Paratope Determination of the Antithrombotic Antibody 82D6A3 Based on the Crystal Structure of Its Complex with the von Willebrand Factor A3-Domain*

The antithrombotic monoclonal antibody 82D6A3 is directed against amino acids Arg-963, Pro-981, Asp-1009, Arg-1016, Ser-1020, Met-1022, and His-1023 of the von Willebrand factor A3-domain (Vanhoorelbeke, K., Depraetere, H., Romijn, R. A., Huizinga, E., De Maeyer, M., and Deckmyn, H. (2003) J. Biol. Chem. 278, 37815–37821). By this, it potently inhibits the interaction of von Willebrand factor to collagens, which is a prerequisite for blood platelet adhesion to the injured vessel wall at sites of high shear. To fully understand the mode of action of 82D6A3 at the molecular level, we resolved its crystal structure in complex with the A3-domain and fine mapped its paratope by construction and characterization of 13 mutants. The paratope predominantly consists of two short sequences in the heavy chain CDR1 (Asn-31 and Tyr-32) and CDR3 (Asp-99, Pro-101, Tyr-102 and Tyr-103), forming one patch on the surface of the antibody. Trp-50 of the heavy and His-49 of the light chain, both situated adjacent to the patch, play ancillary roles in antigen binding. The crystal structure furthermore confirms the epitope location, which largely overlaps with the collagen binding site deduced from mutagenesis of the A3-domain (Romijn, R. A., Westein, E., Bouma, B., Schiphorst, M. E., Sixma, J. J., Lenting, P. J., and Huizinga, E. G. (2003) J. Biol. Chem. 278, 15035–15039). We herewith further consolidate the location of the collagen binding site and reveal that the potent action of the antibody is due to direct competition for the same interaction site. This information allows the design of a paratope-mimicking peptide with antithrombotic properties.

Platelet adhesion, activation, and aggregation are the three major steps in primary hemostasis, a process that is necessary for the cessation of bleeding at sites of vascular injury. The same process is also responsible for pathologic thrombus formation, particularly in arteries at sites of high shear stress. At high shear stress, binding of von Willebrand factor (VWF) 3 to collagen that is exposed in the subendothelial matrix after injury, and the subsequent interaction of the immobilized VWF with the platelet glycoprotein (GP) Ib/IX/V-receptor complex, is a prerequisite for proper adhesion (1)(2)(3). As a result of the reversible VWF-GP Ib/IX/V interaction, platelets roll over the injured vessel wall and slow down (3). Subsequent firm adhesion and activation of platelets are mediated by their collagen receptors, integrins ␣2␤1 and GPVI, respectively (4 -8). Activated platelets expose high affinity GPIIb/IIIa receptors that mediate aggregation by binding to fibrinogen and, to a lesser extent, to VWF (9 -11).
Collagen type VI and fibrillar collagens types I and III are abundant in the subendothelium (12). VWF interacts with the fibrillar collagens mainly via its A3-domain (13,14), whereas its A1-domain has been implicated in binding to collagen type VI (15,16). The latter interaction only mediates platelet adhesion under low shear conditions (17,18). The A3-domain is homologous in sequence and structure with the I-domains of collagen binding integrin ␣-chains, such as the I-domain of ␣2␤1 (19 -21). These domains adopt a dinucleotide binding fold, or Rossman fold, that consists of a central hydrophobic parallel ␤-sheet flanked on both sites with amphipathic ␣-helices (22)(23)(24). Despite their structural similarity, they bind collagen in distinctly different ways. The I-domain of ␣2␤1 binds collagen in a hydrophilic groove at its top face that contains a metal ion-dependent adhesion site (25,26). In contrast, the collagen-binding site of the A3-domain is located at its front face and is rather flat and hydrophobic (23,(27)(28)(29).
