Improving the Affinity and the Fine Specificity of an Anti-Cortisol Antibody by Parsimonious Mutagenesis and Phage Display1 Patrick Chames,2 Ste´phane Coulon, and Daniel Baty3 Immunoassays are widely used to determine steroid concentrations. However, they are limited by the specificity of antisteroid mAbs. We used the phage display system combined with molecular modeling and site-specific randomization to improve the affinity and the fine specificity of an anti-cortisol mAb. Using parsimonious mutagenesis, we have generated a library of mutant Ab fragments (scFv) derived from this Ab by randomizing five amino acids chosen by molecular modeling and Ab-hapten contact structural analysis. Anti-cortisol Ab fragments were selected from the library in the presence of steroid analogues to block cross-reacting binders. Specific elution with free cortisol allowed the recovery of clones with up to eightfold better affinity and fivefold less cross-reactivity than the wild-type scFv. This approach can be applied to any anti-hapten Ab and represents a useful approach for obtaining highly specific Abs for use in steroid immunoassays. The Journal of Immunology, 1998, 161: 5421–5429.

S

teroid hormones exert numerous and diverse actions on their target tissues. They are involved in many cellular processes controlling metabolism, reproduction, differentiation, and behavior (1). Many clinical symptoms are associated with the variation in steroid levels, and synthetic steroids are widely used as agonists or antagonists for therapy. Thus, measurement of the concentration of various steroids in blood or urine can be of great importance for clinical diagnosis and during therapy. The most frequently used method to measure steroid concentrations is competitive binding assay using Abs with high affinity and specificity and radiolabeled steroids. These assays are based on competition for binding to an mAb between radioactive cortisol and cortisol present in the sample. Obviously the Ab must be very specific to be able to discriminate between different steroids. However, the molecular structures of steroids are very similar, and consequently, the isolation of specific steroid Abs devoid of cross-reaction to steroid analogues is difficult. Another problem restricts the recovery of highly specific anti-steroids. Steroids are small molecules (300 – 400 Da) biosynthesized from cholesterol. They are by themselves incapable of causing an immunologic reaction and must be coupled to immunogenic proteins for immunization. As a consequence, the position of the link to the carrier protein affects the specificity of anti-steroid mAb (2, 3). The arm allowing the link is often involved in the epitope. Consequently,

Laboratoire d’Inge´nierie des Syste`mes Macromole´culaires, Institut de Biologie Structurale et Microbiologie, Centre National de la Recherche Scientifique, Marseille, France Received for publication December 8, 1998. Accepted for publication July 9, 1998.

the Ab has a weaker affinity for the free steroid than for the immunogen. Moreover, the cross-reactivities of Abs with different steroids can be explained by small but significant conformational changes in the Ab paratope that can thereby accommodate the different ligand orientations in the binding site (4). To circumvent the lack of specificity of the anti-steroid Abs, steroids can be extracted with an organic solvent and separated by chromatography to determine their concentrations by immunoassay (5–7). However, this method is laborious, and direct immunoassay from plasma or urine extracts is much more desirable for multiple determinations. A different approach may be to modify the characteristics of an available anti-steroid Ab to decrease its cross-reactions with analogues. Much progress has been made recently in Ab engineering. The development of methods for the cloning and heterologous expression of Ab variable gene sequences, leading to the synthesis of functional Ab fragments, together with the application of methods for site-directed mutagenesis has greatly facilitated structure-function studies of Ab-combining sites (8 – 10). The value of these methods is further enhanced by the use of molecular modeling, allowing the prediction of Ab-combining site structure. The use of these techniques combined with powerful phage display technology (11) has led to the production of several Abs with improved affinity (12–16). We have used site-specific randomization to build a library of mutant Ab fragments and then phage display to select from this library anti-cortisol Abs with the required high specificity and affinity.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. 1 This work was supported by the Centre National de la Recherche Scientifique Grant ACC-SV5 9505207 (to D.B.).

Materials and Methods Bacterial strain

2

Current address: University Hospital Maastricht, P. Debyelaan 25, P.O. Box 5800, 6202 AZ Maastricht, The Netherlands. 3

Address correspondence and reprint requests to Dr. Daniel Baty, UPR 9027, Institut de Biologie Structurale et Microbiologie, Centre National de la Recherche Scientifique, 31 chemin Joseph Aiguier, 13402 Marseille Cedex 20, France. E-mail address: [email protected] Copyright © 1998 by The American Association of Immunologists

The Escherichia coli strain TG1 (D(lac-pro), supE, thi, hsdD5/F9, traD36, proAB, LacIq, lac ZDM15) was used as the bacterial host for the preparation of phagemids and as the host for bacteriophage M13KO7. E. coli HB2151 (D(lac-pro), ara, nal r, thi/F9 proAB, LacI q, lacZDM15) was used for expression of soluble scFv. 0022-1767/98/$02.00

5422

SPECIFICITY IMPROVEMENT OF AN ANTISTEROID ANTIBODY

FIGURE 2. Building of the library by overlap PCR. To build the library, four PCR were performed, two of them incorporating mutations in the CDR H3 and L3. Purified products were then linked using overlap PCR to obtain full-size mutated genes. From left to right, the primers used are 5PelB, RPMH3, PMH3, LINKINF, LINKSUP, RPML3, PML3, and 3CMYC. FIGURE 1. Production of scFv 5A4 and scFv-pIII fusions. A, After induction for 3 h at 30°C at the indicated concentration of glucose or IPTG, HB2151 cells harboring the pHEN-5A4 vector were centrifuged and subjected to an osmotic shock. Samples corresponding to the periplasmic fraction (p) and the remaining non periplasmic fraction (np) were subjected to SDS-PAGE, and the proteins were transfered to nitrocellulose. B, HB2151 or TG1 cells harboring pHEN-5A4 vector were induced for 1 h at 30°C with 100 mM IPTG. Entire cells were subjected to SDS-PAGE and transfered to nitrocellulose. The scFv or scFv-pIII fusion proteins were revealed using the anti-c-Myc tag Ab.

Construction of the wild-type anti-cortisol scFv4 To construct the 5A4 scFv, VH and VL genes were amplified by PCR from hybridoma cDNA prepared as previously described (17). The 39 primer used to amplify VH contained a part of the coding sequence for the linker GSTSGSGKPGSGEGSTKG (18), and the 59 primer used to amplify the VL contained the rest of the sequence (with 15 overlapping bases). The scFv was assembled by splice overlap extension PCR (19). The primers used resulted in two restriction enzyme sites (NcoI and EagI) being introduced at the 59 and 39 ends of the scFv gene. These sites were used to ligate the fragments into the pHEN1 phagemid vector (20) to give pHEN-5A4.

