Thiosulfate Dehydrogenase (TsdA) from Allochromatium vinosum

Background: TsdA oxidizes thiosulfate to tetrathionate and shows unusual histidine-cysteine axial heme coordination. Results: Characterization of TsdA variants provides various snapshots of both heme environments. Conclusion: Heme 1 was identified as the catalytic heme along with a substrate binding pocket near Cys96, and a Lys208/Met209 ligand switch is observed at heme 2 upon reduction. Significance: A novel mechanism for thiosulfate oxidation is proposed. Although the oxidative condensation of two thiosulfate anions to tetrathionate constitutes a well documented and significant part of the natural sulfur cycle, little is known about the enzymes catalyzing this reaction. In the purple sulfur bacterium Allochromatium vinosum, the reaction is catalyzed by the periplasmic diheme c-type cytochrome thiosulfate dehydrogenase (TsdA). Here, we report the crystal structure of the “as isolated” form of A. vinosum TsdA to 1.98 Å resolution and those of several redox states of the enzyme to different resolutions. The protein contains two typical class I c-type cytochrome domains wrapped around two hemes axially coordinated by His53/Cys96 and His164/Lys208. These domains are very similar, suggesting a gene duplication event during evolution. A ligand switch from Lys208 to Met209 is observed upon reduction of the enzyme. Cys96 is an essential residue for catalysis, with the specific activity of the enzyme being completely abolished in several TsdA-Cys96 variants. TsdA-K208N, K208G, and M209G variants were catalytically active in thiosulfate oxidation as well as in tetrathionate reduction, pointing to heme 2 as the electron exit point. In this study, we provide spectroscopic and structural evidence that the TsdA reaction cycle involves the transient presence of heme 1 in the high-spin state caused by movement of the Sγ atom of Cys96 out of the iron coordination sphere. Based on the presented data, we draw important conclusions about the enzyme and propose a possible reaction mechanism for TsdA.

mg Ϫ1 ). Reduced methylviologen served as artificial electron donor in these in vitro assays (2). Taken together, these findings indicate that the A. vinosum enzyme is especially adapted to catalyzing thiosulfate oxidation.
In c-type cytochromes like TsdA, the heme moieties are covalently bound to the apoprotein via thioether linkages. These are formed by two cysteine residues mostly present in a Cys-XX-Cys-His motif (9,10). The histidine serves as a proximal axial ligand to the iron encapsulated in the porphyrin macrocycle. A few cytochromes show variations of the heme binding motif to CX 15 CH or even CXXCK (11)(12)(13). In the CXXCK motif, the lysine acts as a proximal heme ligand, such as in the catalytically active site of cytochrome c nitrite reductase (14). In six-coordinate c-type cytochromes, the distal axial ligand is most commonly a second histidine or a methionine (15). The combination of histidine with other ligands is relatively rare. It includes tyrosine in cytochrome f (16), the side chain of lysine in cytochrome c nitrite reductase (17) and its physiological partner NrfH (18), asparagine (19) in sphaeroides heme protein (19), and also cysteine. The latter was first discovered in the SoxXA protein of Rhodovulum sulfidophilum (20). In the catalytic subunit SoxA, the active site heme-ligating cysteine is modified by a further sulfur atom (i.e. it is present as a cysteine persulfide possibly arising from a SoxA-thiocysteine S-sulfate intermediate formed during the catalytic cycle) (6,21). Other examples for His/Cys-ligated hemes in c-type cytochromes are the DsrJ protein from A. vinosum (22) and Desulfovibrio desulfuricans (23), the green heme protein from Halochromatium salexigens (24), a triheme cytochrome c from R. sulfidophilum (25), and the PsbV2 cytochrome from the cyanobacterium Thermosynechococcus elongatus (26).
In this work, we crystallized several recombinant forms of A. vinosum TsdA and determined the three-dimensional structures of the "as isolated," dithionite-reduced, and tetrathionatesoaked proteins as well as of variants with replacements of putative active site and heme iron-ligating residues. Thereby, we confirmed the proposed axial histidine/cysteine ligation of heme 1 and found a distal axial lysine ligation of the second heme in the oxidized state of the protein. Remarkably, the axial ligand of heme 2 changes to methionine upon reduction, as revealed by several x-ray structures. Assessment of enzymatic activities and electronic absorption properties for wild type TsdA and its variants provided further evidence for unexpected plasticity of heme ligation at both hemes. We identified the substrate-binding site in the vicinity of heme 1, close to Cys 96 and surrounded by other strictly conserved residues. A catalytic mechanism is herein proposed.

EXPERIMENTAL PROCEDURES
Bacterial Strains, Media, and Growth Conditions-The bacterial strains and plasmids used are described in Table 1. Escherichia coli strains were cultivated in NZCYM (27), LB, or 2ϫ YT (28) medium. E. coli DH5␣ was used for molecular cloning.
Recombinant DNA Techniques-All general molecular genetics techniques were described previously (29). Chromosomal DNA of A. vinosum was obtained by standard methods as described previously (30). Restriction enzymes, T4 ligase, and Pfu DNA polymerase were obtained from Thermo Scientific (Schwerte, Germany) and used according to the manufacturer's instructions. Oligonucleotides for cloning were obtained from Eurofins MWG (Ebersberg, Germany).
Construction of Expression Plasmids and Site-directed Mutagenesis-The A. vinosum tsdA gene coding for the mature protein without the signal peptide was amplified and cloned as described earlier (1). Point mutations were introduced into tsdA by overlap extension (31) using standard PCR with Pfu DNA polymerase (Thermo Scientific) and pPRIBAAvtsdA (1) as the template. For the TsdA-K208G exchange, two fragments were amplified with the following primers: for the first fragment, TsdA_XbaIf and LysGly_rev; for the second fragment, LysGlyfw and IBA1hind_rev. Both fragments were used as templates for amplification of the complete tsdA gene carrying the desired point mutation. In this step, TsdA_XbaIfopt and IBA1hind_revopt served as primers. The resulting fragment was restricted with XbaI and HindIII and cloned into pPRIB-AAvtsdA, resulting in plasmid pPR-IBAAvtsdAsK208G. The following plasmids were generated applying the same general strategy: pPR-IBAAvtsdAsK208N, pPR-IBAAvtsdAsC96G, pPR-IBAAvtsdAsC96H, and pPR-IBAAvtsdAsC96M (Table 1).
