7 6.0 3.0 bservations 94,338 49,290 47,414 48,959 (4374) (189) (435) (646) Unique refls 30,564 22,828 22,997 24,156 (1738) (163) (394) (589) Completion 0.82 0.55 0.60 0.63 (0.47) (0.024) (0.062) (0.093) Av I/σ(I) 11.0 14.3 12.4 13.1 (3.8) (2.3) (2.1) (2.0) R merge 0.072 0.048 0.058 0.052 (0.21) (0.44) (0.20) (0.28) R anom 0.053 0.064 0.058 R disp 0.035 vs. 0.031 vs. 0.046 vs. peak edge remote *Data for high-resolution shells in parentheses: native 2.48 2.38 Å; selenomethionine 2.44 2.30 Å. R merge = Σ Iav Ii /Σ Ii, where Iav is the average of all individual observations, Ii. R anom = Σ F + F /Σ Fav for anomalous differences, where Fav is the average of Friedel amplitudes at a single wavelength. R disp = Σ Fλi Fλj /Σ Favλ for dispersive differences, where Favλ is the average amplitude at two wavelengths. determined by the MAD method using version 1.15 of the program SLVE [34]. The Patterson analysis found 10 of the 12 possible selenium atoms (11 expected from mass spectrometry). The average selenium occupancy and B factor were 0.95 and 43 Å 2, respectively. The successful heavy-atom solution had a Z-score of 76.3. The resulting electron-density map at 2.5 Å resolution had a standard deviation in the local rms density equal 0.34, and the correlation of rms density in neighboring regions was 0.40. The unmodified MAD-phased map was much better than an earlier three-derivative multiple isomorphous replacement (MIR) map (figure of merit = 0.58 at 3 Å resolution). The MAD map showed 95% of the backbone chain for monomer 1 and density for its ligands, and about 90% of the chain of monomer 2. Missing in monomer 2 was density for the omega loop (245 266) near the catalytic site, possibly because of the absence of ligands in the binding site of this monomer (see below). Density-modified maps based on solvent flattening or twofold symmetry averaging were also examined, but did not improve the density of the omega loop in monomer 2. The sequence [35] of the 377-residue protein was manually fitted using the programs CHAIN [36] and FRD [37]. The 2 molecules formed a noncrystallographic dimer (Figure 2a) and were fitted independently. refinement Simulated annealing refinement was initiated at 1,500 K with X-PLR version 3.8.5.1 [38]. Data from 8.0 2.5 Å with F obs greater than 2σ(F) were used, with 4% of the data excluded for calculating R free [39]. Individual B factors were initially set to 15 Å 2 for protein atoms. Water molecules were excluded from this early refinement, as were ligands and several residues from the N and C termini of each molecule. Monomer 2 also lacked 22 residues in the omega loop. Examination of 2F o F c and F o F c maps of the dimer allowed the fitting of ADP and phosphinophosphate 2 in the binding site of monomer 1, but not in monomer 2. Various annealing and energy minimization protocols failed to bring up density for the ligands or omega loop of monomer 2. Final cycles of refinement used all 2.38 Å data from the native crystal, which had better completeness in the higher resolution shells. For a model that includes 360 residues in monomer 1 (3 362), 347 residues in monomer 2 (numbered 402 644 and 667 770), one ADP and one phosphinophosphate 2 ligand in monomer 1, and 213 water molecules

470 2000, Vol 8 No 5 (B factors < 60 Å 2 ), the resulting crystallographic R factor is 0.184 for all F > 0 data from 100 2.38 Å (27,948). The R free value is 0.257. Deviations of the protein model from ideality are 0.008 Å and 1.4 for covalent bonds and angles, respectively. The electron density of ADP and 2 is seen in Figure 5. Accession numbers Atomic coordinates have been deposited in the Protein Data Bank (with the accession code 1EHI) at Rutgers University. Acknowledgements We thank the staff at station x12c at the National Synchrotron Light Source, Brookhaven National Laboratory, for help in MAD data collection, and the staff at the MacCHESS facility at the Cornell High Energy Synchrotron Source for help in collecting native data. PC Moews, MA McDonough and GV Crichlow assisted with data collection and processing. We are indebted to DI Roper and GG Dodson (York University) for kindly providing an unpublished manuscript. VLH is a Howard Hughes Medical Insititue Fellow. This work was supported by the US National Institutes of Health with grants to CTW and JRK. References 1. Knox, J.R., Moews, P.C. & Frere, J.