A shape-shifting redox foldase contributes to Proteus mirabilis copper resistance

Copper resistance is a key virulence trait of the uropathogen Proteus mirabilis. Here we show that P. mirabilis ScsC (PmScsC) contributes to this defence mechanism by enabling swarming in the presence of copper. We also demonstrate that PmScsC is a thioredoxin-like disulfide isomerase but, unlike other characterized proteins in this family, it is trimeric. PmScsC trimerization and its active site cysteine are required for wild-type swarming activity in the presence of copper. Moreover, PmScsC exhibits unprecedented motion as a consequence of a shape-shifting motif linking the catalytic and trimerization domains. The linker accesses strand, loop and helical conformations enabling the sampling of an enormous folding landscape by the catalytic domains. Mutation of the shape-shifting motif abolishes disulfide isomerase activity, as does removal of the trimerization domain, showing that both features are essential to foldase function. More broadly, the shape-shifter peptide has the potential for ‘plug and play’ application in protein engineering.

T he Escherichia coli periplasmic Dsb protein family is the best characterized bacterial oxidative folding system. E coli DsbA is a monomeric thioredoxin-fold oxidase 1 that introduces disulfide bonds into protein substrates 2 , whereas E. coli DsbC is a dimeric V-shaped thioredoxin-fold protein disulfide isomerase 3 that proof-reads and shuffles incorrect disulfide bonds 4 . E. coli DsbG is also a dimeric V-shaped thioredoxin-fold protein disulfide isomerase and cysteine reducing system 5 that protects cysteines of periplasmic proteins from inappropriate oxidation 6 .
Here, we characterize a putative DsbA-like protein from P. mirabilis and show it plays a key role in virulence by enabling swarming during copper stress. We find that it is not DsbA-like, as it does not catalyse disulfide bond formation. Rather, it is a disulfide isomerase that shuffles incorrect disulfide bonds in proteins. Further, we show that this thioredoxin fold protein is unlike any other characterized to date-it is trimeric-and we demonstrate that its function depends on a shape-shifting motif that could potentially be used as a 'plug and play' peptide module.

Results
Proteus mirabilis UniProt C2LPE2 is a predicted thioredoxin-fold protein, and the purified protein exhibits redox properties characteristic of the thioredoxin fold family. It is highly oxidizing ( À 108 mV), has an acidic active site cysteine (pK a 3.1; compared to a typical cysteine thiol pKa of B8-9) and the oxidized disulfide form of the active site destabilizes the protein compared with the reduced form ( Supplementary Fig. 1a-c).
The encoded protein plays an important role in P. mirabilis virulence. Copper intoxication is a component of nutritional immunity having a number of detrimental effects on bacterial cells. Copper can react with host-generated hydrogen peroxide to generate free radicals, which damage biological molecules 7 ; cycling between Cu(I) and Cu(II) can lead to detrimental redox reactions in bacteria; and copper can bind to protein thiols in place of other metal co-factors 8,9 . Overcoming these antibacterial effects of copper is a key bacterial defence mechanism. The gene encoding P. mirabilis C2LPE2 is located within a cluster of four predicted suppressor of copper sensitivity (Scs) genes and we proposed this the encoded protein (hereafter called PmScsC) would contribute to bacterial copper resistance. Indeed, inactivation of scsC in two independent P. mirabilis clinical isolates significantly inhibited swarming motility (a key virulence trait of P. mirabilis) in the presence of copper (Fig. 1b, Supplementary Figs 2-4). This phenotype could be complemented by introduction of a plasmid encoding wild-type PmScsC, but not a plasmid encoding an inactive PmScsC (Fig. 1a,b).
Although it is annotated DsbA-like, PmScsC is a powerful protein disulfide isomerase (Fig. 1c) that is able to refold and reactivate the scrambled disulfide form of the model substrate RNase A just as rapidly as the archetypal disulfide isomerase E. coli DsbC. It has negligible DsbA-like dithiol oxidase activity ( Supplementary Fig. 1d).
The best-characterized protein disulfide isomerases-EcDsbC and eukaryotic protein disulfide isomerase (PDI)-each have two thioredoxin-fold catalytic domains. DsbC is a dimer, and PDI is a modular 4-domain protein with two catalytically active thioredoxin domains. The presence of two thioredoxin catalytic domains is thought to contribute to highly efficient disulfide shuffling activity 10,11 . The amino acid sequence of PmScsC encodes a single thioredoxin fold, and we therefore expected that its disulfide isomerase activity would be a consequence of dimerization to generate the necessary two catalytic domains. We also expected that the protein would adopt the same dimeric  architecture as V-shaped EcDsbC. However, we were wrong on both counts. Unexpectedly, evidence from chemical cross-linking ( Supplementary Fig. 5), small angle X-ray solution scattering (Fig. 2a,b) and multi-angle light scattering (Fig. 2c) all indicated that PmScsC is trimeric. We confirmed that PmScsC is a trimer by determining three independent crystal structures. The three crystal structures-which we refer to as compact, transitional and extended ( Fig. 3a-c, Table 1)-reveal an extraordinary range of motion in this protein. Importantly, the compact and transitional structures have multiple protomers in the asymmetric unit that adopt the same overall structures in each case (8 compact trimers, RMSD B1.5 Å for 645 Ca; two transitional trimers, RMSD 0.5 Å for 653 Ca) so that crystal symmetry does not appear to impact on the observed conformation. Moreover, the compact and transitional conformations were determined from crystals grown under similar conditions (2.85 M sodium malonate pH 5.8, 20°C with either 0.1 M cobalt or copper added). Nevertheless, comparison of the eight compact and the two transitional trimers in these two crystal structures, reveals an unprecedented level of conformational re-arrangement (RMSDs420 Å). The third structure was determined from crystals grown in the presence of 2 mM copper chloride (32% Jeffamine M-600, 0.1 M HEPES pH 8, 20°C) and reveals a fully extended trimer with one protomer in the asymmetric unit.
