Disulfide isomerase activity of the dynamic, trimeric Proteus mirabilis ScsC protein is primed by the tandem immunoglobulin-fold domain of ScsB

Correct disulfide bond formation is essential for proper folding of many proteins, including bacterial virulence factors. The suppressor of copper sensitivity (Scs) proteins have roles in dithiol/disulfide interchange and the bacterial response to copper stress. Encoded in a four-gene cassette (ScsABCD) present in many Gram-negative bacteria, the Scs proteins are enigmatic and poorly characterized. Here, we show that the periplasmic α-domain of the membrane protein ScsB in the Gram-negative bacterium Proteus mirabilis forms a redox relay with the soluble periplasmic protein PmScsC. We also found that the periplasmic α-domain is sufficient to activate the disulfide isomerase activity of PmScsC. The crystal structure of PmScsBα at a resolution of 1.54 Å revealed that it comprises two structurally similar immunoglobulin-like folds, one of which includes a putative redox-active site with the sequence CXXXC. We confirmed the importance of these cysteine residues for PmScsBα function, and in addition, we engineered cysteine variants that produced a stable complex between PmScsC and PmScsBα. Using small-angle X-ray and neutron scattering analyses with contrast variation, we determined a low-resolution structure of the PmScsC–PmScsBα complex. The structural model of this complex suggested that PmScsBα uses both of its immunoglobulin-like folds to interact with PmScsC and revealed that the highly dynamic PmScsC becomes ordered upon PmScsBα binding. These findings add to our understanding of the poorly characterized Scs proteins.

The correct formation of disulfide bonds is an essential component in the folding of many proteins, including bacterial virulence factors. In bacteria, disulfide bond-forming (Dsb) 5 proteins are responsible for introducing disulfide bonds into substrate proteins (DsbA and -B) (1, 2) as well as reducing and isomerizing (proofreading and shuffling) disulfide bonds that have been incorrectly introduced (DsbC and -D) (3,4). Dsb proteins function through the redox action of two catalytic cysteines that are often embedded in a thioredoxin fold. These systems are well characterized in Escherichia coli; however, homologues of Dsb proteins exist in a range of Gram-negative bacteria. A group of related and poorly studied proteins found most commonly in Proteobacteria are the suppressor of copper sensitivity (Scs) proteins (5). The Scs proteins contribute to the bacterial virulence trait of resistance to copper stress (6,7).
Originally identified in Salmonella enterica serovar Typhimurium (6), the scs locus encodes four proteins, ScsA-D, all of which have predicted catalytic motifs consisting of two cysteines (CXXC or CXXXC). Three of these (ScsB-D) incorporate a predicted thioredoxin fold, and one of the four (ScsA) has a predicted copper-binding motif. ScsD has a predicted N-terminal membrane anchor linked to a predicted periplasmic thioredoxin-fold domain (6) reminiscent of proteins involved in cytochrome c biogenesis, such as DsbE/CcmG (8,9). The best studied of the four Scs proteins is the soluble periplasmic protein ScsC (5,7,10). An extraordinary feature is the variation in structure and function of ScsC proteins across different bacteria. Caulobacter crescentus ScsC (CcScsC) is reported to be a dimeric disulfide isomerase (5), S. enterica serovar Typhimurium ScsC (StScsC) is a monomeric redox protein with no disulfide isomerase activity (7), and Proteus mirabilis ScsC (PmScsC) is a highly dynamic "shape-shifting" trimeric disulfide isomerase (10).
The present work focuses on the structure and function of P. mirabilis ScsB, a putative redox partner of P. mirabilis ScsC. ScsB resembles E. coli DsbD in many respects (5). EcDsbD is the membrane protein partner of the archetypal disulfide isomerase E. coli DsbC. Like EcDsbD, the ScsB protein is predicted to comprise three domains: an N-terminal periplasmic domain (DsbD␣/ScsB␣), a transmembrane domain (DsbD␤/ScsB␤) of eight helices, and a periplasmic C-terminal thioredoxin-fold domain (DsbD␥/ScsB␥) (Fig. 1, A and B). Each of the three domains in both EcDsbD and ScsB have two catalytic cysteine residues. In DsbD, these cysteine pairs in each domain form a reduction cascade that originates from thioredoxin reductase and NADPH in the cytoplasm, is passed through thioredoxin to DsbD␤, DsbD␥, and then DsbD␣, which reduces the cysteines in specific substrate proteins, such as DsbC (11,12) (Fig. 1A). The similarity between the predicted architectures of EcDsbD and ScsB suggests the ScsB protein forms a similar redox relay system. Indeed, C. crescentus ScsB has been shown to interact with CcScsC and to maintain CcScsC in its reduced state, which is necessary for disulfide isomerase activity (5). A major difference between EcDsbD and ScsB is the size of their N-terminal ␣-domains. DsbD␣ comprises ϳ165 residues, whereas ScsB␣ is much larger at ϳ255 residues, and there is no detectable sequence identity between the ␣-domains of DsbD and ScsB. Because the ␣-domain of EcDsbD interacts directly with substrates, such as EcDsbC, the different size and sequence of the ScsB ␣-domain suggests that the proteins interact with different substrates (5).
In the present study, we investigated the structure and function of P. mirabilis ScsB␣, showing that it forms a functional redox relay with the highly dynamic, trimeric disulfide isomerase PmScsC. We also report the first crystal structure of any ScsB␣, revealing unexpectedly the presence of tandem immu-noglobulin folds in PmScsB␣ and showing that this same arrangement is likely to be shared in ScsBs encoded by other organisms. We created a stable PmScsC-PmScsB␣ (trimermonomer) complex and analyzed small-angle X-ray scattering (SAXS) data and small-angle neutron scattering (SANS) data with contrast variation to produce a low-resolution model of the complex. This model revealed that the highly dynamic PmScsC protein becomes more symmetric and ordered upon interaction with PmScsB␣ and that both immunoglobulin folds of PmScsB␣ interact with one protomer of PmScsC.

