Comparative Sequence, Structure and Redox Analyses of Klebsiella pneumoniae DsbA Show That Anti-Virulence Target DsbA Enzymes Fall into Distinct Classes

Bacterial DsbA enzymes catalyze oxidative folding of virulence factors, and have been identified as targets for antivirulence drugs. However, DsbA enzymes characterized to date exhibit a wide spectrum of redox properties and divergent structural features compared to the prototypical DsbA enzyme of Escherichia coli DsbA (EcDsbA). Nonetheless, sequence analysis shows that DsbAs are more highly conserved than their known substrate virulence factors, highlighting the potential to inhibit virulence across a range of organisms by targeting DsbA. For example, Salmonella enterica typhimurium (SeDsbA, 86 % sequence identity to EcDsbA) shares almost identical structural, surface and redox properties. Using comparative sequence and structure analysis we predicted that five other bacterial DsbAs would share these properties. To confirm this, we characterized Klebsiella pneumoniae DsbA (KpDsbA, 81 % identity to EcDsbA). As expected, the redox properties, structure and surface features (from crystal and NMR data) of KpDsbA were almost identical to those of EcDsbA and SeDsbA. Moreover, KpDsbA and EcDsbA bind peptides derived from their respective DsbBs with almost equal affinity, supporting the notion that compounds designed to inhibit EcDsbA will also inhibit KpDsbA. Taken together, our data show that DsbAs fall into different classes; that DsbAs within a class may be predicted by sequence analysis of binding loops; that DsbAs within a class are able to complement one another in vivo and that compounds designed to inhibit EcDsbA are likely to inhibit DsbAs within the same class.


Introduction
Antibiotic resistance has increased dramatically over the last decade and the consequent lack of treatment options poses a major threat for public health [1]. One approach to develop new chemical classes of antibacterials is to target virulence factors that cause disease in antibiotic resistant organisms [2]. Most pathogenic Enterobacteriaceae encode an oxidative folding pathway essential for virulence factor production [2][3][4][5]. Typically, the oxidative folding machinery includes a soluble thioredoxin-fold protein, DsbA, and an integral membrane protein partner, DsbB [6][7][8]. The disulfide form of DsbA is highly oxidizing and donates its disulfide bond to unfolded substrate proteins [9], leaving DsbA in the inactive reduced form. The inner membrane protein DsbB, in concert with its cofactor ubiquinone, interacts with reduced DsbA to oxidize the active site cysteines and convert DsbA to its functionally competent disulfide form [10]. Inhibition of the interaction between DsbA and substrate proteins or between DsbA and its partner DsbB could constitute a means of blocking virulence factor formation and thereby of inhibiting virulence of bacterial pathogens. Supporting this notion, deletion of DsbA homologues in pathogenic organisms results in diminished virulence in infection models [2,11] and deletion of dsbA or dsbB in uropathogenic E. coli (UPEC) severely attenuated its ability to colonize the bladder [11,12].
The characteristic properties of EcDsbA include: an active site CPHC motif that forms a destabilizing disulfide (T m reduced EcDsbA 350 K; T m oxidized EcDsbA 342 K) [13]; the more Nterminal of the two cysteines is nucleophilic and highly acidic, pK a 3.3 (usual value for a cysteine is 8-9) [9]; and EcDsbA is highly oxidizing (redox potential -122 mV) [9]. The past 5 years has seen the characterization of DsbA enzymes from many other bacteria including DsbAs with varying degrees of sequence identity to EcDsbA such as Neisseria meningitidis DsbA1 (NmDsbA1, 23% identity), Pseudomonas aeruginosa DsbA (PaDsbA, 30%) and Vibrio cholerae DsbA (VcDsbA, or TcpG, 40%). These DsbAs share a similar structural fold with EcDsbA though their surface properties vary [14] and they exhibit a wide range of redox properties (Table 1). Importantly, the EcDsbA hydrophobic groove that interacts with its essential partner EcDsbB is considerably truncated in NmDsbA1, PaDsbA and VcDsbA [15][16][17]. This modification and other surface changes in these DsbAs indicate that they fall into a separate class, distinct from EcDsbA, and that inhibitors designed against EcDsbA may not inhibit members of this class of DsbA. Conversely, DsbAs closely related to EcDsbA should be susceptible to the same mode of chemical inhibition.
Here we tested how close the sequence relationship must be to produce similar redox properties and binding interactions. We investigated two well-characterised DsbAs sharing 86% sequence identity, from E. coli K-12 strain (EcDsbA) and S. enterica Typhimurium DsbA strain SL1344 (SeDsbA), by applying comparative structural, sequence and redox analyses to identify properties conserved across these two enzymes. The results allow us to place DsbAs of five other Gramnegative bacteria Enterobacteriaceae, namely Shigella flexneri c. [43] and [14] d. [54] e. [51] doi: 10.1371/journal.pone.0080210.t001 8401 (SfDsbA, 100% sequence identity to EcDsbA), Enterobacter cloacae SCF-1 (EnDsbA, 84%), Citrobacter koseri ATCC BAA-895 (CkDsbA, 84%), Cronobacter sakazakii SP291 (CsDsbA, 82%) and K. pneumonia 342 (KpDsbA, 81%) into the same DsbA cluster as SeDsbA and EcDsbA. To assess whether the redox and structural properties are maintained in this DsbA group we focused on KpDsbA, which shares the lowest sequence identity with EcDsbA. We determined the high resolution crystal structure of reduced KpDsbA and the NMR solution structure of oxidized KpDsbA, and we measured the redox properties of this enzyme. As expected, the redox properties, surface characteristics and binding properties of KpDsbA are similar to those of EcDsbA suggesting that inhibitors developed against EcDsbA are likely to also be effective against other members of this DsbA subclass.

