Alteration of the Reduction Potential of the [4Fe-4S]2+/+ Cluster of Azotobacter vinelandii Ferredoxin I*

The [4Fe-4S]2+/+ cluster ofAzotobacter vinelandii ferredoxin I (FdI) has an unusually low reduction potential (E 0′) relative to other structurally similar ferredoxins. Previous attempts to raise thatE 0′ by modification of surface charged residues were unsuccessful. In this study mutants were designed to alter theE 0′ by substitution of polar residues for nonpolar residues near the cluster and by modification of backbone amides. Three FdI variants, P21G, I40N, and I40Q, were purified and characterized, and electrochemical E 0′measurements show that all had altered E 0′relative to native FdI. For P21G FdI and I40Q FdI, theE 0′ increased by +42 and +53 mV, respectively validating the importance of dipole orientation in control ofE 0′. Protein Dipole Langevin Dipole calculations based on models for those variants accurately predicted the direction of the change in E 0′ while overestimating the magnitude. For I40N FdI, initial calculations based on the model predicted a +168 mV change in E 0′while a −33 mV change was observed. The x-ray structure of that variant, which was determined to 2.8 Å, revealed a number of changes in backbone and side chain dipole orientation and in solvent accessibility, that were not predicted by the model and that were likely to influence E 0′. Subsequent Protein Dipole Langevin Dipole calculations (using the actual I40N x-ray structures) did quite accurately predict the observed change inE 0′.

Iron-sulfur ([Fe-S]) proteins contain clusters composed of iron and inorganic sulfide atoms ligated to the protein primarily by cysteine residues. They are ubiquitous, and have diverse functions ranging from electron transfer to regulation of gene expression (for recent reviews, see Refs. [1][2][3][4][5][6][7]. In order to carry out these different functions, individual proteins can dramatically alter the reactivity of [Fe-S] clusters in a number of ways. For example, by adding or subtracting iron and sulfide atoms to vary the cluster type (1-7), by bridging a cluster between two subunits (8,9), by introducing non-cysteine ligands (8, 10 -12), by bridging an [Fe-S] cluster to another prosthetic group (e.g. Ref. 13), by grouping multiple clusters in a particular order as revealed by the recent hydrogenase structures (13)(14)(15)(16) or by adding other metals or organic groups as occurs in the [Mo-7Fe-9S-homocitrate] FeMo cofactor sites of nitrogenase (17).
For this study it is especially important to note that, even without modification of [Fe-S] type and organization, proteins are still able to control the reactivities of the clusters they contain. For example, [4Fe-4S] clusters with four cysteine ligands can utilize three different redox couples. The ϩ3/ϩ2 couple is used in a class of proteins designated the high potential iron proteins (18 -20), the ϩ2/ϩ couple is used in most ferredoxins and redox active enzymes (21), and the ϩ/0 couple has recently been reported for the iron protein (Fe protein) of nitrogenase (22,23). Even when a particular [4Fe-4S] redox couple has been selected the reactivity of these proteins can be extended further by protein modulation of the reduction potential (E 0Ј ) of a particular redox couple. Thus, high potential iron proteins have potentials ranging from 90 to 450 mV (18 -21, 24 -29), while ferredoxins that contain structurally indistinguishable [4Fe-4S] 2ϩ/ϩ clusters have reduction potentials ranging from Ϫ280 to Ϫ715 mV in different native proteins (21,30).
This study is focused on the question of how a protein could control the E 0Ј of a [4Fe-4S] 2ϩ/ϩ cluster that is ligated via a typical CysXXCysXXCys motif and one remote Cys ligand. Early studies of protein control of [4Fe-4S] 2ϩ/ϩ E 0Ј focused on three structurally characterized, related proteins. The [4Fe-4S] 2ϩ/ϩ cluster of Azotobacter vinelandii ferredoxin I (AvFdI) 1 has an unusually low E 0Ј of ϳ Ϫ630 mV at pH 8 (31), while the analogous clusters in Peptostreptococcus asaccharolyticus 2 ferredoxin (PaFd) and Clostridium acidiurici ferredoxin (CaFd) have E 0Ј ϳ Ϫ430 mV (32). Thus the E 0Ј for the [4Fe-4S] 2ϩ/ϩ clusters contained within these proteins vary by over 200 mV. Early comparisons of the structures and sequences for these three proteins showed that the peptide folding around the analogous clusters is highly conserved with respect to the location of the four Cys ligands, the Cys dihedral angles, and the eight amide groups H-bonded to sulfur atoms of the cluster (33). These similarities have also been confirmed by the new 1.4-Å structures of AvFdI (34,35) and by the 0.95-Å structure of CaFd (36). Thus, these factors do not appear to be responsible for the observed differences in reduction potential among these proteins that all use the same [4Fe-4S] 2ϩ/ϩ couple.
