Elimination of All Charged Residues in the Vicinity of the Active-site Helix of the Disulfide Oxidoreductase DsbA

Disulfide oxidoreductases are structurally related proteins that share the thioredoxin fold and a catalytic disulfide bond that is located at the N terminus of an α-helix. The different redox potentials of these enzymes varying from −270 mV for thioredoxin to −125 mV for DsbA have been attributed to the lowered pK a values of their nucleophilic, active-site cysteines and the difference in thermodynamic stability between their oxidized and reduced forms (ΔΔG ox/red). The lowered pK a of the nucleophilic cysteine thiols was proposed to result from favorable interactions with the helix dipole and charged residues in their vicinity. In this study, we have eliminated all charged residues in the neighborhood of the active-site disulfide of DsbA from Escherichia coli to analyze their contribution to the physicochemical properties of the protein. We show that the conserved charge network among residues Glu24, Glu37, and Lys58 stabilizes the oxidized form of DsbA and thus does not cause the high redox potential of the enzyme. The pK a values of the nucleophilic cysteine (Cys30) and the redox potentials of the DsbA variants E24Q, E37Q, K58M, E24Q/K58M, E37Q/K58M, E24Q/E37Q, E24Q/E37Q/K58M, and E24Q/E37Q/E38Q/K58M are similar to those of DsbA wild type. The redox potentials of the variants neither correlate with the Cys30pK a values nor with the ΔΔG ox/red values, demonstrating that the relationship between these parameters is far more complex than previously thought.

The formation of disulfide bridges is essential for folding and stability of many secretory proteins. Numerous in vitro studies have shown that disulfide bond formation is usually slow and rate-limiting for folding (1)(2)(3)(4)(5), whereas proteins lacking disulfides and cis X-proline peptide bonds may even fold within milliseconds (6,7). In vivo, disulfide bond formation is catalyzed by disulfide oxidoreductases. All members of this class of proteins possess an active-site disulfide bridge and the thioredoxin fold (8,9). The catalytic disulfide with the consensus sequence CXXC (where X represents any amino acid) is located at the amino terminus of an ␣-helix. In eukaryotic proteindisulfide isomerase (PDI) 1 and in the bacterial enzymes DsbA, DsbC, and thioredoxin, only the more N-terminal of the two active-site cysteines is solvent-exposed and acts as nucleophile in disulfide exchange reactions (10 -14).
The disulfide oxidoreductase DsbA is mainly responsible for the efficient formation of disulfide bonds in proteins that are exported to the periplasm of Escherichia coli (15,16). The three-dimensional x-ray structure of oxidized DsbA was determined to 2.0-Å resolution and revealed that the enzyme consists of two domains, the thioredoxin domain and an ␣-helical domain of unknown function ( Fig. 1; Ref. 17). However, the structure of oxidized DsbA provides no hints concerning its unusual physicochemical properties. These are reflected by the extremely low pK a of the nucleophilic active-site cysteine, Cys 30 , which has a value of about 3.5 (18 -20). The low pK a of Cys 30 qualitatively explains that reduced DsbA is more stable than oxidized DsbA (21,22), that it is the strongest oxidant in the family of disulfide oxidoreductases (21,23), and that it undergoes extremely fast reactions with thiol substrates (21,24). Although the residues of the XX dipeptide have been shown to influence the pK a of DsbA Cys 30 thiol (19), it is still unclear whether charged residues in the vicinity of the activesite helix or the helix dipole alone are responsible for the low pK a of Cys 30 in reduced wild type DsbA. In the case of thioredoxin, values between 6.7 and ϳ9 have been published for the pK a of the nucleophilic cysteine Cys 32 (10,(25)(26)(27)(28). Besides charged residues in the vicinity of the active-site helix, other explanations have been proposed for a lowered pK a of the nucleophilic thiol in human and E. coli thioredoxin such as the dipole of the ␣-helix (29 -31), hydrogen bonding between the Cys 32 thiolate and the amide of Cys 35 (30,32,33), and a shared proton between the two sulfurs of Cys 32 and Cys 35 (26,34).
