Phage Display-selected Sequences of the Heavy-chain CDR3 Loop of the Anti-digoxin Antibody 26-10 Define a High Affinity Binding Site for Position 16-substituted Analogs of Digoxin*

The heavy-chain CDR3 region of the high affinity (K a = 1.3 × 1010 m − 1) anti-digoxin monoclonal antibody 26-10 was modified previously to shift its specificity, by substitution of tryptophan 100 by arginine, toward binding analogs of digoxin containing substitutions at position 16. To further change specificity, two 5-mer libraries of the randomly mutagenized phage-displayed 26-10 HCDR3 region (positions 94–98) were panned against digoxin-bovine serum albumin (BSA) as well as against 16-acetylgitoxin-BSA. When a mutant Fab that binds 16-substituted analogs preferentially was used as a parent sequence, clones were obtained with affinities for digoxin increased 2–4-fold, by panning on digoxin-BSA yet retaining the specificity shift. Selection on 16-acetylgitoxin-BSA, however, resulted in nine clones that bound gitoxin (16-OH) up to 150-fold higher than the wild-type 26-10, due to a consensus mutation of SerH95 to GlyH95. The residues at both position H95 (serine) and position H100 (tryptophan) contact hapten in the crystal structure of the Fab 26-10-digoxin complex. Thus, by mutating hapten contact residues, it is possible to reorder the combining site of a high affinity antibody, resulting in altered specificity, yet retain or substantially increase the relative affinity for the cross-reactive ligand.

The utility of antibodies (Abs) 1 as therapeutic agents or diagnostic reagents depends upon the degree of specificity and affinity for the target ligand or protein antigen and/or a lack of cross-reactivity with related antigens. Monoclonal Abs produced by somatic cell fusion from elicited Abs, commonly murine, are often not monospecific, and in some instances Abs against certain antigens are not easily elicited. Immuni-zation of humans for purposes of obtaining a therapeutic Ab is not normally feasible. The ability to clone genes encoding Ab binding fragments (Fab or Fv) into filamentous bacteriophage vectors, which incorporate foreign DNA fused to the bacteriophage coat proteins pIII, pVI, or pVIII and which express functional Ab-combining sites on their surface (1)(2)(3)(4)(5)(6) has revolutionized antibody engineering. Rare phage-carrying V genes encoding a desired specificity are selected directly with antigen by panning and enriched by several cycles of affinity selection. The sequence of the cloned DNA associated with the target specificity can be obtained directly on phagemid DNA. This approach has largely been used to select Abs of desired specificity against clinically important protein antigens from naive human V gene combinatorial libraries (7,8). Alternatively, Abs are selected from libraries obtained from naive or preimmunized mouse lymphocytes; the resultant combining site sequences, the "complementarity-determining" regions of the variable region, can then be engineered into "human frameworks" attached to human constant regions to render them less immunogenic for therapeutic use (9 -12). The affinities of Abs of desired specificity thus recruited from these libraries may be increased by mutagenesis in complementarity-determining regions, in particular HCDR3, by "chain shuffling" (13)(14)(15)(16) and by random variable region mutagenesis.
Antibody specificity has been altered by saturation mutagenesis of complementarity-determining regions or segments thereof in phage-displayed Abs selected using cross-reactive antigens (17)(18)(19)(20). In occasional instances, site-directed mutagenesis has been successful (21). Unwanted specificities for cross-reacting steroids, for example, which interfere in clinical immunoassays, were decreased or removed using saturation mutagenesis of binding site segments in anti-testosterone Abs displayed on bacteriophage (22).
We have been using digoxin-specific Abs as a model system for the study of the interaction of Ab-combining sites with "small" ligands. Digoxin (digoxigenin tridigitoxose; see Fig. 1), when coupled to proteins, elicits Abs of high affinity and varying specificity for cardiac glycoside analogs of defined structure (23,24). The structure of the Fab fragment of the monoclonal murine Ab 26-10 (K a ϭ 1.3 ϫ 10 10 M Ϫ1 ) was determined, as well as the structure of the 26-10 Fab:digoxin complex (25). 2 The high affinity binding is due virtually exclusively to shape complementarity with this hydrophobic, relatively rigid hapten. In particular, comparison of the liganded and unliganded Fab shows no evidence for induced fit upon binding. We set out to examine the degree to which the specificity of Abs in this relatively "rigid" system may be altered yet the high affinity maintained. We focused on heavy-chain CDR3, which includes four contact residues to digoxin, three of which (positions 100, 100a, and 100b; Kabat numbering) (26) were contiguous. Ab 26-10 binds digoxin with high affinity but binds analogs substituted at position 16 of the cardenolide (see Fig. 1) less well (27). The relative decrease in binding correlates with the bulk of the 16-substituent (ϪOH, gitoxin; ϪCHO, 16formylgitoxin; ϪCOCH 3 , 16-acetylgitoxin). In a previous study, 3 we mutagenized five contiguous residues in HCDR3, spanning positions 99 -101 (Kabat numbering), which include three contact residues (tryptophan 100-alanine 100a-methionine 100b), each of which contacts digoxin in the crystal structure of the complex (25). Following panning against digoxin and three different analogs substituted at position 16 in the cardenolide D ring, mutants were recruited, which demonstrated that consensus sequences containing arginine instead of tryptophan at position 100 shifted the specificity of the mutants toward improvement in binding to position 16-substituted analogs. Although the resulting mutants bound gitoxin better than digoxin, this occurred at the cost of decreased affinity. The finding that a substitution at HCDR3 improved binding to analogs with C-16 substituents was counterintuitive, since the HCDR3 segment involved lay on the side of the digoxin ligand opposite to the location of position 16. Position 16 in the D ring is involved in close complementarity with contact residues from the heavychain CDR1, yet mutagenesis in HCDR1 (28) did not result in altered specificity.
