Burkholderia pseudomallei d‐alanine‐d‐alanine ligase; detailed characterisation and assessment of a potential antibiotic drug target

Burkholderia pseudomallei is a serious, difficult to treat Gram‐negative pathogen and an increase in the occurrence of drug‐resistant strains has been detected. We have directed efforts to identify and to evaluate potential drug targets relevant to treatment of infection by B. pseudomallei. We have selected and characterised the essential enzyme d‐alanine‐d‐alanine ligase (BpDdl), required for the ATP‐assisted biosynthesis of a peptidoglycan precursor. A recombinant supply of protein supported high‐resolution crystallographic and biophysical studies with ligands (AMP and AMP+d‐Ala‐d‐Ala), and comparisons with orthologues enzymes suggest a ligand‐induced conformational change occurring that might be relevant to the catalytic cycle. The detailed biochemical characterisation of the enzyme, development and optimisation of ligand binding assays supported the search for novel inhibitors by screening of selected compound libraries. In a similar manner to that observed previously in other studies, we note a paucity of hits that are worth follow‐up and then in combination with a computational analysis of the active site, we conclude that this ligase represents a difficult target for drug discovery. Nevertheless, our reagents, protocols and data can underpin future efforts exploiting more diverse chemical libraries and structure‐based approaches.


Introduction
Burkholderia pseudomallei, a serious pathogen of humans and animals, causes melioidosis. This Gram-negative bacterium is widely distributed in the environment and is classified as a Category B biowarfare agent by the U.S. Centers for Disease Control and Prevention (CDC, www.cdc.gov). Treatment of melioidosis involves intravenous high-intensity application of cephalosporins for up to 2 weeks then an eradication phase to prevent recurrence that can involve the use of antifolate combinations for up to 20 weeks. This is an expensive treatment often complicated by side effects and a further issue is that strains resistant to the current drugs have been identified [1,2]. New antibacterial strategies are being considered [3] and to this end, we sought to investigate the potential of key enzymes as new therapeutic targets and to generate the data, reagents and protocols that support an assessment of these targets for early stage drug discovery.
The peptidoglycan (PG) layer contributes an important protective shield for Gram-negative organisms and antibiotics, such as b-lactams and glycopeptides, indeed the key drug for melioidosis, ceftazidime, target one or more enzymes required for its biosynthesis [4][5][6][7]. Due to the success of targeting parts of the PG biosynthetic pathway [7], we focused on the assessment of D-alanine-D-alanine (D-Ala-D-Ala) ligase as a potential target for the development of new drugs for B. pseudomallei infections.
Structures of DdlB from several bacteria have been determined in the apo-form and ligand-bound complexes [14,15]. Ddl presents the ATP-grasp family fold [16], and the catalysed reaction follows a sequential ordered mechanism [17,18]. Known inhibitors of Ddl include D-cycloserine (DCS) [17][18][19], phosphinates [15,20,21], diazenocarboxamides [22], hydroxyethylamine [23] and boron derivatives of D-Ala [24]. However, there has been little progress in developing highly potent inhibitors of Ddl with efficient engagement of the cellular target. This might be due to the polar nature of these inhibitors that would require active transport for uptake.
Our objective was to characterise and assess BpDdl as a potential antibacterial drug target and as part of that assessment to initiate a search for new inhibitor scaffolds. We report an efficient recombinant protein production system, purification and crystallisation protocols, two high-resolution ligand-bound BpDdl structures (AMP and AMP+D-Ala-D-Ala) together with comparisons with orthologues, the optimisation of two enzymatic assays for the functional characterisation of the enzyme and development of high throughput (HTP) assays that were applied to selected compound libraries. We used biolayer interferometry (BLI) and enzymatic assays to assess the outcomes of the HTP library screens and a computational assessment of the BpDdl active site properties.

Results and Discussion
Overall subunit structure Two high-resolution crystal structures have been determined, BpDdl:AMP and BpDdl:AMP:D-Ala-D-Ala, with statistics in Table 1. Ddl is a member of the ATP-grasp superfamily presents a characteristic and highly conserved fold [16,[25][26][27][28]. ATP-grasp proteins contain three well-defined domains [16]. In the case of Ddl, these are the N-terminal, central and C-terminal domains (Fig. 1). BpDdl secondary structure comprises 10 a-helixes, one 3 10 -helix, one p-helix and 12 bstrands (Fig. 1A). The N-terminal segment presents an a/b-domain whilst the central and C-terminal domains comprise the ATP-grasp fold; two a/b-domains that generate a narrow pocket for the ATP binding. The D-Ala binding part of the active site is located in the Cterminal domain. Two loops are implicated in the binding of the co-factor and the substrate. The P-loop (residues 149-159) is located in the central domain, between b-strands 5 and 6, and interacts with the c-b phosphates of ATP. The x-loop is located in the C-terminal domain and presents a p-helix configuration involving residues 198-222 ( Fig. 1A). At several positions, gross structural differences are noted when comparing subunits of the two BpDdl structures (Fig. 1B). The position of the central domains and several helices on the N-terminal and C-terminal domains, that line the active site, adopt different positions. The Ca comparison illustrates the notable conservation of these structural changes in both BpDdl models (Fig. 1C).