PmScsC also exhibits dynamic conformational flexibility in solution. No single PmScsC crystal structure is consistent with the experimentally determined SAXS scattering curve. However, an ensemble of structures fits the experimental data closely (Fig. 2a,b) supporting the notion that the trimer is highly dynamic in solution. Moreover, the solution scattering curve is unchanged by modification of the pH (range 6.0-8.0) or ionic strength (150-1,500 mM NaCl) ( Supplementary Fig. 6) indicating the dynamic motion is an inherent property of the protein.
We can pinpoint the region responsible for the extraordinary range of motion in this protein, to an 11-residue peptide linking the trimerization and catalytic domains (residues 39-KADEQ-QAQFRQ-49) (Fig. 3). The 31 crystallographically characterized PmScsC protomers have the same thioredoxin-fold catalytic domains (Fig. 4a,b) (RMSD 0.2-1.2 Å for 129 Ca). Similarly, all 31 PmScsC protomers share the same trimeric right-handed coiled coil stalk formed from the N-terminal residues (Fig. 3a-c). At the secondary structure level, it is only the 11-residue linker peptide that changes significantly across the crystal structures ( Fig. 4c, Supplementary Movie 1).
The linker is a shape-shifting peptide that can adopt helical, strand or loop conformations. In the 24 protomers present in the compact PmScsC crystal structure, the linker forms a loop that positions the three catalytic domains close to the trimerization stalk (Figs 3a and 4c). In the extended crystal structure the linker is helical, rotating and translating the catalytic domain away from the trimerization domain relative to the compact structure (Figs 3c and 4c). In the transitional PmScsC crystal structure, each of the two trimers incorporates protomers in three conformations: compact, extended-like and intermediate (Fig. 3b). The two compact protomer conformations of the transitional structure are similar to those in the compact crystal structure (RMSD 0.9-1.8 Å for 188 Ca). The two extended-like protomer conformations of the transitional structure are similar to the protomer in the extended crystal structure though they bend at different points in the linker helix, giving rise to somewhat different catalytic domain placements (Fig. 4c). The two intermediate conformations in the transitional crystal structure have a linker that forms a short b-strand, and positions the catalytic domain directly above the trimerization stalk (Fig. 4c).     As a consequence of the linker peptide flexibility, the three crystal structures describe extraordinarily different PmScsC quaternary shapes and dimensions. The eight compact PmScsC trimers are mushroom-shaped with dimensions 80 Â 80 Â 50 Å 3 (Fig. 3a). The two intermediate PmScsC crystal structures are asymmetric flat triangles with dimensions 105 Â 40 Â 70 Å 3 (Fig. 3b). The extended PmScsC crystal structure is three-leafclover shaped with dimensions 75 Â 80 Â 80 Å 3 (Fig. 3c).
We hypothesized that the dynamic motion suggested by the crystal structures, and supported by solution data, is critical to the mechanism of PmScsC. Indeed, replacement of the shape-shifting peptide with a rigid helical peptide (PmScsC RHP, 39-AEAAAKEAAAKA-50) 12 reduced the flexibility of the protein as evidenced by small angle scattering data ( Supplementary Fig. 7), and abolished activity in the isomerase assay (Fig. 1c). Furthermore, we showed that trimerization of PmScsC is critical for activity because removal of the N-terminal residues (PmScsCDN) also abolishes in vitro disulfide isomerase activity (Fig. 1c) and in vivo complementation assays (swarming in the presence of copper) (Fig. 1b).
Curiously, Nature has also performed the ScsCDN experiment. PmScsC shares 58% sequence identity with Salmonella enterica serovar Typhimurium StScsC 13 ( Supplementary Fig. 8) and both proteins contribute to copper resistance. Whereas we showed above that PmScsC is required for swarming motility under copper stress, StScsC is required for growth of S. Typhimurium in rich media in the presence of copper. By contrast, deletion of scsC in P. mirabilis had no effect on growth in rich media whether copper was present or absent ( Supplementary Fig. 2c). These findings imply that the two proteins have different target substrates and/or different molecular functions. Indeed, StScsC and PmScsC are very different proteins. StScsC lacks the N-terminal residues of PmScsC, it is monomeric rather than trimeric, and it has no isomerase activity. Intriguingly, the catalytic domains of the two proteins (as assessed by crystal structures) are very similar (PDB: 4GXZ) (RMSD 1.0-1.6 Å for 129 Ca). On the basis of this comparison, we predicted that removal of the N-terminal trimerization residues from PmScsC would not only convert the trimer into a monomer and abolish isomerase activity, it might also convert the enzyme into a dithiol oxidase. Indeed, we confirmed that this was the case ( Supplementary Fig. 1d).