Production of scFv and phAb HB2151 cells harboring the phagemid pHEN-5A4, PHEN1, or pHEN coding for scFv mutants were grown at 30°C to an OD600 of 0.8 in 2YT containing 100 mg/ml ampicillin and 2% glucose (2YTAG). The culture was centrifuged for 5 min at 3000 3 g and 25°C, and the cells were resuspended in the same volume of 2YT containing 100 mg/ml ampicillin and 100 mM IPTG to induce expression. The culture was incubated at 30°C for 180 min. Cell fractionation was performed as previously described (21) (Fig. 1A). For equilibrium dialysis, periplasmic extracts were dialyzed 12 h against 0.1 M sodium phosphate (pH 7.2), 10 mM azide, and 0.1% gelatin (phosphogel buffer) just before use. For phAb production, TG1 cells harboring phagemid were grown to an OD600 of 0.5 in 2YTAG at 37°C. Helper phage M13KO7 was then added to a multiplicity of infection of 10. After 30 min at 37°C without agitation, cells were centrifuged as described above and resuspended in 2YT containing 100 mg/ml ampicillin and 50 mg/ml kanamycin, and the culture was incubated at 30°C overnight with agitation. The phAb were recovered in the supernatant and used directly or purified by polyethylene glycol precipitation as previously described (20). 4

Abbreviations used in this paper: scFv, single-chain variable fragment; VH, immunoglobulin heavy chain variable region; VL, immunoglobulin light chain variable region; phAb, phage antibodies; IPTG, isopropyl-thio-b-D-galactopyranoside; cortisol-3-CMO-BSA, cortisol 3-(O-carboxymethyl)oxime-bovine serum albumin; MPBS, 4% skimmed milk powder in phosphate-buffered saline; supE, strain of Escherichia coli that carries a glutamine-inserting amber (UAG) suppressor transfer ribonucleic acid; CDR, complementarity-determining region; H1, complementarity-determining region 1 of the heavy chain V region; H2, complementarity-determining region 2 of the heavy chain V region; H3, complementarity-determining region 3 of the heavy chain V region; L3, complementarity-determining region 3 of the light chain V region.

Construction of the mutant library Overlap PCR was used to build the library (Fig. 2). Four PCR were performed with pHEN-5A4 DNA as the template and with primer pairs 5PelB and PMH3, RPMH3 and LINKINF, LINKSUP and PML3, and RPML3 and 3CMYC (Eurogentec, Seraing, Belgium; see below for primer sequences). The PCR products were purified on polyacrylamide gels, mixed, and used as template in a final PCR using primers 5PelB and 3CMYC. The final PCR product (5 mg) was inserted into the pHEN1 phagemid (2 mg). Fifty electroporations allowed us to obtain a library of 2.5 3 107 clones. Some of the clones were tested both by DNA Miniprep and restriction for the presence of an insert of expected size and also for the production of a 30-kDa product using the anti-tag Ab as previously described (20). The library was rescued as previously described (20) using helper phage M13KO7.

Primer sequences The primer sequences were: 5PelB, CTC GCK GCS CAG CCG GCC ATG GC (K 5 G/T, S 5 G/C; the NcoI site is in italic); PMH3, GCC ATC TAT TAC TGT GCA AGA 123 AGT GTC TAT GGT AGC AGC 453 CCC 663 GAT TCC TGG GGC CAA GG; RPMH3, CTT GCA AGT AAT AGA TGG C; LINKINF, TC ACC TGA ACC AGG TTT ACC AGA ACC TGA GGT AGA ACC TGA GGA GAC GGT GAC; LINKSUP, CT GGT AAA CCT GGT TCA GGT GAA GGT AGT ACT AAA GGT GAC ATT GTG CTG AC; PML3, GCC ACT TAT TAC TGC CAG CAG TGG AGT AGT 223 CCA 553 ACG TTC GGT GCT GGG ACC; RPML3, GCT GGC AGT AAT AAG TGG C; and 3CMYC, C AAG CTT ACT AGT TTA TGC GGC CCC ATT CAG ATC C (1: A 7%, C 79%, G 7%, T 7%; 2: A 79%, C 7%, G 7%, T 7%; 3: G 21%, T 79%; 4: A 70%, C 10%, G 10%, T, 10%; 5: A 10%, C 70%, G 10%, T 10%; 6: A 7%, C 7%, G 7%, T 79%).

Selection of a phAb library The phAb were selected by binding to Ag-coated immunotubes (Maxisorp, Nunc, Naperville, IL). The Ag cortisol-3-CMO-BSA (Sigma, St. Louis, MO) was used to coat the tubes overnight at 4°C at 10 mg/ml in PBS and was then saturated with 4% milk PBS (MPBS) for 1 h at 37°C. For the first round of selection, 1013 titrated units of the phAb library in a total volume of 2 ml of MPBS was used per immunotube. Before binding phAb particles were also saturated for 2 h at 25°C in MPBS. Immunotubes were washed, and bound phAb were eluted by incubation for 10 min with 1 ml of 100 mM triethylamine (and immediately neutralized with 0.5 ml of 1 M TrisHCl, pH 8.0) or by 50 mM free cortisol in PBS for 120 min at 25°C on a rotator. After each round of enrichment, E. coli TG1 were reinfected with eluted phAb, and the phAb were rescued and used for the next round of panning. Selection was stopped when the recovery of eluted phAb increased significantly. The cortisol concentration used for elution was decreased to 10 mM for the second round and to 1 mM for subsequent steps as indicated in the text. For competitive binding protocols, the analogue (prednisolone or dexamethasone) was added to the MPBS buffer used to saturate the phAb particles. Analog concentrations used during first, second, and subsequent rounds were 1, 10, and 50 mM, respectively.

ELISA screening of clones Single ampicillin-resistant colonies resulting from infection of E. coli TG1 with eluted phAb were used to inoculate 150 ml of 2YT medium containing

The Journal of Immunology 100 mg/ml ampicillin and 2% glucose in 96-well plates. The phAb was prepared as previously described (20). Supernatants containing phAb were tested for binding by ELISA in Falcon 96-well plates (Becton Dickinson, Oxnard, CA) coated overnight with cortisol-3-CMO-BSA at 10 mg/ml in PBS and saturated with MPBS. PhAb binding was detected with a horseradish peroxidase anti-M13 Ab conjugate (Amersham Pharmacia Biotech, Uppsala, Sweden). For a first approximation of the affinities and specificities of phAb in solution, the ELISA method of Friguet et al. (22) was used. Free cortisol at concentrations between 100 nM and 1 mM was mixed with a fixed amount of phAb (determined by ELISA titration of a polyethylene glycolpurified preparation) in MPBS. After incubation for 20 h at 25°C, the free phAb titer was measured by ELISA as described above. The Kd value was calculated using the following equation: A0/(A0 2 A1) 5 Kd/at1 1 1, where at1 is the given concentration of Ag, A0 is the ELISA signal obtained without cortisol in the preincubation, and A1 is the ELISA signal obtained following preincubation of phAb with cortisol at at1. Each condition was performed in duplicate. The absorbance of each well of the ELISA plates was measured when the 2,29-azino-bis(3-ethylbenzthiazoline-6-sulfonic acid) reaction was still in the linear range, a fact that was confirmed by taking several time points per plate. To determine the percentage of crossreaction, a fixed amount of phAb was incubated in ELISA wells (coated with 10 mg/ml cortisol-3-CMO-BSA) in the presence of cortisol or an analogue at concentrations ranging from 10 nM to 1 mM for 4 h at 25°C in MPBS. After washing, bound phAb was revealed as described above. From the inhibition curve, concentrations allowing 50% inhibition (IC50) were determined. The percentage of cross-reaction was given by the equation: (IC50 cortisol/IC50 analogue) 3 100. The absorbance of each well of the ELISA plates was measured when the 2,29-azino-bis(3-ethylbenzthiazoline-6-sulfonic acid) reaction was still in the linear range (confirmed by taking several time points per plate).