Overproduction, Purification, and Preparation of Recombinant TsdA Wild Type and Mutant Proteins-E. coli BL21(DE3) cells containing pPR-IBAAvtsdAs or one of the tsdA mutant expression plasmids and pEC86 were cultured in 700 ml of NZCYM medium containing ampicillin (100 mg ml Ϫ1 ) and chloramphenicol (25 mg ml Ϫ1 ) in 1-liter Erlenmeyer flasks at 37°C and 180 r.p.m. At A 600 of 0.6, the culture was switched to 25°C, and the cells were harvested after an additional 16 -20 h. Cells were resuspended in 50 mM BisTris-HCl buffer, pH 6.5, and lysed by sonication. After the removal of insoluble cell material by centrifugation (10,000 ϫ g for 25 min at 4°C), TsdA wild type or TsdA mutant proteins were purified by Strep-Tactin affinity chromatography and gel filtration as described before (32).
Protein Techniques-SDS-PAGE was performed as described previously (33). Heme staining in acrylamide gels was done as described (34). Protein concentration of purified protein was determined with the BCA kit from Pierce.
Enzyme Kinetics-Thiosulfate-dependent ferricyanide reduction was measured by following the decrease of absorbance at 420 nm (⑀ ϭ 1.09 mM Ϫ1 cm Ϫ1 ). 100 mM ammonium acetate buffer (pH 4.0 or 5.0), 1 mM ferricyanide, and varying concentrations of thiosulfate were preincubated in a 0.5-ml cuvette (Hellma Analytics, Mühlheim, Germany) at 30°C for 5 min. Assays were started by the addition of TsdA, and data were recorded in a Specord 210 spectrophotometer (Analytik Jena, Jena, Germany).
Tetrathionate-dependent methylviologen oxidation was measured by following the decrease of absorbance at 585 nm for the artificial electron donor methylviologen (⑀ ϭ 11.8 mM Ϫ1 cm Ϫ1 ) prereduced with titanium(III) citrate. The latter was prepared as described by Zehnder and Wuhrmann (35). 100 mM ammonium acetate buffer, pH 5.0, 300 M titanium(III) citratereduced methylviologen and 1 g of TsdA were preincubated in a 3-ml cuvette (Hellma Analytics) at 30°C for 5 min. Assays were started by the addition of tetrathionate in varying concentrations, and data were recorded in an Agilent 8453 spectrophotometer (Agilent Technologies, Ratingen, Germany) under anoxic conditions. When necessary, control assays without enzyme were run to calculate the chemical oxidation of methylviologen by tetrathionate.
Spectroscopic Methods-UV-visible spectra were recorded in a Specord 210 spectrophotometer (Analytik Jena) and processed with Microsoft Excel.
Crystallization and Data Collection-TsdA crystallization has been reported previously (32). In summary, TsdA at a concentration of 8 mg ml Ϫ1 in 20 mM BisTris-HCl, pH 6.5, crystallized in a condition comprising 23.5% (w/v) PEG 3350, 0.2 M (NH 4 ) 2 SO 4 , 0.1 M BisTris, pH 6.28, and 0.1 M NaI (as additive). 3-l drops were set up in an Oryx 6 dispensing robot (Douglas Instruments), using an MRC Maxi 48-well crystallization plate (Swissci), by mixing 1.2 l of protein solution with 1.5 l of precipitant and 0.3 l of additive, and equilibrated against a reservoir of 120 l of crystallization condition. TsdA mutants and TsdA incubated with ligands crystallized in similar conditions as the "as isolated" protein. Other TsdA variants were obtained by soaking the WT crystals with an excess of ligands (thiosulfate, tetrathionate, and bisulfite) or reduced with dithionite as described in Table 3. Crystals were then backsoaked with the cryoprotection solution. For cryoprotection, crystals were transferred to a new drop with higher PEG 3350 concentration (25.4% w/v) supplemented with 5% (v/v) PEG 400. Complete x-ray diffraction data sets were collected at several synchrotrons (see Table 2 for details).
Data were indexed, integrated, and scaled using XDS and converted to MTZ format with XDSCONV (36). R free flags for all data sets were created at this stage corresponding to 5% of the measured reflections for each data set. For phasing, a data set was collected at a wavelength of 1.722 to 1.98 Å resolution with an overall R merge of 7.2 and 87% completeness (Table 2) on the XALOC beamline at the ALBA synchrotron (Barcelona, Spain).

Structure Determination and Refinement-The
AutoSol wizard in PHENIX (37) was used to determine the phases by the single anomalous dispersion method, making use of the anomalous scattering properties of the heme irons. phenix.xtriage was used to assess the extent of the anomalous signal that was present to 2.6 Å resolution. A search for two iron atoms yielded one single solution with five sites (corresponding to two fully occupied heme iron atoms and three partially occupied iodides), yielding a Bayes correlation coefficient of 36.6% for map skew and a figure of merit of 37.9%. The electron density map was not readily interpretable and was subjected to one round of thorough density modification with the program RESOLVE (38) as implemented in PHENIX (39). The map was significantly improved, allowing the recognition of the heme prosthetic groups and secondary structure elements, yielding a correlation for local r.m.s.d. electron density of 77%. The Auto-Build wizard in PHENIX was used for automated model building and was able to build 182 residues out of 252 (full construct), with 148 docked to sequence with a map-model correlation coefficient of 45%. Iterative manual model building and refinement were carried out in a cyclic manner with COOT (40) and phenix.refine (41) until a complete model was built and refinement convergence was achieved ( Table 2).
The Ramachandran diagram was plotted with RAMPAGE (42), and the model was validated with MolProbity (43), as implemented in PHENIX.
