How Periplasmic Thioredoxin TlpA Reduces Bacterial Copper Chaperone ScoI and Cytochrome Oxidase Subunit II (CoxB) Prior to Metallation*

Background: α-Proteobacteria, extant relatives of mitochondria, are model organisms for studying assembly of bacterial and mitochondrial metalloenzymes. Results: Periplasmic thioredoxin TlpA is a specific reductant for Cu-chaperone ScoI and cytochrome oxidase subunit II (CoxB). Conclusion: Cysteines in the Cu-binding sites of ScoI and CoxB must be reduced prior to metallation. Significance: Structures of TlpA-ScoI and TlpA-CoxB intermediates reveal mechanistic details of the reduction process. ABSTRACT Two critical cysteine residues in the copper-A site (Cu A ) on subunit II (CoxB) of bacterial cytochrome c oxidase lie on the

containing enzyme complex, the question of how its subunits plus heme and copper cofactors are assembled in the membrane has been answered only partially. Particularly intricate is the biogenesis of the membrane-anchored subunit II (CoxB) because it carries a binuclear Cu-Cu center (CuA) on a domain that faces the periplasmic side of the cytoplasmic membrane in Gram-negative bacteria or the intermembrane space in mitochondria (1,2). Copper ions must therefore be delivered to these compartments for Cu A assembly. The CuA center is the entry point of electrons coming from reduced cytochrome c, which are then guided to the membrane-integral subunit I via the low-spin heme A to the high-spin heme A 3 -Cu B center, the site of oxygen reduction (4,5). The two copper ions in the oxidized form of the CuA center have an overall valence of +3 (+1.5/Cu ion), and cytochrome c reduces them to +2 (+1/Cu ion) (6). Subunit II possesses a highly conserved amino-acid-sequence motif for copper binding (HX34CXEXCX3HXXM) where the six ligands (underlined) are provided by the sulfur atoms of the methionine and the two cysteines, one imidazole nitrogen atom each of the two histidines, and the peptide-bond carbonyl oxygen from the glutamate (1,2).
When the CuA-binding domain of subunit II is secreted from the cytoplasm to the periplasm in the course of bacterial protein synthesis and membrane insertion, the two neighboring cysteine thiols (CX3C) in the aforementioned motif are prone to oxidation because the periplasm is an oxidizing environment (7,8). Regardless of whether intramolecular disulfide-bond formation occurs by air oxidation or is catalyzed by a DsbAlike dithiol oxidase (9,10), such a constellation precludes complexation with Cu ions. Enzymecatalyzed re-reduction of such a disulfide bond (an off-pathway intermediate to Cu A assembly) to the dithiol form would thus appear to be necessary for copper ion binding to subunit II. A strong candidate for a periplasmic CoxB reductase is the thioredoxin-like protein TlpA (11,12), which has been discovered first in Bradyrhizobium japonicum, a member of the α-proteobacteria and one of the model organisms for studying bacterial and mitochondrial hemo-and metalloproteins (13)(14)(15). TlpA has a negative redox potential (E 0 ' = -256 mV) which predicts it is a reductant in vivo (16,17). The protein is anchored to the cytoplasmic membrane, with its thioredoxin domain facing the periplasm and, hence, ideally positioned to reduce membrane-associated substrates in this compartment. A tlpA knock-out mutant of B. japonicum is defective for cytochrome oxidase activity (11). In this work, we provide compelling functional and structural evidence that the oxidized CoxB protein is a substrate of the disulfide reductase TlpA, and that CoxB reduction is an essential step in the biogenesis of active COX.
Incidentally, a CX3C site such as in CoxB is also present in the B. japonicum ScoI protein (14), which is a homolog of the mitochondrial copper chaperone Sco1 (18,19). The Cu 2+ -binding domain of the membrane-anchored ScoI and Sco1 proteins faces the periplasm and intermembrane space, respectively, akin to CoxB. We had shown recently that TlpA serves as a reductant for ScoI (20). With few exceptions, the biogenesis of the Cu A center in eukaryotic mitochondria and in prokaryotes depends on the presence of Sco1/ScoI (18,19). One could argue, therefore, that the cytochrome oxidase deficiency in B. japonicum tlpA mutant cells is simply the consequence of a defect in forming a sufficient amount of intracellular Cu-ScoI complex for the synthesis of CuA on CoxB. As shown here, however, this explanation alone does not hold true, because not only ScoI but also CoxB is a direct substrate of TlpA. Furthermore, the requirement of ScoI for cytochrome oxidase activity, but not that of TlpA, can be bypassed by supplying high environmental copper concentrations. This strongly suggests that Cu A can be assembled on CoxB only after the latter had been reduced by TlpA.
Expression plasmids. The bacterial expression plasmids for production of CoxBPD and TlpAS and variants thereof were constructed using standard molecular cloning techniques. Detailed information on the individual plasmids including the coding nucleotide and the corresponding protein sequences is available from authors on request. For production of ScoI S C74S the previously described plasmid pRJ8336 (20) was used.