Protein production
Codon-optimized K. pneumoniae dsbA (GenBank® accession number ACI08793), lacking the sequence coding for the predicted signal sequence (19 aa), was cloned into a modified pMCSG7 (Midwest Center for Structural Genomics) vector compatible with ligation-independent cloning. This modified vector encoded a leader sequence consisting of an Nterminal His 6 -tag followed by a linker containing the tobaccoetch virus protease (TEV) recognition sequence. KpDsbA was expressed in BL21(DE3)pLys cells using autoinduction medium [18] and purified with Talon® resin (Clontech, Australia). The His 6 -tag was removed by TEV protease, leaving the engineered KpDsbA with two additional amino acids (S-1 and N0) at the N-terminus. A final size-exclusion chromatography step using a Superdex75 column (GE Healthcare, USA) yielded highly purified KpDsbA, as judged by SDS-PAGE. Oxidized or reduced KpDsbA was prepared using a 25-fold molar excess of copper-(II)-1,10-phenanthroline or DTT, respectively. Oxidizing/reducing agent was then removed and the protein buffer-exchanged into 10 mM HEPES, pH 7.4 in one step using GE-25 Sephadex desalting resin for crystallization and biochemical experiments.

KpDsbA Complementation of EcDsbA
The ability of KpDsbA to rescue non-motile E. coli dsbAnull (JCB817) and dsbA -/dsbBdouble-null (JCB818) strains was assessed in a cell-swarming assay as described previously [16]. The mature KpDsbA coding sequence was cloned into pBAD33 under an arabinose inducible promotor with the EcDsbA periplasmic signal sequence. A wild-type EcDsbA cloned into pBAD33 vector was used as a positive control.

KpDsbA Disulfide Reductase Activity
Under mild reducing conditions, DsbA proteins can reduce the intermolecular disulfide bonds formed between insulin chains A and B [3]. The rate of disulfide bond reduction can be spectroscopically followed at OD 650nm by an increase in turbidity resulting from production of the insoluble insulin chain B [20]. Samples were prepared in 1 cm cuvettes containing 10 μM of protein (KpDsbA, EcDsbA or EcDsbC), 0.33 mM DTT and 2 mM EDTA in 100 mM NaH 2 PO 4 / Na 2 HPO 4 titrated to pH 7.0. Catalysis was initiated by the addition of 0.131 mM insulin (I0516, Sigma-Aldrich, Australia) to the sample mixture. The assay was repeated three times and data were plotted showing standard deviations.