Thioredoxin, PDI, and DsbA homologues possess a conserved amino acid pair (Asp 26 /Lys 57 in E. coli thioredoxin, Glu 47 /Lys 80 and Glu 391 /Lys 424 in human PDI, and Glu 24 /Lys 58 in E. coli DsbA (Fig. 2)) that is likely to form a buried salt bridge in the vicinity of the active-site disulfide (17,(35)(36)(37)(38). The hydrophobic environment of the salt bridges gave rise to speculation that they may directly or indirectly contribute to the lowered pK a values of the nucleophilic cysteines by electrostatic interactions (38). In addition, elevated pK a values of 7.5 and Ն9 were determined for Asp 26 of both oxidized and reduced thioredoxin in the presence or absence of the salt bridge, respectively (35, 36, 39 -42), which was assumed to be directly linked to the redox properties of thioredoxin. Fig. 2 shows the conserved charged residues near the redox-active center of proteins with the thioredoxin fold. Despite a low overall sequence homology, the charges of the side chains corresponding to Glu 24 and Lys 58 of E. coli DsbA are invariant in the known DsbA sequences and are also conserved in thioredoxins (with very few exceptions, e.g. human thioredoxin, which lacks a corresponding lysine) and PDIs.
To study the influence of the Glu 24 /Lys 58 salt bridge on the physicochemical properties of DsbA from E. coli, we constructed the DsbA variants E24Q, K58M, and E24Q/K58M. In contrast to thioredoxin, a second acidic residue, Glu 37 , is located close to the Lys 58 side chain in the structure of E. coli DsbA (Fig. 1). An acidic residue corresponding to Glu 37 of E. coli DsbA is present in 6 of the 10 known DsbA sequences (Fig.  2). In the x-ray structure of oxidized E. coli DsbA, Glu 37 is even closer to the Lys 58 side chain than Glu 24 (the nearest carboxylate oxygen of Glu 37 is 3.4 Å apart from the ⑀-amino group of Lys 58 compared with 4.0 Å in the case of Glu 24 ). We recently characterized the physicochemical properties of the DsbA variant E37Q, which are similar to those of DsbA wild type (20). To further analyze the function of Glu 37 in the context of the neighboring residues Glu 24 and Lys 58 , we now investigated the properties of the double variant E37Q/K58M, the triple variant E24Q/E37Q/K58M, and the quadruple variant E24Q/E37Q/ E38Q/K58M, which lacks all negatively charged residues less than 13 Å apart from the sulfur of Cys 30 .
Plasmid Construction and Mutagenesis-The plasmid pDSBA2 was derived from plasmid pTHZ11 by deletion of the rop gene after digestion with NspI and religation. pDSBA2 contains the dsbA wild type gene under the control of the trc promotor, the f1 and colE1 origins, the gene coding for ampicillin resistance, and the lacI q gene. Site-directed mutagenesis was performed with single-stranded, uridylated DNA from pDSBA2 according to Kunkel et al. (43) using the Muta-Gene phagemid kit from Bio-Rad (Glattbrugg, Switzerland). The following oligonucleotides were used for mutagenesis (S corresponds to C or G): E24Q, Mutations were identified by restriction analysis and verified by dideoxy sequencing using the Thermo Sequenase sequencing kit from Amersham (Zü rich, Switzerland).
Overproduction and Purification of DsbA Variants-Cells of E. coli THZ2 (dsbA::kan, recA::cam, malF-lacZ102) 2 were transformed with pDSBA2 harboring the respective dsbA mutations. 4.5-liter cultures were grown in 2 ϫ YT medium supplemented with 100 mg/liter ampicillin and 50 mg/liter kanamycin at 30°C. Production of DsbA was induced with 1 mM IPTG at an optical density at 546 nm (A 546 ) of 1.0, and the cells were grown for another 6 h (final A 546 ϭ 3-3.5). The cells were harvested by centrifugation and suspended in 150 ml of buffer (10 mM MOPS/NaOH, pH 7.0, 5 mM EDTA, 150 mM NaCl, 1 mg/ml polymyxin B sulfate). After gentle stirring on ice for 2 h and centrifugation (35,000 ϫ g for 30 min), the supernatant (periplasmic extract) was dialyzed against 10 mM MOPS/NaOH, pH 7.0, or, in case of the variants E24Q/E37Q and E24Q/E37Q/E38Q/K58M, against 10 mM Tris/ HCl, pH 8.6. The extracts were applied to a DE52 cellulose column (40 ml). Proteins were eluted with a linear gradient (500 ml) from 0 to 250 mM NaCl in the same buffers. Fractions containing DsbA were pooled, dialyzed against 10 mM acetic acid/NaOH, pH 4.5, and applied to a CM52 cellulose column (40 ml). Proteins were eluted with a linear gradient (500 ml) of 0 -200 mM NaCl or 0 -500 mM NaCl (in case of the variants E24Q/E37Q and E24Q/E37Q/E38Q/K58M). Fractions containing DsbA were pooled, dialyzed against distilled water, and stored at Ϫ20°C. The yield of pure DsbA was 10 -15 mg/liter⅐A 546 . The molecular mass of each DsbA variant was confirmed by electrospray mass spectrometry (accuracy Ϯ2). The amino acid replacements E24Q, E37Q, and E38Q were also confirmed by N-terminal sequencing of residues 1-40. All DsbA variants were completely oxidized after purification, since no free thiols were detectable by Ellman's assay (44). Protein concentrations were determined by absorbance (A 280, 1 mg/ml, 1 cm ϭ 1.05).
