J. Mol. Biol. (1995) 254, 392–403 CDR Walking Mutagenesis for the Affinity Maturation of a Potent Human Anti-HIV-1 Antibody into the Picomolar Range Wei-Ping Yang1, Kimberly Green1, Sally Pinz-Sweeney1 Amelia T. Briones1, Dennis R. Burton1,2 and Carlos F. Barbas III1* 1 Department of Molecular Biology, The Scripps Research Institute, 10666 North Torrey Pines Road, La Jolla, CA 92037, USA 2 Department of Immunology The Scripps Research Institute, 10666 North Torrey Pines Road, La Jolla, CA 92037, USA We describe the investigation of methodologies for the creation of very high affinity human antibodies. The high affinity human antibody b4/12 was optimized for its affinity to the human envelope glycoprotein gp120 of human immunodeficiency virus type 1 (HIV-1). Five libraries of b4/12 were constructed by saturation mutagenesis of complementarity-determining regions (CDRs). Libraries of antibody Fab fragments were displayed on the surface of filamentous phage and selected in vitro for binding to immobilized gp120. Sequential and parallel optimization strategies of CDRs were examined. The sequential CDR walking strategy consistently yielded b4/12 variants of improved affinity in each of the four different optimization sequences examined. This resulted in a 96-fold improvement in affinity. Additivity effects in the antibody combining site were explored by combining independently optimized CDRs in the parallel optimization strategy. Six variants containing optimized CDRs were constructed. Improvement of affinity based on additivity effects proved to be unpredictable but did lead to a modest improvement in affinity. Indeed, only one of the six combinations demonstrated additivity. The highest affinity Fab prepared using this strategy was improved 420-fold in affinity. The affinity of this Fab was 15 pM as compared to 6.3 nM for b4/12. Examination of the kinetics of Fab binding to gp120 revealed that improvements in affinity were dominated by a slowing of the off-rate of the Fab. The methodology presented here provides a route for the improvement of the affinities of antibodies typical of tertiary immune responses into the picomolar range. Such improvements may have profound effects on the utility of antibodies as therapeutic and prophylatic agents. 7 1995 Academic Press Limited *Corresponding author Keywords: phage display; human immunodeficiency virus; structure/function; passive immunization; additivity effects Introduction Antibodies represent the most molecularly diverse and structurally studied family of proteins (Padlan, 1994). Their therapeutic use in humans has been studied for over a hundred years (Behring, 1893) and their role as adaptable binding molecules of defined specificities has aided the flowering of molecular biology in the latter half of this century. Abbreviations used: CDR, complementaritydetermining region; HIV-1, human immunodeficiency virus type 1; gp, glycoprotein; sCD4, soluble CD4, PCR, polymerase chain reaction; Ig, immunoglobulin. 0022–2836/95/480392–12 $12.00/0 Hence, antibodies play a central role in basic science and in our own survival. Our interest has been to develop a more thorough understanding of antibodies through the purposeful manipulation of their affinity and specificity. The current HIV-1 pandemic demands the development of potent anti-viral molecules. In this regard we have utilized phage display of combinatorial antibody Fab fragment libraries to select a large number of human anti-HIV-1 antibodies (Barbas et al., 1993a; Burton & Barbas, 1994). The most promising of these is antibody b4/12, which when prepared as an IgG1 has been demonstrated to efficiently neutralize a large number of primary 7 1995 Academic Press Limited 393 Affinity Maturation of a Human Antibody isolates of HIV-1 (Burton et al., 1994). To improve the likelihood that this molecule could succeed in prophylactic and therapeutic application we have developed methods for the improvement of its affinity. A number of methods and reports have been published which are devoted to the improvement of antibody affinity (for a review, see Burton & Barbas, 1994). We have sought methods for the development of very high affinity antibodies wherein modifications to the parental antibody are constrained to the complementarity determining regions. Changes in primary sequence in these regions are less likely to generate immunogenic antibodies than changes in the more sequence-constrained framework regions. In a previous report we improved the affinity of b4/12 for gp120 about eightfold and demonstrated that affinity was correlated with neutralization potency (Barbas et al., 1994). In this report we extend and develop our preliminary examination of the CDR walking mutagenesis strategy. Sequential and parallel optimization strategies have been explored and the general applicability of additivity effects in the combining site examined. We have improved the affinity of Fab b4/12 420-fold in its affinity for the HIV-1 envelope protein gp120. This study suggests a route to the creation of very high affinity antibodies. Results Design of antibody libraries No specific structural information on the antibody b4/12 or its antigen gp120 was available to guide the design of Fab libraries. Antibodies, however, constitute one of the most structurally studied classes of proteins (Padlan, 1994; Wilson & Stanfield, 1993). General information regarding the chemistry of this class of proteins, together with previous results obtained by chain shuffling experiments with this antibody, has served to guide the experiments under consideration here (Barbas et al., 1993b). Five segments within four CDRs were targeted for saturation mutagenesis and affinity optimization using phage display. Within the heavy chain, the entire CDR1-H region consisting of residues H31 to H35 was targeted in library H1 (numbering convention of Kabat et al. (1991)). The CDR3-H region of b4/12, 18 amino acid residues long, is significantly longer than the average observed in human antibodies, which is approximately 12 residues (Wu et al., 1993). Due to technical limitations in the size of libraries which may be constructed and surveyed with confidence using the phage display approach, six codons or less are generally targeted for saturation mutagenesis using NNK or NNS doping strategies (Lowman & Wells, 1993; Barbas & Barbas, 1994). For this reason, the nine residues at the core of the VDJ fusion region were targeted in a two-step sequential optimization strategy; H97-100 and H100A-E. Chain shuffling experiments suggested this region to be a natural hotspot in the somatic evolution of this antibody (Barbas et al., 1993b). In previous studies, a sequential selection of library H1 followed by randomization and selection of residues H97-100 of CDR3-H yielded a clone designated 3B3 with a 8.2-fold improvement in affinity (Barbas et al., 1994). As detailed below, the increase in affinity of the 3B3 clone was due almost exclusively to optimization of the H97-100 region. In this report, the adjacent residues in CDR3-H, H100A-E, were chosen for optimization in the context of the clone 3B3; library H3B3. In the light chain, CDRs 1 and 3 were targeted for random mutagenesis. Interaction of CDR2-L with antigen is, in general, less frequent and was not considered for optimization here (Wilson & Stanfield, 1993). In CDR1-L six residues predicted to be solvent exposed, L27A to 32, were targeted for random mutagenesis in library L1. In CDR3-L residues L90, and L92 to 95 were targeted for randomization. Residue L91 which is a highly conserved tyrosine in light chains in general (Wu et al., 1991) and in