Y13C Azotobacter vinelandii Ferredoxin I

Ferredoxins that contain [4Fe-4S]2+/+ clusters often obtain three of their four cysteine ligands from a highly conserved CysXXCysXXCys sequence motif. Little is known about the in vivo assembly of these clusters and the role that this sequence motif plays in that process. In this study, we have used structure as a guide in attempts to direct the formation of a [4Fe-4S]2+/+ in the [3Fe-4S]+/0location of native (7Fe) Azotobacter vinelandiiferredoxin I (AvFdI) by providing the correct three-dimensional orientation of cysteine ligands without introducing a CysXXCysXXCys motif. Tyr13 ofAvFdI occupies the position of the fourth ligating cysteine in the homologous and structurally characterized 8Fe ferredoxin fromPeptococcus aerogenes and a Y13C variant ofAvFdI could be easily modeled as an 8Fe protein. However, characterization of purified Y13C FdI by UV-visible spectra, circular dichroism, electron paramagnetic resonance spectroscopies, and by x-ray crystallography revealed that the protein failed to use the introduced cysteine as a ligand and retained its [3Fe-4S]+/0cluster. Further, electrochemical characterization showed that the redox potential and pH behavior of the cluster were unaffected by the substitution of Tyr by Cys. Although Y13C FdI is functional in vivo it does differ significantly from native FdI in that it is extremely unstable in the reduced state possibly due to increased solvent exposure of the [3Fe-4S]0 cluster. Surprisingly, the x-ray structure showed that the introduced cysteine was modified to become a persulfide. This modification may have occurred in vivo via the action of NifS, which is known to be expressed under the growth conditions used. It is interesting to note that neither of the two free cysteines present in FdI was modified. Thus, ifNifS is involved in modifying the introduced cysteine there must be specificity to the reaction.

monomer that contains one [3Fe-4S] ϩ/0 cluster and one [4Fe-4S] 2ϩ/ϩ cluster. The sequence for the NH 2 -terminal half of AvFdI is homologous to the sequences of the much smaller clostridial-type ferredoxins that contain two [4Fe-4S] 2ϩ/ϩ clusters (1). Comparisons of these and many other [Fe-S] protein sequences have shown that the presence of a highly conserved CysXXCysXXCys motif can be used to predict that a [4Fe-4S] 2ϩ/ϩ cluster will be present (for reviews, see Refs. [2][3][4][5][6]. During the evolution of the 7Fe ferredoxins from the 8Fe ferredoxins, two residues were inserted into this motif between the second and third cysteines to form a CysXXCysXXXXCys motif (1). This insertion moved the second cysteine out of reach of the cluster resulting in the inability to form a [4Fe-4S] 2ϩ/ϩ cluster and the appearance of a [3Fe-4S] ϩ/0 cluster in that position (7)(8)(9)(10)(11)(12)(13).
The structural consequences of inserting the two additional residues between the second and third cysteines are shown in Fig. 1, which compares the structure of native AvFdI (7)(8)(9) to that of Peptococcus aerogenes ferredoxin (PaFd) (10 -13) in the [3Fe-4S] ϩ/0 cluster region of AvFdI, which has the sequence Cys 8 XXCys 11 XXXXCys 16 . The comparison in Fig. 1 reveals that the chain trace of PaFd is puckered out in AvFdI, leaving the AvFdI Cys 11 removed away from the position of the ligating cysteine in PaFd. Cys 11 could therefore not be used as a cluster ligand without substantial structural rearrangement. A further structural difference in this region is that the COOHterminal half of AvFdI, which is absent in the much smaller PaFd, wraps around the AvFdI residue 8 -16 loop shielding it from solvent (9). In this study, we have used structure as a guide in attempts to direct the formation of a [4Fe-4S] 2ϩ/ϩ cluster in the [3Fe-4S] ϩ/0 location of native AvFdI by providing the correct three dimensional orientation of cysteine ligands. As shown in Fig. 1, Tyr 13 of AvFdI occupies the position of the fourth ligating cysteine in PaFd so here we report the purification and characterization of Y13C FdI.

EXPERIMENTAL PROCEDURES
Materials-Native AvFdI was purified as described previously (14). Ammonium sulfate was from Fisher and all other materials were obtained from the vendors listed previously (15).
