Crystal Structure of von Willebrand Factor A1 Domain Complexed with Snake Venom, Bitiscetin

Bitiscetin, a platelet adhesion inducer isolated from venom of the snake Bitis arietans, activates the binding of the von Willebrand factor (VWF) A1 domain to glycoprotein Ib (GPIb) in vitro. This activation requires the formation of a bitiscetin-VWF A1 complex, suggesting an allosteric mechanism of action. Here, we report the crystal structure of bitiscetin-VWF A1 domain complex solved at 2.85 Å. In the complex structure, helix α5 of VWF A1 domain lies on a concave depression on bitiscetin, and binding sites are located at both ends of the depression. The binding sites correspond well with those proposed previously based on alanine-scanning mutagenesis (Matsui, T., Hamako, J., Matsushita, T., Nakayama, T., Fujimura, Y., and Titani, K. (2002) Biochemistry 41, 7939–7946). Against our expectations, the structure of the VWF A1 domain bound to bitiscetin does not differ significantly from the structure of the free A1 domain. These results are similar to the case of botrocetin, another snake-derived inducer of platelet aggregation, although the binding modes of botrocetin and bitiscetin are different. The modeled structure of the ternary bitiscetin-VWF A1-GPIb complex suggests that an electropositive surface of bitiscetin may interact with a favorably positioned anionic region of GPIb. These results suggest that snake venom proteins induce VWF A1-GPIbα binding by interacting with both proteins, and not by causing conformational changes in VWF A1.

Von Willebrand factor (VWF) 1 plays a key role in hemostatic plug formation at sites of vascular injury by interacting with subendothelial matrix proteins and platelet glycoprotein Ib (GPIb) (1). VWF exists as a disulfide-linked multimer composed of identical ϳ250-kDa subunits, each of which contains three adjacent A domains (A1, A2, and A3). The A1 and A3 domains have major binding sites for GPIb and collagens (types I and III), respectively, and the A2 domain contains the cleavage site for the plasma VWF-cleaving protease (ADAMTS13), which regulates the functional multimer size of VWF (2).
Binding between VWF A1 domain and GPIb␣ is evident only when blood is exposed to high shear stress in vivo (3,4). This binding activation mechanism is proposed to reflect structural changes in the A1 domain (4,5). A structural change in the Asp 560 -Gly 561 region was observed in the structure of the VWD type 2B "gain-of-function" A1 domain mutant, which binds spontaneously to GPIb under physiological conditions (6). Binding of VWF to platelets in vitro occurs in the presence of various modulators, such as the antibiotic ristocetin (7) or snake venom proteins bitiscetin (8) or botrocetin (9). Snakes employ bitiscetin and botrocetin to disrupt hemostasis and thereby to kill or weaken their prey; these modulators have been used extensively as tools to study the complicated mechanisms involved in hemostasis.
Bitiscetin and botrocetin are members of C-type lectin-like proteins (CLPs). The first determination of the complete amino acid sequence of CLPs was carried out using coagulation factors IX-and X-binding protein (IX/X-bp) from Trimeresurus flavoviridis (habu snake) venom (10). Up to the present time, various CLPs have been sequenced and characterized with a variety of activities that affect plasma proteins, platelets, endothelial cells, and subendothelial structures. Thus, CLPs with diverse activities appear to be derived from a common ancestor. In this connection, IX/X-bp and IX-bp from habu snake venom have diversified their amino acid sequences in an accelerating manner (11) as also observed for crotalinae snake venom gland phospholipases (12) and serine proteases (13). As shown for the digestive enzymes in venom, which vary predictably in response to differences in diet (14), CLPs have evolved possibly to gain functional diversity in response to important components in the hemostatic system of their prey.
CLPs are heterodimeric proteins consisting of homologous subunits A (␣) and B (␤) linked by a disulfide bond. Both subunits show a similarity to carbohydrate recognition domains of the classic C-type lectins. The crystal structure of IX/X-bp (15) shows that the two subunits are tightly associated by domain swapping, and this dimerization results in the creation of the concave surface predicted to function as a coagulation factor binding site. Examination of the crystal structure of the complex between X-bp and the ␥-carboxyglutamic acid (Gla) domain of factor X has revealed that the binding site is, in fact, the concave surface (16). The structures of CLPs with various functions have been solved by x-ray crystallography (15,(17)(18)(19)(20). CLPs have many structural similarities despite having distinct functions, suggesting that electrostatic interactions, as well as size and shape matching, are important for ligand binding.