Spectroscopic Techniques-Spectroscopic measurements were performed with a Hitachi F-4500 fluorescence spectrophotometer, a Varian Cary3E spectrophotometer, and a Jasco J-710 spectropolarimeter. A temperature of 25°C was applied for all measurements, and buffer solutions were filtered before use (0.2-m pore size). Far-UV and near-UV circular dichroism (CD) spectra of oxidized and reduced DsbA were recorded as described elsewhere (20,23).
Unfolding/Refolding Equilibria-Unfolding/refolding experiments were essentially performed as described by Hennecke et al. (20). The exact GdmCl concentrations were determined from the refractive index of the solutions according to Nozaki (46). The original data of the transitions were analyzed according to the two-state model of folding using the six-parameter fit according to Santoro and Bolen (47). ionization of the Cys 30 thiol was followed by the specific absorbance of the thiolate anion at 240 nm (18,48,49). As a control, the pH-dependent absorbance of the oxidized form of each DsbA variant was recorded. To avoid precipitation artifacts and to minimize buffer absorbance, a buffer system consisting of 10 mM K 2 HPO 4 , 10 mM boric acid, 10 mM sodium succinate, 1 mM EDTA, and 200 mM KCl (containing 100 M ␤-mercaptoethanol for the reduced proteins) was used. The pH (initial value, 8.5) was lowered to 2.2 by stepwise addition of aliquots of 0.1 M HCl, and the absorbance at 240 and 280 nm was recorded and corrected for the volume increase. Samples had an average initial protein concentration of 25 M. The pH dependence of the thiolate-specific absorbance signal (S ϭ (A 240 /A 280 ) reduced /(A 240 /A 280 ) oxidized ) was fitted according to the Henderson-Hasselbalch equation, in which S AH is the corrected signal of the fully protonated and S A Ϫ is that of the fully deprotonated form. Evaluation of the pK a of Cys 30 of DsbA yielded reliable results between pH 7.8 and 2.2, and further acidification resulted in denaturation of DsbA. Deviations above pH 7.8 may have resulted from increasing deprotonation of tyrosine residues. Investigation of Polypeptide Specificity-Reduced, unfolded RNase A and hirudin (A 280, 1 mg/ml, 1 cm ϭ 0.56 and 0.37, respectively) were prepared as described previously (24). The quantitative reduction of the proteins was verified by Ellman's assay (44). Oxidation of RNase A and hirudin by DsbA was analyzed fluorometrically at 25°C with an SX-17MV stopped-flow reaction analyzer (Applied Photophysics) using an excitation wavelength of 295 nm and emission wavelengths Ն320 nm. Pseudo-first-order conditions were applied using 1 M oxidized DsbA and 200 M free thiols of the respective substrate protein or DTT. Buffer conditions were 100 mM formic acid/NaOH, pH 3.0, 100 mM formic acid/NaOH, pH 4.0, 100 mM acetic acid/NaOH pH 5.0, 100 mM MES/ NaOH, pH 6.0, 100 mM sodium phosphate, pH 7.0, 100 mM Tris/HCl, pH 8.0, 100 mM Bicine/NaOH, pH 9.0, and 100 mM glycine/HCl, pH 10.0. All buffers contained 1 mM EDTA and were degassed and flushed with nitrogen before use. Polypeptide specificity is defined as the ratio between the apparent second-order rate constant of the oxidation of the respective substrate protein and the apparent second-order rate constant of the oxidation of DTT.

Construction and
Purification of DsbA Variants-We constructed the plasmid pDSBA2, which contains the dsbA wild type gene under control of the IPTG-inducible trc promotor, the lacI q gene, and the phage f1 origin (see "Experimental Procedures"). Thus, the plasmid combines features necessary for in vivo complementation, efficient overexpression of dsbA, and production of single-stranded DNA for site-directed mutagen-esis (43). The following DsbA variants were generated: E24Q, K58M, E24Q/K58M, E24Q/E37Q, E37Q/K58M, E24Q/E37Q/ K58M, and E24Q/E37Q/E38Q/K58M. All variants could be overproduced in the dsbA-deficient E. coli strain THZ2 (see below) to approximately the same extent as the wild type protein. They were purified to homogeneity by conventional chromatographic techniques in the absence of reducing agents and were all obtained in the oxidized form with yields of about 50 mg/liter bacterial culture.
