Paratope and Epitope Mapping of the Antithrombotic Antibody 6B4 in Complex with Platelet Glycoprotein Ibα*

The monoclonal antibody 6B4 has a potent antithrombotic effect in nonhuman primates by binding to the flexible loop, also known as the β-switch region (amino acids 230-242), of glycoprotein Ibα (GPIbα). This interaction blocks, in high shear stress conditions, the specific interaction between GPIbα and von Willebrand factor suppressing platelet deposition to the damaged vessel wall, a key event in the pathogenesis of arterial thrombosis. To understand the interactions between this antibody and its antigen at the amino acid level, we here report the identification of the paratope and epitope in 6B4 and GPIbα, respectively, by using computer modeling and site-directed mutagenesis. The docking programs ZDOCK (rigid body docking) and HADDOCK (flexible docking) were used to model the interaction of 6B4 with GPIbα and to delineate the respective paratope and epitope. 6B4 and GPIbα mutants were constructed and assayed for their capacity to bind GPIbα and 6B4, respectively. From these data, it is found that the paratope of 6B4 is mainly formed by five residues: Tyr27D, Lys27E, Asp28, and Glu93 located in light chain CDR1 and -3, respectively, and Tyr100C of the heavy chain CDR3. These residues form a valley, where the GPIbα flexible loop can bind via residues Asp235 and Lys237. The experimental results were finally used to build a more accurate docking model. Taken together, this information provides guidelines for the design of new derivatized lead compounds with antithrombotic properties.

Platelets are a key factor in hemostasis (1). However, in some pathological situations, such as stroke or myocardial infarction, shear rate increases, causing platelet activation and thrombus formation, leading to vessel occlusion. This process is dependent on the binding of the platelet glycoprotein Ib␣ (GPIb␣) 4 to von Willebrand factor (VWF), which is bound to the collagen matrix exposed to the flowing blood upon vessel damage.
The structure of GPIb␣ consists of a globular N-terminal region, a sialomucin core, an anionic sequence, a transmembrane region, and a cytoplasmic tail. The N-terminal region (residues 1-282) consists of eight leucine-rich repeats (LRRs) and contains the binding sites for VWF, ␣-thrombin, P-selectin, Mac-1, high molecular weight kininogen, and coagulation factors XI and XII (2)(3)(4). Under nonliganded conditions, platelets present the flexible loop (residues 230 -242) within the N-terminal domain of GPIb␣ into a ␤-switch conformation, which changes upon binding of VWF into a ␤-hairpin conformation, extending the existing VWF antiparallel ␤-sheet (5). The GPIb␣ gain-of-function mutations G233V and M239V found in platelet-type von Willebrand disease stabilize the ␤-hairpin conformation and increase the affinity of GPIb␣ for VWF 5-6-fold (4,6). The globular domain is presented well above the plasma membrane by the sialomucin core, which is connected by a flexible hinge domain: the anionic sequence. The cytoplasmic tail of GPIb␣ contains binding sites for filamin A and 14-3-3 , which play an important role in intracellular signaling upon ligand binding (3,7,8).
Previously, we prepared and characterized a murine monoclonal antibody (mAb) targeting the human GPIb␣, designated as 6B4 (9). This mAb inhibits platelet adhesion under high shear stress conditions, as was shown in flow chambers (10). Injection of 6B4-Fab fragments has a potent in vivo antithrombotic effect in baboons (9,11) but also on inhibiting ex vivo ristocetin-induced platelet aggregation (9). Contrary to most antithrombotic drugs, 6B4-Fab administration did not induce a significant prolongation of the bleeding time. The epitope recognized by 6B4 was mapped previously, using human/canine chimeric rGPIb␣, to be within the C-terminal flanking region, between residues 201 and 268 (10), containing the flexible loop (residue 230 -242) within the N-terminal domain of GPIb␣. An indication that upon binding of 6B4, this loop might not assume the ␤-hairpin conformation, as seen upon binding of VWF, comes from the finding that 6B4 no longer binds to the gain-offunction G233V and M239V (5,10). The goal of this study was to further determine which residues are involved in the binding of 6B4 to GPIb␣.