Measurement of KpDsbA Redox Potential
The standard redox potential of KpDsbA was measured using its intrinsic tryptophan fluorescence, as described previously for EcDsbA [6]. Oxidized KpDsbA was incubated for 12 h at 25 °C in degassed 100 mM NaH 2 PO 4 / Na 2 HPO 4 buffer (pH 7.0, 1 mM EDTA, 298K), containing 1 mM oxidized glutathione (GSSG) and varying concentrations of reduced glutathione (GSH) (0-2 mM). KpDsbA (200 µL) from each redox condition was dispensed into a 96-well plate (TPP AG, Switzerland #92096) and tryptophan fluorescence was measured (excitation at 280 nm, emission set to 332 nm) using a microplate reader (Synergy H1 and Gen5 2.0 software, Biotek, USA). Data were normalized and the redox potential was calculated as described for EcDsbA [6]. In brief, the equilibrium constant K eq was calculated using the equation: Y = ([GSH] 2 / [GSSH])/(K eq + ([GSH] 2 / [GSSH])), where Y is the fraction of reduced protein at equilibrium. The redox potential for KpDsbA was calculated from the Nernst equation: E 0' KpDsbA = E 0' GSH/GSSH -(RT/nF)lnK eq where E 0' GSH/GSSH = -240 mV, R is the ideal gas constant 8.314 JK -1 mol -1 , T is the absolute temperature in K, n is the number of electrons transferred (n = 2), F is the Faraday constant 9.648x104 Cmol -1 and K eq is the equilibrium constant derived from the binding equation. All measurements were performed as biological triplicates. The graph shows a plot of the average values including error bars representing the standard deviation for the replicates.

KpDsbA Thiolate Anion pK a Determination
The pH-dependent absorbance of the catalytic thiolate anion of KpDsbA was followed at 240 nm [21] using a CARY 50 UV/VIS spectrophotometer (Agilent Technologies, USA). The pH titration measurements of oxidized or reduced KpDsbA (40 μM) in 2 mL composite buffer (10 mM Tris, 10 mM sodium citrate, 10 mM K 2 HPO 4 , 10 mM KH 2 PO 4 , 200 mM KCl, and 1 mM EDTA) were conducted at 22 °C. Absorbance (λ = 240 and 280 nm) was measured between pH 6.5 and 2.0 in 0.25 increments. The pK a value was calculated from the fitted curves of three replicates using the Henderson-Hasselbalch equation (pH = pK a -log ([A240 ⁄A280]red ⁄ [A240 ⁄A280]oxid)). Experiments were repeated at least three times. Plotted data represent average values and error bars represent the standard deviations across the replicates.

Relative Stability of Oxidized and Reduced Forms of DsbA Enzymes
Temperature-induced unfolding of native SeDsbA and KpDsbA was determined as described previously [13] using a Jasco J-810 circular dichroism (CD) spectropolarimeter (Jasco, USA). The redox state of the protein was confirmed using Ellman's reagent [22]. The largest difference in molar ellipticity for oxidized or reduced enzymes was calculated from initial far-UV CD spectra (from 250 nm to 190 nm) recorded at 25 °C and 95 °C, respectively. The unfolding of oxidized and reduced protein (SeDsbA ox = 220 nm, SeDsbA red = 220.5 nm and KpDsbA ox = 211 nm, KpDsbA red = 209.5 nm) was monitored at a heat rate of 1 K / min from 298 K to 368 K in a 1 mm quartz cuvette. All measurements were carried out with 10 µM protein in 100 mM NaH 2 PO 4 / Na 2 HPO 4 , 1 mM EDTA at pH 7.0. Samples for measurement of reduced enzyme contained 0.75 mM DTT. Raw data were analyzed in Prism and fitted to a twostate unfolding model as described previously [23]. The standard deviation was measured from three replicates.

KpDsbA Dithiol Oxidation Activity
A peptide (CQQGFDGTQNSCK) with a 1,4,7,10tetraazacyclododecane-1,4,7,10-tetraacetic acid (DOTA) group amide-coupled to the N-terminus, and a methylcoumarin amide-coupled to the ε-amino group of the C-terminal lysine, was purchased from AnaSpec (Fremont, CA). Lyophilized peptide was re-suspended in 100 mM imidazole, pH 6, at a concentration of 2 mM. Europium trifluoromethanesulfonate (Sigma Aldrich, Australia) solution (100 mM) was added to the peptide at a molar ratio of 2:1 and incubated for 5 min at room temperature, to allow europium chelation. The peptide solution was then immediately aliquoted, flash frozen in liquid nitrogen and stored at -80°C. An increase in fluorescence occurs upon oxidation of the peptide cysteines to form a disulfide. Thus, fluorescence can be used to monitor the capacity of DsbA enzymes to catalyse dithiol oxidation.
Assays were conducted using a Synergy H1 multimode plate reader (BioTek, USA) with the excitation wavelength set to 340 nm and emission to 615 nm. A 150 μs delay before reading and 100 μs reading time were used for time-resolved fluorescence. The assay was performed in a white 384-well plate (Perkin Elmer OptiPlate-384, Part #: 6007290). The buffer consisted of 50 mM MES, 50 mM NaCl and 2 mM EDTA at pH 5.5. The reaction consisted of a 50 μL solution in each well, containing 160 nM EcDsbA, KpDsbA or SeDsbA, 1.6 μM EcDsbB (crude membrane extracts, containing ubiquinone) and 8 μM peptide substrate added last to initiate the reaction. Samples containing buffer and DsbA or buffer and peptide were used as controls. Data were measured for three replicates and are presented as mean values, with the standard error of the mean indicated by error bars.