These two proteins, trimeric disulfide isomerase PmScsC and monomeric dithiol oxidase StScsC, are encoded in similar loci and are both associated with copper resistance. Yet, they have very different architectures, and different molecular and cellular functions. To examine if the region encoding the N-terminal extension of PmScsC is broadly conserved in P. mirabilis, the scsC gene was PCR amplified from 25 randomly selected clinical isolates and sequenced. We found that the PmscsC sequence was conserved in all 25 isolates-including the region encoding the trimerization domain and shape-shifting motif-except for a single synonymous mutation in the signal sequence. This high degree of conservation suggests a highly conserved functional role. Caulobacter crescentus ScsC also has an extended N-terminal region and is predicted to be dimeric 14 . Altogether with our work, this suggests that the N-terminal region of this protein is necessary for oligomerization and that there are at least two very different molecular and functional classes of ScsC. How does the trimeric and highly dynamic molecular architecture of PmScsC compare with other structurally characterized disulfide isomerases? Until this point, highly efficient disulfide isomerases were thought to function through the presence of two TRX domains (each with a CXXC active site) that can simultaneously interact with two substrate cysteines 15 . The two active sites of V-shaped DsbC and U-shaped PDI are embedded within a dynamic framework that enable shuffling of substrate disulfides [16][17][18][19] (Fig. 5a,b). Trimeric PmScsC also has the potential to interact with multiple substrate cysteines simultaneously. However, the evidence presented here shows a much more expansive range of motion than observed previously for DsbC or PDI, as depicted in morphing movies (Supplementary Movies 2-4) and static structural comparison (Fig. 5c). For example, the distance between active sites of DsbC (PDB: 1JZD-1EEJ) 3,16 or PDI (PDB: 4EKZ-4EL1) 18 in reported crystal structures ranges between 31-39 Å (26% increase) and 28-40 Å (43%), respectively. By comparison, this distance ranges from 16 to 52 Å (225%) in PmScsC.

Discussion
Why is it that other characterized disulfide isomerases do not have the highly flexible, trimeric structure of PmScsC? How does the monomeric S. Typhimurium ScsC-not a disulfide isomerase-confer copper resistance? How do the very diverse ScsC proteins (monomeric, dimeric or trimeric) form redox relay systems with a predicted partner protein ScsB 14 that appears to be highly conserved in bacterial Scs gene clusters? The likely explanation for the diversity in architecture and flexibility of ScsC proteins may relate to differences in their target specificity.
The structural diversity among ScsC homologues may simply reflect the fact that each ScsC homologue has evolved to optimize disulfide reduction and/or isomerization of a distinct protein substrate or set of substrates that are sensitive to copper-induced damage. Identifying and characterizing these specific substrates, and understanding why these are not refolded or reduced by other periplasmic TRX-fold proteins are key questions that will inform our understanding of bacterial copper resistance mechanisms.
In summary, PmScsC is a unique and highly dynamic disulfide isomerase. The N-terminal residues bring three catalytic domains together into a trimer, and a unique flexible linker enables extraordinary twisting and extending motions that impact on the catalytic domain placement. Crystal structure snapshots supported by X-ray scattering data from solution, indicate an almost doubling in length of the protein in one dimension, in concert with considerable catalytic domain rotation. The combination, extent and diversity of these motions would enable mis-folded substrates bound to PmScsC to explore a broad folding landscape, consistent with the redox foldase activity of this key P. mirabilis copper resistance protein. These findings show in much greater detail than ever before how redox proteins with multiple active sites are able to shuffle incorrectly folded substrates. It is unclear whether PmScsC directly interacts with copper or how it interacts with other proteins required for virulence under copper stress. Nevertheless, the structural data may provide a suitable basis for drug discovery targeting a central defence mechanism of an important human pathogen 20,21 . Importantly, the shape-shifting motif may also represent a modular component for dynamic motion that could be used in 'plug and play' protein engineering applications.

Methods
Bacterial strains and growth conditions. BLAST searches of DsbA homologues were performed on the genome of Proteus mirabilis HI4320 (ref. 22). Clinical isolates of P. mirabilis were cultured from the urine or blood of patients with urinary tract infection or bacteremia. The isolates were sourced from the Princess Alexandra Hospital or Sullivan Nicolaides Pathology (Brisbane, Australia). E. coli Top10 (Invitrogen) was used for plasmid manipulations. E. coli and P. mirabilis were cultured in Luria-Bertani (LB) broth (10 g l À 1 tryptone, 5 g l À 1 yeast extract, 10 g l À 1 NaCl) on LB agar (LB medium containing 15 g l À 1 agar). To prevent swarming of P. mirabilis, LB medium containing 30 g l À 1 agar and without NaCl was used. Media were supplemented with chloramphenicol (30 mg ml À 1 ), ampicillin (100 mg ml À 1 ) or kanamycin (25 mg ml À 1 ) as required.