Protein production and purification. For production of CoxBPD and its variants CoxBPD C229S and CoxBPD C233S in cytoplasmic inclusion bodies, Escherichia coli strain BL21(DE3) transformed with pET11a-coxB PD , pET11a-coxB PD C229S or pET11a-coxB PD C233S , respectively, were used. The purification procedure for CoxBPD, CoxBPD C229S and CoxBPD C233S was identical. E. coli BL21(DE3) transformed with pET11a-coxB PD was grown at 37 °C in 2YT medium (tryptone, 16 g/l; yeast extract, 10 g/l; NaCl, 5 g/l) containing ampicillin (100 μg/ml) until an OD600nm of 0.5 had been reached. CoxBPD expression was induced by addition of 0.1 mM IPTG, and cells were further grown at 30 °C for 4 h. Cells were harvested by centrifugation, suspended in 100 mM Tris-HCl pH 8.0, 1 mM EDTA (3 ml/g wet cells), mixed with DNaseI (50 μg/ml final concentration), and lysed with a Microfluidizer M-110L (Microfluidics, Westwood, MA, U.S.A.). After addition of 0.5 volumes of 60 mM EDTA-NaOH pH 7.0, 1.5 M NaCl, 6% (v/v) Triton X-100, the lysate was stirred at 4 °C for 1 h. The inclusion bodies were harvested by centrifugation (30 min, 48,000 x g, 4 °C) and washed five times at 4 °C with 100 mM Tris-HCl pH 8.0, 20 mM EDTA to remove the Triton X-100. The inclusion bodies were solubilized in 100 mM Tris-HCl pH 8.0, 6 M GdmCl, 1 mM EDTA, 100 mM DTT (20 ml per gram inclusion body) at room temperature under stirring for 2 h. Insoluble material was removed by centrifugation, and CoxBPD was refolded from the supernatant by rapid dilution with 50 volumes of 20 mM Tris-HCl, 0.5 M arginine-HCl, 1 mM EDTA, 10 mM DTT (pH 8.0) at room temperature. Refolded, reduced CoxBPD was concentrated to ~1.5 mg/ml and dialyzed against 20 mM Tris-HCl, 1 mM EDTA, 5 mM DTT, pH 8.5 (4 °C). Precipitated protein was removed by centrifugation, and the supernatant was applied at 4 °C to a Resource Q column (GE Healthcare, Glattbrugg, Switzerland) equilibrated with the same buffer. The flow-through, containing reduced CoxBPD, was concentrated and subjected to gel filtration on Superdex 75 (GE Healthcare) in 20 mM Tris-HCl pH 8.5, 0.3 M NaCl, 0.5 mM EDTA, 5 mM DTT at 4 °C. The fractions containing pure, reduced CoxBPD (as judged after Coomassie-stained SDS-PAGE) were stored at -20 °C until further use. The final yield of purified, reduced CoxB PD was 60 mg per liter of bacterial culture. The identity of the protein (including the N-terminal methionine) was verified by ESI mass spectrometry (calculated mass: 15,578.0 Da; found: 15,578.0 Da).
For the use of CoxB PD C233S in crystallization trials, a different expression construct was made which carried an N-terminal His6-tag. E. coli strain BL21(DE3) harboring pProExHTa-His6-coxB PD C233S was grown and induced with IPTG as described above for CoxB PD . In addition to the C233S replacement, CoxB PD C233S differs from CoxBPD by the sequence GAFLELD at the Nterminus (replacing the N-terminal M127 in CoxB PD ) and a 14-residue longer C-terminal end corresponding to the end of the natural CoxB reading frame. Harvested cells were suspended in 20 mM Tris-HCl pH 8.0, 1 mM β-ME (1.5 ml per g of wet cells). One cOmplete TM protease-inhibitor tablet (Roche, Basel, Switzerland) per 50 ml suspension and DNaseI (50 μg/ml final concentration) were added, and cells were disrupted with a Microfluidizer. The lysate was centrifuged, and the pellet was mixed with 2 volumes of buffer A (20 mM Tris-HCl, 0.5 M NaCl, 6 M GdmCl, 2 mM β-ME, pH 8.0) and stirred at 4 °C for 1.5 h. Insoluble material was removed by centrifugation, the supernatant was mixed with imidazole/HCl, pH 8, (final concentration: 10 mM) and loaded onto Ni-NTAagarose (Qiagen, Switzerland) equilibrated with buffer A. The resin was washed extensively with buffer A containing 20 mM imidazole/HCl. His6tagged CoxBPD C233S was eluted with buffer A containing 200 mM imidazole/HCl, and protein fractions were refolded by dilution with 100 volumes 20 mM Tris-HCl, 0.5 M NaCl, 0.5 M arginine-HCl, 5 mM β-ME, pH 8.0, and incubation for 1 h at room temperature. Refolded, His6-tagged CoxBPD C233S was concentrated to 1.5 mg/ml (96 μM), and aggregates were removed by centrifugation. The N-terminal His 6 -tag was then cleaved off by incubation (18 h, 4 °C) with TEV protease (7 μM) in 20 mM Tris-HCl, 0.3 M NaCl, 5 mM β-ME, pH 8.5. The uncleaved protein and the cleaved tag were removed by binding to a Ni-NTA-column as described above. The cleaved protein in the flow-through was concentrated to ~1.5 mg/ml and subjected to gel filtration on Superdex 75. The final yield of purified CoxB C233S was 4.5 mg per liter of bacterial culture, and its mass was verified by ESI mass spectrometry (calculated mass: 17,396.9 Da; found: 17,397.0 Da).