KpDsbA Crystallization and Crystal Structure Determination
After initial screening using the UQ ROCX facilities, crystals of reduced KpDsbA were grown at 20 °C in VDXm 24-well plates (Hampton Research) using the hanging-drop vapor diffusion method. Screening plates were imaged and incubated in a RockImager 1000 (Formulatrix, MA, USA). Drops contained 0.5 μL of 180 mg/mL reduced KpDsbA and 0.5 μL of crystallization solution (0.1 M succinic acid pH 5.3, 25 % (w/v) polyethylene glycol 1500 and 15 % (v/v) 2-methyl-2,4pentanediol). For diffraction data measurement, crystals were frozen in liquid nitrogen without additional cryo-protectant. Diffraction data were measured at the Australian Synchrotron micro-focus MX2 beamline using BlueIce software [24]. Reflections were processed in Mosflm [25] and XDS [26], analyzed and converted to MTZ in Pointless [27] and scaled in SCALA [27]. Phases were obtained by molecular replacement (MR) using PHASER [28] with EcDsbA as template (PDB code: 1DSB) . The initial model was improved by iterative model building in COOT [29] and refinement in PHENIX [30]. However, the progress of refinement was stalled with a high Rfactor/Rfree of 25.7 % / 29.3 %. Diffraction data analysis in Phenix.xtriage indicated that the crystal was merohedrally twinned with a twinning fraction of 0.42. Further refinement cycles were performed using the twin target function as implemented in PHENIX with the twinning operator h,-h-k,-l. Two fold non-crystallographic symmetry (NCS) is present (which does not align with space group axes), though NCS was not used at any stage of refinement. The refinement finally converged after several TLS refinement cycles. No atoms were modeled into additionally spherical density located between chain D (L133) and chain B (T57) because it was not obvious what was bound. The stereochemical quality of the final model was assessed using MolProbity [31]. A summary of the data processing and refinement statistics are provided in Table 2.
Molecular figures were generated in PyMOL (The PyMOL Molecular Graphics System, Version 1.5.0.4 Schrödinger, LLC) and figures of the electrostatic potential were generated using APBS [32]. The surface, including the proportion of carbon atoms lining the hydrophobic groove in KpDsbA, was calculated using the CastP server [33], by averaging over all six molecules within the asymmetric unit. RMSD calculations and structural alignments were conducted using PyMOL as well as FATCAT [34].

NMR Structure Determination of Oxidized KpDsbA
A sample of uniformly 13 C, 15 N labeled oxidized KpDsbA (1.3 mM) was prepared in 50 mM MES (pH 6.5, 10% 2 H 2 O and 90 % 1 H 2 O). NMR experiments were conducted at 303 K on either 600 MHz or 800 MHz spectrometers equipped with cryogenically cooled probes. All spectra were acquired with standard pulse sequences and processed using TOPSPIN3.1 (Bruker BioSpin). H N , N, C α , C α-1 , C β , C β-1 peak lists were generated manually in CARA using 2D [ 15 N, 1 H]-HSQC, 3D HNCA, 3D CBCA(CO)NH and 3D HNCACB spectra and used as the input for automated backbone assignments using UNIO-MATCH. These assignments were refined manually and extended using 3D 15 N-resolved [ 1 H, 1 H]-NOESY. H β , H α assignments were obtained using a 3D HBHA(CBCACO)NH spectrum. H N , N, C α and C β assignments together with H β , H α were provided as input for UNIO-ATNOS/ASCAN for automated side-chain assignments using 3D 15 N-, 13 C ali -and 13 C aro -resolved [ 1 H, 1 H] NOESY datasets [35,36]. Upper limits for distance restraints used in structure calculations were automatically generated from NOESY datasets using UNIO-ATNOS/CANDID and the structure of oxidized KpDsbA was determined using the torsion angle dynamics program CYANA3.0 [37]. Conformers with lowest CYANA target function values were energy minimized using OPALp and validated using structure validation tools (http:/www.pdb.org/ and http:/ www.nihserver.mbi.ucla.edu/). Structures were inspected and  [38]. Table 3 summarizes the NMR statistics.