Mutagenesis of fdxA and Expression and Purification of Y13C FdI-The oligonucleotide used for the mutagenesis had the sequence GCAAGTACTGCCATTGTGTTGGTGCAAGTGCACCGATTG. This sequence differs from the wild-type sequence (16) by the change of codon 14 from TAC to TGC, resulting in the change of FdI residue 13 from a tyrosine to a cysteine. The success of the mutagenesis was confirmed at the DNA level by dideoxy-DNA sequencing and at the protein level by NH 2 -terminal protein sequencing (which was carried out in the protein sequencing facility in the department of Molecular Biology and Biochemistry at the University of California, Irvine) and by x-ray crystal-* This work was supported in part by National Institutes of Health Grant GM-45209 (to B. K. B.), GM-36325 and GM48495 to (C. D. Stout) and by grants from UKEPSRC and BBSRC (to F. A. A.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Atomic coordinates and structure factors (Code 1FTC) were deposited in the Protein Data Bank, Brookhaven National Laboratory.
** To whom correspondence should be addressed. Tel.: 714-824-4297; Fax: 714-824-8551; E-mail: bburgess@uci.edu. 1 The abbreviations used are: Fd, ferredoxin; AvFdI, Azotobacter vinelandii ferredoxin I; PaFd, Peptococcus aerogenes ferredoxin; ␤ME, ␤-mercaptoethanol; CSS, cysteine persulfide residue; PIPES, 1,4piperazinediethanesulfonic acid; Mes, 4-morpholineethanesulfonic lography. Unless otherwise indicated, oligonucleotide-directed in vitro mutagenesis (15), FdI overexpression (17), and cell growth (15) were performed as described previously. The purification of Y13C was carried out anaerobically using a modification of purification method two described by Stephens et al. (14). A. vinelandii containing the overexpression vector for Y13C FdI was grown and cell-free extracts were prepared from 1.1 kg of cell paste. No sodium dithionite was used in the purification of Y13C FdI because it became obvious after repeated attempts to purify the protein according to the native FdI purification scheme that the presence of 2 mM sodium dithionite in the buffers denatured this variant of FdI. All steps, however, were carried out anaerobically using thoroughly degassed buffers. The cell-free extracts were not heat treated as described in the purification of nitrogenase (18) due to the possibility that Y13C FdI would be heat-labile. Cell-free extracts were loaded onto a 5 cm ϫ 20-cm DE52 cellulose column equilibrated with extensively degassed 0.025 M Tris-HCl, pH 7.4. Following loading and washing, the column was chromatographed with a 0.1 to 0.4 M NaCl gradient (1.6 liters of each, 570 ml/h) in the same buffer. After this initial gradient Y13C FdI was eluted from the column batchwise with 0.5 M NaCl. The ferredoxin fraction was immediately diluted 1:2 in the same buffer without salt and loaded onto a 2.5 cm ϫ 10-cm DE52 cellulose column. After loading, the column was washed with 4 liters of 0.025 M Tris-KCl, pH 7.4, 0.12 M KCl, at a rate of 300 ml/h. A gradient from 0.12 to 0.3 M KCl (800 ml of each) followed, and Y13C FdI eluted close to 0.25 M KCl. The protein was concentrated on a DE52 cellulose column, and 10-ml samples of concentrated protein were then loaded onto a 1 m ϫ 2.5-cm Sephadex G75 superfine gel filtration column equilibrated in 0.025 M Tris-HCl, pH 7.4, 0.1 M in NaCl. Following elution from that column, 40 ml of 70 -80% saturated ammonium sulfate in 0.1 M Tris-HCl, pH 7.4, was added slowly with stirring to 13 ml of Y13C FdI. The protein precipitated over the course of 2 h and was centrifuged in airtight centrifuge tubes in a SS34 rotor at 6,000 rpm for 35 min. The A 280 /A 400 ratio of Y13C FdI following ammonium sulfate precipitation was 1.7 and the A 280 /A 260 ratio was 1.2, values very similar to those reported for the native protein (14). The final yield of Y13C FdI was 19 mg.
Spectroscopy-For spectroscopic studies, all samples were prepared anaerobically under argon in a vacuum atmospheres glove box (O 2 Ͻ 1 ppm) using degassed buffers. Where indicated samples were first concentrated using a Centricon-10 microconcentrator and were then buffer exchanged into 0.1 M potassium phosphate buffer, pH 7.4, or into a mixture of 50 mM PIPES and 50 mM TAPS buffers at pH 6.0 or 8.3. Electron paramagnetic resonance spectra were obtained using a Bruker 300 Ez spectrophotometer, equipped with an Oxford Instrument ESR-9002 liquid helium continuous flow cryostat. Circular dichroism spectra were obtained using a JASCO J720 spectropolarimeter, and UV-visible spectra were obtained with a HP 8452A diode array UV-visible spectrophotometer.
