Structure of the Acinetobacter baumannii Dithiol Oxidase DsbA Bound to Elongation Factor EF-Tu Reveals a Novel Protein Interaction Site

Background: DsbA is a master virulence determinant of bacterial pathogens and a target for antivirulence drugs. Results: AbDsbA is a class I dithiol oxidase that binds EF-Tu-derived and DsbB-derived peptides on different enzyme surfaces. Conclusion: Discovery of high affinity peptide interaction sites provides a platform for inhibitor design. Significance: AbDsbA inhibitors could have anti-biofilm activity against multidrug resistant Acinetobacter baumannii. The multidrug resistant bacterium Acinetobacter baumannii is a significant cause of nosocomial infection. Biofilm formation, that requires both disulfide bond forming and chaperone-usher pathways, is a major virulence trait in this bacterium. Our biochemical characterizations show that the periplasmic A. baumannii DsbA (AbDsbA) enzyme has an oxidizing redox potential and dithiol oxidase activity. We found an unexpected non-covalent interaction between AbDsbA and the highly conserved prokaryotic elongation factor, EF-Tu. EF-Tu is a cytoplasmic protein but has been localized extracellularly in many bacterial pathogens. The crystal structure of this complex revealed that the EF-Tu switch I region binds to the non-catalytic surface of AbDsbA. Although the physiological and pathological significance of a DsbA/EF-Tu association is unknown, peptides derived from the EF-Tu switch I region bound to AbDsbA with submicromolar affinity. We also identified a seven-residue DsbB-derived peptide that bound to AbDsbA with low micromolar affinity. Further characterization confirmed that the EF-Tu- and DsbB-derived peptides bind at two distinct sites. These data point to the possibility that the non-catalytic surface of DsbA is a potential substrate or regulatory protein interaction site. The two peptides identified in this work together with the newly characterized interaction site provide a novel starting point for inhibitor design targeting AbDsbA.

The global increase in multidrug-or pandrug-resistant Acinetobacter baumannii infections poses a major risk to public health. A. baumannii is one of the "ESKAPE" pathogens (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Pseudomonas aeruginosa, and Enterobacter species), so-called because of their ability to "escape" the effects of clinically useful antibiotics by biofilm formation and rapid acquisition of resistance genes (1). Collectively, these ESKAPE pathogens are a major cause of hospital-and community-acquired infections, morbidity, and mortality. New targets and strategies are needed to combat infections caused by these pathogens.
A. baumannii (colloquially known as Iraqibacter) is a Gramnegative, opportunistic pathogen that can cause a wide range of diseases, including pneumonia, meningitis, bacteremia, urinary tract infections, and wound infections (2,3). This microbe has been responsible for multiple outbreaks around the world, including Australia and among United States military personal deployed in the Middle East (5)(6)(7)(8)(9). There is a high incidence of A. baumannii infection among immunocompromised patients, especially those who have experienced long stays in hospital (10). Multidrug-resistant A. baumannii such as the AYE strain have accumulated a large cluster of resistance genes via horizontal gene transfer. Both its multidrug resistance and its ability to survive on highly desiccated abiotic surfaces (e.g. plastic and glass medical devices) have been linked to its success as a nosocomial pathogen (11,12). Currently, there is no vaccine against A. baumannii (4).
A. baumannii can invade host cells, but little is known about its pathogenic mechanism. Biofilm formation is a major virulence trait contributing to bacterial colonization on tissue and persistence in the hospital environment (13). Electron and fluorescence microscopy experiments showed that cell surface fimbriae are necessary for A. baumannii attachment and initiation of biofilm formation on abiotic and biotic surfaces (14). Fimbriae are filamentous, multisubunit protein assemblies. Each fimbrial subunit contains an evolutionarily invariant disulfide bridge (15) that requires both the oxidative folding and the chaperone-usher pathways for assembly (13,14,16,17). The disulfide bond-forming system of Gram-negative bacteria typically comprises DsbA (a soluble, periplasmic, thioredoxin-fold dithiol oxidoreductase) and DsbB (an integral inner membrane protein). DsbA catalyzes the formation of disulfide bonds in nascent substrate proteins as they are translocated into the periplasm. Through this reaction, the active site CXXC motif of DsbA becomes reduced and is restored to its functionally competent oxidized state through interaction with DsbB. DsbA is considered a master regulator of virulence and virulence traits (14, 18 -21). Deletion or mutation of dsbA attenuates virulence factor maturation in bacterial pathogens including Proteus mirabilis (22), uropathogenic Escherichia coli (23), Burkholderia pseudomallei (24), Vibrio cholerae (25), Shigella flexneri (26), and Salmonella enterica serovar typhimurium (27). DsbA inhibitors are potential anti-virulence drugs (28).