In the present study, we used a 26-10 mutant containing arginine instead of tryptophan at position 100, which demonstrated the specificity change, as a template for further mutagenesis of a 5-residue segment of HCDR3 (positions 94 -98) displayed on phage and selected mutants by panning against both digoxin and 16-acetylgitoxin. We succeeded in selecting mutants with a further significant shift in specificity toward 16-substituted analogs. Some of these mutants had affinities for gitoxin higher than that of the original WT 26-10 for digoxin. Thus, despite the fact that 26-10 is a high affinity Ab obtained in a secondary immune response and probably has accumulated somatic mutations resulting in affinity maturation, providing close complementarity with the hapten, the Ab nonetheless retains the capacity to differentiate further to bind structurally related ligands with significant affinity. Moreover, the retention of the high affinity binding occurred despite the mutation of hapten contact residues.

EXPERIMENTAL PROCEDURES
Vector Construction-Randomly mutated DNA sequences were introduced into the HCDR3 region, including residues 94 -98 of both the WT anti-digoxin Fab 26-10 and the mutant A2. 3 Libraries of mutant sequences were inserted into the pComb 26-10 AHN vector (see Fig. 3) described previously. 3 Briefly, the vector contains nucleotide sequences encoding 26-10 Fab from H chain position Glu H1 to Arg H228 (26) in the hinge region (27,29), and the entire L chain (Asp L1 to Cys L214 ) was cloned into the pComb3 phagemid vector (2). Fab is displayed as a fusion to the N terminus of the gene 3 protein of bacteriophage M13 (see Fig. 3). Excision of the gene 3 portion using NheI and SpeI resulted in secretion of soluble 26-10 Fab into the culture supernatant (28) as described (2). The vector was modified previously with alterations in HCDR3 designed to prevent contamination of libraries randomized in the HCDR3 region with the parent 26-10 Fab-expressing vector (and its consequent competition in biopanning). 3 A unique HindIII site was earlier introduced by oligonucleotide-directed mutagenesis at the beginning of the HCDR3-encoding nucleotide sequence and changed the H chain amino acid sequence at positions 94 -96 from GSS to KLP. The introduction of a new restriction enzyme site allows linearization of contaminating parent plasmid DNA after ligation of insert (but not plasmids with library sequences inserted) and its subsequent elimination due to low transformation efficiency of the linear DNA form. Amino acid substitutions resulting from introduction of the HindIII site are predicted to prevent antigen binding, further substantially reducing the probability of selection of a contaminating parent construct. A unique NcoI site was eliminated from the DNA sequence 3Ј of HCDR1 and introduced 3Ј to the 3Ј-end of HCDR3 to provide a restriction site for cloning of library DNA fragments close to the mutagenized region in HCDR3. 3 We previously described the mutant A2 of the anti-digoxin Ab 26-10, obtained from a phage display library randomized at HCDR3 positions 99 -101 (26), which demonstrated a shift in specificity favoring binding to C-16-substituted cardenolides. A2 contained the sequence SRALQ at positions 99 -101, compared with the WT 26-10 sequence KWAMD (see Fig. 2). 3 Libraries of randomly mutated sequences at positions 94 -98 in 26-10 HCDR3 were constructed by PCR using the following oligonucleotide primers: 1, 5Ј-CAACCAGCGATGGCCGAGGTC-3Ј; 2, 5Ј-CTGAGGCT-CCATGGCCCCAGTACTGGAGGGCGCGGCT(SNN) 5 TGCACAGTAA-TAGACTGCAGAATCC-3Ј for creating a library on the background of the A2 mutant (SRALQ at positions 99 -101); or 3, 5Ј-CTGAGGCTCC-ATGGCCCCAGTAGTCCATAGCCCACTT(SNN) 5 TGCACAGTAATAG-ACTGCAGAATCC-3Ј for creating a library on the WT 26-10 background (positions 99 -101) (see Fig. 2) (NcoI site underlined).
N represents an equimolar mixture of all four dNTPs, and S represents an equimolar mixture of dCTP and dGTP. Two combinations of primers, 1 and 2 or 1 and 3, were used in a PCR (Taq polymerase from PerkinElmer Life Sciences) with the pComb26-10 AHN template. The 357-base pair product of this reaction extends from the beginning of the V region of the 26-10 chain through CDR3 and the adjacent NcoI site at nucleotide position 714 (Fig. 3).
This 357-base pair fragment was digested with BspEI and NcoI, and the resulting 314-base pair fragment band was excised from a 2% NuSieve agarose gel (FMC, Rockland, ME) and purified by a gel extraction kit (Qiagen). The purified fragment was ligated into an NcoI-BspEI-digested p26-10AHN vector and introduced by electroporation into E. coli XL1 Blue cells. Infection of bacterial cell cultures of each library of HCDR3 mutants with VCSM13 helper phage (XL1-Blue and VCSM13 from Stratagene, La Jolla, CA) generated two libraries of phage with surface-displayed Fab containing a 5-amino acid randomized segment (positions 94 -98) in HCDR3 either on the mutant (SRALQ) or WT (KWAMD) background (positions 99 -101, see Fig. 2).
Before panning, libraries were sequenced in their HCDR3 region to estimate their degree of randomization. Enrichment of bacteriophage expressing specifically bound Fab mutants was achieved by five successive rounds of binding to glycoside-BSA-coated microtiter wells (Costar 3690, Cambridge, MA). To monitor enrichment during the course of panning, we determined the ratios of the titers of eluted phage from antigen-BSA versus BSA-coated wells. Phage titers in an eluate from BSA-coated wells would represent nonspecifically bound phage. Panning was considered successful if antigen-BSA versus BSA eluate phage ratios increased after each round. For example, the digoxin-BSA/BSA phage ratio after five rounds of panning of library 1 increased substantially: from 8:1 after pan 1 to 500:1 after the fifth pan. Washing, elution, reinfection, and growth were done according to the procedure of Barbas and Lerner (31). A 1-g sample of pooled phagemid DNA from the final panning round was cut with NheI and SpeI and religated to remove the gene 3 sequence that serves as a membrane anchor (2).
After religation, the DNA was transformed into XL1-Blue cells and plated on LB (1.0% tryptone, 0.5% yeast extract, 1.0% NaCl, pH 7.0)/ carbenicillin plates. Bacterial colonies were isolated and screened for Fab production and antigen binding by ELISA. The DNA sequences of cardiac glycoside ELISA-positive clones were determined using the dideoxy sequencing method.