Construction of Proteus mirabilis scsC mutants. Mutation of the scsC gene in P. mirabilis PM38 and PM54 was performed using the TargeTron gene knockout system (Sigma-Aldrich) as per the manufacturer's instructions. Briefly, optimal intron insertion sites and primer sequences (5827, 5 0 -aaaaaagcttataattatc cttaagtatcgaagatgtgcgcccagatagggtg, 5828, 5 0 -cagattgtacaaatgtggtgataacagataagtcgaagatgctaacttacctttctttgt, 5829, 5 0 -tgaacgcaagtttctaatttcgattatacttcgatagagg aaagtgtct) were predicted using the Sigma TargeTron online algorithm, followed by a retargeting PCR and cloning of the amplicon into the pACD4-K-C shuttle vector, resulting in the generation of plasmid pACD4-K-CscsC. Plasmid pAC4K-CscsC and pAR1219 (helper plasmid) were transformed into PM38 and PM54; induction and retargeting of the intron containing the kanamycin-resistance gene cassette into scsC was performed according to the manufacturer's instructions. Transformants with the correct intron insertion were selected by growth in the presence of kanamycin. Mutants with the correct insertion were confirmed by PCR and nucleotide sequencing using primers 5914 (5 0 -atgaaaaaaatttggctggcgttagc) and 5915 (5 0 -ttattttttcatcagtaattgattgacgacatcagaataag), and referred to as PM38scsC and PM54scsC, respectively. The reprogrammed intron was inserted between nucleotide 528 and 529 in the sense strand of scsC. Plasmids pACD4-K-CscsC and pAR1219 were cured by passaging on nonselective medium followed by screening for loss of chloramphenicol and ampicillin resistance. Plasmid pScsC was constructed by amplifying the scsC gene from PM54 using primers PMscsC-pSU2718F (5 0 -gcgtctagattaactattctttcagaggctaaaggagcc) and PMscsCpSU2718R (5 0cgaagcttcgttattttttcactttcgccagttgttctttcac), digestion with XbaI-HindIII and ligation into XbaI-HindIII digested pSU2718. Mutation of the CXXC active site (C82A; plasmid pScsC AXXC) and deletion of residues 1-42 (plasmid pScsCDN) was performed by PCR; all constructs were confirmed by sequencing. These mutants retained the periplasmic signal sequence. Complementation was performed by introducing plasmids pScsC, pScsC AXXC, pScsCDN or pSU2718 (vector control) into PM54scsC.
Swarming motility in the presence of CuSO 4 . Overnight cultures of P. mirabilis were diluted in fresh LB broth to an OD 600 of 0.1. Swarming on LB agar was examined in the absence or presence of CuSO 4 (final concentration of 1.5 mM). For assays performed in the presence of CuSO 4 , a 10 ml sample of the diluted P. mirabilis PM54 and PM54scsC strains, respectively, was inoculated onto the centre of an agar plate and allowed to soak into the agar (resulting in equal diameter for each sample). The plates were incubated at 37°C. The diameter of the swarming zone was measured after 14 h incubation, and every one or two hours thereafter for a total of 10 h. Swarming of PM54scsC containing pSU2718 (vector control) or complementation plasmids (pScsC, pScsC AXXC and pScsCDN) was performed as described above, but with the addition of chloramphenicol (30 mg ml À 1 ) and 1 mM isopropyl b-D-1-thiogalactopyranoside for ScsC induction. Swarming motility at 37°C on LB agar was measured after 6 and 8 h of incubation. Three replicate experiments were performed for all assays. Swarming motility of PM38 wild-type, mutant and complemented strains in the presence and absence of copper was performed in the same manner.
Western blotting. ScsC expression in P. mirabilis wild-type, mutant and complemented strains was examined by western blot analysis. Crude cell lysates were prepared from standardized overnight cultures. Samples were resolved by SDSpolyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto a polyvinylidene difluoride membrane. Blots were probed using a PmScsC specific polyclonal antiserum raised in rabbits (WEHI Antibody Facility) at 1:1,000 dilution in phosphate buffer saline with 0.05% v/v Tween-20. The protein mixture was then desalted using a Sephadex G-25 fine 16/60 column connected to an Ä KTA FPLC system (GE Healthcare) and reverse IMAC (2 Â 2 ml of Talon resin) was used to separate cleaved protein from the His-tagged contaminants. PmScsC and PmScsCDN were reduced or oxidized with a 25-fold molar excess of DTT, or a 10-fold molar excess of copper(II)/1,10-phenanthroline, respectively. The final step of purification for the proteins was size-exclusion chromatography using a Superdex S75 column with 10 mM HEPES pH 7.4, 150 mM NaCl. Proteins were concentrated using Amicon Ultra centrifugal filter devices with a 10-kDa cutoff (Merck Millipore, USA). Yields for PmScsC were B80 mg of purified protein per litre of culture. SDS-PAGE (NuPAGE 4-12% BisTris gel, Invitrogen, Australia) with Coomassie blue stain was used to assess the protein quality. Concentration of protein samples was determined using the A 280 of the sample (read using a Thermo Scientific NanoDrop 2000c spectrophotometer) and calculated extinction coefficients from ProtParam 24 .
E. coli DsbA and E. coli DsbC, lacking the periplasmic leader signal were expressed and purified, as described for PmScsC.
PmScsC sequence conservation. All clinical isolates were de-identified, and individual informed consent was not required. Genomic DNA isolation was performed using the UltraClean Microbial DNA Isolation Kit (MO-BIO Laboratories, #12224) according to the manufacturer's instructions. Yields varied between 100 and 500 ng ml À 1 of DNA. PCR amplification of the scsC gene from P. mirabilis strains was performed using a sequence specific primer pair (5 0 -gtgccgttt aaccagatttatg) and (5 0 -cgtagataaatcagtaagttctg) in combination with Phusion High-Fidelity DNA Polymerase (New England BioLabs Inc., MA, USA). DNA was initially denatured in a single step at 98°C for 30 s followed by 30 cycles of denaturation (98°C, 10 s), annealing (50°C, 20 s) and elongation (72°C, 30 s). In a final step fragment elongation was completed at 72°C for 7 min. PCR samples were examined by electrophoresis and subsequent DNA sequencing.