Another long standing idea is that proteins might control [4Fe-4S] 2ϩ/ϩ E 0Ј by introducing or removing charged residues. Thus, removing a negative surface charge near the cluster should make the cluster easier to reduce (raise E 0Ј ), while adding a negative charge should make it more difficult to reduce (lower E 0Ј ) (18,(37)(38)(39)(40)(41)(42). Indeed, the lower potential AvFdI does have more negatively charged residues near its [4Fe-4S] 2ϩ/ϩ cluster than CaFd or PaFd (43). This idea was attractive because it predicted that the formation of salt bridges between redox partners, when they bind to each other, might serve to raise the potential of the electron acceptor while lowering the potential of the electron donor, thus facilitating electron transfer (37). To test this idea a number of site-directed variants of the lower E 0Ј AvFdI were constructed by changing the negatively charged surface residues near its [4Fe-4S] 2ϩ/ϩ cluster to their neutral or positively charged counterparts in the higher E 0Ј PaFd (43). X-ray structures of the mutant proteins proved that the orientations of the residues were the same in the mutant AvFdIs as they were in native PaFd. Surprisingly, however, the E 0Ј of the AvFd [4Fe-4S] 2ϩ/ϩ cluster was unaffected by these mutations (43). The conclusion from that study was that differences in surface charged residues were not responsible for the large differences in reduction potential observed for the [4Fe-4S] 2ϩ/ϩ clusters of AvFdI and PaFd.
Another factor that has been suggested to be important is the relative solvent accessibility of the [4Fe-4S] 2ϩ/ϩ clusters in the two classes of proteins with one group predicting that the higher E 0Ј of PaFd arises from the presence of buried water molecules (21,44,45). A recent 0.95-Å resolution x-ray structure of a protein in the PaFd class, however, failed to reveal the presence of any internal water molecules (36). Comparison of high resolution structures of AvFdI (34,35) and CaFd (36) also failed to reveal any significant differences in the solvent accessibility of the homologous [4Fe-4S] 2ϩ/ϩ clusters contained within the two proteins.
In this study we use site-directed mutagenesis to manipulate side chain and backbone dipoles that are close to the [4Fe-4S] 2ϩ/ϩ cluster of AvFdI (but not directly H-bonded to the sulfur atoms of the cluster) in order to examine an additional recent proposal that these factors are of critical importance in protein control of E 0Ј (21,40,44,45).

EXPERIMENTAL PROCEDURES
Mutagenesis of fdxA, and Expression and Purification of FdI Variants-For mutagenesis T4 DNA ligase and T4 polynucleotide kinase were obtained from Life Technologies, Inc., while all restriction enzymes were from New England Biolabs (Beverly, MA). The in vitro mutagenesis was performed as described previously (43) using a Mu-taGene M13 in vitro mutagenesis kit from Bio-Rad and the following oligonucleotides with the altered base(s) indicated in bold. The sequences are 5Ј-CCGGACGAGTGCCAGGACTGCGCGCTC-3Ј for I40Q, 5Ј-CCGGACGAGTGCAACGACTGCGCGCTC-3Ј for I40N, and 5Ј-GTT-GAAGTCTGCGGCGTAGACTGTTTC-3Ј for P21G. For all the mutants at Ile 34 position, oligonucleotides with a mixed sequence were used to generate the mutants: 5Ј-GGGCCGAACTTCCTGGTC(CA)A(GC) CATCCGGACG-3Ј. The success of the mutagenesis was confirmed at the DNA level by dideoxy-DNA sequencing using the Sequenase version 2.0 DNA sequencing kit from Amersham Pharmacia Biotech. The overexpression of the FdI variants in their native background in A. vinelandii was carried out as described previously (46), except that the parent strain used for the overexpression was A. vinelandii LM100, a strain that does not synthesize native FdI, and electroporation (BTX TransPorator Plus electroporation system; BTX, Inc., San Diego, CA) was used instead of the triparental mating method in the transformation process.
Cell growth and the purification and triclinic crystallization of native FdI and FdI variants was carried out as described previously (43,47). As a precautionary measure, the FdI variants were initially purified anaerobically in the presence of dithionite. The anaerobically purified protein was then exposed to the air to test its air stability. Once it was established that they were air-stable, further experiments including fast protein liquid chromatography (MonoQ, with a linear gradient of 0.15-0.5 M NaCl in Tris-HCl, pH 8.0) and triclinic crystallization (43) were done aerobically.
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. Samples were first concentrated to the desired level using a Centricon-10 microconcentrator in 0.025 M Tris-HCl, pH 7.4. Reductions were carried out by addition of Na 2 S 2 O 4 to 2.0 mM and incubation until no further change in absorption at Ͼ350 nm could be observed (usually this required about 20 min). UV-visible spectra were obtained with a Hewlett-Packard 8452 diode array UVvisible spectrophotometer. CD spectra were recorded using a Jasco J720 spectropolarimeter. EPR spectra were obtained using a Bruker 300 Ez spectrometer, interfaced with an Oxford Instruments ESR-9002 liquid helium continuous flow cryostat.
Reduction Potential Calculations-The E 0Ј calculations were carried out using the program POLARIS, which was developed by Warshel et al. in the Department of Chemistry at the University of Southern California. This program is now commercially available from A. Warshel's group. In this study we used version 6.30 of this program on the spp2000 computer at the University of California, Irvine Office of Academic Computing. POLARIS was developed to calculate the free energies and electrostatic properties of molecules and macromolecules in solution using a Protein Dipole Langevin Dipole (PDLD) model. A detailed description of how the calculations are done for [Fe-S] proteins is found in Ref. 21 and will only be briefly considered here. The calculations begin with a protein structure in a Protein Data Base (PDB) file format. In this case the starting point for calculations was the new 1.35-Å structure of the oxidized state of native FdI (34) (7FDI). A number of PDB files were then created from that structure by modeling the I34N, I34Q, I40N, I40Q, or P21G mutations using the Insight II package (MSI, San Diego). The program replaces the wild-type residue with a mutant residue after local energy minimization. The actual I40N x-ray structures were also used as a starting point for calculations where indicated (accession code 1b0v). Once a structure had been obtained in PDB format, the next step was to convert that structure to a form suitable for PDLD calculations using a program called PREPARE that is included in the POLARIS package. This procedure involved deletion of all unwanted atoms (in this case the crystallographically observed ordered water molecules) and addition of H atoms if necessary. In the case of side chains that are capable of free rotation, the relative (Coulombic) energies of four orientations were evaluated and the configuration of minimum energy was selected (21). For His residues, the relative energies of the N ␦d -and N ⑀e -protonated forms were evaluated similarly, and the configuration of minimum energy was chosen. Then, with the exceptions indicated below, all atoms were assigned charges including the atoms of the [Fe-S] cluster of interest, in this case the [4Fe-4S]Cys 4 cluster in its oxidized and reduced states (21). The iron and inorganic sulfide atoms of the other [3Fe-4S]Cys 3 cluster were treated as uncharged and assigned as zero. Additionally, the ␤-CH 2 and ␣-CH moieties of all ligating Cys residues were treated as uncharged, while non-ligand Cys residues were treated as normal amino acids. All the ionizable residues were treated as uncharged (total charge of the residue is zero).