Isopropyl-␤-D-thiogalactopyranoside (Sigma) was added to the culture, and incubation continued overnight at 30°C. Fab were collected from either culture supernatants or periplasmic extracts according to Barbas and Burton (1). The yield of Fab was determined by ELISA titration (see below) on goat anti-mouse Fab-coated plates using enzymatically prepared (32) 26-10 Fab of known concentration as a standard. Fab produced by E. coli cells containing phagemid with mutant 26-10 sequences terminate at Arg 228 in the hinge region of 26-10 and contain a gene 3-encoded Thr and Ser just 5Ј of the stop codon.
ELISA Immunoassay-Direct binding ELISA was used for initial testing of binding to congeners of digoxin of clones obtained after five rounds of panning and removal of gene 3. The assay was performed as described previously (28). Wells of 96-well microtiter plates (Falcon) were coated with BSA alone, cardiac glycoside-BSA, or goat anti-mouse Fab (Sigma) in 0.1 M bicarbonate, pH 8.6, and blocked with 3% BSA in TBSA (50 mM Tris-HCl, 150 mM NaCl, pH 7.0, with 0.02% NaN 3 ). Bacterial cultures were induced by 1 mM isopropyl-␤-D-thiogalactopyranoside at logarithmic phase and collected overnight at 30°C with shaking. Bacterial supernatants were collected, added to ELISA plates, and incubated for 2 h at room temperature followed by a 30-min incubation at room temperature with peroxidase-labeled Fab-specific goat anti-mouse IgG. 2Ј,2Ј-Azino-bis(3-ethylbenzthiazoline-6-sulfonic acid) diammonium (Amersham Pharmacia Biotech) was used as a substrate for peroxidase. Color development was measured at 405 nM in a Bio-Tek ELISA reader.
Competition ELISA Immunoassay-The specificity (relative affinity) of Fab for different cardiac glycosides was determined using a competition ELISA assay in a 96-well format. Wells were coated with either digoxin-BSA, gitoxin-BSA, or 16-acetylgitoxin-BSA and contained a constant amount of Fab and serial dilutions in TBSA of cardiac glycosides (10 Ϫ3 M stock solution in pyridine). This resulted in a range of 10 Ϫ5 to 10 Ϫ12 M concentrations of inhibitor in the final mixture. The amount of Fab used was equal to the amount needed to reach a half-maximum binding point when titrated on congener-BSA-coated plates in the absence of competing congener. Each mutant Fab was tested in ELISA with the 16-substituted congeners and digoxin as competitors. The values reported are ratios of molar concentrations of inhibitor required to give 50% inhibition of Fab binding to congener-BSA-coated wells, relative to the respective molar concentration of digoxin. The ratios of inhibitory concentrations remained the same whether digoxin-, gitoxin-, or 16-acetylgitoxin-BSA was applied to coat the wells in competition ELISA, provided the appropriate amount of Fab was used as determined by titration on the same congener-BSA-coated plates.
Affinity Determination-Affinities for digoxin were measured on dilute bacterial supernatants or periplasmic extracts using a saturation equilibrium assay (28) using filtration through glass fiber filters to separate bound and free [ 3 H]digoxin (PerkinElmer Life Sciences). Each tube contained a constant amount of Fab, one of 12 serial dilutions of [ 3 H]digoxin (1.1 ϫ 10 Ϫ7 to 3.0 ϫ 10 Ϫ11 M, depending on preliminary screening), and 0.5 g/ml goat anti-mouse Fab antibody. Scatchard analysis was used to calculate affinities. All assays were done twice and repeated to obtain R 2 values of Ն0.95 and bound/free ratios of Յ0.1 K d .
Site-directed Mutagenesis-Variants of two clones obtained from randomized libraries at residues 94 -98 in HCDR3, SD6 (GSDRD at 94 -98) and SA20 (GGDTT at 94 -98), each with three different background sequences at positions 99 -101 (SWALQ, KRAMD, and KWAMD (WT), Table V) were constructed by site-directed mutagenesis. These mutants were produced by cloning of the corresponding PCR fragments into NcoI-BspEI-digested 26-10 vector (see Fig. 3). PCR fragments were made by using 5Ј-end primer 1 (see "Vector Construction") in combination with one of the following 3Ј-end primers containing mutated (compared with parent SD6 and SA20) sequences

RESULTS
We previously 3 constructed a randomized library displayed on bacteriophage, including positions 99 -101 (Kabat numbering (26)) of the HCDR3 of the high affinity anti-digoxin Ab 26-10 3 . From this library we selected by panning a subset of mutants that displayed altered specificity for digoxin and related congeners substituted at the C-16 position of the cardenolide moiety (Fig. 1). One of these mutant Fabs, designated A2, contained the sequence SRALQ at positions 99 -101 of HCDR3 instead of the wild-type (WT) KWAMD. 3 However, the specificity shift that was demonstrated by an 8-fold increase in relative binding to the C-16-substituted cardiac glycoside gitoxin ( Fig. 1) was accompanied by a reduction of affinity for digoxin. The mutant A2 is "heteroclitic" for gitoxin, binding gitoxin with higher affinity than digoxin, in contrast to the WT 26-10, which binds digoxin 5-fold greater than gitoxin (27). We chose the well characterized clone A2 (Fig. 2) as a parent background clone to further mutagenize the HCDR3 of 26-10 at positions 94 -98 (WT sequence: GSSGN; Fig. 2) (library 1) and select mutants that potentially exhibit changes in affinity and specificity.
A second library was made on the WT 26-10 background (residues 99 -101: KWAMD) with the same randomly mutagenized region 94 -98 of HCDR3 as for library 1 (Fig. 2). The vector pComb3 (2) was used for expression. Cloning of the WT 26-10 sequence and modification of pComb3 26-10 for library construction and purposes of expression were described elsewhere. 3 For a description of library construction (Fig. 3), see "Experimental Procedures." Library 1, consisting of residues randomized at positions H94 -98 on the background of the sequence SRALQ at positions 99 -101 of HCDR3, contained 2 ϫ 10 8 independent clones, based on the transformation efficiency. Twenty clones from library 1 were picked at random and subjected to DNA sequencing. All of the clones sequenced were different, and none exhibited deletion, indicating a satisfactory degree of randomization. Library 2 (on the WT background at positions 99 -101) contained 2 ϫ 10 7 independent clones. The expected amino acid complexity for randomization of a 5-residue segment is 3.2 ϫ 10 6 . Library 2 was sequenced as a total pool of DNA; no dominant bands were found, suggesting adequate randomization of sequences in this HCDR3 region.
