Structural consequences of target epitope-directed functional alteration of an antibody. The case of anti-hen lysozyme antibody, HyHEL-10.

Decreased affinity of an antibody for a mutated epitope in an antigen can be enhanced and reversed by mutations in certain antibody residues. Here we describe the crystal structures of (a) the complex between a naturally mutated proteinaceous antigen and an antibody that was mutated and selected in vitro, and (b) the complex between the normal antigen and the mutated antibody. The mutated and selected antibody recognizes essentially the same epitope as in the wild-type antibody, indicating successful target site-directed functional alteration of the antibody. In comparing the structure of the mutated antigen-mutant antibody complex with the previously established structure of the wild-type antigen-wild-type antibody complex, we found that the enhanced affinity of the mutated antibody for the mutant antigen originated not from improvements in local complementarity around the mutated sites but from subtle and critical structural changes in nonmutated sites, including an increase in variable domain interactions. Our findings indicate that only a few mutations in the antigen-binding region of an antibody can lead to some structural changes in its paratopes, emphasizing the critical roles of the plasticity of loops in the complementarity-determining region and also the importance of the plasticity of the interaction between the variable regions of immunoglobulin heavy and light chains in determining the specificity of an antibody.

Often only a few mutations in the epitope of an antigen enable it to evade immunological attack. Major examples occur in viral coat proteins, e.g. hemagglutinin of influenza virus, gp120 of the human immunodeficiency virus, and human hepatitis B virus surface antigen (1)(2)(3)(4)(5). Despite only a few mutations in the antigenic epitope, the specificity and affinity of neutralizing antibodies for the mutated antigen can be completely abolished. The ability to generate specific antibodies directed against the mutated antigenic epitope thus should prove very useful. However, site-directed random mutation, chain shuffling, and DNA shuffling methods have led to only limited success in improving the affinity of antibodies for target antigens (6 -9).
Extensive analyses of antigen-antibody interactions have led to the conclusion that the high specificity and affinity of antibody molecules toward target antigens (10 -14) essentially originate from complementarity between their molecular shapes (10 -16). Thus, structural alterations arising from changes in the antigen and/or interfacial antibody residues reduce antigen-antibody affinity. Several studies of structural changes involved in the ability of influenza virus to evade the immune system have shown that minimal structural changes in the antigen are enough to prevent antibody binding (17)(18)(19)(20).
Recent advances in antibody engineering have made it possible to prepare tailor-made antibody fragments by in vitro selection (21)(22)(23)(24). In particular, enhancement of the affinity of humanized mouse antibodies and of human antibodies selected from naive phage display libraries without alteration of target sites offers scope for the development of functional and useful antibody molecules (25)(26)(27). Based on hundreds of studies of antibody structures, precise procedures for modeling these proteins have been developed. Therefore, a combination of structural information and in vitro selection, e.g. by phage display technology, should be a realistic and reasonable way to achieve epitope-directed improvements in antibody-antigen recognition.
Recently, we reported the selection and functional characterization of an anti-hen egg white lysozyme antibody (HyHEL-10) from a focus library of antibody variable domains (i.e. mini-libraries containing random mutations at four identical sites in the complementarity determining region (CDR) 1 2 of the heavy chain (CDR-H2) region). Phage display was used to * This work was supported in part by grants-in-aid for general research (to I. K.) from the Japan Society for the Promotion of Science, by grants-in-aid for younger scientists (to K. T.) and for priority areas (to I. K.) from the Ministry of Education, Science, Sports, and Culture of Japan, and by Industrial Technology Research Grant Program in 2000 (to K. T.) from the New Energy and Industrial Technology Development Organization of Japan. The experiments using synchrotron radiation were performed under the approval of the Photon Factory Advisory Committee, High Energy Accelerator Research Organization, Japan. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The enhance its specificity toward turkey lysozyme (a naturally mutated antigen) (28). Isothermal titration calorimetric studies have revealed the following (28). 1) Mutants selected have an affinity toward turkey lysozyme with an order of 10 8 and a reduced affinity toward the original antigen, hen lysozyme, with an order of 10 6 -7 . 2) Increase in negative enthalpy change (Ϫ⌬H) has driven the enhancement of an affinity for the target antigen, whereas decrease in Ϫ⌬H led to reduction of the affinity for hen lysozyme.