During the first part of the PDLD calculation, the Coulombic interactions of the [4Fe-4S]Cys 4 cluster, in its oxidized and reduced states, with all other protein atoms was calculated. This included chargecharge interactions and charge-induced dipole interactions. The next part of the calculations involved construction of a Langevin dipole grid representing water molecules around the protein, and the interactions between the grid dipoles and the cluster were calculated. In this case the grid filled a sphere of radius r L ϭ 25 Å and was composed of two sections, an inner section with 1-Å spacing in a 12-Å radius and an outer section with a 3-Å spacing. For the oxidized state of the cluster, the dipoles on the constructed grid were optimized by sampling a set of 30 grids to give the maximum energy, and then the optimized grid was used without reoptimization for the reduced cluster. The final step involved calculation of the interaction of the cluster charges with the bulk media more than 25 Å away from the cluster. In the actual calculation process, the PDLD calculation is not only carried out on the starting protein structure, but also on molecular dynamics generated structures from the program. This is to take advantage of the fact that the average results based on a series of structures generated by molecular dynamics are more accurate than the single result from the x-ray crystal structure (21). The molecular dynamics simulations were done using a program called ENZYMIX that is attached to the POLARIS program. The molecular dynamics simulations were done at 300 K, generating a snapshot structure every 500 fs. A total of 50 structures were generated for each protein and each of them was subjected to PDLD calculation. The final results were averaged (21).
Electrochemical Experiments-Purified water of resistivity ϳ18 M⍀ cm (Millipore, Bedford, MA) was used in all experiments. The buffers MES, HEPES, and TAPS and the co-adsorbate neomycin sulfate were purchased from Sigma. An AutoLab electrochemical analyzer (EcoChemie, Utrecht, The Netherlands) was used to record DC voltammograms. The three-electrode configuration featuring all glass-cells has been described previously (48). The sample compartment (typically holding 500 l) was maintained at 0°C to optimize stability. All E 0Ј values are given with reference to the standard hydrogen electrode. The saturated calomel electrode was held at 22°C which we have adopted as E (saturated calomel reference electrode) ϭ ϩ243 mV versus the standard hydrogen electrode. E 0Ј values from cyclic voltammetry were calculated as the average of the anodic and the cathodic peak potentials, E 0Ј ϭ 1 ⁄2(E pa ϩ E pc ). The pyrolytic graphite "edge" electrode (surface area typically 0.18 cm 2 ) was polished prior to each experiment with an aqueous alumina slurry (Buehler Micropolish; 1.0 m) and then it was sonicated extensively to remove traces of Al 2 O 3 . All experiments were carried out under anaerobic conditions in a Vacuum Atmospheres glove box with an inert atmosphere of N 2 (Ͻ1 ppm).
Prior For the investigation of pH dependence, protein solutions were dialyzed extensively against the buffered solution at the required pH, using an Amicon 8MC diafiltration unit equipped with a microvolume assembly and a YM-3 membrane.
Crystallization and Structure Determination of I40N FdI-Due to the altered solubility properties of the I40N, I40Q, and P21G variants of FdI with respect to the native protein, it was not possible to crystallize these mutants using the conditions for native FdI (49) or seeding procedures (50). A large number of trials were carried out in an anaerobic glove box to screen for new crystallization conditions. For I40N FdI, brown, diamond-shaped, thin plates, approximately 0.3 ϫ 0.3 ϫ 0.05 mm in size, were grown by vapor diffusion from a solution containing 3 l of 10 mg/ml protein in 0.4 M Tris-HCl buffer, pH 7.8, and 3 l of 4.2 M (NH 4 ) 2 SO 4 and 20 mM NaCl in 100 mM Tris-maleate buffer, pH 6.5 (reservoir solution). The crystallization droplets were equilibrated against 1.0 ml of the reservoir solution at room temperature, and crystals appeared after 4 -5 days. Similar screening as well as seeding experiments with the I40Q and P21G mutants of FdI were unsuccessful.
The thin, plate-like I40N FdI crystals exhibited highly mosaic diffraction in the direction parallel with the thin dimension of the crystal when frozen at 100 K using a variety of cryoprotectants. Therefore, data were collected from a single crystal of I40N FdI mounted aerobically in a thin-walled, glass capillary using the reservoir solution as a synthetic mother liquor. Data were collected at 18°C using CuK ␣ radiation from a Siemens SRA x-ray generator operated at 55 kV, 90 mA, and equipped with a graphite monochromator and a Mar Research 34.5-cm diameter image plate scanner. Data were recorded in 1°oscillations through a total rotation range of 162°with an exposure time of 10 min/frame. The data were indexed, integrated, merged, and scaled with Mosflm and Scala (51) ( Table I).