Binding Affinity of DsbA-Interacting Peptides
Crystal structures of the EcDsbA:EcDsbB complex revealed that the P2 loop region of EcDsbB interacts with EcDsbA [39,40]. Two peptides derived from the P2 loop sequences of EcDsbB and KpDsbB (Ec -PSPFATCD and Kp -PSPFQTCD) were synthesized by solid-phase methods using Fmoc deprotection on rink-amide MBHA resin (leading to C-terminal amidation) and capped by N-terminal acetylation. Amidation and acetylation ensure that there are no charges on the peptide termini, as these are not present in the native DsbB

Comparative Sequence and Structural Analyses
The sequence conservation of ten virulence factors previously identified [2] as substrates of DsbA were analyzed here. Sequences from published and validated DsbA substrate virulence factors were taken from the original literature and used to search the publicly available UniProt database [41] for potential homologues in E. coli, S. enterica Typhimurium and K. pneumoniae. Most of the 10 factors were originally identified in those three organisms except YscC and Caf1M, which were initially reported in Yersinia pestis. A protein-protein BLAST search was performed using the UniProt bacterial genome database with a threshold of P < 0.0001. Unless stated otherwise, homologues were identified in pathogenic strains, i.e. E. coli UPEC O6:K15:H31 and EPEC O127:H6 / O55:H7, S. enterica Typhimurium SL1344 and non-motile K. pneumonieae (hvKP1 / MGH 78578 / NTUH-K2044). Sequence identity between homologues was extracted from the UniProt protein BLAST results. All other sequence alignments reported herein (e.g. for Table 1) were conducted using ClustalW2 [42].

Binding Residues of EcDsbA are conserved in SeDsbA and DsbAs of Five Other Enterobacteriaciae
EcDsbA and SeDsbA share 86 % sequence identity and both have been characterized previously [14,43]. SeDsbA can complement EcDsbA [44] in a null mutant motility assay, indicating that SeDsbA is able to interact with the EcDsbA binding partner EcDsbB and with the EcDsbA substrate E. coli FlgI [45]. Both are weak disulfide reductants in the standard insulin reduction assay for redox enzymes [43]. Both are similarly oxidizing enzymes: the redox potentials of EcDsbA and SeDsbA are -122 and -126 mV, respectively [9,43], whereas the range for all DsbAs is -80 to -163 mV ( Table 1). In both EcDsbA and SeDsbA the measured pK a of the nucleophilic cysteine is 3.3 [7,43], though values vary across all DsbAs from 3.0 to 5.1 ( Table 1). Although disulfide bonds generally stabilize folded proteins, the disulfide form of DsbA enzymes is destabilizing [6,7]. The melting temperatures of the oxidized and reduced forms of EcDsbA and SeDsbA are almost identical (reduced 350 K and 351 K; oxidized 341 K and 342 K, respectively) [13] (Figure S1), whereas the range of melting temperatures across all DsbAs varies considerably ( Table 1). Importantly, the crystal structures of EcDsbA and SeDsbA can be superimposed with an RMSD of 0.8 Å for 176 Cα atoms, whereas across all structurally characterized DsbAs the RMSD with EcDsbA varies from 1.3 Å to 2.9 Å (for 122-167 Cα atoms) ( Table 1) [14]. Two catalytically relevant EcDsbA complex structures have been described, a complex between EcDsbA and EcDsbB [39,40,46] and one between EcDsbA and a peptide segment of SigA, an autotransporter protein from Shigella flexneri [47]. Analysis of these structures revealed that the binding interface comprises the N-terminal regions of the active site helix H1, as well as loops L1 (the first of two loops connecting the thioredoxin and helical domains), L2 (the second of two loops connecting the thioredoxin and helical domains, also referred to as the cisPro loop) and L3-H7 (residues in the loop preceding and at the N-terminal region of helix H7) (Figure 1A). A hypothesis is that DsbAs sharing overall high sequence identity with EcDsbA and with highly conserved loop lengths and residues in these regions will share similar binding activities. As shown in Figure 1B, SeDsbA falls into this cluster as does Shigella flexneri (SfDsbA, P52235), Enterobacter cloacae (EnDsbA. E3G5L9), Citrobacter koseri (CkDsbA, A8AL80) and Cronobacter sakazakii (CsDsbA, I2ED40) and K. pneumoniae (KpDsbA) (Figure 1B). Of these, the DsbA with lowest sequence identity to EcDsbA is KpDsbA (81 %) encoded by an important human pathogen responsible for many antibioticresistant nosocomial infections [1,48,49]. To determine whether KpDsbA falls within the same class as EcDsbA, we investigated its structure, surface, redox and binding properties and compared them with EcDsbA.