Equilibrium dialysis of scFv The phagemid corresponding to the phAb that showed interesting affinity measurements by the ELISA method were sequenced, and those that were mutated were used to transform HB2151 cells to prepare scFv extracts. To determine affinity and percentage of cross-reactions of scFv, equilibrium dialysis was performed in a noncommercial equilibrium microvolume dialyzer (Immunotech Coulter Beckman, Marseille, France) in which paired chambers of 150 ml each were separated by a 6000 – 8000 m.w. cut-off Spectra/Por membrane (Spectrum Medical Industries, Houston, TX). For standardization of the scFv concentration, periplasmic extracts containing scFv were diluted (1 up to 64) in phosphogel buffer. One hundred microliters of dilution was mixed with 50 ml of phosphogel and loaded into chambers on one side of the membrane; on the other side, 50 ml of [1,2,6,73 H]cortisol (2.9 TBq/mmol; New England Nuclear, Boston, MA) diluted in phosphogel (0.18 pmol final) was mixed with 100 ml of periplasmic extract from HB2151 cells harboring the pHEN1 vector at the same dilution as that used for the periplasmic extracts containing scFv. Dialysis cells were incubated for 24 h at 4°C on rotator. The radioactivity in each cell was determined by liquid scintillation counting (PCS liquid scintillation, Amersham Pharmacia Biotech). For each periplasmic extract, the dilution corresponding to 50% of the disintegrations per minute bound was chosen to calibrate the scFv concentration for the following experiment. One hundred microliters of the appropriate concentration of periplasmic extract was mixed with 50 ml of different amounts (0.0625 up to 250 pmol in phosphogel) of steroid (cortisol, dexamethasone, and prednisolone) and loaded into chambers on one side of the membrane; on the other side, 150 ml of the mixture described above ([1,2,6,7-3H]cortisol mixed with periplasmic extract) was added. Specific binding was determined by subtracting the amount of radioactivity in control chambers lacking unlabeled steroid from that obtained in the chambers containing steroid. Bound and free cortisol concentrations were calculated, and the equilibrium constant for binding was determined by Scatchard analysis. From the inhibition curves, concentrations allowing 50% inhibition (IC50) were determined. The percentage of cross-reaction was given by the equation: (IC50 cortisol/ IC50 analogue) 3 100. All cells were used in triplicate with three different periplasmic extracts for each scFv.

Molecular modeling The 5A4 three-dimensional model has been described previously (17). The refined Ab variable region structures were predicted using the model-building computer program INSIGHT II (Biosym Technologies, San Diego, CA). The relating structures were energy minimized using the DISCOVER computer program (Biosym Technologies) with its supplied energy parameters.

5423

Results Cloning of the anti-cortisol Ab fragment We cloned the anti-cortisol 5A4 mAb (obtained by immunization with cortisol-3-CMO-BSA conjugate) in the scFv format (VH and VL domains linked by a flexible peptide linker) to display on the surface of filamentous phage. This system allows powerful selection of Ab libraries (20). Oligonucleotides able to amplify murine VH or VL genes were first designed using the Kabat database (23). The VH and VL genes of the 5A4 mAb were amplified using appropriate primers and assembled by overlap PCR. The resulting gene was inserted into the pHEN1 phagemid vector (20) to give pHEN-5A4, which was introduced into E. coli by electroporation. The pHEN-5A4 allows fusion of the PelB signal peptide to the scFv to send the fragment in the oxidizing environment of the periplasm, leading to disulfide bond formation and proper folding. The c-Myc tag is also fused to the C-terminus of the scFv to facilitate detection. Cell fractionation and Western blot analysis revealed that only about 20% of the scFv was produced as a soluble periplasmic protein in HB2151 cells (Fig. 1A). The rest was probably aggregated as periplasmic inclusion bodies. One of the more useful features of the pHEN1 is the presence of an amber codon at the junction between the Ab sequence and that part of the gene encoding the mature capsin protein III of M13 filamentous phage. In supE cells, suppression was approximately 20% efficient, such that both soluble and pIII-linked scFv were produced (Fig. 1B). Infection of these cells harboring pHEN-5A4 by a helper phage allowed production of phage displaying the Ab fragment on their surface (called phAb). For a first approximation, affinity measurements and competitive binding assay of phAb were performed in solution by ELISA (22). In this condition, the relative affinities of wild-type 5A4 phAb toward cortisol, prednisolone, and dexamethasone were 335, 800, and 840 nM (data not shown). The relative affinities of wild-type 5A4 scFv determined by equilibrium dialysis toward the same steroids were 70, 230, and 320 nM, respectively. The large difference observed could be due to the method used and/or the nature of the Ab (phAb or scFv). For the following experiments, we decided to use the scFv format and equilibrium dialysis, which is a more rigorous method in this case. By this method the affinity of 5A4 wild-type scFv to cortisol was 7.0 nM (Fig. 3A), whereas the affinity of 5A4 Fab was 6.6 nM. Crossreactions of the wild-type scFv were similar to those of the 5A4 Fab (prednisolone: 30% for Fab, 29% for wild-type scFv; dexamethasone: 22% for Fab, 21% for wild-type scFv; see Table IV). Prediction of substitution positions The next step of this work was to create a library of scFv mutants from which those with a higher specificity for the cortisol could be selected. We chose to degenerate paratope residues in contact with or close to the hapten. The amounts of soluble and active scFv 5A4 produced by E. coli are very low, so crystallographic studies are not feasible. We therefore tried to obtain structural information from molecular models of the scFv complexed with cortisol. A computer molecular model of the scFv 5A4 was constructed as previously described (17). The position of cortisol inside the paratope was modified according to new energy minimizations and biochemical data; the six CDR loops in the molecular model form a pocket containing several aromatic amino acids (TrpH50, TrpH47, TrpL91, HisH95, TyrH33, PheH100d) that provide a hydrophobic environment around the cortisol (Fig. 4A). The structure of this pocket was similar to the three-dimensional structure of the antiprogesterone DB3 Fab fragment as determined by x-ray crystallography (24). The stacking interactions between TrpH50 and TrpH100 located on either side of the hydrophobic steroid nucleus

5424

SPECIFICITY IMPROVEMENT OF AN ANTISTEROID ANTIBODY positions from our model have been shown to be essential for high affinity binding in other Abs (Table I). Our prediction of contact amino acids was also strongly supported by the study of MacCallum et al. (25), which described positions of contact residues from 10 x-ray Ab fragment structures complexed with their hapten (Table II). Indeed, almost all the positions we chose are described as highly frequent contact positions, with the exception of position H100d, which is only present in long CDR H3. Mutagenesis and building of the mutant library