Quaternary structure
In solution, at low concentrations, BpDdl behaves as a monomer as indicated by SEC. Native-PAGE on a more concentrated sample indicated monomer as well as the presence of a higher molecular mass species, possibly a dimer (Fig. 2). The crystal structures present a dimer in the asymmetric unit with each subunit aligned perpendicularly to the other (Fig. 1A). The BpDdl dimer is formed by about 12-13% of the accessible surface area (ASA) of a subunit, which involves residues from a3, a4, a5 and the C-terminal region of a8. Comparisons with structures from different organisms (Table 2) indicate that the dimer arrangement is shared (data not shown), and the ASA percentage involved in the dimerisation, also called buried surface area (BSA), is similar. These BSA values are lower than expected for a surface with a high probability of forming a multimeric assembly [29,30]. In contrast, the P-value for the probability of having a similar observed solvation free energy gain (D i G) in other surface areas is lower than 0.5, indicating a higher level of hydrophobicity than expected and it is likely this property that drives oligomerisation [29]. Mycobacterium tuberculosis Ddl (MtDdl) [27] has a reduced contact area, with no interactions involving residues on a3 and a8. Also, the P-value is higher for MtDdl than the other structures. Ddl structures can be grouped according to the additional number of amino acids, compared to BpDdl, at the C-terminal end of a10. When extra residues are present, a10 is elongated towards the partner subunit and an increased number of inter-subunit contacts assist dimer formation. This is noted for Bacillus anthracis (BaDdl), Streptococcus mutants (SmDdl) [31], Staphylococcus aureus (SaDdl) [32], Coxiella burnetii (CbDdl) [27] and Xanthomonas oryzae (XoDdl) [12], where between 8 and 20 additional residues are observed. Table 2 also presents the sequence identity shared with proteins of known structure, which ranges from about 30% to 90%. Despite a wide range of overall sequence identity compared to BpDdl we note that residues contributing to  dimer formation are well conserved, with Gly77, Gly81, Val95, Ala99 and Asp103 present in 12-13 structures of the 14 analysed and three of these residues are also located near the active site binding pockets (Gly77, Val95 and Asp103). Despite differences elsewhere in the subunit-subunit interface, the dimer arrangement is a recurrent feature in Ddl.
Co-factor and substrate-binding pockets Co-factor and substrate-binding sites are identified in Figs 1 and 3, the latter also showing electron density associated with the ligands. Whilst the complete AMP is well-ordered the N, Ca, Cb atoms of D-Ala1 are not. The residues that contribute to the formation of these binding sites are highly conserved in Ddl sequences. Indeed, the ATP-binding site residues (Lys104, Gly153, Glu184, Asp262, Glu275 and Asn277) are also conserved within different functional proteins from the ATP-grasp superfamily such as acetyl-CoA carboxylase [16]. The recognition of ATP involves residues from the central and C-terminal domains. These include Lys104, Lys148 and Glu275 binding to the phosphate, and Glu184, Lys185, Ile187, Glu192, Phe214 and Tyr215 that are involved in adenosine recognition. In the case of D-Ala-D-Ala, the key interacting residues are Tyr215, Tyr221, Asp277, Arg260, Gly281, Ser286, Leu287 and also a Mg 2+bound water molecule.
In the substrate-binding pocket, Lys220, Arg260, Asp277 and Gly281 are strictly conserved and Leu287 in 90% of sequences. These are key residues for substrate binding and catalysis according to the proposed reaction mechanism [17,18]. Conservation is also noted for Phe214, Tyr215 and Tyr221 (40-50% identity) with other aromatic residues. These residues appear to be involved in placing ATP and D-Ala in the correct position for catalysis. In the ATP-binding pocket, Lys104, Lys148, Glu184, Glu192 and Glu275 are conserved in between 98% and 100% of the sequences highlighting their importance to co-factor binding.
When BpDdl:AMP:D-Ala-D-Ala was compared with other product and substrate-bound structures (from EcDdl: PDB codes: 1IOV, 4C5B, 4C5C; from TtDdl: 2ZDH and 2ZDQ, including ADP+phosphinate, ADP+D-Ala, ATP+D-Ala-D-Ala, ADP+D-Ala, ATP+D-Ala respectively), differences in the substrate and product binding were found (Fig. 4). The C a overlay of BpDdl:AMP:D-Ala-D-Ala (subunit A) with these ligand-bound structures showed the location of D-Ala-D-Ala, though similar, participates in different interactions compared to previously characterised proteins. Additionally, the relative position of the Mg 2+ is different. In other Ddl structures (e.g. EcDdl, PDB code 4C5C), two Mg 2+ ions interact with Asp262, Glu275 and c-b phosphates from ATP. However in BpDdl: AMP:D-Ala-D-Ala only one Mg 2+ is observed, interacting with His70. The overall BpDdl:AMP:D-Ala-D-Ala conformation presents a more open form than in other structures, suggesting it is an intermediate conformation, but it cannot be distinguished if this corresponds to a substrate binding or product release phase.