Here we characterize DsbA from A. baumannii (AbDsbA) 6 and show that it has an unusually basic surface, with redox and catalytic properties that qualify it as a bona fide dithiol oxidase. We also report an unexpected interaction between AbDsbA and E. coli EF-Tu (EcEF-Tu); the complex can be co-purified from the cytoplasm of the E. coli expression system. This serendipitous discovery allowed the crystal structure determination of the AbDsbA⅐EcEF-Tu complex at 2.15 Å resolution and identification of specific intermolecular interactions.
Although the highly conserved EF-Tu is generally considered to be a cytoplasmic protein, extracellular localization has been reported in bacterial pathogens including A. baumannii (29), P. aeruginosa (30,31), and Mycobacterium tuberculosis (32). Moreover, EF-Tu has been found to bind human fibronectin, platelet-activating factor receptor, plasminogen, and factor H for adhesion, tissue infiltration, and complement system inactivation. Intracellularly, EF-Tu has functions additional to its central role in protein translation. For example, in Bacillus subtilis EF-Tu plays a role in cell shape by interacting with actinlike proteins (33). Bacteriophages Q␤ and T4 both recruit EF-Tu to perform essential functions (34,35). We investigated the interaction by generating peptides derived from EF-Tu and showed that these bind tightly to AbDsbA, providing a platform for the future design of peptidomimetic inhibitors targeting AbDsbA.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-A codon-optimized synthetic gene coding for A. baumannii (AYE strain) DsbA (amino acids 27-205; Uniprot: B0V5X3) lacking the predicted periplasmic leader sequence (amino acids 1-22) was cloned into the pMCSG7 (36) plasmid to enable expression in the bacterial cytoplasm. This expression construct was also mutated to the A. baumannii (SDF strain) DsbA using the QuikChange (Agilent) procedure. These two natural DsbA variants differ by a single amino acid at the CXXC catalytic motif (CPHC in AYE strain and CPLC in SDF strain).
AbDsbA proteins were expressed in BL21(DE3) cells using autoinduction (37). Cell pellets were resuspended in 50 mM Tris, pH 8.0, 100 mM NaCl and lysed in a Cell Disruptor (TS-Series, Constant Systems LTD.). Cell debris was removed by centrifugation. The supernatant was loaded onto Talon resin (Clontech), packed in a disposable gravity column, and equilibrated with 50 mM Tris, pH 8.0, 100 mM NaCl. Unbound proteins were removed with washing buffer (50 mM Tris, pH 8.0, 100 mM NaCl, 10% glucose, and 10% sucrose). Elution was achieved using wash buffer containing 300 mM imidazole. SDS-PAGE analysis of eluent fractions revealed that AbDsbA coeluted with a second protein, identified as E. coli EF-Tu using mass spectrometry analysis (see below). The His 6 affinity tag was removed by tobacco etch virus protease, leaving three vector-borne residues (Ser-Asn-Ala) at the N terminus of AbDsbA. The AbDsbA⅐EF-Tu complex was further purified using a Superdex S200 gel-filtration column (GE Healthcare). For purification of AbDsbA free from EF-Tu, the above procedure was modified to a batch purification step with Talon resin and repeated high salt wash (50 mM Tris, pH 8.0, 1.0 M NaCl). EcDsbA and E. coli membrane extracts containing overexpressed E. coli DsbB (EcDsbB) were prepared as described previously (38) and resuspended in phosphate-buffered saline containing 10% glycerol.
Protein Identification by Mass Spectrometry-Coomassiestained bands corresponding to AbDsbA (21 kDa) and a ϳ40-kDa endogenous E. coli protein were cut from SDS-PAGE and subjected to in-gel tryptic digestion as described previously (39). The extracted peptides were applied to the MALDI target plate with ␣-cyano-4-hydroxycinnamic acid matrix. MS spectra were recorded in positive reflector mode at a laser energy of 3200 using a 4700 Proteomics Analyzer MALDI TOF-TOF (Applied Biosystems, CA). All MS/MS data from the TOF-TOF were acquired using the default positive ion, 1-kV collision energy, at a laser energy of 4400. Thirteen dominant species from the MALDI-MS data for the E. coli protein were queried against the SwissProt E. coli database with a tolerance of 0.2 Da using the MASCOT Peptide Mass Fingerprint server (Matrix Science). The top hit from the search indicated that the unknown protein was EF-Tu. In addition, MALDI TOF-TOF data were used to identify a 2117.2 m/z species as the peptide AIDKPFLLPIEDVFSISGR, an EF-Tu tryptic peptide. The band corresponding to 21 kDa was confirmed as AbDsbA by comparison of the MALDI-MS species detected in its tryptic digest with those expected to arise from the tryptic digestion of AbDsbA.