The structure of I40N FdI was solved by molecular replacement using the structure of native FdI, refined at 1.35-Å resolution (34), as a search model, and the program Amore (52). The rotation search yielded four solutions with correlation values 1.67-1.33 times the background. After arbitrarily placing the first molecule in the P1 unit cell, translation searches yielded clear solutions for the positions of the second, third and fourth copies of the protein. Rigid body refinement of this model resulted in an R-factor of 0.31 for all data to 3.0-Å resolution. The model was modified by substituting Tyr 13 3 Gly and Ile 40 3 Gly, and refined with Xplor (53), version 3.8, by simulated annealing with inclusion of bulk solvent and anisotropic scaling corrections, resulting in an R-factor of 0.21 for all data in the resolution range 20.0 -2.8 Å. Phases from this model, unbiased for the side chains of Tyr 13 and Asn 40 , were used to calculate a 2͉F o ͉ Ϫ ͉F c ͉ map. The map revealed clear density for the Tyr 13 side chain for all four copies of the protein, confirming the correctness of the molecular replacement solution. The Tyr 13 side chain was modeled into the electron density using Xfit (54). In this process, it was unnecessary to adjust the conformation of the main chain atoms, as expected, since Tyr 13 is adjacent to the [3Fe-4S] cluster in the structure and distal from the site of the mutation at residue 40.
The unbiased 2͉F o ͉ Ϫ ͉F c ͉ map was then used to model the conformation of the Asn 40 side chain in all four copies of the protein. Clear electron density at 1.5 was present for three of the copies; in the fourth, it was weaker but interpretable. In order to orient the C␣-C␤ vector of each Asn 40 residue into the electron density, it was also necessary to rotate the main chain ⌿ torsion angle by approximately 20°. This not only allowed the side chain to be fit for all four copies, but it also allowed the carbonyl of Asn 40 to better fit the density in each case. It was also necessary to adjust the torsion angle of Asp 41 in each of the four copies. In other words, the 2.8-Å resolution map clearly indicated that the main chain conformation was altered in the I40N mutant. At the same time the side chain of the adjacent residue, Asp 41 , was observed to change conformation about its 2 torsion angle in the four independent copies of the mutant protein. The position of the carboxamide moiety of Asn 40 ( 2 Ϯ 180°) could be inferred from hydrogen bonding interactions for all four copies. The complete model was refined against all data to 2.8 Å as before, and a 2͉F o ͉ Ϫ ͉F c ͉ map was used to check and adjust the fit of Asn 40 and Asp 41 to the density. No other significant changes were observed in the vicinity of the mutation at this resolution. Refinement statistics for the final model are summarized in Table I. Coordinates for the four independent structures of the I40N mutant of FdI have been deposited with Protein Data Bank with accession code 1b0v.

RESULTS AND DISCUSSION
Selection of Mutants-Factors that have been proposed to influence E 0Ј of [4Fe-4S] 2ϩ/ϩ proteins, that have not been examined experimentally, are the orientation of the side chain and backbone dipoles relative to the cluster and the solvent accessibility of the cluster. Warshel has developed a method for the quantitative modeling of these factors in proteins, which is known as the PDLD method (21,40,(55)(56)(57)(58). This method calculates the differences in free energies of the oxidized and reduced states of a redox-active prosthetic group, modeling the interaction of the group with protein and solvent. As considered in detail elsewhere, the results are expressed as the sum of four It is important to emphasize at the outset that this method can be used to predict the variation of E 0Ј for the same type of group in different proteins but cannot be used to predict absolute E 0Ј . The PDLD method has been used successfully to model the E 0Ј of cytochrome c (57,58), and most recently it has been applied to the study of [Fe-S] proteins (21,44,45). In that case, it was concluded that to a first approximation, the choice of redox couple made by a [4Fe-4S] cluster was largely influenced by the interaction of the cluster with the polar backbone amide groups and that modification of those amide groups is likely to lead to changes in E 0Ј . Those computational studies further predicted that once the [4Fe-4S] 2ϩ/ϩ couple was selected by the overall protein folding, further modification of the E 0Ј was unlikely to occur by changes in charged residues near the cluster (21,44,45). As discussed in the Introduction, this prediction is fully consistent with experimental evidence in the case of site-directed mutation of the [4Fe-4S] 2ϩ/ϩ cluster region of AvFdI (43). Changes in E 0Ј of a protein-bound [4Fe-4S] 2ϩ/ϩ cluster were predicted to be strongly influenced by the presence or absence of internal water molecules or by the interaction of the cluster with polar amino acid side chains (21,44,45).
In a recent review article, Stephens et al. analyzed nine site-directed variants of AvFdI using the PDLD method (21) and the results are summarized in Table II. X-ray structures are available for all of the proteins shown in Table II and these were used for the calculations. Overall, the results of this study were very encouraging with only minor differences between observed and PDLD calculated relative E 0Ј values. The average difference was 29 mV or 0.67 kcal. This led the authors of that review article to suggest to us additional mutants 3 that might be expected to alter the E 0Ј of the [4Fe-4S] 2ϩ/ϩ cluster of AvFdI.