b4/12 variants obtained by chain shuffling and was not targeted. The CDRs which have been targeted are four of the five most heavily utilized contact regions observed in structures of antibody–antigen complexes (Padlan, 1994; Wilson & Stanfield, 1993). These regions are predicted to form a significant portion of the core of the binding surface of the antibody. This is demonstrated in Figure 1 by highlighting the regions under consideration on the combining site of the human antibody Kol, for which a crystal structure has been determined (Marquart et al., 1980). Sequential optimization of CDRs To assess the sequential CDR walking strategy, four sequential CDR optimization experiments were performed. Selected clones were characterized by sequencing of the entire variable domain and determination of affinity of purified Fab for antigen using the BIAcore biosensor from Pharmacia (Karlsson et al., 1991). The first of these experiments extends the previous two-step sequential walk which targeted CDR1-H in the H1 library and residues H97 to 100 of CDR3-H (Barbas et al., 1994). The highest affinity clone that resulted from these studies was clone 3B3. The subsequent step was to CDR1-L in library L1. Library L1 consisted 5 × 108 independent variants in this region built within the context of the 3B3 heavy chain. Following four rounds of selection, eight clones were sequenced following verification of binding activity with the soluble protein. Five clones were identical to the starting clone at the nucleic acid level and resulted from contamination of the original sequence in the construction of the library. Contamination was at a low level since sequencing of ten unselected variants revealed ten unique sequences. The affinities of the three remaining unique clones 394 Affinity Maturation of a Human Antibody Figure 1. The tube structure of a Fab on which the regions targeted for optimization have been highlighted. The antibody Kol has been utilized as a model here because of its extended CDR3-H region which is similar in length to antibody b4/12 under consideration here. The light chain variable region is shown on the left in pink and the heavy chain variable region on the right in blue. The individual regions targeted within the CDRs have been color coded with the key given in the Figure. This view highlights the relative disposition of the CDR regions targeted for optimization. Molecular models were constructed by Mike Pique using AVS software. 3B3/L1.1-3 were characterized. The clones as compared to their parent 3B3 were found to be of lower or similar affinity (Table 1). Selection was then continued for an additional four rounds resulting in the isolation of a single clone following the sequencing of ten clones, 3B3/L1.4. This clone was improved 29-fold in affinity with respect to b4/12 and 3.5-fold with respect to 3B3 (see Table 1). In the next step, five residues within the CDR3-L of 3B3/L1.4 were mutagenized and a library of 2 × 108 variants was constructed; library L3A. Following five rounds of selection, nine clones were characterized by sequence analysis and affinity determination (Table 2). Each clone contained a unique CDR3-L sequence and an affinity for antigen that was generally lower than the parent clone of this series, 3B3/L1.4. To improve the selection of higher affinity variants in the sub-nanomolar range using a solid phase selection format, an off-rate biased selection was developed. In five subsequent rounds of selection, 8 mM 3B3 was added to the selection well following the washing step. The selection plate was then placed at 37°C for three hours, the well washed, and bound phage eluted at low pH. This selection resulted in the isolation of clones 3B3/L.14L3.10 to 12. These clones differ in sequence only at position L95. Each clone was improved in affinity for antigen. The highest affinity clone 3B3/L1.4L3.11 was improved 3.3-fold in affinity above 3B3/L1.4 and 96-fold above b4/12. This clone is the product of a four-step sequential CDR walk. A parallel experiment without off-rate selection with antibody revealed a multitude of clones similar to that seen following the initial five rounds of Table 1. Light chain CDR1 mutants isolated from the L1 library and by chain shuffling Clone b4/12 3B3 3B3/L1.1 3B3/L1.2 3B3/L1.3 3B3/L1.4 CS CS CS CS 27A S S K T P Q N R R N Residue position 28 29 30 31 32 Kd (parent) Kd (mutant) Kd (b4/12) Kd (mutant) I I E V L L I I I I R R R D A S R R R R R R R R R R R R R R 1 8.2 0.16 0.7 1.1 3.5 NA NA NA NA — 8.2 1.3 5.7 9.0 29 ND ND ND ND R R F Y H D R S G W S S G R R G S S S S The clone designated 3B3 has been described. 3B3 contains mutations in CDR1-H and CDR3-H only. Clones isolated from the L1 library contain the L1 designation in their name and in the examples are paired with the heavy chain from 3B3. Dissociation constants were determined with the BIAcore instrument from Pharmacia. The ratio of parent and mutant Kd values gives the fold increase in affinity for this mutagenesis step which is attributed to changes in this region. The ratio of Kd b4/12 and mutant Kd gives the overall improvement in affinity from the starting Fab b4/12. Clones isolated from chain shuffling experiments have been described and characterized (Barbas et al., 1993b) and are given the CS designation. The CS clones contain multiple mutations in the light chain variable region paired with the original heavy chain of b4/12 and are of an affinity similar to b4/12. For these CS clones ratios of Kd values were not applicable (NA) and were not determined (ND) with the BIAcore instrument. 395 Affinity Maturation of a Human Antibody Table 2. Light chain CDR3 mutants isolated from L3 libraries and by chain shuffling Clone 89 90 91 Residue position 92 93 94 95 96 97 Kd (parent) Kd (mutant) Kd (b4/12) Kd (mutant) b4/12 3B3/L1.4 3B3/L1.4L3.1 3B3/L1.4L3.2 3B3/L1.4L3.3 3B3/L1.4L3.4 3B3/L1.4L3.5 3B3/L1.4L3.6 3B3/L1.4L3.7 3B3/L1.4L3.8 3B3/L1.4L3.9 3B3/L1.4L3.10 3B3/L1.4L3.11 3B3/L1.4L3.12 3B3/L3.13 3B3/L3.14 3B3/L3.15 3B3/L3.16 CS CS CS Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q V V V L T T S K M M Q T T T V Q V K T T T Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y G G G G G G G G G G G G G G G G G G G G G S S Q N V G D F D T L S H I S F A T S T S Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y T T T T T T T T T T T T T T T T T T T T S 1 3.5 0.43 0.52 0.18 0.49 0.44 1.0 0.05 0.36 0.50 3.0 3.3 1.7 1 5.6 0.23 3.2 NA NA NA 1 29 12 15 5.2 14 13 29 1.5 10 15 87 96 49 8.2 46 1.9 26 ND ND ND A A W R R W G D G G D R R R A W G G G G G S S S G G S R S R F S G G G S P S G S A S Clones containing CDR1-L mutations contain the L1 designation in their name and in the examples above are paired with the heavy chain from 3B3. Clones containing CDR3-L mutations contain the L3 designation in their name. Dissociation constants were determined with the BIAcore instrument from Pharmacia. The ratio of parent and mutant Kd values gives the fold increase in affinity for this mutagenesis step which is attributed to changes in this region. The ratio of Kd b4/12 and mutant Kd gives the overall improvement in affinity from the starting Fab b4/12. Clones isolated from chain shuffling experiments have been described and characterized (Barbas et al., 1993b) and are given the CS designation. The CS clones contain multiple mutations in the light chain variable region paired with the original heavy chain from b4/12 and are of an affinity similar to b4/12. For these CS clones ratios of Kd values were not applicable (NA) and were not determined (ND) with the BIAcore instrument. selection, indicating the increased efficacy of the off-rate