Complementation of the dsbA-deficient E. coli Strain THZ2-We used the E. coli strain THZ2 to characterize the in vivo function of the DsbA variants. This strain has a Tn10::kan insertion in the chromosomal dsbA gene and possesses a gene coding for a MalF-␤-galactosidase fusion protein (15,19,50). The following criteria were used to determine the ability of the DsbA variants to complement the dsbA Ϫ phenotype of THZ2 after transformation with the respective dsbA expression plasmids: (i) restoration of motility on agar plates containing 0.3% agar, since THZ2 is immotile due to the lack of disulfide bonds in the P-ring protein of the flagellar motor (45), and (ii) generation of a Lac Ϫ phenotype due to oxidative inactivation of periplasmic ␤-galactosidase by functional DsbA (15,19), which results in white colonies on agar plates containing X-gal. Even in the absence of IPTG, all DsbA variants yielded a DsbA ϩ phenotype of transformed THZ2 cells, which was indistinguishable from the phenotype of cells harboring the expression plasmid for wild type DsbA.
Spectroscopic Properties of the DsbA Variants-Like for DsbA wild type (23), a 3-fold increase in tryptophan fluorescence was observed for the variants E24Q, K58M, E24Q/K58M, and E37Q/K58M when the active-site disulfide bond was reduced. However, all variants in which two or three negative charges had been exchanged (E24Q/E37Q, E24Q/E37Q/K58M, and E24Q/E37Q/E38Q/K58M) exhibited an approximately 2-fold higher fluorescence of the oxidized form compared with oxidized wild type. Therefore, their fluorescence increased only about 1.5-fold upon reduction. This difference was still sufficient to determine their thermodynamic stabilities and redox equilibria with glutathione by fluorescence spectroscopy (see below). The emission maxima of all oxidized and reduced variants were identical to those of wild type DsbA (data not shown). Far-UV and near-UV CD spectra were recorded to detect possible conformational changes in the DsbA variants. The spectra of the oxidized and reduced variants were identical to those of oxidized and reduced wild type (data not shown). Thus, the native tertiary structure of oxidized and reduced DsbA was essentially not affected in any of the variants.
Thermodynamic Stabilities of the Oxidized and Reduced DsbA Variants-The free energies of stabilization of the oxidized and reduced forms of the variants were measured fluorometrically by GdmCl-induced equilibrium transitions assuming a two-state model of folding ( Fig. 3; Table I). All transitions were fully reversible, with cooperativities of 20 -27 kJ/mol⅐M for the oxidized variants and 22-27 kJ/mol⅐M for the reduced variants. Only the reduced variant E24Q/E37Q/E38Q/K58M exhibited a strongly reduced cooperativity of 15 kJ/mol and thus may not fold according to the two-state model (51).
A unique property of the DsbA wild type protein is its destabilized oxidized form (21,22). In all variants investigated, the oxidized form was also less stable than the reduced form. Removal of the negative charge of residue Glu 24 in the variant E24Q slightly stabilized both the reduced and oxidized forms about 2 kJ/mol, whereas the exchanges E37Q and K58M strongly destabilized the oxidized forms of the variants 8 -9 kJ/mol and stabilized the reduced forms about 2 kJ/mol. The stabilities of the oxidized and reduced double variants E24Q/K58M, E37Q/K58M, and E24Q/E37Q revealed that the stability changes caused by the single variants were not additive. Analysis of the stabilities according to the concept of double mutant cycles (Ref. 52; Table II) shows that each of the interactions Glu 24 -Lys 58 , Glu 37 -Lys 58 , and Glu 24 -Glu 37 energetically stabilizes the oxidized form of DsbA wild type and should principally counteract its oxidative force. Qualitatively, similar contributions of the interactions Glu 24 -Lys 58 , Glu 37 -Lys 58 , and Glu 24 -Glu 37 to the stability of oxidized and reduced DsbA are observed on the background of the E37Q, E24Q, and K58M variants, respectively. The stabilities of the oxidized and reduced triple variant E24Q/E37Q/K58M also show the whole interaction network between the residues Glu 24 , Glu 37 , and Lys 58 , which stabilizes the oxidized form of the wild type.
Overall, the measured stability differences between the oxidized and reduced variants (⌬⌬G ox/red ) predicted more oxidiz-ing proteins in the case of the variants K58M, E37Q, E37Q/ K58M, and E24Q/E37Q/K58M and more reducing proteins in the case of the variants E24Q/K58M and E24Q/E37Q (Table I).