Our group characterized a monoclonal antibody against human VWF, 82D6A3, that binds with high affinity to the VWF A3-domain and inhibits VWF binding to collagen types I and III (15). Its inhibiting effect on VWF-mediated platelet adhesion becomes more pronounced with increasing shear stress, as is expected because the collagen-VWF-GPIb axis is essential for platelet adhesion under high shear only (30). The discontinuous epitope of 82D6A3, which has recently been unraveled via phage display and binding studies with alanine mutants of the VWF A3-domain, includes several residues of the A3-domain that are important for collagen binding (30). We further demonstrated that 82D6A3 has antithrombotic effects in a baboon arterial thrombosis model (31), proving for the first time that the VWF-collagen interaction is indeed involved in thrombus formation in vivo (32). Interestingly, the antithrombotic effect was not associated with significantly increased bleeding times, even at high doses, demonstrating that administration of 82D6A3 results in a broad therapeutic window. These benefits were also observed in in vivo studies with saratin, another inhibitor of the VWFcollagen interaction (33,34), and with monoclonal antibodies 6B4 (35,36) and AjvW-2 (37,38), which inhibit VWF-GPIb binding. This suggests that interference with the collagen-VWF-GPIb axis could have important advantages over existing antithrombotic agents such as inhibitors of platelet aggregation (GPIIb/IIIa-receptor blockers) or platelet activation (aspirin). Treatment with the latter agents results in smaller therapeutic windows and is frequently associated with bleeding problems and thus requires close monitoring of the patients (39 -41). In addition, compounds that interfere with the collagen-VWF-GPIb interaction could also be more effective in the prevention of intimal hyperplasia after angioplasty and endarterectomy, because they interfere with the first steps in primary hemostasis and thus prevent the release of smooth muscle cell mitogenic (e.g. transforming growth factor ␤1) and attractant (e.g. basic fibroblast growth factor) factors during platelet activation (42,43).
The goal of this study was to determine which residues of 82D6A3 are involved in the interaction with the VWF A3-domain. In the future, this information will be used as a lead for the construction of a peptidomimetic with similar inhibitory characteristics as 82D6A3. On the basis of the crystal structure of the complex between 82D6A3 and the VWF A3-domain determined in this study, we selected amino acids of the complementarity-determining regions (CDR) of the antibody for mutation to alanine. Binding studies with alanine mutants of the single chain variable fragment of 82D6A3 (scFv-82D6A3) resulted in the identification of the paratope.

EXPERIMENTAL PROCEDURES
Materials-The pCANTAB 5 E expression vector, pCANTAB 5 E gene rescue primer R2 and horseradish peroxidase (HRP)-labeled anti-E Tag antibodies (anti-E Tag-HRP) were purchased from Amersham Biosciences. Human full-length VWF purified from donor plasma displaying a normal multimeric pattern was purchased from Red Cross, Belgium. Bacti casein peptone, Bacti yeast extract, and Bacti purified agar were purchased from BioTrading (Keerbergen, Belgium). PCR buffer, dNTP, and primers for the construction of the scFv-DNA and for its ligation in the prokaryotic expression vector were purchased from Invitrogen. Primers for the construction of the mutants were purchased from Eurogentec (Seraing, Belgium).
Preparation of VWF A3-domain and Fab Fragments of 82D6A3-Recombinant VWF A3-domain comprising residues 920 -1111 of mature VWF was expressed and purified essentially as described (22) with one modification. During removal of the His tag with thrombin the A3-domain precipitated. Resolubilization without significant protein loss was achieved by the addition of solid urea to a final concentration of 7 M followed by overnight dialysis against 25 mM Tris/HCl, pH 8.2. Subsequent purification proceeded as described previously.
82D6A3 was raised in mice against human VWF (15). The IgG of 82D6A3 was purified from earlier produced ascites, and Fab fragments were prepared as previously described (30) and dialysed to 25 mM Tris/ HCl, pH 8.2, and 50 mM NaCl.
For data collection, crystals were transferred to reservoir solution supplemented with 20% (v/v) glycerol and flash frozen in liquid nitrogen. Diffraction data were collected at 100 K on beam line ID14 EH4 at the European Synchrotron Radiation Facility (Grenoble, France). Data were collected to 1.9 Å resolution with oscillations of 0.7°to reduce spot overlap. Data were processed using computer programs MOSFLM, SCALA, and TRUNCATE (44,45). Data collection statistics are found in Table 1.
