Engineered variants provide new insight into the structural properties important for activity of the highly dynamic, trimeric protein disulfide isomerase ScsC from Proteus mirabilis

The structure and function of a bacterial trimeric protein disulfide isomerase was investigated by characterising two variants in which key structural regions were deleted. In one case the effect on structure and function was predictable, while in the other we found unintended consequences for both structure and function.


Introduction
Protein disulfide isomerases are enzymes that proofread and shuffle incorrect disulfide bonds in misfolded protein substrates, and are important for the correct folding and function of many secreted proteins (Berkmen et al., 2005;Hiniker & Bardwell, 2004). The prototypical bacterial disulfide isomerase is disulfide-bond (Dsb) protein C from Escherichia coli (EcDsbC; Zapun et al., 1995;Missiakas et al., 1994;Shevchik et al., 1994). EcDsbC functions as a dimer; when the N-terminal dimerization domain is deleted the resulting protein lacks disulfide isomerase activity (Sun & Wang, 2000) and is also unable to interact with its redox partner EcDsbD (Goldstone et al., 2001). Each EcDsbC protomer has a thioredoxin-fold catalytic domain with a redox-active motif consisting of two cysteines separated by two other amino acids (CXXC; McCarthy et al., 2000).
The disulfide-isomerase activity of EcDsbC requires that the catalytic cysteines are in the dithiol-reduced form (Darby et al., 1998), and this form is generated by the interaction of EcDsbC with its redox-partner membrane protein EcDsbD (Goldstone et al., 2001).
Recently, we reported the structural and functional characterization of Proteus mirabilis suppressor of copper sensitivity protein C (PmScsC; Furlong et al., 2017). Like DsbC, PmScsC is a protein disulfide isomerase with a thioredoxinfold catalytic domain (Furlong et al., 2017), redox-active cysteines (Furlong et al., 2017) and a redox-partner membrane protein (PmScsB) that reduces the active-site cysteines (Furlong et al., 2018). Unlike DsbC, PmScsC is trimeric rather than dimeric (Furlong et al., 2017). Three crystal structures and small-angle X-ray scattering (SAXS) analysis also revealed that PmScsC is highly dynamic (Furlong et al., 2017). The region responsible for this conformational flexibility is an 11amino-acid linker within the trimerization stem of the protein.
Replacement of the flexible linker with a rigid helical peptide linker (PmScsC RHP) limits the rigid-body motion of the catalytic domain relative to the trimer stalk, and generates a protein that is inactive as a disulfide isomerase. Moreover, PmScsC lacking the trimerization stem (the N-terminal 41 residues; PmScsCÁN) is also inactive as a disulfide isomerase but does have dithiol oxidase activity (Furlong et al., 2017). Neither the native protein nor the PmScsC RHP variant exhibited dithiol oxidase activity. The difference in the activity profile between the native protein (disulfide isomerase) and PmScsCÁN (dithiol oxidase) was proposed to be a consequence of their different oligomeric states.
Here, we report the structural characterization of PmScsCÁN and the structural and functional characterization of PmScsCÁLinker (a variant in which the 11-amino-acid flexible linker is deleted; Fig. 1). The crystal structure of PmScsCÁN reveals, as expected, a monomeric protein that closely resembles the catalytic domain from the full-length PmScsC protomer. We show that the PmScsCÁLinker variant lacks both disulfide isomerase and dithiol oxidase activities. The crystal structure reveals a trimeric arrangement for PmScsCÁLinker, with the catalytic domains closely packed against each other. Conversely, SAXS analysis indicates that the oligomeric state of PmScsCÁLinker is concentrationdependent, with a monomeric form present at low concentrations and a dimeric form dominating at higher concentrations. For comparison, we analysed a concentration series of full-length native PmScsC by SAXS, showing that the dominant species of the native protein in solution is trimeric even at very low concentrations.

