PROTEINS: Structure, Function, and Genetics 34:383–394 (1999) Structural and Chemical Complementarity Between Antibodies and the Crystal Surfaces They Recognize N. Kessler,1 D. Perl-Treves,1 L. Addadi,1 and M. Eisenstein2* 1Department of Structural Biology, Weizmann Institute of Science, Rehovot, Israel 2Department of Chemical Services, Weizmann Institute of Science, Rehovot, Israel ABSTRACT The sequences of the variable regions of three monoclonal antibodies with different specificities to cholesterol monohydrate and 1,4dinitrobenzene crystals were determined. The structures of their binding sites were then modeled, based on homology to other antibodies of known structure. Two of these antibodies were previously shown to specifically recognize each one welldefined face of one of the crystals, out of a number of crystal faces of closely related structure. The binding site of the antibody which recognizes the stepped (301) face of the cholesterol crystal is predicted to assume the shape of a step with one hydrophobic and one hydrophilic side, complementary to the corresponding crystal surface. Within the step, the hydroxyl groups of five tyrosines are located such that they can interact with the hydroxyl and water molecules on the cholesterol crystal face, while hydrophobic contacts are made between the cholesterol backbone and hydrophobic amino acid sidechains. In contrast, the modeled binding site of the antibody which recognizes the flat (101) face of 1,4-dinitrobenzene crystals is remarkably flat. It is lined by aromatic and polar residues, that can make favorable contacts with the aromatic ring and nitro groups of the dinitrobenzene molecules, respectively. Proteins 1999;34:383–394. r 1999 Wiley-Liss, Inc. Key words: cholesterol; comparative modeling; molecular crystals; molecular recognition; monoclonal antibody INTRODUCTION The knowledge available on the molecular basis of antibody specificity has greatly increased over the past decade with the elucidation of the three-dimensional structures of a large number of antibody-antigen complexes, especially with protein antigens.1 The complexes are stabilized by Van der Waals forces, hydrogen bonds, and occasional ion-pairs over large (600 to 900 Å2) geometrically and electrostatically complementary surfaces.2,3 Specific antibodies can in principle recognize almost any exposed surface area on the protein antigen, although some regions, characterized by higher thermal motion4,5 or higher accessibility,6 appear to be preferred over others. The immense variety of known antigens has been recently enlarged to include the surfaces of molecular crysr 1999 WILEY-LISS, INC. tals.7,8 These are characterized by well-defined structures and highly ordered, repetitive motifs. The molecular component of the crystal is exposed at the different faces of the crystal in different orientations, exhibiting different chemical moieties in a known pattern. An antibody interacting with a crystal surface would recognize an array of molecular moieties in a given arrangement, typically 5–20 depending on the size of the molecule. On each crystal face the potential recognition sites would thus have well-defined chemical characteristics which differ from each other and from that of the individual component molecule. Specific interactions of proteins with defined crystal surfaces have been observed in different systems where crystals are exposed to biological environments, such as in biomineralization9–11 or in common crystal-associated pathological conditions.12–15 The structure of proteins binding to the rigid surface of crystals was solved for two types of anti-freeze proteins, which adsorb to specific planes of ice crystals. The long linear ␣-helical type I antifreeze protein exhibits four repeated ice-binding motifs, the side-chains of which are inherently rigid or restrained by pair-wise interactions to form a flat binding surface.16 In type III anti-freeze proteins, the structure reveals a remarkably flat amphipathic binding site, with good matching of multiple hydrogen bonds. In both cases, the flatness of the ice-binding site is a crucial feature in the ice-binding mechanism.17 We have shown that the exposure of an organism to crystals, such as may occur in various pathological situations, can trigger the amplification of specific antibodies. The presence of antibodies which recognize crystal surfaces was demonstrated in polyclonal antibody populations using a nucleation assay.18 In humans, the presence of monosodium urate monohydrate crystals in the synovial fluid was shown to trigger the production of immunoglobulins that catalyzed the nucleation of the same crystals in Abbreviations: BSA, Bovine Serum Albumin; CDR, Complementarity Determining Region; DNB, Dinitrobenzene; ELISA, EnzymeLinked-Immuno-Sorbent-Assay; Fab, Fragment Antigen Binding; IgG, Immunoglobulin G; IgM, Immunoglobulin M; MAb, Monoclonal antibody; PBS, Phosphate Buffer Saline; PCR, Polymerase Chain Reaction; PDB, Protein Data Bank. Grant sponsor: Minerva foundation; Grant sponsor: Kimmelman Center for Biomolecular Assembly. *Correspondence to: Dr. Miriam Eisenstein, Chemical Services, Weizmann Institute of Science, 76100 Rehovot, Israel. E-mail: email@example.com Received 12 June 1998; Accepted 16 October 1998 384 N. KESSLER ET AL. TABLE I. Specificity of the Selected Antibodies MAb 122B1 36A1 23C1 Recognized crystal Crystal face 1,4-DNB cholesterol cross-reactive (101) (301) — vitro.19 Similarly, the exposure of rabbits to three structurally-related crystals (monosodium urate monohydrate, magnesium urate octahydrate and allopurinol) led to the development of specific immunoglobulin populations, each able to catalyze the nucleation of the crystal to which the animal was exposed.20 We subsequently proposed that antibodies against specific crystal faces contain within their binding site a structured imprint of the crystal. These antibodies then act as stabilizing templates in a new crystallization event. Using the same type of rationale, polyclonal ice nucleating IgG’s were detected in the sera of cold-ocean marine fish, but not in the sera of species which are not exposed to ice.21 Recently, monoclonal antibodies which recognize crystals of cholesterol monohydrate and of 1,4-dinitrobenzene were selected, using a conventional fusion protocol coupled with an adsorption assay, after exposure of mice to either one of the crystals.7,8 