Docking approaches using computer programs such as ZDOCK (12), an algorithm more appropriate as an initial stage docking algorithm to explore vast putative binding areas in cases were the target binding site is unknown or less defined, and the HADDOCK1.3 method, which allows flexibility (13)(14)(15) in both ligand and target, were used to predict interactions between ligand-bound or ligand-free GPIb␣ and various models of 6B4, followed by mutagenesis experiments in an iterative approach. By this method, we identified residues in the complementarity determining regions (CDRs) of 6B4, crucial for the binding and constituting its paratope. In parallel, the epitope mapping on the flexible loop of GPIb␣ was confirmed and refined. An optimized interaction model was finally constructed combining all of the findings.

EXPERIMENTAL PROCEDURES
Materials-Protein A-Sepharose CL-4B and ECL TM were purchased from Amersham Biosciences. The QuikChange XL site-directed mutagenesis kit was provided by Stratagene (La Jolla, CA), and the DNA sequencing was performed by Genome Express (Meylan, France). Primer sequences listed in Table 1 were purchased from Eurogentec (Seraing, Belgium). Cell culture products for the IgG 4 expression and Lipofectamine 2000 TM were provided by Invitrogen. Centriprep-30 and Centricon-100 devices were provided by Millipore (Billerica, MA). The monoclonal antihuman IgG 4 , Fc-specific Ab was from BD Pharmigen (San Diego, CA), the anti-human IgG horseradish peroxidase-labeled Ab was purchased from Imtec Diagnostic (Antwerpen, Belgium), and the goat anti-mouse horseradish peroxidase-labeled Ab and the orthophenylenediamine were from Sigma.
Docking-Docking of the 6B4-Fab to GPIb␣ (amino acids 1-266) was performed using the software ZDOCK (12) and later HADDOCK1.3 (14). Results were analyzed with the Brugel TM modeling package (16). The starting structures for the docking were four computer models of 6B4-Fab; one model was built by Algonomics (17), and three models were retrieved from the Web Antibody Modeling tool (18), the structure of the ligand-free conformation of GPIb␣ (Protein Data Bank code 1M0Z) (5) and the structure of the ligand-bound conformation of GPIb␣ (Protein Data Bank code 1SQ0) (19). The HADDOCK option to use structural information from all models in the same run was selected. As a guide to the docking, a list of possible interaction sites was used. HADDOCK divides the interacting residues into two classes: active residues, which play an important role in binding, and passive residues, which may be indirectly involved in the binding. On the mAb (the ligand) side, it is obvious that the CDRs are involved in the binding. Previous experimental evidence indicated that in GPIb␣, one major loop (the flexible loop at amino acids 228 -242) is involved in the binding (10). The solvent-accessible residues of the CDRs and the GPIb␣ flexible loop were defined as passive residues in the docking experiment. The preliminary ZDOCK (12) and consequent site-directed mutagenesis had indicated that Y27DA and E93A (Kabat numbering (20)), situated in the light chain CDR1 and -3, respectively, contribute directly to the binding. Therefore, these residues were defined as being active. In the docking experiment, only the variable domains of the mAb were used. Since these were not covalently bonded, distance restraints were used to keep the light and heavy chains together during the simulated annealing phase of the docking.
Production and Purification of Monoclonal Antibodies-The mAbs 6B4, 27A10, and 24G10 were developed in mice a Mutated residues are located on the variable domain of the light (VL) or heavy (VH) chain of 6B4-IgG 4 . The name of the mutant includes the WT residue type (single-letter code) followed by its position in Kabat numbering (number with, if needed, an insertion character) followed by the single-letter code of the mutant residue. The three-letter code for the changes and the rationale for the mutation is also given. b Oligonucleotide sequence of the primers used to construct light or heavy chain mutants of 6B4-IgG 4 (17). The sense primer hybridizes with the noncoding strand of either pKaneo-CM30-L var or pKaneo-50-dhfr-H leu var, used for the preparation of 6B4 mutants. The boldface and underlined codons introduce the mutation. against purified human GPIb␣ (9), purified by affinity chromatography with protein A-Sepharose CL-4B, and dialyzed overnight at 4°C against PBS. Antibody purity was checked by SDS-PAGE under nonreducing conditions, followed by Coomassie Brilliant Blue staining. Concentrations were evaluated by optical density at 280 nm, and antibodies were kept at Ϫ20°C before use. Construction and Expression of 6B4 WT and Mutants-Recombinant 6B4 WT and mutants were prepared as chimeric human/murine IgG 4 as previously described (17). Construction of 6B4 mutants was performed with the QuikChange XL site-directed mutagenesis kit according to the manufacturer's instructions using pKaneo-CM30-L var and pKaneo-50-dhfr-H leu var vectors, coding for the respective chimeric light and heavy chain of 6B4, and the appropriate primer couple (Table 1). After DpnI digestion and bacteria transformation, clones positive for the presence of plasmid DNA were selected, and their purified DNA was sequenced. All 6B4 antibodies were expressed in a transient expression system using human embryonic kidney cell line 293T/17 and Lipofectamine 2000 TM as described before (17).