Molecular biology, protein expression and purification
The constructs for the expression of PmScsC and PmScsCÁN were created previously (Furlong et al., 2017) using UniProt sequence B4EV21 without the predicted periplasmic signalling sequence (i.e. residues 22-243). To make the PmScsCÁLinker expression construct, the DNA encoding the flexible peptide linker (residues 39-KADEQQAQFRQ-49) in PmScsC was deleted from the P. mirabilis scsC gene using overlap extension PCR with the primers 5 0 -GCAATCAT GGCTCTGCAGACGAAAGCACTGGCTAGCGAACATG ATGCC-3 0 and 5 0 -GGCATCATGTTCGCTAGCCAGTGCTT TCGTCTGCAGAGCCATGATTGC-3 0 . Ligation-independent cloning was then used to insert the mutated gene into pMCSG7. Sequencing of the construct confirmed that the encoded protein was in-frame with an N-terminal His 6 tag and TEV protease cleavage site, and revealed that residue 6 of PmScsCÁLinker was also mutated from asparagine to lysine. The protein residues are numbered based on the sequence of full-length PmScsC after TEV protease cleavage (starting S 1 N 2 A 3 . . . ). All proteins were expressed in Escherichia coli BL21 (DE3) pLysS cells and purified using immobilized metalaffinity chromatography, TEV protease cleavage and sizeexclusion chromatography as described previously (Furlong et al., 2017). SDS-PAGE with Coomassie staining was used to estimate that the purity of each protein was >95%. The concentration of the proteins was determined by measuring the A 280 of the samples and adjusting for the theoretical A 280 of a 1 mg ml À1 solution of the protein [Abs 0.1%(w/v)]. The theoretical A 280 values for 0.1%(w/v) determined from ProtParam (Gasteiger et al., 2005) were 0.528, 0.643 and 0.558 for PmScsC, PmScsCÁN and PmScsCÁLinker, respectively.

Disulfide isomerase and dithiol oxidase assays
The disulfide isomerase and dithiol oxidase activities of PmScsCÁLinker were determined as described previously (Furlong et al., 2017). Briefly, to determine the disulfide isomerase activity, the ability of the protein to refold scrambled RNase A (scRNase A) was assessed. 0.23 mg ml À1 (10 mM) PmScsCÁLinker was incubated with 40 mM scRNase A; samples were taken at various time points and the RNase A activity was measured in a spectrophotometric assay with a final concentration of 3 mM cytidine 3 0 ,5 0 -cyclic monophosphate (cCMP). Results are reported as a percentage of the activity of the RNase A-only control and are adjusted for the scRNase A-only negative control. The dithiol oxidase activity of PmScsCÁLinker was assessed by measuring the ability of the protein to form a disulfide bond between the two cysteines Schematic representation of the PmScsC variants used in this study. The region corresponding to the catalytic domain is shown in green, the trimerization stem is coloured magenta and the region representing the flexible linker is shown in cyan. The orange circles represent the approximate positions of the redox-active cysteines. in a synthetic peptide over time. The peptide substrate has 1,4,7,10-tetraazacyclododecane-1,4,7,10-tetraacetic acid (DOTA) bound to europium at the N-terminus and lysine methoxycoumarin amide (MCA) at the C-terminus. As the cysteines become oxidized, the termini of the peptide are brought into close proximity, which causes an increase in fluorescence. PmScsCÁLinker (80 nM or 0.002 mg ml À1 ) was incubated with 8 mM of the peptide substrate and the increase in fluorescence over time was monitored using a Synergy H1 Hybrid plate reader (BioTek, Vermont, USA), reading fluorescence at excitation and emission wavelengths of 340 and 615 nm, respectively. The rate of peptide oxidation is reported relative to the activity of the prototypical dithiol oxidase EcDsbA. The disulfide isomerase and dithiol oxidase assays were repeated two and three times, respectively, and the mean and standard deviation for each is reported.

Crystallization
Initial crystallization hits for both PmScsCÁN and PmScsCÁLinker were identified in 96-well hanging-drop vapour-diffusion experiments using commercial crystallization screens set up with a Mosquito robot (TTP Labtech, Melbourn, England) at the University of Queensland Remote Operation Crystallization and X-ray (UQ ROCX) Diffraction Facility. All crystallization experiments were incubated at 293 K and monitored using a Rock Imager with Rock Maker software (Formulatrix, Waltham, Massachusetts, USA).
Well diffracting, rod-shaped crystals of PmScsCÁN were obtained in a 96-well screening plate in a drop consisting of 200 nl PEGRx HT (Hampton Research, Aliso Viejo, California, USA) condition C10 [0.1 M sodium acetate trihydrate pH 4.5, 30%(w/v) PEG monomethyl ether 5000] and 200 nl 38 mg ml À1 PmScsCÁN in 10 mM HEPES pH 7.4. Perfluoropolyether (Hampton Research, Aliso Viejo, California, USA) was used to cryoprotect the crystals before they were flashcooled in liquid nitrogen.