The antibodies generated from exposure to both crystal types presented a wide range of recognition levels, from crystal-specific to cross-reactive. At the highest level of specificity, one of the monoclonal antibodies generated against cholesterol monohydrate crystals and one of the monoclonal antibodies generated against 1,4-dinitrobenzene crystals were shown to preferentially interact each with one crystal face of well defined character.22 Here we report on the sequences and predicted threedimensional structures of three selected monoclonal antibodies (MAbs), raised against crystals of cholesterol monohydrate and 1,4-dinitrobenzene. The models of the two MAbs, each of which recognizes one specific face of cholesterol monohydrate crystals and of 1,4-dinitrobenzene crystals respectively, exhibit a clear-cut geometrical match between the interacting surfaces. It is suggested that binding is stabilized by multiple chemical interactions between the residues exposed on the antibody-binding site and the molecular moieties exposed on the crystal surface. RESULTS Monoclonal antibodies (MAbs) were produced after repeated injections of crystals of 1,4-dinitrobenzene (1,4DNB) and cholesterol monohydrate in mice. The binding of each monoclonal antibody from both mice hybridoma populations was assayed in parallel on crystals of 1,4-DNB and cholesterol monohydrate.7,8 Three of these MAbs (Table I) were selected here for further characterization: one interacts with 1,4-DNB crystals (but not cholesterol monohydrate crystals), one interacts only with cholesterol crystals, and one is cross-reactive (reactive to all crystal surfaces tested). All the selected MAbs belong to the IgM type and use the kappa class of light chains. Sequences of the Selected Antibodies The sequences of the selected antibodies were determined by cloning the cDNA’s of the heavy and light chain variable domains (VH and VK) by the polymerase chain reaction (PCR) and further sequencing those fragments. The genetic elements comprising the VL and VH region were determined, based on sequence homology to known germline sequences of mouse antibodies (see Materials and Methods, Table II). On the basis of the existent library, none of them is germline. As not all the mouse Ig germlines were sequenced and included in the library, the degree of somatic mutation cannot however be established with certainty. The sequences of the Complementarity Determining Regions (CDRs) of the three selected MAbs are reported in Figure 1. Modeling of the Antibody Structure Models for the variable domains of MAbs 36A1, 23C1, and 122B1 were generated by the comparative modeling procedure described in detail below. The modeling procedure was exclusively based on the 3D structure of antibodies with high sequence identity, determined by X-ray diffraction, and did not include any type of information on the antigen structure. A detailed description of the predicted structures and of their putative complex with the appropriate crystal faces (when applicable), follows. MAb 36A1: Proposed structure and interactions with cholesterol crystals Three candidate templates for the Fv light chain of 36A1 were identified in the PDB23,24, entries 1bbd,25 1hil,26 and 1mcp.27 Superposition of the C␣ atoms of their light chains, including CDR regions, showed that they are very similar, i.e. the RMSD (root mean square deviations) did not exceed 0.65 Å. For the heavy chain we found one candidate, entry 1vfa.28 Superposition of the light chain of 1vfa, except L1, onto the group of the selected light chain templates (RMSD ⬍ 0.7 Å) showed that 1mcp and 1vfa have the closest relative disposition of light and heavy chains. The RMSD for C␣ atoms in the framework of both light and heavy chains of 1mcp and 1vfa was 1.6 Å. The superposition is particularly good for the inner ␤-sheets of both chains (RMSD 0.9 Å). Thus, 1mcp and 1vfa were selected as modeling templates. Five of the CDR’s of 36A1 (L1, L2, L3, H1, H2) were modeled together with the framework, using the same templates. H3 is one residue longer in 36A1, relative to 1vfa. Next, several structures in the PDB were compared, with an appropriate stem for H3 and an apex loop of the same length as in 36A1: 1jhl,29 7fab,30 2fbj,31 1jel,32 and 1mfb.33 The conformations of the loops were found to be different in each structure, although, being very short, they occupy the same region in space. The sequences of the H3 apexes also differ from H3 in 36A1. We therefore generated several plausible structures for this loop by Homology and checked them manually. We chose the loop 385 ANTIBODIES RECOGNIZING CRYSTAL SURFACES TABLE II. Genetic Elements of the Variable Regions of the Selected Antibodies and Their Homology to Germline VK gene family 36A1 (cholesterol) 23C1 (cross reactive) 122B1 (1,4-DNB) Homology to germline VK8,a [GLVK50]c 92% VK8, [GLVK50] 96.3% (V) 67% Number of CDR mutations JK 11 J4 4 J4 27 J2 VH gene family VH2, (IB),b [Vm13] VH2, (IB), [Vm13] VH1, (IIB), [clone 28] Homology to germline Number of CDR mutations JH D 100% 0 J4 FL16.2 96% 1 J4 SP2.2 98% 0 J2 SP2.5 VH , VK ⫽ Variable region of heavy (H) and light (K) chain gene. aV and V -gene family as defined by Kofler et al.57 K H bV and V protein subgroup as defined by Kofler et al.57 is given in round brackets. K H cPrototype member for each family is given in square brackets. The data were taken from the genebank library (University of Wisconsin Genetic Computer Group soft-ware package). JK , JH ⫽ Junction region of light (K) and heavy (H) chain gene, according to Max58 (K chain) and Sakano59 (H chain). D ⫽ Diversity region of heavy chain gene according to Kurosawa.60 Fig. 1. Amino acid sequence of the CDRs of antibodies 36A1, 23C1, and 122B1. The symbol (-) denotes sequence identity to MAb 36A1. Amino acid numbering is according to Kabat et al.55 and CDR position is according to Chotia et al.48,56 that did not clash with the rest of the structure and occupied the same spatial region as H3 in the structures listed above. The combining site of 36A1 assumes a peculiar shape of a step with an angle close to 90°, which is roughly 12 Å high and 24 Å wide (Fig 2A, insert). The long L1 loop forms one ‘‘wall’’ of the step, while the short H2 and H3 loops form the ‘‘floor’’ of the step. The step wall is highly hydrophilic, exposing the hydroxyls of 5 tyrosines and 3 serines, the amide group of 1 asparagine and the