Because all TsdA crystals were isomorphous, the remaining structures of TsdA variants were determined by an initial rigid body refinement with BUSTER-TNT (44). The heme prosthetic groups as well as the coordinating residues were removed from the model used in this refinement, and the program macro "Missing Atoms," together with the "-L" flag ("presence of an

RESULTS
Crystallization and Structure Determination-The crystal structure of recombinant TsdA from A. vinosum was solved by iron-single anomalous dispersion. Crystals belong to the monoclinic space C2, with unit cell parameters a ϭ 79.2, b ϭ 69.9, c ϭ 57.9 Å, ␤ ϭ 129.3°. The crystal structure contains one monomer in the asymmetric unit, corresponding to a Matthews coefficient (46) of 2.31 Å 3 Da Ϫ1 and a solvent content around 47%. The "as isolated" structure was refined to 1.98 Å with a R cryst of 13.8% and R free of 19.3%. The model comprises residues Thr 5 -Pro 210 and Ser 216 -Phe 237 , two heme molecules, three iodide ions, one sulfate, and 164 water molecules. In this work, we numbered TsdA residues without integrating the 27-amino acid signal peptide that is removed during transport of the protein into the periplasm (1) (i.e. only the residues present in the mature protein are numbered). The electron density maps are generally of good quality, except for a disordered loop between Leu 211 and Asp 215 and the N-and C-terminal amino acid residues (4 and 15 residues, respectively).
As pointed out previously (32), using NaI as an additive was absolutely crucial to obtain single, good diffraction quality crystals for data collection. Structural analysis shows one iodide in a hydrophobic pocket formed by Pro 186 , Pro 233 , and Pro 235 . The presence of this ion at this position helps to stabilize the interactions that occur with a symmetry-related molecule, namely with residues Pro 11 , Ala 12 /Ala 13 /Ala 14 , Leu 15 /Leu 16 , and Pro 17 . Interestingly, the maps show some disorder in this region with evidence for an alternate conformation of the main chain around Pro 17 . Remarkably, a second iodide ion is located near the cysteine coordinating heme 1 and within 4 Å distance to the side chains of Cys 96 , Arg 82 , Arg 92 , and Ser 100 , the backbone of Gly 194 , and two water molecules. The third iodide is found nested between the side chains of Leu 39 , Pro 40 , and Phe 42 and is only observed in the "as isolated" structure. Final model parameters and relevant statistics for all TsdA structures characterized are shown in Table 2.
Overall Fold of TsdA and Similar Structures-TsdA is a heart-shaped molecule with approximate overall dimensions 35 ϫ 45 ϫ 55 Å 3 . The molecule is organized in two domains related by a pseudo-2-fold symmetry axis, each one displaying where I hl is the Ith observation of reflection h and ͗I h ͘. c CC* as described by Karplus and Diederichs (67).
are the observed and calculated structure factors for reflection h, respectively. e R free was calculated the same way as R factor but using only 5% of the reflections, which were selected randomly and omitted from refinement.
the typical class I c-type cytochrome topology (Fig. 1A). Each domain harbors four ␣ helices (residues 20 -23 (I N ), 49 -51 (IЈ N ), 89 -98 (II N ), and 111-123 (III N )) in the N terminus and four ␣ helices (149 -159 (I C ), 161-164 (IЈ C ), 198 -208 (II C ), and 219 -229 (III C )) in the C terminus that surround each heme molecule (Fig. 1B). The two domains superimpose with an r.m.s.d. of 2.5 Å with 66 structurally aligned residues that correspond to only 16.7% of sequence identity. Helices I N/C and III N/C intersect almost perpendicularly with each other, whereas helices I N/C and II N/C reside on opposite sides of the heme plane (Fig. 1A). Interestingly, there is an insertion of 14 amino acids (residues 35-48), between helices I N and IЈ N corresponding to an extra ␣-helix in the N-terminal domain that is not present in the C-terminal domain. Each domain has an anti-parallel twostranded ␤-sheet, but these do not superimpose with each other, being located in opposite sides of the heme plane. In summary, these observations suggest that TsdA resulted from a gene duplication event.
TsdA coordinates were submitted to the DALI server (47), the highest match being with the SoxA subunit of R. sulfidophilum SoxAX complex (PDB entry 1H31), with a Z-score of 12.0 and r.m.s.d. of 2.5 Å, corresponding to 102 aligned C␣ atoms and 22% sequence identity. Interestingly, the two superimposed domains correspond to the N-terminal domain of TsdA with the C-terminal domain of the SoxA subunit (Fig. 1C). Heme 1 of TsdA superimposes well with heme 2 of the SoxA subunit, both displaying the typical saddle distortion common in c-type cytochromes. Similar to TsdA, SoxAX also displays a pseudo-2-fold symmetry axis relating the two heme domains, as discussed previously (6,21), but this domain arrangement differs from that of TsdA (Fig. 1C). Fig. 1D highlights the pseudo-2-fold symmetry of both TsdA and SoxA protein structures. TsdA also superimposes with other cytochromes, like the C-terminal domain of cbb 3 oxidase subunit III from Pseudomonas stutzeri (PDB entry 3MK7), cytochrome c 6 from the brown alga Hizikia fusiformis (PBD entry 2ZBO), and cytochrome c 6 FIGURE 1. TsdA structure and comparison with SoxA. A, TsdA overall fold with N-terminal domain (residues 5-138) represented in blue and C-terminal domain (residues 139 -237) in red; the heme prosthetic groups are colored by atom type (yellow for carbon, blue for nitrogen, red for oxygen, and dark red for iron; this color code will be used in all figures). B, TsdA domains (colored as in A), superposed with the heme groups displayed in the same colors as their corresponding domains (heme 1 in blue and heme 2 in red). C, structural superposition of A. vinosum TsdA with R. sulfidophilum SoxA, for which the N-terminal domain (residues 51-150) is shown in green (corresponding heme group in dark green), and the C-terminal domain (residues 151-250) is shown in yellow (corresponding heme group in orange). D, structural superposition of A. vinosum TsdA with R. sulfidophilum SoxA individual domains (color code as in B and C).
from the Phaeodactylum tricornutum (PDB entry 3DMI), which are the next closest hits. These proteins display Z-scores of 8.6 (cbb 3 oxidase), and 8.5 (both cytochromes c 6 ), for 77, 74, and 75 structurally aligned C␣ atoms, respectively. Sequence identities range from 23 to 31%, and superposition occurs with the C-terminal domain of TsdA, not with its N terminus as with SoxA.
Heme Coordination in Oxidized TsdA-The structure of the "as isolated" oxidized TsdA shows that the two heme groups are covalently bound to the polypeptide chain through the thioether bonds of Cys 49 and Cys 52 for heme 1 and Cys 160 and Cys 163 for heme 2. The heme 1 iron is hexacoordinated with His 53 as the proximal axial ligand and Cys 96 as the distal one, as proposed previously (1) ( Fig. 2A and Table 3). Remarkably, heme 2 iron is axially coordinated by His 164 and Lys 208 (Fig. 2B and Table 3). This coordination was quite unexpected because this lysine residue is not conserved in other thiosulfate dehydrogenase proteins. Instead, an asparagine residue is usually located at the equivalent position, followed by a strictly conserved methionine (Met 209 in A. vinosum TsdA) (1).