Each library was subjected to five rounds of panning against two different cardiac glycoside congeners coupled to BSA: digoxin and 16-acetylgitoxin ( Fig. 1). Gene III sequences were removed from final pools of plasmid DNAs of selected mutants according to Barbas (2), and resulting clones after panning were tested for soluble Fab production (goat anti-mouse Fab binding) and binding to digoxin-BSA or 16-acetylgitoxin-BSA in an ELISA. Mutants are identified by a two-letter code and clone number: the first letter, S or W, indicates the background sequence used in library construction (SRALQ or WT 26-10, respectively); the second letter, D or A, indicates the congener used in panning (digoxin or 16-acetylgitoxin, respectively).

Library 1
Panning against Digoxin-BSA-The main purpose of this experiment was to determine whether it is possible to select mutant sequences in the region including residues 94 -98 of HCDR3 that would confer increased affinity of Fabs to 16substituted analogs, while retaining the same specificity pattern (favoring 16-substituted congeners) as the 26-10 A2 mutant (SRALQ at positions 99 -101). After five rounds of panning against digoxin-BSA and removal of gene III, 132 clones were picked which proved positive for both digoxin binding and Fab production by ELISA. Clones were regarded as positive if the signal-to-noise ratio in a single point direct binding ELISA was Ͼ4. We observed a significant increase in the ratio of the titers of phage eluted from digoxin-BSA versus BSA (from 8-fold following pan 1 to 550-fold after pan 5 for library 1), indicating enrichment of specific cardiac glycoside-binding clones. DNA sequences were obtained for 41 clones (Table I). Most clones had a unique amino acid sequence in the mutagenized region, except clones SD1, SD6, SD30, SD66, and SD93. Clones SD6, SD30, and SD93 had the same amino acid sequence, and clones SD6 and SD30 had the same nucleotide sequence, which differed from that of SD93. Clones SD1 and SD66 differed in nucleotide sequence and expressed the same protein sequence. The affinities (K a ) for 25 clones were determined ( Table I). All of the clones that occurred more than once were in the group with the highest measured K a for digoxin (Table I). The parental H94 -98 GSSGN sequence was not found among the clones sequenced, despite the selection of many clones with lower affinity to digoxin relative to that of the parental A2 (SRALQ) mutant. Clones SD28 and SD118 differ from the parent sequence only at position H96 (Table I). A large proportion of clones (34 of 41) retained Ser H95 , a known contact residue to digoxin in the crystal structure of the Fab 26-10-digoxin complex (25), and Gly H97 was present in 22 of 41 clones sequenced. Clones with affinities significantly higher (Ͼ2-fold) than the parental SRALQ all retain Gly H94 and Ser H95 , identical to the WT 26-10 sequence; Asp H96 is substituted for the WT serine in all but one of these. Gly H96 in the WT is replaced in all clones of this group by arginine (SD6 and SD66), asparagine (SD40 and SD91), or glutamine (SD50). Either glycine or aspartic acid occurs instead of asparagine at position 98. The remainder of the clones, with K a values approximating that of the parent SRALQ (1.2 ϫ 10 9 M Ϫ1 ) or lower, had varying degrees of difference in sequence compared with the parent GSSGN sequence. None of these had aspartic acid in position 96, as did the group with highest K a . Both Ser H95 and Gly H97 were preserved in a large proportion of clones of this group (21 of 35), probably reflecting the most common motif necessary to maintain sufficient affinity for digoxin, since these clones were selected by panning against digoxin-BSA. The ratio of WT 26-10 to SRALQ affinity in the prior study by Short et al. 3 was 9 ϫ 10 9 , versus 1.3 ϫ 10 9 M Ϫ1 , and here we used 13 ϫ 10 9 versus 1.2 ϫ 10 9 M Ϫ1 . 4 Five clones (SD23, SD24, SD16, SD20, and SD29; Table I) contain neither the WT Ser H95 nor WT Gly H97 that is present in the majority of mutant Fabs, yet affinities measured for two of these were not significantly different from that of the SRALQ parent. However, all of them possess threonine at position 95, which is structurally related to serine.
Clones with increased affinity for digoxin, selected in this experiment using digoxin-BSA, were further studied using competition ELISA with digoxin, gitoxin, and 16-acetylgitoxin as competitors to determine whether the corresponding Fabs retained the specificity shift of the parent mutant A2. The binding of Fabs SD6, SD40, SD50, and SD66 to position 16substituted congeners, relative to digoxin binding, was very similar to the parent SRALQ Fab (Table III). Apparently, the particular amino acid substitutions in the HCDR3 region of positions 94 -98 of these clones led to a 2-3-fold increase in affinity for digoxin, while retaining the same degree of increase in their binding to 16-substituted analogs compared with the WT 26-10, as in the parent SRALQ clone (ϳ7.5-fold higher binding to gitoxin and 3-fold to 16-acetylgitoxin).
Panning against 16-Acetylgitoxin-BSA-This experiment was performed to establish whether the panning of clones with a mutagenized HCDR3 94 -98 region on the background of the 26-10 A2 mutant (SRALQ at positions H99 -101) Fab against 16-substituted congeners would bring further improvement of binding of selected clones to the 16-substituted analogs of digoxin.