Chemical cross-linking. In total 150 mM of purified PmScsC (in 50 mM HEPES buffer pH 8.5) was reacted with 3 mM homobifunctional dithiobis(sulfosuccinimidylpropionate) (DTSSP) (Pierce) for different time intervals (30 s-30 min) at 23°C. DTSSP contains one amine-reactive N-hydroxysulfosuccinimide (sulfo-NHS) ester at each end of its eight carbon spacer arm. The sulfo-NHS group reacts preferentially with amines (from lysine residues) between pH ¼ 7 and 9, to form a stable amide bond. DTSSP contains a disulfide bond in the spacer arm, which can be readily cleaved with dithiothreitol (DTT). Aliquots were removed at various time points and mixed with (NH 4 )HCO 3 (50 mM final), to stop the reaction, through hydrolysis of remaining functional ester groups in DTSSP. The samples were then run on a non-reducing SDS-PAGE and subjected to Coomassie staining. As control, 50 mM DTT was added to selected samples, which led to cleavage of the internal disulfide bond in DTSSP, and consequently disruption of cross-links between ScsC protomers.

MALLS.
A combined approach, using analytical size exclusion chromatography (SEC) and multiangle laser light scattering (MALLS) was utilized to determine and compare the stoichiometry of PmScsC with PmScsCDN in solution. The setting consisted of an LC20 high-performance liquid chromatography (HPLC) system (Shimadzu, Rydalmere, Australia) and a DAWN HELEOS II laser light detector connected to an Optilab T-rEX refractive index detector (Wyatt Technology, Dernbach, Germany). Either a Superdex 200 or Superdex 75 10/300 GL analytical column (GE Healthcare, USA) was connected to the LC20 HPLC system and equilibrated with 25 mM Tris and 150 mM NaCl, pH 7.5 overnight. Purified proteins were injected (500 ml of 3 mg ml À 1 at a flow rate of 1.0 ml min À 1 ) into the SEC-MALLS system for analysis. To calibrate the detector, 500 ml of 5 mg ml À 1 bovine serum albumin (BSA) (Sigma-Aldrich, Australia) was used in 25 mM Tris and 150 mM NaCl, pH 7.5 at a flow rate of 1.0 ml min À 1 . Wyatt Astra V software was used for data collection and analysis.

SAXS.
Small angle X-ray scattering data were collected on the SAXS-WAXS beamline at the Australian Synchrotron 26 . Immediately before loading, all samples were centrifuged at 10,000 g to remove large particles from the solution, and radiation damage was minimized by flowing samples (B100 ml) past the beam in 1.5 mm quartz capillaries (Hampton Research). Data reduction was carried out using the Australian Synchrotron Scatterbrain software 27 correcting for sample transmission and solvent scattering.
Scattering data were collected on reduced PmScsC wildtype and RHP mutant proteins in 25 mM HEPES pH 7.5, 150 mM NaCl, 1 mM DTT. Data were also collected on oxidized wildtype PmScsC, but as no significant differences in the scattering were observed, only data from the reduced forms are presented. To confirm that the pH and ionic strength of the crystallization conditions do not induce conformational changes in PmScsC, scattering data were collected from wildtype PmScsC prepared at 0.50 mg ml À 1 in a gradient of 25 mM HEPES pH (6.0, 6.5, 7.0, 7.5 and 8.0) and NaCl (150, 300, 600, 900 and 1,500 mM). No significant systematic change in the structural parameters was observed at any point of the gradient, hence, it was concluded that neither ionic strength nor pH influence the conformation of the complex in the ranges measured.
Data quality was assessed by inspection of the linearity of the Guinier region of the data (qR g o1.3), estimated molecular mass of the protein complex, and concentration dependence of the scattering. The estimated molecular mass was determined as outlined in 28 , where the contrast and partial specific volume were estimated from the protein sequence 29 . The pair-distance distribution function (p(r)) was generated from the experimental data using GNOM 30 from which I(0), R g and D max were determined. Rigid body modelling of the scattering data from PmScsC wildtype and mutant proteins was performed using CORAL 31 . C 3 symmetry was assumed, and the starting model was oriented such the 3-fold axis was parallel with the z-axis, and passed through the centre of the oligomization domain. Two rigid bodies were then defined for each monomer: residues 3-44 (oligomerization domain); 47-224 (catalytic domain). The position of the oligomerization domain was fixed, and the position and orientation of the catalytic domain was then optimized against the measured scattering data. The program was run 16 times for each protein, but the models with the lowest penalty function still showed small systematic deviations from the experimental data (Wildtype: w 2 ¼ 3.4; CorMap test 32 , 182 points, C ¼ 51, P ¼ 0.000; RHP mutant: w 2 ¼ 5.6; CorMap test, 171 points, C ¼ 28, P ¼ 0.000). Given the crystal structures show NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16065 ARTICLE significant structural diversity, it is likely that these systematic deviations arise from the fact that there is an ensemble of structures present in solution. Hence, ensemble optimization was also performed with the program EOM 33 . From an initial pool of 1,000 structures (the oligomerization domain possessed C 3 symmetry in all structures, but the entire trimer was permitted to adopt either C 1 or C 3 symmetry), a final ensemble of 6 structures was obtained that yielded an excellent fit to the data (Wildtype: w 2 ¼ 1.0; CorMap test, 182 points, C ¼ 12, P ¼ 0.021; Mutant: w 2 ¼ 1.8; CorMap test, 171 points, C ¼ 11, P ¼ 0.076). The ensemble for the wildtype protein is composed of six structures and spans a diverse range of conformations. Thus, the SAXS data supports the notion that the PmScsC trimer is highly dynamic in solution. The ensemble for the mutant protein is composed of four structures, all in extended conformations. The initial pool of random structures was the same for both WT and RHP optimizations, and the number of structures in the final ensemble was not artificially limited. The reduction in size of the ensemble (from 6 for WT to 4 for RHP) is consistent with the RHP mutation rigidifying the flexible helix, and reducing the conformational space sampled by the protein. Details of the data collection and structural parameters are summarized in Supplementary  Table 1.