Purification and Characterization of 6B4 WT and Mutants-The different expressed antibodies were purified on a protein A-Sepharose CL-4B column and dialyzed against phosphatebuffered saline overnight at 4°C. Quality control was performed by SDS-PAGE and Western blot analysis using a monoclonal anti-human IgG 4 -Fc-specific Ab, followed by goat anti-mouse horseradish peroxidase-labeled Ab before revelation using ECL TM . Antibody concentration was estimated, in comparison with an IgG 4 reference, by sandwich ELISA using anti-human IgG 4 -Fc-specific Ab for capture and an anti-human IgG horseradish peroxidase-labeled Ab for detection. mAb concentrations were adjusted to 1 g/ml and kept at Ϫ20°C before use.
Production and Characterization of rGPIb␣ Mutants-WT and mutant rGPIb␣ were produced in a transient expression system using 293T/17 cells and Lipofectamine 2000 TM . After 48 h, rGPIb␣ secreted in the medium was concentrated using Centriprep-30 and Centricon-100 devices. The concentration of each mutant was determined by a two-step ELISA as described before (21).
Binding of 6B4 WT and Mutants to Human Platelets-The capacity of WT and mutant 6B4 to bind to platelets was tested in an ELISA system where the antibody was added, in a dilution series of 1:2, into wells precoated with fixed intact human platelets (17). Revelation was done by incubating with a monoclonal antihuman-IgG 4 antibody (1:4000), followed by a goat anti-mouse horseradish peroxidase-labeled Ab (1:5000), before the addition of H 2 O 2 and orthophenylenediamine, stop with H 2 SO 4 , and optical density determination (490 -630 nm) on a microplate reader. Binding of 6B4 WT at saturation was set as 100%.
Binding of rGPIb␣ WT and Mutants to Monoclonal Anti-GPIb␣ Antibodies-Recombinant GPIb␣ WT and mutants were tested for FIGURE 1. HADDOCK computer docking model of 6B4 to GPIb␣. A, ribbon representation of the HADDOCK model number 21 complexing GPIb␣ (green) and its ␤-switch region in the "ligand-free" conformation (red) with 6B4 light chain (magenta) and 6B4 heavy chain (blue). Amino acid side chains predicted to form hydrogen or salt bridges are represented as sticks. B, focused ribbon representation showing important residues constituting the 6B4 GPIb␣ interface. Hydrogen bonds and salt bridges are represented as dashed lines. Y100C is not shown, since it does not form hydrogen bonds or salt bridges as mentioned in Table 3. Amino acid residues that later were confirmed to be part of the paratope are lettered in red. All images of the three-dimensional models were generated with PyMOL (available on the World Wide Web).

TABLE 3 Hydrogen bonds and salt bridges between 6B4 and GPIb␣ in model 21
Heavy chain Y100C is not in the Paratope Identification of the Anti-GPIb␣ Antibody 6B4 AUGUST 10, 2007 • VOLUME 282 • NUMBER 32 their capacity to bind to coated monoclonal 6B4, 27A10, and 24G10 in an ELISA set up as described previously (21). Statistical Analysis-The binding capacity of 6B4 and its mutants to GPIb␣ as well as the binding of GPIb␣ and its mutants to monoclonal anti GPIb␣ antibodies were compared by Student's t test. The differences were considered statistically significant when p was Ͻ0.05.