Structure Determination-The structure of the complex of the A3-domain and 82D6A3 was solved by molecular replacement using the program CNS (46). One Fab fragment (47) (Protein Data Bank accession code 2MPA) and one A3-domain (22) (Protein Data Bank accession code 1ATZ) could be oriented and placed in the asymmetric unit. Subsequent rigid body refinement led to an initial model with an R-factor of 44.6%. Automatic model building, using the program ARP/ wARP (48), built ϳ80% of the main chain. The model was manually corrected and supplemented in "O" (49). The model was refined with multiple cycles of TLS and restrained refinement using the program Refmac (50,51), followed by manual rebuilding in "O." Solvent molecules were added to the model automatically using ARP/wARP. Waters were placed in electron difference peaks with a height Ͼ3 ⌺ and were checked manually in "O." Cross-validation, using a 5% test set of reflections, was used throughout refinement. Figures were generated using PyMOL (pymol.sourceforge.net/).
Prokaryotic Expression and Purification of Wild Type scFv of 82D6A3 (scFv-82D6A3(WT))-The construction of scFv-82D6A3(WT) (European Bioinformatics Institute Database accession number AJ965435) through splicing by overlap extension, its cloning in the pCANTAB 5 E vector, and its expression as a soluble protein in the periplasm of Escherichia coli HB2151 cells has been previously described (52). The scFv-82D6A3(WT) was purified from the periplasmic extract and the supernatant as described, using a HiTrap anti-E Tag column (RPAS purification module; Amersham Biosciences) (52). After dialysis against PBS, purity was checked in SDS-PAGE using a 10% gel followed by staining with Coomassie Brilliant Blue.
Construction, Expression, and Purification of Alanine Mutants of scFv-82D6A3-Construction of alanine mutants was performed with the QuikChange site-directed mutagenesis kit (Stratagene) using pCANTAB 5 E-scFv-82D6A3(WT)-DNA and the primers listed in Table 2. The DpnI-digested mutagenesis mixture was immediately used to transform E. coli HB2151 cells according to the manufacturer's instructions (RPAS expression module; Amersham Biosciences). Plasmid DNA was prepared using the NucleoSpin Plasmid Quick Pure kit (Machery-Nagel, Düren, Germany). Sequencing, expression, and purification were performed as described (52). After dialysis against PBS, purity was checked as described above.
Western Blot-Equal volumes of supernatant, periplasm, or elution fractions containing scFv-82D6A3(WT) or its alanine mutants were loaded on a 10% SDS-PAGE gel and transferred to a nitrocellulose membrane (Schleicher & Schuell BioScience, Dassel, Germany). Detection of the scFvs with anti-E Tag-horseradish peroxidase antibodies was performed as described (52).
Concentration Determination of the Purified scFvs-The concentration of the purified scFvs was determined according to the Bradford method (53) using the Coomassie Plus protein assay kit (Pierce). Briefly, dilution series of purified scFvs and Albumin Standard in PBS were added to a 96-well plate (Greiner, Frickenhausen, Germany). A fixed volume of Coomassie Plus protein assay reagent was added, and absorbance was measured at 630 nm.
Binding of scFv-82D6A3(WT) and Its Alanine Mutants to VWF-Binding of scFv-82D6A3(WT) and its alanine mutants to VWF was performed as described before (52). Briefly, a 96-well plate was coated with VWF and incubated with dilution series of purified scFv-82D6A3(WT) and its alanine mutants. Next, bound scFv was detected by incubation with anti-E Tag-horseradish peroxidase antibodies. Visualization was performed by adding H 2 O 2 and orthophenylenediamine (Sigma). The plate was washed three times after coating and blocking and nine times elsewhere with PBS, 0.1% Tween 20.
Inhibition of the scFv Binding to VWF by IgG-82D6A3-A 96-well plate was coated with VWF, blocked, and incubated as described above with scFv-82D6A3(WT) and its alanine mutants at their respective EC 50 concentrations as calculated from their binding curves. IgG-82D6A3 (15 g/ml in PBS, 0.3% milk powder) was added to the plate for 1 h at room temperature. Detection of bound scFv, visualization, and washing were performed as described above.

RESULTS
Structure Determination and Refinement-We solved the crystal structure of the complex between the VWF A3-domain and the Fab fragment of 82D6A3 to 1.9 Å resolution ( Fig. 1A) (Protein Data Bank accession code 2ADF). The structure has been refined to an R-factor of 18.5% and a free R-factor of 21.7% and shows good geometry ( Table 1). The final model contains residues 922-1110 of the A3-domain, residues 1-210 of the 82D6A3 light chain, and residues 1-216 of the 82D6A3 heavy chain. The electron density of residues 122-130, 152-156, 181-189, and 199 -201 in the constant domain of the light chain is ill defined, and the interpretation of these residues is tentative. This does not affect the analysis of the A3-82D6A3 interaction, because these residues are not located in the vicinity of the antigen binding surface of the antibody.