PmScsC and PmScsB␣ form a specific redox relay system
From sequence analogy with CcScsB and CcScsC, PmScsB␣ is predicted to be the redox partner of PmScsC. If this functional relationship holds true, PmScsB␣ would specifically reduce oxidized PmScsC to convert it to the active reduced form. To confirm that a redox relay occurs between PmScsC and PmScsB␣, we performed a redox gel-shift assay. First, we mixed PmScsC with PmScsB␣ and then analyzed the redox states of the two proteins after an incubation period. The redox state was defined by analyzing the proteins by SDS-PAGE after reaction with 4-acetamido-4Ј-maleimidylstilbene-2,2Ј-disulfonic acid (AMS) ( Fig. 2A). When AMS binds to a reduced thiol, the mass is increased by 0.5 kDa, thus allowing discrimination between reduced and oxidized cysteine forms of a protein on mass-based gel separation. The results showed that when oxidized PmScsC was incubated with reduced PmScsB␣, both proteins changed their redox states, confirming that disulfide exchange occurs between the two proteins (Fig. 2B). The reverse redox reaction did not occur: reduced PmScsC remained reduced when mixed with oxidized PmScsB␣. Moreover, we were able to confirm the residues involved in the change in redox state because addition of the variant PmScsB␣ C114A to oxidized PmScsC did not catalyze disulfide exchange. However, there was some evidence for formation of a PmScsC-PmScsB␣ C114A complex (Fig. 2B, band marked X).
Control experiments using PmScsC or PmScsB␣ alone showed that, under the conditions of the experiment, these proteins did not spontaneously change their redox state. These results support the notion that PmScsC and PmScsB␣ form a specific redox relay system (Fig. 2B) and that the residues Cys 114 and Cys 118 of PmScsB␣ are responsible for the redox interaction.

PmScsB␣ primes the disulfide isomerase activity of PmScsC
We have previously shown that PmScsC is a disulfide isomerase (10). We tested the hypothesis that the redox interaction between PmScsB␣ and PmScsC might contribute to this function of PmScsC. We measured the disulfide isomerase activity of PmScsC using scrambled RNase A as substrate ( Fig. 2A). Reduced PmScsC, but not oxidized PmScsC, was able to refold scrambled RNase A and thereby activate the enzyme (10) (Fig.  2C). PmScsB␣ on its own (oxidized or reduced) had no effect on the activity of scrambled RNase A. However, in the presence of reduced PmScsB␣, the inactive oxidized PmScsC was able to restore the activity of RNase A. This result supports the hypothesis that PmScsB␣ reduces the disulfides of oxidized PmScsC, Figure 1. Schematic representation of bacterial disulfide isomerases and their redox relay partners. A, EcDsbC is reduced by EcDsbD␣, which obtains its reducing power from a reduction cascade that starts with thioredoxin reductase in the bacterial cytoplasm. B, PmScsC and PmScsB␣ are putative interaction partners, and the source of PmScsB's reducing power is yet to be determined (represented by the question marks). The catalytic cysteines in each protein are shown as orange spheres.

P. mirabilis ScsB␣ and ScsC form a functional redox relay
thereby enabling its disulfide isomerase activity (Fig. 2D). This experiment also confirmed that disulfide exchange in this system does not flow in the reverse direction: reduced PmScsC did not lose its ability to shuffle scrambled RNase A disulfides when it is incubated with oxidized PmScsB␣. This may suggest that reduced PmScsC is unable to interact with PmScsB␣. Furthermore, the cysteine variant in which Cys 114 is mutated to Ala (PmScsB␣ C114A) does not activate oxidized PmScsC, showing that this cysteine is required for the redox relay to operate.