charged end of 1 lysine. The step floor, in contrast, is composed of three aromatic residues with their rings parallel to the step (2 Tyr, 1 Trp), 2 hydrophobic (Ala, Val), and 1 polar (Ser) residues (Fig 2A). It is important, at this point, to discuss the reliability of our model and of its most pronounced feature, the stepshaped combining site. The high percentage of sequence identity between the light chain of 36A1 and the candidate template structures, together with the pronounced structural similarity among the light chains of these templates, suggest that the structure of the modeled light chain is highly reliable. The model of the heavy chain is also based on a template with high sequence identity and the structure is reliable except for H3. However, this is a very short loop in 36A1, with a limited range of conformations all occupying a preferred region in space (based on our comparison of several structures with loops of the same length). Hence, variations in the structure of this loop are not expected to affect the overall shape of the step. Another important consideration is the relative positioning of the light and heavy chains, which were modeled using different templates. The RMSD between the framework C␣ atoms in the ␤-sheets at the light-heavy interface of 1mcp and 1vfa is only 0.9 Å. An error of this magnitude might slightly affect the shape of the combining site, retaining however the overall shape of the step. In summary, the model structure for 36A1, including the features of the combining site, appears to be highly reliable. The binding of 36A1 to the different cholesterol crystal faces has been characterized by immunolabeling techniques.8 Specific recognition was detected for the (301) face. By far lower binding was observed on the other commonly developed side faces of the crystal, (100), (101), (201), (010), and (011), and practically no binding on the large plate faces, (001). The recognized (301) face is characterized by the presence of molecular steps, exposing the 3␤ hydroxyl groups of the cholesterol molecule and lattice water molecules on the walls of the steps, which separate between long ‘‘floors’’ where only the hydrophobic cholesterol backbones are exposed (Fig. 3). This geometry is highly suggestive of a possible interaction with the hydrophilic and hydrophobic regions of the antibodybinding site, respectively. Manual docking of MAb 36A1 on the step of the (301) face shows an excellent geometrical and chemical match between the two surfaces (Fig 2B). During the docking procedure, only the 1 torsion angle of a single tyrosine residue was changed, in the allowed range, to better fit the 386 N. KESSLER ET AL. Fig. 2. (A) Model structure of MAb 36A1 (cyan), with the antigen binding site residues highlighted: red:Tyr-L31, Tyr-L32, Tyr-L92, Tyr-H97, and Tyr-H99 ; yellow:Tyr-L91, Tyr-L94, Trp-H52, Ala-H53, Val-H95 and Ser-H56. A stick representation of the antibody-binding site is shown. The balls mark the hydroxyl oxygens of the tyrosines on the vertical hydrophilic ‘‘wall’’ of the step, while the hydrophobic residues on the ‘‘floor’’ of the step (yellow) are viewed edge-on. The insert highlights the step-like solvent accessible surface of the antibody variable region. The position of the binding site is indicated by an arrow. Here and hereafter, in this view, the light chain is shown on the left and the heavy chain on the right. (B) Docking model of MAb 36A1 (in line representation) onto the (301) face of the cholesterol crystal (in space-filling representation).Yellow: cholesterol backbone, antibody side chains of Tyr-L91, Tyr-L94, Trp-H52, Ala-H53, Val-H95, and Ser-H56; cyan: antibody backbone; red: hydroxyl oxygen atoms of the cholesterol molecule, antibody tyrosines L31, L32, L92, H97 and H99; blue: water molecules. The stepped (301) face with the docked antibody Fv region is presented in a nearly edge-on view. ANTIBODIES RECOGNIZING CRYSTAL SURFACES 387 Fig. 3. Cholesterol crystal packing arrangement on the (101), (010), (100), (001), and (301) faces. The (011) and (021) faces are similar to (010), except that only the cholesterol backbone is exposed as in (100). The relevant surface is represented at the top of each figure, as marked (white dashed lines). Yellow: cholesterol backbone; red: hydroxyl oxygen atoms; blue: water molecules. hydroxyl surface of the cholesterol step. No backbone conformational modifications were allowed, however. Automatic docking was attempted using a modified version (unpublished) of the program Molfit.34,35 Several structures for the crystal-antibody complex were obtained. In all of them the step-shaped combining site matched the step on the crystal surface, but several equivalent positions along the step were obtained, translated at intervals corresponding to the displacement of the cholesterol molecules along the b axis of the crystal. The step in the antibody-binding site is 12 Å high, as is the dimension of one unit cell of cholesterol monohydrate along the a axis, comprising two molecules. The width of the antibody-binding site is 24 Å, which fits four molecules of cholesterol along the b axis, whereas the length of the bottom of the step in the antibody-binding site fits the size of one molecule of cholesterol along the c axis (⬃ 17 Å). In addition to the geometrical match (Fig 2B), the model indicates the possibility of good chemical interactions, with hydrogen bonds being formed between the hydroxyl and water groups on the cholesterol (301) face and the polar residues constituting the antibody-binding site. Notably, the five tyrosines (L31, L32, L92, H97, and H99) on the hydrophilic side of the antibody-binding site (Fig 2A) are spaced such that they can be juxtaposed each to one of five cholesterol hydroxyl groups (not shown). The hydrophobic cholesterol backbones are suggested to interact with the hydrophobic residues (Ala-H53, Val-H95, Tyr-L91, Tyr-L94, Trp-H52) lining the floor of the step in the antibody-binding site. Five structures, representative of the crystal faces commonly developed in cholesterol crystals, are shown in Figure 3. The lower binding of MAb 36A1 to some of these is easily understandable, based on the geometric and electrostatic mismatch between the surfaces. The (100), (011), and (021) faces all expose the hydrophobic cholesterol backbone exclusively. On the (001) face, only the 3␤ hydroxyls and the lattice water are