The Epitope of 82D6A3-82D6A3 binds a discontinuous epitope located at a side face of the A3-domain (Fig. 1B). The interaction buries a solvent-accessible surface of ϳ1640 Å 2 . The epitope consists of residues located in strand ␤3, helix ␣3, and loops ␤2-␤3 and ␣3-␤4. Loop ␣3-␤4 and ␣3 residue Arg-1016 form a protrusion on the surface of the A3-domain that is nicely accommodated by a cleft between the light and heavy chain subunits of 82D6A3. Accordingly, mutations of Arg-1016 and residues Ser-1020, Met-1022, and His-1023 of ␣3-␤4 led to a significant decrease in binding of VWF to 82D6A3 (30). Residues in the A3-domain implicated in collagen binding, as defined by mutagenesis studies, are located in helix ␣3, strand ␤3, and loops ␣2-␣3 and ␣3-␤4 (27)(28)(29). Although it cannot be excluded that the mutations in the A3-domain have caused indirect effects on collagen binding, these residues define a putative collagen binding surface that overlaps extensively with the footprint of 82D6A3 (Fig. 1C), and thus 82D6A3 effectively blocks collagen binding to VWF by direct competition for the same binding site.
82D6A3 Residues Interacting with the A3-Domain-Residues of 82D6A3 that interact with the A3-domain are located in all six CDRs forming five hydrogen bonds and three salt bridges, listed in Table 3, and numerous van der Waals interactions. The heavy chain seems to be the most important in antigen binding with 608 Å 2 of its solvent-acces-   -CA GGG TAT ACC TTT ATA GCT TAT GGA ATG AAC TGG G-3Ј  H32Tyr 3 Ala  5Ј-GGG TAT ACC TTT ATA AAT GCT GGA ATG AAC TGG GTG-3Ј  H50Trp 3 Ala  5Ј-GT TTA AAG TGG ATG GGC GCG AAA AAC ACC AAC ACT GG- sible surface buried in the complex compared with only 216 Å 2 for the light chain. As is frequently observed in antigen binding, CDR3 of the heavy chain is involved in many interactions. A salt bridge is present between H99Asp of 82D6A3 and His-1023 of the A3-domain ( Fig. 2A). Furthermore, H103Tyr interacts through hydrogen bonding with the side chain of Asp-1009 and the main chain nitrogen of Thr-976. Extensive hydrophobic interactions exist between CDR3 residues H101Pro, H102Tyr, and H103Tyr and A3-domain residues Ile-975, Gly-1012, Phe-1013, Arg-1016, and Tyr-1017. In CDR1 the two main interacting residues are H31Asn and H32Tyr (Fig. 2B). H31Asn forms hydrogen bonds to the main chain oxygen atoms of Asp-979 and Ile-978 and makes hydrophobic contacts with Pro-981, while H32Tyr forms a hydrogen bond with the main chain nitrogen of Ile-978. CDR2 contributes little to the interaction (not shown). Distances between atoms of CDR2 residue H54Asn and Asn-983 and Trp-982 of the A3-domain are too long (3.7 and 3.8 Å, respectively) to qualify as hydrogen bonds. Moreover, only one residue of CDR2, H50Trp, is involved in hydrophobic contacts with the A3-domain, more specifically with Met-1022 and His-1023.
Interacting residues of the light chain (Fig. 2C) are mainly located in CDR2. CDR2 residue L49His forms a salt bridge with A3-domain residue Asp-1009, and L56Pro makes hydrophobic contacts with Ile-975. In CDR3, L95Arg makes hydrophobic contacts with Met-1022, and a rather long salt bridge is formed between L92Asp and Arg-1016 of the A3-domain. The only interaction between CDR1 of the light chain and the A3-domain is a hydrophobic interaction between L32Tyr and Arg-1016, respectively.