FIGURE 3. Determination of the affinities of wild-type 5A4 scFv and AsnL94Ser scFv. A, ScFv affinities were determined by equilibrium dialysis and Scatchard analysis (‚, wild-type; f, AsnL94Ser). B, The relative affinities of scFv were determined by equilibrium dialysis in which binding to cortisol was inhibited by the analogues (‚, wild type with cortisol; f, l, and F, AsnL94Ser with cortisol, prednisolone, and dexamethasone, respectively). The IC50 is the concentration of steroid necessary to obtain 50% of the control signal (that without any competition).

in the DB3 paratope appear to correspond to TrpH50 and TrpL91 in our model. mAb 5A4 has a high affinity for cortisol-3-CMO-BSA (the immunogen) but not for cortisol-21-hemisuccinate-BSA. This and the paratope conformation strongly suggest that the D ring (containing carbon 21) is buried in the back of the cavity and that the A ring (containing carbon 3) is located outside the binding pocket. Moreover, 11-deoxycortisol was not recognized by mAb 5A4, suggesting that the 11-hydroxyl group may interact with a paratope residue and should be orientated toward the cavity. On the contrary, scFv 5A4 also recognized cortisol-7-carboxymethyl-BSA, suggesting that C7, in the B ring, is exposed to the solvent. From these biochemical data, cortisol was positioned with its a-face interacting with TrpL47 and TrpL50 and the b-face, which possesses the two methyl groups, interacting with TrpL91. The 21-hydroxyl group of cortisol might form a hydrogen bond with HisH95 or ThrH100b, and the 3-ketone and 11-hydroxyl group might interact with AsnL94 and ProL96, respectively (Fig. 4A). The model predicts that residues TyrH33, SerH35, AsnH58, and PheH100d are also close enough to cortisol to interact. We compared these results with published findings following mutagenesis of some of the paratope residues and affinity measurements to identify positions involved in hapten recognition. These studies are consistent with our model. Many of the contact

For practical reasons, it is difficult to obtain libraries of .108 clones (transformation efficiency being the major limiting factor), which means that the number of degenerate positions (where an amino acid is substituted by one of the 20 other residues) must be limited to 6 (206 5 6.4 3 107). Eleven positions in mAb 5A4 were predicted to be potentially in interaction with the hapten (Table II). We thus had to choose five or six mutagenesis targets among these residues. TrpH50 and TrpL91 are thought to interact by hydrophobic stacking (4, 26) with the steroid nucleus. TrpH47 is also involved in interactions between VH and VL domains. These three residues were thus conserved. Among the eight remaining residues, we decided to restrict the mutagenesis to residues belonging to CDR H3 and L3 (HisH95, ThrH100b, PheH100d, AsnL94, and ProL96) because these CDRs form the center of the paratope and contain positions that have most frequently been shown to be in contact with the Ag, especially in the case of haptens (25). The nature of these targeted residues is probably very important for the binding activity, and multiple mutations per Ab would probably create a large number of functionally useless Abs. To avoid this problem, we used parsimonious mutagenesis (27, 28). The principle of this method is to use mutagenesis codons leading to about 50% amino acids at each degenerated positions being wild type. This has several advantages. First, the total number of substitutions per Ab is reduced, resulting in a larger number of well-folded and potentially active binders in the library. Secondly, if a residue chosen as mutagenesis target is essential for binding, classical mutagenesis methods will frequently substitute it for one of the 20 others, leading to a library in which 95% of the clones are inactive. If two crucial residues are chosen, 99.75% of the library clones will be inactive. In the same case, parsimonious mutagenesis allows the conservation of crucial amino acid in about 50% of clones for one mutation and 25% for two mutations. Nevertheless, each substitution at each mutagenesis position is well represented: the least represented of all substitutions (0.1%) will still be present at 104 copies in a 107 clone library (Table III). This favors the selection of substitutions leading to a stronger affinity or specificity. The library was built using overlap PCR to link four fragments, two of them incorporating the mutations in the CDR H3 or L3 (Fig. 2). Ligation of large amounts of DNA and 50 electroporations allowed the construction of 2.5 3 107 clones. The library quality was then checked using DNA restriction analysis and Western blot analysis of whole cell extract after induction, using the anti-c-Myc tag Ab. Ninety-six percent (46 of 48) of the clones had an insert of the right size, and 29% (14 of 48) gave a detectable signal by Western blots (data not shown). Selection of clones with improved characteristics As a control, a first selection was performed in immunotubes coated with cortisol-3-CMO-BSA. After binding and washing, the bound phAb were eluted using a classical elution method by pH shock (100 mM triethylamine, pH 12.0). After the second round of amplification and selection, we observed a 10-fold amplification of the ratio of output phAb to input phAb. Forty-eight clones were tested for binding to cortisol-3-CMO-BSA, and 43 of 48 were positive. Sequencing of 18 clones indicated that two kinds of phAb

The Journal of Immunology

5425

FIGURE 4. Model of the 5A4 anti-cortisol paratope and structure of cortisol and its analogues. A, Model of 5A4 paratope complexed with cortisol. Amino acids thought to interact with the hapten are labeled. Cortisol is placed inside the cavity; carbon 7 and oxygens 3, 11, and 21 are labeled. The 11-hydroxyl function is close to the residue ProH96. Residues TrpL91 and TrpH50 are thought to interact by stacking with the hydrophobic nucleus of cortisol. B, Comparison of structure between cortisol and its cross-reacting analogues. Arrows indicate the differences: a double bond between carbons 1 and 2 (for both analogues), a fluorine atom on carbon 9, and a methyl group on carbon 16 (for dexamethasone).

had been selected. The first type (11 of 18) had the amino acid sequence of the wild-type Ab. The third base of each of the relevant codons was different from that of the wild-type as expected from the sequence of primers used for the mutagenesis. The other sequences contained a deletion in the scFv gene, leading to the display of truncated VH domains. By ELISA we showed that these clones were able to bind any steroid conjugated to the BSA (probably by a nonspecific hydrophobic interactions) and could not be displaced by free steroid in a competitive binding ELISA (data not shown). To avoid the selection of such clones, we subsequently eluted the specifically bound phAb with a buffer containing free cortisol. Thus, phAb bound by nonspecific interaction are less likely to be recovered and thus amplified. We used lower concentrations of cortisol during the successive rounds of amplification/ selection/elution to favor the recovery of phAb with high affinity for free cortisol (see Materials and Methods). Two new clones emerged from this selection. Both of them had a single mutation at codon L94 (CDR L3). The wild-type asparagine was replaced by a serine or a threonine. Both of these clones had a higher affinity Table I. Residues shown to be in contact with haptens by site-directed mutagenesis studies and affinity determinations Crucial Residuesa