An initial assessment of clones after five rounds of panning against 16-acetylgitoxin-BSA was done by comparing Fab binding to digoxin-BSA and 16-acetylgitoxin-BSA in a single point ELISA. By so doing we sought to rapidly screen the pool of selected clones, looking for candidates with improved binding to 16-substituted analogs. All 40 clones tested were positive for both haptens. Based on ELISA data, clones SA5, SA10, SA12, SA14, SA19, SA20, SA29, and SA33 showed increased relative 4 The calculated affinities varied in relation to the lot of tritiated digoxin used; all assays reported here were done using the same lot.   3 Kabat numbering is used (26). These two parental sequences were used for libraries 1 and 2, which were randomized at positions 94 -98. The one-letter code for amino acids is used (where X represents any amino acid).
binding to 16-acetylgitoxin, compared with WT 26-10 and 26-10 SRALQ controls (data not shown). The DNA sequences and affinities for digoxin of these eight clones and 14 other clones that were positive for both haptens in ELISA are shown in Table II. The binding specificities of a representative 19 clones are shown in Table III. The data reveal a distinct sequence homology pattern among clones with improved relative binding to 16-substituted congeners. The hapten contact residue Ser H95 was replaced by glycine in clones SA5, 10, 12, 14, 19, 20, 29, and 33. (Clone 19 proved to consist of two clones with different sequences. Clones 19-2 and 19-3 were subcloned on agar.) In these eight clones, Gly H97 was also replaced by one of five different amino acid residues (Table II), except in clone SA5. There is a certain degree of homology or consensus among the remaining 14 clones sequenced (Table II). The sequence pattern resembles that for clones obtained in panning against digoxin-BSA ( Table I). All of them except SA22 retain the WT Ser H95 contact residue, while SA22 has threonine at position 95. A significant proportion of clones (8 of 14) retain both Ser H95 and Gly H97 . Clones SA18 and SA41 have the same amino acid sequence as clones with increased affinity for digoxin obtained from library 1 panned against digoxin-BSA (SD66 and SD6, respectively), but their nucleotide sequences are different. The K a values for digoxin binding of clones selected in this panning experiment are approximately the same as those for clones with similar amino acid sequences found in panning against digoxin-BSA (compare Tables I and II).
We measured the relative affinity for 16-substituted congeners of 18 different clones panned from library 1 against 16acetylgitoxin. The specificity patterns segregated into two relatively distinct groups (Table III). Those clones with glycine substituted at position 95 (SA5, SA10, SA12, SA14, SA19.2, SA19.3, SA20, SA29, and SA33; Tables II and III) have affinities for 16-substituted analogs substantially increased (from 2to 20-fold) compared with the parent 26-10 SRALQ (A2) mutant. The affinity for digoxin is decreased 2-4-fold for clones SA5, SA10, SA19.2, SA19.3, and SA20 (Table II) relative to the parent. We could not determine the K a for digoxin for clones SA12 and 14 (Ͻ10 7 ; Ref. 28). One clone, SA33 Fab, had the same K a value as the parent SRALQ mutant. The relative affinity for gitoxin increased ϳ20-fold for clones SA10, SA12, SA19.2, SA19.3, and SA20 (IC 50 ϭ 0.02) compared with the SRALQ mutant (IC 50 ϭ 0.4), and more than 100-fold compared with WT 26-10. Fabs SA29 and 33 showed less increase in affinity for gitoxin (5-6-fold), and clone SA5 showed only a 2-fold increase. SA5 was the only clone from this group with glycine at position H95, the same as for the WT 26-10 or the 26-10 SRALQ mutant. It appears that both introduction of glycine at position 95 of HCDR3 and replacement of glycine at position 97 are necessary for the dramatic shift in specificity toward gitoxin. With regard to the specificity shift toward analogs with bulkier substituted groups at position 16 of the cardenolide moiety, such as 16-formylgitoxin or 16-acetylgitoxin, virtually all of the clones with glycine at position 95 a Dashes indicate amino acid residues identical to the respective residues in WT 26 -10. b Affinities were measured using an equilibrium saturation method with filtration through glass fiber filters for separation of bound and free ligand (28,34). The margin of error between the determinations for a given Fab in different experiments was less than 25%.
c Clone A2, derived from 26 -10 WT, 3 contains the sequence SRALQ at positions 99 -101, and was used as the parent clone for Library 1. d,e Clones with different nucleotide sequences but identical amino acid sequences. In addition, clone SD30 had identical nucleotide sequence to SD6.
f Not measurable in this assay.
showed increased relative affinity for these congeners compared with the WT 26-10 or 26-10 SRALQ mutant. The affinity for 16-acetylgitoxin increased approximately 3-7-fold compared with the SRALQ mutant and 9 -20-fold compared with WT 26-10. In the case of mutant SA5, a 40-fold increase in binding 16-acetylgitoxin was observed compared with the SRALQ mutant and 120-fold compared with the WT 26-10. Binding to formylgitoxin, which has a group at position 16 larger than the hydroxyl group of gitoxin, but smaller than the acetyl group of 16-acetylgitoxin, occupies an intermediate range. Fab SA5 is unique in that its binding to acetylgitoxin is greater than to formylgitoxin. Fab SA5 binds 16-acetylgitoxin equally well as digoxin. For most clones, the consensus sequence of glycine at position 95 of HCDR3 and replacement of WT glycine 97 favor binding to congeners with a less bulky group at position 16 of the cardenolide ring. The specificity for position 16-substituted analogs of Gly H95 mutants was supported by measuring the binding of Fab to ouabain by competition ELISA; none of the mutants showed any change in specificity compared with the parent A2 Fab (data not shown). Competition ELISA, performed for 10 other clones selected in this experiment that did not contain Gly at H95 but instead serine and threonine (Table III), did not reveal any substantial change in specificity pattern compared with the parent 26-10 SRALQ mutant. Some of these clones are identical to their counterparts from panning against digoxin-BSA or have only one amino acid residue different in the HCDR3 sequence. The only clone with glycine at position 95 and a replaced glycine at position 97, SD31, found in the library 1 experiment using digoxin-BSA panning, also demonstrated a further shift of specificity toward gitoxin (Table III).

Library 2
Panning against Digoxin-BSA-This experiment was done to determine the effects of random mutagenesis of residues 94 -98 in HCDR3 of the WT 26-10 on affinity for digoxin, as compared with library 1.
After five pannings against digoxin-BSA, 40 clones expressing soluble Fabs were tested in direct single point ELISA for binding to digoxin-BSA, 16-acetylgitoxin-BSA, and goat antimouse Fab. All 40 clones were positive for both haptens. In contrast to library 1, no clones with increased binding to 16acetylgitoxin were found (data not shown). Random sequencing of 12 digoxin-positive clones revealed the occurrence of repeat DNA sequences (Table IV), with seven unique sequences. Values of the affinity for digoxin of five clones were lower than that of WT 26-10. Ser H95 and Gly H97 were present in all clones, not unexpectedly, since panning was done against digoxin-BSA.
