Crystal Structure of an Unusual Thioredoxin Protein with a Zinc Finger Domain*

Many Gram-negative bacteria have two cytoplasmic thioredoxins, thioredoxin-1 and -2, encoded by the trxA and trxC genes, respectively. Both thioredoxins have the highly conserved WCGPC motif and function as disulfide-bond reductases. However, thioredoxin-2 has unique features: it has an N-terminal motif that binds a zinc ion, and its transcription is under the control of OxyR, which allows it to be up-regulated under oxidative stress. Here, we report the crystal structure of thioredoxin-2 from Rhodobacter capsulatus. The C-terminal region of thioredoxin-2 forms a canonical thioredoxin fold with a central β-sheet consisting of five strands and four flanking α-helices on either side. The N-terminal zinc finger is composed of four short β-strands (S1–S4) connected by three short loops (L1–L3). The four cysteines are at loops L1 and L3 and form a tetragonal binding site for a zinc ion. The zinc finger is close to the first β-strand and first α-helix of the thioredoxin fold. Nevertheless, the zinc finger may not directly affect the oxidoreductase activity of thioredoxin-2 because the zinc finger is not near the active site of a protomer and because thioredoxin-2 is a monomer in solution. On the basis of structural similarity to the zinc fingers in Npl4 and Vps36, we propose that the N-terminal zinc finger of thioredoxin-2 mediates protein-protein interactions, possibly with its substrates or chaperones.

motif; during the catalytic cycle, the two cysteines alternate between reduced and oxidized forms. When thioredoxin reduces a disulfide bridge, it itself becomes oxidized. The disulfide in thioredoxin must therefore be regenerated by a thioredoxin reductase, which in turn uses the reducing equivalents derived from NADPH (2,3). In addition to functioning as an electron donor, thioredoxin is an essential component of the T7 DNA polymerase and is involved in the assembly of a number of filamentous viruses (4).
Bacterial thioredoxin-2 is a recently described cytoplasmic thioredoxin homolog. It shares ϳ30% overall sequence identity with thioredoxin-1 and contains an N-terminal extension of about ϳ40 residues (5)(6)(7)(8). Thioredoxin-2 seems to play a role in the protection against oxidative stress, as the deletion of its gene in Rhodobacter capsulatus results in increased sensitivity to oxidative agents (8,9). Furthermore, under oxidative stress conditions, the expression of thioredoxin-2 is up-regulated by the transcription factor OxyR (6,9), whereas the expression of thioredoxin-1 does not change. The N-terminal extension contains two additional CXXC motifs and binds a zinc ion with high affinity. Based on in vitro experiments with the oxidative agent hydrogen peroxide, it has been proposed that the zinc ion is released from thioredoxin-2 under oxidative stress (7). However, it is not clear whether the zinc ion is released in intact cells particularly because thioredoxin-2 remains in its reduced state even under oxidative stress (6). Mutation of the zinc-binding domain does not affect the oxidoreductase activity of thioredoxin-2 in vivo, at least not for the substrates analyzed (6). However, it has been reported that the deletion of the zinc finger domain decreases the reductase activity in vitro (5). Thioredoxin-2 has been found in ϳ20 different bacterial species. Proteins with the same name in eukaryotes lack a zinc finger domain. Human thioredoxin-2, an ortholog of bacterial thioredoxin-1, resides in mitochondria. Yeast thioredoxin-2 is an almost identical isoform of cytosolic thioredoxin-1 (10,11).
As the zinc finger of bacterial thioredoxin-2 is found in different organisms, it is unlikely to be without function. Similar domains are found in many proteins that are involved in DNA, RNA, or lipid binding; in protein-protein interactions; in protein stability; or in the sensing of oxidative stress (12). The zinc finger motif usually contains ϳ20 -60 residues with cysteines and/or histidines folded into a tetragonal binding site for a zinc ion. The zinc ion is important in stabilizing the structural fold and binds with high affinity. Although all zinc finger domains share the tetragonal zinc coordination, their structures are as diverse as their functions. It is therefore not easy to predict the structure of a zinc finger domain on the basis of the amino acid sequence. Specifically, a BLAST search using thioredoxin-2 does not reveal any zinc fingers with known structures. The thioredoxin-2 zinc finger contains two CXXC motifs separated by 15-16 residues. A similar pattern is found in DksA (CXXCX 17 CXXC), transcription factor IIB (CXXCX 15 CXXC), and ClpX (CXXCX 18 CXXC) (13)(14)(15)(16), but the structures of these zinc fingers are quite different. The DksA zinc finger has the typical ␤␤␣-fold found in many transcription factors that bind directly to DNA; the transcription factor IIB zinc finger has four short strands that are involved in the interaction with RNA polymerase II; and the zinc finger in ClpX has two ␤-hairpins followed by an ␣-helix and is involved in homodimerization and substrate interaction (14,15). It has been proposed that the zinc finger of thioredoxin-2 is similar to that in the redox sensor protein Hsp33 (7), but there is essentially no sequence similarity. The primary sequence of thioredoxin-2 does not reveal much of its function or structure.