Crystal structure of PmScsB␣
After removal of the periplasmic signal sequence, the predicted PmScsB ␣-domain comprises 255 residues (residues 21-275 of the reported PmScsB protein sequence, UniProt entry B4EV20). The crystal structure of this domain was solved to a resolution of 1.54 Å by using selenomethionine (SeMet)labeled protein together with SAD phasing (Protein Data Bank code 6C29; Table 1 and Fig. 3, A-D). The structure revealed a monomer consisting of two immunoglobulin-like folds linked by an ␣-helix. This arrangement was consistent across all three molecules in the asymmetric unit (r.m.s.d., 0.7-1.0 Å, 247 residues aligned). Curiously, the two subdomains, A and B, are reasonably similar in structure ( Fig. 3D; r.m.s.d., 2.5 Å, 81 residues aligned), although the sequence identity for such an alignment is very low (ϳ7%). The subdomains each comprise two ␤-sheets, one with three ␤-strands and the other with four ␤-strands (Fig. 3, A and B). Subdomain A has additional features, including the cysteine motif ( 114 CXXXC 118 ), two ␤-strands forming a ␤-hairpin that links the two ␤-sheets of the canonical immunoglobulin fold, and a loop between ␤2 and ␤3 located on one side of the catalytic cysteines (Fig. 3D, arrows). The cysteines are present in the reduced form in all three molecules in the asymmetric unit.
The presence of an immunoglobulin-like fold in both PmScsB␣ subdomains resulted in many hits in a DALI search against the Protein Data Bank (13). Notably, the top hit was the   (Fig. 4, B and C). These proteins have typical immunoglobulinlike folds with no catalytic cysteines, and like EcDsbD␣ they have one immunoglobulin fold that aligns with subdomain A of PmScsB␣. These three proteins were also hits in a DALI search against PmScsB␣ subdomain B although with much higher r.m.s.d. values (Protein Data Bank code 1JPE: r.m.s.d., 3.8 Å, 75 residues aligned; Protein Data Bank code 2A7B: r.m.s.d., 3.0 Å, 84 residues aligned; Protein Data Bank code 3MNM: r.m.s.d., 2.9 Å, 86 residues aligned). Curiously, there were no high-scoring DALI hits that had a tandem immunoglobulin-fold arrangement like that of PmScsB␣. We did find lower-scoring DALI hits with two immunoglobulin-like folds (mostly antigen-binding fragments of antibodies; e.g. Protein Data Bank code 5JUE; Z-score, 4.6; r.m.s.d., 3.0 Å, 80 residues aligned), but they lacked the connecting ␣-helix of PmScsB␣.

Structure-based sequence alignment
The amino acid sequence of PmScsB␣ was aligned with that of EcDsbD␣ using structure-based alignment and the sequences of CcScsB␣ and StScsB␣ (Fig. 5). ScsB␣ from S. enterica Typhimurium and C. crescentus were chosen as these are the only other organisms where the interaction partner, ScsC, has been characterized. EcDsbD␣ was chosen because of the known structural and functional relationship. PmScsB␣ shares 44% sequence identity with StScsB␣ (280 residues), 21% identity with CcScsB␣ (291 residues), and just 12% The wire mesh represents the likelihood-weighted F o Ϫ F c electron density difference map, which is contoured to 3 and was generated using phenix.refine with the region of interest omitted from the structure. D, superimposition of subdomain A and B structures showing the structural similarity between the domains. Arrows indicate the inserted ␤-hairpin structure and the extended loop of subdomain A. In all panels, subdomain A is colored purple, and subdomain B is magenta with the linking ␣-helix shown in gray. The catalytic cysteines are represented as orange spheres. The N and C termini are labeled in A and B.

P. mirabilis ScsB␣ and ScsC form a functional redox relay
identity with EcDsbD␣ (116 residues of Protein Data Bank code 1JPE). There are several conserved residues across the three ScsB␣ and EcDsbD␣ protein sequences aside from the catalytic cysteines. Other conserved residues include Leu 34 and Ile 58 (PmScsB␣ numbering); their side chains interact with each other in the hydrophobic core of both PmScsB␣ and EcDsbD␣ so their conservation suggests a key role in folding or stability. Interestingly, the tyrosine hydroxyl of the conserved residue Tyr 45 (Tyr 42 in EcDsbD␣) is within 3.5-3.8 Å of the active-site sulfurs of Cys 114 and Cys 118 , suggesting a potential role in catalysis or substrate binding. Two other conserved residues, Tyr 87 and Pro 72 , are located at either end of the ␤-hairpin. The hydroxyl group of Tyr 87 is within hydrogenbonding distance of the main-chain oxygen of Pro 72 (ϳ2.6 Å), suggesting that their conservation may be important in preserving the structure of the ␤-hairpin. The ␤2-␤3 loop sequence is also highly conserved across the ScsB␣ sequences, suggesting an important function.

Formation of a 3:1 PmScsC-PmScsB␣ complex
The interaction between the catalytic cysteines of PmScsC and PmScsB␣ is transient. For this reason, the catalytic cysteine variants PmScsC C87S and PmScsB␣ C114A were produced. These variants allowed the capture of a stable PmScsC-PmScsB␣ complex for structural studies. Size exclusion chromatography (SEC) and subsequent SDS-PAGE analysis suggested that the predominant complex species had a 3:1 ratio of PmScsC to PmScsB␣, corresponding to one PmScsC trimer bound to one PmScsB␣ monomer (Fig. 6). The predicted molecular mass calculated from SAXS and SANS analysis ( Table 2) also provided evidence to support this stoichiometry. A peak corresponding to a larger species was seen on the SEC chromatogram ( Fig. 6A, peak marked Z), suggesting that a smaller amount of 3:2 and/or 3:3 complex may have also formed.