The concave surfaces of many CLPs have been proposed to form the binding site for a variety of highly diverse ligands. In this connection, we previously showed that the electrostatic potentials on the concave surface are significantly different among IX/X-bp, bitiscetin, and botrocetin (20). Recently, Fukuda et al. (21) reported the crystal structure of a botrocetin-VWF A1 complex; they found no large structural change compared with the uncomplexed A1 domain and suggested that the activation of binding by botrocetin is distinct from that of high fluid shear stress. Although bitiscetin and botrocetin have similar effects on VWF function, they are expected to bind VWF A1 quite differently based on the results of alanine scanning studies (22,23) and also based on the differences in pI and in electrostatic potentials on the concave surface as mentioned above.
To better understand the mechanism by which venom modulators affect hemostasis, we solved the crystal structure of bitiscetin-VWF A1 domain complex. The structure refined at 2.85 Å resolution showed that bitiscetin binds to the A1 domain via its concave surface in a manner different from botrocetin. The bitiscetin contact sites on the A1 domain partially overlapped with those of botrocetin, but the direction of the longaxis of bitiscetin was almost perpendicular to that of botrocetin when the bound A1 domains were superimposed. As in the case of the botrocetin-VWF A1, our structure showed no significant change in the VWF A1 domain. Bitiscetin-and botrocetin-VWF A1 complexes and their electrostatic potential maps revealed a positively charged patch on each venom protein, which is located very close to the anionic region of GPIb␣. We propose that this positively charged patch of snake venom modulators plays a key role in the induction of GPIb-binding without inducing a significant conformational change in the GPIb-binding site on VWF.

EXPERIMENTAL PROCEDURES
Materials-Bitiscetin was purified from crude venom of Bitis arietans (purchased from ICN Biomedicals Inc.) as described previously (8). Recombinant VWF A1 domain (residues 497-708), expressed in Escherichia coli and purified as described (24), was further purified with MonoQ (Amersham Biosciences) and TSK-GEL G2000SW gel filtration column (Tosoh). The affinity of VWF A1 binding to immobilized bitiscetin was measured by a surface plasmon resonance assay using Biacore 3000 (Biacore AB), and a high affinity with a K d 2 nM was obtained. Bitiscetin and VWF A1 domain were mixed in a 1:1.3 molar ratio, placed on ice for 2-3 h, and then loaded on Superdex 200 column (Amersham Biosciences) equilibrated with sample buffer (10 mM Tris-Cl, pH 7.5, 0.1 M NaCl). Complex-containing fractions were concentrated by ultrafiltration (Microcon, Millipore) to 6 mg/ml. The concentration of protein complex was estimated by absorbance at 280 nm with a molar extinction coefficient of 96,930 liters/mol⅐cm and a molecular weight of 53,455 for the complex.
Crystallization and Data Collection-Needle-shaped crystals were obtained with a sitting-drop vapor diffusion technique by mixing 3.0 l of protein solution (6 mg/ml) and 2.5 l of reservoir solution. The reservoir solution consisted of 88% Crystal Screen II (Hampton Research) no. 30 solution (0.1 M Hepes-Na, pH 7.5, 10% polyethylene glycol 6000, 5% 2-methyl pentanediol) and 12% Crystal Screen II no. 46 solution (0.1 M Bicine-Na, pH 9.0, 20% polyethylene glycol-monomethyl ester 550). The crystals were typically grown to maximum size in 2 weeks at 285 K. A crystal (0.8 ϫ 0.15 ϫ 0.15 mm 3 ) was transferred with a nylon loop to cryoprotectant solution consisting of 83% reservoir solution and 17% glycerol and was then rapidly frozen by a cold nitrogen stream at 100 K. Diffraction data were collected at Photon Factory BL-6B (Tsukuba, Japan) with an R-AXIS IV ϩϩ imaging plate detector (Rigaku). The data were integrated by using MOSFLM (25) followed by scaling with SCALA (26) in the resolution range of 30 to 2.85 Å.
Diffraction rings from unknown salts were observed in all images at 3.23 Å (data not shown), therefore, the diffraction spots within the 3.30 -3.16 Å resolution range were excluded during the data processing. The crystals belong to tetragonal space group P4 1 or P4 3 with unit cell dimensions of a ϭ b ϭ 89.28 Å, c ϭ 53.39 Å. The V M value (27) was calculated to 1.99 Å 3 /Da assuming one bitiscetin-VWF A1 complex per asymmetric unit. The data collection statistics are summarized in Table  I.