Comparison of Measured and Predicted Redox Potentials of the DsbA Variants-The redox potentials (E 0 Ј ) of the DsbA variants were measured at pH 7.0 and 25°C by determining their equilibrium constants (K eq ) with glutathione using a value of Ϫ240 mV for the standard potential of the GSH/GSSG redox couple (53). The equilibrium constants were measured fluorometrically, assuming no significant equilibrium concentrations of DsbA-glutathione mixed disulfides (23), and yielded reliable data without systematic deviations from theory ( Fig. 4; Table III). Surprisingly, the redox potential of none of the variants differed more than 10 mV from that of the DsbA wild type, although the stability differences between their oxidized and reduced forms (⌬⌬G ox/red ) predicted that the redox potentials should be 26 -77 mV more oxidizing in the case of the variants K58M, E37Q, E37Q/K58M, and E24/E37Q/K58M and 10 -14 mV more reducing in the case of E24/K58M and E24Q/ E37Q (Table III). A qualitative agreement between the measured and the predicted changes in redox potential was only obtained for the variants K58M and E24Q/K58M. The lack of correlation between the measured values of ⌬⌬G ox/red and K eq is best shown by the variant E37Q/K58M, which is 8 mV more reducing than DsbA wild type and predicted to be 77 mV more oxidizing.
Ionization of the Cysteine 30 Thiol-To test whether the measured differences between the redox potentials of the variants and DsbA wild type were caused by changes in the pK a value of the nucleophilic Cys 30 thiols, we determined the pH dependence of the fraction of the Cys 30 thiolate in all reduced variants by their specific absorbance at 240 nm, using oxidized DsbA as a reference (Fig. 5, Table III) (18 -20, 48, 49). A pK a of 3.55 was measured for wild type DsbA, which is identical within experimental error with previously published data (18 -20). The Cys 30 pK a values of the variants varied between 3.19 (K58M) and 4.46 (E24Q/K58M), which again demonstrated that direct or indirect electrostatic interactions between the Cys 30 thiolate and the charged residues in the vicinity of the active-site helix do not significantly affect the reactivity of the nucleophilic cysteine. Due to the general dependence of disulfide exchange equilibria on the pK a values of the involved thiols (54) and assuming that the amino acid replacements only af-  Table III. FIG. 3. GdmCl-dependent folding/unfolding equilibria of oxidized and reduced wild type DsbA and the variant E24Q at pH 7.0 and 25°C. Open symbols (E, Ⅺ, छ, É) correspond to unfolding experiments, and closed symbols (q, f, ࡗ, ç) correspond to refolding experiments. E and q, oxidized DsbA wild type; छ and ࡗ, reduced DsbA wild type; Ⅺ and f, oxidized variant E24Q; É and ç, reduced variant E24Q. Only one-half of each set of data points is shown. The normalized transitions were obtained from the original fluorescence data after a six-parameter fit (solid lines), assuming the two-state model of folding (47). The thermodynamic stabilities of all oxidized and reduced variants studied are summarized in Table I. fected the pK a value of the Cys 30 thiol but not that of the Cys 33 thiol, we expected decreased pK a values of Cys 30 in the more oxidizing variants and increased pK a values in the more reducing variants. However, there was no general correlation between the Cys 30 pK a values and the measured redox potentials (Table III). For example, the most oxidizing variant E24Q/ E37Q even showed an increased pK a of the Cys 30 thiol.