exposed. However, low binding was also observed for the (101) and (010) faces, which exhibit a stepped structure identical or similar to that of the (301) face respectively, but with (on average) one unit cell interval between the steps rather than three. We suspect that the lower binding is related to the shorter distance between the steps on these faces, where the bulky IgM molecule may not be able to bind cooperatively. MAb 23C1: Proposed structure MAb 23C1 is a cross reactive antibody, that binds mainly to imperfections on the crystal faces.8 The sequence identity between 23C1 and 36A1 is very high, 91% for the light chains and 85% for the heavy chains, which differ mostly in the H3 region. Therefore the same templates were used to build the model for 23C1 as for 36A1. We found 4 structures in the PDB with H3 of the same length as in 23C1: 8fab,30 1igm,36 1iai,37 and 6fab.38 Superposition of 388 N. KESSLER ET AL. the heavy chains of these structures revealed two conformations for their H3. Thus, the H3 loops in 6fab and 1iai were similar and those in 1igm and 8fab were similar to one another but different from the first pair. We built two models, with different H3’s, according to 6fab and 1igm. Energy minimization strongly preferred one model, in which H3 was based on 6fab, over the other. Although 23C1 has a high sequence identity and CDR loop similarity (L1, L2, L3, H1 and H2) with 36A1, the structure of their combining sites is markedly different. Due to the longer H3 in 23C1, the step that was observed in the combining site of 36A1 is flatter (not shown). Moreover, MAb 23C1 has 3 charged residues (Arg-H3, Asp-H58, and AspL91) located in a central position, while MAb 36A1 has none. The model structure of 23C1 is less reliable than the model for 36A1, due to the longer H3 loop and its two apparent conformations. Yet, our conclusion regarding the characteristics of the site are sound because the exposed charged residues in the combining site come from the L1 and H2 loops. MAb 122B1: Proposed structure and interactions with 1,4-DNB crystals Three candidate templates were found for the light chain of 122B1: 1ivl,39 1mlb,40and 3hfm.41 The RMSD between all C␣ atoms in these chains were less than 0.7 Å. The candidate templates for the heavy chains were 1fpt,42 1for,43 and 1jhl.29 Superposition of their C␣ atoms, excluding H3, gave RMSD less than 0.9 Å. According to its sequence, the H3 loop of 122B1 is expected to form a non-bulged stem structure.44 The fragment -CAN-, which indicates a non-bulged stem, occurs in 1mrf,45 where indeed the non-bulged H3 stem conformation is found. In 1mrf the buried D101 of the heavy chain forms hydrogen bonds to Trp 103(H) and Tyr 36(L). Notably, in 122B1 these residues are conserved suggesting that similar hydrogen bonds are formed. The relative position of the light and heavy chains is likely to be affected by the conformation of the stem of H3, which is found at the light-heavy interface. The structure of 1mrf was thus selected as a template for the relative positioning of the light and heavy chains. The light chain of 1mlb was superimposed on the light chain of 1mrf (97 atoms excluding L1; RMSD ⫽ 0.63 Å) and the heavy chain of 1jhl was superimposed on the heavy chain of 1mrf (101 atoms excluding H3 and missing atoms in the framework of 1mrf, RMSD ⫽ 0.78 Å). Thus, 1mlb was used as a template for constructing the light chain of 122B1, 1jhl was used as a template for the heavy chain, 1mrf provided a template for the stem of the H3 loop of 122B1, whereas the apex of this loop (4 residues) was generated by Homology and a conformer acceptable with regard to intramolecular clashes was selected. The combining site of 122B1 is flat, with exposed aromatic (Phe, Tyr, Trp, and His), polar (Asn, Gln, and Ser) and hydrophobic (Leu) residues (Fig. 4A). Notably, the overall percentage of the aromatic residues in the CDRs of 122B1 is not exceptionally high, but they appear to be all exposed to the solvent. The model structure for the light chain of this antibody is deemed reliable due to the high sequence identity with the template. Similarly reliable is the model for the heavy chain, H3 excluded. The relative position of the light and heavy chains is determined by the conformation of the stem of H3 and by the interactions of the buried D101(H); 1mrf provided a good template for these interactions and for the relative L/H positioning. The model for the nonbulged stem of H3 is also reliable, whereas the apex of H3 is less reliable, as in all antibody models. MAb 122B1 preferentially recognizes the (101 ) face of the 1,4-DNB crystals.22 Substantially lower reactivity was observed for the (101), (110), and (100) faces. The (101) face is planar and is characterized by aromatic rings organized in a herring-bone structure, with the dinitro-benzene molecule emerging side-on, almost perpendicular to the face. One oxygen atom of each nitro group and the benzene ring profiles are thus exposed at the surface (Fig. 5). The flat surface of MAb 122B1 well matches the (101 ) face of the 1,4-DNB crystal, both geometrically and chemically (Fig. 4). The antibody-binding site is characterized by the presence of aromatic residues, which may interact with the aromatic rings exposed at the (101 ) crystal face, and of polar (Asn, Gln, and Ser) residues which can form hydrogen bonds with the nitro groups. Manual docking of MAb 122B1 on the (101 ) face of the 1,4-DNB crystal shows a striking correspondence between 5 of the antibody tyrosines (50 of the light chain and 32, 97, 98, 100 of the heavy chain) and one tryptophan (94 of the light chain) and the structure of the dinitrobenzene molecules in the crystal (Fig 4B). These side chains appear to be positioned such that they match the crystal lattice, forming a continuation of the herring-bone motif of the crystal into the antibody. The lower binding of MAb 122B1 to the other faces of the 1,4-DNB crystal is easily rationalized by their different structures (Fig. 5). The (101) and (110) faces are rough faces, and expose mainly the nitro groups in a ridge-andchannel motif. The (100) face is not as rough, exposing mainly the nitro moieties and less the aromatic rings. In contrast to the (101 ) face, however, the aromatic rings are oriented towards the (100) face such that their ring plane is exposed. Hence, their packing motif cannot be matched by the antibody aromatic moieties, as for