selection procedure. Previously, an antigenbased off-rate selection was reported by Hawkins et al. (1992). This technique was not considered here due to the very high cost of the gp120 antigen. The second sequential walk involved CDR1-H to CDR3-H to CDR3-L steps. In this experiment, 4 × 108 variants in CDR3-L were produced, paired with the heavy chain from 3B3; library L3B. Following the selective protocol utilized for the L3A library, four unique clones were characterized following sequencing. Clone 3B3/L3.14 was improved 5.6-fold in affinity as compared to 3B3 and 46-fold as compared to b4.12. This clone was the predominant clone obtained after selection as revealed by sequencing. The third sequential CDR walk involved CDR1-H to CDR3-H97-100 to CDR3-H100A-E. This involved the construction of library H3B3 wherein 1 × 108 variants in the H100A-E region were generated in the context of clone 3B3. Following a selective regime involving antibody assisted off-rate selection, seven clones of highly related sequence were isolated and characterized (Table 3). All but one clone was improved in affinity for antigen, with the highest affinity clone h1.3B/h3.33 being improved 7.7-fold as compared to 3B3 and 63-fold as compared to b4/12. The last CDR walk involved a re-investigation of the single step walk to CDR1-H. Previously, library H1 had been generated and selected to yield a sub-library of functional CDR1-H variants of b4/12. The individual clones were not characterized and were utilized as a collection in subsequent mutagenesis of CDR3-H generating among others, clone 3B3 (Barbas et al., 1994). The sequence at position H31 to H33 of 3B3 was Asn-Phe-Thr. This sequence generates a potential glycosylation site in CDR1-H. To circumvent this sequence, as well as to identify a more optimal sequence in CDR1-H, library H1 was selected for an additional six rounds of panning beyond that previously reported. The results of this selection are given in Table 4. Clones lacking the potential glycosylation site were characterized. Clone h1.1 exhibited a 3.9-fold improvement in affinity for antigen. Analysis of selected CDRs Analysis of CDR1-H variants shown in Table 4 demonstrated a strong selection towards a consensus sequence. H31 was exclusively His or Asn, suggesting a functional role for the conservation of the d-amino group, which is a shared element in the side chains of these residues. H32 was exclusively an aromatic residue, most frequently Phe. H33 was exclusively Thr. CDR1-H of b4/12 is predicted to have a class 1 canonical structure. It contains the key residues which define this class: Ala at 24, Gly at 26, Tyr at 27, Phe at 29, Ile at 34, and Arg at 94 (Chothia et al., 1989, 1992). Interestingly, selection at H34 revealed Ile, Leu, or Val, all of which fall within those described for a class 1 structure. Position H35 revealed a selection towards the parental residue 396 Affinity Maturation of a Human Antibody Table 3. Heavy chain CDR3 mutants isolated from the H3B3 library and by chain shuffling b4/12 h1.3B/h3.3(3B3) h3.3 h1.3B/h3.32 h1.3B/h3.33 h1.3B/h3.34 h1.3B/h3.35 h1.3B/h3.36 h1.3B/h3.38 h1.3B/h3.39 CS CS CS CS 95 96 97 98 99 100 Residue position A B C D E F G I J K 101 102 Kd (parent) Kd (mutant) V V V V V V V V V V V V V V G G G G G G G G G G G G G G P E E E E E E E E E E E E P Y W W W W W W W W W W W W Y S G G G G G G G G G T T T T W W W W W W W W W W W W W W D D D E E E H D T D D D D D D D D D D D D D D D D D A D N N N N N N N N N N N N N N Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y M M M M M M M M M M M M M M D D D D D D D D D D D D D D V V V V V V V V V V V V V V 1 ND 7.9 1.8 7.7 5.9 5.6 0.43 4.4 5.5 NA NA NA NA D D D Q M M Q Q Q Q D D M D S S S F F R R R R V S F D S P P P R R R R R R R P P P P Q Q Q F Y F Y Y F Y Q Q Q Q Kd (b4/12) Kd (mutant) 1 8.2 7.9 15 63 48 46 3.5 36 45 ND ND ND ND Heavy chain CDR1 mutants contain the h1 designation in their name and in the examples above all Fabs contain the original light chain from b4/12. Clones containing CDR3-H mutations contain the h3 designation in their name. Clone h3.3 was created by swapping the mutant h1.3B CDR1-H region of 3B3 with the original CDR1-H region of b4/12. Dissociation constants were determined with the BIAcore instrument from Pharmacia. The ratio of parent and mutant Kd values gives the fold increase in affinity for this mutagenesis step which is attributed to changes in this region. The ratio of Kd b4/12 and mutant Kd gives the overall improvement in affinity from the starting Fab b4/12. Clones isolated from chain shuffling experiments have been described and characterized (Barbas et al., 1993b) and are given the CS designation. The CS clones contain multiple mutations in the heavy chain variable region paired with the original b4/12 light chain and are of an affinity similar to b4/12. For these CS clones ratios of Kd were not applicable (NA) and were not determined (ND) with the BIAcore instrument. His at this position. Correlation of affinity with sequence suggests that His at H31 is a key element in the increased affinities of clones h1.1-4. CDR1-H of h1.1 and h1.3B differ in sequence only at position H31 but differ by approximately fourfold in affinity. Variation in the aliphatic side-chains of the residue at H34 had little effect on the affinity of these clones. Previous analysis of CDR3-H variants derived by selection of the H97 to 100 region suggested the Pro Table 4. Heavy Chain CDR1 mutants isolated from the H1 library and by chain shuffling Clone 31 b4/I2 h1.1 h1.3 h1.4 h1.2 h1.5 h1.6 h1.3B/h3.3(3B3) CS CS N H H H N N N N N N Residue position 32 33 34 35 F F F F Y F W F F F V T T T T T T T T T I V L I L L I V V V H H H M Q I M H H H Kd (b4/12) Kd (mutant) 1 3.9 3.2 2.6 ND ND ND 8.2 ND ND Heavy chain CDR1 mutants contain the h1 designation in their name and in the examples above all Fabs contain the original light chain from b4/12. Clones containing CDR3-H mutations contain the h3 designation in their name. Dissociation constants were determined with the BIAcore instrument from Pharmacia. The ratio of Kd b4/12 and mutant Kd gives the overall improvement in affinity from the starting Fab b4/12 and in all cases except clone 3B3 gives the fold increase in affinity for this mutagenesis step which is attributed to changes in this region. Clones isolated from chain shuffling experiments have been described and characterized (Barbas et al., 1993b) and are given the CS designation. The CS clones contain multiple mutations in the heavy chain variable region paired with the original b4/12 light chain and are of an affinity similar to b4/12. For these CS clones ratios of Kd were not determined (ND) with the BIAcore instrument. Clones containing the potential glycosylation site Asn-X-Thr were not characterized with the exception of 3B3. to Glu change to be associated with the higher affinity of clone 3B3. The CDR1-H segment of 3B3 contributes little to the increase in affinity observed in 3B3. This is demonstrated by swapping the CDR1-H region of b4/12 for that of 3B3. The resulting clone h3.3 bound with approximately the same affinity as 3B3 (Table 3). Sequence analysis of the H100A to E region selected in library H3B3 reveals a set of highly related clones. Position H100A was predominately Glu or Asp. H100B was Gln or Met. Met at this position was observed in a clone derived by chain shuffling. H100C was predominately Arg, but Phe and Val were also observed. Phe had been observed at H100C in a clone derived by chain shuffling. Position H100D was exclusively Arg and H100E was either Phe or Tyr. Within this set of clones are several related by