Here, we report the crystal structure of R. capsulatus thioredoxin-2 at 1.92-Å resolution. Thioredoxin-2 has an N-terminal zinc finger and a C-terminal canonical thioredoxin fold. Our results show that the structure of the N-terminal zinc finger is similar to those of Npl4, Vps36, and transcription factor IIB (16 -18). We propose that the zinc finger of thioredoxin-2 mediates protein-protein interactions with its substrates or chaperones during the defense against oxidative stress.

EXPERIMENTAL PROCEDURES
Cloning and Protein Expression-The R. capsulatus thioredoxin-2 gene was amplified from genomic DNA by PCR and cloned into pET28a ϩ between the NdeI and HindIII sites. The recombinant protein has an N-terminal histidine tag and a thrombin site between the tag and thioredoxin-2. The expression vector was transformed into C43(DE3) cells. The transformed cells were grown to A 600 ϭ 0.7 and induced by the addition of isopropyl ␤-D-thiogalactopyranoside to a final concentration of 0.5 mM. Cells were harvested after culturing at 37°C for an additional 4 h. The cell pellet was resuspended in Tris-buffered saline (20 mM Tris, 0.3 M NaCl, and 10% glycerol, pH 8.0), and cells were broken with a Microfluidizer. The recombinant thioredoxin-2 protein was in the supernatant and purified by Ni-NTA 3 chromatography under native conditions. Thrombin was added to the purified sample at a ratio of 10 units of thrombin to 1 mg of recombinant thioredoxin-2. The cleaved sample was concentrated and further purified by size-exclusion fast protein liquid chromatography in a running buffer of 20 mM HEPES and 0.1 M NaCl, pH 7.4. Fractions containing thioredoxin-2 were pooled and concentrated to a final concentration ϳ10 mg/ml.
Crystallization, Data Collection, and Structure Determination-Before trays were set up, dithiothreitol was added to a final concentration of 5 mM, and the protein solution was incubated at room temperature for 1 h. Crystals were grown in 28 -36% polyethylene glycol 400, 10 mM ZnSO 4 , and 50 mM N-(2-acetamido)iminodiacetic acid, pH 6.5, by the hanging drop method at 22°C. Crystals usually grew to 200 ϫ 200 ϫ 50 m 3 in 1-2 weeks. Crystals were looped directly from the drop and flash-frozen in liquid nitrogen. Data were collected using synchrotron radiation at a zinc peak wavelength of 1.2826 Å at beam line ID19 of the Advanced Photon Source. A data set for single-wavelength anomalous diffraction (SAD) phasing was collected. The images were indexed and scaled by HKL3000 (19). Phasing, map calculation and improvement, and automated model building were done in HKL3000. Briefly, using the zinc SAD data set, the SHELX module in HKL3000 found three zinc sites (20). The zinc sites were refined and phases were calculated using mlphare (21). Automated model building was carried out using the ARP/wARP program (22). The electron density map was further improved by NCS averaging using the NCS operator derived from the partially built model (21). The map was used for a second round of ARP/wARP automated model building. Most of the model (ϳ90%) was built. The rest was manually built using Coot (23). The refinement was carried out in CNS (24) against a second data set at 1.92-Å resolution.
Analytical Ultracentrifugation-Analytical ultracentrifugation was carried out on reduced thioredoxin-2 in a Beckman Coulter Optima XL-A ultracentrifuge. Three samples with initial concentrations of ϳ15, ϳ40, and ϳ80 M thioredoxin-2 were loaded on a 6-well cell and centrifuged at three speeds of 24,000, 28,000, and 36,000 rpm. Data were collected after equilibrium was achieved as indicated by no changes in two successive scans separated by 4 h. Data were analyzed by Beckman XL-A data analysis software using MULTIFIT to an ideal selfassociation model (25).
is the same as R cryst for a selected subset (5%) of the reflections that was not included in prior refinement calculations. e r.m.s.d., root mean square deviation.
