Cellular quality control screening to identify amino acid pairs for substituting the disulfide bonds in immunoglobulin fold domains.

We are interested in determining which amino acid pairs can be substituted for the disulfide (S-S) bonds in proteins without disrupting their native structures under physiological conditions. In this study, we focused on the intradomain S-S bonds in Ig fold domains and aimed to determine a simple rule for replacement of their S-S bonds. The cysteines of four different Ig fold domains were mutated randomly, and the amino acid pairs substituted for the S-S bonds were screened by the method utilizing a cellular quality control system. Among the 36 selected mutants, 31 were natively folded without S-S bonds, as judged from the cooperativity of thermal unfolding. In addition, the selected mutant llama heavy chain antibodies retained antigen-binding affinity. At least two of the pairs Ala:Ala, Ala:Val, Val: Ala, and Val:Val were found in the selected mutants for all four different Ig fold domains, and they were stably folded at 30 degrees C. This suggests that examination of these four pairs could be enough to obtain natively folded Ig fold domains without S-S bonds.

Native disulfide (S-S) 1 bonds in proteins stabilize their functional structures, and the removal of an S-S bond results in marked destabilization of these proteins (1)(2)(3)(4)(5)(6)(7)(8). It is of interest to determine which amino acid pairs can be substituted for the S-S bonds without disrupting their native structures under physiological conditions. Many S-S-bonded proteins are often regarded as targets for industrial and pharmaceutical uses. For practical applications, efficient production of these proteins with recombinant technology is essential, and the in vitro formation of correct S-S bonds is one of the most challenging problems for the preparation of functional recombinant proteins. Eliminating even one S-S bond critically decreases the possible S-S-bonding combinations, which results in higher yields of active proteins.
In this study, we focused on the replacement of intradomain S-S bonds in Ig fold domains. Four typical Ig fold domains were used, and amino acid pairs substituted for S-S bonds were selected by cellular quality control screening. Then, the selected mutants were examined for their ability to form the native structures using CD and functional assay. Two of the four Ig fold domains tested here were variable fragments, i.e. the variable region of the heavy chain of camelid heavy chain antibody (VHH) and an engineered mouse V L domain ( graft). The other two were ␤2-microglobulin (␤2-m) and the constant domains of the human light chain (C L fragment). The VHH used here recognizes human chorionic gonadotropin (hCG) as an antigen (9). The graft is based on the sequence of the humanized 4D5-derived V L domain artificially designed to change the antigen specificity and interdomain interactions (10 -12).
Our screening method, which utilizes a cellular quality control system for the secretory pathway in Saccharomyces cerevisiae and is referred to as "cellular quality control screening," was developed to screen for sequences that can fold into the native structures under physiological conditions. This method is based on the efficiency of secretion of target sequences detected by an antibody directed against a generic tag (13,14). In the secretory quality control system, native or correctly folded proteins are secreted, whereas misfolded proteins are retained and degraded within the cells; the efficiencies of secretion of proteins correlate with their structural statuses (14 -18). Thus, cellular quality control screening does not require any prior knowledge of the target protein, such as its enzymatic activity. Taking advantage of this unique feature, we applied an identical screening system to the selection of amino acid pairs able to replace S-S bonds of four different Ig fold domains that do not share any functional properties.

Construction of Libraries of Mutant Ig Fold
Domains-DNA fragments encoding a signal sequence, a pro-sequence, a multicloning site, and a FLAG tag sequence were all inserted sequentially into p415GALS (19) to create p415GSSE (14). Synthetic genes of VHH, C L fragment, and graft were designed to optimize the codon frequencies for yeast using the peptide sequences of the Protein Data Bank accession number 1G9E (from 1 to 117), the sequence published in Ref. 12 (from 1 to 111), and the Protein Data Bank accession number 1CLY (from 114 to 218), respectively. The plasmid containing the cDNA of human ␤ 2 -m was a gift from Y. Goto (Osaka University) and H. Naiki (Fukui Medical University). Mutant Ig fold domains varying in the random mutations at the Cys positions were prepared by PCR using Pfu Turbo DNA polymerase (Stratagene) and oligo DNAs with random sequences at the Cys positions. In the case of ␤ 2 -m, we used three different oligos to introduce the random mutations and to avoid the appearance of Cys at the mutated positions. In these oligos, the DNA sequences at Cys positions were VNN, NHN, and NNR, where V, N, H and R indicate the mixed base pairs of A/C/G, A/C/G/T, A/C/T, and A/G, respectively. The prepared plasmids were transformed into Escherichia coli XL-1 Blue (Stratagene) by electroporation using a Gene Pulser II (Bio-Rad). The VHH, ␤ 2 -m, C L fragment, and graft libraries consisted of 13,000, 47,000, 26,000, and 480,000 colonies, respectively. About 90% of the colonies contained successfully ligated plasmids. In the sequences of randomly picked clones, no clear preference for specific amino acids was observed.