single point changes. Clone h1.3B/h3.35 and h1.3B/h3.36 differ only at H100A, which is His and Asp in the respective clones. They differ in affinity, however, by 13-fold. The affinity can be completely restored as shown in clone h1.3B/h3.39 with the mutation of Arg at H100C to Val. This result highlights the interdependence of residues in this region. CDR1-L sequences shown in Table 1 indicate that a number of the targeted positions accept a wide range of residues with little effect on affinity. Compare 3B3/L1.2 and 3B3/L1.3. Conservation of Arg at L32 is noted in b4/12 and all variants. Chain shuffling had produced variants with Asn or Arg at L27A and Ser, Gly, or Trp at L29. None of these mutations is noted in the variants obtained by saturation mutagenesis. The highest affinity variant differs from the parental sequence in five of the six selected regions with a significant change in the net charge of this region from +3 to 0. CDR1-L of b4/12 and selected variants do not possess sequences 397 Affinity Maturation of a Human Antibody Table 5. Binding kinetics of b4/12, b4/12 mutants and sCD4 Clone kon /104 (M−1 s−1 ) koff /10−4 (s−1 ) sCD4 b4/12 h1.1 h1.3Bh3.33 h1.1h3.33 h1.1h3.33/L1.4L3.14 h1.1h3.33/L1.4L3.11 h1.1h3.33/L3.14 L1.4L3.14 L1.4L3.11 1.7 2 0.03 7.6 2 0.02 14 2 1.0 7.0 2 0.8 15 2 2.6 7.8 2 0.7 9.2 2 0.9 7.0 2 0.1 12 2 1.1 14 2 1.2 1.0 2 0.007 4.8 2 0.02 2.3 2 0.03 0.071 2 0.006 0.034 2 0.001 0.012 2 0.002 0.090 2 0.007 0.083 2 0.005 4.1 2 0.02 4.2 2 0.02 Kd (nm) 5.9 6.3 1.6 0.10 0.022 0.015 0.097 0.12 3.4 3.9 Kd (b4/12) Kd (mutant) NA 1 3.9 63 286 420 65 53 1.9 1.6 DDGI (kcal/mol) 0.094 −1.54 −2.33 −2.02 −1.39 −1.17 Binding kinetics were determined using the BIAcore instrument from Pharmacia. The envelope glycoprotein gp120 was immobilized on the biosensor chip. The Kd value was calculated from koff /kon . The ratio of Kd values indicates the fold increase in binding affinity as compared to the Fab b4/12. DDGI was calculated using equation (1) and DDG as + RT ln(Kd b4/12/Kd mutant). which fall within those described for classification of this region into a canonical structure class (Chothia et al., 1989). CDR3-L sequences demonstrated a striking selection for Gly at L92 in all 16 characterized mutants (Table 2). This position was conserved in all mutants derived by chain shuffling. L93 was predominately Gly or Arg. Gly at this position was the predominant mutant obtained by chain shuffling. Other positions were more permissive. Four clones are related by amino acid changes at a single position, L95. Clone 3B3/L1.4L3.3, which has Val at this position binds gp120 with an affinity 18-fold lower than 3B3/L1.4L3.11, which has His at this position. Conversion of His to either Ser or Ile in 3B3/L1.4L3.10 and 3B3.L1.4L3.12 has little effect on affinity. This highly related set has Thr at position L90, which was the only mutant at this position obtained by chain shuffling. With respect to canonical structures predicted for this region, the parental antibody and all variants except 3B3/L3.14 do not contain conformation defining residues of the appropriate identity required for classification. Selection for Gln at L90 and Pro at L94 in this clone classifies this as a class 2 canonical structure (Chothia et al., 1989). This was the highest affinity clone selected in the CDR1-H to CDR3-H to CDR3-L walk and differs radically in sequence from 3B3/L1.4L3.11, which was the optimal clone selected in the context of the optimized CDR1-L region L1.4. Investigation of additivity effects in the antibody combining site An alternative CDR walking strategy involves a parallel optimization strategy wherein CDRs are independently optimized and recombined in a single clone. This strategy is based on the assumption that in most cases additivity will be observed when non-interacting mutations are combined (Wells, 1990). To assess the applicability of additivity principles to the antibody combining site, six new antibodies were created by shuffling optimized CDR regions. The resulting clones were expressed, purified, and the affinity for antigen determined. The results are shown in Table 5. Combination of the most optimal CDR3-H region with the most optimal CDR1-H region produced clone h1.1h3.33, which was improved 286-fold in its affinity for antigen as compared to b4/12. In this case the measured affinity is in good agreement with that predicted assuming additivity; 286-fold as compared to 246. Recruitment of CDR1-L and CDR3-L segments L1.4 and L3.14 into this clone resulted in only a 1.5-fold increase in affinity, see clone h1.1h3.33/L1.4L3.14. These segments yielded 3.5 and 5.6-fold increases in affinity, respectively, in the context of the antibody in which they were originally selected. If simple additivity had been operative, a 19.6-fold improvement in affinity relative to h1.1h3.3 would have been observed and an antibody with a 5600-fold improvement in affinity would have resulted. These two light chain CDRs were the products of independent selections. To investigate the source of the discrepancy between the observed and predicted results, clone h1.1h3.3/L3.14 was constructed. This clone introduces only a single optimized CDR3-L region. A 5.5-fold decrease in affinity was observed as compared to clone h1.1h3.33. This result demonstrates an incompatibility of the L3.14 CDR3-L region with the h1.1h3.33 heavy chain. In clone h1.1h3.33/L1.4L3.11, the light chain containing the sequentially optimized CDR1 and CDR3 regions L1.4 and L3.11 was combined with the h1.1h3.33 heavy chain. The resulting clone demonstrated an affinity 4.4-fold lower than that observed for h1.1h3.3 paired with the original light chain derived from b4/12. Simple additivity would have produced a 12-fold increase in affinity. The results obtained by combining optimized light chain regions with heavy chain variants are unpredictable. Combination of L1.4L3.14 and L1.4L3.11 light chains with the original heavy chain of b4/12 resulted in approximately a twofold increase in affinity in each case. 398 Changes in the binding kinetics of improved antibodies Insight into the mechanism of affinity improvement is provided by determination of the binding kinetics using the BIAcore instrument. As shown in Table 5, improvements in affinity are dominated by decreases in off-rates. Only a twofold variation in on-rate was observed compared to a 400-fold variation in off-rate. This trend was maintained in the 34 additional clones reported in Tables 1 to 4 (data not shown). Off-rate is expected to be most affected if phage selections are performed under conditions approaching equilibrium (Lowman & Wells, 1993). The kinetics of soluble CD4 (sCD4) binding to gp120 were determined for comparative purposes. CD4 is the natural ligand for gp120. The affinity determined by BIAcore is in close agreement with the reported value of 3 nM (Ryu et al., 1990) This measurement serves to correlate the affinities reported here with those determined by other methods. The apparent Kd value of Fab b4/12 as determined by competition ELISA is 9 nM (Barbas et al., 1992b). Affinity improvements determined with the BIAcore instrument are generally well correlated with those determined with other methods (Lowman & Wells, 1993; Kelley & O’Connell, 1993). Antibody binding kinetics compare favorably with sCD4. Fab on-rate was faster in all cases than that observed for sCD4. The off-rate of sCD4 is 4.8-fold slower than b4/12 and 83-fold faster than the highest affinity Fab h1.1h3.33/L1.1L3.14. Since small quantities of aggregated protein can lead to artifactually slow off-rates (van der Merwe et al., 1993), the dissociation