Crystallization and structure determination. Crystallization screenings were performed at the UQ ROCX facility at the University of Queensland (uqrocx.imb.uq.edu.au) using commercial screens and the hanging-drop vapour diffusion method.
Compact structure. Selenomethionine-labelled PmScsC crystals were grown at 20°C from a drop comprising 200 nl of 30 mg ml À 1 purified protein in 10 mM HEPES pH 7.4 and 200 nl of well solution, 2.85 M sodium malonate pH 5.8 containing 0.1 M cobalt(II)chloride hexahydrate. Crystals grew over a period of 2-3 days. A data set at the Se-edge (wavelength ¼ 0.9792 Å) was measured under cryogenic conditions (100 K) at the Australian Synchrotron MX2 beamline using the BluIce interface 34 . Data were processed in XDS 35 and scaled to space group P2 1 2 1 2 in AIMLESS 36 . A manually guided molecular replacement approach was performed using StScsC as the template (PDB ID: 4GXZ) in MOLREP 37 . Manual selection of solutions based on observed packing, allowed placing of six molecules in the asymmetric unit. This solution was improved in the subsequent MR runs in MOLREP 37 , allowing the addition of eight molecules in the second trial and a total of nine molecules in the third run. A SAD-MR approach in PHENIX 38 , by combining SAD phasing and the partial MR solution (nine molecules), and subsequent density modifications resulted in an electron density map with a skew value of 0.82 (1,826 residues (190 side chain built) in 250 fragments). Successive AutoBuild runs in PHENIX 38 built 1,557 residues (806 side chain) and dropped the R and Rfree values to 38% and 41.5%, respectively. Extensive manual building was required to complete one copy of PmScsC. Molecular replacement using this model as template in PHASER 39 allowed placing of 12 molecules in the asymmetric unit. After several rounds of manual building in Coot 40 and PHENIX 38 refinements R/ Rfree values dropped to 29.1 and 32.3%, but further refinement cycles were unsuccessful. Careful examination of Ctruncate 41 indicated the presence of nearly perfect pseudo-merohedral twinning (-h,-k,l) and led to a reconsideration of space group. Reprocessing the data set to a resolution of 2.6 Å in space group P2 1 yielded lower R-merge (0.072 versus 0.097 for data processed in P2 1 2 1 2). A new molecular replacement run using the reprocessed data in P21 space group and the PmScsC template with a resolution cutoff set to 3.2 Å finally identified 24 copies in PHASER 39 . Rigid body refinement in REFMAC 42 showed that refinement considering the twin operator (-h,-k,l) produced better R-factors (37 versus 42%, no twinning). After manual correction of two chains that were placed inaccurately during molecular replacement by PHASER 39 , and several rounds of refinement in PHENIX 38 and REFMAC 42 the final R/Rfree values are 24.8/28.2% at 2.6 Å resolution. The Ramachandran favoured/outlier statistics for the structure are 95%/0.9%. A stereo image of representative electron density for this structure is show in Supplementary Fig. 9a.
Transitional structure. Crystals were grown at 20°C from drops containing 1 ml of 48 mg ml À 1 oxidized PmScsC in 10 mM HEPES pH 7.4 and 1 ml of 2.85 M sodium malonate pH 5.8 containing 0.1 M copper (II) chloride. Crystals formed after 2-3 days and were cryoprotected with 3.4 M sodium malonate pH 5.8 and flash frozen in liquid nitrogen. Data were collected using the BluIce software 34 on the MX2 beamline of the Australian Synchrotron at a wavelength of 0.9792 Å under cryogenic conditions (100 K). The data were processed with XDS 35 and scaled in AIMLESS 36 using the autoPROC framework 43 . The space group was determined to be I4 and molecular replacement using PHASER 39 , as implemented in the CCP4 suite 44 , was performed with a single catalytic domain from the compact structure (residues 46-224). Five molecules were initially found in the asymmetric unit and after assessing the crystal packing one of these was deleted and a subsequent round of PHASER 39 was run with the altered solution as an additional input model. This resulted in the placement of 6 molecules in the asymmetric unit, which is consistent with the presence of two trimers. Manual building of the trimerization domain was performed in Coot and rounds of autobuild using Buccaneer 45 in CCP4 and refinement using REFMAC 42 and BUSTER 46,47 were performed to improve the structure. Several rounds of manual adjustment in Coot 40 and refinement in PHENIX 38 were performed to yield final R/Rfree values of 17.1 and 22.2%. Validation in MolProbity 48 was used throughout the refinement process, and the final Ramachandran favoured/outlier statistics for the structure are 98%/0.5%. A stereo image of representative electron density for this structure is show in Supplementary Fig. 9b.