One of those mutants 3 was designed to create a cavity in the protein that might accommodate a water molecule. We did not select that mutant, however, both because our experience suggested that cavity mutants were unstable and because more recent information suggests that internal waters are unlikely to be present in naturally occurring low potential ferredoxins. Thus, of the seven x-ray structures 4 and three NMR structures 5 that are now available for [4Fe-4S] 2ϩ/ϩ containing naturally occurring ferredoxins, including one at 0.94 Å (36), none have internal water molecules. Internal water molecules have also not been observed for any of the FdI variants whose structures have been previously reported, nor were they introduced by any of the calculations reported herein.
Three of the suggested mutants 3 involved replacing nonpolar Ile residues near the [4Fe-4S] 2ϩ/ϩ cluster of FdI with polar residues, either Gln or Asn (Fig. 1). One of those residues, Ile 54 , is a fully buried hydrophobic residue with contacts to five other residues and was not selected because of probable (based on our experience) stability and folding problems. The position of the other two Ile residues, Ile 40 and Ile 34 , relative to the [4Fe-4S] 2ϩ/ϩ cluster, is shown in Fig. 2. Four mutants were constructed placing glutamine and asparagine at each of those two positions to give I34N, I34Q, I40N, and I40Q. A final set of six suggested mutants 2 were designed to introduce potential changes in backbone amide orientation by replacing residues with Gly (Fig. 1). These residues were targeted because earlier studies had shown that the contribution of their amide groups to the ⌬ Q term exceeded 4.0 kcal/mol (44). Of these one, P21G, was selected because it was in a region of the protein that we had previously shown to tolerate several mutations without problems (43,50,59,60) and because the amide N of this residue was already pointing toward the cluster in native FdI. In addition, the homologous residue had already been successfully mutated in another protein C. pasteurianum Fd to yield stable products (61,62). The position of Pro 21 relative to the [4Fe-4S] 2ϩ/ϩ cluster of AvFdI is also shown in Fig. 2.
ϩ78 ϩ50 a All calculations were done with molecular dynamics except C20A as reported in Ref. 21 Procedures" and elsewhere (21), those structures (in PDB format) were then converted to a form suitable for PDLD calculations using the PREPARE program in the POLARIS package. PDLD calculations were then carried out on the 50 snapshot pictures generated by molecular dynamics, and the final results were averaged. Table III presents the results of these initial calculations.
For all five FdI variants, large positive changes in E 0Ј were predicted. The calculations in Table III suggest that the predicted changes in E 0Ј are not influenced by the interaction of the cluster with the bulk solvent (⌬ B ) and none of the calculations suggested that water molecules would actually enter the protein. In three cases, I34N, I34Q and I40N, the ⌬ L term indicated that the clusters would become somewhat more accessible to solvent, an event that should cause the E 0Ј to become more negative. For I40Q the ⌬ Q␣ term was very similar to that of native FdI, whereas for all other variants the change was again expected to result in a decrease in E 0Ј with I40N predicted to have the biggest decrease. The dominating factor in the calculations for all of the variants, which overall were predicted to have large increases in E 0Ј , is the ⌬ Q term, reflecting the importance of the orientation of positive end of the introduced side chain dipoles and/or the reorientation of the backbone amides relative to the cluster.
Purification and Characterization of the FdI Variants-Ile 34 , Ile 40 , and Pro 21 are all surface residues close to the [4Fe-4S] 2ϩ/ϩ cluster (Fig. 2). Prior to construction of the FdI variants, modeling studies indicated that all changes could be easily accommodated without interfering with other residues. All of our AvFdI variants are expressed in their native backgrounds in A. vinelandii (46). In previous studies we had modified other surface residues and had not experienced any problems with stability such that the mutant proteins accumulated to the same levels in vivo as native FdI (e.g. 31,43,63). In contrast, two of the mutants, I34N and I34Q, were present in such low levels that they could not be purified. The other three mutants did accumulate to levels similar to those found for the native protein.
As described under "Experimental Procedures," I40N, I40Q, and P21G FdI were successfully purified and their O 2 oxidized UV-visible absorption spectra were indistinguishable from that of native FdI (data not shown). Native FdI is an air-stable protein and the FdI variants also appeared to be air stable in their oxidized states. The addition of dithionite to I40N, I40Q, P21G, or native FdI results in partial bleaching of the spectrum (data not shown). 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 E 0Ј of the [4Fe-4S] 2ϩ cluster is too low to be reduced by dithionite at neutral pH (31,60). For I40N the result was indistinguishable from native FdI, however, for both I40Q and P21G the bleaching on reduction was more extensive than is seen for the native FdI. This result would be consistent with partial dithionite reduction of the [4Fe-4S] 2ϩ cluster caused by raising its E 0Ј in I40Q and P21G relative to native FdI, something that was predicted by the calculations shown in Table III.