Screening of the Random Mutant Libraries-S. cerevisiae strain RY810556-2B (MAT␣ ura3-1 leu2-3, 112 his3-11, 15 can1-100) was a gift from G. Fink of the Whitehead Institute for Biomedical Research (Cambridge, MA). The expression and detection of secreted proteins were carried out as described previously (13,14). After the transformation of libraries into yeast by electroporation, we obtained 22,000 colonies from the VHH library, 4,000 from the ␤ 2 -m library, 9,000 from the C L fragment library, and 79,000 from the graft library. As a positive control, the transformants bearing the wild-type sequences were placed on separate areas of the plates for comparison. The secreted proteins were transferred to nitrocellulose membranes and then detected with an anti-FLAG M2 monoclonal antibody (Sigma-Aldrich) and chemiluminescence (Amersham Biosciences). On the initial on-plate screening, we selected 179, 32, 48, and 144 clones from transformants from the VHH, ␤ 2 -m, C L fragment, and graft libraries, respectively. The candidate clones were cultured for 1 day in natural liquid medium containing 10 g/liter of yeast extract and 20 g/liter of peptone (YEP) with 2% galactose as the carbon source (20). Each culture medium was subjected to SDS-PAGE under non-reducing conditions. The M2 antibody was used to detect proteins after blotting.
Biophysical Measurements of the Selected Mutants-Isolated DNAs encoding the candidate mutants and the original Ig fold domains were cloned into an E. coli expression vector, pAED4 (21). In addition, several control mutants were generated by PCR-based site-directed mutagenesis. They were expressed in E. coli strain BL21 (DE3) pLysS (Stratagene), and the expressed proteins were accumulated in inclusion bodies. These inclusion bodies were dissolved with 6 M guanidine HCl or 8 M urea. Crude ␤ 2 -m and C L fragments in the denaturant were dialyzed against 1% acetic acid and then purified by reversed-phase high performance liquid chromatography (HPLC). For VHHs and grafts, proteins in the denaturant were refolded by a 1/50 dilution with 10 mM sodium acetate (pH 5.3). A Resource S cation exchange column (Amersham Biosciences) equilibrated with 10 mM sodium acetate (pH 5.3) was used to purify crude VHHs and grafts. For the preparation of reduced proteins, the oxidized proteins were incubated with Ͼ20 mM dithiothreitol and 6 M guanidine HCl at pH 8.5. After overnight reaction, the completion of reduction was confirmed by either Ellman's method (VHH and graft) (22) or HPLC (␤ 2 -m and C L fragment) (23). By matrixassisted laser desorption ionization time-of-flight mass spectrometry using a Voyager DE STR-D1 (Applied Biosystems, Foster City, CA), the molecular weights of the purified proteins were confirmed to be identical to the expected values calculated from their amino acid sequences (with an error of Ϯ0.025%). The concentration of protein in the stock solution was determined by measuring the absorbance at 280 nm with a UV-2500PC spectrophotometer (Shimadzu, Kyoto, Japan) (24).
