Structure of Yeast Sulfhydryl Oxidase Erv1 Reveals Electron Transfer of the Disulfide Relay System in the Mitochondrial Intermembrane Space*

Background: Mia40 is regenerated by the sulfhydryl oxidase Erv1 in the disulfide relay system. Results: Crystal structures of the Erv1 core domain and full length of Erv1 were determined. Conclusion: The Erv1 N-terminal amphipathic helix is critical for electron transfer from Mia40 to the core redox center of Erv1. Significance: This is the first structural snapshot of the electron transfer process in Mia40-Erv1 disulfide relay system. The disulfide relay system in the mitochondrial intermembrane space drives the import of proteins with twin CX9C or twin CX3C motifs by an oxidative folding mechanism. This process requires disulfide bond transfer from oxidized Mia40 to a substrate protein. Reduced Mia40 is reoxidized/regenerated by the FAD-linked sulfhydryl oxidase Erv1 (EC 1.8.3.2). Full-length Erv1 consists of a flexible N-terminal shuttle domain (NTD) and a conserved C-terminal core domain (CTD). Here, we present crystal structures at 2.0 Å resolution of the CTD and at 3.0 Å resolution of a C30S/C133S double mutant of full-length Erv1 (Erv1FL). Similar to previous homologous structures, the CTD exists as a homodimer, with each subunit consisting of a conserved four-helix bundle that accommodates the isoalloxazine ring of FAD and an additional single-turn helix. The structure of Erv1FL enabled us to identify, for the first time, the three-dimensional structure of the Erv1NTD, which is an amphipathic helix flanked by two flexible loops. This structure also represents an intermediate state of electron transfer from the NTD to the CTD of another subunit. Comparative structural analysis revealed that the four-helix bundle of the CTD forms a wide platform for the electron donor NTD. Moreover, computational simulation combined with multiple-sequence alignment suggested that the amphipathic helix close to the shuttle redox enter is critical for the recognition of Mia40, the upstream electron donor. These findings provide structural insights into electron transfer from Mia40 via the shuttle domain of one subunit of Erv1 to the CTD of another Erv1 subunit.

intermembrane spaces (IMS) 2 of mitochondria and endoplasmic reticulum are the two major locations for the introduction of disulfide bonds (1). The IMS has a dedicated disulfide relay system to introduce disulfide bonds into the small cysteine-rich substrate proteins (2), such as small Tim proteins and copper chaperone Cox17, which are nuclearly encoded and cytosolically synthesized (3,4). These substrate proteins are characterized by a relatively low molecular mass in the range of 8 -17 kDa and a conserved motif of cysteine pairs. These are twin CX 3 C or twin CX 9 C motifs that are crucial for the import of preprotein and accumulation of mature proteins in the IMS (5)(6)(7). Newly synthesized unfolded substrate proteins would pass through the translocase of the outer membrane and form a mixed disulfide bonded intermediate with Mia40 (mitochondrial intermembrane space import and assay/oxidoreductase 40) in the IMS (8 -10). Mia40 is a conserved oxidoreductase that is soluble in mammals and plants but membrane-anchored in fungi (9,11). It harbors an conserved redox-active motif of -CPC-CX 9 C-CX 9 C- (8,11), using a CPC site to form an intermolecular disulfide bond with substrate proteins (12,13). In the disulfide exchange reaction, a disulfide bond is introduced into the substrate protein, accompanied by the release of the reduced Mia40, which is reoxidized/regenerated to a functional state by the sulfhydryl oxidase Erv1 (essential for respiration and viability/FAD-linked sulfhydryl oxidase 1) (2,14). Thereafter, the reduced Erv1 passes the electron to either cytochrome c or molecular oxygen (15)(16)(17)(18). Together, Mia40 and Erv1 are two essential components of the disulfide relay system that is of crucial importance for mitochondrial biogenesis (2,19,20).
The FAD-linked sulfhydryl oxidase Erv1 (EC 1.8.3.2) is essential for the respiration and vegetative growth of the yeast (21,22). A number of Erv1 homologs have been characterized in plants (23), mammals (24,25), and double-stranded DNA viruses (26,27). The mammalian homologs are called augmenters of liver regeneration (ALRs). All Erv1/ALR family members share a conserved core domain harboring a CXXC motif (the core redox center), juxtaposed with FAD and involved in redox reactions. To date, the structures of core domains have been determined for human ALR (28,29), Arabidopsis thaliana Erv1 (30), Rattus norvegicus ALR (31), and Saccharomyces cerevisiae Erv2 (32). All exist as a homodimer, with each subunit composed of a four-helical bundle that accommodates the isoalloxazine ring of FAD with an additional single-turn helix.