Extended structure. Crystals were grown at 20°C from drops containing 1 ml of 20 mg ml À 1 oxidized PmScsC in 10 mM HEPES pH 7.4, 150 mM NaCl and 2.5 mM copper (II) chloride (added to protein solution immediately before crystallization) and 1 ml of 32% Jeffamine M-600 pH 7 in 0.1 M HEPES pH 8. Crystal formation occurred within minutes of set up and crystals continued to grow until day two. Crystals were cryoprotected with 32% Jeffamine M-600 pH 7, 0.1 M HEPES pH 8 and 20% ethylene glycol and frozen in liquid nitrogen. Data were collected using the BluIce software 34 on the MX2 beamline of the Australian Synchrotron at a wavelength of 0.9537 Å under cryogenic conditions (100 K). Using the autoPROC framework 43 the data were processed with XDS 35 and scaled in AIMLESS 36 . The Wilson B factor for the data was relatively high (Bdouble that of the data for the other two crystal structures). The space group was determined to be H3 2 and molecular replacement using PHASER 39 , as implemented in the CCP4 suite 44 , was performed with a single catalytic domain from the compact structure (residues 46-224). Only one molecule was found in the asymmetric unit (the trimer is generated by crystal symmetry). Manual building of the trimerization domain was performed in Coot with rounds of autobuild (Buccaneer 45 ) and refinement (BUSTER 47 ) performed to improve the structure. Several rounds of manual adjustment in Coot 40 and refinement in PHENIX 38 , including the use of TLS, were performed to yield final R/Rfree values of 25.1 and 26.3%. Validation in MolProbity 48 was used throughout the refinement process and the final Ramachandran favoured/outlier statistics for the structure are 98%/0.5%. A stereo image of representative electron density for this structure is show in Supplementary  Fig. 9c.
Residues in all three crystal structures were numbered based on their position after the TEV cleavage site in the construct. MUSTANG-MR 49,50 was used for structure alignment and RMSD calculations and PyMOL was used to create images of the structures and perform other measurements. The distance range between active sites in DsbC, PDI and PmScsC structures was measured using the Cb positions of the more N-terminal of the two catalytic cysteines. Supplementary Movie 1 was created in Pymol by generating morphs between each of the PmScsC crystal structures, as well as morphs between each of the available structures for E. coli DsbC (PDB: 1JZD-1EEJ) and human PDI (PDB: 4EKZ-4EL1).
Measurement of PmScsC redox potential and pK a values. The redox potential of PmScsC was determined by incubating 2 mM PmScsC with 1 mM GSSG in combination with varying concentrations of GSH (4-15 mM) for B12 h at 24°C in 100 mM sodium phosphate buffer, pH 7.0 and 1 mm EDTA. Trichloroacetic acid, 10% v/v, was used to precipitate the protein samples and the pellets were washed with ice-cold acetone. The samples were then treated with 2 mM 4-acetamido-4 0maleimidylstilbene-2,2 0 -disulfonic acid, in 50 mM Tris-HCl, pH 8.0, and 1% SDS, which blocks free thiols and increases the weight of the reduced protein by B1 kDa. The reduced and oxidized forms of PmScsC were separated on a 12% non-reducing SDS-PAGE (NuPAGE) and stained with Coomassie Brilliant Blue. The fraction of reduced protein, R, was determined via densitometric analysis of the stained SDS-PAGE gel using ImageJ 51 , and the equilibrium constant K eq was calculated via R ¼ ([GSH] 2 /[GSSH])/(K eq þ ([GSH] 2 /[GSSH])) 52 . The Nernst equation, E 0 ¼ E 0 GSH/GSSG -(RT/nF) Á lnK eq , where E 0 GSH/GSSG is the standard potential of À 240 mV (ref. 53), was used to calculate the redox potential of PmScsC.
The pH-dependent absorbance of the catalytic thiolate anion of PmScsC was determined at 240 nm (ref. 54) using a CARY 50 UV/VIS spectrophotometer (Agilent Technologies, USA). The pH titration measurements of oxidized or reduced PmScsC (40 mM) were conducted at 22°C in 2 ml reaction 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). Absorbance (l ¼ 240 and 280 nm) was measured between pH 6.5 and 1.5 in 0.25 increments. The pK a value was calculated from the fitted curves of three replicates using the Henderson-Hasselbalch equation (pH ¼ pK a À log ((A240/A280) red /(A240/A280) oxid )). Redox potential and pK a experiments were each repeated three times.