Thermal unfolding was monitored as the change in ellipticity at either 230 or 235 nm for VHH and graft, 210 nm for ␤ 2 -m, and 232 nm for C L fragment at a protein concentration of 4 M using a J-720 or J-820 spectropolarimeter (Jasco, Tokyo, Japan) and a 1-cm cell. For all measurements, the heating rate was 1°C min Ϫ1 . The buffers used for the experiments on VHH, ␤ 2 -m, and C L fragment contained 20 mM sodium phosphate (pH 7.9). Because the heat-denatured grafts tended to form aggregates at neutral pH, thermal unfolding experiments were carried out at pH 5.3 with a 10 mM sodium acetate buffer. The midpoint temperatures of thermal unfolding (T m ) and enthalpy values of unfolding (⌬H u ) were estimated on the basis of the two-state transition mechanism by curve fitting using Igor Pro (WaveMetrics Inc., Lake Oswego, OR) (25). We assumed that the change in heat capacity between the folded and unfolded states (⌬C p ) was 0, as all analyzed transitions were steep, and the influence of ⌬C p on T m was negligible. In most cases (Ͼ80% of all measurements), Ͼ70% of the CD signals were retained at 10°C after heat denaturation. Even in another 20% of the experiments with relatively low reversibility, at least 40% of the CD signals were retained after cooling down.
Binding Assay of VHH with Antigen hCG-VHH mutants were immobilized on cyanogen bromide-activated Sepharose (Amersham Biosciences) using the standard protocol supplied by the manufacturer. Aliquots of Sepharose beads (20 l), to which about 10 g of VHHs had been immobilized, were incubated with 10 g of hCG (Sigma-Aldrich) in a 20 mM Tris-HCl buffer (pH 8.0) containing 0.15 M NaCl at room temperature for 90 min. After washing the beads three times with the same buffer, samples were boiled for 15 min at 95°C with standard SDS-PAGE sample buffer containing ␤-mercaptoethanol and then applied to the 10 -20% gradient SDS-polyacrylamide gels. The gels were stained with Coomassie Brilliant Blue.

Model Building of Mutant Structures and Calculation of the Contact
Surface Area-Model structures of the mutants were obtained by simulations using dead-end elimination (26) from the structures of wildtype VHH (Protein Data Bank accession number 1HCV), ␤ 2 -m (Protein Data Bank accession number 1HHG), and C L fragment (Protein Data Bank accession number 1CLY). Two-step energy minimization involving the conjugate gradient method was applied for each generated structure using the software PRESTO, version 3 (27,28). In the first step, only the mutated residues were allowed to move freely, the other residues being fixed. In the next step, restricted movement was permitted for other residues using position-dependent restraints.
The area of contact surface between the mutated and surrounding residues was calculated for each model structure using the following procedure. We defined ASA paired as the water-accessible surface area (ASA) calculated for the two residues simultaneously extracted from all of the structural coordinates. Notation ASA paired (␣,␤) contains only the ASA from residue ␣ and does not include the ASA from residue ␤ of ASA paired , which is calculated for the residue ␣ and ␤ pair. ASA single (␣) is defined as ASA for only the residue ␣ extracted from the structural coordinates. The contact surface area for residue i, ⌬ASA mut (i), was calculated for each mutated residue using the equation below.
⌬ASA mut (i) ϩ ⌬ASA mut (j) is calculated for the mutated residue pair i and j as the external contact. The internal contact between mutated residue pair i and j (⌬ASA mutpair (i,j)) is defined in the following equation.

Screening of Stable Ig Fold Mutants without the S-S Bond-
In the first on-plate screening step, we selected 179, 32, 48, and 144 clones from the VHH, ␤ 2 -m, C L fragment, and graft libraries based on the secretion levels of target sequences. As a positive control, we placed the transformants with the wild-type sequences on separate areas of the plates. To prevent the false positives on screening, we introduced multi-step selection, in which the first on-plate screening is followed by secondary liquid culture secretion screening (13,14). In the subsequent liquid culture secretion screening, the supernatants of cultures of the candidates were subjected to SDS-PAGE and immunoblotting. We selected 64 (VHH), 24 (␤ 2 -m), 24 (C L fragment), and 38 ( graft) clones that produced bands of substantial intensity (Fig. 1). DNA sequencing of the candidates revealed 18 (VHH), 3 (␤ 2 -m), 8 (C L fragment), and 11 ( graft) different amino acid sequences ( Fig. 1 and Table I). Those of VHH, C L fragment, and graft included revertants, which had Cys:Cys at the positions of S-S bonds, found in 1, 1, and 11 clones, respectively. In screening of the C L fragment library, the Cys:Val mutation was found in seven clones. As Cys:Val contains cysteine, we did not carry out further analysis of this mutant. In the case of ␤ 2 -m, we used three different oligo DNAs to introduce random mutations and to avoid the appearance of Cys at the mutated positions.