Electrochemistry-Purified water of resistivity ϳ18 ohm/m (Millipore) was used in all experiments, and each of the reagents was at least analytical grade. DC (staircase) cyclic voltammetry was carried out using an Autolab electrochemical analyzer (Eco-Chemie, The Netherlands). The all-glass cell, electrodes, and application of protein film voltammetry have been described previously (15,19). The saturated calomel reference electrode was held at 22°C, and potential values were converted to the standard hydrogen electrode scale by using E (saturated calomel reference electrode) ϭ ϩ243 mV. The pyrolytic graphite edge working electrode (area approximately 0.18 cm 2 ) was prepared for protein film formation by polishing with an aqueous alumina slurry (Buehler Micropolish 1.0 m) and sonicating extensively in water to remove traces of Al 2 O 3 . All voltammetric experiments were carried out in an anaerobic glove box (Belle technology) with the O 2 Ͻ 2 ppm. To prepare films, ϳ1 l of protein solution (typically 100 M in 20 mM Hepes, pH 7.0, containing 0.1 M NaC1 with 200 g/ml polymyxin as co-adsorbate) was applied to the surface of the chilled pyrolytic graphite edge electrode using a glass capillary tip. The buffer-electrolyte solution in the cell contained 0.1 M NaCl, 20 mM mixed buffer (5 mM of acetate, Mes, Hepes, and TAPS) and 200 g/ml polymyxin. Reduction potentials were determined from the average of peak positions measured in the directions of increasing (E pa ) and decreasing (E pc ) potentials.
Crystallization-All crystallization experiments were done in an anaerobic glove box at Ͻ1 ppm O 2 using degassed solutions. Y13C FdI is less soluble than native FdI so we were unable to induce nucleation using seed crystals of native FdI or other mutants as done previously (20,21). Refinement of the native crystallization conditions (8,9) did not yield crystals. Subsequently, crystallization screening was done using the vapor diffusion method and conditions developed by E. A. Stura (22). Clusters of small needle-like crystals were observed using sodium citrate as a precipitant. However, refinement of these conditions did not produce larger crystals, although the crystals could be grown reproducibly. On an attempt to induce nucleation in droplets that had not formed the small needle-like crystals after 18 days, several of the crystallization plates were transferred in sealed containers from the glove box at 27°C to a room at 18°C for 7 days, then to an incubator at 8°C for 10 days, and then returned to the glove box. Temperature shifts are known to induce nucleation of the native protein (8). After five months (nine months total elapsed time) large, black rod shaped single crystals were observed in several of the drops.
Two crystals were mounted in sealed capillaries for data collection (Table I). Both crystals were grown from a 13 mg/ml protein solution in 25 mM Tris-HCl, pH 7.4, and in both cases the droplets consisted of 2 l of protein solution mixed with 2 l of a 1.0-ml reservoir solution. For the first crystal mounted, the reservoir solution was 1.26 M sodium citrate, pH 7.0, containing 0.5 mM ␤-mercaptoethanol (␤ME). For the second crystal mounted, the reservoir solution was 1.32 M sodium citrate, pH 7.0, without ␤ME added.
Structure Determination-Diffraction data were collected using CuK␣ radiation from a Ru200 x-ray generator operated at 40 kV, 80 mA, and equipped with a graphite monochromator and Siemens area detector mounted on a four-circle goniostat. Data were collected by both and scans in increments of 0.25°with exposure times of 400 s/frame. To minimize the effect of decay, the first crystal was translated in the beam twice. The total time for data collection was 9 days for the first crystal (with ␤ME) and 4 days for the second crystal (without ␤ME). The data from both crystals were indexed, integrated, merged, and scaled using the Xengen suite of programs (Table I) (23). The space group of both crystals is P2 1 with a ϭ 39.3, b ϭ 73.7, c ϭ 45.1A, and ␤ ϭ 98.4°and two molecules (A and B) of Y13C FdI (2.67 Å 3 /Da) per asymmetric unit.