TlpAS (TlpA residues 38-221) used for the disulfide exchange equilibrium with CoxB PD , and TlpA S C110S used for crystallization were produced as fusions to the C-terminus of MalE protein and purified after factor-Xa cleavage from MalE as described previously (16,20,23). Alternatively, an analogous expression plasmid (pMal-p/TEV-TlpA S ) encoding the same fusion with a TEVprotease instead of factor-Xa cleavage site was used for production of TlpAS for stopped-flow kinetics. The corresponding fusion protein was digested with TEV protease as described above for CoxB PD C233S , and TlpA S was further purified by anion exchange chromatography as described (23). TlpAS purified with this alternative method bears two additional glycines at its N-terminus, which proved to be necessary for efficient cleavage (calculated mass: 19,634.7 Da, found: 19,632.5 Da). ScoI S C74S was produced and purified as reported recently (20). TEV protease was purified as described (24). Disulfide-exchange equilibrium between TlpA and CoxB. CoxB PD ox and TlpA S red were mixed at three different molar ratios (initial concentrations in μM: 10:5, 5:5, and 5:10, respectively) and incubated for 18 h at 25 °C in 50 mM NaH 2 PO 4 -NaOH (sodium-phosphate), 0.1 mM EDTA, pH 7.0. The equilibria were quenched with formic acid (10% v/v final concentration), and all four redox species (CoxB PD ox , CoxB PD red , TlpA S ox and TlpA S red ) were separated by reversed-phase HPLC on a Zorbax C8 SB-300 column (Agilent Technologies, Basel, Switzerland) at 60 °C using an acetonitrile gradient (35-45%) in 0.1% TFA. Eluted proteins were detected at 220 nm, and their concentration was quantified by peak integration. The equilibrium constant (Keq) and the redox potential of CoxBPD were calculated according to equation (1) and the Nernst equation (2), respectively, using a value of -256 mV for the redox potential (E 0 ') of TlpA S (20).
TlpA S red and CoxB PD red were obtained after reduction with excess DTT (10 mM for CoxBPD and 5 mM for TlpAS) for ≥30 min at room temperature in 20 mM Tris-HCl, 0.3 M NaCl, 0.5 mM EDTA, pH 8 (CoxB PD ) or 10 mM Tris-HCl, pH 8 (TlpA S ), followed by centrifugation of CoxBPD at 100,000 x g for 30 min at 4 °C and removal of excess DTT from both proteins by gel filtration on Hi-Prep 26/10 Sephadex G25 columns (GE Healthcare). CoxB PD ox was obtained after incubation of CoxB PD red (10-20 μM) with a 1.5fold molar excess of DTNB in 50 mM sodiumphosphate, 0.1 mM EDTA, pH 7, at room temperature for 20 min, followed by gel filtration on a Hi-Prep 26/10 column as described above in 50 mM sodium-phosphate, 0.1 mM EDTA, pH 7. Complete oxidation of CoxBPD red by DTNB was verified by reversed-phase HPLC (see above).
Stopped-flow fluorescence kinetics of disulfide exchange between TlpA and CoxB. The kinetics of the reduction of CoxB PD ox by TlpA S red were recorded at 25 °C in 50 mM sodiumphosphate, 0.1 mM EDTA, pH 7. The reactions were initiated with stopped-flow mixing (1:1) in a SX20 stopped-flow instrument (Applied Photophysics, Leatherhead, UK) and followed via the decrease in tryptophan fluorescence of TlpAS at 320 nm upon oxidation (excitation at 280 nm) (20). A constant, initial TlpA S red concentration of 1 μM was used, while CoxBPD ox was used in excess, with initial concentrations of 5, 10, or 15 μM. The kinetic profiles were fitted globally with Berkeley-Madonna TM (version 8.3.18), with a fixed value of 7.1 for the ratio between the forward (k2) and reverse (k-2) reactions based on the determined equilibrium constant (Keq = k2/k-2).
CD spectra and thermal unfolding transitions of CoxB PD . CD spectra of CoxB PD (0.2 mg/ml) in 10 mM sodium-phosphate, pH 7.0 were recorded at 25 °C with a J-715 spectropolarimeter (Jasco, Easton MD, USA; 0.1 cm path length). Thermal unfolding transitions of oxidized and reduced CoxB PD in 10 mM sodium-phosphate pH 7.0 (supplemented with 5 mM DTT in the case of reduced CoxBPD) were followed via the increase in the far-UV CD signal at 218 nm upon unfolding (heating rate: 1 K/min), and fitted and normalized according to a reversible thermal unfolding equilibrium (equation 3): where S represents the observed CD signal at 218 nm, y f and y u are the y-axis intercepts at zero K and mf and mu the slopes of the pre-and posttransition baselines, respectively, T is the temperature, Tm is the melting temperature, and ΔH m is the enthalpy change of unfolding at T m .