Conformational changes
Ligand-induced conformational changes can be observed when comparing the AMP loaded structure with an apo-form extracted from the PDB (Fig. 5). Binding of AMP generates a more compressed conformation where the central domain moves towards the centre of the structure and the x-loop rotates and generates the active site pocket.
ATP binding therefore is likely to stabilise the xloop conformation enforcing interactions with Tyr215 to create a closed conformation. Phe214 is able to interact with a hydrophobic area of the central   is disordered and not included in the model but in the AMP-bound structures this loop is ordered with an average B-factor of 36.7 A 2 (the overall average B-factor value is 23.9 A 2 ). This higher B-value suggests a greater degree of flexibility in the loop.
There are apo and ligand-bound structures for SaDdl (2I8C, 2I80, apo: 2I87) [32] and TtDdl (2YZM, 2YZN, 2ZDH, 2ZDG, 2ZDQ, apo; 2YZG). In both cases, the conformational changes noted when comparing different states do not appear as pronounced as in BpDdl. Movement of the TtDdl central domains involves an a7 translation of 4.7 A, whereas no differences in SaDdl are observed. This tends to suggest that the conformational changes are distinct in different organisms. As indicated in ESC library screening in Results and Discussion, purified BpDdl contained AMP and we were unable to obtain an apo-BpDdl crystal form. With this in mind, electron density maps of the apo-BpDdl, SaDdl and TtDdl structures were inspected to check for missed co-factors but in all three there was no evidence of anything bound in the ATP pockets.
A comparison of the electrostatic surfaces in and around the active sites of apo-BpDdl and BpDdl:AMP reveals significant differences (Fig. 6). The D-Ala pocket is more electropositive in the AMP-bound structure. Therefore, at pH 7.5, D-Ala (pI = 6.01) would preferably bind to the BpDdl:AMP conformation over the apo-form. This suggests that the D-Ala pocket is formed after the x-loop conformational change. The residues involved in determining the electrostatic properties of the different conformations (Glu22, Asp262, Glu275 and Asn277 in apo-BpDdl, and His70, Arg260 and Asn277 in BpDdl:AMP) are highly conserved. Also, as described previously, His70, Arg260 and Asn277 are key amino acids for Ddl catalysis. This is consistent with the hypothesis that the generation of a favourable environment for catalysis is dependent on ATP binding, in agreement with the sequential ordered mechanism previously characterised [17,18] where ATP is the first component to bind Ddl.

Enzyme kinetics
Enzyme activity was investigated using a coupled spectrophotometric assay and a BIOMOL Green assay at    (Table 3). Using the coupled assay, inhibition of the enzyme activity was observed at high concentrations of ATP and K i,ATP was also determined. In the BIOMOL Green assay, high ATP concentrations decrease the signal-to-noise ratio due to nonenzymatic hydrolysis and consequently, ATP concentrations did not exceed 1 mM. Nevertheless the K i, ATP could not be accurately determined with the BIO-MOL assay. The values for K m2 were consistent between the coupled and the BIOMOL assays. Comparing the K m values with other Ddl enzymes, we note more similarity between BpDdl and MtDdl or EcDdl than Yersinia pestis Ddl or HpDdl (Table 1). Indeed, HpDdl seems to be the least related Ddl, not only displaying a lower sequence identity but also different kinetic parameters. We showed that our BpDdl samples had AMPbound, so the K ATP determination is compromised. No information about the presence of AMP or similar compounds in HpDdl, EcDdl and MtDdl has been reported. Purification of MtDdl [17] and HpDdl [26] was carried out using nickel-affinity columns whereas EcDdl used ion exchange [20,33]. It is possible that MtDdl and HpDdl also co-purified with the co-factor as similar protocols to those we used for BpDdl were employed; however direct confirmation would have to be sought.
The pH optimum for BpDdl was determined. Differences in the activity due to the change in the pH are observed in the K m2 and the V max determinations. At a lower pH there is an increase of 1.2 times in the K m2 and 1.7 times decrease in the V max . To further explore this, activity was measured over a pH range of 5.1-9.5. Regarding this, a test at different pH values has been carried out (Fig. 7). The peak of the maximum activity is observed at pH 8 (100% relative activity) and the activity decreases around 1.7 times at pH 7.5 (56% relative activity). This is consistent with the results from the K m experiments.
Determination of the K i and the mechanism of inhibition for DCS were assessed (Fig. 8). DCS appears to be a competitive inhibitor of substrate with a K i,DCS value of 100 lM. At the highest DCS concentration tested, 1 mM, a different V max was observed what may   [19] but not at a high concentration (1000 lM). This DCS concentration dependent behaviour appears to also apply to BpDdl.