-M. (1996). Molecular evolution of bacterial β-lactam resistance. Chem. Biol. 3, 937-947. 2. Massova, I. & Mobashery, S. (1998). Kinship and diversification of bacterial penicillin-binding proteins and β-lactamases. Antimicrob. Agents Chemother. 42, 1-17. 3. Groves, P., Searle, M.S., Mackay, J.P. & Williams, D.H. (1994). The structure of an asymmetric dimer relevant to the mode of action of the glycopeptide antibiotics. 2, 747-754. 4. Walsh, C.T. (1989). Enzymes in the D-alanine branch of bacterial cell wall peptidoglycan assembly. J. Biol. Chem. 264, 2393-2396. 5. Arthur, M. & Courvalin, P. (1993). Genetics and mechanisms of glycopeptide resistance in enterococci. Antimicrob. Agents Chemother. 37, 1563-1571. 6. Healy, V.L., Park, I.-S. & Walsh, C.T. (1998). Active-site mutants of the VanC2 D-alanyl D-serine ligase, characteristic of one vancomycinresistant bacterial phenotype, revert towards wild-type D-alanyl Dalanine ligases. Chem. Biol. 5, 197-207. 7. Lessard, I.A.D., Healy, V.L., Park, I.-S. & Walsh, C.T. (1999). The omega loop of the D-ala D-lactate ligase VanA from vancomycinresistant enterococci: determinants for differential effects on D-ala Dlactate vs D-ala D-ala formation. Biochemistry 38, 14006-14022. 8. Marshall, C.G., Broadhead, G., Leskiw, B.K. & Wright, G.D. (1997). D-Ala D-ala ligases from glycopeptide antibiotic-producing organisms are highly homologous to the enterococcal vancomycin-resistance ligases VanA and VanB. Proc. Natl Acad. Sci. USA 94, 6480-6483. 9. Handwerger, S., Pucci, M.J., Volk, K.J., Liu, J. & Lee, M.S. (1994). Vancomycin-resistant Leuconostoc mesenteroides and Lactobacillus casei synthesize cytoplasmic peptidoglycan precursors that terminate in lactate. J. Bacteriol. 176, 260-264. 10. Park, I.-S. & Walsh, C.T. (1997). D-Alanyl D-lactate and D-alanyl Dalanine synthesis by D-alanyl D-alanine ligase from vancomycinresistant Leuconostoc mesenteroides. J. Biol. Chem. 272, 9210-9214. 11. Fan, C., Moews, P.C., Walsh, C.T. & Knox, J.R. (1994). Vancomycin resistance: structure of D-alanine D-alanine ligase at 2.3 Å resolution. Science 266, 439-443. 12. Park, I.-S., Lin, C.-H. & Walsh, C.T. (1996). Gain of D-alanyl D-lactate or D-lactyl D-alanine synthetase activities in three active-site mutants of the Escherichia coli D-alanyl D-alanine ligase B. Biochemistry 35, 10464-10471. 13. Huyton, T. & Roper, D.I. (1999). Crystallization and preliminary X-ray characterization of VanA from Enterococcus faecium BM4147: towards the molecular basis of bacterial resistance to the glycopeptide antibiotic vancomycin. Acta Crystallogr. D 55, 1481-1483. 14. Fan, C., Park, I.-S., Walsh, C.T. & Knox, J.R. (1997). D-alanine Dalanine ligase: phosphonate and phosphinate intermediates with wildtype and the Y216F mutant. Biochemistry 36, 2531-2538. 15. Shi, Y. & Walsh, C.T. (1995). Active-site mapping of E. coli D-ala D-ala ligase by structure-based mutagenesis. Biochemistry 34, 2768-2776. 16. Ellsworth, B.A., Tom, N.J. & Bartlett, P.A. (1996). Synthesis and evaluation of inhibitors of bacterial D-alanyl D-alanine ligases. Chem. Biol. 3, 37-44. 17. Yamaguchi, H., et al., & Katsube, Y. (1993). Three-dimensional structure of the glutathione synthetase from E. coli B at 2.0 Å resolution. J. Mol. Biol. 229, 1083-1100. 18. Fan, C., Moews, P.C., Shi, Y., Walsh, C.T. & Knox, J.R. (1995). A common fold for peptide synthetases cleaving ATP to ADP: glutathione synthetase and D-alanine:D-alanine ligase of E. coli. Proc. Natl Acad. Sci. USA 92, 1172-1176. 19. Murzin, A.G. (1996). Structural classification of proteins: new superfamilies. Curr. pin. Struct. Biol. 6, 386-394. 20. Galperin, M.Y. & Koonin, E.V. (1997). A diverse superfamily of enzymes with ATP-dependent carboxylate-amine/thiol ligase activity. Protein Sci. 6, 2639-2643. 