In addition to the conserved core redox center, Erv1/ALR proteins, except for the viral homologs, possess another cysteine pair. This is at the N-terminal domain (NTD) in fungi and mammals and at the C-terminal segment in plants (33). Genetic studies demonstrated that the N-terminal CXXC motif of yeast Erv1 was required for in vivo functions (34). In fact, the NTD of yeast Erv1 is necessary and sufficient for interaction with Mia40. Moreover the N-terminal cysteine pair is required for the formation of a mixed disulfide intermediate with Mia40 (35). Because of its role in forwarding electrons from Mia40 to the C-terminal core domain (CTD), the NTD is termed the shuttle domain, and the CXXC motif at the NTD is termed the shuttle redox center (16,20,33,35).
To gain insights into the structural basis of this electron transfer process, we determined the structure of the CTD at 2.0 Å resolution and the structure of the C30S/C133S double mutant of full-length Erv1 (Erv1FL) at 3.0 Å resolution. The structure of the N-terminal shuttle domain and its interactions with the CTD led us to propose a putative model of electron transfer from Mia40 via the shuttle domain of one subunit to the CTD of another subunit of Erv1.

EXPERIMENTAL PROCEDURES
Overexpression and Purification of Erv1 and Mutants-The coding sequences of the intact yeast ERV1/YGR029W and the C-terminal core domain (Asp 86 -Glu 189 , designated as Erv1CTD) were amplified by PCR using S. cerevisiae S288c genomic DNA as the template and cloned into a pET28aderived vector, respectively. The constructs add a hexahistidine tag to the N terminus of the recombinant protein, which were overexpressed in Escherichia coli BL21 (DE3) (Novagen, Madison, WI) strain using 2ϫ YT culture medium. Expression was started by adding 0.2 mM isopropyl-␤-D-thiogalactoside, and the cells continued growing for another 20 h at 16°C before harvesting. The cells were harvested by centrifugation at 8000 ϫ g for 10 min and resuspended in lysis buffer (20 mM Tris-HCl, pH 8.0, 200 mM NaCl). After 5 min of sonication (power-on for each 1 s with an interval of 3 s in a total time of 20 min) and centrifugation at 12,000 ϫ g for 25 min, the supernatant containing the soluble target protein was collected and loaded to a HiTrap nickel-chelating column (GE Healthcare) equilibrated with binding buffer (20 mM Tris-HCl, pH 8.0, 200 mM NaCl). The target protein was eluted with 300 mM imidazole buffer and further loaded onto a Superdex 75 column (GE Healthcare) equilibrated with 20 mM Tris-HCl, pH 8.0, 200 mM NaCl (20 mM sodium citrate, pH 5.38, 50 mM NaCl for Erv1CTD). Fractions containing the target protein were pooled and concentrated to 10 mg/ml by ultrafiltration (Millipore; 10-kDa cut-off). The purity of proteins was estimated by SDS-PAGE, and the proteins were stored at Ϫ80°C. For expression of the N-terminal domain of Erv1 (Met 1 -Asp 83 , designated as Erv1NTD), the nucleotide sequence was PCR-amplified and cloned in pGEX-4T-2 expression vector. The Erv1NTD was expressed and purified as previously described (36). The mutant proteins were expressed, purified, and stored in the same manner as the wild type.