Structural information of the complexes would give several insights into the molecular mechanism for target site-directed functional alteration of antibodies. Here we report structural analyses of complexes between the naturally mutated antigen and an artificially mutated and selected antibody and between the normal antigen and the same antibody. On the basis of the results obtained, we discuss the critical effects on antibody specificity of the plasticity of the CDR loops (including the roles of some antibody residues that support the loop structures) and also of the plasticity of the interaction between the variable regions of immunoglobulin heavy and light chains.

EXPERIMENTAL PROCEDURES
Materials-Hen egg white lysozyme (HEL) was obtained from Seikagaku-Kogyo Inc. (Tokyo, Japan) and turkey egg white lysozyme (TEL) from Sigma. The expression vector for the HyHEL-10 variable region fragment (Fv) (wild-type and an SFSF mutant in which sites 53, 54, 56, and 58 are Ser, Phe, Ser, and Phe, respectively) was described previously (28). All other reagents were of biochemical research grade.
Expression and Purification of Soluble Fv Fragments-A transformed Escherichia coli strain, BL21 (DE3) (29), harboring an expression vector was grown at 28°C in 2ϫ YT (30) supplemented with 200 g ml Ϫ1 ampicillin, until the early stationary phase. To induce the expression of the soluble Fv fragment, isopropyl ␤-D-thiogalactopyranoside was added to a final concentration of 0.1 mM, and the culture was grown overnight at 28°C. The bacterial supernatant and periplasmic fractions were separated from 200 ml of the culture as follows (31). The culture sample was centrifuged at 6000 ϫ g for 15 min at 4°C; the culture supernatant was removed, and the cell pellet was resuspended in 10 ml of 20 mM Tris⅐HCl (pH 7.5), 0.5 M sucrose, and 0.1 mM EDTA and was incubated for 5 min at room temperature. Then 40 ml of water was added to give an osmotic shock, and the cells were left on ice for 30 min. The cells were collected by centrifugation at 7000 ϫ g for 60 min at 4°C, and the supernatant was saved as the periplasmic sample.
The supernatant and periplasmic samples were salted out with am-

TABLE II Amino acid residues participating in antigen-antibody interactions
Maximum interatomic distances between antibody and antigen amino acid residues considered to be "in contact" were as follows: C-C, 4.1 Å; C-N, monium sulfate at 80% saturation, and the precipitates were collected by centrifugation at 7000 ϫ g for 30 min at 4°C. The protein precipitates were dissolved in phosphate-buffered saline and were dialyzed against the same buffer for 2 days at 4°C. The precipitates that formed during dialysis were removed by centrifugation at 10,000 ϫ g for 15 min. The supernatant was loaded onto a TEL-Sepharose column, in which about 10 mg of TEL per milliliter of gel was bound to CNBractivated Sepharose 4B (Amersham Bioscience) previously equilibrated with phosphate-buffered saline. The column (inner diameter, 10 mm ϫ 5 cm) was washed with 100 mM Tris⅐HCl (pH 8.5) containing 500 mM NaCl; the adsorbed protein was then eluted with 100 mM glycine buffer (pH 2.0). The eluate was quickly neutralized with 1 M Tris⅐HCl (pH 7.5). The Fv fragment obtained by affinity chromatography was further purified on a Superdex 75pg column (inner diameter, 10 mm ϫ 100 cm), equilibrated with 50 mM Tris⅐HCl (pH 7.5) containing 200 mM NaCl, and finally dialyzed overnight against phosphate-buffered saline at 4°C.
Crystallization, Data Collection, and Structure Determination-Although the complex of the HyHEL-10 SFSF Fv fragment with HEL was crystallized as described previously (32), the crystal of the SFSF⅐HEL complex most suitable for further analyses could be grown from 100 mM HEPES buffer (pH 7.6 -7.8), 10% w/v polyethylene glycol 8000, 10% ethylene glycerol. The crystal of the SFSF⅐TEL complex suitable for further analyses was grown from 200 mM ammonium acetate, 100 mM tri-sodium citrate dihydrate buffer (pH 6.6), 20% w/v polyethylene glycol 4000, 10% ethylene glycerol. They were elongated, bipyramidshaped crystals.