Quantification of AMP
Characterisation of structures derived from crystals grown in the presence of different ligands, or none, revealed all had AMP-bound. This, we presume is derived from the expression host and we decided to quantify the level of AMP in the purified sample. Denatured and folded samples were compared. Firstly, controls BaKynB, BaKynB in the presence of 2 lM AMP and BpDdl itself were denaturated using (a) pH extremes, (b) an elevated temperature and (c) chymotrypsin digestion. The experimental and expected results are shown in Fig. 9. The negative control (in the absence of protein and AMP) and positive control (10 lM AMP in the absence of protein) results were similar to expected values. BaKynB likewise, confirmed there is no AMP-bound [34]. The BaKynB data indicate that the different sample treatments are not interfering with the AMP-glo assay. The results for BpDdl demonstrate the sample has been co-purified with AMP. The presence of AMP without pretreatment of BpDdl indicates there is some free material in the protein solution (around 0.4 lM) probably due to AMP exchange/release. The sample treated with chymotrypsin presents an AMP concentration of about 2 lM that indicates 1.6 lM (approximately 15 nmol) AMP remains bound to BpDdl. Taking into account that the concentration of BpDdl is 25 mgÁmL À1 (about Fig. 7. Effect of pH on enzyme activity. The relative activity is set to 100% at maximum level, at pH 8. 20 nmol), then the ratio AMP:BpDdl suggests almost full occupancy of BpDdl by the monophosphate. This is consistent with the structural data that show one, well-ordered AMP per BpDdl subunit. The presence of AMP in the purified enzyme, presumably acquired from the bacterial expression system has implications for our analyses. The K ATP determination and the compound screens therefore have been done mainly with the binary complex BpDdl:AMP instead of apo-BpDdl. Additionally, during the spectrophotometric assay, a retarded phase was observed at the onset of the reaction (Fig. 10). This could be due to diffusion of AMP that has to first occur prior to ATP loading. Also, the fragment screen using BLI was carried out with the closed BpDdl conformation and that might explain why a relatively low ATP-binding signal was measured.
An apo-BpDdl structure (PDB code 4EGQ) has been determined by the Seattle Structural Genomics Center for Infectious Disease. Electron and difference density maps in the ATP-binding sites were inspected to ascertain if a ligand might be present. There were no significant features so we conclude this is a bone fide apo-BpDdl form. It is unclear why in our structures AMP is present and this may be due to different purification or crystallisation protocols. The PDB entry stipulates that the apo-form crystal was grown at a basic pH (0.1 M CHES pH 9.5, 1 M Na + -K + tartrate and 0.2 M Li 2 SO 4 ) whereas the crystals obtained in our work were obtained at pH 5.5. There are no details of the apo-form purification. Whilst it is not uncommon for a purified protein to acquire ligands during recombinant protein purification as exemplified by human PPAR-b/d binding endogenous fatty acids of E. coli [35], it is intriguing in this case. We speculate that this might be indicative of a mechanism whereby the levels of AMP:ATP could be networked to the control exercised by the second messenger cyclic di-AMP to regulate peptidoglycan biosynthesis [36]. The role of AMP and derivatives is coming under increased scrutiny in this aspect [37].

Compound library screens
We screened three compound libraries. BLI was used in conjunction with a fragment library whilst the malachite green-based assay was applied to the ESC and small diversity set libraries.