The structure was solved by molecular replacement using the data set from the first crystal and the Xplor version 3.1 suite of programs (24). The native FdI structure from tetragonal crystals (8,9) with residue 13 modeled as a glycine was used as a search model. Rigid body, positional, and isotropic B factor refinement of the structure representing the two best solutions to the rotation and translation searches resulted in an R factor of 0.23 for all data in the resolution range  (Fig. 2). A model for a cysteine persulfide residue (CSS) was constructed and fit to the density using the XtalView suite of programs (25). The model consists of a cysteine amino acid side chain (C␣, C␤, S␥) with an additional sulfur atom (S␦) bonded at 2.02 Å to S␥). The C␤-S␥-S␦ bond angle is tetrahedral. The CSS residues were refined in Xplor using charge and energy restraints of a disulfide for the S␥-S␦ bond and a protonated cysteine sulfur for the terminal S␦ atom. The side chains of a total of 34 Glu, Asp, Lys, Arg, and Gln residues on molecules A and B were adjusted to fit the electron density in 2F o -F c maps due to changes in their conformation in the monoclinic crystal form relative to the tetragonal form. The final model was refined against all observed data to 2.35 Å (Table I). There are no outliers in the Ramachandran plot except for four glycines in both molecules A and B. Co-ordinates have been deposited with the Protein Data Bank.
Because the first crystal was grown from a droplet containing ␤ME, the possibility existed that the additional density observed on the cysteine introduced at position 13 was due to ␤ME, if slow oxidation of the cysteine had occurred over the extended crystallization time. Although no additional density was observed for carbon and oxygen atoms of ␤ME on either molecule A or B, it was possible that these atoms could have been disordered while sulfur bonded to Cys 13 was not. Therefore, a second data set was collected from a crystal grown in the absence of ␤ME (Table I). The data from this crystal was scaled to the first, and a F o -F c map was calculated using data in the range 20.0 -3.0 Å and phases based on the refined model of Y13C FdI, including CSS residues on both molecules A and B. The difference Fourier map was featureless at residue 13, indicating no difference in the crystals grown without ␤ME versus ␤ME. Subsequently, a 2F o -F c map calculated with all data in the range 20.0 -2.67 Å revealed a very similar density for the CSS residue on molecules A and B as observed in the first crystal (Fig. 2). Therefore, we conclude that ␤ME is not a factor in the formation of monoclinic Y13C FdI crystals, and it is not present as a modification of Cys 13 . The Y13C FdI structure was also refined against all observed data to 2.67 Å (Table I).
Crystallographic Evidence for a Cysteine Persulfide-In addition to the shape and height of the electron density for both CSS13 and CSS213 in both the structure with and without ␤ME added (Fig. 2), the presence of a cysteine persulfide was demonstrated by a series of refinement calculations in which each sulfur or oxygen of each non-ligand cysteine and serine was modeled as oxygen and sulfur, respectively (Table II). For example, Cys 11 was modeled as "Ser 11 ," and Ser 56 was modeled as "Cys 56 ." Similarly, the terminal atom on the Cys 13 side chain was modeled as sulfur (the CSS residue) or oxygen, a possible Cys 13 -SOH residue. The only other free Cys or Ser in FdI is Cys 24 , which was also modeled as "Ser 24 ." The corresponding residues in molecule B were modified in the same way. A series of refinement calculations were carried out using the 2.3 Å data from the crystal grown with ␤ME (Table I), Xplor3.1, and the same charge and energy restraints used to refine the Y13C FdI structure (Table I). For the test Cys 13 -SOH side chain, the residue was defined to have a S␥-O␦ bond of 1.8 Å, tetrahedral angle for C␤-S␥-O␦ and OH group as in serine. The eight models, in which only one sulfur or oxygen atom at a time was changed, were independently refined. Three of these models with Ser 11 , Ser 24 , and Cys 56 were also refined against the native FdI data (9) at the same resolution by the same protocol as a further control, since the native FdI crystallizes in a different space group. Finally, the correct native structure with Cys 11 , Cys 24 , and Ser 56 was refined at 2.3 Å for comparison. Table II summarizes the results of the test refinement calculations. The height of the electron density of the atom in question in a 2F o -F c map, calculated in each case with all data in the range 20.0 -2.3 Å, is given in intervals of 1 of the electron density, and the isotropic temperature factor (B factor) for each atom in question is given. In each and every case where a sulfur atom was modeled as an oxygen, the B factor refined to 2.0 Å 2 , the minimum allowed by the program, indicative that the model had insufficient electron density at that site. In particular, a model with oxygen bonded to S␥ of Cys 13 refines to 2.0 Å 2 on both molecules A and B, while the S␥ atom retains a B factor similar to that observed with a persulfide (CSS residue) as a model. Conversely, when an oxygen is modeled as a sulfur (Ser 3 "Cys") the B factor always becomes much larger, indicative of excess electron density at the site in the model. Thus, the fact that a S␦ atom on CSS13 and CSS213 refines to a reasonable B factor indicates that it is not oxygen. Finally, the electron density in each and every case is 4 -5 for sulfur, whether modeled as sulfur or oxygen, and 1.5 for oxygen in serine, whether modeled as sulfur or oxygen. Together, these data demonstrate that Cys and Ser can be discriminated at 2.3 Å resolution, and that the atom bonded to S␥ of Cys 13 is more massive than oxygen and has electron density comparable to that of sulfur in other cysteine residues with comparable B factors. In principle, this atom could be aluminum, silcon, phosphorus, chlorine, or argon in terms of x-ray scattering; however, none of these atoms are likely or able to bond singly to sulfur. Therefore, we conclude that the CSS residue is a cysteine persulfide.