SAXS and SANS provide structural insights into the PmScsC-PmScsB␣ interaction
SANS contrast variation experiments were performed on the 3:1 PmScsC-PmScsB␣ complex formed using unlabeled PmScsC C87S and deuterium-labeled PmScsB␣ C114A ( D PmScsB␣ C114A) (Fig. 7). Deuterium labeling changes the neutron scattering length density of PmScsB␣. It is possible to tune the neutron scattering length density of the solvent by changing its deuterium content such that it matches one of the components of the protein complex. This is said to be the match point of that component, and the measured scattering data can then be interpreted as being from the unmatched component of the complex alone. Data collected close to the D PmScsB␣ C114A match point (100% D 2 O, where PmScsC C87S dominates the scattering) indicate that the highly dynamic PmScsC protein becomes comparatively rigid and adopts a conformation that, at low-resolution, appears symmetrical upon interaction with PmScsB␣. Specifically, the pairdistance distribution function (p(r)) is bimodal (Fig. 8A, green curve) with two well defined peaks and similar to the p(r) generated from the extended symmetrical crystal structure (Protein Data Bank code 5ID4) (Fig. 8A, black dashed curve) (10). An asymmetric arrangement of the PmScsC trimer would yield an asymmetrically shaped second peak in the p(r), whereas significant flexibility would see this same peak broaden.
Analysis of the SANS data showed that a plot of the scattering intensity at zero angle (I(0)) normalized by protein concentration versus deuterium content of the solvent was well represented by a parabolic function, indicating that the samples were pure (Fig. 8B). A Stuhrmann plot revealed that the D PmScsB␣ C114A molecule was located toward the center of the complex (Fig. 8C). A rigid body model of the complex was optimized simultaneously against the SAXS data set and five SANS contrast variation data sets. The resulting model provided an excellent fit to all of the scattering data (Fig. 9A). The configuration adopted by PmScsB␣ and PmScsC in the model precludes the simultaneous binding of two (or three) PmScsB␣ molecules to PmScsC, helping to explain the preferential formation of a trimer-monomer complex (Fig. 6). The model also highlights two potential interfaces between PmScsC and PmScsB␣ (Fig.  9A). The most obvious interface is that between chain A of PmScsC and subdomain A of PmScsB␣ for which the intermolecular disulfide bond is modeled. The resolution of the model is too low to be definitive, but the interface includes residues that could form hydrogen bond interactions. The second interface is between chain A of PmScsC and subdomain B (the second immunoglobulin domain) of PmScsB␣.

Discussion
Suppressor of copper sensitivity proteins play a role in the response to copper stress of the important human pathogens S. enterica Typhimurium and P. mirabilis (7,10). At least two of

P. mirabilis ScsB␣ and ScsC form a functional redox relay
the four Scs proteins are homologues of Dsb proteins (6), which are essential for the correct formation of disulfide bonds in virulence factors of many Gram-negative bacteria (14 -17). P. mirabilis ScsC was recently shown to be a trimeric, highly dynamic disulfide isomerase (10). The present study aimed to characterize PmScsB␣, the predicted interaction partner of PmScsC, and to investigate the nature of the interaction between the two.
The domain architecture of full-length ScsB resembles that of E. coli DsbD, the major difference being the size of the periplasmic N-terminal domain. This difference suggested that ScsB␣ would be structurally distinct from DsbD␣ perhaps because they interact with different substrates. We confirmed that PmScsC and PmScsB␣ are redox partners and are therefore functionally similar to EcDsbC and EcDsbD␣. We also reported the structure of PmScsB␣, showing that it consists of two immunoglobulin-like folds connected by an ␣-helix. The structures of the two immunoglobulin-like subdomains of PmScsB␣ are broadly similar with the exception that subdomain A has a catalytic CXXXC motif as well as a ␤-hairpin and extended loop that both cover the CXXXC active site. The positioning of the ␤-hairpin and the loop could potentially shield the active site from nonspecific interactions or facilitate interactions with partner proteins. Unexpectedly, we found that domain A of PmScsB␣ shares the same features of the immunoglobulin fold of EcDsbD␣ ( Fig. 4; r.m.s.d., 2.5 Å, 103 residues aligned) despite a sequence identity of just 12%. The key difference between the structures of subdomain A and EcDsbD␣ is that EcDsbD␣ lacks the ␤2-␤3 loop of PmScsB␣. The sequence of this ␤2-␤3 loop is conserved in S. enterica Typhimurium and C. crescentus ScsB␣, suggesting that it plays a key functional role. A residue that is conserved across the three ScsB␣s and EcDsbD␣ is Tyr 45 (PmScsB␣ numbering). In EcDsbD␣, the equivalent Tyr 42 hydroxyl is thought to be involved in nucleophilic attack of Cys 103 , which resolves the intermolecular disulfide formed between EcDsbD␣ and its substrate (e.g. EcDsbC) (18). The nucleophilic attack in EcDsbD␣ is proposed to also require Asp 68 (18), a residue that is not conserved in Pm-, Cc-, or StScsB␣. It is unclear from the PmScsB␣ structure which residue, if any, might play a role equivalent to that of Asp 68 .
The discovery of tandem immunoglobulin-like folds in PmScsB␣ raises the question as to the function of the second or noncatalytic immunoglobulin-like fold. EcDsbD␣ has a single immunoglobulin-like fold, which, in the E. coli DsbC-DsbD␣ crystal structure, spans the central cleft of the V-shaped DsbC substrate to interact with both domains of the disulfide isomerase (Fig. 9B) (19). This binding mode is thought to favor binding to dimeric EcDsbC; EcDsbD␣ is unable to interact with monomeric EcDsbC (3,19). By comparison, the SANS-derived model of the PmScsC-PmScsB␣ complex suggests that PmScsB␣ subdomain A interacts with just one of the three PmScsC protomers in the trimer. Subdomain B of PmScsB␣ forms a secondary binding site to PmScsC that may stabilize the primary interaction or add specificity (Fig. 9A). Subdomain B could also be required for interactions with other proteins that have yet to be identified.
EcDsbC-EcDsbD␣ is the redox relay complex formed between dimeric EcDsbC and monomeric EcDsbD␣. The crystal structure revealed a 2:1 (rather than a 2:2) complex between the dimer of EcDsbC and a monomer of EcDsbD␣ (3,19). On this basis, we hypothesized that PmScsC-PmScsB␣ would form a 3:1 complex between trimeric PmScsC and monomeric