All crystallographic data were collected at 100 K by using synchrotron radiation on beam line 6A of the Photon Factory (Tsukuba, Japan) with a Weissenberg camera (33). The diffraction images were integrated with the hkl program DENZO, and the intensity data were processed with SCALA and AGROVATA in the CCP4 suite (34). The crystallographic data and statistics are summarized in Table I.
The crystal of SFSF Fv complex with HEL was isomorphous with crystals of the wild-type (WT) Fv complex with HEL (32), whereas the crystal of SFSF Fv complex with TEL was grown in the different crystal form. The model coordinates of the SFSF complexes were derived from those of the wild-type complex structure (Protein Data Bank ID code 1c08).
The structure of the SFSF⅐HEL complex was refined with the pro- gram CNS (35). Although symmetry of diffraction intensities of SFSF⅐TEL showed Laue group of 4/mmm, the corresponding unit cell cannot contain one molecule in an asymmetric unit under the space group symmetry. Test for hemihedral twinning (36) has shown this crystal as a perfect twinning crystal, i.e. the twin fraction was estimated as 0.5. The structures of this crystal were determined by a molecular replacement method with the program AMoRe (37) in the CCP4 suite. Refinement was carried out by program SHELX97 (38), introducing TWIN and BASF options.
The atomic coordinates of the mutant Fv-lysozyme complexes were deposited in the Protein Data Bank (ID codes 1UA6 and 1UAC for HEL⅐SFSF and TEL⅐SFSF, respectively).

RESULTS
Overall Structure of Mutant Lysozyme-HyHEL-10 Complexes-Earlier, we reported the selection and functional characterization of an anti-hen egg white lysozyme antibody (Hy-HEL-10) from a focus library of antibody variable domains (i.e. mini-libraries containing random mutations at four identical sites in the CDR-H2 region). Phage display was used to enhance its specificity toward turkey egg white lysozyme (28). Several mutants were selected. Here we report structural analyses of one of them (SFSF), complexed with TEL (essentially a naturally mutated antigen) and with the wild-type antigen HEL. The contact residues in each complex are summarized in Table II.
Superposition of the C␣ backbones of the TEL⅐SFSF, HEL⅐SFSF, and HEL⅐WT complexes indicated that, despite four mutations in the CDR-H2 loop of SFSF relative to WT and four mutations in TEL relative to HEL, the overall structures of the TEL⅐SFSF, HEL⅐SFSF, and HEL⅐WT complexes were almost identical (Fig. 1), except for some residues in the flexible loops.
The crystal structure of HEL⅐WT indicates that the following noncovalent bonds occur in the interface between antigen and antibody: 16 hydrogen bonds, 2 salt bridges, and 98 van der Waals interactions (32). In TEL⅐SFSF, 14 hydrogen bonds, 2 salt bridges, and 97 van der Waals interactions were found; and in HEL⅐SFSF, 16 hydrogen bonds, 2 salt bridges, and 94 van der Waals interactions were found.
Differences in amino acid residues participating in the interactions were also observed (Table II). In the TEL⅐SFSF complex, Thr 30 VH, Trp 62 TEL, Lys 73 TEL, and Asn 77 TEL were found to be direct contact residues in the antigen-antibody interface, but Ser 93 VL, Tyr 96 VL, and Ser 52 VH were not. In the HEL⅐SFSF complex, Thr 30 VH, Trp 62 HEL, and Arg 73 HEL were included in the direct contact residues, but Ser 93 VL and Trp 94 VL were not.
Shape Complementarities-Shape complementarities of the molecular surfaces between antibody Fv fragment and two antigen lysozymes have been estimated using program SC (15) in the CCP4 program suit. Sc values for HEL⅐WT, HEL⅐SFSF, and TEL⅐SFSF were 0.702, 0.729, and 0.705, respectively, suggesting that the interface of these antigen-antibody complexes have almost identical shape complementarity to each other. On the other hand, shape complementarities of the molecular surfaces between VH and VL are calculated to be 0.700, 0.733, and 0.695, for HEL⅐WT, HEL⅐SFSF, and TEL⅐SFSF, respectively.