Paratope Definition Using Alanine Mutants of scFv-82D6A3(WT)-To define more precisely the role of individual amino acids of 82D6A3 in binding to VWF-A3, we constructed alanine mutants of 82D6A3 and measured their interaction with VWF. Because, as previously reported (52), the scFv of 82D6A3 binds to VWF and inhibits its interaction with collagen, the DNA coding for this antibody fragment was used for the construction of the alanine mutants. We mutated all 82D6A3 residues that form hydrogen bonds or salt bridges with residues of the A3-domain (Table 3) as well as residues that are located within 4.0 Å of the A3-domain and have a buried solvent-accessible surface area in the complex of at least 20 Å 2 . A total of thirteen residues were mutated to alanine, eight in the heavy chain (H31Asn 3 Ala, H32Tyr 3 Ala, H50Trp 3 Ala, H54Asn 3 Ala, H99Asp 3 Ala, H101Pro 3 Ala, H102Tyr 3 Ala, and H103Tyr 3 Ala) and five in the light chain (L32Tyr 3 Ala, L49His 3 Ala, L56Pro 3 Ala, L92Asp 3 Ala, L95Arg 3 Ala). All mutated proteins were expressed in E. coli HB2151 cells. Expression levels varied between 84 and 103% of the expression level obtained with scFv-82D6A3(WT) (data not shown). All proteins were subsequently purified from the supernatant and periplasmic extract to Ͼ90% purity (Fig. 3). Next, the effect of the individual mutations on the binding of the scFv to VWF was studied by enzyme-linked immunosorbent assay (Fig. 4). Except for H54Asn 3 Ala, all mutations in the heavy chain of 82D6A3 substantially decrease the affinity for VWF (Fig. 4A). Mutations in the light chain, however, only have a moderate, or no, effect on binding (Fig. 4B). Moreover, the binding of wild type and mutant scFv-82D6A3 to VWF was inhibited by IgG-82D6A3 (data not shown), proving the specificity of the binding.

DISCUSSION
Inhibitors of the interaction of VWF with collagen types I and III as new antithrombotic compounds can have important advantages over currently used antithrombotic agents, such as GPIIbIIIa antagonists. Because the VWF-collagen interaction only mediates platelet adhesion and thrombus formation in high shear stress conditions, overall hemostasis is only minimally affected by inhibition of this interaction, which might result in a reduced risk for bleeding complications. Moreover, by interfering with the first step in primary hemostasis, inhibitors of the VWF-collagen interaction also prevent platelet activation, a process that mediates intimal hyperplasia after angioplasty or endarterectomy by the release of smooth muscle cell-proliferating and -attracting factors (34,54).
Indeed, previous in vivo studies demonstrated that 82D6A3, a monoclonal antibody that inhibits the VWF-collagen interaction through binding to the VWF A3-domain, is a powerful antithrombotic agent. Importantly, administration of 82D6A3 did not significantly prolong the bleeding time, even at high doses (31). To maximally exploit the in vivo effects of 82D6A3, developing an orally active compound with the same characteristics as the monoclonal antibody is a prerequisite. As a first step in the development of such a new antithrombotic peptidomimetic related to 82D6A3, we determined its paratope by resolving the crystal structure of the complex between a Fab fragment of 82D6A3 and the VWF A3-domain combined with mutagenesis studies.
The crystal structure of the complex shows that most of the A3 residues previously reported as being part of the epitope (Arg-963, Pro-981, Asp-1009, Arg-1016, Ser-1020, Met-1022, and His-1023) (30) do indeed interact with 82D6A3. This is not true, however, for Ser-1020. Mutation of this residue to Ala affects binding to 82D6A3, but in the complex no direct interactions involving this residue are observed. Apparently, Ser-1020 plays an indirect role in complex formation. It may do so by hydrogen bonding to His-1023, an important residue for 82D6A3 binding (30), thereby maintaining a conformation critical for recognition by 82D6A3. In addition, no direct interactions involving Arg-963 are observed in the complex, whereas mutation of this residue decreases the binding to 82D6A3. However, because the decrease in binding is subtle (25%) (30), this suggests a minor role in complex formation at most, which is now confirmed by the crystal structure. For Asp-979 the opposite is true. Mutation of this residue has no effect on binding, although it  is involved in direct interactions with 82D6A3 residue H31Asn. This apparent discrepancy is explained by the fact that H31Asn hydrogen bonds to the main chain carbonyl of Asp-979, an interaction that is not affected by mutagenesis.