KpDsbA Complements EcDsbA in vivo
The E. coli protein FlgI is required for E. coli motility and, in turn, FlgI requires the DSB machinery of E. coli to function. FlgI function is impaired in E. coli dsbAdeficient (JCB817) and dsbA -/dsbBdouble-mutant (JCB818) strains due to the absence of EcDsbA mediated dithiol oxidase activity [50]. As a consequence, these E. coli strains are non-motile. Intriguingly, K. pneumoniae is non-motile and does not encode a FlgI homologue. We tested whether KpDsbA was able to catalyse disulfide bond formation of E. coli FlgI using an in vivo complementation strategy [3]. We demonstrated that KpDsbAlike SeDsbA [44] -can fully restore the motility of dsbAdeficient strains, but not in the double dsbA -/dsbBmutant cells ( Figure S2). This experiment shows that KpDsbA is able to oxidize FlgI cysteines and this requires the presence of EcDsbB.

KpDsbA has redox properties almost identical to those of EcDsbA and SeDsbA
EcDsbA exhibits weak insulin reductase activity in the presence of dithiothreitol [52] whereas the E. coli disulfide isomerase EcDsbC is highly active in this assay. Reduction of the intermolecular disulfide bonds between the A and B chains of insulin results in precipitation of the B chain and this can be monitored by measuring the OD 650nm . We found that purified recombinant KpDsbA has the same weak insulin reductase activity as EcDsbA (Figure 2A) and SeDsbA [43]. The activity of other characterized DsbA enzymes varies. NmDsbA1, for example, has a much weaker activity than that of EcDsbA [15], and DsbA from Mycobacterium tuberculosis (MtbDsbA) is inactive in this assay [53]. In contrast, TcpG (VcDsbA) from Vibrio cholerae catalyses insulin reduction much faster than EcDsbA [54].
The pK a value of the nucleophilic cysteine in the active site CXXC motif is a key determinant of DsbA reactivity towards substrate proteins. We measured the pK a value for the nucleophilic cysteine of KpDsbA using pH-dependent thiolate absorbance at λ = 240 nm ( Figure 2C). The pK a Cys30 for KpDsbA was found to be 3.2, nearly identical to that of EcDsbA and SeDsbA (3.3) compared with the observed range for other DsbAs (3.0-5.1).
We also confirmed that reduced KpDsbA (T m red 347.1 ± 0.2 K) is more stable than oxidized KpDsbA (T m ox 335.8 ± 0.3 K) ( Figure 2D) Figure S1). Again, the range reported for all DsbAs is much wider (T m red 337-357 / T m ox 331-341 K) [51,54]. We then tested the dithiol oxidase activity of KpDsbA using a fluorescently labeled peptide substrate. The activity was monitored by the increase in europium fluorescence resulting from cyclization of the substrate peptide through formation of an intramolecular disulfide bond. In the presence of EcDsbB, we found that the rate for KpDsbA and SeDsbA catalyzed disulfide bond formation was almost indistinguishable from that of EcDsbA measured at the same concentration of enzyme (Figure 3). This result suggests that KpDsbA (and SeDsbA) is able to interact in the same way as EcDsbA with the peptide substrate and with EcDsbB. TcpG has a similar activity to EcDsbA in this assay [54], whereas MtbDsbA is inactive in the presence of EcDsbB [53].

Crystal structure of reduced KpDsbA
We determined the crystal structure of reduced KpDsbA (PDB: 4MCU) at 1.99 Å resolution by molecular replacement, using EcDsbA as the template. As expected, the structure is very similar to that of EcDsbA ( Figure 4A). The asymmetric unit contains six KpDsbA molecules each adopting the typical DsbA fold. Structural superposition of these six independent copies yielded a root mean square deviation (RMSD) < 0.45 Å for 176 Cα atoms between residues Gly6 -Val181. Likewise, structural  [16]. These higher values are a consequence of structural deviations including a truncated helix H7 and a shortened hydrophobic groove.
The structure of the catalytic site of KpDsbA is strictly conserved with that of EcDsbA, comprising the active site motif 30 Cys-Pro-His-Cys 33 located at the N-terminal end of helix H1 and the adjacent cisPro (Val-Pro 151 ) L2 loop (Figure 4B). The cysteine residues (Cys30 and Cys33) are present in the reduced state in the crystal structure. A hydrophobic patch and a large groove surrounds the nucleophilic Cys30, as also occurs in EcDsbA and SeDsbA ( Figure 4C). As expected, these surface features are lined with residues contributed from the L1, L2 and L3 loops.
The six independent copies of KpDsbA in the crystal structure allow an analysis of conformational variability of the loop residues forming the binding surface. This revealed that the side chains of His32, Phe63, Leu64, Gln147, Thr167 and Met170 adopt various rotamer conformations, whereas there is no evidence of conformational variability in Tyr29, Cys30, Pro31, Val149, Pro150, and Phe173 ( Figure 5A). The side chain variations do not influence the surface accessibility of the hydrophobic groove, which was calculated to be 371 ± 32 Å 2 by CastP [33] across the 6 molecules. Moreover, the hydrophobic nature of the groove is unaffected by the side chain conformational variability as indicated by the proportion of carbon atoms lining this groove (69 ± 3 %) [33].