It should be noted that the levels of secretion of the graft mutants on plates and in liquid medium were very low compared with the case of the wild-type Cys pair. Thus, for the first screening, we selected the clones with relatively weak signals, along with those with intense signals, which were similar to that from clones with the wild-type sequence. The DNA sequencing of candidates with intense signals revealed that all of these mutants had the Cys:Cys pair at the position of the S-S bond. For the second screening of graft mutants, we cultured the candidates at 24°C instead of the normal growth temperature of yeast (30°C) (Fig. 1D). Because the levels of secretion of graft mutants, which do not include revertants, were still low even at 24°C, we compared the levels of secretion of mu-tants with a 1/5 dilution of the wild-type protein (Fig. 1D, lane 1). The graft mutants were the most aggregation-prone among the Ig fold domains studied here, and this could be related to the low levels of secretion of graft mutants in liquid cultures.
Thermal Unfolding of the Obtained Mutants-Almost 90% of the selected mutants were unfolded in a cooperative manner, as judged from the thermal transition detected by means of CD, indicating that they retained the ability to form the native structures (Fig. 2). The Trp:Ile G17D mutant of graft was not soluble after refolding, and thus, we examined 35 mutants. Among them, 31 showed cooperative thermal unfolding curves. Because ⌬H U reflects the steepness of the thermal transition, this value is a good index of the cooperativity of unfolding. The ⌬H U values at the T m of these mutants were 240 -390 kJ/mol for VHH, 240 -310 kJ/mol for ␤ 2 -m, 180 -230 kJ/mol for C L fragment, and 180 -260 kJ/mol for graft. The ⌬H U values of oxidized wild-type VHH, ␤ 2 -m, C L fragment, and graft at their T m values were 380, 260, 270, and 290 kJ/mol, respectively. Considering the difficulty in estimating ⌬H U from one transition curve, the ⌬H U values of the mutant and wild-type proteins were similar. In addition, the ⌬H U values of mutants were reasonable values as compared with those of typical globular proteins with molecular weights between 8,000 and 15,000 (from 160 to 470 at 50°C) (29).
Some of the mutants obtained for VHH (Trp:Ala, Ala:Ala, and Trp:Pro G10D), ␤ 2 -m (Val:Val, Val:Ala, and Ala:Val), and C L fragment (Ala:Val, Ala:Ile, and Val:Val) showed higher T m values than those of the corresponding reduced wild-type proteins (Table I). The largest difference in T m values between the selected mutants and the reduced wild-type protein was 6°C. This indicates that a Cys:Cys pair at the position of an S-S bond does not always best fit this position under reducing conditions. As the Ig fold domains used here were evolved to function under oxidized conditions, it is reasonable that their sequences are not optimal for reducing environments.
The difference in T m values between oxidized and reduced wild-type grafts was 9°C (Table I), which was the smallest among the Ig fold domains examined here. Loop entropy calculations (30) and the enthalpy values of oxidized proteins allow prediction of the T m values of the reduced proteins: 41°C for VHH, 39°C for ␤ 2 -m, 33°C for C L fragment, and 31°C for graft. In this calculation, ⌬C p was assumed to be 6 kJ mol Ϫ1 K Ϫ1 , because the ⌬C p values of eight typical small globular proteins, of which the molecular weights are between 8,000 and 15,000, are reported to range from 4 to 9 kJ mol Ϫ1 K Ϫ1 , and the average ⌬C p is 6 Ϯ 2 kJ mol Ϫ1 K Ϫ1 (29). The predicted T m values are close to the observed T m values of reduced wild-type VHH (41°C), ␤ 2 -m (37°C), and C L fragment (32°C). However, in the case of graft, the observed T m was about 14°C higher than the predicted value. This suggests that reduction of the S-S bond in graft triggers new and preferable effects, such as hydrogen bond formation or the relaxation of main chain distortion caused by the S-S bond.