Panning against 16-Acetylgitoxin-BSA-After five rounds of panning against 16-acetylgitoxin-BSA, only three of 32 clones tested in single point ELISA showed binding to either digoxin or 16-acetylgitoxin. None of these three clones demonstrated an increase in 16-acetylgitoxin ELISA signal, as found for library 1 panned against 16-acetylgitoxin. Sequencing revealed that two of them (WA12 and WA34) had identical DNA sequences (encoding LSDPQ). Both of the clones with different sequences, WA34 and WA7, had serine at position 95 and Gly H97 replaced (by proline or valine). In competition ELISA (Table III), both clones demonstrated a modest specificity shift toward 16-substituted analogs as compared with WT 26-10 Fab. The low number of binders selected in this experiment may be partially attributed to limited initial representation of sequences with affinity sufficient for panning against 16-acetylgitoxin from this library under conditions used (the binding of parental 26-10 to 16-acetylgitoxin is reduced 130-fold compared with digoxin).

DISCUSSION
Monoclonal Ab 26-10, with high affinity for the cardiac glycoside digoxin (digoxigenin tridigitoxose), has been the subject of studies of the structural basis of affinity and specificity, as the three-dimensional structure of the 26-10 Fab in complex with digoxin was determined, as well as the structure of the unliganded form (25). Digoxin binds to 26-10 in a deep, ellipsoidal pocket that is predominantly hydrophobic. The lactone ring of digoxin (Fig. 4) is buried at the bottom of the pocket. The cardenolide A-D rings are packed between residues from HCDR1 and HCDR2 on one side of the pocket and residues from HCDR3 and LCDR1 on the other side (Fig. 4). The cardenolide is rigid, with rotation limited to the C 3 -sugar bond, the sugar-sugar bonds, and the bond to the lactone ring. The specificity of Ab 26-10 for digoxin and 33 structurally related cardiac glycosides was determined (27). Substitutions at position 16 of the cardenolide D ring result in a decrease in binding directly related to the size of the substituent (-OH, -CHO, or -COCH 3 ). The binding of Ab 26-10 becomes sensitive to the nature of the (sugar) substituents at C 3 when position 16 substitutions are present. We postulated (27) that the C-16-substituted analogs were "shifted" in the site to accommodate the more bulky group at C-16, resulting in new contacts with the sugars, not present in the 26-10-digoxin complex.
a Dashes indicate amino acid residues identical to the respective residues in WT 26 -10. b Affinities were measured using an equilibrium saturation method with filtration through glass fiber filters for separation of bound and free ligand (28,34). The margin of error between the determinations for a given Fab in different experiments was less than 25%.
c Clone A2, derived from 26 -10 WT, 3 contains the sequence SRALQ at positions 99 -101 and was used as the parent clone for library 1.
d These two clones are identical in amino acid sequence to clones SD1 and SD6 (Table I)  In previous work, we selected mutants from a phage display library that had been randomized at five contiguous HCDR3 positions, 99 -101 (Kabat numbering) (KWAMD). 3 Three consecutive residues (Trp 100 , Ala 100a , and Met 100b ) make contact with the hapten (Fig. 4) (25). The residue Trp 100 interacts extensively with the B and D rings of digoxin. A consensus sequence at residues 99 -101 was evident among mutants obtained by panning against digoxin-BSA, formylgitoxin-BSA, and 16-acetylgitoxin-BSA. When 16-substituted analogs were used in panning, Trp 100 was uniformly substituted by Arg 100 . Mutants containing Arg 100 proved to have increased specificity for 16-substituted congeners. In particular, these mutants were heteroclitic for gitoxin, albeit at the cost of a loss of digoxin affinity of 1 log. These results were interpreted as arising from a shift in position of the hapten, allowing more space at the cardenolide C-16 position, due to substitutions of the less bulky arginine 100 on the opposite side of the hapten (Fig. 4). 3 In the work reported here, we looked for further modulation of affinity and specificity of 26-10 by selecting phage-displayed random mutants at HCDR3 positions 94 -98 on the background of both the WT sequence of 99 -101 and a mutant sequence, SRALQ at 99 -101, which had conferred improved binding to 16-substituted analogs. 3 The relevant sequences of HCDR3 are shown in Fig. 2. Each of these libraries was selected using digoxin-BSA and 16-acetylgitoxin-BSA.
Library 1 was mutagenized at residues H94 -98 on the SRALQ background at positions 99 -101 (Fig. 2), and library 2 was mutagenized on the WT KWAMD background at positions 99 -101. We wished to see what influence the SRALQ back- 1.0 64 a Digoxin contains a 12-hydroxyl, while the other three congeners listed lack a 12-hydroxyl. b Affinities were measured using an equilibrium saturation method with filtration through glass fiber filters for separation of bound and free ligand (28,34). The margin of error between the determinations for a given Fab in different experiments was less than 25%.
c Clone A2, derived from 26 -10 WT, 3 contains the sequence SRALQ at positions 99 -101, and was used as the parent clone for library 1. d SD, clones selected from library 1 (SRALQ background) panned against digoxin; SA, clones selected from library 1 panned against 16acetylgitoxin; WA, clones selected from library 2 (WT background) panned against 16-acetylgitoxin. e NM, not measurable in this assay. f Clone 19 proved to be composed of a mixture of two mutants, which were subsequently separated and characterized, designated 19.2 and 19.3.

M S T G H
a Affinities were measured using an equilibrium saturation method with filtration through glass fiber filters for separation of bound and free ligand (28,34). The margin of error between the determinations for a given Fab in different experiments was less than 25%. ground would have on specificity and affinity of mutants, selected by panning against digoxin and position 16 analogs. Is a "stepwise" approach, in which first HCDR3 residues 99 -101 are mutated and thereafter the contiguous residues 94 -98 are mutated, useful in selecting desirable mutants with shifted specificities and increased affinities? The stepwise approach is one solution to the practical limitations of library size due to the limitation of transformation efficiency of Escherichia coli.