In vitro reconstitution of the Cu A center in CoxBPD. The CuA center in CoxBPD was reconstituted via the reaction scheme deduced from the recently described Cu A reconstitution in subunit II of the ba 3 -type cytochrome oxidase from Thermus thermophilus (6): Reduced CoxBPD (50 μM) was incubated with CuCl 2 (500 μM) in 50 mM Bis-Tris-HCl, pH 7.0, overnight at room temperature. Excess CuCl 2 was removed by gel filtration in 50 mM Bis-Tris-HCl, pH 7.0, the solution was concentrated to ~150 μM protein, and precipitated material was removed by ultracentrifugation. The measured absorbance spectrum with its maxima at 367 nm, 479 nm and 813 nm confirmed the formation of the Cu A center. Preparation and crystallization of the TlpA-CoxB mixed disulfide complex. The TlpA S C110S -CoxB PD C233S complex was generated as described above for TlpA S C110S -ScoI S C74S . The final reaction that led to the mixed-disulfide complex was performed with equimolar concentrations (15 μM each) of CoxB PD C233S and TlpA S C110S -TNB in 10 mM Tris-HCl, pH 8 (1 h at room temperature). The TlpAS C110S -CoxBPD C233S complex was purified by gel filtration on HiLoad Superdex 75 16/60 as described above, and finally concentrated to 12 mg/ml in 10 mM Tris-HCl, pH 8.0. ESI mass spectrometry and SDS-PAGE analysis revealed that seven C-terminal residues in CoxBPD C233S (residues 273-279) had been cleaved off by an unknown protease (calculated mass of the cleaved mixed-disulfide complex: 36,284.5 Da; found: 36,287.5 Da). The TlpA S C110S -CoxB PD C233S stock solution (3 μl) was mixed with 1 μl of precipitant solution (25% (w/v) PEG 2000 monomethyl ether, 0.8 M formic acid-NaOH, 0.1 M sodium cacodylate, pH 6.5). Rod-like crystals grew at 4 °C within 13 weeks (sitting-drop vapor diffusion). Crystals were cryo-protected in precipitant solution containing 15% glycerol and flash-frozen in liquid nitrogen.
X-ray crystallography and structure determination. The crystal structure of the TlpA S C110S -ScoI S C74S mixed disulfide was solved by molecular replacement with PHASER (25), using the crystal structure of the soluble domain of B. japonicum TlpA (PDP entry 1jfu) and the crystal structure of Bacillus subtilis Sco1 (PDB entry 1xzo) as search models. The asymmetric unit of the TlpA-ScoI crystal structure contains four complexes, where chains A and B form complex 1, chains C and D complex 2, chains E and F complex 3, and chains G and H complex 4. For the analysis of the structure, mostly complex 2 (chains C+D) was employed. The structure of the TlpAS C110S -CoxBPD C233S complex was also solved by molecular replacement with PHASER (25), using the structures of B. japonicum TlpA (PDB entry 1jfu) and subunit II of cytochrome c oxidase from Rhodobacter sphaeroides (PDB entry 1m56), in which the N-terminal membrane helices (residues 1-130), residues 151-191, the Cterminal helix (residues 270-289), and the loop (residues 250-261) containing the two copperbinding cysteines are absent. PHENIX (26) and Refmac (27) were used for refinement and model validation. COOT (28) was used for manual model building. Secondary structure elements were assigned with DSSP (29). T-Coffee (30) was used for multiple sequence alignments. The Pdb2pqr web-server (31) and APBS (32) were employed for analysis of electrostatics, PyMOL (The PyMOL Molecular Graphics System, Version 1.5.0.4 Schrödinger, LLC) for generation of structural figures, and ALINE (33) for producing figures of sequence alignments. Interface analysis of the structures was performed with EPPIC (34). PISA (35) was used to localize hydrogen bonds and salt bridges in the dimer interfaces.

RESULTS
Excess environmental Cu 2+ rescues cytochrome oxidase activity in a scoI but not in a tlpA deletion mutant of B. japonicum. B. japonicum scoI and tlpA mutants share similar phenotypes including a defect in COX activity; nevertheless, respiration in these strains is possible owing to the presence of quinol oxidases (14). We noticed previously that excess Cu 2+ ions in growth media can compensate for the requirement of the Cu 2+ -specific copper chaperone ScoI in forming active COX (14). The precise mechanistic reason for this bypass is not known. Surprisingly, we now observed that a tlpA mutant cannot be compensated in a similar way by high external copper supplementation (Fig. 1). Therefore, TlpA in cells ought to have at least a second function apart from serving as a reductant to the CX3C motif in ScoI (20). This result prompted us to inquire into the possibility that TlpA interacts with CoxB because this COX subunit also carries a CX3C motif as part of its CuA binding site.