BLI: Fragment library screen
BpDdl was interrogated with a fragment library of around 700 compounds, using a BLI binding assay. A total of 11 hits were selected for a secondary inhibition assay at 500 lM. A single compound (Fig. 11) gave enzyme inhibition with a K i value of 10 mM. Attempts to obtain a structure of the hit in complex with the enzyme by co-crystallisation failed. Fig. 9. Quantification of AMP. Neg: is the negative control (without enzyme, without AMP). Pos: is the positive control from the standard curve (without enzyme with 10 lM AMP). KynB: is the BaKynB sample. WT: means no pretreatment is performed. DT and DpH: indicates variations in the temperature or pH were carried out. Ch: indicates digestion with chymotrypsin was used and + is used when 2 lM AMP was added. The error bars correspond to the SD derived from three experiments.

ESC library screening
The optimised HTP assay had a Z 0 score of 0.8 and a signal-to-noise ratio of 5 after 80 min of reaction. Initially, the experiments were run in the absence of TCEP and Tween-20. To target the substrate and cofactor binding sites, a concentration of D-Ala and ATP close to the K m values was used. A library of 6775 compounds, comprising well-known drugs, and, for control purposes, known aggregators, was tested. A total of 74 hits were selected and re-tested using the BIOMOL Green assay. Results showed no interference by any of the compounds but that 61 appeared to be false positives leaving 13 compounds for further investigation. The dose-response experiment for the hits was performed in the presence of 1 mM TCEP and 0.01% (w/v) Tween-20 and no inhibition was observed. Thus the redox environment may affect the stability of the compounds and/or induce aggregation. Indeed, inspection of the chemical structures suggested they could also be aggregators [38] and this was further supported by the dose-response profile from one of the compounds (SAM001247071, Fig. 12), which is characteristic of this issue [39]. Although our assay appears to be reliable no useful hits were obtained.

DDU small diversity set screening
Due to the high similarity of the ATP-binding pockets between different functional proteins (glutathione synthetase [40], ATP-citrate lyase [41], succinyl-CoA ligase [42] and kinases [43]), targeting this pocket might result in inhibition of human enzymes with toxic side effects. To avoid this, the inhibitor search was focused on obtaining new inhibitors targeting the D-Ala binding pocket. The DDU diversity set screen was selected to inform on this part of the active site. A new optimisation of the HTP assay was carried out to establish screening conditions to target this pocket. The reaction was linear for 20 min, the signal-to-noise ratio 6 and the Z 0 score 0.8-0.9. From the 15 667 compounds tested, four hits were selected on the basis of a 30% inhibitory effect. However, the follow-up dose-response experiment identified them as false positives.

Recombinant protein production
An oligonucleotide encoding BpDdl (Unitprot code: Q63QJ9, protein mass 33 kDa) was purchased from GeneWiz (Takeley, Essex, UK) and inserted into a modified pET15b vector (Novagen, Madison, WI, USA), that adds an N-terminal hexahistidine tag followed by a tobacco etch virus (TEV) protease cleavage site before the resulting target protein. The restriction enzymes used were NdeI (at the 3 0 DNA end) and BamHI (5 0 end). The integrity of the construct was confirmed by sequencing. Recombinant protein production was achieved with BL21 (DE3) cells, 1 mM IPTG (final concentration) for induction and incubating overnight (O/N) at 20°C. Purification involved use of metal ion affinity chromatography with a Ni 2+ -NTA column (GE Healthcare, Chicago, IL, USA) using buffer A (20 mM Tris-HCl pH 7.4, 200 mM NaCl, 5 mM MgCl 2 ) for sample loading and buffer B (20 mM Tris-HCl pH 7.4, 200 mM NaCl, 5 mM MgCl 2 , 800 mM imidazole) for elution. The hexahistidine tag was cleaved, during dialysis back into buffer A, using 1 mg of TEV protease per 10 mg of BpDdl, incubated O/N at 4°C. A second round of affinity chromatography with a Ni 2+ -NTA column removed the tag and the TEV protease, itself carrying a hexahistidine tag, and fractions carrying the purified protein were pooled. Protein purity and molecular weight were assessed by SDS/PAGE. Protein concentration was determined using the theoretical molar extinction coefficient e(BpDdl) = 27 515 M À1 Ácm À1 as calculated in PROTPARAM [44]. Quaternary structure was investigated using native-PAGE (Novex, ThermoFisher Scientific, Waltham, MA, USA) and size-exclusion chromatography (SEC) using buffer A and the HiLoad 16/60 superdex 200 prep grade column (GE Healthcare).