Needle-like crystals of PmScsCÁLinker were obtained in well F12 of the commercial screen JCSG-plus (Molecular Dimensions, Newmarket, England) and were then optimized in a 24-well plate grid screen around the original condition. This optimization resulted in more robust and well diffracting crystals. Data were collected from a crystal grown using 1 ml 0.1 M HEPES pH 7, 32%(v/v) Jeffamine M-600 and 1 ml of 40 mg ml À1 PmScsCÁLinker in 10 mM HEPES pH 7.4, 150 mM NaCl. The crystal was cryoprotected in the crystallization condition plus 20% ethylene glycol before being flashcooled in a cryostream (100 K).

Data collection, structure solution and refinement
Data for the PmScsCÁN structure were collected using a wavelength of 0.95370 Å at 100 K on the MX2 beamline at the Australian Synchrotron using the Blu-Ice software (McPhillips et al., 2002). The data were processed and scaled in the trigonal space group P312 using XDS (Kabsch, 2010), POINTLESS (Evans, 2006) and SCALA (Evans, 2006). Phases were obtained by molecular replacement in Phaser (McCoy et al., 2007) using Salmonella enterica serovar Typhimurium ScsC (PDB entry 4gxz; Shepherd et al., 2013) as the model. The initial model was improved by iterative model building in Coot (Emsley et al., 2010) and refinement in PHENIX (Adams et al., 2010). The quality of the final PmScsCÁN model was assessed using MolProbity  and had an R free value of 20.8% and an R work value of 17.5% at a resolution of 2.15 Å . The coordinates and structure factors have been deposited in the Protein Data Bank under the code 6nen, and raw data images are available from https://doi.org/10.14264/uql.2018.843.
Data for the PmScsCÁLinker structure were collected at UQ ROCX under cryogenic conditions (100 K) using a Rigaku FR-E SuperBright X-ray generator, producing X-rays at a wavelength of 1.54187 Å , and a Rigaku Saturn 944 CCD area detector. The data collection was performed with the Rigaku CrystalClear (v.2.0) program. Data were processed in iMosflm (v.7.1.3; Battye et al., 2011) with space group H32 (also known as R32:h) and were scaled in AIMLESS (Evans & Murshudov, 2013) as implemented in the CCP4 suite (v.6.5.008; Winn et al., 2011). Owing to detector-distance constraints at UQ ROCX, diffraction data were limited to a highest resolution of 2.08 Å . The crystals diffracted strongly, resulting in very high mean I/(I) values (27.2 overall and 11.8  in the highest resolution shell). The structure was phased by molecular replacement using Phaser (McCoy et al., 2007) and the catalytic domain of the compact PmScsC crystal structure (residues 47-224; PDB entry 4xvw) was used as the search model. The trimerization domain (without the 11-amino-acid flexible linker) was then manually built into the electron density in Coot (Emsley et al., 2010). The structure was refined using rounds of phenix.refine and manual adjustment in Coot, with reference to MolProbity . The final R free and R work values are 21.0% and 17.4%, respectively, at a resolution of 2.08 Å . The coordinates and structure factors have been deposited in the Protein Data Bank under the code 6mhh, and raw data images are available from https://doi.org/ 10.14264/uql.2018.842. All structural figures were produced in PyMOL (v.1.6; Schö dinger) and structural alignments were performed in Coot using the 'superpose' command (Emsley et al., 2010) which aligns C atoms by least squares. The CONTACT program from the CCP4 suite (Winn et al., 2011) was used to analyse the interactions between the trimerization stems of PmScsC and PmScsCÁLinker. Data-collection and refinement statistics are given in Table 1.