Because Val:Ala was not selected on the screening of C L fragment mutants, although this pair was found on the other three screenings, we prepared a Val:Ala mutant of C L fragment and compared it with the selected mutants. This comparison revealed a preference, or polarity, for Ala:Val and Val:Ala mutations in C L fragment. The differences in T m between the Ala:Val and Val:Ala mutants of VHH, ␤ 2 -m, and graft were 4, 1, and 1°C, respectively. On the other hand, the T m difference between the Ala:Val and Val:Ala mutants for C L fragment was 8°C. The occurrence of polarity implies that the pairs substituted for the S-S bonds were determined not only by the structural similarity to the Cys pair but also by the environment around the positions of the replaced Cys residues. For comparison, and because Ala:Ala has often been used to replace S-S bonds (5, 31, 32), we made Ala:Ala mutants for ␤ 2 -m, C L fragment, and graft. In the case of C L fragment, the T m was Ͼ10°C lower than that of the most stable selected mutant, Ala:Val, indicating that Ala:Ala is not always an appropriate pair to replace an S-S bond. Although Val:Val was selected from ␤ 2 -m, C L fragment, and graft libraries, we could not find this pair in the selected VHH mutants. We prepared a Val:Val mutant as a control and compared the stability. The Val:Val mutant of VHH was 2-13°C less stable than the selected pairs and thus was not selected on the screening.
Antigen Binding of the Obtained VHH Mutants-The wildtype VHH used here was the variable domain of a llama heavy chain antibody raised against the ␣-subunit of hCG. To determine whether S-S bond substitution affects the antigen-binding ability, Sepharose-immobilized VHH variants were prepared and incubated with hCG at room temperature (Fig. 3). The amounts of hCG bound to immobilized Ala:Ala, Ala:Ser, Gly:Val, Ala:Ile, Ala:Val, Trp:Pro G10D, Trp:Ala, Trp:Gly, and Ser:Ser were similar to those of wild-type VHH, indicating that these mutants retain antigen-binding affinity. Ala:Leu, Ala:Phe, Val: Ala, Gly:Ile, Ser:Ala G97V, Gly:Phe, Gly:Ala, and Gly:Leu also bound hCG; however, the amounts of bound hCG were smaller The panels showing the western blots of VHH, C L fragment, and graft mutants were prepared for comparison using selected candidates and false positive clones, which were selected on the first screening but did not show a considerable level of secretion on the second screening. The panel for ␤ 2 -m was used to select candidates. The clones with mutants and wild-type sequences were cultured for 1 day in YEP medium with 2% galactose as the carbon source. Each culture medium was subjected to SDS-PAGE under non-reducing conditions. The name of the mutant pair is shown above each lane. CC denotes the wild-type sequence. The secretion levels of the graft mutants were very low, thus the supernatant of the wild-type graft culture was diluted at 1/5 before electrophoresis. than those for wild-type VHH. hCG did not interact with the Sepharose alone (Fig. 3, lanes 2 and 13). The Val:Val mutant, which was prepared as a negative control and exhibited lower stability, bound only a small amount of hCG (Fig. 3, lanes 5 and  15). About 10% of the Val:Val mutant was unfolded at 25°C, as observed in the thermal unfolding curve (Fig. 2A). Thus, it is reasonable to conclude that the equilibrium between the unfolded and folded molecules of the Val:Val mutant greatly reduced the antigen-binding affinity at room temperature. We also carried out competition experiments on the immobilized (7 g) and soluble (5 g) wild-type and Ala:Ile VHHs. The addition of soluble protein reduced the amount of hCG bound to the immo-bilized wild-type and Ala:Ile VHHs by 90 and 50%, respectively (data not shown). This finding further corroborates the presence of specific interactions between hCG and VHH mutants.
False Positives on the Cellular Quality Control Screening-There were some examples that did not satisfy the basic assumption behind our screening method based on a cellular quality control system. For graft, we could not observe the cooperative transition for the Val:Val, Ile:Val, Phe:Tyr, and Trp:Val mutants, and Trp:Ile V17D mutant was insoluble after refolding, suggesting that these mutants have non-native or aggregated structures. The Gly:Phe, Gly:Leu, and Gly:Val pairs of C L fragment exhibited significant levels of secretion in the second screening (Fig. 1C, lanes 8, 11, and 16) and showed cooperative thermal unfolding curves. However, their T m values were Ͻ30°C, indicating that more than half of the molecules are unfolded at the temperature for yeast culture. The occurrence of these false positives suggests that the level of secretion is determined not only by the structural status and stability but also by other unknown factors. DISCUSSION It is notable that at least two of the pairs Ala:Ala, Ala:Val, Val:Ala, and Val:Val were found in the selected mutants for all four different Ig fold domains and that they were stably folded at 30°C (Table I and Fig. 2). This suggests that examination of only four mutations, i.e. Ala:Ala, Ala:Val, Val:Ala, and Val:Val, could be enough to obtain natively folded Ig fold domains without S-S bonds. The number of pairs using Ala and/or Val is far smaller than ones using all 20 amino acids, reducing the efforts necessary to identify amino acid pairs to replace S-S bonds without disrupting the native structures under physiological conditions. Therefore, our finding makes it easy and practical to replace the S-S bonds of preexisting antibodies.