Crystallization and Structure Determination-AbDsbA⅐EF-Tu complex crystals were grown at 20°C by setting drops with 250 nl of protein solution (10 mg/ml in 25 mM HEPES, pH 7.4, and 100 mM NaCl) and 250 nl of precipitant solution (200 -300 mM KCl and 4.5-6% PEG 3350) using a Mosquito robot (TTP Labtech) at the University of Queensland Remote Operation Crystallization and X-ray (ROCX) diffraction facility. Crystals were cryoprotected (in 250 mM KCl, 30 mM HEPES, pH 7.4, 15% PEG3350, and 25% glycerol) and flash-frozen in liquid nitrogen. Diffraction data from the cryo-cooled crystal were collected at the Australian Synchrotron MX2 beamline with an ADSC Quantum 315r detector controlled by BLU-ICE (40). Reflections were indexed and integrated in XDS (41), analyzed in Pointless (42), and scaled in SCALA (42) from the CCP4 suite (43).
Cell content analysis indicated one copy of AbDsbA⅐EF-Tu complex in the crystallographic asymmetric unit. Molecular replacement using E. coli EF-Tu (PDB ID 1EFC), truncated into its three individual domains, yielded a solution in PHASER (44). The initial electron density map after Phenix refinement guided the location of AbDsbA and allowed tracing 33% of the AbDsbA sequence (59 of 178 residues). Iterative manual rebuilding, density modification, and refinement in PHENIX (45) and COOT (46) permitted tracing the entire AbDsbA sequence. Density corresponding to bound diphosphate nucleoside (GDP) and Mg 2ϩ was modeled in the guanosine binding domain of EF-Tu. The first seven residues of EF-Tu and the last residue of AbDsbA were not modeled due to a lack of electron density in these regions. The final R-factor/R-free was 17.2%/21.5%. Data processing and refinement statistics are provided in Table 1. The coordinates and structure factors have been deposited to the Protein Data Bank and assigned the identifier 4P3Y.
Molecular Modeling-Molecular figures were prepared using PyMOL Molecular Graphics System, Version 1.6 Schrödinger, LLC. Electrostatic surface potential calculations were carried out in APBS (47). Interaction analysis of the interface residues and the buried surface area were determined using PISA (48). Molecular modeling and rotamer analysis were performed using Coot model building tools (46), and geometry idealization was carried out in Phenix.pdbtools (45). Membrane protein topology was analyzed in TOPSCON (49).
EcDsbA Complementation-A chimera gene encoding for a mature AbDsbA with an E. coli DsbA signal peptide or a wildtype EcDsbA was cloned under an arabinose-inducible promoter in the pBAD33 plasmid (50). DsbA null E. coli cells (JCB817; Ref. 51) harboring the EcDsbA/pBAD33 or AbDsbA/ pBAD33 plasmid were spotted onto a soft M63 minimal agar plate containing 40 mg/ml concentrations of each amino acid (except L-cysteine) and 0.1% arabinose. Cell swarming was analyzed after incubating the plates for 4 -5 h at 30°C. The experiment was repeated two times, and the agar plates were photographed.
Determination of Standard Redox Potential-Oxidized AbDsbA (2 M) in degassed 100 mM phosphate buffer, pH 7.0, containing 1 mM EDTA and 1 mM oxidized glutathione (GSSG) was incubated with a range of reduced glutathione (GSH) concentrations (0.01 M-1 mM) for 24 h at 25°C. AbDsbA was precipitated with 10% trichloroacetic acid, and the pellet was washed with 100% ice-cold acetone. Free thiols in AbDsba were labeled with 4 mM 4-acetamido-4Ј-maleimidylstilbene-2-2Јdisulfonate in 50 mM Tris, pH 7.0, and 1% SDS. Reduced and oxidized forms of AbDsbA were separated on a NuPAGE 12% Bis-Tris gel (Invitrogen) and stained with Coomassie. Intensities of the reduced protein were analyzed using ImageJ (Version 1.44) (52). The fraction of the reduced protein was plotted against the ratio of [GSH] 2 /[GSSG], and the equilibrium constant K eq was calculated using a binding equation where Z is the fraction of reduced protein at equilibrium. The redox potential was calculated using the Nernst equation: is the standard potential of Ϫ240 mV (53), R is the universal gas constant 8.314 JK Ϫ1 mol Ϫ1 , T is the absolute temperature in K, n is the number of electrons transferred, F is the Faraday constant 9.648 ϫ 10 4 Cmol Ϫ1 , and K eq is the equilibrium constant. The redox equilibria measurements were repeated two times, and the resulting mean values are plotted including S.D. for each measurement.