pH Dependence of the Polypeptide Specificity of the DsbA Variants-It was shown previously that DsbA possesses a peptide binding site (17). This explains its specificity toward unfolded polypeptide substrates that are randomly oxidized by the enzyme (20, 21, 24, 55-57). DsbA's polypeptide specificity is higher at acidic pH and independent of the isoelectric point of the polypeptide substrate (20). To test whether the charge network Glu 24 -Glu 37 -Lys 58 affects the polypeptide specificity of DsbA, we measured the reactivities of the oxidized variants TABLE I Thermodynamic stabilities (⌬G Stab ) of oxidized and reduced wild type DsbA and the DsbA variants at 25°C and pH 7.0 GdmCl-dependent unfolding/refolding equilibria were measured as described in the legend to Fig. 3. ⌬⌬G WT corresponds to the difference between ⌬G Stab of an oxidized or reduced DsbA variant and the corresponding redox form of the wild type (⌬G Stab (variant) Ϫ ⌬G Stab (wild type)). ⌬⌬G ox/red represents the difference between the free energy of folding of the oxidized and the reduced forms of a DsbA variant. An entropic destabilization of the oxidized, unfolded proteins by the disulfide bond was not considered.   10 and compared them with their reactivities toward DTT. Since both hirudin and RNase A lack tryptophan residues, the reactions were followed by the increase in DsbA's tryptophan fluorescence in a stopped-flow fluorescence spectrometer. The polypeptide specificity of DsbA was defined as the ratio between the apparent second-order rate constant of random oxidation of a reduced substrate protein and the apparent secondorder rate constant of oxidation of DTT (k protein /k DTT ). Fig. 6 shows the pH dependence of the polypeptide specificities of all variants for the substrates RNase A and hirudin. Like in DsbA wild type, the peptide specificity of variants is most pronounced at acidic pH and disappears at pH values above 9. Overall, the values of k protein /k DTT do not differ more than a factor of 4 from those of wild type DsbA. The individual values of k protein and k DTT were also very similar to those of wild type DsbA (data not shown). Interestingly, the peptide specificity of the variant E24Q is present over a broader pH range compared with the wild type and all other variants and is still significantly pronounced between pH 7 and pH 8 (Fig. 6). The only variant that appears to be sensitive toward the charge (pI) of the polypeptide substrate proved to be the variant E24Q/K58M, which exhibits the highest k protein /k DTT values for RNase A and the lowest k protein /k DTT values for hirudin at acidic pH (Fig. 6).

DISCUSSION
The nucleophilic active-site cysteine thiols of Cys 30 in DsbA and Cys 32 in thioredoxin have very low pK a values of approximately 3.5 (18 -20) and 7.5 (27,60), respectively, compared with 9.5 for a normal cysteine (61). This difference has been proposed to be the main determinant of the redox properties of these enzymes (18). The thiolate anion that is located at the N terminus of an ␣-helix has been suggested to be stabilized by the partial positive charge of the helix dipole and by additional electrostatic interactions between charged residues and the active-site disulfide (18,27,31,38,62). In this study we focused on the interaction network of all charged residues (Glu 24 , Glu 37 , Glu 38 , and Lys 58 ) in the vicinity of the active site of DsbA because at least Glu 24 and Glu 37 may form an ion pair with Lys 58 in the structure of oxidized DsbA (17).
Thermodynamic Stabilities of the DsbA Variants and Function of Glu 24 -The residues corresponding to Glu 24 and Lys 58 in DsbA, e.g. Asp 26 and Lys 57 in E. coli thioredoxin, are in general conserved in disulfide oxidoreductases (Fig. 2). In thioredoxin, an acidic residue corresponding to Glu 37 or Glu 38 in DsbA is absent, and thus a buried salt bridge is very likely to be formed solely between Asp 26 and Lys 57 . Importantly, an increased pK a value of Asp 26 was observed, which was attributed to the hydrophobic environment of this residue (27, 35, 36, 39 -42). An abnormal pK a of an acidic residue in a native protein influences protein stability pH-dependently as shown by the Equation 3 (35,36), in which ⌬⌬G is the energy by which the acidic residue stabilizes or destabilizes the protein, and pK a (f) and pK a (u) are the pK a values of the acidic residue in the native and unfolded protein, respectively. Accordingly, the oxidized variant D26A was demonstrated to be 15.5 kJ/mol more stable than the wild type at pH 7.0 (35,36). On the other hand, the pK a of Asp 26 does not explain why oxidized thioredoxin is thermodynamically favored over the reduced form by 10 kJ/mol at pH 7.0 (63), since the pK a was demonstrated to be invariant with a value of 7.5 in both oxidized and reduced thioredoxin (40). Removal or absence of the salt bridge Asp 26 /Lys 57 in the variant K57M or in human thioredoxin (which lacks the residue corresponding to Lys 57 ),  Table III.   