RESULTS
First Model Using ZDOCK-The 6B4 paratope was tentatively determined by constructing computer models of the variable regions of 6B4 bound to different crystal structures of GPIb␣. In a first approach, a 6B4 computer model (17) was docked to the ligand-free conformation of GPIb␣ (1M0Z.pdb) using the ZDOCK program (12). The resulting model was characterized by one major binding site in which the mAb binds to the flexible loop of GPIb␣. The model suggested that both the light and heavy chain of 6B4 contribute to the binding to GPIb␣.
Based on this model, seven 6B4 residues were selected for mutation to Ala: four on the light chain (Y27D, K27E, Val 92 , and Glu 93 ) and three on the heavy chain (Ser 56 , Asn 58 , and Ile 100 ). Mutations of Val 92 and Ser 56 to Ala were included as negative control, since no major effect was expected. Only Y27DA and E93A were found to affect the binding of 6B4 to GPIb␣ (Fig. 2,  A and B).
Second Model Using HADDOCK-Since the rigid body docking method ZDOCK only allow the prediction of two interacting residues, we, in a second approach, used the results from the first round as input to construct a new docking model using HADDOCK1.3 that allows for flexible docking in both 6B4 and GPIb␣. We used the four available structures of the mAb plus the ligand-free and ligand-bound structures of GPIb␣ simultaneously in the docking experiment. The docking of 6B4 to the ligand-bound conformation of GPIb␣ (1SQ0.pdb) did not produce any acceptable model, since hardly any hydrogen bridges between the two proteins were found that furthermore mainly occurred between main chain elements (data not shown).
The docking results with the ligand-free GPIb␣ structure were grouped into clusters, which are defined as an ensemble of at least two conformations displaying a backbone root mean square deviation at the interface smaller than 1.0 Å (14). Of the energetically best models in each of the three clusters thus obtained (Table 2), docking model 21 in cluster 1 was selected, because both Y27D and Glu 93 are predicted to bind GPIb␣, which is in agreement with our previous experimental data (Fig.  1). Based on this docking model, light chain K27E, Asp 28 , and Tyr 94 and heavy chain Ser 97 and Y100C were selected for mutagenesis (Table 1).
In this model, Asp 28 forms four interactions, two of which are salt bridges, with the side chains of Arg 64 and His 86 located in the GPIb␣ LRR2 and LRR3, respectively. To disrupt these interactions, Asp 28 was not only mutated to Ala but also to the positively charged Arg (Table 1). Furthermore, also in this model (and in most generated models) K27E is predicted to play an important role in binding, since it forms two ionic interactions with Asp 83 and Asp 106 of GPIb␣ (Table 3, lines 3 and 5). In the first round, however, somewhat unexpectedly, mutation of K27E to Ala did not inhibit the binding of 6B4 to GPIb␣ ( Fig.  2A), so also here we made a second mutant in which the negatively charged Asp is introduced instead. All 6B4 mutants, except for Y94A, which only was expressed in very low quantities, were tested for their capacity to bind to immobilized human platelets ( Fig. 2A). Of the five new mutants tested, only S97T still bound normally to GPIb␣ on platelets. Fig. 2B clearly shows that six mutants (Y27DA and E93A from round 1 and Shown is a surface representation of the variable domains of 6B4 (C) with the residues that are critical for binding (red), the residues where a mutation did not affect the binding (green), and Tyr 94 (blue), which was suggested but which could not be tested because of low expression levels. The image of the three-dimensional model was generated with PyMOL (available on the World Wide Web).
K27EE, D28A, D28R, and Y100CA from round 2) at 0.25 and 0.5 g/ml maximally allowed 25% binding of 6B4 to GPIb␣, whereas the other mutants (K27EA, V92A, S56A, N58A, and I100A from round 1 and S97T from round 2) show nearly normal binding. Based on the mutagenesis data, we can conclude that residues Tyr 27D , Lys 27E , and Asp 28 of CDR L1, Glu 93 of CDR L3, and Tyr 100C of CDR H3 are part of the paratope of 6B4 (Fig. 2C). The three other CDRs do not seem to be involved in the binding.
Epitope Determination on GPIb␣-Based on the docking model of 6B4 bound to GPIb␣, a number of residues were predicted to be part of the epitope. All residues present at the antigen surface and forming hydrogen bonds or ionic interactions with residues of 6B4 could be putative residues of the epitope ( Table 3). Several of these residues are located in the LRR: Arg 64 , Asp 83 , His 86 , and Asp 106 . Some other residues involved (Lys 231 , Asp 235 , and Lys 237 ) are part of the ␤-switch region, the key element that changes its conformation during ligand binding (5).