H94 -98 Mutagenesis on WT H99 -101 KWAMD Background (Library 2)-We used selection on a WT KWAMD background at positions 99 -101 as a control experiment to compare the pattern of mutations in the region H94 -98 to that for the mutant SRALQ background at 99 -101. The affinities of all mutants selected from the library on the WT KWAMD background were lower than that of the parent 26-10 Fab (Table  IV). Panning on digoxin-BSA produced clones demonstrating conservation of Ser H95 and Gly H97 and modest variation at the other residues. Selection of Ser H95 is consistent with the crystal structure of the 26-10-digoxin complex. Ser H95 , located adjacent to the lactone and close to the steroid D ring, contacts the hapten and forms a hydrogen bond with Asn H35 , an HCDR1 hapten contact residue (25). Gly H97 , although completely solvent-exposed, is likely to be conserved to preserve the backbone conformation. The and values for Gly H97 in the digoxin-26-10 Fab complex are ϩ54 and Ϫ175, respectively. In the WT Fab 26-10, the conformation of the HCDR3 backbone is a critical determinant of the placement of Trp 100 and high affinity binding of digoxin. The side chain of the residue at position 94 projects away from the hapten and into the hydrophobic core of the V H domain. Small hydrophobic side chains seem likely to be tolerated, while large side chains would cause disruption, possibly affecting HCDR3 and backbone. It is difficult to explain the presence of Ser H94 in mutants WD9 and WD5 (Table IV). The presence of larger hydrophobic residues at this position appears acceptable, but perhaps at a cost in affinity. Diversity at position 96 is less than expected. The side chain of Ser H96 points away from the hapten and is partially solvent-exposed in the WT 26-10-digoxin complex. It seems possible for larger side chains to fit here, projecting into the solvent or making hydrogen bonds with other nearby polar groups. Ile H96 potentially makes hydrophobic interactions with Phe H32 and Tyr H27 . Position H98 residues in the mutants selected are polar or charged, which agrees well with the structural data indicating that Asn H98 is solvent-accessible and would not be predicted to affect antigen binding or backbone conformation. In summary, for library 2 the selection by digoxin-BSA panning of mutants in the H94 -98 region (on the background of WT sequence at H99 -101) does not result in unique affinity or specificity differences. When the same library was panned repeatedly against 16-acetylgitoxin-BSA, only a few binding clones were recruited. That only a few clones were found is not surprising, since the background at positions 99 -101 was WT, and the WT Ab binds 16-acetylgitoxin poorly compared with clones having a mutated background at positions H99 -101, such as the mutant SRALQ. In contrast, the results with library 1 (see below) using 16-acetylgitoxin-BSA panning were successful in recruiting clones that secreted Fab with altered specificity.
H94 -98 Mutagenesis on H99 -101 SRALQ Background (Library 1)-The mutant A2 (SRALQ at positions 99 -101) 3 used as the parental sequence in Library 1 binds gitoxin better than digoxin. We panned library 1 (Fig. 2) against digoxin-BSA to determine whether it is possible to raise the affinity of mutant Fabs to digoxin while simultaneously retaining the specificity shift toward gitoxin. Seven mutants (Table I) with five unique amino acid sequences had affinities to digoxin increased 2-4fold compared with the parent A2 Fab (Table I, mutants SD1,  SD6, SD40, SD50, SD66, SD91, and SD93). These mutants maintained the same specificity shift toward position 16-substituted analogs as exhibited by the parent A2 mutant (Table   FIG. 4. A, combining site of the anti-digoxin Ab 26-10, including the HCDR3 loop (residues 94 -101, Kabat numbering). The H chain backbone is shown in green and the L chain backbone in red. Side chains are shown explicitly for certain labeled residues in HCDR3, HCDR1, and LCDR1. The side chains for HCDR3 loop residues are shown in yellow. The hapten contact residues for HCDR3 include Ser 95 , Trp 100 , Ala 100a , and Met 100b . The contact residues Asn H35 in HCDR1 and Tyr H50 in HCDR2 are shown explicitly. Other contact residues are omitted for clarity (Tyr H47 , Tyr H33a , Thr L91 , and Pro L96 ). The hapten digoxigenin monodigitoxose (cf. Fig. 1) is in purple. The lactone ring is at the bottom of the binding pocket. C-16 is located on the right in the cardenolide D ring, which is seen en face, where hydroxyl, formyl, and acetyl groups are substituted in the analogs gitoxin, 16-formylgitoxin (gitaloxin), and 16-acetylgitoxin, respectively. Asn H35 is involved in close complementarity to the C-16 of the cardenolide and also hydrogen-bonds with Ser H95 and Tyr H47 (25). B, orthogonal view of the binding site with the hapten (purple) in the center, seen end-on. Only HCDR3 residues are shown explicitly. Color coding is the same as in A. The images were produced using MOLSCRIPT (38) and Raster3D (39). III). Thus, it proved possible to find mutations in the region H94 -98 responsible for raising the affinity for both digoxin and position 16-substituted analogs, provided SRALQ is used as a parent sequence at positions H99 -101. Ser H95 was preserved in most of the mutants (Table I) or replaced by threonine in a few clones, reaffirming its important role in the structure of the binding pocket of 26-10. At position H94, mostly glycine or hydrophobic residues were observed, and at position H98 generally charged or polar residues were observed, as expected from the crystal structure. Although small hydrophobic residues seem to fit better at position H94 to preserve loop structure, larger hydrophobic residues are allowed in these mutants, possibly because of simultaneous introduction of hydrophobic residues at position H96 (Table I) that can pack against the hydrophobic core (both H94 and H96 side chains point approximately toward the same area, away from the hapten) (Fig. 4). There is a cluster of charged residues at positions H96 -98 among the best binding mutants, such as the H96 -98 sequence DRD in the mutant SD6 (Table I). We cannot offer a simple explanation for this observation; there are no charged residues nearby to stabilize the structure by interacting with the DRD cluster. We noted above that Gly H97 seems to be essential for maintaining the loop conformation. Indeed, it is present in many of the mutants (the group characterized by mutant SD126, etc.) (Table I), where position 98 is a hydrophobic residue. However, Gly H97 is replaced by charged residues in the DRD-type mutants; this may be possible because of the presence of arginine at position 100 instead of tryptophan, as in the WT sequence. Arg 100 would make the loop more flexible, allowing greater latitude in moving the CDR backbone. The position of Trp H100 is stabilized by interaction with Tyr L32 on the opposite side from the hapten (Fig. 4). If this were true, if Arg H100 was replaced by Trp, the affinity of mutants in the SD6 group of clones would be expected to be decreased. However, the binding to digoxin of such a mutant produced by site-directed mutagenesis of SD6 (GSDRDSWALQ, K a ϭ 3.4 ϫ 10 9 M Ϫ1 ) ( Table V) was not lower than that of the parent SD6 Fab. Thus, the influence of the DRD-like sequences on the loop structure cannot be so simply rationalized.