Removing the S-S bonds from antibodies simplifies their in vitro folding and thus enables their production at low cost using a bacterial expression system. In addition, reducing the number of S-S bonds of antibodies is a key issue for extending their applications, such as the broad use of engineered multivalent antibodies and "intrabodies," another form of engineered antibody that is a recombinant antibody used to block the function of a target protein within the cell (33)(34)(35)(36)(37)(38)(39)(40)(41). A single chain variable fragement of an antibody (scFv), which consists of variable domains of heavy and light chains, comprises a small antibody fragment with antigen-binding ability. scFvs with different antigen specificities can be fused to produce high affinity multivalent antibodies. Although in principle any number of scFvs can be connected, the difficulty in the correct formation of many S-S bonds limits the number of scFvs that can be fused. A known problem of intrabodies is improper folding in the cell caused by the reductive environment of the cytosol, in which the S-S bonds of antibodies are hardly formed (42).
As an S-S bond in an Ig fold domain is critical for its structure and stability (2,7,8), replacement of the S-S bond was considered to be challenging. So far, only two artificial pairs (Val:Ala and Ala:Tyr) have been found to be functional replacements for the S-S bonds in Ig fold domains (3,40,43,44). It has also been reported that one of the Cys residues involved in an S-S bond can be replaced by Tyr, Ser, or Val (41,45,46). In this work, we found 21 different mutation pairs that allowed native folding, and these pairs comprised combinations of 10 amino acids, i.e. Gly, Ala, Val, Ile, Leu, Phe, Tyr, Trp, Pro, and Ser (Table I). This indicates that the S-S bonds of Ig fold domains can be substituted by various amino acid pairs; however, hydrophobic amino acids are employed in most cases. The frequent appearance of hydrophobic amino acids is consistent with the fact that S-S bonds in Ig fold domains are buried inside the hydrophobic core.  b The third mutation, G10D, did destabilize the protein. W/P without the third mutation showed a higher T m (44°C) than W/P G10D.
The most surprising mutant was Trp:Pro G10D of VHH. Proline is a strong ␤-sheet breaker (47)(48)(49), and Cys-96 of VHH is at the center of the eighth strand. Thus, we first thought that Trp:Pro G10D was a false positive in the screening. However, this mutant bound to the antigen hCG, indicating that the introduction of proline at position 96 did not cause unfolding of VHH. Furthermore, the T m of this mutant was the second highest among the VHH mutants. Because the third mutation, G10D, would contribute to the stability of this mutant, we prepared a Trp:Pro mutant without the third mutation and found that the G10D mutation rather decreased the T m by about 2°C. Thus, the Trp:Pro pair alone is a suitable pair for replacing the S-S bond in VHH.
The pairs able to replace the S-S bonds might be simply explained by the extent to which the cavity caused by removing the cysteines is filled. To test this possibility, we estimated the areas of contact between the mutated residues and the amino acids around the S-S bond (Fig. 4A). In addition, we calculated In each panel, the unfolding curves for the oxidized and reduced wild-type proteins are shown by solid black lines and dots, respectively. Control mutants (Ctrl), which were not selected on screening but prepared for comparison, are indicated next to the mutant pair names in the panels. As representative examples, we have shown only four mutants for VHH, C L fragment, and graft. The other mutants, not shown here, also unfolded cooperatively, except the Ile:Val, Phe:Tyr, and Trp:Val mutants of graft. The wavelengths for detection were chosen by comparing the spectra for folded and unfolded proteins so as to obtain reasonable signal differences between these two states.