Crystallization, Data Collection, and Processing-Before crystallization, Erv1FL was incubated with 5 mM GSSG:GSH at a molar ratio of 3:1 for 1 h at 4°C. Crystals of Erv1CTD and Erv1FL were grown by hanging drop vapor diffusion at 30 and 16°C, respectively, with the initial condition by mixing 1 l of the 10 mg/ml protein sample with equal volume of reservoir solution (Erv1CTD: 25% (w/v) polyethylene glycol 3350, 0.2 M ammonium acetate, 0.1 M Tris-HCl, pH 8.5; Erv1FL: 15% (w/v) polyethylene glycol 6000, 0.1 M HEPES, pH 8.0). The crystals of Erv1CTD appeared in 3 days, whereas that of the Erv1FL appeared in ϳ2 days. The crystals were transferred to cryoprotectant (reservoir solution supplemented with 25% (v/v) glycerol) and flash-cooled at 100 K in liquid nitrogen. Both data were collected at a radiation wavelength of 0.979 Å at the Shanghai Synchrotron Radiation Facility (Shanghai Institute of Applied Physics, Chinese Academy of Sciences), using the Beamline BL17U at 100 K with a MX-225 CCD (Marresearch). The data sets of Erv1CTD was processed using the HKL2000 package (37), and that of Erv1FL was processed with iMosflm.
Structure Determination and Refinement-Both crystal structures were determined by the molecular replacement method with MOLREP (38) in the CCP4 suite (39) using the coordinates of R. norvegicus FAD-dependent sulfhydryl oxidase (Protein Data Bank code 1OQC) (31) as the search model. Refinement was carried out using REFMAC5 (40) and COOT (41). The overall assessment of model quality was performed using MOLPROBITY (42). The final atomic coordinates and structure factors were deposited in the Protein Data Bank under the accession codes 4E0H and 4E0I. The crystallographic parameters of the structures are listed in Table 1. All of the structure figures were prepared with PyMOL (43).

RESULTS AND DISCUSSION
Overall Structure of the Highly Conserved Erv1CTD-The yeast Erv1 has two domains. The highly conserved CTD follows a flexible N-terminal domain (20). We first determined the structure of Erv1CTD, which is from Asp 86 to Asp 188 . Each asymmetric unit consists of one Erv1CTD molecule, which adopts an all-␣ overall structure (Fig. 1A), similar to previous core domain structures. For example, the root mean square deviation (RMSD) between Erv1CTD and human ALR is only 0.69 Å over 104 C␣ atoms. Erv1CTD consists of a four-helix bundle (helices ␣1-␣4) and an additional single-turn helix ␣5 perpendicularly packed against the bundle. The C terminus of ␣4 extends out from the N terminus of ␣3, and the FAD binds at the center of the four-helix bundle. The core redox center (Cys 130 -Cys 133 ) is in proximity to the isoalloxazine ring of FAD, and the structural disulfide pair Cys 159 -Cys 176 is proximal to the adenine ring of FAD (32,34).
Erv1CTD form a noncovalently linked homodimer along a crystallographic 2-fold axis (Fig. 1A). Gel filtration chromatography also suggests that Erv1CTD most likely exists as a dimer in solution with an estimated mass of ϳ27 kDa (Data not shown). Helices ␣1 and ␣2 (Asp 86 -Ser 104 and Asp 111 -Ile 127 ) from each subunit form a bundle of four helices with a buried interface of 1830 Å 2 . Residues Leu 90 , Leu 97 , Val 101 , Phe 120 , Ile 123 , Phe 124 , Ile 127 , and Pro 129 from each subunit form hydrophobic patches at the center of the dimeric interface (Fig. 1B). In addition, residues Ser 104 , Glu 116 , Gln 119 , His 126 , and Pro 129 contribute to the intermolecular hydrogen bonds and salt bridges (Fig. 1C). Multiple sequence alignment showed that most hydrophobic and polar residues involved in the dimeric interface are conserved (Fig. 1D). This conserved dimer interface provides the structural basis for the intersubunit electron transfer, in agreement with previous results that the dimerization promotes the functionally essential intersubunit disulfide exchange reaction (20,30,32).