of Fabs b4/12 and h1.1h3.33 were studied following gel filtration chromatography. In both cases the off-rates determined for these proteins were within the values determined for the proteins that had not been subjected to this additional purification step. In an additional study the binding kinetics of Fab h1.1h3.33 binding to gp120 were determined in the reverse orientation, i.e. with Fab immobilized and gp120 as the soluble ligand. The on-rate of this interaction was 2.8-fold slower and the off-rate was 1.3-fold faster than that determined with gp120 immobilized. This variation is within the range expected since it is difficult to match the densities and orientations of the proteins which are immobilized. Thus, the differences in off-rates reported herein are not artifacts resulting from protein aggregation. Discussion Antibody b4/12 represents the most potent and broadly neutralizing human anti-HIV-1 antibody yet described (Burton et al., 1994). This antibody has been demonstrated to neutralize 75% of 36 primary isolates of HIV-1 tested at concentrations which could be achieved by passive immunization. This extends earlier work that demonstrated potent neutralization of laboratory adapted isolates of Affinity Maturation of a Human Antibody HIV-1 with monovalent Fab of this antibody (Barbas et al., 1992b). The Fab fragment of b4/12 was originally isolated from a phage display library prepared from an individual who had been HIV-1 positive for six years but had no symptoms of disease. The antibody binds a highly conformational epitope in the CD4-binding region of the HIV-1 envelope glycoprotein gp120. The epitope within the CD4 binding region of gp120 recognized by b4/12 is unique as compared to five other antibodies directed against this region and isolated from the same individual (Roben et al., 1994). To potentially improve the therapeutic or prophylactic efficacy of this antibody, we investigated strategies for the evolution of its affinity for antigen. Phage display provides a convenient format for the refinement of the contact between receptor and ligand or antibody and antigen which may result from unpredictable sequence changes in the region of interest. The limitations of the phage display approach have been considered elsewhere (Lowman & Wells, 1993; Barbas & Barbas, 1994). We have proposed that repeated introduction of diversity into CDRs followed by stringent selections should allow for the refinement of human antibodies to levels of affinity far beyond those generated by the immune response (Barbas et al., 1994). CDR targeted mutagenesis is advantageous since optimization of these regions is most likely to improve affinity and least likely to create problems of immunogenicity. CDR walking involves targeted mutagenesis of CDR regions and selection for fitness with monovalent phage display (Barbas et al., 1994). The free energy changes that result from the combination of sets of mutations may be expressed by the equation: DDGAB = DDGA + DDGB + DDGI (1) where DDGA and DDGB are changes in the free energy of mutants A and B, DDGAB is the change in free energy for the protein that combines the mutations of A and B, and DDGI is the change in the interaction energy between the two sets of residues (after Lowman & Wells, 1993). In accord with equation (1), two strategies are evident for the application of CDR walking, either sequential or parallel optimization of CDRs. The most appropriate route to the optimization of antibody affinity is dependent on the term DDGI . In this study we have investigated these two CDR walking strategies. Sequential CDR walking takes into account that DDGI may not always be negligible and that optimal binding may result from the interdependence of CDR loops. Such interdependence could result from coordinated structural changes on binding antigen and is supported by recent evidence which suggests induced-fit mechanisms may best describe the binding observed in some antibody–antigen complexes (Wilson & Stanfield, 1993). Together with a previous study we have performed five sequential CDR walking experiments. In the previous study a library of CDR1-H variants, H1 library, with five Affinity Maturation of a Human Antibody residues randomized was selected by panning against gp120. The collection of clones which resulted was then used in the construction of a CDR3-H library where four additional residues were randomized. The resulting library was selected by an additional six rounds of panning. A variety of clones with improved binding were produced. Correlations within this set of selected progeny were examined to establish functional trends which may be predictive of the activity of subsequent generations of b4/12 (Barbas et al., 1994). With the gp120 envelope protein derived from the MN isolate of HIV-1, affinity was well correlated with neutralizing ability. Correlation of affinity with neutralizing ability in the HIV-1 system is very dependent on the epitope recognized (Roben et al., 1994) and has been observed elsewhere (Nakamura et al., 1993). A 54-fold improvement in neutralizing potency was determined for the highest affinity clone 3B3. The ability of these evolved monovalent Fabs to neutralize with potencies equivalent or better than sCD4 is unique. In a recent multi-center study of human and mouse anti-HIV-1 antibodies, no bivalent antibody has demonstrated such potency (D’Souza et al., 1994). Furthermore, four primary clinical isolates of HIV-1 not neutralized by the parent b4/12 were neutralized by the highest affinity clone selected, 3B3. Clone 3B3 was improved 8.2-fold in binding gp120 for a final affinity of 0.77 nM. These experiments suggest that improvement in the affinity of b4/12 should improve its therapeutic potency. To dissect the relative contribution of the mutant CDR1-H and CDR3-H of 3B3, CDR1-H was converted to wild-type sequence in clone h3.3. This clone retained the binding affinity of 3B3. This experiment suggested that the use of the pool of selected CDR1-H variants in this experiment was not the best starting point for further optimization. With stringent selection higher affinity antibodies from the original H1 library were selected as shown in Table 4. Based on this result all subsequent CDR walks reported in this study use a single optimized clone on which diversity is incorporated, as opposed to a collection of clones of selected but ill-defined binding characteristics. In every case examined, saturation mutagenesis within the CDRs followed by selection using phage display resulted in the improvement of affinity. Selection of clones with sub-nanomolar binding affinities was facilitated by incorporating an off-rate biased selection in the panning step (vida supra). Affinity gains at each step of the walks were generally in small increments, about fourfold, with the exception of the two steps in CDR3-H which produced about an eightfold improvement in each case. Three different walks yielded clones of similarly high affinity 137-65 pM; 3B3/L1.4L3.11, 3B3/L3.14, and h1.3/ h3.33. The net improvement in affinity of these clones was 46 to 96-fold. If the free energy change of a multiple mutant (in this case several optimized CDRs combined) is nearly equal to the sum of the free energy changes 399 observed in the point mutants (or singly optimized CDRs) the free energy changes are said to be additive; DDGI is negligible. The reader should note that additivity in free energy changes results in multiplicative changes in Ka ; a fourfold improved affinity in CDR1 combined with a fourfold improved affinity in CDR2 results in a clone with a 16-fold improvement in affinity. In