The epitope now observed in the crystal structure supports our previous conclusion from binding studies that the collagen and 82D6A3 binding sites on A3 overlap substantially (Fig. 1C) (30). A much smaller overlap with the putative collagen binding site is observed in the crystal structure of the VWF A3-domain bound to RU5 (27) (Protein Data Bank accession code 1FE8), an antibody that also inhibits collagen binding. Although 82D6A3 extensively interacts with the collagen binding face of A3, RU5 interacts with this face only to a very limited extent and predominantly engages the bottom face. Whereas the RU5 and 82D6A3 binding sites are distinctly different, both molecules, when bound to A3, do occupy a small common volume of space, which explains why they compete for binding to VWF.
On the basis of the crystal structure of the A3-82D6A3 complex we selected amino acids for mutagenesis to determine the paratope more precisely. Selection criteria used were involvement in an intermolecular salt bridge or hydrogen bond or a decrease in solvent-accessible surface area of Ͼ20 Å 2 . The thirteen residues that were mutated account for ϳ85% of the total surface area of 82D6A3 buried in the complex. For mutagenesis, we used the scFv of 82D6A3, because the DNA coding for the scFv is much smaller and easier to handle than the DNA coding for the IgG. This is justified because the binding and inhibition characteristics of the scFv and the IgG are very similar (52). As the protein yield after purification was not the same for all mutants, despite the use of an anti-E Tag column, we cannot exclude that the alanine substitutions might have an influence on this. However, as the expression level of all mutants was comparable with that of scFv-82D6A3(WT) and varied between 84 and 103% of the expression level of scFv-82D6A3(WT), this indicates that the alanine substitutions did not markedly affect the structural integrity of the molecule by inducing global conformational changes. We studied the binding of the alanine mutants using fulllength VWF instead of the A3-domain, because this is more physiologically relevant. In the binding studies we observed that 5 of 13 mutations tested do not affect VWF binding significantly, suggesting that our selection of residues has not been overly restrictive and that it likely includes all residues that contribute to the interaction significantly.
Residues identified as being essential for antigen binding are H31Asn, H32Tyr, H99Asp, H101Pro, H102Tyr, and H103Tyr. These residues are located in two loops and form one continuous patch on the surface of the antibody. H50Trp and L49His, situated adjacent to this patch, play ancillary roles in binding (Fig. 5).
Before we succeeded in resolving the crystal structure of the 82D6A3-A3-domain complex, we computationally determined the structure of the complex between the crystal structure of the VWF A3-domain (22) and the homology-modeled Fv-82D6A3 (52) using the automated docking program DOT (www.sdsc.edu/CCMS/DOT/). Interestingly, the position of 82D6A3 on the A3-domain in the obtained model corresponded strikingly well with the one in the co-crystal, implying that docking successfully predicted the overall structure of the 82D6A3-A3domain complex. As a consequence, similar amino acids interacting with the A3-domain were identified both in the co-crystal and the docking model (H102Tyr, H103Tyr, L32Tyr, and L92Asp), of which two indeed turned out to be part of the paratope (H102Tyr and H103Tyr). Although the overall structure of the homology-modeled Fv-82D6A3, which was used in the docking procedure, corresponded well with the structure of the Fv-82D6A3 in the co-crystal, subtle differences in the conformation of the CDR loops were present. As a consequence, additional 82D6A3 residues interacting with the A3-domain were identified in the co-crystal.
The properties of the dominant region of the paratope, two short loops that form a continuous surface patch, are typical for antigen recognition segments of antibodies. Although it will be a challenge to construct a mimic peptide based on the paratope, techniques have been developed to mimic the tertiary structure of multiple loops in order to direct the critical amino acids into the same conformation and orientation as in the bio-active surface of the native protein (55,56), and successful design of mimic peptides derived from CDRs has been reported (57,58). The construction of such a peptide resembling the paratope in both charge distribution and conformation and investigation of its binding and inhibiting characteristics will be the next step in the development of an orally active antithrombotic compound with potentially important advantages over the existing antithrombotic agents. FIGURE 5. The paratope of 82D6A3. A, ribbon diagram of the variable domains of 82D6A3 with amino acids of the paratope shown in stick representation and colored according to the extent of residual VWF binding observed if mutated to alanine: 0 -40%, red; 40 -80%, yellow; 80 -100%, green. B, surface representation of the variable domains of 82D6A3 with surfaces corresponding to residues in the paratope color coded as in panel A.