H33 H33 H33 L32 L91 L96 H33 L32 L91 L96 H35 H50 H99 H105 L96 H50 H35 H47 H50 H95 H100b L96 H35 H95 L32 L91 L96 H35 H33 H35 H33 L32 L91 L92 L96 H55 H95 H97 L96 H33 H35 H50 H100 L94 L96 a

for cortisol (Table IV). The affinities determined by equilibrium dialysis and Scatchard analysis for the wild-type scFv and the AsnL94Ser scFv mutant were 7.0 and 2.8 nM, respectively (Fig. 3A). Interestingly, the relative affinity for analogues was also higher but to a lesser extent, leading to a reduction of cross-reaction by a factor of 2 (Table IV). The relative affinity determined by equilibrium dialysis of the AsnL94Ser scFv mutant for analogues (prednisolone and dexamethasone) is shown in Fig. 3B. Use of competitive binding and pre-elution in presence of analogues We thought that we could select more specific binders by blocking cross-reacting phAb with analogues before the selection with cortisol. The specific elution was conserved in the following selections, but before selection on a solid surface coated with cortisol-3 CMO-BSA, the phAb library was incubated in a buffer containing Table II. Amino acids predicted by molecular modeling to be involved in the anti-cortisol binding site Positiona

Localization

Residue

Contact Frequencyb

H33 H35

CDR H1

Tyr Ser

6/10 6/10

H47 H50 H58

CDR H2

Trp Trp Asn

2/10 5/10 2/10

H95 H100b H100dc

CDR H3

His Thr Phe

7/10 3/10 0/10

L91 L94 L96

CDR L3

Trp Asn Pro

8/10 7/10 10/10

Ref.

37 38 39 40 41 42 43 44 15 45 46 35 47

Positions involved in contact with cortisol in the anti-cortisol model are in bold.

a

Positions chosen for the parsimonious mutagenesis are in bold. Number of contacts made at each chosen position determined by the study of 10 x-ray structures of anti-hapten antibodies (25). c This position is only present in larger CDR H3. b

5426

SPECIFICITY IMPROVEMENT OF AN ANTISTEROID ANTIBODY free cortisol. After four rounds of amplification/selection, the use of prednisolone as analogue led to the isolation of a single clone, as shown by sequencing. This clone had one mutation in the CDR H3 (ThrH100bAla). The affinity of this scFv for cortisol (6.8 nM) was similar to the affinity of the wild-type scFv (7.0 nM). However, its relative affinity for prednisolone was reduced (Table IV). We obtained similar results with selection in the presence of dexamethasone. A single clone was obtained with a mutation in the CDR H3 (PheH100dTyr). The affinity of the scFv for cortisol (7.2 nM) was not significantly different from that of the wild-type Ab, but the relative affinity for dexamethasone was nearly halved, leading to a 13% cross-reaction instead of 21% in the case of the wild-type scFv (Table IV). For both mutants, the relative affinity for the second analogue was also reduced.

Table III. Predicted frequencies of amino acids at each position subjected to parsimonious mutagenesis VH CDR3 H95

H

Ala Cys Asp Glu Phe Gly His Ile Lys Leu Met Asn Pro Gln Arg Ser Thr Val Trp Tyr Stopb

H100b

T

VL CDR3 F

H100d

N

L94

PL96

123a

453a

663a

223a

553a

0.49 0.39 4.37 1.16 0.39 0.49 49.30 0.39 1.16 5.63 0.10 4.37 5.53 13.11 5.63 0.88 0.49 0.49 0.10 4.37 1.16

7.00 0.79 0.79 0.21 0.79 1.00 0.79 5.53 1.47 1.21 1.47 5.53 7.00 0.21 2.47 12.53 49.00 1.00 0.21 0.79 0.21

0.49 4.37 0.39 0.10 49.30 0.49 0.39 4.37 0.10 18.64 1.16 0.39 0.49 0.10 0.59 5.92 0.49 5.53 1.16 4.37 1.16

0.49 0.39 4.37 1.16 0.39 0.49 4.37 4.37 13.11 0.59 1.16 49.30 0.49 1.16 1.65 4.86 5.53 0.49 0.10 4.37 1.16

7.00 0.79 0.79 0.21 0.79 1.00 5.53 0.79 0.21 7.21 0.21 0.79 49.00 1.47 7.21 7.79 7.00 1.00 0.21 0.79 0.21

Double mutants

a Nucleotide mixtures for parsimonious mutagenesis primers used for construction of the phAb library (1: A 7%, C 79%, G 7%, T 7%; 2: A 79%, C 7%, G 7%, T 7%; 3: G 21%, T 79%; 4: A 70%, C 10%, G 10%, T 10%; 5: A 10%, C 70%, G 10%, T 10%; 6: A 7%, C 7%, G 7%, T 79%). Predicted frequencies of amino acids are indicated by numerals b Indicates the frequency of stop codons.

the analogue. Analog concentrations were increased at each round from 1 to 50 mM (see Materials and Methods). After binding to the solid surface (in the presence of analogues) and washing, a preelution was performed with 50 mM analogue in washing buffer to remove all phAb still able to bind the analogue efficiently. The bound phAb was then recovered by specific elution with 50 mM

To determine whether the observed effects of the mutations were additive, we combined the mutation AsnL94Thr independently with ThrH100bAla or PheH100dTyr. As a control the mutation AsnL94Thr was also introduced into the wild-type scFv, and the results obtained were undistinguishable from those obtained with binders selected by competitive elution (data not shown). Both clones Asn L94 Thr/ThrH100bAla and Asn L94Thr/PheH100dTyr had a very high affinities for cortisol (0.9 and 1.6 nM) and were more specific than the single mutant AsnL94Thr for prednisolone and dexamethasone (Table IV).

Discussion What is the limit of the exquisite specificity of mAbs? Steroids are a simple and useful model for answering this question. The structural difference between two steroids can be very small. For example, estradiol and estrone differ from one another by the presence of a hydroxyl or ketone group, respectively, at C17; cortisol is obtained from its precursor, the 11-deoxycortisol, by hydroxylation at C11. We chose to work with an anti-cortisol Ab that mainly cross-reacts with two steroid analogues, prednisolone and

Table IV. Affinities and specificities of wild-type and mutant scFv

Prednisolonea

scFv

Fabe Wild type AsnL94Ser AsnL94Thr ThrH100bAla PheH100dTyr AsnL94Thr, ThrH100bAla AsnL94Thr, PheH100dTyr a

Dexamethasonea

Cortisola Affinityb (nM)

Cortisol IC50c (nM)

IC50c (nM)

Cross-reactiond (%)

IC50c (nM)

Cross-reactiond (%)

6.6 7.0 2.8 1.3 6.8 7.2 0.9

70 75 40 20 70 80 10

230 260 170 130 370 420 110

30 29 24 15 19 19 9

320 350 260 210 460 600 150

22 21 15 10 15 13 7

1.6

20

320

6

450

4

Reported results are the means of three independent experiments made in triplicate. Affinities for cortisol were determined in solution by equilibrium dialysis and Scatchard analysis. The margin of error, given for affinities for cortisol, never exceeded 10%. c The relative affinities were determined by equilibrium dialysis in which binding to cortisol was inhibited by free analog. The IC50 is the concentration of steroid necessary to obtain 50% of the control signal (that without any competition). d The percentage of cross-reactions was calculated as follow: (IC50 cortisol/IC50 analog) 3 100. e The values are reported for the Fab obtained from the 5A4 mAb after papaine proteolyse. b