Crystallisation
The protein sample was concentrated to 25 mgÁmL À1 in buffer A. Initially, thin needle shaped crystals grew in sitting drops plates at 18°C after incubating the protein (at different protein : reservoir ratios) with a reservoir containing 0.  combinations (e.g. ATP, DCS, D-Ala+ADP, D-Ala-D-Ala+AMP, DCS+AMP, DCS+AMPPNP). Numerous crystals were analysed but only two structures are reported, a binary complex with AMP and a ternary complex with D-Ala-D-Ala+AMP. In all other cases only AMP was observed in the active site and this will be discussed further below.

Diffraction, data collection and structural analyses
Optimised crystals, grown in the presence of ATP, were tested in-house using a M007HF X-ray generator with a Saturn 944HG+ CCD detector (Rigaku-Americas, Woodlands, TX, USA) and subsequently used to record a 2.0 A resolution data set on ESRF beamline BM30-A. The data were indexed and integrated using XDS [45], then scaled and analysed using AIMLESS [46]. The Matthews coefficient (V m of 2.27 A 3 ÁDa À1 with bulk solvent of about 50%) suggested two molecules in the asymmetric unit. Two subunits were positioned with PHASERMR [47] using only the protein components of the 2.2 A resolution apo-BpDdl structure (PDB code 4EGQ, subunit A). Rigid body refinement followed by cycles of restrained refinement using REFMAC [48], electron and difference density map inspections, and model manipulations were made with COOT [49]. AMP, not ATP, was observed in the active site and the end result was a 2.0 A resolution structure of this binary complex. Other crystals, of putative apo-enzyme and grown in the presence of different ligands provided 13 data sets on Diamond beamline I04-1. Data were processed using XDSgui and IMOLSFLM [50], and scaled and merged with AIMLESS. All crystals were isomorphous with the initial structure. These data produced two structures. In one case, that of a crystal obtained in the presence of ADP and D-Ala, data extending to 1.3 A resolution were recorded and molecular replacement was performed with the refined subunit A from the binary complex. This structure only showed AMP-bound in the active site and hence, due to a superior resolution, supersedes the first; this structure is BpDdl:AMP. A 1.5 A resolution structure of a ternary complex of BpDdl:AMP: D-Ala-D-Ala was the second structure. This was refined starting from the BpDdl:AMP model. Tight, noncrystallographic symmetry (NCS) restraints were employed at the onset of the refinements and these were released once solvents and ligands were included. Side-chain rotamers, water molecules and ligands, for example, ethylene glycol, were added when the protein models were completed and then anisotropic thermal parameters were included in the model. In the case of the ternary complex, a difference Fourier map suggested the presence of D-Ala-D-Ala. A model with full occupancy for the complete ligand resulted in negative difference density at the N-terminus of the dipeptide. However, if only the second D-Ala was refined, positive difference density was observed. We conclude that the first D-Ala (D-Ala1) is not fully ordered and although included in the model we assigned zero occupancy. A Mg 2+ ion is present near His70 (subunit A). Two strong difference density features, in BpDdl:AMP (near residues 293 from subunit A, 221 from subunit B) were interpreted as SO 2À 4 ions. Three such examples were noted in BpDdl:AMP: D-Ala-D-Ala (near residues 293 from subunit A, 221 and 290 from subunit B). The sulphates near residue 221 occupy the D-Ala binding pocket. Refinement was concluded when no improvements resulted for R free and R work values, when density maps indicated no further changes were justified and that satisfactory model geometry was obtained. For the other data sets, omit maps were inspected after some refinement and AMP was noted in all structures, irrespective of the ligands that had been added or when an apo-structure was sought. This indicated that AMP might be present in the protein solution, a point we had to investigate (see below).

Model analyses and informatics
MOLPROBITY [51] was used to assess model geometry. Secondary structure was inspected using DSSP [30], the dimer interface was analysed using PISA [29,52] and MOLSOFT ICM (www.molsoft.com), and B-factors analysed using Baverage from the CCP4I suite. Figures were prepared using ALINE [53], PYMOL (www.pymol.org) and MOLSOFT ICM. The DALI server [54] was used for comparative purposes whilst electrostatic surfaces and properties of the enzyme active site were determined using MOLSOFT ICM. A comparison of Ddl sequences was carried out. First, a search for relevant sequences in UniProt (www.uniprot.org) was conducted. The identified sequences were grouped by 50% identity and aligned using CLUSTAL OMEGA [55] with group representatives (based on 588 sequences).