For PmScsC (SASBDB ID SASDER4) and PmScsCÁLinker (SASBDB ID SASDEQ4), the data were modelled as a linear combination of model scattering curves to quantify the proportions of different oligomeric states in solution (Hu et al., 2012). Structures of the PmScsC monomer (chain A), dimer (chains A and B) and trimer (chains A, B and C) were taken from a CORAL model previously optimized against the SAXS data for PmScsC (SASBDB ID SASDB94; Furlong et al., 2017). Structures of the PmScsCÁLinker monomer (chain A), dimer (chains A and B) and trimer (chains A, B and C) were taken from the crystal structure of PmScsCÁLinker (PDB entry 6mhh .2 kDa (10.0 mg ml À1 ). Masses from the Porod volume are used as concentration estimates, yielding masses from I(0) inconsistent with the expected mass. The 0.5 mg ml À1 data were used for subsequent data analysis. Svergun et al., 1995). The calculated scattering curves (with units of e 2 ) were multiplied by r e 2 (cm 2 e À2 ) Â N A (mol À1 ) Â 10 À3 (l cm À3 ) = 4.78181 Â 10 À5 (cm À1 e À2 mol À1 l). This normalization permits the coefficients arising from a fit to the experimental data on an absolute scale to be interpreted as the molar concentration of each oligomer present in solution. For PmScsC, a linear combination of monomer and trimer scattering curves was fitted to the scattering data at each concentration. Additionally, to reduce the number of free parameters, the concentration of monomer for each SAXS curve was optimized together with a common equilibrium constant, K t (3 Â monomer ! K t trimer), and the concentration of trimer was calculated from K t and the momomer concentration. For PmScsCÁLinker, a linear combination of monomer, dimer and trimer scattering curves was fitted to the scattering data at each concentration. Again, to reduce the number of free parameters the concentration of monomer for each SAXS curve was also optimized together with two common equilibrium constants, K d (2 Â monomer ! K d dimer), and K t (monomer þ dimer ! K t trimer). The concentration of dimer was calculated from K d and the monomer concentration, while the concentration of trimer was calculated from K t and the concentrations of monomer and dimer. From the model concentrations of each component, a total protein concentration was also calculated for each SAXS data set and these are reported in Table 3. These calculated concentrations are systematically lower than the measured protein concentrations.

PmScsCDN crystal structure and SAXS analysis
The design and biochemical characterization of PmScsCÁN has previously been reported (Furlong et al., 2017). The present work reports the crystal structure of this variant. Crystals of the PmScsCÁN mutant diffracted to a resolution of 2.15 Å at the Australian Synchrotron. The crystal structure of the mutant was solved using molecular replacement and there was only one PmScsCÁN molecule in the asymmetric unit (Table 1, Figs. 2a and 2b). In the crystal packing, PmScsCÁN makes contact with six symmetry-related molecules, but none involve packing against the catalytic motif or the positively charged surface patch that surrounds it. The structure of PmScsCÁN is very similar to the structure of the catalytic domain present in each of the three previously reported PmScsC structures (r.m.s.d. of 0.51-0.68 Å for the C atoms of 174 residues aligned between chain A of each structure; Fig. 2c). Small-angle scattering (  Table 3 Oligomer fractions and equilibrium constants for dimer and trimer formation (K d and K t , respectively) from analysis of the SAXS data.
Values obtained from the global optimization are shown first and those from the independent optimization are given in parentheses.  . This superimposition shows that the structures are similar whether or not the trimerization domain is present. (Fig. 3). The scattering data for PmScsCÁN are representative of a monomeric globular particle and are consistent with the scattering curve calculated from the crystal structure of PmScsCÁN (PDB entry 6nen). The crystal structure also shows good agreement with the dummy-atom model derived from the scattering data (Fig. 3c).

Activity of PmScsCDLinker
A key structural feature of native PmScsC is an 11-aminoacid flexible linker that is situated in the trimerization stem and links to the catalytic domain. This short region facilitates the conformational variations observed in the crystal structures and the dynamic motion of the trimeric protein observed in SAXS analyses (Furlong et al., 2017). To better understand the role of the flexible linker, we designed a PmScsCÁLinker variant in which the 11-residue linker is deleted and characterized its function in standard assays. We first assessed the activity of PmScsCÁLinker in the scrambled RNase A assay and found that it was poorly active, with baseline activity similar to that of a typical dithiol oxidase such as EcDsbA or PmScsCÁN (Fig. 4a). We then assessed its activity in the model peptide dithiol oxidase assay, in which EcDsbA and PmScsCÁN are both active. However, unlike the monomeric proteins EcDsbA and PmScsCÁN, PmScsCÁLinker was also inactive in the dithiol oxidase assay (Fig. 4b). Disulfide isomerase activity is thought to be associated with the oligomerization of thioredoxin-fold proteins (Furlong et al., 2017;Ke et al., 2006;Sun & Wang, 2000), so we were interested to determine whether PmScsCÁLinker forms a trimer, like the full-length native protein, or a monomer, like the truncated PmScsCÁN variant.