TlpA is a specific reductant of CoxB. The possible target for TlpA-catalyzed thiol-disulfide exchange is the cysteine pair C229/C233 in the periplasmic domain of CoxB. This domain (residues 128-265, hereafter termed CoxBPD) was expressed in Escherichia coli in the form of cytoplasmic inclusion bodies and purified after in vitro refolding under reducing conditions (Fig. 2). The dithiol of C229/C233 in CoxB PD could be quantitatively converted to the disulfide form by oxidation with DTNB. Both reduced and oxidized CoxB PD showed almost identical, cooperative thermal unfolding transitions, arguing for an intact tertiary structure (Fig. 2B,C). This was further confirmed by the successful in vitro reconstitution of the CuA center in reduced CoxBPD by addition of Cu 2+ ions (Fig. 2D) (6). Oxidized CoxB PD failed to assemble the Cu A center.
Assuming a function of TlpA as a CoxB reductase, we predicted (i) that the cysteine pair C229/C233 in CoxBPD is less reducing than the active-site cysteine pair of TlpA (C107/C110), and (ii) that reduction of CoxB by TlpA would have to proceed fast, i.e., with a rate constant above the threshold (~10 3 M -1 s -1 ) for physiological disulfide exchange reactions between thioredoxin-like oxidoreductases and their substrates (36). We first determined the equilibrium constant (K eq ) of the disulfide exchange reaction at pH 7.0 and 25 °C between CoxBPD and the well-characterized, soluble TlpA variant TlpAS, which lacks the Nterminal membrane anchor (12). Reduced TlpA S was mixed with CoxB PD ox at different ratios, and the reaction products were separated and quantified by reversed-phase HPLC after acid quenching (Fig. 3A). The equilibrium constants, deduced from the individual HPLC runs after peak integration, proved to be independent of the initial CoxB PD :TlpA S ratio within experimental error, demonstrating that equilibria had been attained. The deduced Keq value of 7.1±0.6 translates into a redox potential (E 0 ') of -231±1 mV for CoxBPD. This shows that reduction of CoxB PD by TlpA S (E 0 ' = -256 mV) is thermodynamically favorable. The equilibrium is, however, far less on the side of oxidized TlpAS compared with the reduction of ScoIS by TlpAS, which has an equilibrium constant of 1740 (20).
Next, the kinetics of the attainment of the disulfide-exchange equilibrium between CoxB PD and TlpAS were followed after stopped-flow mixing of reduced TlpAS with a 5-, 10-, or 15-fold excess of oxidized CoxB PD , where we made use of the strong decrease in the tryptophan fluorescence of TlpA S upon oxidation (16,20). The obtained rate constants of 8.4 x 10 4 M -1 s -1 and 1.2 x 10 4 M -1 s -1 for the forward and reverse reaction, respectively, demonstrate rapid disulfide exchange between both proteins (Fig. 3B). The rate constant of CoxB PD reduction is thus very similar to that of the reduction of ScoIS by TlpAS (9.4 x 10 4 M -1 s -1 ) (20) and a strong hint that both CoxB and ScoI are physiological substrates of the periplasmic reductase TlpA.
Trapping and crystallizing the TlpA-ScoI and TlpA-CoxB heterodisulfides. Disulfide exchange between thioredoxins and their substrates occurs via mixed-disulfide intermediates, in which the more N-terminal, active-site cysteine of a thioredoxin forms the intermolecular disulfide with the substrate protein (37). In accordance with this reaction scheme, the active-site thiol of C107 in TlpA forms a transient mixed disulfide with C78 of ScoI (20). A mixed disulfide is kinetically unstable because it is rapidly attacked by a neighboring cysteine thiol. Therefore, a stable mixed disulfide between TlpAS and ScoI S was prepared by using the singlecysteine variants TlpA S C110S and ScoI S C74S . Activation of TlpAS C110S with DTNB and subsequent reaction with ScoIS C74S yielded the TlpAS C110S -ScoIS C74S complex.
To determine the cysteine residue in CoxB PD , which forms the mixed-disulfide intermediate with TlpA, the single-cysteine variants CoxB PD C229S and CoxB PD C233S were tested for their reactivity with DTNB-activated TlpAS C110S in a similar way as described previously with ScoI (20). The result was that predominantly C229 of CoxB reacted with C107 of TlpA. Based on this finding, a stable TlpAS C110S -CoxBPD C233S mixed-disulfide complex was prepared as described above for the TlpA S C110S -ScoI S C74S complex.
The folds of TlpA, ScoI, and CoxB in the context of mixed-disulfide complexes. In both mixed-disulfide complexes, the structure of TlpAS C110S is identical to that determined previously for wild-type TlpA S , which shows a characteristic thioredoxin-like fold with a central β-sheet flanked by five α-helices and the activesite cysteine pair at the N-terminal end of helix α2 (12) (Figs. 4 and 5). In the TlpAS C110S -ScoIS C74S complex, ScoI also displays a thioredoxin-like fold, composed of 9 β-strands and 6 α-helices with a β-hairpin extension between α-helix 4 and βstrand 8, which is characteristic of Sco-like proteins (38). Unlike the active-site cysteines in bona fide thioredoxins, however, the Cu 2+ ionliganding cysteine pair C74/C78 is located at a different position in the thioredoxin fold, namely in the loop segment (residues 71-80) between strand β3 and helix α1 (Figs. 4A and 5). Given that physiological disulfide-exchange reactions between thiol-disulfide oxidoreductases with a thioredoxin fold have not been observed and are kinetically restricted (7,36), the reaction between TlpA and ScoI is unusual. This interaction is probably due to the different location of the C74/C78 cysteine pair in ScoI, which might facilitate the accessibility of the C74-C78 disulfide bond in oxidized ScoI for rapid, nucleophilic attack by C107 of TlpA.