NMR Solution Structure of KpDsbA is Similar to the Crystal Structure
Previous studies have demonstrated that there are minimal differences between reported structures (crystal and NMR) of oxidized and reduced EcDsbA. To determine if this was also  (Figure S4 A/B). It was not possible to assign several backbone amide resonances corresponding to residues in the β1-β2 loop (Ile16, Gly18, Glu19, Gln21, Val22, Leu23), so that this region appears to be largely disordered in the NMR ensemble compared with the rest of the structure. The backbone (N, Cα, C') and all-heavy atom RMSD for the 179 well-defined residues (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15) of the 20 KpDsbA conformers were 0.67 ± 0.17 Å and 1.03 ± 0.13 Å, respectively. Structural statistics are summarized in Table 3. As observed for other DsbA structures, the individual thioredoxin and helical domains can be superimposed with higher precision than the entire structure. This is most likely due to inter-domain motion, which has also been reported in the structures of EcDsbA [55] and VcDsbA [56]. Residues which fall into disallowed Ramachandran regions include the unassigned residues Glu19, Gln21, Val22, and His32, and residues in loop regions, i.e. Lys55, Phe63, Leu64, Asn155 and Met170.
The overall conformation of the NMR structure of oxidized KpDsbA is similar to that of the crystal structure of reduced KpDsbA ( Figure 5C). For example, superposition of molecule A in the crystal structure of reduced KpDsbA with the first  [57,58].
The structures of the catalytic sites and hydrophobic surface features are similar, considering that the cysteines of the CXXC motif are oxidized in the NMR structure and reduced in the crystal structure ( Figure S4C). As has been noted previously for other DsbA solution and crystal structures [56,59], L3 of KpDsbA is a relatively flexible part of the protein in both NMR and crystal structures (Figure 5B and C). Thus, overall the structures of oxidized and reduced KpDsbA are similar, notwithstanding the different conditions and approaches used for structure determination.

Binding Affinity of DsbB peptides is similar for KpDsbA and EcDsbA
The similar surface features and similar predicted binding residues of KpDsbA and EcDsbA suggested that these enzymes would interact with binding partners with similar affinity. The crystal structures of the EcDsbA:EcDsbB complex showed that the second periplasmic loop P2 of EcDsbB binds directly to EcDsbA [39,40]. The binding residues are 98-PSPFATCD-104 and these are highly conserved in KpDsbB (98-PSPFQTCD-104). These two P2 peptides were synthesized and isothermal titration calorimetry (ITC) was used to assess their binding affinity for both enzymes. KpDsbA and EcDsbA were found to bind to PSPFATCD and PSPFQTCD with similar affinities (K d 11-18 µM, Table 4, Figure S3A). We investigated the interaction of KpDsbA with PSPFQTCD by structural superposition of KpDsbA onto the structure of EcDsbA in the EcDsbA:EcDsbB complex structure. Residue Ala of EcDsbB PSPFATCD was mutated in silico to PSPFQTCD, using the most commonly observed rotamer for glutamine. The superimposed model showed that the P2 loop matched the surface of KpDsbA very well, with no clashes apparent between the P2 residues and KpDsbA ( Figure S3B).