To examine this issue further EPR experiments were carried out. In general, the presence of an oxidized [3Fe-4S] ϩ cluster in any protein is easily identified by the appearance of a characteristic g ϭ 2.01 EPR signal. The EPR spectra of oxidized I40N, I40Q, and P21G FdI were found to be both qualitatively and quantitatively indistinguishable from that of native FdI (data not shown). As is the case for native FdI, all three FdI variants were EPR-silent in the dithionite-reduced state in the perpendicular mode. This is consistent with the formation of [3Fe-4S] 0 but inconsistent with the reduction of [4Fe-4S] 2ϩ to the paramagnetic 1ϩ oxidation state, if that cluster is S ϭ 1 ⁄2. 6 Thus, the bleaching observed in the UV-visible experiment is unlikely to arise from reduction of the [4Fe-4S] 2ϩ cluster and is more likely due to a lowered stability of the reduced protein. This possible instability of the reduced protein was confirmed by re-oxidation of the proteins by exposure to air. For I40N and native FdI the bleaching of the UV-visible spectrum was completely reversible, whereas for I40Q and P21G it was not. The cause for the relative instability of the reduced protein is currently unknown. The wavelength dependence and form of the CD spectra of oxidized and reduced native FdI and I40N, I40Q, and P21G were extremely similar (data not shown). In previous studies of C20A and C20S mutants, where the [4Fe-4S] 2ϩ/ϩ cluster had undergone a ligand exchange and structural rearrangement, the CD spectra were dramatically different (50,59). Thus, the CD data, combined with the UV-visible and EPR data, strongly suggest that the overall protein folding and the environment around the clusters has not been significantly perturbed in any of the altered proteins.
Reduction Potential Measurements-As was the case for all of the proteins shown in Table II, the [4Fe-4S] 2ϩ/ϩ E 0Ј for I40N, I40Q, P21G, and native FdI were measured under identical conditions, at the same time, using direct electrochemical methods as described under "Experimental Procedures." Results obtained at pH 7.0, for ferredoxin solutions containing 4 mM neomycin, are shown in Fig. 3. Two pairs of well-defined oxidation and reduction peaks are observed in each case; currents vary proportionally with (scan rate) 1/2 up to 20 mV s Ϫ1 , as expected for a reaction involving protein molecules diffusing to a planar electrode surface. The [4Fe-4S] 2ϩ/ϩ cluster E 0Ј is known to be weakly pH-dependent (31,60), and the same pH dependence was observed for all of the variants examined herein. Table IV summarizes the data collected at pH 7.0 and compares the results to the ⌬E 0Ј values predicted from the PDLD calculations. For both I40Q and P21G, E 0Ј became more positive, as predicted by the calculations, but the magnitude of the change was substantially less than predicted. The big surprise in this study was that I40N whose [4Fe-4S] 2ϩ/ϩ cluster was predicted to have an E 0Ј 168 mV more positive than native FdI actually had an E 0Ј 33 mV more negative.
Attempts to Bring the I40N Calculations Closer to the Observation-The E 0Ј information shown in Tables III and IV was available prior to the solution of the x-ray structure of I40N FdI. Therefore, the starting point for the initial calculations was a model for I40N based on the native structure. As described above, in this procedure the PDB file for the 1.35-Å resolution structure of oxidized FdI was modified using the Insight II program by substitution of Ile by Asn. There are six allowable orientations for the Asn side chain generated by the program, and the program uses energy minimization to select the lowest energy orientation. This structure is then used as the starting point, and, as described under "Experimental Procedures" and elsewhere (21), it is then further modified, molecular dynamics is performed and finally the ⌬E 0Ј calculations are completed. Because the critical factor in control of E 0Ј in this case was expected to be the orientation of the positive ends of the side chain and backbone dipoles of the introduced Asn, we decided to fix the residue in different orientations at the outset rather than use the orientation selected by local energy minimization from the program. 7 Fig. 4A shows the two extreme orientations. Orientation one has the side chain pointing toward the cluster and is the one chosen by energy minimization, while orientation six is the other extreme with the side chain pointing away from the cluster. As shown in Table V, the change in E 0Ј predicted for orientation one was 168 mV more positive than that of native FdI whereas the prediction for orientation six was only 19 mV more positive, bringing the calculations in much closer agreement with the observation shown in Table IV. As shown in Table V, the large difference between the calculations for the two orientations arose from the ⌬ Q terms, with the model based on orientation six being in much closer agreement with the native protein.
X-ray Structure of I40N FdI-Because of the extreme difference between the theory and the experimental results, the I40N mutant was chosen as a target for crystallization. The Ile 40 3 Asn mutant of FdI proved difficult to crystallize, and once crystals were obtained they could not be frozen at 100 K without introducing severe mosaicity into the diffraction pattern. Consequently, data collection was limited to ambient temperature, resulting in a 2.8-Å resolution data set (Table I). Nevertheless, this resolution proved sufficient to define the orientation of the Asn 40 side chain in all four independent copies of I40N FdI in the asymmetric unit (Fig. 5). These data revealed that four different orientations were chosen for the Asn 40 side chain and Fig. 4 compares those orientations to those chosen in the models. As shown in Fig. 4, the actual structures did not correspond to either orientation one or six, both of which represent extreme cases. The actual orientations of the dipoles were all somewhere in the middle. Additionally, both our modeling and the PDLD program failed to predict the change in the backbone conformation of the protein as it was revealed by the x-ray structure.
In addition to the change in the orientation of the side chain at position 40, the unbiased electron density map revealed significant alteration in the conformation of the main chain at residues 40 -41, as well as change in the conformation of the Asp 41 side chain (see "Experimental Procedures"). A superposition of the four independent structures of I40N FdI onto native FdI is shown in Fig. 6. Overall, the structures are very similar with root-mean-square differences of only 0.21-0.25 Å for all 527 N, C␣, C, O, and C␤ atoms in the protein following least squares fit of each copy of I40N FdI (molecules A, B, C, and D) onto native FdI. However, displacements of 0.22-0.81 and 0.13-0.59 Å occur at the C␣ atoms of Asn 40 and Asp 41 , respectively, while the side chains of these residues adopt a unique Ϫ577 ϩ42 ϩ86 conformation in each independent copy of the protein (Fig. 6).