FIG. 3. Binding of purified VHHs with the antigen human hCG. Sepharose beads carrying immobilized VHH mutants and control VHHs (10 g) were incubated with 10 g of hCG in 20 mM Tris-HCl buffer (pH 8.0) containing 0.15 M NaCl. After washing, hCG was dissociated from the beads by boiling for 15 min with SDS and ␤-mercaptoethanol, and then the supernatants were subjected to SDS-PAGE. The gels were stained with Coomassie Brilliant Blue. The name of the mutant pair is shown above each lane. CC denotes the oxidized wild-type protein. Sepharose alone mixed with hCG was applied to lanes 2 and 13, which are denoted by beads. hCG alone (5 g) was applied to lane 25 as a control. The molecular weights of markers are shown on the left of the panel. It should be noted that clear separation of the ␣ and ␤ subunits of hCG, which form a heterodimer, required treatment with a reducing reagent before SDS-PAGE. the areas of internal contact between pairs (Fig. 4B). Structural minimization of the introduced pair and the shell formed by the amino acids around the S-S bond was carried out as described under "Experimental Procedures." As the van der Waals contact is supposed to stabilize the protein structure, an increase in the contact between amino acids should result in an increase in stability. However, there was no clear correlation between stability and the contact surface area. In a cavity filling experiment on T4 lysozyme, introduction of a bulky amino acid into the hydrophobic core caused bond angle distortion and unfavorable van der Waals contacts (50). These unfavorable strains could explain why the amount of the contact area did not correlate with the stability. The lack of correlation observed here indicates that empirical testing is required to identify the suitable pairs for replacing the S-S bonds in Ig fold domains.
Systematic replacement of the S-S bonds of other types of folds has been reported. Pairs that can be substituted for the S-S bond between Cys-14 and Cys-38 of bovine trypsin inhibitor (BPTI) have been searched for using cellular quality control screening (13). In this experiment, nine pairs were obtained, and Gly:Val and Gly:Met A27V showed the highest T m values. Although Val:Val was also selected on screening, its T m was about 5°C lower than those of Gly:Val and Gly:Met A27V. In the present study, the amino acid pairs using Gly were also selected on screening (Table I). However, in contrast to the S-S bond between Cys-14 and Cys-38 of BPTI, the T m values of these mutants were 4 -13°C lower than those of the most stable mutants selected. This inconsistency would be a result of the solvent accessibility of the S-S bond replaced, because the S-S bond between Cys-14 and Cys-38 of BPTI is solvent-accessible, and the S-S bonds examined here are buried in the hydrophobic core. Indeed, for the buried S-S bond in BPTI, which connects Cys-30 and Cys-51, Ala:Ala and Val:Ala were the two most stable pairs (51). Their T m values were Ͼ10°C higher than those of the other six pairs studied, i.e. Gly:Ala, Thr:Ala, Ser:Ala, Ala:Ser, Ser:Ser, and Gly:Met. Thus, it is possible that four pairs of Ala and/or Val could be substituted for the buried S-S bonds in proteins other than Ig fold domains.
We have demonstrated that cellular quality control screening allows the identification of amino acid pairs for replacing S-S bonds in Ig fold domains. The screening showed a low frequency of false positives; the selected mutants exhibited cooperative thermal unfolding and were folded well at 30°C, except for the several false positives noted under "Results." Moreover, the selected VHH mutants showed antigen-binding affinity under our experimental conditions. Our screening system is independent of the functions of proteins and thus can be applied to other proteins with different folds. Therefore, this method is especially useful for comprehensive research on the ability of many different sequences to fold into the native structures. FIG. 4. Plots of external (A) or internal (B) contacts against ⌬T m . ⌬T m is defined as the difference in T m values between the mutants and the corresponding reduced wild-type protein. Green, red, and blue characters indicate VHH, ␤ 2 -m, and C L fragment mutants, respectively. In this figure, CC indicates the reduced wildtype protein. External contact indicates the area of contact between mutated amino acids and the residues around them, which was calculated by summation of the ⌬ASA mut (i) values for the mutated amino acid pairs as described under "Experimental Procedures." The internal contact between the mutated pairs is defined as ⌬ASA mutpair (i,j), as described under "Experimental Procedures." As Ser: Ala G97V of VHH had a third mutation, we did not include this mutant in this analysis. The linear correlation coefficients between ⌬T m of all proteins and contact areas were 0.43 (external contact) and 0.36 (internal contact).