RESULTS AND DISCUSSION
Ferredoxins are small acidic electron transfer proteins that contain [Fe-S] clusters attached to the polypeptide via cysteine residues. When a ferredoxin contains a [4Fe-4S] 2ϩ/ϩ cluster three of the four cysteine ligands are usually supplied by a highly conserved CysXXCysXXCys sequence motif (1,6). Little is known about the in vivo assembly of these clusters and the role that the sequence motif plays in that process. Where x-ray structures are available it is clear that the clusters are buried (7)(8)(9)(10)(11)(12)(13), which suggests that they may be assembled during the protein folding process rather than after the protein is fully folded. For the simpler ferredoxins the protein can fold in vitro from apoprotein with the correct [4Fe-4S] 2ϩ/ϩ in place (26,27), but in vivo other proteins are likely to be involved in the delivery of iron and sulfide (28). Whether the normal assembly process requires a specific sequence motif or just a specific three dimensional orientation of protein cysteine ligands is not known.
AvFdI is a 7Fe protein that contains a [3Fe-4S] ϩ/0 cluster ligated by cysteinyl residues at positions 8, 16, and 49 (7-9). Although there is a free cysteine at position 11, that cysteine is far removed from the [3Fe-4S] ϩ/0 cluster and could not be used as a fourth ligand to the cluster in the absence of substantial structural rearrangement. In this study the x-ray structures of AvFdI and PaFd (7-13) were used as a guide in attempts to obtain the correct three-dimensional orientation of four cysteines without creating a CysXXCysXXCys sequence motif. Fig. 1 shows the structures of one of the CysXXCysXXCys loops of PaFd (11,12), the corresponding Cys 8   X-ray Structure Determination-As described under "Experimental Procedures" and shown in Table I, the x-ray crystal structures for two crystals of Y13C, one grown with ␤ME and one grown without ␤ME, were solved. Both crystals had two molecules (A and B) in the asymmetric unit. Fig. 2 shows the electron density for the refined Cys 13 residue of Y13C in both  (Table I)   molecules A and B in the crystallographic asymmetric unit in both a crystal grown with or without ␤ME present. Two conclusions arise from analysis of these data. First, the engineered Cys 13 fails to serve as a cluster ligand and second, it has been modified to become a cysteine persulfide. The structure of Y13C FdI refined against the higher resolution, more complete data set collected from a crystal grown in the presence of ␤ME (Table I) will be used as a basis for further discussion. The geometry of the persulfides is normal and their conformations are very similar in molecule A and B: S␥-S␦ bonds 2.00, 2.01 Å and C␤-S␥-S␦ angles 104.4°and 108.8°, respectively. The B factors for the persulfides are also similar, being 4.3, 7.3, and 11.9 Å 2 for C␤, S␥, and S␦ respectively, on molecule A, and 6.9, 9.9, and 13.6 Å 2 for these atoms on molecule B. Overall molecules A and B are very similar with an root-mean-square deviation for 106 C␣ atoms after least squares fit of 0.126 Å and a very similar distribution of average B factor per residue over the length of the polypeptide chain. The two copies of FdI in the asymmetric unit are also very similar to native FdI in tetragonal crystals (9) with rms deviations for 106 C␣ atoms after least squares fit of 0.179 and 0.183 Å. However, the conformations of several surface side chains, in particular Lys 98 , are rearranged due to the alternate crystal packing. In the monoclinic unit cell similar contacts occur between residues at the C terminus and the [4Fe-4S] cluster binding loop as in tetragonal crystals, while a local 2-fold axis relating molecule A and B is orthogonal to b, inclined 55°from a, but does not intersect the 2 1 -screw axis. This interaction creates favorable contacts involving Lys 10 , Thr 82 , Glu 83 , Asp 93 , Gly 96 , Lys 98 , Gln 102 , and His 103 . Fig. 3 shows the environment of the cysteine persulfide residue relative to Tyr 13 in native FdI. The persulfide retains hydrophobic contacts to Cys 11 , Pro 50 , Pro 87 , and Ala 91 present in native FdI for Tyr 13 ; a hydrogen bond involving Tyr 13 and Asp 95 is lost. The van der Waals surface of the persulfide occupies a similar volume as the tyrosine, except that, by virtue of being a shorter side chain, the space filled by the tyrosine hydroxyl and C atoms and some of the space filled by the C⑀1 and C⑀2 atoms, is vacant. Consequently, the [3Fe-4S] cluster remains shielded from solvent in Y13C FdI, but is closer to the solvent accessible surface of the protein. In effect, residue 13 in both structures fills a small cleft formed by the cysteine containing loops of residues 8 -16 and 47-54 on either side of the [3Fe-4S] cluster (Fig. 3).