In the P1 triclinic unit cell, the four molecules of I40N FdI pack as two asymmetrically related dimers (A-B and C-D); each dimer involves a Trp 94 -Trp 94 stacking interaction between pseudo-twofold related monomers. Consequently, each of the four copies of I40N FdI occupies a unique environment in the crystal lattice, accommodating the four unique conformations of Asn 40 -Asp 41 . An explanation for this crystal packing arrangement is that the Ile 3 Asn replacement destabilizes the protein, so that the Asn 40 -Asp 41 peptide cannot adopt a single, low energy conformation. Therefore, a lower symmetry unit cell with four copies of the protein in the asymmetric unit is favored, in contrast to native FdI, which crystallizes readily in a high symmetry space group with one molecule in the asymmetric unit (45). This would account for the difficulty in obtaining crystals of the I40N mutant.
In native FdI Ile 40 occurs in the [4Fe-4S] 2ϩ/ϩ cluster binding loop of residues Cys 39 -Ile-Asp-Cys-Ala-Leu-Cys 45 where Ile 40 lies on the protein surface and shields the cluster from solvent. The Ile 40 side chain is also in contact with Pro 21 and Val 22 . The amides of residues 40 and 41 provide NH . . . S hydrogen bonds to S4 of the cluster and S␥ of Cys 39 , respectively (34). Together, the four independent Asn 40 side chains display a range of positions and conformations for their carboxamide groups (Fig.  6). The conformations are such that they roughly occupy the space occupied by the C␥1 and C␦ atoms of Ile 40 (the -CH 2 CH 3 moiety). However, they also fall into two groups. In molecules A and D, the C␣-C␤ bond direction for Asn 40 is similar to that for Ile 40 in native FdI, but in molecules B and C it is quite differ-ent, so that the C␤ atoms are displaced ϳ1.8 Å from C␤ of Ile 40 (Fig. 6). Nevertheless, in these copies the 1 torsion angles place the carboxamide groups close to those of the Asn 40 side chains in molecules A and D. In none of the Asn 40 conformations is a new NH . . . S or any other kind of hydrogen bond formed. In contrast, the side chain exhibits relatively unfavorable contacts with either Pro 21 or Val 22 . For example, in molecule A the N␦2 atom is 3.8 Å from Val 22 , in molecule B the O␦1 atom is 3.0 Å from Pro 21 , in molecule C the O␦1 atom is 3.5 Å from Val 22 , and in molecule D the N␦2 atom is 3.6 Å from Val 22 . Apparently, the range of conformations observed for Asn 40 reflects relatively unfavorable interactions with either of two nearby hydrophobic residues, and a lack of any other favorable new interaction. Although Asp 41 displays variation in the position and conformation of the carboxyl group in the four independent structures, all four conformations allow two hydrogen bonds to the N terminus at Ala 1 as observed in native FdI.
PDLD Calculations Based on the I40N Structures-Once the I40N structures were available, all four structures were used as a starting point for the PDLD calculations and the data are summarized in Table V. The average calculated ⌬E 0Ј for the four orientations is Ϫ1.8 mV, within 31 mV or 0.7 kcal of the observed ⌬E 0Ј , with structures A and D being in very close agreement. In both of these cases, the calculations predicted a somewhat increased exposure to solvent as was suggested by the structures and also reflected the importance of the changes in the side chain and backbone amides that were observed. It should be noted that we do not know whether or not all four orientations are present in solution or if one is preferred. The electrochemical measurements for this mutant have an error of ϳ Ϯ10 mV (for a given conformation), so if more than one structure is present in solution they all must have E 0Ј values that are very close together.
Conclusions-The [4Fe-4S] 2ϩ/ϩ cluster of native AvFdI has an unusually low E 0Ј (31,60). In previous studies we have shown that this E 0Ј is insensitive to changes that occur at the [3Fe-4S] ϩ/0 site of the protein, which is 6.7 Å away at its closest point of contact (e.g. Refs. 31 and 63). We have also shown that the E 0Ј is insensitive to modification of surface charged residues near the [4Fe-4S] 2ϩ/ϩ cluster (43). The reduction potential did become substantially more negative in two variants C20A FdI and C20S FdI that underwent [4Fe-4S] 2ϩ/ϩ ligand exchange, with a major change in ligand torsion angle, and an accompanying structural rearrangement (Table II) (50,59). In this study we have extended the range of E 0Ј without such a structural rearrangement. I40Q and P21G exhibit E 0Ј 53 and 42 mV more positive than native FdI, respectively, while I40N has E 0Ј more negative by 33 mV. To date, the 12 FdI variants that have been constructed in the region of the [4Fe-4S] 2ϩ/ϩ cluster have a 153-mV range in E 0Ј (Tables II and IV). It should be noted that none of these mutations affected the reduction potential of the [3Fe-4S] ϩ/0 cluster eliminating the other cluster as an important factor in control of E 0Ј .
In a recent review article (21), Stephens et al. examined nine FdI variants whose structures and E 0Ј were known (Table II) and found that their observed [4Fe-4S] 2ϩ/ϩ E 0Ј values were in close agreement with ⌬E 0Ј calculated using the PDLD methodology. This encouraged the approach of trying to predict the [4Fe-4S] 2ϩ/ϩ E 0Ј of FdI variants prior to their construction.