Disulfide Reductase Assay-Insulin turbidity as a result of disulfide bond reduction was monitored using a spectrophotometer at 650 nm for 60 min (54). The reaction mixture contained 131 M insulin, 10 M AbDsbA or EcDsbA or EcDsbC in 100 mM phosphate buffer, pH 7.0, containing 0.33 mM DTT and 2 mM EDTA. The mean values obtained for each data point from three repeats are plotted, and the standard deviations are displayed as error bars.
Cysteine Thiol Oxidation Assay-Labeled synthetic peptide substrate (CQQGFDGTQNSCK) with a europium DOTA (1,4,7,10-tetraazacyclododecane-1,4,7,10-tetraacetic acid) at the N terminus and a methylcoumarin at the C-terminal lysine side chain was prepared as previously described (55). The assay was performed in a 50-l reaction volume in the presence of 2 mM oxidized glutathione or 1.6 M membrane-solubilized E. coli DsbB (EcDsbB) in a 384-well plate (PerkinElmer Life Sciences OptiPlate). The reaction mixture contained 50 mM MES, pH 5.5, 50 mM NaCl, 2 mM EDTA, and 80 or 120 nM AbDsbA or AbDsbA⅐EF-Tu complex. Disulfide bond formation of the peptide (7 M), which was added at the end to initiate the reaction, was followed using time-resolved fluorescence with excitation at 340 nm and emission at 615 nm, with a delay of 150  (85) percentile is the best among the structures of comparable resolution. The percentile (bracket) and the number of structures included in the comparison (N) are given in parentheses within bracket. The two Ramachandran outliers, residues Asp-142 and Arg-333, were also observed in the previous EF-Tu structure (PDB ID 1EFC).

Data collection
Wavelength ( s and reading time of 100 s in a Synergy H1 multimode plate reader (Biotek) as described before (56). The reaction in the absence of enzyme served as the control. The mean values obtained from three repeat experiments were plotted, and the S.E. are shown for each data point unless otherwise indicated. Isothermal Titration Calorimetry (ITC)-Synthetic peptides were purchased from Genicbio Ltd. Peptides were dissolved in 25 mM HEPES, pH 7.4, 100 mM NaCl (ITC buffer) to obtain 4 mM stock solutions, which were flash-frozen and stored at Ϫ80°C. The isothermal titration experiments were performed using either a Microcal Tm ITC200 or an autoITC200 (GE Healthcare). The syringe was loaded with titrant peptide at concentrations of 2 or 4 mM (DsbB P2 peptide) or 200 M (EF-Tu switch I peptide). The sample cell was loaded with oxidized AbDsbA with concentrations of 100 M (for titrations with DsbB P2 peptide) or oxidized/reduced/mixed AbDsbA of 10 M (for titrations with EF-Tu switch I peptide) in ITC buffer. To assess whether EF-Tu switch I peptide competes for the DsbB P2 peptide binding site, 100 M oxidized AbDsbA was incubated with 125 M EF-Tu switch I peptide for 60 min in ITC buffer before titration with DsbB P2 peptide at 2 mM concentration as described above. Titrations were executed at 25°C using 19 titrations of 2 l each separated by 180 s and at a constant stirring speed of 1000 rpm. A pre-injection of 0.5 l was performed to limit slow diffusion of titrant into the cell before the titration, and the corresponding data point was removed from subsequent analysis. Thermodynamic parameters were obtained from non-linear curve fitting with the MicroCal TM Origin software (Version 7.0552) using a 1:1 bind-ing model. Each titration experiment was repeated twice, and the resulting mean and standard deviations are tabulated.

AbDsbA Exhibits the Typical Characteristics of a Dithiol
Oxidase-A. baumannii exhibits extensive genetic variation even among closely related strains due to gene transfer, mobilization of insertion sequences, recombination, gene loss, and mutation (57). Nevertheless, DsbA is highly conserved with sequence identities of Ն99% among 122 A. baumannii strains analyzed using BLAST against the Uniprot Acinetobacter Knowledgebase. One prominent variation in AbDsbA sequences is located in the active site CXXC motif of an avirulent (SDF) strain. Specifically, the active site motif 34 CPHC 37 in virulent AYE strain (AbDsbA H36 ) is substituted for 34 CPLC 37 in the SDF strain (AbDsbA L36 ). Because variations in the active site motif can directly influence redox potential and disulfide-bond forming activity of DsbA enzymes (58, 59), we measured the redox potentials of both AbDsbA natural variants (Fig. 1, A and B). We found that the redox potential of AbDsbA H36 is more oxidizing (Ϫ101 Ϯ 1.4 mV) than that of AbDsbA L36 redox potential (Ϫ134 mV Ϯ 1.2 mV), comparable to the reported value of Ϫ94 mV for P. aeruginosa DsbA (PaDsbA) (60). Conversely, the redox potential of AbDsbA L36 was more similar to the Enterobacteriaceae DsbAs (Ϫ116 to Ϫ129 mV) (61).