The Journal of Immunology dexamethasone. The structural differences between these analogues and cortisol are few. For instance, in the case of prednisolone the only difference is the presence of a double bond between the carbons 1 and 2 (Fig. 4B). Obviously it is very difficult, if not impossible, to obtain by classical hybridoma methods and a simple screen an mAb capable of discriminating between these two molecules. Consequently, we investigated whether it was possible to tailor the paratope of an Ab to suppress the unwanted cross-reactions. In the absence of an x-ray structure of the Ab, we decided to use a strategy to improve its characteristics by randomization of few positions chosen with the help of molecular modeling (29, 30) and a powerful selection by phage display. Ab structures are very conserved, and even hypervariable loop folding can be efficiently predicted from the primary sequence (31). Schiweck et al. (32) succeeded in designing a completely new paratope using a molecular model, and subsequent x-ray crystallographic analysis confirmed their model. It should be emphasized that the aim of our model was only to identify amino acid positions close to the steroid. Determining the side chain orientation, the most difficult task in molecular modeling, was not very important in our case. Furthermore, the choice of mutagenesis targets was strongly supported by an analysis of biochemical and structural data for Ab-hapten complexes. We decided to substitute only five amino acids of the paratope to limit the theoretical diversity. After cloning and electroporation we obtained a library of 2.5 3 107 clones. The theoretical diversity (205 5 3.2 3 106) was thus less than the number of clones. Only about 30% of clones produced an scFv as assessed by Western blot analysis. Similar results have already been obtained (15, 33). This low yield of expressing clones may be due to the successive PCR amplifications leading to incorporation of deletions and stop codons, for example. Nevertheless, the number of producing clones was still higher than the diversity. Selecting for anti-cortisol Abs in this library in the absence of any selection pressure, we obtained clones that have conserved the original amino acid sequence. This is not surprising, since in our library, 3% of clones are theoretically wild type and can obviously bind to the cortisol. However, we also selected clones lacking the VL domain plus a part of the VH domain. These clones are expected to present a hydrophobic interface that is usually packed with the VL interface. This could be responsible for the nonspecific binding observed with these clones, leading to their selection. Surprisingly, these deleted clones are very well expressed in E. coli and thus have a selective advantage over the entire scFv-producing clones during each amplification (data not shown). To avoid the selection of such “sticky” clones, we eluted phAb with a buffer containing free cortisol (50 mM). We thereby isolated mutant phAb with a higher affinity for cortisol (2.5-fold for the mutation AsnL94Ser and 7.8-fold for the mutation AsnL94Thr). These clones carried mutations at the position L94, a residue close to the ketone function on position C3 of both cortisol and analogues in the molecular model. These clones may interact with this ketone group, available only on the free cortisol (but absent in the cortisol-3CMO-BSA used for immunization and selection). Indeed, both selected residues (Ser and Thr) have a hydroxyl group able to form a strong hydrogen bond. Interestingly, the affinities of the different clones (wild-type , AsnL94Ser , AsnL94Thr) seem to correlate with the strength of this putative bond (amide-ketone , primary alcohol-ketone , secondary alcohol-ketone). This interaction can be also found in the analogues, but in these steroids, the electrons of the ketone group are delocalized by a double bond between carbons 1 and 2 of the A ring, leading to a weaker putative interaction with the Ser or Thr residue. This could explain why these mutants cross-reacted up to twice less strongly with the analogues. It should be emphasized that these clones were isolated by com-

5427 petitive elution with free cortisol. This elution step is not involved in protocols allowing the recovery of mAbs, and, conversely to phAb, selection of mAbs interacting with the group used to link the carrier protein is highly improbable, thereby leading to Abs with lower affinities and specificities. Selection in the presence of analogues led to the isolation of two mutants, ThrH100bAla and PheH100dTyr, with prednisolone or dexamethasone, respectively. Both these clones have a affinity for cortisol similar to that of the wild-type scFv. However, the relative affinity of these clones for analogues is lower than that of the wild-type Ab (Table IV). Note that competitive elution, selecting good cortisol binders, leads effectively to clones with a higher affinity for cortisol, whereas competitive selection, selecting clones unable to bind analogues efficiently, leads to clones with a wildtype affinity for cortisol but with a lower relative affinity for the analogues. The mutation PheH100dTyr reduces only slightly the affinity for cortisol. Nevertheless, it seems that this mutation causes a steric hindrance that has a larger effect on analogue binding than on cortisol binding. When this mutation was combined with the mutation AsnL94Thr, we observed the same effect; the affinity for cortisol was sightly modified (from 1.3 to 1.6 nM), and the relative affinity for analogues was reduced by a factor of .2 (from 130 to 320 nM for prednisolone and from 210 to 450 nM for dexamethasone). The effects of mutations AsnL94Thr and PheH100dTyr were identical in both single- and double-mutant contexts, and their effects were additive. Consequently, the cross-reactions of this double mutant for prednisolone and dexamethasone were fivefold lower than that for the wild-type Ab. Mutation ThrH100bAla also reduced the relative affinity for analogues but more weakly than mutation PheH100dTyr. However in this case, the effect of this mutation was completely different when combined with the mutation AsnL94Thr. The double mutant AsnL94Thr/ThrH100bAla had higher relative affinities for analogues than the single mutant AsnL94Thr. The affinity for cortisol was also increased. Thus, in this case it seems that the presence of mutation ThrH100bAla increases the effect of mutation AsnL94Thr. The effects of these mutations were not additive, but there was nevertheless an interaction between the two mutations. Interestingly, substitution ThrH100bAla alone reduced the relative affinity for both analogues but to a lesser extent in the case of dexamethasone (6% instead of 10% for prednisolone), and this was also the case for the double mutant AsnL94Thr/ThrH100bAla (3% for dexamethasone instead of 6% for prednisolone; Table IV). Although this effect is difficult to explain (because the only difference between prednisolone and cortisol, the double bond between the carbons 1 and 2, is also present in dexamethasone), it is interesting to note that this clone was selected for a lower relative affinity toward prednisolone but not toward dexamethasone. From this point of view, these observations are logical. Is it possible to describe a model explaining these findings? The results for mutations AsnL94Ser or AsnL94Thr suggest the creation of a new hydrogen bond between the ketone groups on the A ring of steroids and the substituted amino acids. The effects of mutations selected in the presence of analogues are more difficult to explain. Both mutations decrease relative affinity for prednisolone and dexamethasone. In our model, positions H100b and H100d are near the steroid D ring. The D ring of prednisolone is identical with that in cortisol. Therefore, a new direct interaction involving this D ring is improbable. However, the tyrosine hydroxyl group available on the mutant may interact with another paratope residue, inducing a small rearrangement or maintaining CDR loop conformations through this atomic interaction. The hindrance caused by this change may be stronger for the analogues and thereby lead to reduced cross-reactions. This kind of paratope modification has