Apart from two exceptions (vide infra), CoxB PD C233S in the mixed-disulfide complex with TlpAS C110S adopts essentially the basic fold that had been observed in the structure of the periplasmic CuA center-containing domain of subunit Cox2 in the aa 3 -type cytochrome oxidases from Paracoccus denitrificans (1,39), bovineheart mitochondria (2,40), and Rhodobacter sphaeroides (41). Specifically, CoxB PD C233S adopts an extended Greek key-like 5-stranded β-sheet (Greek key plus one additional, antiparallel strand), a 3-stranded β-sheet, a β-hairpin and a Cterminal α-helix (Figs. 4B and 5). The intermolecular disulfide bond in TlpA S C110S -CoxBPD C233S was photoreduced during X-ray data collection (Fig. 4B, lower right inset), with a final refined distance of 2.9 Å between the two S γ atoms (which compares with an ideal distance of 2.05 Å). A proper disulfide between C107 (TlpA) and C229 (CoxBPD), however, can be modeled upon a slight conformational change of both cysteine side chains without changing the overall geometry of the interface.
TlpA displays non-overlapping surface areas for recognition of different substrates. The individual monomers in both mixed-disulfide complexes are oriented such that their N-termini are located on the same side of the heterodimers (Fig. 6A). Due to the presence of an N-terminal transmembrane domain in full-length CoxB and the N-terminal membrane anchors in wild-type TlpA and ScoI, the structures are fully consistent with a topologically suitable positioning on the periplasmic side of the cytoplasmic membrane for disulfide exchange between TlpA and ScoI or CoxB in vivo.
Remarkably, the solved structures of the two mixed-disulfide complexes uncovered two separate TlpA interfaces for interaction with either ScoI or CoxB (Figs. 4 and 6), i.e., both interfaces overlap only in a comparatively small area. The interface area of TlpA S C110S -CoxB PD C233S (997 Å 2 ) is more spacious than that of TlpA S C110S -ScoI S C74S (775 Å 2 ). An upper-bound estimate for the shared overlap area is about 350 Å 2 (measured on the TlpA surface with program EPPIC; 34). Herein, the TlpA residues C107, K171 and M179 are involved in similar types of interactions with both, ScoI and CoxB (Table 2): while C107 forms the respective intermolecular disulfide bond, K171 forms a salt bridge with E84 of ScoI (Fig. 4A, left inset) or with the Cu A -coordinating E231 of CoxB (Fig. 4B, lower right inset), and M179 forms a twofold backbone-backbone interaction with C229 of CoxB (Fig. 4B, lower right inset), or with C78 of ScoI (Fig. 4A, left inset) where the backbone-amide and -carbonyl of M179 of TlpA are in close proximity to the backbone-carbonyl and backbone-amide of C229 of CoxB and C78 of ScoI, respectively ( Table 2).
All other interactions at the proteinprotein interfaces are specific to the individual mixed-disulfide complexes (Table 2; for example, see Fig. 4A, lower right inset). The interface between TlpAS C110S and ScoIS C74S is characterized by a unique cluster of π-stacking interactions between four aromatic residues (W106 and F167 of TlpA, and F83 and F123 of ScoI (Fig. 4A, upper right inset). W106 of TlpA and F123 of ScoI are conserved in TlpA homologs and ScoI-like proteins, respectively (alignments are available from authors on request). Additional minor interactions are listed in Table 2.
The specific interface between TlpA S C110S and CoxBPD C233S features eight unique hydrogen bonds (for example, see Fig. 4B, left inset) and additional hydrophobic interactions of P115 and P198 of TlpA with the side chains of the leucine pair L168/L169 of CoxB (Fig. 4B, upper right inset), in which L169 is fully conserved ( Table 2). The backbone amides of these leucines are hydrogen-bonded with the carboxylate of E200 in TlpA (Fig. 4B, upper right inset).
A striking difference between the heterodimer interfaces of TlpA-ScoI and TlpA-CoxB is revealed by the surface electrostatics of the individual molecules (Fig. 6A). The possible implications of this observation will be discussed below.