Recombinant expression of a DsbA can rescue the motility of DsbA null E. coli cells because DsbAs typically have a broad protein substrate specificity (62,63). However, we found that AbDsbA only weakly restores the motility of E. coli DsbA-null cells in the cell-swarming assay (Fig. 1C).
To assess whether AbDsbA has in vitro disulfide bondforming activity, we used a time-resolved fluorescence assay using a substrate peptide labeled with a europium chelate and a coumarin donor (see "Experimental Procedures."). In the presence of EcDsbB as a source of DsbA oxidant, we found that AbDsbA enzymes were only marginally active compared with EcDsbA (Fig. 1D). However, using oxidized glutathione as the DsbA oxidant (Fig. 1E), we found that AbDsbA L36 and EcDsbA dithiol oxidase had comparable activity and that AbDsbA H36 was more active again, under the same conditions. These results suggest that AbDsbA is an oxidase, but it does not interact with EcDsbB.
We then assessed for the ability of AbDsbA to catalyze disulfide reduction using insulin as the general substrate in the presence of dithiothreitol as a mild reducing agent (51). In this experiment AbDsbA L36 reduced the intermolecular disulfide bonds of insulin more rapidly than AbDsbA H36 but much more slowly than the dedicated disulfide isomerase EcDsbC (Fig. 1F).
The results of these experiments demonstrated that both variants of AbDsbA are bona fide dithiol oxidases. The data also indicated that the interaction surface for DsbB may be different from that of EcDsbA. The lack of AbDsbA interaction with EcDsbB was somewhat surprising given that PaDsbA, which shares 44% sequence identity to AbDsbA, does interact with EcDsbB (60). However, AbDsbA has a calculated pI of 9.5 compared with 5.8 for PaDsbA, and this may contribute to differing interaction propensities.
AbDsbA Directly Interacts with EF-Tu-Recombinant Histagged AbDsbA (AbDsbA H36 , hereafter referred to as AbDsbA, for simplicity) expressed in the cytoplasm of E. coli and purified by His-affinity chromatography consistently co-purified with a 40-kDa contaminant that we identified as an endogenous E. coli protein ( Fig. 2A). Binding of the E. coli protein by His-immobilized AbDsbA was unchanged up to 1.1 M NaCl or 0.5% v/v Triton X-100. Even after metal affinity purification, AbDsbA and the E. coli protein co-eluted as a major fraction in size exclusion chromatography (Fig. 2B). Analysis of samples in SDS-PAGE run under non-reducing conditions indicated that the interaction between AbDsbA and EF-Tu was not thiol-mediated ( Fig. 2A). Using in-gel digestion and mass spectrometric analysis, we identified the co-eluant as the E. coli elongation factor Tu (EF-Tu). Notably, the dithiol oxidase activity of the AbDsbA⅐EF-Tu 1:1 complex was reduced by ϳ2/3 compared with the activity of AbDsbA alone (Fig. 2C).
Structure of the AbDsbA⅐EF-Tu⅐GDP Complex-To elucidate the structural basis for the recognition of EF-Tu by AbDsbA, we determined the structure of AbDsbA⅐EF-Tu complex. Before crystallization the protein complex was purified by size exclusion chromatography. The crystal structure of the complex was phased by molecular replacement using the coordinates of E. coli EF-Tu (PDB ID 1EFC) as the search model (Fig. 3A) and refined to 2.15 Å resolution.
The crystal structure revealed a 1:1 complex and showed that EF-Tu is captured in its guanosine diphosphate (GDP)-bound state (Fig. 3, A and E). EF-Tu⅐GDP folds into three distinct domains: the N-terminal guanine nucleotide binding catalytic domain (Domain 1, also known as the G domain) connected to 7-stranded and 6-stranded ␤-barrel domains (Domains 2 and 3). The global architecture of EF-Tu⅐GDP in complex with AbDsbA is similar to that of the previously reported EF-Tu⅐GDP structure except for a minor movement of Domains 2 and 3 and a major local conformational rearrangement in the catalytic Domain 1 (Fig. 3, B-C).
Previous structural studies showed that the EF-Tu catalytic domain can undergo nucleotide-dependent rearrangements primarily involving two flexible regions, known as switch I (residues 51-65) and switch II (residues 84 -100). In the GTP- The switch II helix alters both its orientation and length between these two nucleotide binding states. In contrast, in the present complex, switch I binds to AbDsbA in an extended conformation (Fig. 3, B and D), whereas the switch II conformation and the GDP binding mode are as described previously for EF-Tu⅐GDP (PDB ID 1EFC; Ref. 64).