The crystal structure reveals a trimeric form of PmScsCDLinker
The three crystal structures of native PmScsC reported previously show that the linker residues can adopt helical (extended-conformation), strand (intermediate-conformation) or loop (compact-conformation) secondary structures (Furlong et al., 2017). In the extended-conformation crystal structure the linker adopts a helical conformation, thereby creating a very long helix (35 residues, approximately 9.5 turns; Fig. 5a) extending from the trimerization stalk through into the catalytic domain of PmScsC.
We solved the crystal structure of PmScsCÁLinker at a resolution of 2.08 Å using molecular replacement (Table 1, Figs. 5b and 5c). The variant crystallized in the same space group (H32) and crystallization conditions as the extendedconformation PmScsC (PDB entry 5id4; Furlong et al., 2017). However, the unit-cell dimensions of PmScsCÁLinker are shorter than those of the extended PmScsC structure (a = b = 64.0, c = 299.8 Å compared with a = b = 86.7, c = 330.0 Å ). The crystal structure of PmScsCÁLinker, like the crystal structure of the extended form of PmScsC, is trimeric through crystallographic symmetry, with a single protomer in the asymmetric unit (Fig. 6a). The regions on either side of the linker region reported to be helical in the three native PmScsC Small-angle X-ray scattering data for PmScsCÁN. (a) Measured scattering data for PmScsCÁN. The scattering profile of a rigid-body model is shown as a solid black line overlaid on the scattering data for PmScsCÁN [ 2 = 1.12; CorMap test (Franke et al., 2015), 302 points, C = 14, P = 0.018]. Inset: Guinier plot for PmScsCÁN (R 2 = 0.991). (b) The pair-distance distribution function, p(r), derived from the scattering data is indicative of a globular structure for PmScsCÁN (multiplied by a factor of four for clarity) with a maximum dimension of $55 Å . (c) Probable shape of PmScsCÁN obtained from the filtered average of 16 dummy-atom models (green envelope): 2 = 1.079 AE 0.001; NSD = 0.522 AE 0.007; resolution = 19 AE 2 Å . The image in (c) was generated using PyMOL, where the grey shape represents the total volume encompassed by the aligned dummy-atom models and the crystal structure is shown as a ribbon diagram aligned with the filtered model (dark green). crystal structures are directly connected in PmScsCÁLinker and are also helical.
The crystal structure of PmScsCÁLinker is similar to that of extended PmScsC except for a shortening of the long helix by almost three turns (Figs. 5a and 5b). This shortened helix in the crystal trimer of PmScsCÁLinker brings the three catalytic domains into much closer proximity in the variant compared with the native extended conformation. This proximity is evident from the distance between the catalytic cysteines of the protomers: 15 Å in PmScsCÁLinker and 34 Å in the extended native structure (measured between the C atoms of Cys84; Figs. 6a and 6b).
The PmScsCÁLinker crystal structure also reveals interactions between each of the catalytic domains of the trimer. The surface interface is formed between a positively charged patch close to the CXXC motif on the catalytic domain of one protomer and a neutral/hydrophobic patch on another (Fig. 6c). This interface is not formed in the extended PmScsC   Fig. 1. The N-terminus of each structure is labelled and the positions of the residues flanking the flexible linker in the native structure are indicated. (c) Electron density around the deleted region in the PmScsCÁLinker structure. The wire mesh represents the F o À F c electron-density difference map (generated in phenix.refine with the region of interest omitted from the structure) contoured to 3 and shown within a 2 Å radius of each atom.

Figure 4
Biochemical activity of PmScsCÁLinker. (a) The ability of PmScsCÁLinker to isomerize the incorrect disulfide bonds in scrambled RNase A is poor; it is even lower than that of the dithiol oxidase EcDsbA. (b) PmScsCÁLinker has poor dithiol oxidase activity; it is much lower than that of monomeric PmScsCÁN and similar to those of the protein disulfide isomerases EcDsbC and PmScsC. Asterisks indicate data published previously (Furlong et al., 2017). Error bars in both panels represent the standard deviation of the mean. structure, which has both of these regions exposed to solvent, and it is also not formed by crystal packing in the PmScsCÁN structure.
The trimerization stems of all three native PmScsC crystal structures are -helical with a hydrophobic core and with hydrogen bonds between surface-exposed polar/charged residues. These features are conserved in the crystal structure of PmScsCÁLinker (Figs. 6d and 6e). However, it should be noted that PmScsCÁLinker has an additional unanticipated mutation (Asn6Lys) at the N-terminus. In the three native crystal structures the side chain of the Asn6 residue forms an intramolecular hydrogen bond to the main chain of Gln9. In the PmScsCÁLinker crystal structure the side chain of Gln9 adopts the same conformation as in all three native PmScsC structures, but the mutated residue Lys6 does not form an equivalent hydrogen bond to the main chain of Gln9. The lack of this specific hydrogen bond could potentially increase the N-terminal disorder, although the crystal structure suggests there is little impact on the hydrogen bonds and hydrophobic contacts that contribute to trimerization.