To validate our model, a set of 38 single to triple GPIb␣ mutants (21) containing 62 charged residues mutated to Ala (Fig. 3) was tested for the binding to wild type 6B4. As an additional control, two other inhibitory anti-GPIb␣ mAbs were tested, namely 24G10, which competes with 6B4 for the binding to human platelets (10), and 27A10 (22), which does not compete (Fig. 4A).
A marked impairment of the binding of mAb 6B4 was seen to GPIb␣ mutants D83A/H86A (Fig. 4A, lines 3 and 4) 6 and 7), which contain all of the predicted residues except for Arg 64 (line 2) and Lys 231 (line 10), mutation of which did not affect binding. However, since the binding of all three antibodies (and others) 5 to the first four mutants was similarly decreased, it is possible that these induce a conformational change in GPIb␣, as previously hypothesized (21) and hence might have an indirect effect on the antibody binding. These residues, therefore, are being treated with caution in the description of the epitope, which on the other hand clearly involves Asp 235 and Lys 237 of the ␤-switch region (Fig. 4B).
Refined Docking Model-All of the results from the mutagenesis experiments on both 6B4 and GPIb␣ were combined and used for another docking round with HADDOCK1.3, using again the four 6B4 models and the ligand-free and ligandbound GPIb␣ conformation. Results were ranked by interface area and energy contribution to the complex formation and visually analyzed. Also, here the only acceptable docking models could be made with the ligand-free GPIb␣ conformation, of which the best model, number 61 ( Fig. 5 and supplemental material), is very close to the previous model 21. In model 61, 6B4 is rotated and translated into the N-terminal direction, resulting in an increased interaction area from 917.2 to 1164.8 Å 2 . On the GPIb␣ side, only the ␤-switch region has a different conformation, leading to a difference in root mean square deviation of 3.4 Å between the mAbs and the ␤-switch region. The root mean square difference between the two docking models, 21 and 61, is 8.2 Å when we consider the CDRs but only 7.7 Å when we restrain to the residues that are common in the interface.
A close comparison of the two docking models reveals that some of the amino acids of 6B4 that were involved in the interaction area of the docking model 21 no longer are (i.e. L27C). On the other hand, new residues are now situated in the contact surface (i.e. Met 51 and Phe 71 ). In model 61, CDR L2 with residues Met 51 , Ser 52 , Thr 53 , and Arg 54 is now part of the interacting surface from which it was absent in model 21. In both docking models, the light chain has a dominant role in the binding to GPIb␣, whereas CDR H1 is not contributing. The interaction of GPIb␣ in model 61 does not differ from the one in model 21; the N-terminal flanking region, the five first LRRs, and the ␤-switch region are all involved in the binding to 6B4.

DISCUSSION
We have developed a monoclonal antibody, 6B4, targeting the human GPIb␣, which is responsible for the binding of platelets to the exposed collagen in damaged vessels via von Willebrand factor under high shear stresses. After promising results in animal models of arterial thrombosis, we have developed a 5 M. Yamashita and T. Matsushita, unpublished results.  Red, involved in 6B4 binding; green, not involved; blue, inconclusive due to likely misfolding of the mutant; orange, gain-of-function mutations that disrupt 6B4 binding. A, binding of monoclonal antibodies to mutant GPIb␣. After expression and purification, each GPIb␣ was tested for the binding to coated mAb 6B4, 24G10, or 27A10 as described under "Experimental Procedures." Each bar represents the mean with S.E. value obtained for at least two independent duplicate assays. *, statistically different, with p Ͻ 0.05. Results with line numbers in boldface type are discussed under "Results," and 1-7 corresponds to the numbering in Table 3. B, surface representation of the N-terminal part of GPIb␣ with predicted residues involved or not in 6B4 binding. Gly 233 and Met 239 , for which the gain-of-function mutation to Val disrupts the binding of 6B4, are in orange. The images of the three-dimensional models with the surface representation were generated with PyMOL (available on the Word Wide Web). recombinant and humanized Fab fragment of 6B4. Importantly, at effective antithrombotic doses, 6B4-Fab does not prolong the bleeding time; nor does it induce thrombocytopenia (9,11). To take full advantage of the in vivo effects of blocking GPIb␣, a compound fit for oral administration is a prerequisite for a broad and prophylactic use. As a first step toward that goal, we here identified the paratope of 6B4 that confers the inhibitory properties of the molecule. In this study, we mapped both the paratope and the epitope of 6B4 by combining computer docking models with mutagenesis studies on both 6B4 and GPIb␣.