In this complex with EF-Tu, AbDsbA adopts the canonical DsbA fold comprising a non-contiguous thioredoxin fold with a helical domain insertion (Fig. 4A). The AbDsbA structure is similar to that of PaDsbA with a root mean square deviation of 1.28 Å for 164 C ␣ atoms. The structure encompasses a fivestranded mixed ␤-sheet surrounded by 7 helices (H1-H7) with the crucial CXXC motif located at the N terminus of helix H1. Structural comparison of AbDsbA and PaDsbA shows that the catalytic residues formed by the CXXC ( 34 CPHC 37 ) motif and the adjacent essential cis-Pro motif (Val 150 -Pro 151 ) are identical (Fig. 4B). In AbDsbA, the distance between sulfur atoms of the first and the second of the two active site cysteines measures 3.6 Å, which falls within the range observed in other reduced DsbA structures (3.3-3.8 Å). The interatomic distances of key atoms of the active site residues of AbDsbA are similar to those of PaDsbA, suggesting that EF-Tu binding does not induce changes at the AbDsbA active site. Catalytic residues of DsbA enzymes are surrounded by three surface loops (Fig. 4A): the ␤3-H2 connecting loop (L1), the H6-␤4 connecting loop (L2, the well characterized cis-Pro loop), and the ␤5-H7 connecting loop (L3). In E. coli DsbA these three loops create a catalytic surface targeted by substrate and partner proteins (63,66,67). In the present structure, however, the switch I-mediated EF-Tu interaction does not engage the catalytic surface but instead interacts with the solvent-exposed non-catalytic surface of AbDsbA (see below).
The Positively Charged Non-catalytic Surface of AbDsbA Engages EF-Tu-We previously identified two DsbA classes based on the topological arrangements of strand ␤1: class I DsbA (3-2-4 -5-1) and class II DsbA (1-3-2-4-5) (61). AbDsbA falls into the class I topology, with ␤1 forming hydrogen bonds to ␤5, creating a ␤-sheet topology of 3-2-4 -5-1. The structural consequence of the class I DsbA topology (Fig. 4C) is the formation of a large groove between the thioredoxin and helical domains on the non-catalytic surface. This groove is truncated in the class II DsbAs in which ␤1 interacts with ␤3.
The surface shape and electrostatic features of the non-catalytic surface groove are variable among class I DsbA members (Fig. 4, D-G). In AbDsbA, the non-catalytic surface groove has a distinct positive electrostatic potential, and it is this region that is engaged by the EF-Tu switch I segment (Fig. 4G). Analysis of the interactions between the two proteins shows that the EF-Tu switch I segment (residues 47-56) adopts an extended conformation, occupying the groove between H1-to-␤3 and H3 on the non-catalytic surface of AbDsbA (Fig.  5A). The C terminus of switch I (residues 51-56) forms extensive hydrogen bonds and electrostatic interactions with four basic residues (Arg-56, Arg-59, Arg-84, Lys-85) and a polar residue (Gln-45) of AbDsbA (Fig. 5B). Asp-51 at the center of the EF-Tu switch I segment makes contact with five AbDsbA residues in the groove. In addition, EF-Tu residues Phe-47, Gln-49, Ala-53, and Pro-54 form van der Waals contact with Leu-48, Ile-51, His-88, Leu-89, Pro-90, and His-93 of AbDsbA. These polar and hydrophobic interactions bury 40% (683 Å 2 ) of the solvent-accessible surface area of the EF-Tu switch I segment (residues 47-56).
On the basis of this structurally characterized interaction, we designed an 11-residue peptide derived from the E. coli EF-Tu switch I sequence and assessed its binding to AbDsbA using ITC. The binding study showed that the switch I peptide segment ( 46 AFDQIDAPEE 56 ) binds to AbDsbA with an affinity range of 74 -162 nM ( Table 2). Binding of the switch I peptide is unaffected by the redox status of AbDsbA. In contrast, we did not detect binding of this same switch I peptide to EcDsbA under the same conditions.