the (101 ) face. The model thus suggests a plausible explanation for the observed binding preference of 122B1. We note, however, that the conformation of the aromatic side chains cannot be accurately modeled, in particular those of Tyr-97 and Tyr-98 in the apex of the H3 loop. DISCUSSION In this study we determined the amino acid sequences and modeled the 3D structures of antibodies which recognize specific crystal surfaces of cholesterol monohydrate and 1,4-dinitrobenzene. In two cases the models were subsequently matched to the recognized crystal surfaces. The binding preference may be explained first in terms of the geometrical fit between the two interacting surfaces. In ANTIBODIES RECOGNIZING CRYSTAL SURFACES addition, a striking correspondence was found between the antibody residues exposed at the binding site and the molecular moieties exposed at the appropriate crystal surface. The antibodies considered here were selected based on their affinity for crystal surfaces. None of these has any appreciable binding to the individual molecule comprising the crystal.7,8 This may be explained when considering that the exposed area of a single molecule on any crystal surface (approximately 50 Å2), is one order of magnitude smaller than the interface area of an antibody-binding site (600–900 Å2). Moreover, an analysis of binding site topographies showed that the combining sites of antibodies that bind small haptens are usually concave,46 in sharp contrast with the step-like and planar binding sites of the crystal-binding antibodies considered here. The interaction of one single molecule with the flat surface of the antibody-binding sites is thus too weak, and binding occurs rather to a number of specifically ordered molecules, held together by the crystal lattice forces. One of the antibodies examined here, 36A1, selectively interact with cholesterol monohydrate. One, 122B1, with 1,4-DNB crystals. Both show specific interaction with one crystal face, out of several expressed faces of either cholesterol monohydrate or 1,4-DNB. The available information on the binding patterns of these antibodies offers thus the rare advantage of a systematic approach to binding of antibodies to a series of organized surfaces comprising the same component molecule. Although the accuracy in the modeling of protein structures is still limited in general, the reliability of modeling of antibody structures is increased by three essential advantages: the number of high quality crystal structures of antibody fragments available, which provides a substantial base for model building, the conservation of the antibody structure in the framework regions,47 and the small repertoire of main chain conformations for five of the six hypervariable regions48 and the stem fragment of the sixth.44 Comparisons of models with subsequently obtained X-ray data of the same variable regions showed that the accuracy levels in the position of most of the predicted CDR loops were better than 1.0 Å for backbone atoms.39,49,50 The positions of the side chains do occasionally differ, varying from one of the allowed conformations to the other. This is frequently observed to occur however between independent crystal structures of the same antibody, as well as between the structures of free and complexed antibodies. Thus, reasonably accurate models can be generated, which enable identification of at least the prominent structural features of the binding sites. As noted, immunolabeling indicates a clear recognition of MAb 36A1 for the (301) face of the cholesterol crystals. The step of ⬃90° in the binding site of 36A1 is due to the very short H3 and long L1 loops. It fits very well the step on the (301) face of the cholesterol crystal, both geometrically and chemically, allowing the establishment of a net of hydrogen bonds and hydrophobic interactions with the crystal surface. In particular, the array of five tyrosines on the hydrophilic wall of the step appears to be optimally 389 located to form hydrogen bonds with the hydroxyls of a corresponding array of cholesterol molecules on the crystal surface. In addition, the structural water molecules of the cholesterol monohydrate crystals, exposed on the step on the (301) face, form an ordered array. They might be conserved in part within the antibody-antigen complex, and participate in additional hydrogen bonds to the serines and tyrosines of the antibody-binding site. This type of hydrogen bond between polar amino acid residues and structured water arrays are well represented in the complex between antifreeze proteins and ice crystals.16,17 The area of the binding site of MAb 36A1 involved at the interface with the crystal is ⬃800 Å2, in good agreement with typical values of buried surfaces in antibody complexes with proteins. The model structure of MAb 122B1 is very different from MAb 36A1, both in shape and in chemistry. The antibodybinding site is extremely flat, due to both H3 and L1 loops being short. Only one serine residue emerges slightly from this flat surface, and this is within the error limit of the model. Five aromatic residues are positioned such that they can interact with the aromatic rings organized in a herring-bone structure at the surface of the flat (101) face of 1,4-DNB, and with polar (Asn and Gln) residues which are suitable for interacting with the nitro groups. It is impossible to establish the conformation of the exposed aromatic side chains with certainty. The possibility to arrange six aromatic rings such that they appear to continue the crystal lattice, is however highly suggestive of the possible optimization of aromatic interactions in this antibody-antigen complex. The area of the antibodybinding site buried in the complex is ⬃1000 Å2. Docking of both 36A1 and 122B1 was performed onto the appropriate crystal surfaces, considered as locally perfect. Although crystal surface layers are known to be often composed of islands and terraces and decorated by imperfections and dislocations, there is plenty of evidence indicating that in molecular crystals the average structure of the face is strikingly similar to that obtained by cutting the bulk structure along the relevant plane.51 Besides, islands have probably to be larger than 1,000 Å2 in order to be stable. The presence of steps and imperfections on the surface is in fact the factor that may explain why some antibody reactivity is observed on crystal faces of average structure different from that specifically recognized. Determining the specific binding properties of an antibody with no specific recognition is, by definition, impossible. In this study, however, two cross-reactive antibodies were selected, one of which, 23C1, by coincidence has a high sequence homology with 36A1. Comparison between the two antibodies may be useful in this case, for understanding their very different binding behavior. Very evident is the presence of four charged moieties in the binding site of 23C1, that are absent in 36A1. 