Enzymatic assays
A coupled spectrophotometric assay was initially optimised for characterising the enzyme. To support the HTP screening of compound libraries, a malachite green-based assay was developed.
A coupled spectrophotometric assay K m determination The activity of BpDdl was measured using a spectrophotometric assay [17]. The assay buffer contained 100 mM Tris-HCl pH 8, 100 mM KCl, 10 mM MgCl 2 , 2.5 mM phosphoenolpyruvate (PEP), 0.4 mM NADH, 5 lLÁmL À1 phosphoglycerate kinase/lactate dehydrogenase (PK/LDH) (Merck & Co, Kenilworth, NJ, USA) and 30 lM BpDdl. Additionally, to determine K m for D-Ala and ATP, 500 lM ATP and variable concentrations of D-Ala (0.1, 0.3, 0.5, 0.7, 1, 1.5, 2, 2.5, 3, 5, 10, 20 and 100 mM) or 10 mM D-Ala and variable concentrations of ATP (0.1, 0.2, 0.5, 1.2 and 3 mM) were added respectively. Ten times (109) concentrated D-Ala/ATP stocks were prepared for each assay so the volume added was constant. The final reaction volume was 100 lL, and the reaction started by adding the enzyme. All components were incubated for 30 min at 30°C and then the assay was performed in triplicate. Negative controls included the master mix with BpDdl and without substrate, and the master mix without BpDdl and with D-Ala/ ATP. No decrease in absorbance was detected in the negative controls during the assay period. The positive control was the PK/LDH combination and instead of ATP, 0.4 mM ADP was added. Initial velocities were obtained by measuring the steady-state rate for 60 s and determination of the K m values [K m1 (D-Ala), K m2 (D-Ala), K m (ATP)] was carried out as previously described [17]. Inhibition of the enzyme activity at high ATP concentrations was observed, and the K i,ATP calculation was carried out with GRAPHPAD (www.graphpad.com).

BpDdl inhibition
Inhibition experiments were performed in the presence of different inhibitor concentrations to determine K m2 (D-Ala). In the case of DCS the concentrations used were 100, 200, 250, 300, 500 and 1000 lM. Stocks (109) for each inhibitor concentration were prepared. Each K i determination was conducted for about 12 h. To check for time-dependent loss of BpDdl activity, an assay without inhibitors was performed before and after the inhibition studies. K i calculations were performed using double reciprocal plots [17].
To check for inhibition of any of the coupled enzymes, a cross-validation assay was performed. The assay buffer in this case contained 100 mM Tris-HCl pH 8, 100 mM KCl, 10 mM MgCl 2 , 2.5 mM PEP, 0.4 mM NADH, 5 lLÁmL À1 PK/LDH mixture and 0.5 mM ADP. Concentrations of 0.5 and 10 mM DCS (from stock 1 M) and a positive control (without inhibitor) were tested. No inhibition was detected. A potential inhibitor was identified (see below) and the K i determination was carried out as for DCS using final concentrations of 200, 500, 2000 and 5000 lM. compound was not a factor and K ATP was determined using the Michaelis-Menten equation. Regarding the K m2 (D-Ala) and K m (ATP) results, the chosen HTP assay conditions were as follows: 100 mM Tris-HCl pH 8, 10 mM KCl, 10 mM MgCl 2 , 0.4 nM BpDdl, 200 lM ATP and 5 mM D-Ala.
To determine whether the assay was linear under the chosen assay conditions, a time course experiment was carried out. A total of 24 reactions were started at time points 2.5 min apart during 90 min, and eight replicates included. Additionally, the same experiment was done in the absence of enzyme. These results were used to determine the Z 0 score (Eqn 1) and the signal-to-noise ratio of the experiment.
Determination of DMSO tolerance and D-cycloserine IC 50 Using the HTP assay conditions, a dose-response experiment was performed using variable concentrations of DMSO (from 0% to 27%) or DCS (0-300 lM). A 2% inhibition of enzyme activity was noted at 6% DMSO.