Favored Disulfide Bond between NTD and CTD-Erv1 contains two redox centers (Cys 30 -Cys 33 at the NTD and Cys 130 -Cys 133 at the CTD), both of which are indispensable for the disulfide relay/oxidation activity (17). The importance of the four cysteine residues of the two CXXC motifs has been assessed, and a mutation of Cys 30 to Ser 30 was not harmful to the in vivo function of Erv1 (20). This indicates that Cys 33 of NTD might form a mixed disulfide with the core redox center in the electron transfer process. However, the cysteine of the core redox center (Cys 130 or Cys 133 ) that is favored by Cys 33 remains unknown. To assign this cysteine, we overexpressed and purified three mutants of the NTD and CTD: NTD-C30S, CTD-C130S, and CTD-C133S. The reaction efficiency of the two CTD mutants toward NTD-C30S was semiquantatively compared using SDS-PAGE. A complex of NTD-C30S with CTD-C133S ( Fig. 2A, eleventh lane), but not CTD-C130S ( Fig.  2A, tenth lane), was observed. Moreover, the oxidized Erv1C30S/C133S mutant formed more mixed disulfide dimers than the Erv1C30S/C130S mutant (Fig. 2B). Thus, we concluded that Cys 130 is the favored cysteine for the disulfide intermediate with the N-terminal Cys 33 . These results are in agreement with the somewhat buried position of Cys 133 , whereas Cys 130 is relatively exposed to the solvent, as shown in the CTD structure (Fig. 1A). Thus, it should be easier for the NTD to access Cys 130 than Cys 133 .
Overall Structure of Erv1FL-The failure of our initial attempt to crystallize the full-length Erv1 might be due to the high flexibility of the NTD. Because the NTD might be fixed by the core domain via a mixed disulfide between Cys 33 and Cys 130 , the C30S/C133S double mutant of full-length Erv1 was incubated with 5 mM GSSG:GSH at a molar ratio of 3:1. This provides an oxidation environment for forming a mixed disulfide bond between the favored cysteine residue Cys 33 and Cys 130 from a neighboring subunits. The sample was crystallized, and the structure was determined at 3.0 Å resolution. Each asymmetric unit contains three molecules of Erv1 (subunit A, B, and C). Subunits A and B could form a homodimer along a noncrystallographic 2-fold axis with an interface of 1870 Å 2 , mainly contributed by the conserved core domain, and the interface is the same as the dimer interface in the Erv1CTD structure. Subunits A and C has a buried interface of 640 Å 2 that might be due to the crystal packing. Moreover, subunit C forms a dimer with a subunit related by a crystallographic 2-fold axis. Residues Glu 71 -Asp 188 could be traced in all three molecules, whereas the segment with residues Glu 14 -Ser 47 could be fitted to the electron density map of only a single molecule. This segment of the NTD is folded into an ␣ helix (␣0, Leu 36 -Ser 47 ) following a long defined loop (Glu 14 -Thr 35 ), and locates proximally to the core redox center of subunit B (Fig. 3A). The shuttle redox center (Cys 30 -Cys 33 ) is on a loop very close to first turn of helix ␣0. Although residues Glu 48 -Asp 83 form a flexible linker between the CTD and helix ␣0, residues Met 1 -Gln 13 and Gln 48 -Ser 70 are not visible in the final 2F o Ϫ F c electron density map, presumably because of their high flexibility. Based on cross-linking results shown in Fig. 2B and previous reports on electron transfer from the reduced shuttle redox center to the core redox center of another subunit in the dimer (20,30), we assign the visible NTD to subunit A (Fig. 3A). Helix ␣0 of subunit A is linked to the core domain of subunit B via an intermolecular disulfide bond (Cys 33 -Cys 130 Ј).
Structural Basis for the Intersubunit Electron Transfer-The structure of Erv1FL captures an intermediate state of electron transfer from one NTD to the CTD of another subunit (designated CTDЈ) via a mixed disulfide bond between Cys 33 and Cys 130 Ј (Fig. 3A). The 2F o Ϫ F c electron density map clearly displays a disulfide bond-linked complex in the asymmetric unit (Fig. 3B). NTD binds to CTDЈ at the surface adjacent to the

The Intersubunit Electron Transfer of Yeast Erv1
OCTOBER 12, 2012 • VOLUME 287 • NUMBER 42 core redox center with a total interface area of 880 Å 2 (450 Å 2 for NTD and 430 Å 2 for CTDЈ). This interface has a typical area for redox protein complexes because of their short-lived interactions (44). In addition to the intersubunit disulfide bond, five hydrogen bonds are involved in stabilizing the conformation of NTD (Fig. 3C). In particular, the carbonyl oxygen of Ser 32 and Asn 34 forms hydrogen bonds with the amide nitrogen of Val 87 Ј, respectively. The amide nitrogen of Leu 36 forms a hydrogen  bond with Asp 86 Ј-O␦2, whereas Arg 31 -N1 makes a hydrogen bond with the carbonyl oxygen of Pro 84 Ј. These four hydrogen bonds close to the shuttle redox center form a network of seven residues, five of which contribute with the main chain atoms, whereas residues Arg 31 and Asp 86 Ј donate their side chain atoms. Sequence alignment showed that Arg 31 and Asp 86 are conserved in Erv1/ALR and homologs (data not shown). In addition, Asp 24 -O␦2 forms a hydrogen bond with Trp 132 Ј-N⑀1 of the CTDЈ. Moreover, hydrophobic contacts between a hydrophobic patch (Val 87 Ј and Trp 132 Ј) of the CTDЈ and the complementary side (Ile 21 , Ile 22 , and Thr 35 ) of the NTD also contribute a part to the interface (Fig. 3D).