P. mirabilis ScsB␣ and ScsC form a functional redox relay
PmScsB␣, and the SANS structure of the complex appears to preclude higher stoichiometry. Nevertheless, we did find in vitro evidence suggestive of PmScsC forming 3:2 and 3:3 complexes with PmScsB␣ (Fig. 6A), showing that different arrangements might form with PmScsB␣ in the absence of the PmScsB ␤and ␥-domains. We expect that these higher order complexes would not form in the context of full-length PmScsB.
How do our findings on PmScsB␣ and PmScsC inform on interactions between ScsC and ScsB in other organisms? A functional interaction between ScsC and ScsB has already been reported for C. crescentus (5). With two examples now confirmed, the ScsB-C redox relay system is likely to be present in all organisms that encode the four-gene scs cassette. However, there is likely to be a wide diversity of interactions. ScsC is reported to be monomeric for StScsC (7), dimeric for CcScsC (5), and trimeric for PmScsC (10). By contrast, ScsB shares similar features across the same organisms, suggesting that StScsB␣ and CcScsB␣ both incorporate the tandem immunoglobulin-fold arrangement present in PmScsB␣. As indicated above, both subdomains of PmScsB␣ interact with one PmScsC protomer, suggesting that this tandem immunoglobulin-fold arrangement could facilitate the interaction with ScsCs regardless of the ScsC oligomeric state. Future studies will be needed to establish whether it is the tandem immunoglobulin-fold architecture that permits interactions with monomeric, dimeric, and trimeric PmScsCs. Curiously, a BLAST search of E. coli DsbD or S. enterica Typhimurium DsbD against the P. mirabilis genome did not identify any hits, suggesting the possibility that PmScsB may also act as the redox partner for P. mirabilis DsbC proteins.

Sequence analysis
The TMHMM Server (version 2.0) was used to determine the region of the ScsB proteins that formed the N-terminal periplasmic domain. EMBOSS Needle (20, 21) with residues 1-275, 1-279, and 1-290 of UniProt entries B4EV20 (PmScsB), AAL20046 (StScsB), and Q9ABL0 (CcScsB), respectively, was used to calculate the sequence identities reported among these proteins. The DALI sever (13) was used to calculate r.m.s.d. values and structure-based sequence identities. Coot (22) was used to create visuals of the structural alignments. The sequences of all four proteins were aligned using PROMALS3D (23) with the Protein Data Bank files of PmScsB␣ and EcDsbD␣ (code 1JPE) instead of the sequence alone. Residue numbers quoted in the text were taken from the deposited Protein Data Bank files for the PmScsC and PmScsB␣ protein structures, which are numbered based on the tobacco etch virus proteasecleaved protein sequence (starting S 1 N 2 A 3 . . . ).

Molecular biology
A codon-optimized gene encoding P. mirabilis ScsB␣ without the predicted signal peptide (UniProt entry B4EV20, residues 21-275; Integrated DNA Technologies) was inserted into pMCSG7 (24) using ligation-independent cloning. The PmScsC (UniProt entry C2LPE2/B4EV21) construct used throughout this study was created previously (10). The constructs for the variants PmScsC C87S and PmScsB␣ C114A were created using the QuikChange Lightning site-directed mutagenesis kit (Agilent Technologies). The WT PmScsC and PmScsB␣ constructs were used as templates, and primers 5Ј-gcggatcgaaacgtttagagtacgggcaattatag-3Ј, 5Ј-ctataattgcccgtactctaaacgtttcgatccgc-3Ј, 5Ј-gtcagaatgcatacattgctagcggtactcaatgtaaggacacc-3Ј, and 5Ј-ggtgtccttacattgagtaccgctagcaatgtatgcattctgac-3Ј, respectively, were used to introduce the point mutations. PmScsB␣ was also subcloned into pET24a for deuterated bioreactor expression using restriction enzymes EcoRI and NdeI. All constructs contained an N-terminal His 6 tag followed by a tobacco etch virus protease cleavage site and were confirmed by sequencing.

P. mirabilis ScsB␣ and ScsC form a functional redox relay
respectively. SDS-PAGE (NuPAGE system, 4 -12% BisTris gel, Invitrogen) with Coomassie Blue stain was used to assess the protein quality.