The greatest agreement between initial calculation and observation in this study was in the analysis of P21G FdI. This variant was purified and characterized spectroscopically, and appears to be structurally homologous to the native protein. P21G FdI was designed to alter the orientation of the backbone amide relative to the [4Fe-4S] 2ϩ/ϩ cluster. It was predicted to have a ⌬E 0Ј 86 mV more positive than native FdI, whereas the  Table V. observed ⌬E 0Ј was 42 mV more positive. Thus, the direction of the change was accurately predicted while the magnitude was overestimated, in this case by 44 mV or 1 kcal. As shown in Table II for eight of the nine FdI variants examined previously, the calculated changes in E 0Ј were also larger than the observations on average by 29 mV or 0.67 kcal. A previous study of C. pasteurianum Fd showed that the substitution of the analogous Pro by Lys, Asn, Met, or Thr led to much smaller changes in E 0Ј (Յ13 mV) (61,62). The PDLD calculations suggest that adding one NH . . . S interaction by substitution of Pro by Gly should increase E 0Ј by ϩ86 mV. However, when we take into account the fact that the PDLD calculations appear to consistently overestimate ⌬E 0Ј (see above), that value is much closer to the observed ⌬E 0Ј of ϩ42 mV. This in turn suggests that there has been essentially no conformational change in the backbone for P21G and that the amide dipole should be oriented as it is in the model.
I40Q was designed to change the [4Fe-4S] 2ϩ/ϩ cluster E 0Ј by replacing a nonpolar residue with a polar residue near the cluster. The observed ⌬E 0Ј was 53 mV, whereas the predicted ⌬E 0Ј was 116 mV. In this case, the direction of the change was again accurately predicted and the magnitude was again an overestimate. Nonetheless, the magnitude of the change observed for this mutation is greater than has been observed for alteration of surface charged residues in any low potential ferredoxin (43,61,62), thus providing validation for the conclusion that [4Fe-4S] 2ϩ/ϩ reduction potential can be controlled by the orientation of polar residues on the surface of the protein near the cluster. This in turn brings up the interesting possibility that conformational changes induced upon docking of electron transfer partners could cause reorientation of a surface polar residue in such a way as to alter the E 0Ј of one of the partners to facilitate electron transfer.
I40N was also designed to influence the E 0Ј of the [4Fe-4S] 2ϩ/ϩ cluster by introducing a polar residue in the vicinity of the cluster. In this case, a very large positive ⌬E 0Ј was initially predicted (ϩ168 mV) and a negative ⌬E 0Ј of Ϫ33 mV was observed. Further calculations showed that, even with molecular dynamics, the starting orientation of the polar residue had a very large effect on the outcome of the PDLD calculations. Thus, in cases where mutant x-ray structures are being modeled based on native structures, it appears to be important to test every allowable orientation rather than rely on the lowest energy orientation prior to beginning the ⌬E 0Ј calculations.
Although it was possible to bring the I40N calculations closer to the observation by modeling different orientations of the side chain, the actual structure of I40N FdI showed three features that could affect the E 0Ј of its [4Fe-4S] 2ϩ/ϩ cluster in addition to the orientation of its side chain. First, the conformation of the adjacent charged residue, Asp 41 , is affected (Fig. 6). Previous   studies have shown that the presence versus absence of adjacent charged groups on the FdI protein surface had very little affect on E 0Ј (43), so it is unlikely that this change causes the observed Ϫ33 mV change. Second, the conformational changes resulting from the mutation include significant changes in the main chain at residues 40 and 41, and this in turn affects two favorable NH . . . S interactions in native FdI. In particular, the residue 40 amide to S4 distance changes from 3.46 Å in native FdI to 3.4, 3.4, 3.0, and 3.5 Å in molecules A, B, C, and D, respectively. In addition, the linearity of the interaction is reduced in molecules B and D. At the same time, the residue 41 amide to Cys 39 S␥ distance changes from 3.4 Å to 3.7, 3.4, 3.9, and 3.3 Å in molecules A, B, C, and D, respectively, while the linearity of the interaction is significantly reduced for molecule C. Together, these changes indicate both more and less favorable interactions, but are overall relatively small. Consequently, their effects upon the E 0Ј may cancel out. Third, the solvent exposure of the cluster is increased in the I40N mutant. Although the Asn 40 side chain is positioned, on average, as the Ile 40 side chain (Fig. 2), it is considerably less bulky (C 2 N 1 O 1 H 4 versus C 4 H 9 ). Solvent accessibility to the cluster will also be enhanced by the fact that the Asn 40 is polar and would interact more favorably with H 2 O at the protein surface. Further, the apparent conformational flexibility of residues 40 and 41 in the mutant may facilitate access of H 2 O molecules to the cluster. These increases in solvent interactions were predicted by the original calculations shown in Table III and were expected to lower E 0Ј , as was observed.
Taken together the data presented above show that when an x-ray structure is available, as in the case of I40N and previously described variants (21), the current version of the PDLD calculations program POLARIS is able to quite accurately determine the difference in E 0Ј for site-directed variants of FdI. Further, the individual components of the calculations give insight into the causes for the changes in E 0Ј . The failure in this case was in the ability of InsightII to accurately model in advance a structure for a FdI variant, even though only a single surface residue was being modified. This difficulty arises when the amino acid replacement, in this case Ile 3 Asn, destabilizes the protein, as evidenced by the lack of a single, discrete conformation for I40N FdI. Consequently, the main chain angles of the protein are affected and this has a major effect on the position of the introduced side chain. This may be a general problem with small [Fe-S] proteins whose structures are dominated by the clusters they contain, and points out that in these cases data interpretation is likely to be critically dependent upon determining actual x-ray structures of protein variants.