DISCUSSION
Owing to the unique subunit and cofactor composition and the peculiar topology of cytochrome oxidase in the bacterial cytoplasmic membrane or mitochondrial inner membrane, the biogenesis of this enzyme is a very complex, multifactorial process. Although a substantial number of assembly factors for heme and copper insertion have so far been identified, our mechanistic understanding about how they assemble and act in concert is still incomplete. Results reported here add an important piece to the puzzle by showing how TlpA prepares ScoI and cytochrome oxidase subunit II (CoxB) for Cu-ion insertion and formation of the CuA center. The epistatic nature of TlpA function became evident from the fact that the cytochrome oxidase defect in a B. japonicum scoI mutant, but not in a tlpA mutant, could be corrected phenotypically by supplying excess Cu 2+ to the growth medium. Such a rescue of COX activity had been observed previously in Sco-deficient organisms including human cell lines (14,(42)(43)(44). How this rescue works remains enigmatic in view of the fact that intracellular metal levels are usually regulated tightly, and chelators adjust them to appropriate concentrations in vivo (18). It is often typical for chaperones to be important especially under conditions of stress or nutrient limitation. In fact, ScoI requirement for cytochrome oxidase activity in B. japonicum seems to be most distinct under Cu-limiting growth conditions (14,15). Moreover, the bypass of ScoI by Cu-replete conditions finds a possible explanation in the ease with which we could reconstitute the Cu A center in vitro simply by incubating reduced CoxB in the presence of Cu 2+ (Fig. 2D), as it had been worked out by Chacon and Blackburn (6). Assuming that such a self-assembly at high external Cu 2+ concentrations also works in vivo, this would explain why ScoI may become dispensable. By contrast, it is absolutely essential for the success of the in vitro reconstitution of the Cu A center that CoxB is provided in reduced form. Extrapolating this sine qua non to the in vivo situation means that CoxB must always be kept reduced for Cu ion insertion, which is precisely the function we ascribe to TlpA. Obviously, the TlpA requirement for the biogenesis of functional COX cannot be bypassed by Cu-replete conditions. Our in vitro experiments performed both with oxidized ScoI (20) and oxidized CoxB as substrates (this work) clearly support the function of TlpA as a reductase. TlpA has a more negative redox potential than ScoI and CoxB, and the rates of reduction of both substrate proteins are fast and lie in a physiologically relevant range. This was already a strong hint for a specific protein-protein interaction between TlpA and ScoI as well as TlpA and CoxB, a postulate that was fully confirmed by the high-resolution crystal structures of the mixed-disulfide TlpA-ScoI and TlpA-CoxB complexes. Apart from the detailed contact areas already described in Fig. 4 (see above), TlpA S C110S possesses a distinct "dipolar" interface in the region where ScoIS C74S and CoxBPD C233S bind. While the TlpA interface shared with ScoI is of basic nature, the interface shared with CoxB is mostly acidic. ScoI and CoxB complement this dipole as each protein provides the opposite charge to the interface (Fig. 6B). In conclusion, we suggest a model in which electrostatic interactions between TlpA and its substrates are also governed by an overall charge complementarity. Such a reciprocal electrostatic complementarity might contribute to the fast rates of ScoI and CoxB reduction by TlpA (45). Interestingly, although the CX3C motif is the common target in both substrate proteins, TlpA attacks the more N-terminal cysteine in CoxB (i.e., C229 in the C 229 X 3 C 233 sequence) as opposed to the more C-terminal cysteine in ScoI (i.e., C 78 in the C 74 X3C 78 sequence). This might also be a reflection of the bipolar asymmetry on TlpA which positions each of two protein substrates from different angles on top of its active site.
A bioinformatics analysis has shown that TlpA is widespread in aerobic bacteria which respire with cytochrome oxidases (available from authors on request). The peculiar, exposed location of the Cu A center on the outer side of the membrane, there being confronted with an oxidizing environment, appears to make a TlpAlike reductase function mandatory for the biogenesis of cytochrome oxidase. Whether or not this prerequisite also applies to mitochondria is less obvious. Recent work has shown, however, that dithiol-disulfide oxidoreductase relay systems do exist in mitochondria, both in the oxidizing as well as reducing direction (8,46), and thioredoxinlike proteins were found in the mitochondrial intermembrane space (47). The existence of a ScoI-and CoxB-specific reductase in the periplasm of Gram-negative bacteria inevitably leads to the question, where the reducing power is ultimately derived from. Although E. coli does not possess a cytochrome oxidase for aerobic respiration, a system similar to that of E. coli DsbD, which relays reducing power from the cytoplasm over the membrane into the periplasm (9,48), is a valid option also for B. japonicum.
Another unsolved problem for future research is where the Cu ions for the biogenesis of the Cu A center are derived from, and how Cu A assembly proceeds. Although the Sco1/ScoI protein is widely accepted to be the Cu 2+ carrier to CuA (18,19), the direct transfer of Cu 2+ to apo-CoxB has not yet been demonstrated with in vitro experiments. Furthermore, formation of the Cu A center as an electron-delocalized Cu 1.5+ -Cu 1.5+ system implicates not only Cu 2+ but also Cu + in the metallation process (6). A good candidate for a Cu + carrier to CuA is the PCuAC protein (called PcuC in B. japonicum) (15,49). How the ScoIand PcuC-like copper chaperones cooperate in the assembly of Cu A on the TlpA-reduced CoxB subunit of cytochrome oxidase remains to be elucidated.