Expression of SeMet-labeled PmScsB␣
To express SeMet-labeled PmScsB␣, needed for experimental phasing in crystallographic structure determination, a 10-ml starter culture of E. coli BL21 (DE3) pLysS cells containing pMCSG7-PmScsB␣ was grown in LB medium supplemented with the appropriate antibiotics at 37°C for 4 -5 h at 220 rpm. This culture was then used to inoculate 1 liter of M9 minimal medium supplemented with antibiotics and amino acids (100 mg of L-methionine, L-lysine, L-threonine, and L-phenylalanine and 50 mg of L-leucine, L-isoleucine, and L-valine) and then incubated overnight at 37°C at 220 rpm. Cells were harvested from the overnight culture, washed with 1 liter of sterile water, and resuspended in water to an A 600 of ϳ1. M9 minimal medium containing antibiotics but no amino acids was then inoculated with the cells to a final A 600 of 0.1. The cells were cultured at 30°C at 220 rpm for 6 -8 h until an A 600 of 0.5 was reached. Amino acids, including 100 mg of L-selenomethionine instead of L-methionine, were then added to the culture flasks, and after 15 min at 30°C isopropyl 1-thio-␤-D-galactopyranoside was added to a final concentration of 1 mM to induce expression. The cultures were incubated at 30°C at 220 rpm overnight. Cells were harvested, and the protein was purified as described previously (10) with the addition of 5 mM DTT to all purification buffers. The yield of SeMet-labeled PmScsB␣ was 5 mg/liter of culture (ϳ5 g of cell pellet).

Expression of deuterated PmScsB␣ C114A
Deuterated PmScsB␣ C114A was expressed for SANS analysis of the complex between PmScsC C87S and PmScsB␣ C114A. The deuterated protein PmScsB␣ C114A was produced at the National Deuteration Facility, Australian Nuclear Science and Technology Organisation in 1 liter of "ModC1" medium (26) containing 90% (v/v) D 2 O and unlabeled glycerol (40 g/liter). Miniprep plasmid DNA "pET24a-PmScsB␣ C114A" was used to transform 50 l of Invitrogen competent BL21*(DE3) cells, which were then incubated with 250 l of SOC medium for 2 h. The 300-l culture was added to 10 ml of ModC1 medium containing 50% D 2 O and 40 g/ml kanamycin, shaking at 200 rpm at 37°C. After 16 h (A 600 , 1.2), 9 ml was added to 36 ml of fresh medium with kanamycin at 100% D 2 O (i.e. 45 ml at 90% D 2 O). After another 5 h (A 600 , 0.9), the culture was diluted with fresh 90% D 2 O medium to 102 ml. After another 5 h, 100 ml (A 600 , 0.87) was used to inoculate 900 ml of fresh 90% D 2 O medium with kanamycin in a 2-liter Real Time Engineering bioreactor aerated with air at 0.5 liter/min and with pH held to a minimum of 6.2 by addition of 28% NH 4 OH in H 2 O (Sigma). Dissolved oxygen tension was set to 75% and was  There is a small systematic change in R g but no systematic change in M; thus, it was deemed that the complex is largely free of concentration-dependent effects up to a concentration of ϳ5.0 mg/ml. b The Porod volume calculation is not valid for systems where the ⌬ of the particle is not homogeneous (such as is the case for the neutron contrast variation experiment reported here). Instead, the composite scattering functions (Fig. 8D) were summed together to give the scattering curve of the protein complex with homogeneous contrast (i.e. I homogeneous ϭ I 11 ϩ I 22 ϩ I 12 ), and the Porod volume was determined from this curve. The molecular mass derived from the I(0) was also taken from this curve.

P. mirabilis ScsB␣ and ScsC form a functional redox relay
above 60% throughout. At an A 600 of 14.2, the temperature was lowered to 20°C, and expression was induced by the addition of 1 mM isopropyl 1-thio-␤-D-galactopyranoside. At 17 h postinduction, with an A 600 of 32, the culture was harvested with a wet weight yield of 79.4 g and with excellent expression as evident by SDS-PAGE. The deuteration level was determined by partial trypsin digest MALDI-TOF comparison of unlabeled and labeled samples and was found to be 70.8%. Deuterated PmScsB␣ C114A was purified as described for the unlabeled protein except the His 6 tag was not cleaved as it was needed for the purification of the complex. The yield of purified deuterated PmScsB␣ was ϳ240 mg/liter of bioreactor culture (ϳ40 g of cell pellet).

Determination of protein concentration
The reported protein concentrations were determined using the A 280 of the sample (read using a Thermo Scientific Nano-Drop 2000c spectrophotometer) and calculated extinction coefficients from ProtParam (27).