5428

SPECIFICITY IMPROVEMENT OF AN ANTISTEROID ANTIBODY

been demonstrated in the case of an anti-hapten affinity maturation (34 –36). Wedemayer et al. (36) demonstrated by crystallographic studies that none of nine somatic mutations responsible for a large increase of affinity (30,000-fold) was situated in the hapten binding site; the mutations are all responsible for rearrangements and stabilization of CDR positions. The case of the mutation ThrH100bAla is different because this mutation affects the mutation AsnL94Thr effect. This result may be explain by a modification of the steroid position in the binding site, due to the local modification of the paratope near the steroid D ring, induced by the substitution. The new position of the hapten may cause steric hindrance for the analogues but not for the cortisol. However, in the double mutant the new position adopted by analogues may allow a stronger interaction between the Thr residue and the 3-ketone group on the steroid A ring, enhancing the effect of the AsnL94Thr mutation. It would be very interesting to compare the crystallographic structures of the two double mutants complexed with cortisol as this may help elucidate the effects of the mutations at a molecular level. Interestingly, all the selected clones have wild-type amino acids at positions L96 (Pro) and H95 (His). These residues are probably essential for cortisol binding or efficient folding of the Ab. In our model, the proline L96, a position always found in interaction with the hapten in 10 x-ray structures of Ab-hapten complexes (25), may interact with the 11-hydroxyl function of cortisol and its analogues. This interaction is very important, since the cortisol analogue lacking this function (11-deoxycortisol) is not recognized by mAb 5A4. This would explain the conservation of this residue in this study. Similarly, the conservation of the HisH95 residue may have been due to an essential interaction with the cortisol D ring. The conservation of these two residues in all selected clones emphasizes the importance of using parsimonious mutagenesis for this kind of study. Which of these new anti-cortisol Ab fragments is the best for diagnostic purposes? In normal humans, cortisol levels exhibit a diurnal rhythm, and the plasma concentration varies between 130 – 650 nM (1). For the cortisol immunoassay, affinities between 1–10 nM are well suited. The best Ab is thus the most specific, which is Ab AsnL94Thr/PheH100dTyr. In conclusion, we show that using guided randomization for diversity generation and phage display for selection, it is possible to obtain anti-hapten Abs with improved affinity and specificity even if there is only a very small difference between the Ag and its analogues (Fig. 4B). The specificity could probably be further increased by conserving these mutations and creating secondary libraries with randomization of other contact positions. An alternative solution is to mutate amino acids responsible for the position of entire CDR, thereby allowing small changes in the paratope cavity, which could be useful for the fine tuning of specificity. Finally, this work demonstrates that the development of recombinant Abs may be a solution of choice for obtaining the highly specific Abs required for steroid immunoassays.

Acknowledgments We thank Drs. C.-Y. Cuilleron and E. Mappus for helpful discussion, Prof. M. Delaage and Dr. J. Chauveau for the gift of the hybridoma 5A4 and helpful discussion, M. Chartier for technical assistance, Dr. E. Loret for help in molecular modeling, Drs. D. Duche´ and V. Ge´li for the critical reading of this manuscript, and Dr. S. Hufton for critical comments and help in preparing the manuscript.

References 1. Milgrom, E. 1990. Steroid hormones. In Hormones: From Molecules to Disease, E. E. Baulieu and P. A. Kelly PA, eds. Hermann, Paris, p. 385.

2. Yoon, D. Y., M. J. Choi, I. S. Choe, T. W. Chung, and S. M. Byun. 1993. Influence of the conjugation site on the specificity of monoclonal antibodies to progesterone and on the performance of direct enzyme immunoassay. Biochem. Mol. Biol. Int. 31:553. 3. Gani, M., J. Coley, J. Piron, A. S. Humphreys, J. Arevalo, I. A. Wilson, and M. J. Taussig. 1994. Monoclonal antibodies against progesterone: effect of steroid-carrier coupling position on antibody specificity. J. Steroid Biochem. Mol. Biol. 48:277. 4. Arevalo, J. H., C. A. Hassig, E. A. Stura, M. J. Sims, M. J. Taussing, and I. A. Wilson. 1994. Structural analysis specificity: detailed comparison of five Fab9-steroid complexes. J. Mol. Biol. 241:663. 5. Barkley, M. S., B. L. Lasley, M. A. Thompson, and C. H. Shackleton. 1995. Equol: a contributor to enigmatic immunoassay measurements of estrogen. Steroids 46:587. 6. Agasan, A. L., B. J. Stewart, and T. G. Watson. 1994. Development of a radioimmunoassay method for ethynylestradiol in plasma using a monoclonal antibody. J. Immunol. Methods 177:251. 7. Wudt, S. A., U. A. Wachter, J. Homoki, and W. M. Teller. 1995. 17a-Hydroxyprogesterone, 4-androstenedione, and testosterone profiled by routine stable isotope dilution/gas chromatography-mass spectrometry in plasma of children. Pediatr. Res. 38:76. 8. Roberts, S., J. C. Cheetham, and A. R. Rees. 1987. Generation of an antibody with enhanced affinity and specificity for its antigen by protein engineering. Nature 328:731. 9. Sharon, J. 1990. Structural correlates of high antibody affinity: three engineered amino acid substitutions can increase the affinity of anti-p-azophenylarsonate antibody 200-fold. Proc. Natl. Acad. Sci. USA 87:4814. 10. Kussie, P. H., B. Parhami-Seren, L. J. Wysocki, and M. N. Margolies. 1994. A single engineered amino acid substitution changes antibody fine specificity. J. Immunol. 152:146. 11. Marks, J. D., H. R. Hoogenboom, A. D. Griffiths, and G. Winter. 1992. Molecular evolution of proteins on filamentous phage: mimicking the strategy of the immune system. J. Biol. Chem. 267:16007. 12. Glaser, S. M., D. E. Yelton, and W. D. Huse. 1992. Antibody engineering by codon-based mutagenesis in a filamentous phage vector system. J. Immunol. 149: 3903. 13. Deng, S. J., C. R. MacKenzie, J. Sodowska, J. Michniewicz, N. M. Young, D. R. Bundle, and S. A. Narang. 1994. Selection of antibody single-chain variable fragments with improved carbohydrate binding by phage display. J. Biol. Chem. 269:9533. 14. Barbas, C. F., D. Hu, N. Dunlop, L. Sawyer, D. Cababa, R. M. Hendry, P. L. Nara, and D. R. Burton. 1994. In vitro evolution of a neutralizing human antibody to human immunodeficiency virus type 1 to enhance affinity and broaden strain cross-reactivity. Proc. Natl. Acad. Sci. USA 91:3809. 15. Casson, L. P., and T. Manser. 1995. Evaluation of loss and change of specificity resulting from random mutagenesis of antibody VH region. J. Immunol. 155: 5647. 16. Schier, R., A. McCall, G. P. Adams, K. W. Marshall, H. Merritt, M. Yim, R. S. Crawford, L. M. Weiner, C. Marks, and J. D. Marks. 1996. Isolation of picomolar affinity anti-c-erbB-2 single-chain Fv by molecular evolution of the complementarity determining regions in the center of the antibody binding site. J. Mol. Biol. 263:551. 17. Le Calvez, H., J. Fieschi, J. M. Green, N. Marchesi, J. Chauveau, and D. Baty. 1995. Paratope characterization by structural modelling of two anti-cortisol single-chain variable fragments produced in E. coli. Mol. Immunol. 32:185. 18. Whitlow, M., B. A. Bell, S. L. Feng, D. Filpula, K. D. Hardman, S. L. Hubert, M. L. Rollence, J. F. Wood, M. E. Schott, D. E. Milenic, T. Yokota, and J. Schlom. 1993. An improved linker for single-chain Fv with reduced aggregation and enhanced proteolytic stability. Protein Eng. 6:989. 19. Horton, R. M., H. D. Hunt, S. N. Ho, J. K. Pullen, and L. R. Pease. 1989. Engineering hybrid genes without the use of restriction enzymes: gene splicing by overlap extension. Gene 77:61. 20. Hoogenboom, H. R., and G. Winter. 1992. By-passing immunisation: human antibodies from synthetic repertoires of germline VH gene segments rearranged in vitro. J. Mol. Biol. 227:381. 21. Pluckthun, A., and A. Skerra. 1989. Expression of functional antibody Fv and Fab fragments in Escherichia coli. Methods Enzymol. 178:497. 22. Friguet, B., A. F. Chaffotte, L. Djavadi-Ohaniance, and M. E. Goldberg. 1985. Measurements of the true affinity constant in solution of antigen-antibody complexes by enzyme-linked immunosorbent assay. J. Immunol. Methods 77:305. 23. Kabat, E. A., T. T. Wu, H. M. Perry, K. S. Gottesman, and C. Foeller. 1991. Sequences of Proteins of Immunological Interest, 5th Ed. U.S. Department of Health and Human Services, Bethesda. 24. Arevalo, J. H., M. J. Taussig, and I. A. Wilson. 1993. Molecular basis of crossreactivity and the limits of antibody-antigen complementarity. Nature 365:859. 25. MacCallum, R. M., A. C. Martin, and J. M. Thornton. 1996. Antibody-antigen interactions: contact analysis and binding site topography. J. Mol. Biol. 262:732. 26. Parhami-Seren, B., P. H. Kussie, R. K. Strong, and M. N. Margolies. 1993. Conservation of binding site geometry among p-azophenylarsonate-specific antibodies. J. Immunol. 150:1829. 27. Balint, R. F., and J. W. Larrick. 1993. Antibody engineering by parsimonious mutagenesis. Gene 137:109. 28. Schier, R., R. F. Balint, A. McCall, G. Apell, J. W. Larrick, and J. D. Marks. 1996. Identification of functional and structural amino-acid residues by parsimonious mutagenesis. Gene 169:147. 29. Eigenbrot, C., M. Randal, L. Presta, P. Carter, and A. A. Kossiakoff. 1993. X-ray structures of the antigen-binding domains from three variants of humanized anti-