The Fe-Fe distance between the two hemes is around 15.5 Å, with the closest atomic distance being 4.7 Å. This is close to the range seen in other two-domain cytochrome c proteins. For hemes at a distance of Ͻ14 Å, electron transfer has been reported to be rapid and essentially independent of the environment between the hemes (48). The two hemes form a diheme perpendicular stacked arrangement, a typical motif in multiheme cytochromes (49). Furthermore, both TsdA hemes are quite solvent-exposed with positive electrostatic potential around the cleft (see below; Fig. 3). Together, these findings confirm that heme 1 and heme 2 constitute a redox-linked pair with facile electron flow between them.
Structural Properties of Heme 1-Several three-dimensional structures of TsdA were obtained, in different redox states and complexed with ligands, as well as structures of several TsdA variants (see below). The heme coordinations in these structures are summarized in Table 3. The PDB codes mentioned below refer to this table.
Comparison of the characterized structures revealed substantial conformational and chemical heterogeneity around Cys 96 , the distal ligand of heme 1, which is in agreement with previous EPR data (1). We found that in some TsdA structures (e.g. in the TsdA-K208N structure shown in Fig. 4A) (PDB entry  4WQC; Table 3), this residue is present in two different conformations. In the main conformation, Cys 96 is directly coordinating the iron through its S␥, whereas in the minor conformer, this S␥ is tilted away by ϳ50°from the heme-coordinating position and is no longer bound to the iron. The movement of the sulfur is due solely to rotation of the cysteine C␣-C␤ bond and does not involve any major backbone rearrangements.
In several structures, Cys 96 is in a persulfurated state, similar to what has been observed previously in the crystal structure of SoxA (6,7,50). The persulfuration is complete, for example, in the structure where the protein was co-crystallized with tetrathionate (PDB 4WQA) or from a crystal soaked with sodium bisulfite (Fig. 4B, PDB entry 4WQB). However, in some cases, the persulfuration of Cys 96 is only partial (e.g. in one of the two TsdA-K208G structures characterized) (PDB entry 4WQD). Moreover, when the protein is isolated and crystallized in the presence of tris(2-carboxyethyl)phosphine, the persulfuration is not observed (data not shown).
Further heterogeneity was detected for Cys 96 as in some structures it is present in a double oxidized sulfinate state (Table 3). The Cys 96 sulfinate state probably results from oxidative damage during the long crystallization process at room temperature, as also observed in other enzymes (51). In the tetrathionate-soaked crystal, the partially persulfurated cysteine is also oxidized to the sulfinate state (PDB entry 4WQ8), whereas in the dithionite-reduced crystal, Cys 96 is found in two different states, either as a sulfinate or as the unmodified cysteine, and a sulfide was modeled coordinating the heme, which is not bound to Cys 96 (Fig. 4C, PDB entry 4WQ9). This interpretation is based on the shape and contour of the electron density maps, its biochemical plausibility, and refinement results. The presence of a reductant (dithionite) in aerobic conditions leads to generation of reactive oxygen species that are particularly damaging and may produce the sulfinate. On the other hand, if the sulfinate state was already present before dithionite treatment, it will not be reduced because sulfinate is in an irreversible oxidation state. The sulfide ion coordinating the heme in the dithionite-reduced structure (PDB entry 4WQ9) induces a third conformation for Cys 96 (Fig. 4C). This sulfide most likely originates from dithionite, which is known to usually contain a mixture of sulfur compounds.
Close to heme 1, near Cys 96 , a conspicuous cleft is present that is lined by Arg 82 , Arg 92 , and Arg 197 , which lie on the same side of the heme, with Arg 197 closer to the cleft's entrance (Fig.  3B). Arg 197 is not strictly conserved but is replaced by serine in some TsdA sequences (1). This basic cleft is readily solventaccessible, providing access to Cys 96 and heme 1. It is noteworthy that the side chain of strictly conserved Arg 92 is only 3.6 Å away from the S␥ of Cys 96 and is also close to Arg 82 . A positive charge appears to be important at the latter position because sequence comparisons reveal conservation of either arginine or lysine (1). Because Cys 96 is essential for TsdA catalytic activity (1), we propose that this positive groove is the ligand binding pocket that provides access of thiosulfate to the active site. A very similar cluster of basic residues has been found in the SoxA structure and was also suggested to be the active site channel for anionic substrates (6).
In some structures, the anomalous difference map showed a strong blob of electron density (above 5 contour) inside this active site cavity and close by Cys 96 . This extra density was modeled as an iodide ion (from the crystallization buffer), and the refinement outcome was reasonable. Soaking crystals of wild type protein with thiosulfate was not successful in providing a structure with substrate bound to the active site, most likely because the high solvent accessibility is linked to a low residence time of bound substrate. However, a partially occupied thiosulfate is observed in three TsdA structures ( Table 3): that of the K208N variant (Fig. 4A, PDB entry 4WQC), one of the K208G variants (Fig. 5C, PDB entry 4WQE), and the dithionite-reduced wild type protein (Fig. 4C, PDB entry 4WQ9). In all three cases, the thiosulfate is stabilized by hydrogen bonds to Arg 82 , Arg 92 , and/or Ser 100 . Arg 92 is a strictly conserved residue, Arg 82 is conservatively substituted by Lys in some TsdAs (and in SoxA), and Ser 100 is strictly conserved among TsdAs and is replaced by Thr in SoxA (1).
In the TsdA-K208N structure (PDB entry 4WQC), the thiosulfate molecule was modeled with 75% occupancy (Fig. 4A). In this structure, the S1-S2 bond of thiosulfate points toward helix II N (where Cys 96 lies), and the oxygen atoms are closer to the heme. The S␥ of Cys 96 was modeled in two alternate conformations, as described above. The iron-coordinating S␥ is 3.6 Å away from S2 and 4.2 Å from the closest oxygen atom of S 2 O 3 2Ϫ . In the dithionite-reduced structure (PDB entry 4WQ9), the S1-S2 bond of thiosulfate points toward the heme (Fig. 4C). This thiosulfate conformation is rotated by almost 180 o relative to that in the TsdA-K208N (Fig. 4A). We consider this conformation where the thiosulfate S1-S2 bond points toward the heme the more likely physiological position for the substrate (see below).