The Ni-NTA beads were washed three times with 1 ml of the same buffer and then washed three times with 1 ml of the same buffer containing 30 mM imidazole. The proteins were eluted with 300 mM imidazole. Samples eluted from the Ni-NTA columns were run on SDS-polyacrylamide gel under reducing conditions. The bands in the SDS gel were cut out and analyzed by liquid chromatography/tandem mass spectrometry (Taplin Mass Spectrometry Facility, Harvard Medical School).

RESULTS AND DISCUSSION
Overall Structure of Thioredoxin-2-R. capsulatus thioredoxin-2 was crystallized, and its structure was determined using SAD originating from the bound zinc ion (Table 1). Thiore-doxin-2 has an N-terminal zinc finger motif and a C-terminal canonical thioredoxin fold (27). The thioredoxin fold consists of a five-stranded ␤-sheet with flanking ␣-helices (Fig.  1A). The secondary structural elements are ␤␣␤␣␤␣␤␤␣, as shown in Fig. 1B. The distance between the sulfur atoms of Cys 73 and Cys 76 at the active site is 3.7 Å, indicating that, in the crystal, thioredoxin-2 is in its reduced state. This is expected because 5 mM dithiothreitol was added to the protein solution during crystallization. Nevertheless, the structure can be superimposed on the oxidized form of E. coli thioredoxin-1 with a backbone root mean square deviation of 0.93 Å (Fig. 1C), in agreement with the previous observation that the reduced and oxidized forms of thioredoxin-1 are almost identical (28). The active sites of thioredoxin-1 and -2 are also very similar (Fig. 1D). The highly conserved WCGPC motif is located at the loop following the central ␤-strand (␤2) and the beginning of the second ␣-helix. The active-site cysteine (Cys 73 in thioredoxin-2 and Cys 32 in thioredoxin-1) is at the beginning of helix ␣2 and is solventexposed. This cysteine has a relatively low pK a (29). The deprotonated thiolate is presumably stabilized by the helical dipole moment of helix ␣2 (30). The second conserved cysteine (Cys 76 in thioredoxin-2 and Cys 35 in thioredoxin-1) has a higher pK a . It is likely to be deprotonated through a general base catalysis by an aspartic residue in the ␤2-strand (Asp 67 in thioredoxin-2 and Asp 26 in thioredoxin-1). Because the carboxyl group of this aspartic residue is ϳ5 Å away from the second sulfur atom, the deprotonation is probably mediated by the polarization of a water molecule. Another common feature among the thioredoxins is a positively charged residue (Arg 77 in thioredoxin-2 and Lys 36 in thioredoxin-1) that interacts with the backbone carbonyl group of the residue preceding the conserved tryptophan (Trp 72 and Trp 32 , respectively). This interaction might be important to maintain the active-site conformation. Overall, thioredoxin-1 and -2 likely use the same mechanism to catalyze disulfide reduction of their substrates.
The N-terminal Zinc Finger-The N-terminal extension of thioredoxin-2 is folded into a zinc-binding finger, consisting of four short strands (S1-S4) connected by three short loops (L1-L3) ( Fig. 2A). The S1/S2 strands are hydrogen-bonded with each other and connected by loop L1. The S3/S4 strands are shorter than the S1/S2 strands with fewer hydrogen bonds formed between them; they are connected by the short loop L3. The two CXXC motifs are located in the loops (L1 and L3) at the end of the four strands. The four cysteines form a tetragonal zinc-binding site. The zinc ion is clearly visible in the anomalous difference map (Fig. 2B). The S1/S2 strands are associated with the S3/S4 strands through the zinc ion and through hydrophobic interactions. A number of hydrophobic residues (Leu 12 and Val 23 in the S1/S2 strands and Pro 32 , Ile 42 , and Leu 43 in the S3/S4 strands) are involved in packing these two structural elements together (Fig. 2A). These residues are relatively well conserved. Although the hydrophobic interactions are likely to be important, the zinc ion probably plays a dominant role in keeping the S1/S2 and S3/S4 structural elements together. The distance between the zinc ion and the sulfur atoms in the cysteines is Ͻ2.3 Å, indicating a strong interaction.
In addition to the zinc ion in the zinc finger domain, a second zinc ion, coordinated by His 111 and Asp 58 , is observed in one protomer in the crystal (Fig. 2B). The zinc ion at this site has a relatively low occupancy and is probably bound with low affinity.