Model Building and Refinement-The initial model was built by the molecular replacement method using bitiscetin (Protein Data Bank entry 1JWI) and wild type VWF A1 domain (Protein Data Bank entry 1AUQ) as search models. Molecular replacement calculations with AMoRe (28) were applied independently using each model in the resolution range of 10 to 4 Å. The highest solution from each model indicated that the space group of the crystal was P4 3 , with correlation coefficients of 0.353 (VWF A1) and 0.184 (bitiscetin) and R-values of 0.482 (VWF A1) and 0.530 (bitiscetin). Applying the corresponding transformation matrices, a proper bitiscetin-VWF A1 complex model could be easily built. Molecular replacement calculations using the complex gave a correlation coefficient and R -value of 0.693 and 0.331, respectively. Subsequent rigid-body refinement, simulated annealing, and individual B-factor refinement were done with the program CNS (29). After several cycles of refinement, the R-factor and R-free (12% of the reflections) decreased to 19.4 and 27.6%, respectively. The final model includes residues 3-127 (␣-subunit) and 3-125 (␤-subunit) of bitiscetin, 501-702 of VWF A1 domain, and 40 water molecules. The refinement statistics are summarized in Table I.

RESULTS AND DISCUSSION
Overall Structure of Bitiscetin-VWF A1 Complex-Bitiscetin, as well as other CLPs, has a heterodimeric structure, and VWF A1 domain binds to the concave surface of bitiscetin (Fig. 1a). The concave surface often functions as a ligand-binding site as shown previously for the structure of X-bp complexed with the factor X ␥-carboxyglutamic acid domain (16). Helix ␣5 of the A1 domain is positioned near the center of the concavity, near the junction between the bitiscetin ␣ and ␤ subunit, but interacts mainly with the ␤-subunit, leaving a solvent-filled space between discontinuous binding sites at both ends of the concave surface of bitiscetin. Helix ␣4 of the A1 domain contacts the ␣-subunit of bitiscetin, burying a solvent accessible surface of ϳ177 Å 2 , and sheet ␤6 and helices ␣5 and ␣6 contact the ␤-subunit, burying a solvent accessible surface of ϳ431 Å 2 (Fig. 1a).
A comparison with the structure of the botrocetin-VWF A1 complex shows that the botrocetin-and bitiscetin-binding sites overlap at A1 helices ␣4 and ␣5. The direction of the long-axis of bitiscetin is almost perpendicular to that of botrocetin when the bound A1 domains are superimposed. Although bitiscetin (516 Å 2 ) and botrocetin (520 Å 2 ) make contact with A1 domains, bitiscetin binds more tightly than botrocetin, with K d values of 2 and 12 nM (24), respectively. This may be due to a number of solvent-mediated interactions and comfortable fitting of the A1 domain to the concave surface of bitiscetin compared with the case of botrocetin.
Interaction of Bitiscetin and VWF A1 Domain-At the contact region on the ␣-subunit (contact 1), direct hydrogen bond interactions are observed between Tyr 64 of bitiscetin and Arg 632 of the VWF A1 domain (hereafter, the amino acid residues of bitiscetin and VWF A1 will be listed in the same order) and between the main chain carboxyl of His 106 and Gln 639 , and indirect water-mediated interactions are observed between Asp 65 and Gln 628 and between the O atom of Val 63 and Asp 119 (Fig. 1b). In the contact region on the ␤-subunit (contact 2), hydrophobic and hydrogen bond interactions are found. A hydrophobic patch on the ␤-subunit composed of Leu 58 , Val 64 , Leu 65 , Phe 104 , and Ile 109 contacts a hydrophobic surface of the A1 domain that is composed of Pro 655 , Leu 659 , Ile 662 , and Val 676 . Adjacent to the hydrophobic region, hydrogen bonded interactions are found between Trp 111 and Lys 117 in the ␤-subunit and Glu 666 in A1, between Gln 110 and Arg 663 , between Glu 22 and Lys 673 , between Lys 20 and both Gln 686 and Glu 689 , and between Arg 115 and the backbone oxygens of Glu 666 , Lys 667 , and Ala 669 (Fig. 1c). In addition, a hydrogen bond between Glu 96 in the bitiscetin ␣-subunit and Lys 660 in the A1 domain is observed.