Discussion
We have shown that the structural, surface, redox and binding properties of EcDsbA, SeDsbA and KpDsbA enzymes are highly conserved, and that these three DsbAs and four other DsbAs (from Enterobacter cloacae, Citrobacter koseri, Shigella flexneri and Cronobacter sakazakii) might be considered an Enterobacteriaceae subclass of DsbA. Carbapenem-resistant Enterobacteriaceae are responsible for a large proportion of difficult to treat community-and hospitalacquired infections [60] and there is an urgent need to develop novel therapeutic strategies to tackle these so-called 'super bugs' [61].
One approach to generate new classes of antibacterials is to target virulence rather than viability of bacteria. An antivirulence approach is predicted to lead to less selective  [30] and is contoured at 1.0 σ C. Electrostatic surface representation of EcDsbA, SeDsbA and KpDsbA (left, middle, right). Positive and negative electrostatic potentials are contoured from blue (+7.5 kT/e) to red (-7.5 kT/e). The hydrophobic grooves of all three enzymes are indicated by a dashed oval [8,43]. doi: 10.1371/journal.pone.0080210.g004 pressure for resistance development, since most virulence traits are not essential for survival [62]. Targeting virulence may also expand the repertoire of antimicrobial targets, preserve the endogenous host microbiome and extend the lifespan of conventional antibiotics [61]. Most antivirulence strategies developed to date target individual virulence factors [61][62][63][64][65] and this has yielded some successes [66,67]. However, the armory of DsbA substrate virulence factors expressed in different Enterobacteriaceae varies (Figure 6), so that drugs targeting specific virulence factors may not be effective against all Enterobacteriaceae. On the other hand, DsbA itself catalyzes assembly of many virulence factors [68][69][70] and DsbA knockouts severely attenuate virulence in infection models [12]. Targeting DsbA is therefore a compelling approach for the development of anti-virulence agents, because DsbA inhibitors should inhibit a range of virulence traits. Significantly, our findings point to the opportunity to , with C α atoms colored by temperature factor (B-factor). Molecule D was selected as its temperature factor distribution is the most pronounced due to minimal crystallographic contacts. In particular, the high B-factor of loop L3 indicates mobility in that region, consistent with the NMR data C. Stereo diagram of representative states of reduced (X-ray, cyan) and oxidized (NMR, yellow) structures of KpDsbA. Red arrows highlight differences in the structures at N-terminal and L3 loop regions. doi: 10.1371/journal.pone.0080210.g005 develop a single antivirulence drug effective against DsbAs encoded by at least seven Enterobacteriaceae pathogens.
The crystal structure and NMR solution structure of KpDsbA (the latter derived by semi-automated approaches) reported here are in excellent agreement. The availability of structural data for KpDsbA opens up the possibility of using structurebased approaches to generate DsbA inhibitors. Moreover, the close similarity of the crystal and NMR structures, and the use of semi-automated NMR, highlights how NMR can be used as an efficient first screen in e.g. drug-like fragment campaigns. By contrast, the six molecules in the asymmetric unit of KpDsbA crystal structure is far from ideal for rapid fragmentscreening, but is nevertheless advantageous for follow up analysis.
Taken together, our data show that DsbA enzymes sharing >80% sequence identity with EcDsbA also share almost identical redox and surface properties and can thus be categorized as a distinct DsbA subclass. Further analyses will be required to determine how many subclasses of DsbA exist, and whether DsbAs with lower than 80% sequence identity will fall into the EcDsbA-like class. Importantly, our results suggest that compounds designed to inhibit EcDsbA will likely inhibit all DsbAs within the same class. Finally, we propose that compounds that bind KpDsbA might be identified rapidly using semi-automated NMR approaches, and that development of 'hits' to optimise potency can be achieved using a pipeline comprising biochemical and structural assays similar to those outlined herein. Figure S1. Thermal unfolding of SeDsbA. A. Temperatureinduced unfolding of oxidized (ox, ν) and reduced (red, θ) SeDsbA was monitored by far-UV CD spectroscopy. Unfolding was monitored in 1 K steps from 298 K to 368 K. Normalized average data points of three measurements were fitted to a two-state folding model. The reduced state of SeDsbA (351.2 +/-0.2 K) is 9 K more stable than its oxidized (342.8 +/-0.  Table 4. B. Model of Figure 6. Conservation of DsbA substrate virulence factors. Comparison of the sequence conservation of DsbA oxidoreductases from E. coli (Ec), S. enterica Typhimurium (Se) and K. pneumonia (Kp) and of characterized DsbA substrate virulence factors. Sequence identities relative to the characterized substrate protein are represented in different colours, as shown in the key. White squares indicate the lack of a sequence homologue in the specific bacteria. * YscC and Caf1M were identified as DsbA substrate proteins in Yersinia pestis [71,72]. a [68], b [73], c [44], d [74], e,f [75], g [5], h,k [71,72], i [45], j [69].