protein engineering studies this is generally the case unless residues interact with each other by direct contact or indirectly through electrostatic or structural perturbations (Wells, 1990). This (interaction) will certainly be the case for optimization within the long CDR loops of CDR1-L, CDR2-H and CDR3-H which require optimization of more than five residues and at least two sequential optimizations within the given CDR. Parallel CDR walking makes the assumption that the optimized loops will exhibit additivity in free energy changes when the individually optimized loops are combined. The advantage of a parallel approach is the speed with which the final desired clone might be obtained. In a study guided by a high resolution X-ray structure and extensive alanine scanning mutagenesis an additivity-based approach has been convincingly demonstrated in the human growth hormone system with the generation of a hormone with a 380-fold improvement in affinity for its receptor (Lowman et al., 1991; Lowman & Wells, 1993). Within the present system we are guided by the data base of available antibody structures and hints at somatic hotspots observed in chain shuffling experiments. Within antibodies of known structure, CDR-CDR contacts are well documented. However, from crystallographic studies it is not clear how significant these interactions are or to what extent change in sequence in interacting regions will affect affinity for antigen. Examples of the utilization of additivity in the affinity optimization with antibodies have been reported (Foote & Winter, 1992; Hawkins et al., 1992, 1993; Riechmann & Weill, 1993). The extent of mutation in the combined segments in these reports is generally much lower than that reported here. Encouraged by these results, a parallel CDR walking mutagenesis strategy was examined where independently optimized CDRs were combined. Table 5 summarizes the binding data for six combinations of CDRs. Only in the combination of CDR1-H with CDR3-H was additivity observed. This result could not be rationalized by knowledge of structures of antibodies, since these CDRs are known to form many van der Waals’ contacts. Changes within the CDR1-H in h1.1 are, however, modest in comparison to selected changes in other CDRs. Recently interchain CDR-CDR contacts in antibodies of known structure have been tabulated (Padlan, 1994). This study revealed that CDR3-L generally forms van der Waals’ contacts with all the CDRs of the heavy chain whereas CDR3-H generally interacts with all the CDRs of the light chain. CDR1-L, however, contacts predominately CDR3-H. Overall the results of combinations were 400 unpredictable but did lead to some modest improvements over the clones derived from sequential selection. The large negative interaction energies observed for most combinations given in Table 5 are sufficient to undermine the small 0.82 kcal/mol increases which would accompany the about fourfold increases in affinity of the individual optimized CDRs. The failure of the additivity-based strategy is most likely due to direct interactions though long-range non-additive effects are also possible (LiCata & Ackers, 1995). The −2.02 kcal/mol interaction energy observed in the construction of h1.1h3.33/L3.14 suggests that to a large degree non-additivity is observed in these cases due to the incompatibility, perhaps due to steric interaction, directly or through antigen, of the optimized CDR3-L regions with independently optimized CDR3-H region. The highest affinity clone produced by parallel optimization, h1.1h3.3/ L1.4L3.1 was improved 420-fold in affinity for antigen. This is only a 4.4-fold improvement over 3B3/L1.4L3.11, the highest affinity clone obtained in the sequential approach, and would likely have been achieved in the next sequential optimization step. The highest affinity Fab generated in these studies, h1.1h3.3/L1.4L3.14, bound gp120 with a 15 pM dissociation constant. Since most of the CDR sequence of this antibody remains to be optimized, the limit to which one may improve affinity remains to be reached. Examination of the kinetics of binding using the BIAcore instrument revealed that improvements in affinity were dominated by slowing of the off-rate. The highest affinity clone demonstrated a 400-fold decrease in off-rate as compared to b4/12. The BIAcore sensorgrams for Fabs b4/12 and h1.1h3.3/ L1.4L3.14 are given in Figure 2. The slowing in off-rate observed here is in agreement with that observed in the affinity maturation of human growth hormone (Lowman & Wells, 1993) and is expected since experimentally selective pressure is provided to enrich for the most stable Fab/gp120 Affinity Maturation of a Human Antibody complex. In comparison to the natural ligand for CD4, the Fab fragments bound with a four to eightfold faster on-rate than sCD4 and in the best case, with an off-rate 83-fold slower than sCD4. Further improvements in off-rates cannot be measured with the BIAcore instrument which is limited to off-rates faster than 10−7 s−1 This study demonstrates that the sequential CDR walking strategy is an effective route to the generation of very high affinity antibodies. The large affinity increases observed in the optimization of the CDR3-H region suggest that the most expeditious route to improve affinity may be optimization of this region. This is likely to be particularly true for an antibody such as b4/12, which has an extended CDR3-H of 18 residues. Since human and murine CDR3-H regions have mean lengths of approximately 12 and 9 residues, respectively (Wu et al., 1993), which are generated by mechanisms most analogous to saturation mutagenesis (Sanz, 1991), it should not be surprising that the humoral response has not selected the combinatorially most optimized segment in this region. The ability to create very high affinity antibodies has implications for their use as catalysts, since optimization of affinity for the transition-state of a reaction is directly correlated with enhanced catalytic efficiency (Lerner et al., 1991). For therapeutic and prophylactic application, very high affinity antibodies have important theoretical advantages if potency, for example in viral neutralization assays, is strictly correlated with affinity. In therapeutic applications the quantity of protein required could be drastically reduced. For example, if a serum level of 25 mg/ml were required for a given indication, a 420-fold improvement would reduce the quantity of protein required to 0.06 mg/ml. In prophylactic applications, increased potency could dramatically extend the time in which the individual is protected from disease. Given the 21 day half-life of an antibody in the body Figure 2. Comparison of BIAcore sensorgrams obtained for the binding of Fabs b4/12 and h1.1h3.33/ L1.4L3.14 to immobilized gp120. RU, resonance units. 