23C1 was observed to bind to imperfections in the crystal. We suspect that the presence of charged residues in the antibody-binding site may be connected to its lack of selectivity, and to its nonspecific binding to imperfections. 390 N. KESSLER ET AL. Fig. 4. (A) Solvent-accessible surface representation of MAb 122B1 (white). The exposed residues of the antibody-binding site are highlighted. Yellow: aromatic (Phe, Tyr, and Trp); orange: Asn, Gln and Ser; green: aliphatic residues (Leu, Ile, Val, Ala, Met and Cys). (B) Model of the complex of MAb 122B1 and the (10°) face of the 1,4-DNB crystal (in space-filling representation). MAb 122B1 is shown in stick representation, except for the aromatic rings of Tyr-H32, Tyr-H97, Tyr- H98 and Trp-L94 which are shown in space-filling representation to highlight their location at lattice sites that appear to continue the crystal stacking motif. Two other residues, Tyr-L50 and Tyr-H100b which are found behind Tyr-H97 and Tyr-H32 respectively, are also located at positions matching crystal lattice sites. ANTIBODIES RECOGNIZING CRYSTAL SURFACES 391 Fig. 5. Crystal packing arrangement on the (101), (110), (101), and (100) faces of 1, 4-DNB. Yellow: carbon atoms; green: nitrogen atoms; red: oxygen atoms. The relevant surface is represented at the top of each figure, as marked (white lines). We suggest in conclusion that there may be antibodies which have an intrinsic capability to bind to all kinds of surfaces, such as those of molecular crystal addressed here, in a wide range of degrees of recognition. It appears that, as in protein-protein complexes, the recognition is modulated by sterical and chemical interactions. These may be analyzed systematically in depth, because the antigen has a repetitive homogeneous structure, known at the atomic level. The level of reliability of the antibody models built here is obviously not sufficient to allow Ångstrom-scale resolution of fine structural features. It would be thus preposterous, for example, to claim definition of the structure down to the conformation of the amino acid side-chains. However, in contrast to other instances where these very features assume a mechanistical importance, such as in catalytic antibodies, in the present system such fine tuning is not essential, inasmuch as it would not modify the general shape and character of the antibody-binding site. We are presently testing the degree of recognition of the selected antibodies for defined molecular arrays, using progressively more relaxed structures, such as monolayers of cholesterol at the air-water interface. This type of studies should further our understanding of the molecular and structural requirements of antibody-antigen surfaceto-surface interactions. MATERIALS AND METHODS Preparations Of Monoclonal Antibodies The production, purification and partial characterization of the monoclonal antibodies was previously described.7,8 cDNA Cloning and Sequencing The mRNA was purified from mouse hybridoma cells using the Guanidine Isothiocyanate (GIT) technique for preparation of total RNA.52 cDNA was prepared using Reverse Transcription kit (Promega), and amplified by the polymerase chain reaction (PCR). The following consensus oligonucleotide primers for the J region and the amino terminus of the variable region were used.53 58 terminus: 58 VK - CCCAAGCTTGACATTGTGGTGACCCAGTCTCCA 58 VH - GCGAATTCGTCGACSAGGTSMARCTGCAGSAGTCWGG 38 terminus: 58 JK - CCCGAATTCTTAGATCTCCAGCTTGGTCCC 58 kappa constant - GCGCCGTCTAGAATTAACACTCATTCCTGTTGAA 58 JH - TGARGAGACGGTGACCAGGGTBCC 58 IgM constant - GCACTAGTGCACATGTGGAGGACACG 392 N. KESSLER ET AL. TABLE III. Parent Structures Used as Templates for Modeling of the Framework Regions (VK, VH) and CDRs (L1, L2, L3, H1, H2, H3) of the Various Antibodies La (% identity) 36A1 (cholesterol) 23C1 (cross reactive) 122B1 (1,4-DNB) aVariable bVariable 1mcp, (84%) 1mcp, (91%) 1mlb, (77%) L1 L2 L3 1mcp 1mcp 1mlb 1mcp 1mcp 1mlb 1mcp 1mcp 1mlb Hb (% identity) H1 H2 H3 1vfa, (84%) 1vfa, (85%) 1jhl, (88%) 1vfa 1vfa 1jhl 1vfa 1vfa 1jhl Stem: 1vfa apex: designed 6fab Stem: 1mrf apex: designed light domain excluding the J region. heavy domain excluding the D and J regions. Amplification was performed with VentR DNA Polymerase Recombinant (New England Biolabs) in an automated thermal cycle: 1 cycle of 3 min at 94°C. 5 cycles of 3 min denaturation, at 94°C, 30 sec annealing at 45°C and 1 min extension at 72°C. 25 cycles of 30 sec denaturation, at 94°C, 30 sec annealing at 60°C and 1 min extension at 72°C. The PCR fragments were purified by Isolation kit-DNA (Biological industries Co., Beit Haemek, Israel) or by centricon-100 (Amicon). PCR fragments were sequenced using the same primers as for PCR amplification and Ampli Taq DNA Polymerase, FS cycle sequencing was performed on an ABI PRISM 377 DNA Sequencer. Characterization of the MAb Sequences Sequences were analyzed using the University of Wisconsin Genetic Computer Group software package. Stringsearch was applied on the genebank library in order to generate a library of the selected patterns: ‘‘immunoglobulin,’’ ‘‘germline.’’ and ‘‘mouse.’’ Homology comparison (FASTA) with the sequences of the selected library was performed and germline sequences with the best sequence identity were determined for each MAb. Computer-Aided Molecular Modeling Models of the Fv regions of the antibodies were constructed on the basis of experimental structures from the Protein Data bank (PDB).23,24 Initial lists of possible templates were prepared for the light and heavy chains separately, including all the structures with high percentage of sequence identity (⬎ 75%) for the framework and CDRs L1, L2, L3, H1, and H2. The relative position of the heavy and the light chains was also considered. Thus, templates with high sequence identity both for the light and the heavy chains of the modeled sequence were preferred. Where this was not possible, the structures of the candidates for the light and heavy chains were superimposed to find a pair of templates with similar relative positioning at the interface. The template structures selected for each of the antibody framework regions and CDRs are listed in Table III. For CDR loops L1, L2, L3, H1 and H2 we selected templates according to the canonical forms described by Chothia et al.48 In most cases, the same templates were used for these loops and for the framework. Otherwise, we looked for examples in the PDB for the given loop and superimposed them on the template chosen for the framework. The stem fragments of the H3 loop (defined as in Reference 44) were modeled together with the framework except for 122B1 (see Results). Templates for the apex of H3 were selected, for 23C1, based on the length of the loop, whereas for 36A1 and 122B1, where H3 is short, loops were generated using the Homology module of MSI/ Biosym (MSI/Biosym Inc, San Diego, California). Initial coordinates for each model structure were generated using the Homology module of MSI/Biosym. The new side chains inherited equivalent torsion angles from the side chains of the template structures, where possible. The models were manually checked for correct van der Waals distances, and severe clashes were fixed by changing the rotamers of amino acid side chains too close to one another. Finally, the initial models were energy minimized using program ENCAD.54 The positions of the C␣ atoms were constrained in the minimization to their initial values such that the final structure was very similar to the initial one and the overall fold of the protein was not disrupted. The same scaffold was used for the framework regions and for some of the CDRs of MAbs 36A1 and 23C1, because of the high sequence similarity in their VH (excluding H3), and VL regions. The suggested docking models were generated by manually juxtaposing the antibody model structures and the corresponding crystals surface structures, using MSI/ Biosym. ACKNOWLEDGMENTS We are grateful to S. Cabilly for providing us with primers and helping us with the cloning of the antibodies, and to D. Schindler for useful advice. L.A. is the incumbent of the Dorothy and Patrick Gorman professorial chair and N.K. is the recipient of the Jeaninne Klueger scholarship. This project was supported by a research grant from the Minerva foundation and by the Kimmelman Center for Biomolecular Assembly. REFERENCES 1. Wilson IA, Stanfield RL. Antibody-antigen interactions: New structures and new conformational changes. Curr Opin Struct Biol 1994;4:857–867. 2. Davies DR, Padlan EA, Sheriff S. Antibody-antigen complexes. Annu Rev Biochem 1990;59:439–473. ANTIBODIES RECOGNIZING CRYSTAL SURFACES 3. Wilson IA, Stanfield RL. Antibody-antigen interactions. Curr Opin Struct Biol 1993;3:113–118. 4. Westhof E, Altschuh D, Moras D, Bloomer AC, Mondragon A, Klug A, Van Regenmortel MHV. Correlation between segmental mobility and the location of antigenic determinants in proteins. Nature 1984;311:123–126. 5. Tainer JA, Getzoff ED, Alexander H, et al. The reactivity of anti-peptide antibodies is a function of the atomic mobility of sites in a protein. Nature 1984;312:127–134. 6. Davies DR, Sheriff S, Padlan EA. Antibody-antigen complexes. J Biol Chem 1988;263:10541–10544. 7. Kessler N, Perl-Treves D, Addadi L. Monoclonal antibodies that specifically recognize crystals of dinitrobenzene. FASEB J 1996;10: 1435–1442. 8. Perl-Treves D, Kessler N, Izhaky D, Addadi L. Monoclonal antibody recognition of cholesterol monohydrate crystal faces. Chem Biol 1996;3:567–577. 9. Addadi L, Weiner S. Interactions between acidic proteins and crystals: Stereochemical requirements in biomineralization. Proc Natl Acad Sci USA 1985;82:4110–4114. 10. Albeck S, Weiner S, Addadi L. Polysaccharides of intercrystalline glycoproteins modulate calcite growth in vitro. Chem Eur J 1996;2:278–284. 11. Hunter GK. Interfacial aspects of biomineralization. Curr Opin Solid State Mater Sci 1996;1:430–435. 12. Kozin F, McCarty DJ. Molecular orientation of immunoglobulin G adsorbed to microcrystalline monosodium urate monohydrate. J Lab Clin Med 1980;95:49–58. 13. Cherian PV, Schumacher HR. Immunochemical and ultrastructural characterization of serum proteins associated with monosodium urate crystals (MSU) in synovial fluid cells from patients with gout. Ultrastruct Pathol 1986;10:209–219. 14. Shiraga H, Min W, VanDusen WJ, et al. Inhibition of calcium oxalate crystal growth in vitro by uropontin: Another member of the aspartic acid-rich protein superfamily. Proc Natl Acad Sci USA 1992;89:426–430. 15. Bertrand JA, Pignol D, Bernard JP, Verdier JM, Dagorn JC, Fontecilla-Camps JC. Crystal structure of human lithostathine, the pancreatic inhibitor of stone formation. EMBO J. 1996;15: 2678–2684. 16. Sicheri F, Yang DS. Ice-binding structure and mechanism of an antifreeze protein from winter flounder. Nature 1995;375:427– 431. 17. Jia Z, DeLuca CI, Chao H, Davies PL. Structural basis for the binding of a globular antifreeze protein to ice. Nature 1996;384: 285–288. 18. Perl-Treves D, Kam M, Addadi L. Interactions between antibodies and crystal surfaces. Mol Cryst Liq Cryst 1996;278:1–15. 19. Kam M, Perl-Treves D, Caspi D, Addadi L. Antibodies against crystals. FASEB J 1992;6:2608–2613. 20. Kam M, Perl-Treves D, Sfez R, Addadi L. Specificity in the recognition of crystals by antibodies. J Mol Recognit 1994;7:257– 264. 21. Verdier JM, Ewart KV, Griffith M, Hew CL. An immune response to ice crystals in north atlantic fishes. Eur J Biochem 1996;241:740– 743. 22. Bromberg R, Kessler N, Addadi L. Antibody recognition of specific crystal faces of 1–4 dinitrobenzene. J Cryst Growth 1998;193:656– 664. 23. Bernstein FC, Koetzle TF, Williams GJ, et al. The protein data bank: A computer-based archive file for macromolecular structures. J Mol Biol 1977;112:535–542. 24. Abola EE, Bernstein FC, Bryant SH, Koetzle TF, Weng J. Protein data bank. In: Allen FH, Bergerhoff G, Sievers R, editors. Crystallographic databases-information content, software systems, scientific applications. Bonn/Cambridge/Chester: Data Commission of the International Union of Crystallography; 1987. p 107–132. 25. Tormo J, Stadler E, Skern T, et al. Three-dimensional structure of the Fab fragment of a neutralizing antibody to human rhinovirus serotype 2. Protein Sci 1992;1:1154–61. 