The Zinc Finger Is Not Near the Active Site of Thioredoxin-2-In our structure, the zinc finger is not close to the WCGPC active site. Rather, it is located at one side of the thioredoxin fold defined by the first strand (␤1) and the first ␣-helix (␣1) (Fig. 1, A and B). The zinc ion is almost at the opposite end of the active site, with a distance of ϳ30 Å between them. It is thus quite unlikely that the zinc finger can directly affect the reactivity of the active site within the same molecule. It is also unlikely that the zinc finger would directly compete for substrate binding within a protomer. The known structures of complexes between a thioredoxin fold and a substrate (DsbD␣⅐DsbD␥ (31), DsbA⅐DsbB (32), and thioredoxin⅐NF-B (33)) indicate that the substrate-binding region is defined by the WCXXC active site at the ␤2-␣2 region and the beginning of the ␤4-strand where the Gly-Ile-Pro conserved motif is located (Fig. 1D). These regions are far away from the zinc finger domain.
Dimer in Crystal and Monomer in Solution-Although the zinc finger is not close to the WCGPC active site or the substrate-binding groove within a protomer, it is conceivable that it could be close to one of these in an oligomer. In our crystals, the two protomers in the asymmetric unit are arranged such that the zinc finger of one protomer is close to the active site of  the second protomer (Fig. 3A). Notably, the side chain of the conserved residue Arg 77 forms a hydrogen bond with the backbone carbonyl of Cys 37 in loop L3 of the zinc finger. In addition, the side chain of Gln 102 is hydrogen-bonded with the carbonyl of Ala 16 in loop L1. Finally, residues in loops L1 (Leu 15 , Ala 16 , and Cys 17 ) and L3 (Cys 37 , Gly 38 , and Ala 39 ) are in van der Waals contact with residues around the active site (Pro 71 , Trp 72 , Cys 73 , Pro 75 , and Arg 77 ) (Fig. 3A). These interactions suggest that the zinc finger might affect the oxidoreductase activity of the thioredoxin fold in a dimer or higher oligomer. However, several lines of evidence suggest that thioredoxin-2 is monomeric in solution. First, the buried surface between the protomer is relatively small, ϳ1016 Å 2 (7.5% of the total surface). Second, the reduced form of R. capsulatus thioredoxin-2 runs at the position of a monomer in gel filtration experiments (data not shown). Similar results have also been reported for both the reduced and oxidized forms of E. coli thioredoxin-2 (7). To exclude that thioredoxin-2 dimerizes with low affinity, we performed analytical ultracentrifugation experiments, which can measure K d values as low as 1 mM. Thioredoxin-2 at three different concentrations (ϳ15, ϳ40, and ϳ80 M) was used and centrifuged to equilibrium at three different speeds. The data were fit to a selfassociation model. The best fit was obtained with a monomer species of 14.3 Ϯ 0.6 kDa (theoretical molecular mass of 15.5 kDa) (Fig. 3B). The deviations from the theoretical values (Fig. 3B, upper panels) were close to zero and randomly distributed. The results indicate that thioredoxin-2 is a monomer in solution even at concentrations as high as 80 M. The dimer observed in the crystal is thus likely to be a nonphysiological artifact. It should be noted that in other thioredoxin crystals, dimers or oligomers were also observed in the asymmetric unit (34 -36). Given that the interfaces between the monomers are different in these structures and that the existence of dimers or oligomers in solution has not been demonstrated for any of the wild-type thioredoxins, it appears that all members of the family function as monomers.
The Zinc Finger of Thioredoxin-2 Is Homologous to Those That Mediate Protein-Protein Interactions-We performed a Dali search using the zinc finger of thioredoxin-2 (residues 7-44) to find homologous structures. Four structures (Vps36 (Protein Data Bank code 2J9U), Npl4 (code 1NJ3), 30 S ribosomal protein S27e (code 1NVH), and YAF2 (code 2D9G)) were found with Z scores Ͼ2 (37). The structures are considered to be similar if the Dali Z score is Ͼ2. All of them contain two CXXC motifs that form a tetragonal zinc-binding site. The length between the CXXC motifs varies from 10 residues in Npl4 to 16 residues in thioredoxin-2. To illustrate the structural similarity, the zinc fingers of Vps36 (code 2J9U) and Npl4 (code 1NJ3) were superimposed on that of thioredoxin-2 with an overall backbone root mean square deviation of ϳ1.9 Å (Fig.  4A) (16 -18). As shown in Fig. 4A, the positions of the zinc ions are very close to each other. Loop L2 is short in Npl4 (shown in yellow) and relatively long in thioredoxin-2 and Vps36 (shown in pink and blue, respectively), yet the overall structure at the zinc-binding site is similar.