Relative stability of oxidized and reduced PmScsCDN. Unfolding of PmScsCDN was monitored using the change in the far-ultraviolet circular dichroism (CD) signal 55 . The largest difference in molar ellipticity for oxidized and reduced enzymes was calculated from initial far-ultraviolet CD spectra (from 250 to 190 nm) recorded at 25 and 95°C using a Jasco J-810 circular dichroism (CD) spectropolarimeter (Jasco, USA), respectively. The unfolding of oxidized and reduced protein (PmScsCDN ox ¼ 222.5 nm, PmScsCDN red ¼ 219.5 nm) was monitored at a constant heat rate of 1°C min À 1 starting from 25°C and increasing to 95°C in a 1 mm quartz cuvette. Unfolding reactions were performed using 10 mM protein in 100 mM NaH 2 PO 4 /Na 2 HPO 4 , 1 mM EDTA at pH 7.0. Reduced enzyme samples contained 0.75 mM DTT. The CD data were converted to a fraction of folded protein via a folded (T) ¼ (y(T) À y unfolded )/(y folded À y unfolded ) and a folded (T) was then fitted using a Boltzmann sigmoid function 1-1/(1 þ e N(Tm À T) ), where N is a scaling factor that describes the nature of the transition and T m is the protein melting temperature. The thermal unfolding experiment was repeated three times for both the oxidized and reduced protein.
Protein disulfide isomerase assay. The bovine pancreatic ribonuclease RNase A contains four disulfide bonds that are necessary for its enzymatic function. Starting with 'scrambled' RNase A (in which the cysteines have formed non-native disulfide bonds), one can determine the ability of other enzymes to restore its function by spectroscopically monitoring the hydrolysis of 3 0 ,-5 0 -cyclic monophosphate (cCMP), a synthetic substrate of the ribonuclease. Thus, in vitro disulfide isomerase activity of PmScsC was monitored by using scrambled RNase A as substrate 56 . Inactive scrambled RNase A was produced by treating 70 mg of native RNase A (Sigma-Aldrich) with 6 M guanidinium chloride and 150 mM DTT in 50 mM Tris pH 8, incubated at room temperature for 16 h. After adjusting the solution to pH 8, the unfolded protein was then desalted and buffer exchanged into 100 mM acetic acid/NaOH pH 4. The presence of reduced cysteines was confirmed spectrophotometrically by mixing an unfolded RNase A sample with eight-fold higher molar concentration of 5,5-dithio-bis-(2-nitrobenzoic acid) (DTNB or Ellman's reagent), which reacts with thiols to absorb at 412 nm. The unfolded RNase A was then air-oxidized over 5 days in 6 M guanidinium chloride, 50 mM Tris pH 8.5 at room temperature in the dark. After concentration of the protein and readjustment of the pH to 8.5, the scrambled RNase A was buffer exchanged into 100 mM acetic acid/NaOH pH 4 and concentrated to 11.5 mg ml À 1 for use in assays. The absence of reduced cysteines was confirmed, as described above. The total reaction volume for the scrambled RNase A assay was 750 ml containing 100 mM sodium phosphate buffer pH 7.0, 1 mM EDTA, 8.2 mM DTT and 10 mM PmScsC, PmScsCDN, PmScsC RHP, EcDsbC or EcDsbA and 40 mM of scrambled RNase A. At various time intervals, 50 ml sample was removed and mixed with 150 ml of 3 mM cytidine 3 0 ,-5 0 -cyclic monophosphate (cCMP) to measure RNase A activity. Hydrolysis of the cyclic phosphate ester bond in cCMP was monitored at 296 nm using a Synergy H1 multimode plate reader (BioTek, USA). Samples containing native RNase A and scrambled RNase A in the absence of any other enzyme served as positive and negative controls, respectively. The disulfide isomerase assays were repeated twice for each protein.
PmScsC dithiol oxidation activity. The potential of PmScsC/PmScsCDN to catalyse disulfide bond formation in vitro was assessed using a synthetic model peptide (CQQGFDGTQNSCK), as described before 57 . Dithiol oxidation was determined fluorometrically, using a Synergy H1 multimode plate reader (BioTek Instruments, USA) with the excitation wavelength set to 340 nm and emission to 615 nm. A 150 ms delay before reading and 100 ms reading time were used for timeresolved fluorescence. The reaction buffer contained 50 mM MES, 50 mM NaCl and 2 mM EDTA at pH 5.5. The reaction volume was 50 ml in each well of a white 384-well plate (Perkin Elmer OptiPlate-384, Part #: 6007290), including increasing concentrations (80-320 nM) of EcDsbA, EcDsbC, PmScsC or PmScsCDN and 2 mM glutathione (GSSG) as oxidant. At last, 8 mM peptide substrate was added to initiate the reaction. EcDsbA served as positive control. As negative control, reactions in the absence of proteins were used. The initial linear portion of the raw data was used to calculate the rate of oxidation as fluorescence increase per minute. The dithiol oxidation activity experiments were repeated three times for each protein.
Data availability. The UniProt accession code for the P. mirabilus ScsC protein described in this paper is C2LPE2 and the accession codes for other proteins described in this study are H9L4C1 (StScsC), P0AEG4 (EcDsbA) and P0AEG6 (EcDsbC). Coordinates and structure factors for the three ScsC crystal structures were deposited in the protein data bank (PDB) with accession codes 4XVW (compact), 5IDR (transitional) and 5ID4 (extended). In addition to the PDB accession codes listed above, PDB accession code 4GXZ (StScsC) was used as a molecular replacement template and 1EEJ and 1JZD (EcDsbC), and 4EKZ and 4EL1 (hPDI) were used to generate Fig. 5. Scattering data and models have been deposited in SASBDB 58 with accession codes SASDB94 (wildtype PmScsC) and SASDBW6 (RHP mutant). All other data are available from the corresponding authors upon reasonable request.