It is interesting to note that in the three structures where full persulfuration of Cys 96 is observed (TsdA-tetrathionate co-crystal, bisulfite soak, and K208G variant II; Table 3), a sulfite molecule is present at the basic cleft instead of thiosulfate and is in the same spatial position as the sulfonate group of the thiosulfate molecule in the dithionite-reduced structure. Moreover, the position of the S1 atom of thiosulfate is occupied in these structures by a water molecule. Therefore, the geometry and spatial arrangement of the sulfite and the water molecule resemble a thiosulfate molecule (Figs. 4B (PDB entry 4WQB) and 5C (PDB entry 4WQE)). Here, the water oxygen is at a distance of 3.8 and 5.4 Å from the Cys 96 S␥ and S␦ atoms, respectively. The amino acid residues stabilizing the sulfite are identical to those stabilizing thiosulfate. Furthermore, in TsdA- K208G variant II, Cys 96 is persulfurated with a partially occupied sulfite in its vicinity and a thiosulfate further away, close to the cleft's entrance (Fig. 5C, PDB 4WQE). One oxygen of this thiosulfate is only 2.7 Å away from the water molecule that is close to the sulfite and at the same spatial position as the S1 atom of thiosulfate present in the two structures depicted in Fig. 4, A and C (PDB entries 4QWC and 4QW9). Based on these structures with various ligands, a tetrathionate was manually fitted into the TsdA active site (Fig. 5D).
Site-directed mutagenesis studies of Cys 96 -The structural data proved our hypothesis that Cys 96 acts as the sixth distal ligand of heme 1. We have previously reported studies with a TsdA-C96G variant in which the heme ligating cysteine was exchanged by glycine, an amino acid not capable of heme coordination (1). Now, we also replaced Cys 96 with histidine and methionine, which should be capable of heme iron coordination. TsdA-C96H and TsdA-C96M were produced successfully and analyzed by gel filtration. All three TsdA cysteine mutant proteins behaved as monomers and showed the expected size of 28 kDa upon SDS-PAGE (not shown). Electronic absorption spectra strongly indicated that methionine and histidine acted as ligands to heme 1 in the respective variant proteins. First, the Soret peak shifted to a longer wavelength and increased in intensity ( Fig. 6 and Table 4), consistent with replacement of Cys 96 with a stronger field ligand in both cases (52). Second, spectral features characteristic for five-coordinate high-spin heme (52) were not present. An absorbance band at 695 nm, usually indicating methionine ligation, was not observed for the TsdA-C96M variant (Fig. 6B) (53). It should be noted, however, that some methionine-ligated hemes do not exhibit a peak at this wavelength (54). As opposed to the other two variants, TsdA-C96G exhibited a low-intensity spectral feature indicative of the presence of pentacoordinated high-spin heme (Fig.  6D) (i.e. a band at 630 nm in the oxidized state) (53).
Activity in the thiosulfate-oxidizing as well as in the tetrathionate-reducing direction was completely abolished in all of the TsdA Cys 96 variants (Table 5). In addition, it was not possible to reduce any of the TsdA cysteine variants with thiosulfate (not shown). All of these findings clearly indicate Cys 96 as an essential active site residue.  APRIL 3, 2015 • VOLUME 290 • NUMBER 14

JOURNAL OF BIOLOGICAL CHEMISTRY 9231
Heme 2 Undergoes a Ligand Switch upon Reduction-Similarly to heme 1, heme 2 is also quite solvent-exposed, although only one of its propionate groups (bound to the pyrrole ring D), is oriented toward the solvent with Arg 50 and His 51 in its vicinity (Fig. 3D). Toward the interior of the protein, the second propionate arm and pyrrole ring B stack against a wall of hydrophobic residues, such as Pro 4 , Leu 179 , Phe 180 , and Leu 211 . Interestingly, Leu 179 and Phe 180 shape a protuberance over the heme, making this entrance somewhat narrower than in heme 1. Lys 208 , the distal axial ligand to heme 2 in oxidized TsdA, and the neighboring Met 209 are located close to the surface of the protein followed by a flexible loop (Pro 210 -Leu 216 ) that is disordered in all data sets analyzed.
Most remarkably, when crystals of A. vinosum TsdA are reduced by soaking with bisulfite or dithionite, Lys 208 is replaced by the adjacent Met 209 as the distal axial ligand (Fig.  4D). When Met 209 coordinates heme 2, Lys 208 switches its position, with the side chain now pointing toward a turn between two helices, helix II N and another helix not part of the classical class II c-type cytochrome. The side chain of Lys 208 is now within hydrogen bonding distance to the carboxylate of Glu 63 and main chain carbonyl oxygen of Ala 64 . In some other TsdA structures, the side chain of Glu 63 shows a different rotamer or an alternate conformation, suggesting flexibility around this residue. Interestingly, no major backbone conformational changes are observed upon ligand switch; only the end of helix II C , where Lys 208 and Met 209 sit, unwinds along with a positional adjustment of the following residues (Figs. 2B and 4C). The loop comprising residues Leu 211 -Leu 216 is not ordered.
Site-directed Mutagenesis of Lys 208 -TsdA variants were constructed in which the unusual lysine ligand of heme 2 was replaced not only by glycine (TsdA-K208G) but also by asparagine (TsdA-K208N), thereby creating the situation observed in the vast majority of TsdAs from other organisms. An asparagine residue at the Lys 208 equivalent position is highly conserved among proteobacterial thiosulfate dehydrogenases and, for example, found in TsdA from the ␤-proteobacteria Cupriavidus metallidurans, Thiomonas intermedia, and Sideroxydans lithotrophus as well as in the pathogenic ⑀-proteobacterium Campylobacter jejuni (1). Even Marichromatium purpuratum, a member of the same ␥-proteobacterial family as A. vinosum, possesses a TsdA homologue (AHF02870.1) with an Asn residue instead of Lys 208 . The ⑀-proteobacterium Wolinella succinogenes is a notable exception and contains a TsdA homologue (NP_906283.1) with a Thr residue instead of the highly conserved asparagine.