TABLE III Measured pK a values of the Cys 30 thiol, equilibrium constants with glutathione, and redox potentials of DsbA wild type and the DsbA variants at 25°C and comparison with predicted redox potentials The pK a values of the thiol of Cys 30 were determined by absorbance measurements at 240 nm as described in Fig. 5. Equilibrium constants (K eq ) with glutathione were measured at pH 7.0 as described in the legend to Fig. 4. A value of Ϫ240 mV (53) was used as the standard redox potential of GSH/GSSG to calculate the redox potential of DsbA. The changes in redox potential compared with the wild type (⌬EЈ 0 WT ) were predicted from the measured pK a values of the Cys 30 thiol according to the theory of Szajewski and Whitesides (54) assuming a constant pK a of Cys 33 of 9.5. ⌬EЈ 0 WT was also predicted from ⌬⌬⌬G ox/red , the energy difference between ⌬⌬G ox/red of a DsbA variant and ⌬⌬G ox/red of DsbA wild type (see Table I). In both cases ⌬EЈ 0 is obtained from the equation ⌬EЈ 0 ϭ Ϫ(RT/2F) ⅐ ln(K variant /K WT ), with K variant and K WT representing the predicted equilibrium constants with glutathione of a DsbA variant and the wild type, respectively. respectively, results in an increased pK a of 9.4 for Asp 26 (41,42) and should therefore destabilize the protein (35,36). Nevertheless, assuming a pK a (u) of 3.9 (61), a ⌬⌬G of only ϩ0.7 kJ/mol at pH 7.0 (Equation 3) is predicted, and consequently, the pK a of Asp 26 cannot account for the stability difference between the oxidized and reduced form of thioredoxin at physiological pH. Interestingly, the exchange K57G had no effect on the pK a of Asp 26 in oxidized thioredoxin (27). The same considerations principally apply to the buried side chain of Glu 24 in DsbA and the putative salt bridge with Lys 58 . Glu 24 was proposed to be the residue titrating at pH 6.7 and increasing the reactivity of the Cys 30 thiolate with iodoacetamide about 4-fold (18). Assuming a pK a (f) of 6.7 (18) and a pK a (u) of 4.4 (61), the removal of the negative charge is predicted to stabilize the variant E24Q 12.1 kJ/mol at pH 7.0 (Equation 3). However, we found only a slight stabilization of 2.2 kJ/mol for reduced and 2.3 kJ/mol for oxidized E24Q (Table  I). If the pK a of Glu 24 , as is the case for Asp 26 of thioredoxin, increased about 2 units in the variant K58M, a ⌬⌬G of ϩ2.7 kJ/mol at pH 7.0 should be expected. Instead, values of ϩ8.3 and Ϫ3.0 kJ/mol were observed for the oxidized and reduced variant, respectively (Table I), which cannot be explained by an altered pK a of Glu 24 . The same considerations apply for Glu 37 , which could also form a salt bridge with Lys 58 and which is an alanine in thioredoxin (Fig. 2).
In contrast to the D26A exchange in thioredoxin, which caused an increase of the Cys 32 pK a of 0.4 -0.9 units (27,41), the pK a of Cys 30 in the DsbA variant E24Q was unchanged (Table III). The function of Glu 24 in DsbA therefore appears to be different from that of Asp 26 in thioredoxin. This might result from the presence of Glu 37 , but the double variant E24Q/E37Q is destabilized 5.4 and 8.0 kJ/mol for the oxidized and reduced forms, respectively, although the Cys 30 pK a is increased 0.3 units (Tables I and III).
It can be concluded that, in agreement with the data available for thioredoxin, the pK a values of Glu 24 and Glu 37 most likely do not contribute to the stability difference between the oxidized and reduced form of DsbA either in the presence or in the absence of the buried salt bridge with Lys 58 . Nevertheless, altered stabilities of both forms can be detected with the DsbA variants. Whether Glu 24 indeed modulates the reactivity of Cys 30 in DsbA still has to be tested by the pH dependence of the reactivity of Cys 30 in the E24Q variant.
The analysis of the thermodynamic stabilities of the variants with the concept of double mutant cycles (52) reveals another unexpected feature of the supposed ion pairs Glu 24 -Lys 58 and Glu 37 -Lys 58 because they appear to counteract the oxidative force of wild type DsbA by thermodynamically stabilizing the oxidized form. The same is valid for the interaction between Glu 24 and Glu 37 (Table II).