We also investigated whether EF-Tu from A. baumannii (AbEF-Tu) might interact with AbDsbA. Sequence comparison of switch I of E. coli EF-Tu with that of AbEF-Tu showed that the two sequences are similar but not identical (Table 2). Positions 1 and 3 of EcEF-Tu switch I are not conserved in AbEF-Tu, but neither of these residues interacts with AbDsbA in the crystal structure (Fig. 5). However, Asn at position 7 of EcEF-Tu does interact with Gln-45 of AbDsbA, and this residue is replaced by Ser in AbEF-Tu. Side-chain modeling suggested that substitution of Asn for Ser at position 7 should maintain the hydrogen bond interaction with Gln-45 of AbDsbA. As expected, the switch I peptide of AbEF-Tu did bind to AbDsbA, although with a 3-fold reduced affinity of ϳ500 nM compared with the EcEF-Tu peptide ( Table 2). Removal of the 3 residues at the N terminus of the AbEF-Tu switch I peptide reduced binding by a further ϳ3.5-fold, suggesting that the interaction of Phe/Tyr at position 2 with His93 of AbDsbA is very favorable.
AbDsbA Binds a Peptide Derived from AbDsbB-Dithiol oxidase activity of AbDsbA relies on interaction with the redox cycling enzyme DsbB (Fig. 1, D-E). Structural studies of EcDsbA⅐EcDsbB complex revealed that a transient interaction between the cysteine-containing second periplasmic loop (P2) of the 4-helix bundle protein EcDsbB and the surface surround-ing the catalytic cysteine of EcDsbA is responsible for maintaining EcDsbA in a functionally active oxidized state. The interaction of EcDsbA with EcDsbB is mediated in part by three loops, L1, L2, and L3, that define the catalytic surface. In AbDsbA, the L1 loop comprises five residues and is conformationally similar to that of P. aeruginosa DsbA (Fig. 4A), whereas the L3 loop is much shorter in AbDsbA, formed from just 3 residues (Gln 162 -Gly 163 -Glu 164 ). As a consequence, AbDsbA has small discontinuous pockets (Յ50 Å 3 ) on the catalytic surface between the cis-Pro loop (L2) and L3 as opposed to a large continuous groove found in this region in EcDsbA.
The A. baumannii genome encodes for a protein homologous to EcDsbB (25% sequence identity, Uniprot: B0V9V0). The high secondary structural similarity of the predicted transmembrane helical regions suggested that the putative AbDsbB retains the same four-helix topology as EcDsbB with two loops pointing into the periplasm. In the structure of the EcDsbA⅐EcDsbB complex, Cys-104 from the EcDsbB P2 loop segment ( 98 PSPFATCDFM 107 ) forms a mixed disulfide bond with catalytic residue Cys-30 of EcDsbA, and other residues form hydrophobic interactions and three hydrogen bonds with the cis-Pro loop (Fig. 5C). The corresponding P2 loop of AbDsbB comprises residues 95 PDQVPSCGPG 104 .
Analyzing the crystallographic packing interface of the AbDsbA⅐EcEF-Tu complex showed that residues 262 RKLLD 266 of a neighboring EF-Tu domain 2 interact with the catalytic surface of AbDsbA, placing the EF-Tu Leu-264Ј side chain very close to the catalytic cysteine of AbDsbA (3.8 Å from Cys-34 S␥). Hydrogen bond interactions are formed between EF-Tu Leu-264Ј and Asp-266Ј and the AbDsbA cis-Pro loop in a manner similar to that observed between the EcDsbA cis-Pro loop and the P2 loop of the EcDsbA⅐DsbB crystal structure (Fig. 5D). Modeling the AbDsbB P2 loop segment onto the observed crystal structure conformation of EF-Tu RKLLD residues suggested that 7-8 AbDsbB residues could bind to the catalytic surface of AbDsbA. On the basis of this molecular model, we designed a heptapeptide (PSCGPGL) from the AbDsbB P2 sequence and showed that it bound to AbDsbA with a K D of 7.9 M (Table 2). Similar to a previous study (56), binding of the AbDsbB peptide is cysteine-dependent in that substitution of Cys for Leu in the peptide eliminated binding. Importantly, binding of the AbDsbB P2 peptide to AbDsbA is independent of EF-Tu peptide binding, supporting the notion that EF-Tu and AbDsbB bind to AbDsbA at distinct sites (Table 2).

Multidrug resistant
A. baumannii is emerging as an agent of serious nosocomial and community-acquired infection (68). The current economic loss of preventable hospital-acquired infections ($5.7-6.7 billion) is now comparable to the costs of stroke ($6.7 billion), diabetes mellitus ($4.5 billion), and chronic obstructive lung disease ($4.2 billion) in the United States (69). Anti-virulence therapy is an attractive strategy to combat rapidly spreading infections caused by multidrug-resistant A. baumannii (28,70,71).