401 Affinity Maturation of a Human Antibody (Capon et al., 1989), an antibody which provided protection through a single half-life would, if improved 420-fold and given in the same quantity, provide protection for six months. Materials and Methods Reagents, strains and vectors Oligonucleotides were from Operon Technologies (Alameda, CA). Escherichia coli, phage, and the phagemid vector pComb3H are as described (Barbas et al., 1991; Barbas & Wagner, 1995)). Restriction enzymes were obtained from Boehringer Mannheim. Taq polymerase and DNA ligase were obtained from Promega and Gibco-BRL, respectively. The recombinant protein gp120 IIIB was purchased from Intracel Corp. (Cambridge, MA). Reagents for surface plasmon resonance experiments were obtained from Pharmacia. Construction and selection of libraries The construction of library H1 has been described (Barbas et al., 1994). This library was selected for gp120 binding by an additional six panning cycles beyond the four cycles performed in the previous report. For selections, gp120 IIIB was immobilized on Costar 3690 microtiter wells. One well, coated with 1 ag of gp120 overnight, was used for each round of selection. To construct library H3B3 clone 3B3 was utilized as a template for overlap PCR mutagenesis as described (Barbas et al., 1994). The two fragments required for this procedure were obtained with oligonucleotide primer pairs FTX3 (Barbas et al., 1992a) and 3B3C350B (5' - CCAGACGTCCATATAATAATTGTCMNNMNNMN NMNNMNNCCAACCCCACTCCCCCACTCT - 3') and with R3B (Barbas et al., 1993b) and H4H3OF (5'-GACAATTATTATATGGACGTCTGG-3'), respectively. N = A, C, G, or T,: M = A or C which gives K = T or G in the complementary strand (NNK doping strategy). Following fusion and gel purification, the fragment was digested with restriction enzymes XhoI and SpeI, purified and ligated into the phagemid vector pComb3H containing the b4/12 light chain from which a 300 base-pair stuffer fragment in the heavy chain cloning sites XhoI and SpeI had been removed. The ligation products were introduced into XL1-Blue cells by electrotransformation to yield 1 × 108 independent transformants. The library was selected by panning against gp120 IIIB for four cycles. In the fifth through the tenth cycles of selection, 8 mM Fab 3B3 was added to the well following removal of unbound phage by washing. The plate was then incubated with the Fab for three hours at 37°C (off-rate selection). The well was then washed five times and bound phage were eluted and propagated as usual. Library L1 was prepared in a similar procedure using 3B3 as a template for overlap PCR mutagenesis with oligonucleotide primer pairs KEF (Barbas et al., 1993b) and HIVCDR1-OV-B (5'-AGGTTTGTGCTGGTACCAGGCTACMNNMNNMNNMNNMNNMNNGTGACTGGACCTACAGGAGAAGGT-3') and T7B (Barbas et al., 1993b) and HIVCDR1-FO (5'-GTAGCCTGGTACCAGCACAAACCT-3'). Following fusion and gel purification, the fragment was digested with restriction enzymes SacI and XbaI, purified and ligated into the phagemid vector pComb3H containing the 3B3 heavy chain wherein a 1.2 kb stuffer fragment between SacI and XbaI had been removed. The ligation products were introduced into XL1-Blue cells by electrotransformation to yield 5 × 108 independent transformants. Libraries L3A and L3B were prepared with primer pairs KEF and H4KCDR3-B0 (5'-CAGTTTGGTCCCCTGGCCAAAAGTGTAMNNMNNMNNMNNATAMNNCTGACAGTAGTACAGTGCAAAGTC-3') and T7B and HVKFR4-FO (5'TACACTTTTGGCCAGGGGACCAAACTG-3'). Library L3A was prepared with 3B3/L1.4 as the PCR template and library L3B was prepared with 3B3 as the template. Library sizes were 2 × 108 and 4 × 108, respectively. Library L3A was selected for binding to gp120 with five cycles of panning followed by five cycles of panning with 3B3 used in an off-rate selection as described above. Library L3B was selected with three panning cycles followed by four cycles with Fab 3B3 off-rate selection. Expression of soluble Fab Following selection, phagemid DNA was isolated and digested with SfiI and the approximately 1500 bp cassette was ligated with SfiI-digested pPhoA. pPhoA was constructed by insertion of the phoA promoter and antibody encoding cassette between the AatII and HindIII sites of pBR322. The phoA promoter was isolated from genomic E. coli DNA by PCR. The antibody encoding expression cassette was derived from pComb3H. The pPhoA vector was designed to be compatible with pComb3H and is based on the previously reported pAK19 expression vector by Carter et al. (1991). This vector is designed for use in fermentation, though for the experiments reported here fermentation was not necessary. Expression of Fab in shaker flasks with this vector was achieved by phosphate starvation as described (Carter et al., 1992). In some cases transfer to pPhoA was not performed. In these cases NheI/SpeI restriction enzyme digestion and religation of the pComb3H vector leads to efficient expression of soluble Fab as described (Barbas et al., 1991). Following screening for binding activity in an enzyme linked immunosorbant assay (ELISA), the complete amino acid sequences of the variable regions of ELISA positive clones were deduced by dideoxy sequencing as described. Fab was purified by affinity chromatography as described (Barbas et al., 1992b). To examine if protein aggregates are isolated using this protocol, an additional gel filtration purification step was performed on Fabs b4/12 and h1.1h3.33. Gel filtration chromatography was performed using a standardized HiPrep 16/60 Sephacryl S-100 column equilibrated and run (0.5 ml/min) in PBS buffer. Fabs eluted at 50 kDa and were collected and used for BIAcore studies. Recruitment of optimized complementarity determining regions Shuffling of optimized CDR regions was facilitated by an encoded SacII site in framework three of the heavy chain and a KpnI site in framework two of the light chain. The heavy chain h1.1h3.33 was constructed from the 210 bp XhoI/SacII fragment of h1.1 and the 470 bp SacII/SpeI fragment of h1.3B/h3.33. Light chain L14L3.11 was constructed from the 100 bp SacI/KpnI fragment of L1.4 and the 550 bp KpnI/XbaI fragment of L3.14. Combinations of mutant light and heavy chains were constructed in the pPhoA vector. Surface plasmon resonance The kinetic constants for the binding of Fab to gp120 IIIB were determined by surface plasmon resonance- 402 based measurements using the BIAcore instrument from Pharmacia. The sensor chip was activated for immobilization with N-hydroxysuccinimide and N-ethyl-N '-(3-diethyl aminopropyl) carbodiimide. The protein, gp120, was coupled to the surface by injection of 50 ml of a 50 mg/ml sample. Excess activated esters were quenched with 15 ml of 1 M ethanolamine (pH 8.5). Typically 4000 resonance units were immobilized. Binding of Fab fragments to immobilized gp120 was studied by injection of Fab in a range of concentrations (0.5 to 10 mg/ml) at a flow rate of 5 ml/min. The binding surface was regenerated with HCl, 1 M NaCl (pH 3) and remained active for 030 measurements. Each immobilization surface was standardized by determination of the binding kinetics of clone 3B3 so that relative values could be determined. The association rate (kon ) was determined by studying the rate of binding of the protein to the surface at five different protein concentrations ranging from 10 mg/ml to 0.5 mg/ ml. The dissociation rate (koff ) was determined by increasing flow rate to 50 ml/min after the association phase. The koff value is the average of three measurements. The kon and koff values were calculated using BIAcoreTM kinetics evaluation software. The derived equilibrium dissociation constants (Kd ) were calculated from the rate constants. To exclude protein aggregation as a cause for the slowing of the off-rates of mutant Fabs, the off-rates of Fabs b4/12 and h1.1h3.33 were determined following gel filtration purification to remove any aggregated protein. The binding of gp120 to immobilized Fab b4/12 was also studied. Fab was immobilized and kinetics determined as described above using five concentrations of gp120 ranging from 20 mg/ml to 1.25 mg/ml to determine association rate and 50 mg/ml injections to determine off-rate. Acknowledgements We thank Mike Picque for molecular graphics, Richard Lerner and Robyn L. Stanfield for helpful discussions, and Ray Sweete for sCD4. We thank a reviewer for calling the van der Merwe et al. reference to our attention. C.F.B. is a Scholar of the American Foundation for