We also panned library 1 on 16-acetyldigoxin-BSA to select mutants with improved binding to 16-substituted analogs relative to digoxin. Not surprisingly, many selected mutants had sequences and binding properties (Tables II and III) similar to  those selected using digoxin-BSA (compare Tables I and II). One group of mutants, such as SA30, SA2, etc., had Ser H95 and Gly H97 mostly preserved, as in the group obtained from digoxin-BSA panning (SD126, SD7, etc.; Table I). The mutants with Ser H95 (Table II), selected using 16-acetylgitoxin-BSA, retain the same specificity profile as the parent A2 clone (Table III).
However, a group of nine clones with glycine at position H95 was selected only in panning on 16-acetylgitoxin-BSA (Tables   II and III), except for a single clone with Gly H95 (SD31) selected using digoxin-BSA panning (Table I). Competition ELISA (Table III) showed a further dramatic shift of specificity of these Gly H95 mutants toward position 16-substituted analogs. The affinity for digoxin is the same for clone SA33 as that of the parent A2, while the other Gly H95 clones show a 2-10-fold reduction in affinity. Based on measured K a for digoxin and competition ELISA data, we estimate affinities for gitoxin of some of the Gly H95 mutants (e.g. SA20 and SA19.3) to be equal to the affinity of WT 26-10 for digoxin. At the same time, their affinity for digoxin is 50 times lower than the affinity for gitoxin. Apparently, the Gly H95 mutation is responsible for further specificity shifts, up to 20-fold, toward C-16-substituted analogs as compared with the parent A2. Moreover, here we find a significant increase in affinity to C-16-substituted analogs, unlike previous studies on 26-10 mutants in which the specificity shifts were attributed mostly to selective decrease in affinity to digoxin, while binding to analogs remained at the same level (28,33,34). For most mutants, the size of the substituent at position 16 of the cardenolide ring had the same influence on analog binding as for the parent A2: i.e. the bulkier the group, the weaker the binding. The only exception was clone SA5, where the specificity shift was more dramatic for 16-acetylgitoxin than for gitoxin. Binding of mutant SA5 Fab clones to 16-acetylgitoxin relative to digoxin binding increased 130-fold compared with WT 26-10. Clone SA5 is the only mutant in this group in which glycine is not replaced at position H97 when there is a glycine replacement for serine at position H95. Based on the K a values and competition ELISA data, we estimate that the affinity for 16-acetylgitoxin of mutant SA5 is increased 42 times compared with the parent A2 Fab.
We attempted to test to what degree the H94 -98 sequence containing Gly H95 determines the observed specificity shift independent of the rest of the HCDR3 loop. We therefore constructed three mutants of the original SA20 (Table II) sequence (GGDTTSRALQ) as follows: SA20 -1, GGDTTSWALQ; SA20 -2, GGDTTKRAMD; and SA20 -3, GGDTTKWAMD (Table V). Mutant SA20 -1 differs from SA20 by the substitution of Trp for Arg at position H100. Mutant SA20 -2 consists of the sequence of SA20 at positions 94 -98 and the WT sequence at 99 -101, except for the substitution of tryptophan for arginine at position 100. Mutant SA20 -3 has the sequence of SA20 at 94 -98 on the WT 99 -101 background. On the basis of competition ELISA, all of these mutants showed a specificity shift toward gitoxin (data not shown) similar to the parent SA20 clone. The only effect of introducing mutated 99 -101 sequences was that the affinity for digoxin and its analogs dropped 10-fold compared with the parent SA20 clone. Apparently, as in the case of the SD6 mutant (Table V) 3 Clones SD6 and SA20 were obtained from library 1 (Fig. 2) and derived from the A2 parent. Clones SD6-1, -2, and -3 and SA 20-1, -2, and -3 are mutants of SD6 and SA20, respectively, produced by site-directed mutagenesis.
loop and, in particular, independent of the residue at position H100. With both the SD6 and SA20 (SA20 -1) mutants, we did not observe any influence on specificity by reintroducing Trp H100 , thus replacing Arg 100 , and converting SRALQ to SWALQ, in the case of both the SD6 and SA20 mutants. The same was true for the completely WT H99 -101 KWAMD background (mutant SA20 -3: GGDTTKWAMD). Based on the three-dimensional structure of the Fab 26-10digoxin complex, we argue that the most probable cause of the specificity shift for Gly H95 mutants is the extra space at the H95 position due to the removal of the side chain of Ser H95 . This space is required to accommodate the 16-hydroxyl, -formyl, or -acetyl groups, thus increasing their complementarity with the Ab binding pocket. It was proposed previously (35) that the hydrogen bond between Asn H35 and Ser H95 was important in maintaining the binding pocket of 26-10. This hydrogen bond is lost in Gly H95 mutants. It is possible that the position 16 group of congeners is now closer to the Asn H35 side chain and makes a hydrogen bond with it, replacing the hydrogen bond to Ser 95 . This might explain why the specificity shift is so dramatic in the Gly H95 mutants. The results of the 26-10 H94 -98 randomization experiments on an H99 -101 SRALQ background indicate that it is possible to dramatically shift the specificity of the Fab toward 16-substituted digoxin analogs. The SRALQ background at H99 -101 seems to provide a sufficient threshold affinity for binding to 16-substituted analogs of digoxin so that mutants containing Gly H95 , resulting in a further specificity shift toward binding 16-substituted analogs, may be repeatedly selected. In the light of the crystal structure of the complex, it is likely that most of these mutant structures look similar to 26-10 WT, with some local changes and in some cases compensatory mutations. Nevertheless, using phage display, we were able to dramatically change Fab specificity by changing the amino acid residues at two positions that were digoxin "contact" residues in the original structure of the loop and in some cases retain or improve affinity.