SAXS reveals that the oligomeric state of PmScsCDLinker is concentration dependent
We had previously observed that the PmScsC RHP variant (in which the flexible linker is replaced by a rigid helix) forms a trimer, like the native protein, but is inactive in the protein disulfide isomerase assay (Furlong et al., 2017). This lack of activity was thought to be a consequence of the reduced catalytic domain motion observed in solution using smallangle scattering (Furlong et al., 2017). This finding led to the hypothesis that potent protein disulfide isomerase activity requires at least two catalytic domains (for example a dimer or trimer) which have the ability to move independently of each Trimerization of PmScsCÁLinker. (a) Side and top views of the catalytic domain in the crystal structure of PmScsCÁLinker. The crystal structure reveals a trimer through crystallographic symmetry. In many respects this crystal structure resembles that of native extended PmScsC shown in (b) (PDB entry 5id4). However, the extended helix linking the trimerization domain to the catalytic domain is shorter in PmScsCÁLinker owing to deletion of the linker [cyan in (b)]. In both (a) and (b) the proteins are coloured using the scheme used in Fig. 1. (c) The electrostatic surface potential of the catalytic domain of the PmScsCÁLinker protomer structure. Electrostatic calculations were performed using APBS (Jurrus et al., 2018) and were contoured at À7.5 kT e À1 (red) and +7.5 kT e À1 (blue). A solid orange circle indicates the position of the catalytic cysteines; black ovals surround the regions of the catalytic domains in close contact with adjacent catalytic domains in the PmScsCÁLinker crystal structure. (d) The hydrophobic core of the trimerization stem of PmScsCÁLinker; side chains of residues contributing to the hydrophobic core are coloured pink. (e) Amino-acid residues mediating side-chain polar and electrostatic interactions between the trimerization stems of two PmScsCÁLinker protomers. Black dashed lines represent hydrogen bonds and the residues involved are labelled.
other (Furlong et al., 2017). Our present results show that PmScsCÁLinker also forms a trimer and is also inactive as a protein disulfide isomerase. We therefore anticipated that the lack of protein disulfide isomerase activity was a consequence of reduced rigid-body motion of the catalytic domain, and sought to confirm this in solution using small-angle scattering.
Unexpectedly, the initial SAXS data for PmScsCÁLinker displayed a strong dependence on concentration, consistent with an equilibrium between different oligomeric forms of PmScsCÁLinker (monomer, dimer and trimer). This result raised the possibility that at low concentrations the wild-type protein may also be present as a mixture of monomers, dimers and trimers, and this might also impact on its function or its regulation. To investigate this possibility further, SAXS data were collected from native PmScsC (Fig. 7) and PmScsCÁLinker (Fig. 8) at concentrations between 0.1 and 10.0 mg ml À1 (Table 2). At the lowest concentration (0.1 mg ml À1 ) the shape of the p(r) function for native PmScsC differed from the shapes of those at higher concentrations, indicating some dissociation of the wild-type protein. A number of approaches were trialled for the analysis, but fitting a linear combination of monomer and trimer scattering to the data at each concentration was found to perform best ( Fig. 7 and Table 3). The results presented in Table 3 show generally good agreement for the two modelling approaches that were taken (global fit of a single equilibrium constant to all data sets and fitting of each data set independently). However, it can be seen that there is some disagreement at low concentrations, where the global fit predicts a much lower proportion of monomer in solution than does the independent fitting. Thus, it appears that the higher concentration data sets with higher signal-to-noise ratios may have a disproportionate impact on the magnitude of the determined equilibrium constant. Nonetheless, both methods of analysis show that the bulk of the material is present as the trimeric form at all concentrations. For PmScsCÁLinker, the results are very different; the shape of the p(r) curve changes significantly with each concentration evaluated. Again, we trialled a number of approaches to fit these data, but fitting a linear combination of monomer, dimer and trimer scattering to the data at each concentration was found to perform best ( Fig. 8 and Table 3). While the analysis worked well, it is limited by the fact that the structures of the monomer and dimer forms in solution are not known. In each case, we made a simplistic assumption that a monomer adopted the structure of a single protomer from the corresponding crystal structure and a dimer could be represented by removing a single protomer from the trimer crystal structure. We also know that native PmScsC is dynamic in solution (Furlong et al., 2017). While the model curves provide a reasonable fit to the scattering data, there are systematic SAXS data for native PmScsC. (a) Scattering data collected at $0.1 mg ml À1 (blue; 2 = 2.55), $0.5 mg ml À1 (orange; 2 = 7.05), $2.0 mg ml À1 (green; differences between the experimental data and the model, especially at higher concentrations. The limitations of the modelling (lack of accurate models, dynamic motion) explain most of this variation.