Studies from animal models suggested that the absence of DsbA or DsbB attenuates infection and pathogenesis of Gramnegative bacteria (23,24,73). Because DsbA/B is not essential for bacterial survival (51) but is required for virulence (17), DsbA/B is a target for the development of novel anti-virulence agents against Gram-negative bacteria (17). For example, biofilm formation is essential for colonization and infection by A. baumannii (13,74,75). Interference with this process by inhibiting DsbA could impair the ability of bacteria to establish infection.
Our characterization of AbDsbA showed that it is a member of the DsbA Ib sub-class, which also includes DsbAs from Neisseria and Pseudomonas (61). Using ITC, we demonstrated that an AbDsbB-derived peptide binds to AbDsbA via its catalytic cysteine, supporting the idea that DsbA and DsbB form a functional redox pair in A. baumannii. The absence of an interaction between AbDsbA and EcDsbB suggests that the truncated catalytic surface of AbDsbA may engage AbDsbB by a different binding mode to that described for EcDsbB and EcDsbA.
We also reported the detailed interaction between AbDsbA and EF-Tu. EF-Tu is highly conserved (85% sequence identity between AbEF-Tu and EcEF-Tu) and is best known for its catalytic role in the elongation cycle of bacterial protein synthesis. However, EF-Tu also acts as a response protein in abiotic and biotic stresses (76,77), and it has roles in biofilm development (78) and immunospecific immunity (79). EF-Tu is generally considered to be a cytoplasmic protein, but there is evidence for localization on the surface and outer membrane vesicles of bacteria (29,80,81). It is also a relatively ubiquitous interactor with other proteins (actin like MreB (33) and fibronectin (32)). The association between AbDsbA and EF-Tu provides new avenues for understanding the interaction of EF-Tu with other proteins; for example, the Switch I region may interact tightly with cytoplasmic proteins that exhibit similarly strong positive surface TABLE 2 Thermodynamic parameters for peptides derived from EF-Tu or AbDsbB binding to AbDsbA or EeDsbA K D is apparent dissociation constant, ⌬H is enthalpy change, T is absolute temperature (K), ⌬S is entropy change, N is apparent stoichiometry, and n.b. is no detectable binding. ϪT⌬S is calculated from the free energy equation (⌬G ϭ ⌬H Ϫ T⌬S ϭ ϪRT ln K D ). The averaged values from two ITC experiments are tabulated with standard deviation. A. baumannii EF-Tu-derived peptides are differentiated from EcEF-Tu peptides by grey shading. The redox status of AbDsbA used in the titration experiment is identified by: ox, oxidized; red, reduced; mix, mixed.
charge. The switch I region is also the binding site for other proteins including ribosome (82). The PDB entry 4IW3 (yet to be published) shows the EF-Tu switch I region interacts with Pseudomonas putida prolyl-4-hydroxylase, although in a somewhat different conformation. Overall, these data suggest that the switch I region of EF-Tu is a flexible protein interaction loop.
It is tempting to speculate a physiological/pathological role for the association between AbDsbA and EF-Tu that might involve regulation of cell growth in response to nutrient deprivation or oxidative stress in the host environment or chaperoning of EF-Tu by AbDsbA. Both AbDsbA and EF-Tu are important proteins in their own right, and both have been identified as antibacterial targets. Although we showed that AbEF-Tuderived peptides do bind to AbDsbA, we have no direct evidence for an AbDsbA⅐AbEF-Tu interaction in A. baumannii. Consequently, we cannot rule out the possibility that the interaction between EcEF-Tu and AbDsbA is simply an artifact.
Nevertheless, the detailed interactions derived from the AbDsbA⅐EF-Tu structure provides a potential new approach for AbDsbA inhibitor development. The DsbA catalytic surface is widely accepted as the major target surface for substrate/ partner protein interactions (63,66,67). This interaction is essential for formation of intermediate mixed disulfides with substrate proteins. However, the non-catalytic surface may also contribute to the folding of substrates or to the interaction with as-yet unidentified regulatory proteins.
Members of class I DsbAs all have a large charged non-catalytic surface groove (Fig. 4, D-G). In EcDsbA the equivalent acidic surface was identified several decades ago as a potential interaction site (83). The variable molecular landscape and nature of this surface among DsbAs may reflect the extraordinary variability of substrates (61,84). The high resolution structure of the AbDsbA⅐EcEF-Tu complex supports the idea that the pronounced non-catalytic surface groove of class I DsbA enzymes serves as an additional protein interaction site. Whether or not this is the case remains to be elucidated. Nevertheless, the reduced activity of AbDsbA in the presence of bound EcEF-Tu suggests compounds designed to target this non-catalytic site would inhibit AbDsbA enzyme activity.
In summary, we have characterized AbDsbA and identified two peptides that bind to non-overlapping AbDsbA surfaces. These distinct peptides provide the starting point for peptidebased inhibitors targeting AbDsbA activity.