The overall structure of CTDЈ of Erv1FL subunit B is quite similar to that of subunit A and the isolated CTD, with RMSD of 0.32 and 0.69 Å over 104 C␣ atoms, respectively. However, when approaching the NTD, helix ␣3 (Cys 130 -Glu 143 ) at the core redox center rotates outwards against the isoalloxazine ring of FAD at an angle of ϳ6.0°along its C terminus (Fig. 3E). In addition, the other three helices of the four-helix bundle also shift slightly outwards. These conformational changes lead to a wider bundle mouth to interact with the approaching NTD. The segments that locate on the top of the four helices of CTDЈ constitute a platform that plays a crucial role in the NTD interaction.
Although the NTD of subunit A is cross-linked to the CTD of subunit B, the traceable segment Glu 14 -Ser 47 exhibits relatively high B factors. This structural flexibility is considered necessary for recognition of both Erv1CTD and Mia40 (29). Moreover, the linker between the shuttle redox center and the core domain displays a much higher flexibility, and most of the linker could not be traced in the electron density map. This highly flexible linker enables the shuttle redox center to easily flip between the Erv1CTD and Mia40.
A Putative Binding Model between Erv1 and Mia40-After determining the structure of Erv1NTD, we attempted but failed to obtain a crystal of Erv1NTD in complex with Mia40 by a similar cross-linking strategy. Alternatively, we simulated a model of the Mia40-Erv1NTD complex using the program HADDOCK (45), based on our Erv1FL structure and the previously reported structure of Mia40 (Protein Data Bank code 2ZXT, without the maltose-binding protein tag). This was driven by interaction restraints between the active site residues of Erv1 and Mia40, as defined by the program WHISCY (46). Among the 15 output clusters, the cluster of lowest energy with eight members satisfied the best interaction restraints and had the largest buried solvent-accessible interface area of ϳ1050 Å 2 (595 Å 2 for Erv1NTD and 455 Å 2 for Mia40). The overall backbone RMSD of 0.5 Ϯ 0.3 Å for the eight members indicated that

The Intersubunit Electron Transfer of Yeast Erv1
OCTOBER 12, 2012 • VOLUME 287 • NUMBER 42 the model of Mia40-Erv1NTD was somewhat reliable. In the model, helix ␣0 makes extensive interactions with the hydrophobic cleft of Mia40 (Fig. 4A). This cleft is composed of a cluster of highly conserved hydrophobic residues Met 302 , Phe 311 , Phe 315 , Phe 318 , and Met 337 (Fig. 4B). These residues are also involved in recognizing the CX 9 C and CX 3 C substrates (10), suggesting that Mia40 uses the same site to bind both protein substrates and the electron acceptor Erv1. Notably, this model is in accordance with the previous report that Erv1 competitively binds to the substrate-binding site on Mia40 (29).
The docked interface on Erv1NTD that comprises hydrophobic residues Leu 36 , Leu 37 , and Phe 39 (Fig. 4B) is in agreement with the results of Banci et al. (29). They used NMR titration to determine that the CRACVDFKTWM segment of ALR (homologous to the Erv1 CRSCNTL 36 L 37 DF 39 Q segment) is critical for recognition of Mia40. Using mutagenesis in combination with complementation assays, they confirmed that the hydrophobic residues downstream of the Erv1 shuttle redox center (Leu 36 , Leu 37 , and Phe 39 ) play a vital role in complex formation with Mia40 in vitro and in vivo. These residues on the amphipathic helix ␣0, as shown in the structure of Erv1FL, are solvent-exposed before being recognized by Mia40. Once fitted into the hydrophobic cleft of Mia40, helix ␣0 will enable Cys 33 of Erv1NTD to come as close as ϳ6.3 Å to Cys 298 of Mia40.