Overall, the SAXS analysis suggests that only 1-21% of native PmScsC is present as monomers in solution at the lowest concentration of 0.1 mg ml À1 ; we conclude that the trimer is the dominant form of the native protein in solution even at low concentrations. For the variant PmScsCÁLinker it is a different story; the protein is in equilibrium between different oligomerization states. At 0.1 mg ml À1 approximately 97% of the PmScsCÁLinker solution is present as monomers and $3% is present as dimers, with negligible trimers. At 10 mg ml À1 these proportions change to $40% monomers, $48% dimers and $12% trimers.

Discussion
This study focused on characterizing two variants of PmScsC, which in its native form is a highly dynamic, trimeric disulfide isomerase. The native PmScsC protomer consists of an N-terminal trimerization stem with an 11-amino-acid flexible linker connected to a C-terminal thioredoxin-fold catalytic domain harbouring a CXXC active site. To better understand the role of the trimerization stem and flexible linker, we designed the variants PmScsCÁN and PmScsCÁLinker. PmScsCÁN lacks the first 41 amino acids of the native PmScsC construct; its loss of disulfide isomerase activity and gain of dithiol oxidase activity has been reported previously (Furlong et al., 2017). PmScsCÁLinker is newly reported here and lacks the 11 amino acids that form the flexible linker. Here, we have reported the structure of PmScsCÁN and analysed the structure and biochemical function of PmScsCÁLinker.
The X-ray crystal structure of PmScsCÁN agrees with previous MALS data showing that this variant is monomeric (Furlong et al., 2017). The crystal structure is also consistent with solution SAXS data and the low-resolution model obtained from SAXS analysis. Although the PmScsCÁN crystal structure is very similar to the structures of the catalytic domains in each of the three crystal structures of native PmScsC, unlike the native enzyme it lacks disulfide isomerase (disulfide-shuffling) activity and instead has dithiol oxidase (disulfide-forming) activity. The PmScsC catalytic domain has an acidic catalytic cysteine and an energetically unfavourable disulfide (Furlong et al., 2017), which are also features of the archetypal but monomeric dithiol oxidase EcDsbA, so it is perhaps not surprising that the removal of the N-terminal trimerization residues leads to oxidase activity in a model assay. Moreover, the structural similarity between monomeric SAXS data for PmScsCÁLinker. (a) Scattering data collected at $0.1 mg ml À1 (blue; 2 = 1.27), $0.5 mg ml À1 (orange; 2 = 2.19), $2.0 mg ml À1 (green; 2 = 6.89), $5.0 mg ml À1 (red; 2 = 37.59) and $10.0 mg ml À1 (grey; 2 = 177.29). The overlaid line (black) on each scattering curve is the fitted linear combination of monomer and trimer scattering curves. (b) The pair-distance distribution function, p(r), derived from the scattering data normalized by concentration shows that the oligomeric composition of the solution is highly dependent on concentration. (c) The fraction of protein present as monomer, dimer and trimer, derived from the fitted global optimization models, shows that the monomeric species is dominant in solution at low concentrations. The concentration of trimer rises slowly and and is predicted to reach a protein fraction of $0.2 at 38 mg ml À1 , which is the concentration at which crystallization was performed.
PmScsCÁN and the catalytic domains of native trimeric PmScsC supports the notion that it is the N-terminal stem, which defines whether the catalytic domain is monomeric or trimeric, that is the determinant of isomerase or oxidase activity. Put simply, the differential activity that we observe between native PmScsC and variant PmScsCÁN is not a specific feature of the catalytic domain because the same catalytic domain has the propensity for oxidase (monomeric protein) or isomerase (trimeric protein) activity depending on its context. The isomerase activity of PmScsC requires both its N-terminal trimerization domain and its C-terminal catalytic domain, in the same way that the isomerase activity of EcDsbC requires its dimerization domain and its catalytic domain (Sun & Wang, 2000).