Interfacial Water Molecules-In the TEL⅐SFSF complex, five water molecules bridged the imperfect complementarity of the antigen-antibody chains, and in HEL⅐SFSF, seven interfacial water molecules were observed. The hydrogen bond networks (via interfacial water molecules) were almost identical in the TEL⅐SFSF and HEL⅐SFSF complexes, except for the two additional interfacial water molecules in HEL⅐SFSF.
Local Structural Changes in Mutated CDR-H2-Superposition of the three complexes indicated that the orientation of CDR-H2 changed in the mutants (Fig. 2). A major change occurred around Phe 54 VH. The epitope recognized by residue 54 has a flexible structure, suggesting induced conformational changes depending on whether WT or SFSF antibody binds to it.
As shown in Fig. 3a, Phe 54 VH of SFSF interacts with Trp 63 HEL, Leu 75 HEL, and Asp 101 HEL, leading to clustering of the residues in the HEL⅐SFSF complex. In contrast, Phe 54 VH of SFSF interacts only with Trp 62 TEL in the TEL⅐SFSF complex. The different orientations of Phe 54 VH in TEL⅐SFSF and in HEL⅐SFSF may lead to other changes in orientation of the CDR-H2 loop in the complex structure.
Other local structural changes were observed around Phe 58 VH. In the HEL⅐SFSF complex, the amide groups of Arg 21 HEL (gray) make hydrogen bonds with the hydroxyl group of Tyr 50 VH (red), and the aromatic ring of Tyr 58 VH (red) interacts with Arg 21 HEL (gray) via an amino-aromatic interaction (Fig. 3, b and c). In the TEL⅐SFSF complex, one of the amide groups in Arg 21 TEL (green) makes a hydrogen bond with the hydroxyl group of Tyr 50 VH (violet) and also interacts with the aromatic ring via an amino-aromatic interaction (Fig. 3, b  and c). In addition, contacts between Tyr 58 VH and Phe 50 VH in TEL⅐SFSF are increased compared with the HEL⅐WT and HEL⅐SFSF complexes. Isothermal titration calorimetric studies have suggested an enthalpic contribution of Phe 58 VH in lysozyme-mutant antibody interactions (28). Deletion of the hydroxyl group of Tyr 58 VH from HyHEL-10 WT may lead to reorientation of the aromatic ring of VH at site 58 toward the loop consisting of residues 99 -102, which might enhance the affinity of the antibody for lysozyme.
Local Structural Changes in Other CDR Loops-Even though there was no mutation in the other CDR loops of SFSF, crystallographic studies have revealed several local differences between the TEL⅐SFSF and HEL⅐SFSF complexes.
CDR-L3-In TEL⅐SFSF, the amide side chain of Asn 92 VL made hydrogen bonds with the main chain carbonyl of Asn 19 TEL (3.16 Å) and the side chain of Asn 32 VL. In HEL⅐SFSF, the length of the hydrogen bond between Asn 92 VL and Asn 19 HEL is 3.27 Å, and one interfacial water molecule mediates the contacts with the main chain carbonyls of Asp 18 HEL and Gly 16 HEL via hydrogen bonds.
CDR-H1-In TEL⅐SFSF and HEL⅐SFSF, hydrogen bond formations between Ser 31 VHO␥ and the main chain carbonyl of Arg 73 HEL (or Lys 73 TEL) and between the hydroxyl group of Tyr 33 VH and the main chain carbonyl of Lys 97 HEL (or TEL) were observed. In addition, the side chain carboxyl group of Asp 32 VH made a hydrogen bond with the side chain of Asn 77 TEL in TEL⅐SFSF. Asn 77 TEL participated in a hydrogen bond network via an interfacial water molecule located in the interface between TEL⅐SFSF and CDR-H3 (Fig. 4), which was not observed in the HEL⅐SFSF and HEL⅐WT complexes.