The occurrence of persulfides in protein structures is not without precedent. In the Q206C mutant of subtilisin, a persulfide is observed at the introduced cysteine (29). In this case, as well as in human superoxide dismutase (30), a persulfide apparently arises as an artifact of the purification procedure. In bovine liver rhodanese, a persulfide is observed at the catalytically essential Cys 247 in the active site; the electron density of this residue at 2.5 Å resolution is very similar to that shown in Fig. 2 (31). In Desulfovibrio gigas ferredoxin II (FdII) additional electron density is observed on Cys 11 , a residue analogous to Cys 11 in AvFdI (32). In D. gigas FdII Cys 11 is not a ligand and a [3Fe-4S] cluster is present in the protein; the additional density was modeled and refined as methane thiol (32). In Y13C FdI there is no additional electron density beyond the S␦ atoms of the persulfides. At this point it is not clear how the Y13C FdI persulfide arose but one possibility is that it was formed in vivo via the action of NifS or a NifS-type enzyme (33). These pyridoxal phosphate-containing enzymes appear to assist [Fe-S] cluster assembly by donating sulfide (33), possibly by formation of persulfides on ligand cysteines. It is interesting to note that neither of the free cysteines present in FdI is modified, whereas the engineered Cys 13 is a persulfide. This suggests that if NifS is involved in modifying Cys 13 there must be specificity to the reaction, either involving a Cys ligand motif or a so far unrecognized structural motif. Fig. 4 shows the immediate environment of the CSS13 residue on molecule A of Y13C FdI. The structure of molecule B is virtually identical. Considering the three divalent, inorganic sulfur atoms of the [3Fe-4S] cluster, S␥ of Cys 11 and CSS13, there are six sulfur atoms in close proximity in this structure, and seven interatomic distances less than 6 Å. The conformation of Cys 11 is very similar to that in native FdI. S␥ of CSS13 has the most contacts (four), consistent with the expectation that, if residue 13 were a cysteine, then this sulfur atom could potentially be a ligand to a fourth iron atom, forming a [4Fe-4S] cluster. In particular, the contact between this S␥ and S12 of the [3Fe-4S] cluster is short, 3.44 Å, or 0.26 Å less than the combined van der Waals radii of two sulfur atoms. This distance is comparable to those between S9, S11, and S12 (average, 3.47 Å) and within the range of S-S distances observed in other [3Fe-4S] and [4Fe-4S] clusters (32). For comparison, the contact between Cys 24 S␥, a non-ligand cysteine, and inorganic sulfur of the [4Fe-4S] cluster in native FdI is 3.42 Å. Therefore, under oxidizing conditions it could be expected that a S-S bond could form between CSS13 S␥ and S12, as indeed occurs between Cys 24 and the [4Fe-4S] cluster in the presence of ferricyanide (34). The structure does not, however, explain the observation described below that Y13C FdI is unstable under reducing conditions.
Absorbance and CD Spectra-The UV-visible absorption spectra of Y13C FdI are shown in Fig. 5 compared with those of native FdI. In the O 2 oxidized state these spectra are indistinguishable. These spectra were also unchanged by the anaerobic addition of 2 mM dithiotheritol, which was added in an attempt to reduce the persulfide. Native FdI is an air-stable protein.
Y13C is also fairly air stable when compared with some other FdI mutants (20), but it is not as stable as the native protein.
For example, Y13C FdI could be stored at room temperature in air for about 24 h without cluster destruction, but longer storage eventually led to complete bleaching of the protein. Therefore the protein was purified anaerobically. As shown in Fig. 5 the addition of dithionite to either native or Y13C FdI resulted in partial bleaching of the spectrum. For the native protein this is known to arise from the reduction of the [3Fe-4S] ϩ cluster to the 0 oxidation state, because the reduction potential of the [4Fe-4S] 2ϩ cluster is too low to be reduced by dithionite at this pH (35). The data in Fig. 5 show that the Y13C FdI dithionite reduction behavior is parallel to that of native FdI; however, we were surprised to observe that the Y13C FdI variant was extremely unstable in the dithionite-reduced state with complete destruction of both clusters occurring within 90 min of the addition of dithionite. Thus, although the protein was purified anaerobically, dithionite was not added to any of the buffers. It should be noted that the instability on reduction is not dependent upon the presence of dithionite as it is also observed in the direct eletrochemical experiments described below.