21. Denessiouk, K.A., Lehtonen, J.V. & Johnson, M.S. (1998). Enzyme mononucleotide interactions: three different folds share common structural elements for ATP recognition. Protein Sci. 7, 1768-1771. 22. Wright, G.D. & Walsh, C.T. (1993). Identification of a common proteasesensitive region in D-alanyl D-alanine and D-alanyl D-lactate ligases and photoaffinity labeling with 8-azido ATP. Protein Sci. 2, 1765-1769. 23. Duncan, K. & Walsh, C.T. (1988). ATP-dependent inactivation and slow binding inhibition of Salmonella typhimurium D-alanine D-alanine ligase (ADP) by (aminoalkyl)phosphinate and aminophosphonate analogues of D-alanine. Biochemistry 27, 3709-3714. 24. McDermott, A.E., Creuzet, F., Griffin, R.G., Zawadzke, L.E., Ye, Q.-Z. & Walsh, C.T. (1990). Rotational resonance determination of the structure of an enzyme-inhibitor complex: phosphorylation of an (aminoalkyl)phosphinate inhibitor of DD-ligase by ATP. Biochemistry 29, 5767-5775. 25. Carlson, H.A., Briggs, J.M. & McCammon, J.A. (1999). Calculation of the pka values for the ligands and side chains of E. coli D-alanine Dalanine ligase. J. Med. Chem. 42, 109-117. 26. Creighton, T.E. (1993). Proteins: s and Properties. WH Freeman and Co., New York, NY. 27. Gold, H.S. & Moellering, R.C. (1996). Antimicrobial drug resistance. New Engl. J. Med. 335, 1445-1453. 28. Frere, J.-M. (1995). β-lactamases and bacterial resistance to antibiotics. Mol. Microbiol. 16, 385-395. 29. Hoeffler, D. & Zimmermann, U. (1997). Methicillin-resistant Staphylococcus aureus and vancomycin-resistant enterococci. Lancet 350, 737-739. 30. Iwakawa, K., Tsusaka, N., Kizu, J. & Yamamoto, K. (1999). Selection of Methicillin-resistant Staphylococcus aureus following increased use of vancomycin despite restricted use of third-generation cephem. Jpn. J. Infect. Dis. 52, 221-222. 31. Neuhaus, F.C. & Hammes, W.P. (1981). Inhibition of cell wall biosynthesis by analogues of alanine. Pharma. Therapy 14, 265-319. 32. Leahy, D.J., Erickson, H.P., Aukhil, I., Joshi, P. & Hendrickson, W.A. (1994). Crystallization of a fragment of human fibronectin: introduction of methionine by site-directed mutagenesis to allow phasing via seleomethionine. Proteins: Struct. Funct. Genet. 19, 48-54. 33. twinowski, Z. & Minor, W. (1997). Processing of X-ray diffraction data collected in oscillation mode. Methods Enzymol. 276, 307-326. 34. Terwillinger, T.C. & Berendzen, J. (1999). Automated structure solution for MIR and MAD. Acta Crystallogr. D 55, 849-861. 35. Evers, S., Casadewall, B., Charles, M., Dutka-Malen, S., Galimand, M. & Courvalin, P. (1996). Evolution of structure and substrate specificity in D-alanine:D-alanine ligases and related enzymes. J. Mol. Evol. 42, 706-712. 36. Sack, J.S. (1988). CHAIN: a crystallographic modeling program. J. Mol. Graph. 6, 224-225. 37. Jones, T.A. (1985). Interactive computer graphics: FRD. Methods Enzymol. 115, 157-170. 38. Brünger, A.T. (1992). X-PLR: A System for X-ray Crystallography and NMR. Version 3.8.5.1. Yale University Press, New Haven, CT. 39. Brünger, A.T. (1992). Free R value: a novel statistical quantity for assessing the accuracy of crystal structures. Nature 355, 472-475. 40. Marshall, C.G. & Wright, G.D. (1997). The glycopeptide antibiotic producer Streptomyces toyocaensis NRRL 15009 has both D-alanyl- D-alanine and D-alanyl-D-lactate ligases. FEMS Microbiol. Lett. 157, 295-299. 41. Carson, M. (1991). Ribbons 2.0. J. Appl. Crystallogr. 24, 958-961. 42. Robinson, A.C., Kenan, D.L., Sweeney, J. & Donachie, W. (1986). Further evidence for overlapping transcriptional units in an E. coli cell envelope-cell division gene cluster: DNA sequence and transcriptional organization of the ddl ftsq region. J. Bacteriol. 167, 809-817. 43. Elisha, B.G. & Courvalin, P. (1995). Analysis of genes encoding D-alanine:D-alanine ligase-related enzymes in Leuconostoc mesenteroides and Lactobacillus spp. Gene 152, 79-83. 44. Kraulis, P. (1991). MLSCRIPT: a program to produce both detailed and schematic plots of protein structures. J. Appl. Crystallogr. 24, 946-950.