CDR-H3-No major conformational changes occurred between the three complexes. The contact area of TEL⅐VL in the TEL⅐SFSF complex, however, increased. Additionally, a hydrogen bond network in TEL⅐SFSF via an interfacial water molecule, distinct from that in HEL⅐SFSF, was observed (Fig. 4), suggesting that the interfacial water molecules control the complementarity of the antigen-binding site of the antibody, as was suggested for anti-steroid hormone antibody 4155 (40) and other antigen-antibody and protein-ligand interactions (41)(42)(43).
CDR-CDR and VH-VL Interactions-Although the contact area of the SFSF antibody participating in binding the TEL (ϳ819 Å 2 ) was almost the same as in the HEL⅐SFSF complex (ϳ817 Å 2 ), the relative ratio of each CDR loop to the total areas of the antibody molecule covered by the antigen changed (Fig.  5). The ratio of CDR-H2 is reduced due to significant decreases in contact area at residue 54VH in SFSF (Fig. 3). However, some changes in the local structures of CDR-L2, -L3, and -H1 via hydrogen bond formation and removal of interfacial water molecules lead to different ratios of areas covered by each CDR loop in the WT⅐HEL complex.
The contact area between VL and VH in TEL⅐SFSF increased by 50 and 148 Å 2 relative to the HEL⅐SFSF (1528 Å 2 ) and HEL⅐WT (1430 Å 2 ) complexes, respectively, mainly because of an increase in the contact of CDR-H3 with the VL chain (Fig.  6). These results suggest that an increase in the VL-VH contact area may drive the target site-oriented functional alteration of the antibody.
In the HEL⅐SFSF complex, the side chain amide group of Arg 71 VH makes hydrogen bonds with the main chain carbonyl of Ser 53 VH and also with the main chain carbonyls of Ile 29 VH and Asp 32 VH via one water molecule (W78, Fig. 7A). Similarly, in the HEL⅐WT complex, the NH 2 of Arg 71 VH makes hydrogen bonds with the main chain carbonyl of Asp 32 VH via one water molecule. In addition, in both these complexes, the side chain O␦2 makes hydrogen bonds with the main chain of Asn 94 VH, whose side chain supports the loop structure of CDR-H3 via hydrogen bonding with the main chain amides of Gly 100 VH, Asp 101 VH, and Tyr 102 VH (Fig. 7B). These observations clearly indicate that the loop structures of the VH CDRs interact with each other via hydrogen bond networks.
On the other hand, although a hydrogen bond network of Asn 94 -Gly 100 -Asp 101 -Tyr 102 was observed in the TEL⅐SFSF complex (Fig. 7B), the main chain carbonyl of Asp 32 VH makes a hydrogen bond with the side chain of Trp 34 VH, and no hydrogen bonds of Arg 71 VH with the CDR loops via water molecules exist (Fig. 7A). In addition, the side chain of Asp 32 VH makes a hydrogen bond with the side chain of Asn 77 TEL (Fig.  4), leading to breakage of the hydrogen bond with the side chain of Asn 97 VH. These results indicate that interaction of CDR-H1 with CDR-H3 in TEL⅐SFSF is weakened by mutations in CDR-H2, which might lead to stronger interaction of CDR-H3 with the VL chain in TEL⅐SFSF than in HEL⅐SFSF or HEL⅐WT. The hydrogen bond networks occurring via interfacial water molecules differ between TEL⅐SFSF and the HEL complexes and may be due to the structural changes in CDR-H1 and CDR-H3. DISCUSSION Here we report the crystal structure of two antigen-antibody complexes. One is a complex between a mutant antibody frag-ment and its target antigen (the mutant antibody was selected from focus libraries containing random mutations at four sites via phage display using TEL as a natural mutant for the HEL antigen). The other is a complex between the same mutant antibody fragment and the original (HEL) antigen. To the best of our knowledge, this is the first report of the structures of complexes of engineered antibody fragments with their target (mutated) and wild-type antigens.
Introduction of amino acids rarely found in immune proteins into the antibody loops often causes structural disruption of other parts of the antibody molecule. The results reported here emphasize the critical role of the CDR loops, via subtle but critical structural rearrangements of the antibody loops because of mutations in several sites of a different loop, and also the importance of the plasticity of the VL-VH interaction in antibody specificity. We should emphasize that only four mutations in CDR-H2 result in subtle but significant structural changes in other CDR loops.