With slight conformational changes, these two cysteine residues can form a transient mixed disulfide bond.
Universal Mode of Electron Transfer from Mia40 to Erv1 Shuttle Domain in Animals and Fungi-Erv1 homologs have a highly conserved core domain but variable shuttle domains (33). To find the probable original shuttle domain and its distribution, the sequence of yeast Erv1 was used in a BLAST search against the nonredundant protein sequences database. We chose 10 representatives of various species and compared the residues around the shuttle redox centers (Fig. 5). Both fungi and animals have an N-terminal shuttle domain with a highly conserved redox center (CXXC) but variable linkers between the shuttle and the CTD. Similar to the residues that constitute the amphipathic helix ␣0 in the yeast Erv1NTD, the corresponding residues in the animal homologs were also predicted to have a high propensity to form a helix (residues FKTWM in human ALR) (29). In yeast and human Erv1/ALR, the hydrophobic residues are somewhat aligned on one side of the amphipathic helix ␣0, indicating a universal hydrophobic interaction pattern between Mia40 and Erv1/ALR from fungi and animals.
In fungi and animals, the N-terminal shuttle redox center of Erv1/ALR functions as an antenna stretched from the core domain. This antenna is held by the CTD with a flexible linker of varied lengths. Of note, the linker is gradually truncated during evolution from lower to higher organisms (Fig. 5). The linker of yeast Erv1 is composed of 38 amino acids, whereas the linker of human ALR is of only 14 amino acids. A shorter linker might be helpful in enhancing electron transfer efficiency from the shuttle to the core redox center.
A Working Model of Electron Transfer Process in the Mia40-Erv1 Disulfide Relay System-Based on our structural analyses and previous reports, we propose a working model of electron transfer in the Mia40-Erv1 disulfide relay system (Fig. 6). To simplify the illustration, we show that only one of two shuttle domains of the Erv1 dimer is working with a molecule of Mia40 during the catalysis cycle. In our model, 1) after the oxidation and release of a given substrate protein, the hydrophobic cleft of the reduced Mia40 is exposed (12). Meanwhile, the amphipathic helix ␣0 at the shuttle domain of the oxidized Erv1 is recognized by the hydrophobic cleft of Mia40. 2) The shuttle redox center forms an intermolecular disulfide bond with Mia40 (between Mia40-Cys 298 and Erv1-Cys 33 ) (12,20). Consequently, this transient disulfide bond is exchanged upon the attack of Mia40-Cys 296 to release oxidized Mia40. Regenerated Mia40 is ready to oxidatively refold another substrate protein.
The electron transfer from Mia40 to the shuttle domain of Erv1 requires mechanisms to overcome the thermodynamically unfavorable redox gradient. This might be driven by conformational changes of Erv1NTD at different redox states (17).

The Intersubunit Electron Transfer of Yeast Erv1
OCTOBER 12, 2012 • VOLUME 287 • NUMBER 42 JOURNAL OF BIOLOGICAL CHEMISTRY 34967 regenerate an oxidized shuttle redox center. Simultaneously, the core redox center of Erv1 is reduced. This process is spontaneously driven by differences in redox potential between the shuttle and core redox centers (17,20). 5) Electron transfer from the core redox center to FAD does not require a conformational change (the distance from the flavin C4␣ to the thiol group of Cys 133 is ϳ3.3Å) but must overcome the unfavorable redox gradient (17). This process might be driven by coupling to the efficient downstream electron flow to cytochrome c (47). 6) The reduced FAD efficiently transfers the electron to the most favorable physiological electron acceptor, cytochrome c, which passes the electron to the respiratory chain (15,17,18). Thus, Erv1 is ready for another electron transfer cycle.
Conclusions-This work captured an intermediate structure of the N-terminal shuttle domain cross-linked to the core domain of Erv1 via an introduced disulfide bond between Cys 33 of one subunit and Cys 130 Ј of another subunit. The frozen position of the highly flexible shuttle domain enabled us to determine a model of electron transfer from the upstream Mia40 to the downstream cytochrome c via a recognition pattern similar to the interaction between Mia40 and its substrate protein. These findings provide for the first time structural insights into the overall Mia40-Erv1 disulfide relay system.