Correlation of Stabilities, pK a Values of the Cys 30 Thiol, and Redox Potentials-In principle, the following predictions can be made for the relationship between the difference of the thermodynamic stabilities of the oxidized and reduced form (⌬⌬G ox/ red), the pK a of the Cys 30 thiol, and the equilibrium constant with glutathione (redox potential) of DsbA. (i) The ⌬⌬G ox/red at a given pH should be predictable if it is assumed that the oxidation of DsbA causes nothing but the removal of the negative charge of Cys 30 (18). (ii) Equilibrium constants of disulfide exchange reactions can be predicted if the pK a values of all thiols involved are known (54). Consequently, a change in the pK a of Cys 30 in a DsbA variant should alter the redox potential (⌬E 0 Ј WT ) (18,19). (iii) ⌬⌬G ox/red should also directly correlate with ⌬E 0 Ј WT . Although a correlation between changes in the pK a of Cys 30 and the redox potentials could recently be demonstrated for a series of DsbA variants in which the XX dipeptide between the active-site cysteines had been randomly mutated (19), Table III shows that the above assumptions are not consistent with the results obtained for DsbA variants devoid of charges in the vicinity of the active site. The situation seems to be far more complex. The most extreme example is the variant E37Q/K58M, which is 8 mV more reducing than the wild type, but the ⌬⌬G ox/red value predicts an increase of the redox potential of 77 mV (Table III). Overall, the measured redox potentials of the DsbA variants correlate neither with the values of ⌬⌬G ox/red nor with the Cys 30 pK a values. What are the possible explanations for these deviations from theory? First, the assumption that the pK a of the buried active-site cysteine FIG. 6. pH dependence of the polypeptide specificity of wild type DsbA and the DsbA variants at 25°C. The polypeptide specificity of DsbA was defined as the ratio between the apparent second-order rate constant of the random oxidation of a reduced substrate protein and the apparent second-order rate constant of oxidation of DTT (k protein /k DTT ). Reduced RNase A (A) and reduced hirudin (B) were used as examples for proteins with basic and acidic pI, respectively. The reactions were performed under pseudo-first-order conditions and followed by the increase in tryptophan fluorescence caused by the generation of reduced DsbA (for details see "Experimental Procedures"). q, DsbA wild type; छ, E24Q; E, K58M; Ⅺ, E24Q/K58M; ࡗ, E37Q/K58M; f, E24Q/E37Q; å, E24Q/E37Q/K58M; ç, E24Q/E37Q/E38Q/K58M. Cys 33 is normal and not affected by residue replacement (18) may not be justified. Since the absorbance of DsbA at 240 nm changes strongly at alkaline pH, we were not able to measure the pK a of Cys 33 accurately by the difference in A 240 between the oxidized and reduced variants. Single exchanges in the charge network Glu 24 -Glu 37 -Glu 38 -Lys 58 of DsbA are likely to change the pK a values of the other residues in the network, i.e. the loss of a negative charge at one position may be compensated by a decreased pK a of another acidic residue. Second, at least some of the exchanged, charged residues may only affect the stabilities of the oxidized and reduced forms of the protein but not the pK a values of the active-site cysteines. In this context, the properties of the DsbA variant K58M are remarkable. Although the removal of the positive charge of Lys 58 is expected to cause an accumulation of negative charges in the vicinity of the active site, the pK a of Cys 30 is even further decreased to 3.2 (Table III), which is, to our knowledge, the lowest pK a of a cysteine thiol in a native protein known so far. Therefore, we believe that especially the pH dependence of ⌬⌬G ox/red and the equilibrium constants with glutathione will have to be measured for DsbA wild type and some of the variants to gain further insights into the relationship between ⌬⌬G ox/red , E 0 Ј , and the pK a of the Cys 30 thiol. In Vivo Activity and Polypeptide Specificity of the DsbA Variants-It is obvious that neither charge of the residues Glu 24 , Glu 37 , Glu 38 , and Lys 58 is required for efficient recycling of oxidized DsbA by DsbB in vivo, since all variants restored the DsbA ϩ phenotype of the dsbA-deficient E. coli strain THZ2. The broad polypeptide specificity of DsbA was found to be most pronounced at acidic pH but was largely unaffected in the variants. Only the variants E24Q and E24Q/K58M showed higher polypeptide specificity also between pH 7 and 8 and a slightly increased preference for RNase A compared with hirudin, respectively (Fig. 6). The x-ray structure of oxidized DsbA supports the view that charges near the active site are not critical for peptide recognition (17). DsbA has a conserved groove within the thioredoxin domain, which is located next to the active-site disulfide and composed of hydrophobic and uncharged polar residues.
Conservation of the Buried Salt Bridge-The results presented in this paper raise the question of why the residues corresponding to Glu 24 and Lys 58 in DsbA are so well conserved within the disulfide oxidoreductase family. The reactivity of oxidized DsbA toward reduced polypeptides and DTT and the other redox properties investigated are practically independent of the presence of Glu 24 and Lys 58 or even of the interaction between Glu 24 and Glu 37 . One possible explanation could be that the Glu 24 -Lys 58 ion pair has a distinct function in the folding pathway of thioredoxinlike proteins. In the case of barnase, a buried salt bridge has been shown to direct the folding pathway without stabilizing the native structure (64,65). Thus, as in barnase, ionic interactions could play an important role in the early condensation of the disordered polypeptide chain of disulfide oxidoreductases by restricting the number of possible conformations of early folding intermediates, which therefore may result in the acceleration of folding.