CD spectra of oxidized and reduced native FdI and Y13C FdI obtained at pH 7.4 are compared in Fig. 6. The wavelength dependence and form of the oxidized spectra are again extremely similar, strongly suggesting that the two proteins should have the same cluster composition. For the dithionitereduced protein, the wavelength dependence is also similar but not identical. However, the amount of time required to prepare the samples and collect the data was sufficiently long that some cluster destruction may have occurred in the reduced state so that the minor differences may not be significant. To obtain more information about the reduced [3Fe-4S] 0 cluster, we carried out the CD measurements at both high and low pH. The wavelength dependence and form of the CD spectra of native AvFd are quite different depending upon the pH due to the direct protonation of the reduced cluster at low pH (15,36). The same changes in CD are observed for the Y13C FdI on going from high to low pH, indicating that its reduced [3Fe-4S] 0 can also be directly protonated (data not shown). However, again the experiments must be carried out quickly, because the Y13C protein is extremely unstable in the reduced state.
Electrochemistry-Further confirmation of the similarities between the native and Y13C variant of FdI was obtained by electrochemical characterization. Despite the instability of the protein, it was possible to determine reduction potentials over a limited pH range using protein-film voltammetry in which a monolayer or less of sample molecules contacting different solutions can be examined in a short period of time (19). Fig. 7 shows base line-corrected voltammograms, measured at 20 mV s Ϫ1 for the direction of increasing potential, which reveal the general manner in which the different cluster redox transitions in the Y13C FdI vary with pH. The most prominent feature, by reference to previous studies (37), is assigned to the novel pH-dependent two-electron couple [3Fe-4S] 0/2Ϫ and is designated CЈ. Likewise, the signal with the most positive potential is designated AЈ and is assigned to the well documented couple [3Fe-4S] 1ϩ/0 , while the remaining signal (BЈ) overlaid completely by CЈ at pH Ͻ 7.0 corresponds closely to the potential values of the couple [4Fe-4S] 2ϩ/1ϩ in the native protein. Fig. 7 shows the pH dependence of reduction potentials determined from the average of the oxidative and reductive peak potentials. No differences were observed between values obtained at scan rates 20, 50, and 100 mV s Ϫ1 . Notwithstanding the narrow range over which data were obtained, the curve for couple AЈ could be fit to a scheme consistent with protonation of the "0" oxidation level at pK ϭ 7.2 Ϯ 0.3, and E 0Ј (alkaline form) ϭ Ϫ417 Ϯ 10 mV. The line for couple CЈ is very steep, and, overlaying couple BЈ, it was not realistic to attempt analysis beyond linear regression, which gave a slope of Ϫ74.5 mV/pH unit. At pH 7.0, couples AЈ, BЈ, and CЈ have reduction potentials Ϫ402 Ϯ 5, Ϫ640 Ϯ 10, and Ϫ740 Ϯ 10 mV, respectively, values previously obtained for the native protein. These data therefore indicate that the replacement of tyrosine by a cysteine persulfide does not affect either the pH dependence or reduction potential of the [3Fe-4S] cluster. Attempts were made to detect iron uptake into the [3Fe-4S] 0 cluster using the procedure previously described (38) for observing cluster transformations in FdIII from Desulfovibrio africanus. No changes in signals were observed in experiments carried out at pH 7.0, with Fe(II) concentrations of 370 and 690 M in the buffer electrolyte and holding the potential at Ϫ460 mV for various times before cycling.
EPR-In general, the presence of a [3Fe-4S] ϩ cluster in any protein is easily identified by the appearance of a characteristic g ϭ 2.01 EPR signal (14). As shown in Fig. 8 the EPR spectrum of oxidized Y13C FdI was found to be both qualitatively and quantitatively indistinguishable from that of native FdI. As is the case for native FdI, Y13C FdI is EPR silent in the dithionite-reduced state, consistent with the formation of [3Fe-4S] 0 . No evidence for the conversion of that cluster to a [4Fe-4S] 2ϩ/ϩ cluster was obtained when the reduction was carried out in the presence of excess iron.