Both A. vinosum TsdA Lys 208 variants behaved as monomers upon analytical gel filtration and exhibited the expected size of 28 kDa in SDS-PAGE (not shown). Similar values for Soret red / 280 nm for TsdA Lys mutants and TsdA wild type protein proved full heme loading (Table 5). Both TsdA Lys 208 variants were catalytically active in thiosulfate oxidation as well as in tetrathionate reduction (Table 5), thus supporting heme 1 as the active site and suggesting heme 2 as the electron exit point.
In the crystal structures of the TsdA-K208N and TsdA-K208G variants, His 164 and Met 209 are coordinating heme 2 (Fig. 5, A and B); the environment around the heme 2 distal ligand is highly disordered and, although modeled, displays high thermal motion parameters (B factors) and some unexplained electron density, which we attribute to different conformations of the loop. It is plausible to assume that the introduction of an amino acid residue without any side chain conferred on the terminal portion of helix II C an extra degree of freedom (note that the disordered loop in the "as isolated" structure starts at Pro 210 ), making this loop extremely flexible and difficult to model. Nevertheless, there is no doubt that Met 209 is the distal ligand in this TsdA variant. This is corroborated by electronic absorption spectroscopy, which indicated Met 209 as the sixth ligand to heme 2 when Lys 208 is replaced by either glycine or asparagine ( Fig. 7 and Table 4). This was evident by the new charge transfer band around 695 nm indicative of Met-coordinated heme iron (53) in the oxidized variant pro-   (Fig. 7, F and G). The finding that Met 209 serves as the ligand for the heme 2 iron in oxidized TsdA-K208N shows that the weak field ligand asparagine cannot functionally replace lysine.
In a last experiment, we replaced Met 209 by glycine. Electronic absorption spectroscopy indicated the presence of fivecoordinate high-spin heme in this variant due to the presence of a band at 630 nm (Fig. 7H) (53). Activity was still present in both the thiosulfate-oxidizing and tetrathionate-reducing directions ( Table 5). A conspicuous difference was observed with regard to S 0.5 for thiosulfate. This value rose by a factor of 5 at pH 5.0 (Table 5). A similar effect was caused by the Lys 208 3 Gly replacement (Table 5). Both observations hint at a possible cooperativity between the two hemes that needs to be further elucidated in the future.
pH Dependence of Activity and Electronic Absorption Spectra-Redox transformations of cysteine-bound hemes are often accompanied by protonation or deprotonation reactions of the essential cysteine ligand. Such reactions cannot be assessed by structural analysis alone, and we therefore performed an inde-pendent set of experiments analyzing the effect of pH on TsdA activity and electronic absorption spectra.
Consistent with earlier results (1), V max for thiosulfate-dependent reduction of the artificial electron acceptor ferricyanide proved to be about 5 times lower at pH 5.0 than at pH 4.0 (4470 versus 28,600 units/mg; Table 5), whereas S 0.5 for thiosulfate decreased from 1.1 mM at pH 4.0 to 0.14 mM at pH 5.0 This results in a 15% higher value for k cat /S 0.5 at pH 5.0 than at pH 4.0 ( Table 5), implying that TsdA is even slightly more efficient in converting thiosulfate to tetrathionate at pH 5.0. At pH 5.0, we also analyzed the tetrathionate reduction ability with reduced methylviologen as artificial electron donor of the TsdA wild type protein. The enzyme exhibited very low specific activity (V max of 82 units/mg) and a relatively high S 0.5 of 1.98 mM for tetrathionate, resulting a k cat of 37 s Ϫ1 and k cat /S 0.5 ϭ 18.7 (Table 5). These findings agree with earlier work that identified A. vinosum TsdA wild type as a more efficient thiosulfate dehydrogenase than tetrathionate reductase (2).
In a second approach, we used electronic absorption spectroscopy to collect experimental evidence for protonation/de- protonation reactions of the active site cysteine in TsdA. This is possible because changes in the protonation state of thiol (thiolate)-bound hemes are often associated with large shifts in the Soret spectral region (54). Indeed, for the reduced A. vinosum enzyme, an increase in pH significantly altered the electronic spectrum (Fig. 8). The Soret band maximum of the reduced protein was red-shifted (from 417 to 420 nm), whereas the intensity of the band significantly decreased with increasing pH (Fig. 8B). This is indicative for the presence of a stronger field ligand of the heme iron at higher pH (52). At a lower pH, the cysteine thiolate is probably protonated, and the Cys 96 thiol would be a weaker ligand to the heme (54). For the oxidized protein, the Soret band maximum remained at 406 nm even at pH 8.0 (Fig. 8A), indicating retention of thiolate ligation.
In the oxidized state of the TsdA wild type protein, a peak at 634 nm was observed at pH 6.5 and pH 8.0 but not at pH 5.0 (Fig. 8A, inset). In the completely reduced state of the protein, the band was not present (Figs. 6D and 8B, inset). A band at the same position has been reported to be indicative for thiolateligated heme (55). This interpretation is corroborated by the fact that the 634 nm absorbance band is absent in all TsdA variants with replacements of Cys 96 (Fig. 6D and Table 3) but present (albeit at about 620 instead of 634 nm) in the TsdA Lys 208 mutant proteins (Fig. 7, F and G, and Table 4).
The 634 nm band became more pronounced when wild type TsdA was partially reduced with its natural substrate thiosulfate or by the addition of 1 mM sodium dithionite (Fig. 8C), pointing at the transient presence of high-spin heme (53), possibly caused by movement of Cys 96 out of the iron coordination sphere. This conclusion is supported by the presence of a structural conformer in which Cys 96 is tilted away from the hemecoordinating position, as described above.
In contrast, the 634 nm intermediate state band observed for the TsdA wild type protein was completely absent in the Cys 96 mutant proteins, unambiguously allocating this spectral feature to heme 1 ( Fig. 6 and Table 4).