Our first ZDOCK-based docking approach using 6B4 and GPIb␣ in its ligand-free conformation (1M0Z.pdb) identified only light chain residues Y27D and Glu 93 , of the seven selected residues, to be critical for the binding. This relatively poor result might be due to the fact that ZDOCK is an algorithm developed for rigid body docking. GPIb␣, however, is a molecule with several known conformations, and also antibodies rearrange their residue side and/or main chains to improve the affinity for their targets. One way to overcome this problem would be to perform molecular dynamics simulations followed by calculation of relative free binding energies with the molecular mechanics Poisson-Boltzmann surface area as, for example, Wu et al. (23) did to resolve the interaction between the scorpion toxin ScyTx and the small conductance calcium-activated potassium channel Rsk2. An alternative strategy, which we followed, is using a docking strategy method that allows flexibility in both ligand (6B4) and target (GPIb␣), such as HADDOCK1.3. Docking tasks were submitted, including both the ligand-free and ligand-bound conformation of GPIb␣ in combination with 6B4 models constructed with the Web Antibody Modeling program, next to the model that we previously used to prepare a humanized 6B4-Fab fragment (17). After ranking the hits and visual inspection, no good candidate models were identified with the ligand-bound conformation of GPIb␣. This finding is in total agreement with our previous result, where the binding of 6B4 was some 6-fold lower to the gain of function (G233V and M239V) GPIb␣, as compared with the wild type (10). These gains of function are found in platelettype von Willebrand disease and enhance the affinity of GPIb␣ for VWF (6,24). Furthermore, the structure of GPIb␣ carrying either one of these mutations is similar to its conformation in complex with the VWF A1 domain (19).
Next, 6B4 residues involved in the binding to ligand-free GPIb␣, as deduced from the docking model 21, were expressed as single mutants. The chimeric human/mouse 6B4-IgG 4 was chosen because it has the same characteristics (17) as the parental IgG and is easy to manipulate and to produce. Indeed, production of the mutants in quantities sufficient for the binding studies was possible for all 13 mutants except for Y94A. Binding experiments of 6B4 WT and its mutants were performed on whole fixed platelets of healthy volunteers, thereby presenting GPIb␣ in the GPIb⅐IX⅐V complex. All together we positively identified three paratope residues in CDR L1 and one in CDR L3 and CDR H3 each, which all together are in close spatial proximity on the antibody surface.
To further validate docking model 21, we next explored the role of every charged residue in GPIb␣ by using the Ala scan library previously described (21). Mutation of most of the predicted interacting residues caused deficient binding. However, since a number of these mutations are suspected to induce a conformational change in GPIb␣ (21), which also here resulted in a decreased binding of two other anti-GPIb␣ antibodies with different epitopes (10), we cannot make a definitive statement on these, in contrast to residues Asp 235 and Lys 237 located in the ␤-switch region, which specifically affected binding of 6B4. These results are furthermore in perfect agreement with the study of Cauwenberghs et al. (10), where we used human/canine chimeric GPIb␣ to map the epitope.
Finally, we performed a new docking experiment, including all known experimental data, which yielded a final model in total agreement with all of the mutagenesis results.
In conclusion, by identifying the crucial residues involved in the paratope-epitope interaction of 6B4 with GPIb␣, a detailed model of the complex at the atomic level is proposed. This information allows for a better understanding of the antithrombotic action of 6B4 and will be helpful in the design of improved molecules.
Furthermore, and in a broader perspective, the iterative method we used here to identify the interacting amino acid residues by alternating docking and mutagenesis experiments allows for a more rational identification of relevant residues than what a classical laborious mutagenesis scan can provide and can be readily applied to other protein-protein interactions for which sufficient structural information is available.