The significant contribution of VL-VH interactions to the specificity and affinity of antibody molecules was first pointed out by Colman (interface adaptor hypothesis) (44) and was experimentally demonstrated by Stanfield et al. (45) using uncomplexed and ligand-complexed forms of anti-human immunodeficiency virus type 1 peptide antibody (Fab 50.1). The antibody fragment has a smaller VL-VH contact area, perhaps because of the shorter length of CDR-H3, which may lead to changes in VL-VH contact upon complexation, i.e. induced fitting upon complexation. Light chain shuffling of an antibody molecule resulted in drifts in epitope recognition, supporting the importance of cooperativity between variable regions to the high affinity of antibodies (46). Recently, limited antibody pairings of domains have been proposed (47), and utilization of VL-VH interactions for immunoassay has been reported (48). Interestingly, the VL-VH contact area of anti-phosphocholine antibody (McPC603) is ϳ1700 Å 2 , despite the smaller area of contact between the antibody and the antigen (ϳ290 Å 2 ), than the case of protein antigen-antibody interaction (49).
Therefore, one may conclude that an increase in affinity of our mutant antibody for a mutant target antigen is due not so much to improvements in complementarity around the mutated CDR-H2 loop as to subtle structural rearrangements of antigen, VL, and VH, which might lead to an increase in covered areas of CDR-L2, -L3, and -H1 upon complexation. We can conclude that these structural rearrangements increase the covered areas of VL-VH interfaces, resulting in an enhancement of the affinity of the mutant antibody fragment for the mutated antigen. The significant increase in interfacial areas of VL-VH interfaces in CDR-H3 should be emphasized, because antibodies that are highly diverse in CDR-H3 are created via genetic recombination in vivo (50,51).
Complementarity of VH-VL in the SFSF⅐TEL complex is less than in SFSF⅐HEL. However, the covered area of VH-VL interface of SFSF⅐TEL upon complexation is larger than SFSF⅐HEL, and SFSF has more than 5-fold greater affinity for TEL than for HEL (28). This suggests that an increase in complementary association of VH with VL does not always lead to the enhancement of the affinity of antibodies. Cooperativity between variable regions upon complexation might not lead to improvement of interfacial complementarities, despite grafting high affinity of antibodies for the targets.
The structural analyses of three antigen-antibody complexes also highlight the critical role of residue 71 of VH in the tolerance of structural changes in CDR-H2 during induced fitting. A critical role of Arg 71 VH in the packing of CDR-H1 and -H2 and in the orientation of CDR-H2 has been pointed out (52), and its role has been emphasized in reports of the structural consequences of humanizing mouse antibody molecules (53,54). A series of structural changes in CDR loops may originate from distinct packing of the CDR structures in response to the orientation of the side chain of Arg 71 VH.
Studies of hundreds of structures of antibody-antigen complexes have indicated that amino acid residues in CDRs participate in direct recognition of target antigens (10 -14), suggesting that improvements in local complementarity may enhance the affinity of an antibody for its natural or mutated antigen. The results reported here, however, highlight the critical contribution to the function of antibody molecules of subtle but significant conformational changes in the antibody molecule due to mutations that affect, for example, the orientation and cooperative locations of CDR loops and VL-VH interactions. It may be significant that replacements of amino acids during affinity maturation in the immune system often occur in the framework region rather than in the CDR loops (55)(56)(57). For improvement of the affinity of an antibody for its target antigen, not only directly contacting regions but also other regions changed by mutations, including framework regions, will be especially important. For example, some CDRs tolerate the inclusion of naturally rare amino acids via structural rearrangement of the loop structures, again emphasizing that a focus library containing mutation in specificity-determining regions (58) or hotspot residues (59) can be efficiently utilized for target site-oriented functional alteration of antibody molecules.
Finally, we should point out the relative insensitivity of the antibody structure to the introduction of amino acids rarely used in the immune system into an antigen. The target sitedirected functional alteration of antibody molecules via focus libraries containing random mutations at several sites should be promising for researchers in various fields. Our results should encourage researchers, especially in the fields of evolutionary engineering, antibody engineering, and rational protein design. Additional accumulation of structural data on interactions between engineered proteins, including antibody molecules, is especially important.