Taken together these data demonstrate that, despite the introduction of an additional Cys residue in the correct structural location, a [4Fe-4S] 2ϩ/ϩ cluster fails to form in that location.
Why does Y13C FdI not form a [4Fe-4S] 2ϩ/ϩ cluster in the [3Fe-4S] ϩ/0 location? One possibility is that a CysXXCysXXCys motif is actually required for the in vivo assembly of a [4Fe-4S] cluster in that location. Another is that the presence of the persulfide may prevent Cys 13 from serving as a ligand. Alternatively it may be that the presence of the persulfide indicates that the assembly process was initiated by addition of a sulfide to Cys 13 , but that the subsequent use of that residue as a  cluster ligand was somehow prevented. For example, the [3Fe-4S] cluster may already have had its full complement of inorganic sulfur and therefore have been unable to use CSS13 to insert an inorganic sulfur.
Analysis of the structure also provides clues as to why Cys 13 might not be able to serve as a cluster ligand even if the persulfide could be cleaved in vivo. Thus, residue 13 and its neighbors are involved in a number of contacts with the surrounding protein that may restrain Cys 13 from undergoing small but necessary conformational changes. That some adjustment is needed is indicated by the S␥ to [3Fe-4S] cluster divalent sulfur distances which cannot be made symmetrical by simple rotation of the Cys 13 1 torsion angle. In the Y13C FdI structure these distances at best range from 2.5 to 4.2 Å, whereas in the PaFd structure where an iron atom of the [4Fe-4S] cluster is ligated, the corresponding distances are 3.5 to 4.1 Å. Therefore, even if the additional sulfur of the persulfide were removed Cys 13 would have to shift ϳ1 Å to ligate Fe as Cys 11 does in PaFd. In the structure of AvFdI (9) at least 11 hydrogen bonds may restrain such a shift; the carbonyls and amides of residues 11-15 are involved in hydrogen bonds with the main chain atoms of Cys 8 , Val 17 , Glu 18 , and Leu 88 , and with the side chains of Thr 14 and Lys 84 ; and the side chains of Lys 12 , Thr 14 , and Asp 15 are involved in hydrogen bonds with the carbonyl of Gly 28 , amide of Lys 85 , and the side chains Glu 27 and Lys 84 , forming two salt bridges. In previous studies we have shown that the [4Fe-4S] 2ϩ/ϩ cluster of AvFdI is able to switch one remote Cys ligand for another Cys residue (23,38,42). It should be noted, however, that in the structures of those C20A and C20S FdI variants considerable rearrangement takes place to accommodate a new ligand (Cys 24 ) to a [4Fe-4S] cluster (20,39). In those cases the loop undergoing the conformational change, Cys 20 XXXCys 24 , is on the surface of the protein and is involved in only two hydrogen bonds with other residues.
Physiological Consideration-The data just described demonstrate that Y13C FdI is extremely similar to native FdI in its cluster organization and redox properties, but that it differs from native FdI in being unstable in the reduced state and in the presence of oxygen. The protein also accumulates to only about 5% of wild-type levels in vivo, which may reflect either an initial folding problem or increased oxygen sensitivity leading to protein destruction. In native FdI, Tyr 13 is partially exposed on the surface of the protein but also is in contact with the [3Fe-4S] cluster (9). As indicated above, the replacement of this large, aromatic residue with a cysteine persulfide creates a cavity adjacent to the cluster. Although the cluster is not directly exposed to solvent, the smaller CSS13 chain is unlikely to be as effective in shielding the cluster, perhaps explaining the increased oxygen sensitivity of this mutant and its instability in the reduced state. The Tyr 13 side chain also has hydrophobic contacts with Cys 11 , Pro 50 , Pro 87 , and Ala 91 , and is involved in a hydrogen bond with Asp 95 . The loss of these contacts may expose additional hydrophobic surface area in Y13C FdI as evidenced by the reduced solubility of the mutant protein at high ionic strength, as observed in our early unsuccessful crystallization trials. The exposure of this additional hydrophobic surface might also contribute to the lowered stability of the Y13C FdI variant.
In previous studies we have shown that AvFdI has an electron-transfer function that is important for cell growth (35,40) and a regulatory function in controlling the expression of the fpr gene product (40 -42). Y13C FdI appears to be able to carry out both of these functions as evidenced by the normal growth of the Y13C overproduction strain and by the observation that FPR levels are the same in strains overproducing the native and Y13C variants of FdI. If the in vitro instability of the reduced protein discussed above is due to increased solvent exposure, then something appears to prevent this exposure from becoming a problem in vivo.