26. Rini JM, Schulze-Gahmen U, Wilson IA. Structural evidence for induced fit as a mechanism for antibody-antigen recognition. Science 1992;255:959–965. 393 27. Satow Y, Cohen GH, Padlan EA, Davies DR. Phosphocholine binding immunoglobulin Fab McPC603. An X-ray diffraction study at 2.7 Å. J Mol Biol 1986;190:593–604. 28. Bhat TN, Bentley GA, Boulot G, et al. Bound water molecules and conformational stabilization help mediate an antigen-antibody association. Proc Natl Acad Sci USA 1994;91:1089–1093. 29. Chitarra V, Alzari PM, Bentley GA, et al. Three-dimensional structure of a heteroclitic antigen-antibody cross-reaction complex. Proc Natl Acad Sci USA 1993;90:7711–7715. 30. Saul FA, Poljak RJ. Crystal structure of human immunoglobulin fragment Fab New refined at 2.0 Å resolution. Proteins 1992;14: 363–371. 31. Suh SW, Bhat TN, Navia MA, et al. The galactan-binding immunoglobulin Fab J539: An X-ray diffraction study at 2.6-Å resolution. Proteins 1986;1:74–80. 32. Prasad L, Sharma S, Vandonselaar M, et al. Evaluation of mutagenesis for epitope mapping. Structure of an antibodyprotein antigen complex. J Biol Chem 1993;268:10705–10708. 33. Zdanov A, Li Y, Bundle DR, et al. Structure of a single-chain antibody variable domain (Fv) fragment complexed with a carbohydrate antigen at 1.7-Å; resolution. Proc Natl Acad Sci USA 1994;91:6423–6427. 34. Katchalski-Katzir E, Shariv I, Eisenstein M, Friesem AA, Aflalo C, Vakser IA. Molecular surface recognition: Determination of geometric fit between proteins and their ligands by correlation techniques. Proc Natl Acad Sci USA 1992;89:2195–2199. 35. Eisenstein M, Shariv I, Koren G, Friesem AA, Katchalski Katzir E. Modeling supra-molecular helices: Extension of the molecular surface recognition algorithm and application to the protein coat of the tobacco mosaic virus. J Mol Biol 1997;266:135–143. 36. Fan ZC, Shan L, Guddat LW, et al. Three-dimensional structure of an Fv from a human IgM immunoglobulin. J Mol Biol 1992;228: 188–207. 37. Ban N, Escobar C, Garcia R, et al. Crystal structure of an idiotype-anti-idiotype Fab complex. Proc Natl Acad Sci USA 1994;91:1604–1608. 38. Strong RK, Campbell R, Rose DR, Petsko GA, Sharon J, Margolies MN. Three-dimensional structure of murine anti-p-azophenylarsonate Fab 36–71. 1. X-ray crystallography, site-directed mutagenesis, and modeling of the complex with hapten. Biochemistry 1991;30:3739–3748. 39. Essen LO, Skerra A. The de novo design of an antibody combining site. Crystallographic analysis of the VL domain confirms the structural model. J Mol Biol 1994;238:226–244. 40. Braden BC, Souchon H, Eisele JL, et al. Three-dimensional structures of the free and the antigen-complexed Fab from monoclonal anti-lysozyme antibody D44.1. J Mol Biol 1994;243: 767–781. 41. Padlan EA, Silverton EW, Sheriff S, Cohen GH, Smith-Gill SJ, Davies DR. Structure of an antibody-antigen complex: crystal structure of the HyHEL-10 Fab-lysozyme complex. Proc Natl Acad Sci USA 1989;86:5938–5942. 42. Wien MW, Filman DJ, Stura EA, et al. Structure of the complex between the Fab fragment of a neutralizing antibody for type 1 poliovirus and its viral epitope. Nat Struct Biol 1995;2:232–243. 43. Liu H, Smith TJ, Lee WM, et al. Structure determination of an Fab fragment that neutralizes human rhinovirus 14 and analysis of the Fab-virus complex. J Mol Biol 1994;240:127–137. 44. Morea V, Tramontano A, Rustici M, Chothia C, Lesk AM. Conformations of the third hypervariable region in the VH domain of immunoglobulins. J Mol Biol 1998;275:269–294. 45. Pokkuluri PR, Bouthillier F, Li Y, Kuderova A, Lee J, Cygler M. Preparation, characterization and crystallization of an antibody Fab fragment that recognizes RNA. Crystal structures of native Fab and three Fab-mononucleotide complexes. J Mol Biol 1994;243: 283–297. 46. MacCallum RM, Martin ACR, Thornton JM. Antibody-antigen interactions: Contact analysis and binding site topography. J Mol Biol 1996;262:732–745. 47. Kabat EA, Wu TT, Bilofsky H. Unusual distribution of amino acids in complementarity-determining (hypervariable) segments of heavy and light chains of immunoglobulins and their possible roles in specificity of antibody-combining sites. J Biol Chem 1977;252:6609–6616. 394 N. KESSLER ET AL. 48. Chothia C, Lesk AM, Tramontano A, et al. Conformations of immunoglobulin hypervariable regions. Nature 1989;342:877– 883. 49. Bajorath J, Sheriff S. Comparison of an antibody model with an X-Ray structure: The variable fragment of BR96. Proteins 1996;24: 152–157. 50. Eigenbrot C, Randal M, Presta L, Carter P, Kossiakoff AA. X-ray structures of the antigen-binding domains from three variants of humanized anti-p185HER2 antibody 4D5 and comparison with molecular modeling. J Mol Biol 1993;229:969–995. 51. Weissbuch I, Addadi L, Lahav M, Leiserowitz L. Molecular recognition at crystal interfaces. Science 1991;253:637–645. 52. Chirgwin JM, Przybyla AE, MacDonald RJ, Rutter WJ. Isolation of biologically active ribonucleic acid from sources enriched in ribonuclease. Biochemistry 1979;18:5294–5299. 53. Orlandi R, Gussow DH, Jones PT, Winter G. Cloning of immunoglobulin variable domains for expression by the polymerase chain reaction. Proc Natl Acad Sci USA 1989;86:3833–3837. 54. Levitt M. Molecular dynamics of native protein,I: Computer simulation of trajectories. J Mol Biol 1983;168:595–617. 55. Kabat EA, Wu TT, Perry HM, Gottesman KS, Foeller C. Sequences of proteins of immunological interest. Fifth edition. Washington D.C.: U.S. Department of Health and Human Services, National Institutes of Health Publication:91–3242. 1991. 56. Chothia C, Lesk AM. Canonical structures for the hypervariable regions of immunoglobulins. J Mol Biol 1987;196:901–917. 57. Kofler R, Geley S, Kofler H, Helmberg A. Mouse variable-region gene families: Complexity, polymorphism and use in nonautoimmune responses. Immunol Rev 1992;128:5–21. 58. Max EE, Seidman JG, Leder P. Sequences of five potential recombination sites encoded close to an immunoglobulin k constant region gene. Proc Natl Acad Sci USA 1979;76:3450–3454. 59. Sakano H, Maki R, Kurosawa Y, Roeder W, Tonegawa S. Two types of somatic recombination are necessary for the generation of complete immunoglobulin heavy-chain genes. Nature 1980;286: 676–683. 60. Kurosawa Y, Tonegawa S. Organization, structure and assembly of immunoglobulin heavy chain diversity DNA segments. J Exp Med 1982;155:201–218.