Our SAXS analysis showed that native PmScsC is a trimer in solution, even at low concentrations. However, SAXS analysis of PmScsCÁLinker showed that it exists predominately as a monomer or a low-affinity dimer (K d of $200 mM) in preference to a trimer in solution, even though it crystallizes in the trimeric form. Together, these data suggest that removal of the flexible linker of PmScsC between the trimerization stem and catalytic domains has the unintended consequence of reducing the propensity to form a stable trimer in solution.
The crystal structure of PmScsCÁLinker provides insight into why the trimeric form of the variant may be energetically unfavourable compared with the monomeric and dimeric forms in solution. The trimerization stem interactions of PmScsCÁLinker are similar to those observed in the crystal structures of the native enzyme, but all three catalytic motifs are in closer proximity to each other than in any of the native PmScsC crystal structures. This close packing suggests the possibility of unfavourable contacts between catalytic domains in PmScsCÁLinker that may contribute to the instability of the trimer. Indeed, the crystal structure of PmScsCÁLinker reveals a close association of a positively charged patch near the catalytic motif on the accessible surface of one protomer with a neutral/hydrophobic patch on another. In the crystal packing of the PmScsCÁN structure (which does not have the constraint of the trimerization stem), there are no contacts involving the positively charged surface patch near the catalytic motif, suggesting that packing against this site is not favoured. Thus, the formation of the PmScsCÁLinker trimer is likely to be dependent on the balance of favourable interactions formed between the trimerization stems and unfavourable interactions between the catalytic domains. At lower concentrations the unfavourable interactions between the catalytic domains may override the favourable interactions, but at higher concentrations, with more molecules in close contact, it may be more thermodynamically favourable for the hydrophobic portions of the trimerization helices to interact with each other. This interplay between interactions of the catalytic domains and trimerization stems could explain why the oligomeric state of PmScsCÁLinker in solution alters with concentration of the protein. Taken together, the crystal structure and SAXS analysis of PmScsCÁLinker suggests that the 11-amino-acid linker plays a dual role by (i) acting as a spacer between the trimerization and catalytic domains to enable the formation of a stable trimer and (ii) providing the necessary functional dynamics and cooperativity between the catalytic domains (Furlong et al., 2017).
Removal of the 11-amino-acid flexible linker from PmScsC impacts on the activity of the protein: unlike the native protein, PmScsCÁLinker has no disulfide isomerase activity. Structural characterization of PmScsCÁLinker revealed two potential reasons for this result. Firstly, at the concentration used in the assay (10 mM, $0.2 mg ml À1 ) it is likely that most of the protein would be monomeric (Fig. 8c), and we know from PmScsCÁN variant studies that monomeric protein has low protein disulfide isomerase activity (Furlong et al., 2017). Secondly, even if sufficient trimeric PmScsCÁLinker were present in solution, the crystal structure suggests there would be limited space between the catalytic domains to bind a misfolded protein substrate and possibly limited motion to allow disulfides to be shuffled. It is interesting to note that PmScsCÁLinker is inactive in the dithiol oxidase assay. Given that the protein would be predominantly monomeric at the concentration tested in this assay (80 nM or 0.002 mg ml À1 ; Fig. 8c) we expected that, like monomeric PmScsCÁN, it would have oxidase activity. One explanation for this result is that the N-terminal hydrophobic regions of PmScsCÁLinker somehow occlude the substrate-binding site that is available in monomeric PmScsCÁN.
Overall, our structural and functional characterization of PmScsCÁN and PmScsCÁLinker provides new insight into the importance of unique structural features of the trimeric disulfide isomerase PmScsC. We confirmed previous assumptions that disulfide isomerases are similar to but differentiated from dithiol oxidases by requiring both an oligomerization domain and a catalytic domain. We further propose that the flexible linker of PmScsC is important not just as a shapeshifting peptide, but also as a spacer to enable functional flexibility between the catalytic domains. We suggest that this functional flexibility between the catalytic domains is necessary for the cooperative enzymatic activity of the disulfide isomerase PmScsC. Finally, on a more general note, we found that deleting amino acids from a protein can have unintended consequences. In this case, the deletion of the flexible linker unexpectedly changed the oligomerization kinetics of the protein, even though the region directly involved in trimerization remained intact.