DISCUSSION
In this work, we performed a thorough structural and biochemical analysis of a purple bacterial thiosulfate dehydrogenase. The protein was found to contain two heme groups, both with unusual axial coordination. In the oxidized state, heme 1 has a His 53 /Cys 96 ligation, whereas heme 2 is coordinated by His 164 /Lys 208 . All tested TsdA Cys 96 variants were catalytically inactive both in the thiosulfate-oxidizing as in the tetrathionate-reducing direction. In contrast, all TsdA Lys 208 variants retained activity. This identifies Cys 96 as an active site residue and emphasizes a general important role for cysteine in proteins containing His/Cys-ligated hemes (6,8,22,23). One other example where replacement of the heme-ligating cysteine inactivated the protein in vivo is the A. vinosum triheme cytochrome DsrJ (22). Most data are available for the heterodimeric SoxXA protein. Its SoxA subunit is a diheme protein in R. sulfidophilum and Paracoccus pantotrophus (8), whereas only the heme corresponding to TsdA heme 1 is present in SoxA from Starkeya novella. In all cases, the His/Cys-ligated heme equivalent to TsdA heme 1 has been shown to participate in the SoxXA-catalyzed reaction (7,20,50,56). However, it is still a matter of debate to what extent the heme-ligating cysteine is involved in the reaction and how it exerts its function. The His/Cys ligation corresponds to an extremely low redox potential of the corresponding heme in SoxA and DsrJ (23,(57)(58)(59). Values of Ϫ432 Ϯ 15 and Ϫ479 Ϯ 10 mV (versus NHE at pH 7.0) have been reported for the SoxA hemes in P. pantotropus and S. novella SoxA (12,14). Thus, considering the E m of ϩ24 mV for the thiosulfate/tetrathionate couple (60), the Cys coordination of heme 1 as the electron entry point appears somewhat surprising because it would imply a highly unlikely electron transfer from thiosulfate oxidation to TsdA heme 1.
A positively charged cavity found next to heme 1 and Cys 96 probably constitutes the active site of TsdA. In two structures, we observed a bound thiosulfate in the proposed active site region. This region appears to be rather similar to that described for SoxA (6,7,50,61), where in both cases, several arginine residues are expected to create a strong positive electrostatic field suited for binding the anionic substrate thiosulfate. Bamford et al. (21) suggested that the active site cysteine in SoxA is stabilized in its thiolate form (or the persulfide thereof) by its coordination to the ferric c-type heme. Indeed, a thiolate can be expected to be preferred over a thiol as a ligand for an oxidized heme, where the metalloporphyrin dianion unit has a core charge of ϩ1 (54). Furthermore, it was proposed that the basic residues in the vicinity of the active site contribute to stabilization of the cysteinate form (6). For R. sulfidophilum SoxXA, it has been discussed that the thiolate ligand temporarily dissociates from the heme and covalently binds the thiosulfate substrate, resulting in a SoxA-thiocysteine S-sulfate intermediate state, accompanied by reduction of the catalytic heme and a second heme present in the SoxX subunit (6). In this mechanism, the oxidized thiosulfate would subsequently be transferred to a cysteine residue present in the SoxY subunit of the SoxYZ carrier protein. Catalysis by SoxA would thus essentially follow a rhodanese-like mechanism involving a thiosulfate transfer reaction. On the other hand, an S. novella SoxXA variant with a Cys-Met exchange was still catalytically competent in vitro, and it was concluded that the cysteine is not absolutely crucial for SoxXA activity, making a purely rhodanese-like SoxXA reaction mechanism unlikely (7).
The crystallographic data acquired in our study are in principle consistent with different catalytic mechanisms. The thiosulfate molecule observed in two of the structures (Fig. 4, A and C) closely approaches both Cys 96 and the heme 1 iron, and the possibility of a sulfur transferase-like mechanism similar to what has been suggested for SoxXA cannot be discounted (6). However, different interpretations are possible for the results reported herein. The importance of Cys as an active site ligand may lie in its ability to act as a switch. It could keep the heme in a low-redox potential oxidized state but would be a weak enough ligand that it could be moved out of the coordination sphere when thiosulfate enters the active site, thus leading to a strong increase in the heme's redox potential. This suggestion corresponds to the alternative Cys 96 conformation observed in some structures. Combination of two thiosulfate molecules at the active site, liberating two electrons, could then proceed with reduction of the two hemes of TsdA. This suggestion conforms in part to an alternative reaction mechanism discussed recently for SoxXA (57). Whereas Bradley et al. (57) promoted the idea that SoxXA might mediate the direct coupling of thiosulfate to SoxY without dissociation of any heme ligands, we consider the possibility that Cys 96 temporarily exits the heme 1 iron coordination during the reaction cycle, leading to a change in the redox potential such that electron transfer into heme 1 is facilitated. The observed switch at heme 2 from His/Lys to His/Met ligation upon reduction, leading to greater disparity in redox potential between heme 1 and heme 2 (discussed in more detail below), could be a further means to drive the reaction to tetrathionate production. We showed in this work that the S␥ of Cys 96 can adopt a second conformation tilted away from the heme 1 iron and out of its coordination sphere. TsdA exhibits higher specific activity at low pH, which is consistent with the idea that in its protonated form, the S␥ of Cys 96 is a weaker ligand to heme iron that should be more prone to move into the non-iron-coordinating alternate position, promoting catalysis. Furthermore, we showed that when we replaced Cys 96 by the stronger field ligands methionine and histidine, the activity of the enzyme was completely abolished. For the R. sulfidophilus SoxXA, it has also been proposed that the thiolate ligand to the active site heme is protonated upon reduction (20,57). It is difficult to envision how a protonated cysteine could be a suitable acceptor for covalent attachment of thiosulfate.
The structural flexibility of Cys as heme ligands has been reported for the several proteins where they are present, including SoxA (7,8,23). It is a matter of discussion whether this flexibility is a prerequisite for catalysis (59). In the case of TsdA, the flexibility of the active site Cys observed by EPR spectroscopy (1), and especially the movement of S␥ of Cys 96 out of the heme coordination sphere documented in this work by structural and spectroscopic evidence, is highly likely to enable the transient formation of high-spin iron. Our finding that the TsdA-C96G variant has no residual activity contradicts a reaction mechanism that mainly relies on the transient formation of a pentacoordinated state of the catalytic heme, because glycine should per se not be able to act as a heme ligand. On the other hand, this protein variant presents only a low-intensity high-spin signal in the electronic absorbance spectra, indicating that a major heme population may have coordination by water or another exogenous ligand, resulting in a protein state that is not catalytically competent. Further experiments are necessary to clarify this question.