In this context it is noteworthy that our work allowed for the first time a description of the structural differences between apo-and holo-CoxB. A superposition of the B. japonicum CoxBPD C233S structure onto the structure of P. denitrificans subunit Cox2 of cytochrome oxidase, which carries an intact Cu A center (PDB entry 3hb3) is shown in Fig. 7. The largest difference concerns the segment 229-240 which contains four of the six CuA-coordinating residues, namely C229, E231, C233 and H237 (Fig. 7A). The positions at the C α -atoms deviate substantially from those of the Cu A -coordinating residues of P. denitrificans Cox2, ranging from 6.5 to 12.1 Å. By contrast, the positions of M240 and H194 do not change with respect to Cox2. This finding suggests that the CuA loop 229-240 in CoxB undergoes a major conformational change in the course of metallation. Furthermore, before metallation, the two cysteines C229 and C233 in the segment 229-240 are surface-exposed such that a disulfide bond between them is accessible for reduction by TlpA. After reduction, the two cysteine thiols may bind Cu ions, and the entire loop may undergo a structural rearrangement to form the mature Cu A center in holo-CoxB. Besides the Cu A loop, we noticed one other structural difference, which concerns the segments L159-D172 of B. japonicum CoxB and L137-D159 of P. denitrificans Cox2. In Cox2, that region contains two short α-helices which contribute to the interface with the neighboring subunit I, whereas these helices are absent in B. japonicum CoxB (Fig. 7B). Fig. 1. Whole-cell cytochrome oxidase activity is rescued in B. japonicum scoI but not tlpA mutant by excess Cu 2+ in the growth medium. The respective B. japonicum mutants were grown aerobically in PSY medium with or without 50 µM CuCl 2 and spotted onto a filter paper soaked with the redox indicator TMPD. Active COX in cells such as the wild type (WT) converts TMPD to indophenol blue. Mutants of subunits I (coxA) and II (coxB) of COX served as negative controls.  Reduced TlpA (TlpAS red) and oxidized CoxB (CoxBPD ox) were mixed at initial concentrations (in μM) of 5 and 10, 5 and 5, or 10 and 5, respectively, and incubated for 18 h. The reactions were quenched with formic acid. All four redox species were separated by reversed-phase HPLC, detected via absorbance at 220 nm, and quantified by peak integration. The equilibrium constants deduced from the individual HPLC runs proved to be identical within experimental error, demonstrating that the disulfide-exchange equilibrium had been attained. (B) Stopped-flow fluorescence kinetics of the reduction of CoxB by TlpA. TlpAS red (1 μM) was mixed with a 5-, 10-, or 15-fold excess of CoxBPD ox, and the reaction was followed via the decrease in tryptophan fluorescence upon oxidation of TlpA S . Data were fitted globally (solid lines) with Berkeley-Madonna TM according to the indicated disulfide-exchange equilibrium, with a fixed ratio of 7.1 between the second-order rate constants of the forward and reverse reactions (k 2 and k -2 , respectively), and normalized. The estimated error of the rate constants is about 20%. Fig. 4. Caught in the act: X-ray structures of TlpA in complex with ScoI or CoxB. The two mixed disulfides consisting of (A) TlpAS C110S (grey) and ScoIS C74S (orange), or (B) TlpAS C110S and CoxBPD C233S (cyan) are shown in the same orientation with respect to TlpAS C110S , highlighting the existence of two distinct binding interfaces in TlpA for the two substrates. The cysteines involved in formation of the intermolecular disulfides (C107 of TlpA S C110S , C78 of ScoI S C74S , and C229 of CoxB PD C233S ) are shown as sticks. The intermolecular disulfide bond of the TlpA S C110S -CoxB PD C233S complex was lost due to radiation damage. The inserted boxes (see text) present detailed views of specific, polar interactions. In some of the views, the structures were rotated by the indicated angle. Residues are shown as sticks and hydrogen bonds as red dotted lines. Yellow sticks: sulfur atoms; red sticks: oxygen atoms; blue sticks: nitrogen atoms; yellow broken line in B, lower right inset: distance between cysteine 229 of CoxB PD C233S and cysteine 107 of TlpA S C110S , indicating the position of the intermolecular disulfide that was lost due to radiation damage; N: N-terminus of the respective polypeptide chain.  5. Primary structure, regular secondary structure elements, and residues forming specific interactions at the protein-protein interface in the mixed disulfide complexes of TlpAS C110S , CoxBPD C233S and ScoIS C74S . The cysteines forming intermolecular disulfide bonds are highlighted in yellow. Interacting residues are highlighted in cyan blue and orange for the TlpA-CoxB and the TlpA-ScoI complex, respectively. Magenta-shaded TlpA residues participate in interactions with both, CoxB and ScoI.  Sequence identity is 44%. The cysteines liganding the copper ions in the CuA center are highlighted in yellow, the other copper-coordinating residues undergoing a large conformational rearrangement upon formation of the Cu A center are highlighted in purple, while the copper-coordinating residues that undergo no conformational rearrangement upon CuA formation are highlighted in cobalt blue. Secondary-structure elements (β-strands: blue arrows; α-helices: red cylinders) were assigned as indicated by DaliLite (50).  [ii] 2 Ramachandran outliers at the border of the allowed region [P76 of ScoIS C74S (TlpAS C110S -ScoIS C74S , chains B and F), P141 of TlpAS C110S (TlpAS C110S -CoxPD C233S , chain A), and A236 of Cox PD C233S (TlpA S C110S -Cox PD C233S , chain B)]. 2F o -F c density however mandates this conformation of the residues.