Crystal structure of PmScsB␣
A number of conditions from commercial screens resulted in PmScsB␣ crystals. These conditions were replicated in 24-well plates, and the best diffracting SeMet-labeled PmScsB␣ crystals were obtained in a drop containing 1 l of 43 mg/ml protein in 10 mM HEPES, pH 7.4, 5 mM DTT and 1 l of 0.1 M MES monohydrate, pH 6, 6% (v/v) Tacsimate, pH 6, 24% PEG 4000. A commercial additive screen (Hampton Research) was used to optimize crystals, and those grown over 2 days to ϳ400 m with the addition of 4% 2,5-hexanediol or 4% 1,3-butanediol resulted in the best diffraction. Crystals were cryoprotected in the crystallization condition plus 20% glycerol and flash frozen in liquid nitrogen.
Because of the lack of a suitable model for molecular replacement, experimental phasing was required to solve the crystal structure of PmScsB␣. A fluorescence scan was used to determine the absorption edge of SeMet-labeled crystals, and a data set (Table 1) was collected near the peak absorption wavelength (0.9786 Å) at 100 K. Data processing in autoPROC (28), which utilizes XDS (29) and AIMLESS (30), determined the space group to be P2 1 2 1 2 1. SAD phasing using the SHELX workflow (31) in HKL2MAP (32) found 14 selenium sites with an occupancy Ͼ0.2. We expected only 12 selenium sites; the extra two sites were positioned very close to other sites, suggesting alternate conformations, although only one SeMet was modeled with an alternate conformation in the final structure. SAD phasing resulted in an initial incomplete alanine model of the protein structure (figure of merit, 0.42). This initial model was then used for molecular replacement against the same data in Phaser (33) in the CCP4 suite (34). Two rounds of automated building and refinement using Buccaneer (35) with REFMAC (36), as implemented in CCP4 (34), completed the structure, revealing three molecules in the asymmetric unit. Then several rounds of manual adjustment in Coot (22) and refinement in PHENIX (37) were performed with reference to the validation program MolProbity (38). Multiple residues in the crystal structure have alternate conformers, and occupancies of the selenium atoms in the SeMet residues were all Ͻ1.0. The electron density for some loop regions is poor in chains B and C, particularly residues 79 -82, 99 -100, and 51-52 (chain B only). R/R free values are 17.3 and 19.8% at a resolution of 1.54 Å. All structural figures were made in PyMOL (39).

Redox interaction assay
To confirm the redox interaction between PmScsC and PmScsB␣, oxidized PmScsC and reduced PmScsB␣ were incubated together at final concentrations of 10 M in 100 mM sodium phosphate, pH 7, 1 mM EDTA for 2 h. As controls, oxidized and reduced forms of both proteins, a sample containing reduced PmScsC and oxidized PmScsB␣, and a sample with oxidized PmScsC with PmScsB␣ C114A were also incubated for 2 h. All samples were then treated with 10% TCA to precipitate the protein, and after centrifugation the pellets were washed with 100% cold acetone and dried. The protein pellets were resuspended in a solution of 200 M AMS in 1% SDS, 50 mM Tris, pH 7.0. AMS binds to the reduced cysteines of pro-

P. mirabilis ScsB␣ and ScsC form a functional redox relay
teins, increasing the molecular weight and allowing the differentiation between the oxidized and reduced forms of a protein when run on a gel. Therefore, all samples were run on a nonreducing, nondenaturing 12% NuPAGE BisTris gel for 2.5 h at 100 V. The gel was Coomassie-stained and imaged using an Odyssey Fc imaging system (LI-COR Biosciences) with the 700-nm channel.

RNase A isomerase assay
To determine whether PmScsB␣ activates the isomerase PmScsC, an isomerase assay utilizing scrambled RNase A (scRNase A) and the RNase A model substrate cCMP was performed. Bovine RNase A (purchased from Sigma-Aldrich) was scrambled as described previously (40). The various oxidized and reduced protein samples described above for the gel-shift redox interaction assay were again prepared but with the addition of 40 M scRNase A to each sample. After 2 h, 50 l of the protein/scRNase A reactions was taken and added to 150 l of 4 mM cCMP in a UV-Star 96-well plate (Greiner Bio-One, Austria). The increase in absorbance at 296 nm was monitored over 3 min with a Synergy H1 multimode plate reader (BioTek) and used to calculate the rate of cCMP hydrolysis. Native RNase Aand scRNase A-only samples were included as controls and used to determine the relative activity of the RNase A in each where R m is the radius of gyration of that object with homogenous contrast and ␣ and ␤ are related to the contrast variations within the object. The values obtained from a fit to the plot (solid black line) are: R m ϭ 37.5 Ϯ 0.4, ␣ ϭ Ϫ210 Ϯ 50, and ␤ ϭ 520 Ϯ 100. The negative value of ␣ reveals that the region with higher contrast (i.e. D PmScsB␣ C114A) lies toward the center of the molecule. Error bars represent the S.E. D, the composite scattering functions determined from the neutron contrast variation data. The I 11 curve (green; left y axis) corresponds to scattering from PmScsC C87S, the I 22 curve (blue; left y axis) corresponds to scattering from D PmScsB␣ C114A, and the I 12 curve (gray; right y axis) is related to the arrangement of D PmScsB␣ C114A relative to PmScsC C87S. I homogeneous ϭ I 11 ϩ I 22 ϩ I 12 (red; left y axis) is the scattering curve of an object with the same shape as the PmScsC C87S-D PmScsB␣ C114A complex but with homogeneous contrast and is used for estimating the Porod volume and molecular mass from the SANS data.

P. mirabilis ScsB␣ and ScsC form a functional redox relay
sample. The results shown are the average and S.D. of three technical replicates.