Among the four proteins, the functions of two zinc fingers (Vps36 and Npl4) have been well studied. Vps36 is a component of the ESCRT-II complex (endosomal sorting complex required for transport) that is involved in the sorting and transport of ubiquitinated proteins to multivesicular bodies. The Vps36 zinc finger binds to Vps28, an ESCRT-I component, with high affinity (nanomolar range). Residues involved in the association (Ile 122 , Cys 123 , Asn 145 , and Cys 146 ) are located in two loops (L1 and L3) of the zinc finger domain (Fig. 4B) (17). Mutations in the binding interface abolish the binding to Vps28. The zinc finger of Npl4 has been shown to bind to ubiquitin with low affinity (with a K d of ϳ100 -200 M). A conserved TF (Thr 13 -Phe 14 ) motif immediately following the first CXXC motif has been implicated in ubiquitin binding as shown by chemical shift perturbation and mutagenesis studies. In addition, Met 25 near the second CXXC motif is also important for ubiquitin binding. The TF binding surface is not exactly the same as that of Vps36 (Fig. 4B) (16). Nevertheless, both are located at loops L1 and L3.
We noticed two surfaces of the thioredoxin-2 zinc finger that are involved in protein-protein contacts in the crystal. The first binding surface is the one in the dimerization interface (Figs. 3A and 4B). This binding surface resembles that of Vps36 and, to a lesser extent, that of Npl4. The second binding surface is shown in Fig. 4C. It involves the unstructured N terminus (residue 1-7) of one protomer interacting with one side of the zinc finger domain of a symmetry-related molecule. Met 2 and Lys 5 in the N-terminal segment form four hydrogen bonds with Ala 20 , Lys 22 , and Lys 33 in the zinc finger (Fig. 4C). On the basis of these interactions and the similarity to other zinc finger domains, we propose that this type of zinc finger mediates protein-protein interactions. Specifically, the zinc finger of thioredoxin-2 may be involved in the binding of substrates and/or chaperones. Thioredoxin-2 has been found to interact with a number of proteins, including the DnaJ and DnaK chaperones, in a largescale search for interacting proteins in E. coli (38). Our attempts to confirm the interaction with DnaJ by direct pulldown experiments have been unsuccessful (data not shown). However, the interaction with DnaK could be confirmed by us in experiments in which we pulled down His-tagged thioredoxin-2 with Ni-NTA beads and identified bound proteins by mass spectrometry (Fig. 5). A thioredoxin-2 mutant lacking the zinc domain appeared to bind less DnaK (Fig. 5, lane 3 versus lane 4). These data suggest that the zinc domain may interact with the DnaK chaperone, but they do not exclude the possibility that a fraction of this domain may be unfolded and thus be bound by DnaK.
The zinc finger of thioredoxin-2 is different from that of Hsp33, a redox sensor, and from those of DNA-binding proteins. These results indicate that thioredoxin-2 does not act as a redox sensor, is not involved in protecting DNA against oxidative damage, and is not involved in transcription, yet conservation of this domain suggests that it has an important function. On the basis of the structural homology to the zinc fingers of Npl4 and Vps36, we propose that the zinc finger of thioredoxin-2 mediates protein-protein interactions.
Thioredoxin-2 Is Conserved among Proteobacteria, and Its Zinc Finger Is Seen in Other Proteins-Although thioredoxin-2 was initially found only in E. coli and Corynebacterium nephridii (5), we found 85 homologs using the zinc finger domain (residues 1-40) of the protein from R. capsulatus in a BLAST search. All of these proteins are from Gram-negative pro- teobacteria, representing all five (␣-⑀) classes. The only exception is the protein from C. nephridii, a Gram-positive actinobacterium. We then performed a pattern search using CXXCX 10 -20 CXXCX 30 -40 WCXXC in which we allowed 10 -20 residues between the two CXXC motifs in the zinc finger and 30 -40 residues between the second CXXC motif and the WCXXC active site of thioredoxin. We identified 12 additional homologs from Gram-negative proteobacteria, four from Gram-positive actinobacteria (Streptomyces avermitilis, Arthrobacter aurescens, Mycobacterium vanbaalenii, and Mycobacterium flavescens), and one from a Deinococcus-Thermus species (Thermus thermophilus). Thus, thioredoxin-2 is conserved in Ͼ100 different bacteria. It is interesting that a zinc finger motif homologous to that of thioredoxin-2 is also present in the Paraquat-inducible membrane protein Pqi and in a diguanylate cyclase/phosphodiesterase from Marinobacter aquaeolei VT8. The zinc finger may thus have evolved from a common ancestor and be used in various proteins as a functional module. This is consistent with our proposal that the N-terminal zinc finger is involved in protein-protein interactions.