AIDS Research and recipient of an Investigator Award from The Cancer Research Institute. This work was supported by the National Institutes of Health grant RO1 AI 37470 to C.F.B. References Barbas, C. F., III & Wagner, J. (1995). Synthetic human antibodies: selecting and evolving functional proteins. Methods, 8, 94–103. Barbas, C. F., III, Kang, A. S., Lerner, R. A. & Benkovic, S. J. (1991). Assembly of combinatorial antibody libraries on phage surfaces: The gene III site. Proc. Natl Acad. Sci. USA, 88, 7978–7982. Barbas, C. F., III, Bain, J. D., Hoekstra, D. M. & Lerner, R. A. (1992a). Semisynthetic combinatorial antibody libraries: a chemical solution to the diversity problem. Proc. Natl Acad. Sci. USA, 89, 4457–4461. Barbas, C. F., III, Björling, E., Chiodi, F., Dunlop, N., Cababa, D., Jones, T. M., Zebedee, S. L., Persson, M. A. A., Nara, P. L., Norrby, E. & Burton, D. R. (1992b). Recombinant human Fab fragments neutralize human type 1 immunodeficiency virus in vitro. Proc. Natl Acad. Sci. USA, 89, 9339–9343. Affinity Maturation of a Human Antibody Barbas, C. F., III, Amberg, W., Simoncsits, A., Jones, T. M. & Lerner, R. A. (1993a). Selection of human anti-hapten antibodies from semisynthetic libraries. Gene, 137, 57–62. Barbas, C. F., III, Collet, T. A., Amberg, W., Roben, P., Binley, J. M., Hoekstra, D., Cababa, D., Jones, T. M., Williamson, R. A., Pilkington, G. R., Haigwood, N. L., Cabezas, E., Satterthwait, A. C., Sanz, I. & Burton, D. R. (1993b). Molecular profile of an antibody response to HIV-1 as probed by combinatorial libraries. J. Mol. Biol. 230, 812–823. Barbas, C. F., III, Hu, D., Dunlop, N., Sawyer, L., Cababa, D., Hendry, R. M., Nara, P. L. & Burton, D. R. (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, 3809–3813. Barbas, S. M. & Barbas, C. F., III (1994). Filamentous phage display. Fibrinolysis, 8, 245–252. Behring, E. A. (1893). Die Geschichte der Diphtherie. pp. 186–187. Thieme, Leipzig. Burton, D. R. & Barbas, C. F., III (1994). Human antibodies from combinatorial libraries. Advan. Immunol. 57, 191–280. Burton, D. R., Pyati, J., Koduri, R., Sharp, S. J., Thornton, G. B., Parren, P. W. H. I., Sawyer, L. S. W., Hendry, R. M., Dunlop, N., Nara, P. L., Lamacchia, M., Garratty, E., Stiehm, E. R., Bryston, Y. J., Cao, Y., Moore, J. P., Ho, D. D. & Barbas, C. F., III (1994). Efficient neutralization of primary isolates of HIV-1 by a recombinant human monoclonal antibody. Science, 266, 1024–1027. Capon, D. J., Chamow, S. M., Mordenti, J., Marsters, S. A., Gregory, T., Mitsuya, H., Byrn, R. A., Lucas, C., Wurm, F. M., Groopman, J. E., Broder, S. & Smith, D. H. (1989). Designing CD4 immunoadhesins for AIDS therapy. Nature, 337, 525–531. Carter, P., Kelley, R. F., Rodrigues, M. L., Snedecor, B., Covarrubias, M., Velligan, M. D., Wong, W. L. T., Rowland, A. M., Kotts, C. E., Carver, M. E., Yang, M., Bourell, J. H., Shepard, H. M. & Henner, D. (1991). High level Escherichia coli expression and production of a bivalent humanized antibody fragment. Bio/Technology, 10, 163–167. Chothia, C., Lesk, A. M., Tramontano, A., Levitt, M., Smith-Gill, S. J., Air, G., Sheriff, S., Padlan, E. A., Davies, D., Tulip, W. R., Colman, P. M., Spinelli, S., Alzari, P. M. & Poljak, R. J. (1989). Conformations of immunoglobulin hypervariable regions. Nature, 342, 877–883. Chothia, C., Lesk, A. M., Gherardi, E., Tomlinson, I. M., Walter, G., Marks, J. M., Llewelyn, M. B. & Winter, G. (1992). The structural repertiore of human VH segments. J. Mol. Biol. 227, 799–817. D’Souza, M. P., Geyer, S. J., Hanson, C. V., Hendry, M., Milman, G. & Collaborating Investigators (1994). Evaluation of monoclonal antibodies to HIV-1 by neutralization and binding assays: an international collaboration. AIDS, 8, 169–181. Foote, J. & Winter, G. (1992). Antibody framework residues affecting the conformation of the hypervariable loops. J. Mol. Biol. 224, 487–499. Hawkins, R. E., Russell, S. J. & Winter, G. (1992). Selection of phage antibodies by binding affinity mimicking affinity maturation. J. Mol. Biol. 226, 889–896. Hawkins, R. E., Russell, S. J., Baier, M. & Winter, G. (1993). The contribution of contact and non-contact residues of antibody in the affinity of binding to antigen. J. Mol. Biol. 234, 958–964. 403 Affinity Maturation of a Human Antibody Kabat, E. A., Wu, T. T., Perry, H. M., Gottesman, K. S. & Foeller, C. (1991). Sequences of Proteins of Immunological Interest. 5th edit., US Department of Health and Human Services, Public Health Service, National Institutes of Health (NIH Publication no. 91-3242). Karlsson, R., Michaelsson, A. & Mattson, A. (1991). Kinetic analysis of monoclonal antibody-antigen interactions with a new biosensor based analytical system. J. Immunol. Methods, 145, 229–240. Kelley, R. F. & O’Connell, M. P. (1993). Thermodynamic analysis of an antibody functional epitope. Biochemistry, 32, 6828–6835. Lerner, R. A., Benkovic, S. J. & Schultz, P. G. (1991). At the crossroads of chemistry and immunology: catalytic antibodies. Science, 252, 659–667. LiCata, V. J. & Ackers, G. K. (1995). Long-range, small magnitude nonadditivity of mutational effects in proteins. Biochemistry, 34, 3133–3139. Lowman, H. B. & Wells, J. A. (1993). Affinity maturation of human growth hormone by monovalent phage display. J. Mol. Biol. 234, 564–578. Lowman, H. B., Bass, S. H., Simpson, N. & Wells, J. A. (1991). Selecting high-affinity binding proteins by monovalent phage display. Biochemistry, 30, 10832– 10838. Marquart, M., Deisenhofer, J., Huber, R. & Palm, W. (1980). Crystallographic refinement and atomic models of the intact immunoglobulin molecule Kol and its antigen-binding fragment at 3.0 Å and 1.9 Å resolution. J. Mol. Biol. 141, 369–391. Nakamura, G. R., Byrn, R., Wiles, D. M., Fox, J. A., Hobbs, M. R., Hastings, R., Wessling, H. C., Norcross, M. A., Fendly, B. M. & Berman, P. W. (1993). Strain specificity and binding affinity requirements of neutralizing monoclonal antibodies to the C4 domain of gp120 from human immunodeficiency virus type 1. J. Virol. 67, 6179–6191. Padlan, E. A. (1994). Anatomy of the antibody molecule. Mol. Immunol. 31, 169–217. Riechmann, L. & Weill, M. (1993). Phage display and selection of a site-directed randomized single-chain antibody Fv fragment for its affinity improvement. Biochemistry, 32, 8848–8855. Roben, P., Moore, J. P., Thali, M., Sodroski, J., Barbas, C. F., III & Burton, D. R. (1994). Recognition properties of a panel of human recombinant Fab fragments to the CD4 binding site of gp120 that show differing abilities to neutralize human immunodeficiency virus type 1. J. Virol. 68, 4821–4828. Ryu, S.-E., Kwong, P. D., Truneh, A., Porter, T. G., Arthos, J., Rosenberg, M., Dai, X., Xuong, N.-H., Axel, R., Sweet, R. W. & Hendrickson, W. A. (1990). Crystal structure of an HIV-binding recombinant fragment of human CD4. Nature, 348, 419–426. Sanz, I. (1991). Multiple mechanisms participate in the generation of diversity of human H chain CDR3 regions. J. Immunol. 147, 1720–1729. van der Merwe, P. A., Brown, M. H., Davis, S. J. & Barclay, A. N. (1993). Affinity and kinetic analysis of the interaction of the cell adhesion molecules rat CD2 and CD48. EMBO J. 12, 4945–4954. Wells, J. A. (1990). Perspectives in biochemistry: additivity of mutational effects in proteins. Biochemistry, 29, 8509–8517. Wilson, I. A. & Stanfield, R. L. (1993). Antibodyantigen interactions. Curr. Opinion Struct. Biol. 3, 113–118. Wu, T. T., Johnson, G. & Kabat, E. A. (1993). Length distribution of CDRH3 in antibodies. Protein: Struct. Funct. Genet. 16, 1–7. Edited by F. E. Cohen (Received 10 April 1995; accepted in revised form 5 September 1995)