Fragment library screen using BLI
A fragment library from the Drug Discovery Unit (DDU, University of Dundee) was screened using an OctetRED384 (ForteBio-Molecular Devices, Fremont, CA, USA). The protein was immobilised on hexahistidine tag high affinity sensors (ForteBio), and the experiment carried out in black plates (Greiner). We used TEV protease as a control and to identify promiscuous binders, and ATP as the control ligand. During the assay performance, we noticed a low binding signal (lower than 0.1 nm) in the presence of ATP. Selected hits were tested for enzyme inhibition using the coupled enzymatic assay at a fixed concentration of 500 lM, in the presence of 2 mM D-Ala. One compound with an inhibition effect greater than 20% was followed up for K i determination.
ESC library screen using the malachite green-based assay A small selection of compounds from the BioAscent compound cloud and the NCC library of drug compounds from the NIH were tested at a concentration of 10 lM. Each screening plate included positive (normal reaction without compounds added) and negative controls (without enzyme). Compounds were dispensed using a Labcyte Echo acoustic dispenser. The assay was carried out using the HTP conditions and allowed to proceed for 1 h before stopping the reaction by adding BIOMOL green. Data were analysed using ActivityBase XE (www.idbs.com). A compound was considered a hit when the normalised percentage effect was higher than the median plus three times the standard deviation. The final DMSO concentration during screening reached 0.075% so the assay has not been compromised.
Selected hits were re-tested using the HTP assay conditions in the presence and absence of enzyme with a final concentration of 20 lM to filter false positives. Compounds inhibiting more than 30% of BpDdl activity were selected for IC 50 determination. The dose-response experiment was carried out using the same conditions and in the presence or absence of 0.01% (w/v) Tween-20 and 1 mM TCEP. Data were analysed using ActivityBase XE and Microsoft EXCEL.
Small diversity set screen using the malachite green-based assay A small diversity set library was screened under HTP assay conditions modified to target the D-Ala pocket, at pH 7.5, and including 0.01% (w/v) Tween-20 and 1 mM TCEP in the assay buffer. Consequently, further assay optimisation was required. K m2 (D-Ala) and K m (ATP) at pH 7.5 were determined as described before using 1 nM BpDdl. The time course experiment was performed with 3, 2 and 1 mM D-Ala and 500 lM ATP. The compound set was used at a final concentration of 30 lM. The reaction was carried out with 1 mM D-Ala and 0.5 mM ATP, and incubated for 12 min then absorbance was measured using a PHERAstar (BMG Labtech, Ortenberg, Germany). Data were analysed using ActivityBase XE and Microsoft EXCEL and any compound providing 30% inhibition was considered a hit. Validation of hits was performed by dose-response experiments using different concentrations (1/10 dilutions starting at 100 lM) in duplicate.

AMP detection
The crystallographic analyses revealed the presence of AMPbound to the protein even when no nucleotide was added to the sample. Therefore, the AMP-Glo kit from Promega (Madison, WI, USA) was used to quantify the presence of AMP in the purified BpDdl. The reaction was performed in a 96-well white plate (Greiner), using 25 lL of sample, in triplicate. The signal was measured with a FLUOstar OPTIMA microplate reader (BMG Labtech). The control protein, selected because it does not bind AMP/ADP/ATP, was an amidase called BaKynB [34]. Samples were denaturated before AMP quantification using three different treatments. These included heating (20 min at 100°C prior to the assay performance), exposure to extreme pH (2 min incubation at pH 2, then pH 10, then 7) and digestion with chymotrypsin. The samples used were BpDdl, BaKynB and BaKynB in the presence of 2 lM AMP (BaKynB-AMP) as a control. Chymotrypsin was added to a final ratio chymotrypsin : sample of 1 : 20 and the mixtures incubated O/ N at 25°C. Then, the samples were filtered using an Amicon ULTRA 3K concentrator (Merck & Co). The protein concentration was 4 mgÁmL À1 for BaKynB and 25 mgÁmL À1 for BpDdl. Data are presented in Fig. 9.

Conclusions
We set out to assess, and potentially validate, BpDdl as a target for the development of the new antibacterial drugs for B. pseudomallei. Two high-resolution crystal structures of BpDdl in complex with AMP and AMP+ D-Ala-D-Ala were obtained. Comparative studies indicate aspects of Ddl that are well conserved allow us to describe conformational differences that result from ligand binding. The BpDdl structures provided templates for a computational druggability assessment, but also identified that the enzyme was retaining AMP during isolation. This was confirmed using an AMP detection assay. The co-purification with AMP suggests that we have assayed a binary complex not the apo-enzyme. This had not been reported previously and might also apply in other Ddl studies. The BpDdl activity was measured using two different enzyme assays and showed similar kinetic parameters to two other Ddl enzymes but different from a third. We noted a retarded phase during assay that was probably due to the presence of AMP which first has to vacate the co-factor binding site prior to catalysis. High-throughput binding and enzymatic assays were developed to conduct compound library screens. Despite screening more than 22 500 compounds, we did not identify any good starting points for further study. Whilst our study was underway, AstraZeneca reported a larger compound library screen for Streptococcus mutans Ddl but no lead compounds were found [56], which is consistent with our results.
Our overall assessment is that BpDdl represents a challenging target for drug discovery. However, our progress in enabling ligand-binding studies of the enzyme means that in future a structure-based rational approach [57] combining crystallography with computational docking and modelling calculations would be feasible. In similar fashion, further HTP screens can be prosecuted but the composition of the library would have to be carefully considered.