Complex formation
The PmScsC and PmScsB␣ cysteine variants were designed following the same approach used for the EcDsbC-EcDsbD␣ interaction (19). Mutation of cysteines in EcDsbC and EcDsbD␣ resulted in a stable complex between the two proteins (3). Cysteines Cys 87 of PmScsC and Cys 114 of PmScsB␣ are the least accessible of the two cysteines in each active site, suggesting they facilitate resolution of intermolecular disulfide bonds. Variants PmScsC C87S and PmScsB␣ C114A were therefore prepared and used in complex formation studies.
The 3:1 PmScsC C87S-PmScsB␣ C114A complex was formed by mixing the two proteins together in a 3:1 ratio at concentrations of ϳ200 mg/ml and incubating the mixture at 37°C for 3 days. To separate the complex and free PmScsB␣ from PmScsC, PmScsB␣ was left His-tagged, and the protein mixture was then run over 2 ml of fresh TALON resin twice. Columns were washed with 25 mM Tris, pH 7.4, 150 mM NaCl, and the complex and free PmScsB␣ were eluted with 25 mM Tris, pH 7.4, 150 mM NaCl, 250 mM imidazole. This was then run over a Superdex 200 10/300 Increase size exclusion column on an ÅKTA system to separate PmScsB␣ from complex. A peak on SEC eluted earlier than for ScsC, suggesting that a complex between the two proteins had formed (Fig. 6A). Running the fractions from the peak corresponding to the 3:1 complex on nonreducing SDS-PAGE resulted in three bands (Fig.  6B). The lower bands correlated to free PmScsC C87S (24.8 kDa) and a small amount of PmScsB␣ C114A (30.3 kDa), and the upper band corresponded to the molecular mass of one PmScsC C87S molecule bound to one PmScsB␣ C114A molecule (55.1 kDa). The upper band dissociated when reducing agent was added, leaving two bands correlating to monomeric PmScsC C87S and PmScsB␣ C114A in an approximate 3:1 ratio (Fig. 6B). There was also evidence for the formation of small amounts of complex containing PmScsC C87S and PmScsB␣ C114A in ratios of 3:3 and 3:2, but only those SEC fractions containing 3:1 complex were collected for further analysis. The complex formation process was repeated multiple times to gain enough sample for SAXS/SANS analysis as the overall yield of 3:1 complex was ϳ2.5%.

Small-angle solution scattering
Small-angle neutron scattering data shown in Table 2 were collected on the QUOKKA instrument at the Australian Nuclear Science and Technology Organisation ( Table 2) (41). Two aliquots of PmScsC C87S-D PmScsB␣ C114A complex at ϳ5 mg/ml were dialyzed, one against a buffer with a D 2 O content of 0% and the other against a buffer with a D 2 O content of ϳ100%. Samples with D 2 O contents of 20, 42, and 80% were obtained by mixing the two original solutions at the appropriate ratios. The two-dimensional data were normalized to a common incident neutron count and corrected for sample transmission, background radiation, empty cell scattering, and detector sensitivity and radially averaged to produce I(q) versus q profiles. Scattering data from the two different sampledetector distances were then merged, and buffer scattering data were then subtracted from the protein ϩ buffer data to give the protein scattering profiles (the 20, 42, and 80% buffer scattering curves were taken as a linear combination of the 0 and 100% buffer scattering curves). To correct for the effects of incoherent scattering by 1 H-rich samples, backgrounds levels were adjusted by a small constant such that the high-q scattering displayed q Ϫ4 dependence.
Small-angle X-ray scattering data were collected on the SAXS-WAXS beamline at the Australian Synchrotron (Table  2) (42). Serial dilutions of the stock solution of PmScsC C87S-D PmScsB␣ C114A complex were made to give four concentrations between 0.55 and 4.5 mg/ml that were centrifuged at 10,000 ϫ g immediately before loading into a 96-well plate. To reduce radiation damage, samples (ϳ100 l) were drawn into a capillary from the 96-well plate and flowed past the beam. All measured two-dimensional data were averaged and corrected for sample transmission, solvent scattering, and detector sensitivity and radially averaged to produce I(q) versus q profiles using Scatterbrain (version 2.7.1) (43).
The estimated molecular masses were calculated using values for contrast and partial specific volume predicted from the protein sequence using MULCh (version 1.1) (44). Data processing and Guinier analysis were performed using PRIMUS (version 3.2) (45). The p(r) was generated from the experimental data using GNOM (version 4.6) (4) from which I(0), R g , and D max were determined. For the SANS data, analysis of the dependence of R g upon contrast and calculation of the composite scattering functions were also calculated using MULCh (Fig.  8D). Corefinement of a rigid-body model against SANS and SAXS data was performed using SASREF (version 7.d) (46) where the crystal structure of PmScsC (Protein Data Bank code 4XVW, chain A) was broken into three rigid sections: 1) residues Ala 3 -Gln 43 were fixed to preserve the structure of the trimerization domain observed in the crystal structure, 2) residues Gln 44 -Phe 47 were defined as a short linker with a restraint between residues Gln 43 and Gln 44 to keep them in close proximity, and 3) residues Arg 48 -Lys 224 were defined as the catalytic domain free to move with a restraint between residues Phe 47 and Arg 48 to keep them in close proximity. The additional PmScsC chains were generated by symmetry (assuming a C 3 point group). The crystal structure of PmScsB␣ (residues Ala 2 -Ala 248 ) was included in the optimization with a distance restraint between Cys 84 of a PmScsC monomer and Cys 118 of PmScsB␣ of 8 Å. Thus, although the complex overall is not symmetrical, the PmScsC portion of the model possesses 3-fold symmetry. The optimization was run 16 times, and the model that provided the best global fit to the scattering data is reported here; however, all models obtained are similar in structure.