The Journal of Immunology

30.

31.

32.

33.

34.

35.

36.

37.

38.

p185HER2 antibody 4D5 and comparison with molecular modeling. J. Mol. Biol. 229:969. Barry, M. M., C. D. Mol, W. F. Anderson, and J. S. Lee. 1994. Sequencing and modeling of anti-DNA immunoglobulin Fv domains: comparison with crystal structures. J. Biol. Chem. 269:3623. Chothia, C., A. M. Lesk, A. Tramontano, M. Levitt, S. J. Smith-Gill, G. Air, S. Sheriff, E. A. Padlan, D. Davies, W. R. Tulip, et al. 1989. Conformations of immunoglobulin hypervariable regions. Nature 342:877. Schiweck, W., and A. Skerra. 1997. The rational construction of an antibody against cystatin: lessons from the crystal structure of an artificial Fab fragment. J. Mol. Biol. 268:934. Vaughan, T. J., A. J. Williams, K. Pritchard, J. K. Osbourn, A. R. Pope, J. C. Earnshaw, J. MacCafferty, R. A. Hodits, J. Wilton, and K. S. Johnson. 1996. Human antibodies with sub-nanomolar affinities isolated from a large non-immunized phage display library. Nat. Biotechnol. 14:309. Lavoie, T. B., W. N. Drohan, and S. J. Smith-Gill. 1992. Experimental analysis by site-directed mutagenesis of somatic mutation effects on affinity and fine specificity in antibodies specific for lysozyme. J. Immunol. 148:503. Xiang, J., Y. Sha, L. Prasad, and L. T. Delbaere. 1996. Complementarity determining region residues aspartic acid at H55, serine at H95 and tyrosines at H97 and L96 play important roles in the B72.3 antibody-TAG72 antigen interaction. Protein Eng. 9:539. Wedemayer, G. J., P. A. Patten, L. H. Wang, P. G. Schultz, and R. C. Stevens. 1997. Structural insights into the evolution of an antibody combining site. Science 276:1665. Allen, D., T. Simon, F. Sablitzky, K. Rajewsky, and A. Cumano. 1988. Antibody engineering for the analysis of affinity maturation of an anti-hapten response. EMBO J. 7:1995. Glockshuber, R., J. Stadlmuller, and A. Pluckthun. 1991. Mapping and modification of an antibody hapten binding site: a site-directed mutagenesis study of McPC603. Biochemistry 30:3049.

5429 39. Mian, I. S., A. R. Bradwell, and A. J. Olson. 1991. Structure, function and properties of antibody binding sites. J. Mol. Biol. 217:133. 40. Denzin, L. K., G. A. Gulliver, and E. W. Voss, Jr. 1993. Mutational analysis of active site contact residues in anti-fluorescein monoclonal antibody 4-4-20. Mol. Immunol. 30:1331. 41. Ruff-Jamison, S., and J. R. Glenney. 1993. Molecular modeling and site-directed mutagenesis of an anti-phosphotyrosine antibody predicts the combining site and allows the detection of higher affinity interactions. Protein Eng. 6:661. 42. Schildbach, J. F., R. I. Near, R. E. Bruccoleri, E. Haber, P. D. Jeffrey, S. C. Ng, J. Novotny, S. Sheriff, and M. N. Margolies. 1993. Heavy chain position 50 is a determinant of affinity and specificity for the anti-digoxin antibody 26-10. J. Biol. Chem. 268:21739. 43. Sompuram, S. R., and J. Sharon. 1993. Verification of a model of a F(ab) complex with phenylarsonate by oligonucleotide-directed mutagenesis. J. Immunol. 150:1822. 44. Stewart, J. D., V. A. Roberts, N. R. Thomas, E. D. Getzoff, and S. J. Benkovic. 1994. Site-directed mutagenesis of a catalytic antibody: an arginine and a histidine residue play key roles. Biochemistry 33:1994. 45. Short, M. K., P. D. Jeffrey, R. F. Kwong, and M. N. Margolies. 1995. Contribution of antibody heavy chain CDR1 to digoxin binding analyzed by random mutagenesis of phage-displayed Fab 26-10. J. Biol. Chem. 270:28541. 46. Sotriffer, C. A., K. R. Liedl, R. H. Winger, A. M. Gamper, R. T. Kroemer, D. S. Linthicum, B. M. Rode, and J. M. Varga. 1996. Heteroligation of a mouse monoclonal IgE antibody (La2) with small molecules, analysed by computeraided automated docking. Mol. Immunol. 33:129. 47. Burks, E. A., G. Chen, G. Georgiou, and B. L. Iverson. 1997. In vitro scanning saturation mutagenesis of an antibody binding pocket. Proc. Natl. Acad. Sci. USA 94:412.