Previously, a critical role for a post-translational cysteine persulfide modification of the SoxA heme ligand for SoxXA has been suggested (6). Whereas in the two structures of SoxXA from R. sulfidophilum and P. pantotrophus, the cysteines at the active site hemes were found to be quantitatively modified to cysteine persulfide, in the S. novella structure, the coordinating residue was modeled as an equal mixture of Cys and cysteine persulfide (7). The detection of the persulfide has been taken as an indication that sulfur substrates, such as thiosulfate, could indeed become temporarily bound to the heme-ligating residue, and it has been proposed that incomplete catalysis (i.e. transfer of solely the thiosulfate sulfone sulfur instead of the complete thiosulfate moiety to SoxYZ) is the cause for persulfuration of the active site cysteine (6). It should however, be noted that even recombinant SoxA that had been produced in E. coli in the absence of thiosulfate or other Sox pathway proteins carried the modification, at least in part of the protein molecules (7,57). The A. vinosum TsdA and all its variants studied here were recombinantly expressed in E. coli, and we also observed partial persulfide modification in some structures. It is noteworthy that in some of these structures, we observe a sulfite molecule at the active site, which we interpret as the product of reaction with thiosulfate, possibly present in the growth medium, leading to formation of the Cys persulfide and sulfite. The possibility exists that the modification is not crucial for catalysis and that it arises from a side reaction, at least in TsdA. The Cys 96 thiolate may be prone to modification when a suitable electron acceptor is not present.
In the structure derived from a dithionite-reduced crystal, thiosulfate was evident in the active site pocket with the sulfane group oriented toward the heme plane (Fig. 4C). Furthermore, in some other structures, the same position is occupied by a sulfite and a water molecule. Taken together, their arrangement closely resembles the spatial organization of thiosulfate with the S1-S2 bond (or the S1 atom of sulfite and the oxygen atom of water) oriented such that the sulfane group is pointing toward the heme plane (Fig. 5C). In none of the structures did we observe a covalent adduct between the thiosulfate molecule and Cys 96 .
Although the back-reaction has never been discussed for the SoxXA-catalyzed reaction, and it is indeed irrelevant in vivo in this case, it is proven that TsdAs can be even more adapted for catalysis of tetrathionate reduction, as in, for example, the enzyme from C. jejuni (2). Although it is not clear in detail which modifications lead to the observed changes in catalytic efficiencies, it is clear from sequence comparisons that this does not involve major changes in the active site residues (1, 2). A mechanism including formation of a covalent thiocysteine S-sulfate intermediate from tetrathionate would involve two serial cleavages of sulfur-sulfur bonds, the first occurring between the two central atoms of tetrathionate and the second occurring between S␥ of Cys 96 and the thiosulfate moiety. Our structural data neither directly exclude nor support such a mechanism, although it is difficult to envision a molecule with the size of tetrathionate approaching Cys 96 close enough for direct reaction with the thiolate.
In one of the structures for the TsdA-K208G variant (PDB 4WQE), a thiosulfate molecule mimicked by sulfite and a water molecule is oriented close to Cys 96 with the help of Arg 82 and Arg 92 , and a second thiosulfate is present in close vicinity (Fig. 5C). It is striking that a tetrathionate molecule modeled into this structure superimposes almost exactly on the thiosulfate plus water and sulfite molecules (Fig. 5D). Thus, this structure reveals how two thiosulfate molecules could fit in the active site before the covalent bond is formed between their sulfane sulfur atoms.
It is clear from the structural analyses presented here that TsdA heme 2 is ligated by Lys 208 in the oxidized state of the enzyme. Oxidation of two thiosulfate molecules to tetrathionate releases two electrons that can be accommodated by the two heme groups of TsdA. We showed that a switch occurs at heme 2 from Lys to Met axial ligation upon reduction. Many six-coordinate heme proteins are known to swap ligands upon reduction/oxidation, and such changes affect the redox potential of the hemes (15,62). Ligand motifs that favor binding to Fe(III) heme generally result in more negative reduction potential values than ligand motifs that favor binding to Fe(II). Soft ligands, such as methionine, stabilize reduced iron, which results in a positive shift in E m values (15,62). On the basis of these considerations, heme 2 is predicted to be more positive in potential in the His/Met than in the His/Lys-ligated state. This should lead to marked hysteresis in electrochemical measurements (63) and can be experimentally verified in the future. The ligand change could thus guide the catalytic electron transfer event such that the gate for the back-reaction is closed upon reduction of heme 2, because the positive redox potential of the His/Met ligated heme hinders its reoxidation.
Overall, our kinetic, spectroscopic, and structural data lead us to propose the following mechanism for thiosulfate oxidation (Fig. 9). Two thiosulfate molecules enter the active site, which probably induces a movement of the S␥ of Cys 96 out of the iron coordination sphere. This ligand movement results in an increase of the redox potential of heme 1, thus allowing the sequential uptake of the two electrons resulting from the conversion of the two thiosulfates to tetrathionate, leading to the reduction of both hemes. Upon reduction, heme 2 undergoes a ligand switch, which increases its redox potential and hinders the back-reaction. The high-potential iron-sulfur protein (HiPIP) is a likely electron acceptor for the A. vinosum TsdA in vivo (64).

CONCLUSIONS
In summary, we provide a thorough functional and structural characterization of A. vinosum TsdA and propose a catalytic mechanism for this enzyme. We established the His/Cys-ligated heme 1 and the His/Lys-ligated heme 2 as the catalytic and electron transfer hemes, respectively. The heme 1-ligating Cys 96 was shown to be essential for catalysis, and heme 2 was shown to undergo a ligand switch with Met 209 replacing Lys 208 as an axial ligand to heme 2 in the reduced state. Future experiments, including magnetic circular dichroism spectroscopy or synchrotron-based x-ray-induced photoreduction experiments coupled with single crystal microspectrophotometry may shed light on short lived intermediates and redox states of the enzyme and may help to clarify the reaction mechanism. . Proposed mechanism of TsdA activity. A reaction scheme for the TsdA-catalyzed oxidation of two thiosulfate molecules to tetrathionate is shown. Two thiosulfate molecules enter the active site, inducing the S␥ atom of Cys 96 to tilt away from the iron coordination sphere. This is probably accompanied by protonation of S␥. The two thiosulfates react directly with each other, liberating two electrons that reduce heme 1 and heme 2. Upon reduction, heme 2 undergoes a shift from His/Lys to His/Met ligation. Tetrathionate is released, and the hemes are oxidized. HiPIP (high-potential iron-sulfur protein) is a likely in vivo electron acceptor for A. vinosum TsdA (64). Cys 96 is deprotonated, and S␥ is re-established as the ligand to heme 1 iron.