Comparison of the Bitiscetin-VWF A1 Structure with Mutagenesis Results-Recently, Matsui et al. (23) reported the localization of bitiscetin-binding sites on the VWF A1 domain by alanine-scanning mutagenesis. According to this report, three A1 domain mutations on helix ␣5 (K660A, E666A, and K673A) and one mutation on helix ␣4 (R632A) significantly reduced binding to bitiscetin. These results are in good agreement with our structural data, because all of these residues are involved in hydrogen bond interactions with bitiscetin ( Fig. 1, b  and c). Interestingly, the A1 mutant R663A showed 1.5-fold higher binding activity relative to the wild type A1 domain (23). The side chain of Arg 663 interacts with Gln 110 in the ␤-subunit of bitiscetin, one of the major interactions between the A1 domain and bitiscetin (Fig. 1c), and mutation of Arg 663 might be expected to reduce the affinity of binding. This contrary result may be explained as follows. The orientation of Arg 663 in bitiscetin-free A1 structure is different from that in the bitiscetin-complexed structure; when A1 domains are superimposed, the Arg 663 in unbound A1 collides with Trp 111 of the bitiscetin ␤-subunit (Fig. 1c, inset). Moreover, Arg 663 makes intramolecular hydrogen bonds with Glu 666 in the free A1 structure that must be broken when bitiscetin binds. Therefore, the mutation R663A may promote binding to bitiscetin by avoiding the need to rearrange the hydrogen bonds and orientation of Arg 663 , which may be costly in terms of energy.
Comparison with Other von Willebrand Factor A1 Domain Structures-Several studies suggest that structural changes in the A1 domain can increase the affinity of binding to GPIb (4,5). For example, VWD type 2B mutations in the VWF A1 domain, such as I546V, promote binding to GPIb independent of fluid shear stress (4). The crystal structure of the I546V mutant A1 domain revealed a small structural change in residues Asp 560 -Gly 561 at the edge of strand ␤3 (6,21), and this change was proposed to cause the increase in affinity for GPIb (6). In the crystal structure of GPIb␣-A1 complex, hydrogen bonding interaction are observed between the main chain carboxyl of Asp 560 and side chain O␥ of Thr 240 in GPIb␣ (31), suggesting that the conformation of Asp 560 -Gly 561 could change upon GPIb␣ binding. In the bitiscetin-complexed A1 structure, however, the Asp 560 -Gly 561 region has the same conformation as in wild type uncomplexed A1 (Fig. 2b). Fur-thermore, when the wild type or I546V A1 structures were superimposed onto the bitiscetin-complexed A1 structure, no significant structural changes were found at either the bitiscetin-or GPIb␣-binding sites (Fig. 2a). Large structural changes were observed only in the N-terminal and C-terminal regions of A1 domain that flank the Cys 509 -Cys 695 disulfide bond. These regions are expected to be changeable because of their location outside of the disulfide loop, and indeed their flexibility is evident from the crystallographic B-factors in these regions.
Predicted Model of the Bitsicetin-VWF A1-GPIb␣ Ternary Complex-The crystal structure of the VWF A1 (R543Q)-GPIb␣ (M239V) complex reveals two binding sites on the A1 domain for GPIb: a major site at the ''top'' of A1 involves the ␤3 strand, ␣3 helix, and part of the ␣3-␤4 loop; a minor site at the "bottom" of the domain involves loops ␣3-␤4, ␤3-␣2, and ␣1-␤2. In the structures of A1 complexed with bitiscetin or botrocetin, however, no significant changes are observed in these regions. Moreover, in the structure of the botrocetin-A1 (I546V) complex, the main chain carboxyl of Asp 560 has the same conformation as the uncomplexed wild type A1 domain, despite the use of a gain-of-function A1 mutant (21). These observations suggest that bitiscetin and botrocetin stimulate the binding of A1 to GPIb by a mechanism that does not require allosteric conformational changes in VWF A1.
The modeled structure of the bitiscetin-VWF A1-GPIb␣ ternary complex (Fig. 3) indicates that the binding sites of bitiscetin and GPIb␣ on the A1 domain do not overlap, so that GPIb␣ could bind easily to the bitiscetin-VWF A1 complex. In this model, the ␤-subunit of bitiscetin is well separated from GPIb␣, whereas the ␣-subunit of bitiscetin and the C terminus of the GPIb␣ fragment are approximated. This model is consistent with the observation that monoclonal antibodies to the ␣-subunit inhibit GPIb binding to the bitiscetin-A1 complex, but antibodies to the ␤-subunit do not (23).
Implications for the Mechanism of GPIb-A1 Domain Binding Step 3 (bottom), the anionic region acts as an anchor, which enhances the GPIb␣ to bind to A1 domain.
Induced by Venom Proteins-The electrostatic potential of bitiscetin-A1 complex shows that there is a positively charged patch close to the C terminus of GPIb fragment in the modeled ternary complex (Fig. 4, left). The positively charged patch of bitiscetin is constructed from ␣-subunit residues Lys 17 , Lys 20 , Lys 21 , Lys 61 , and Lys 120 . A similar positively charged patch is also found in the botrocetin-A1 complex (Fig. 4, right), which consists of ␣-subunit residues Arg 98 and Lys 101 and ␤-subunit residues Lys 102 , Lys 107 , Trp 109 , Arg 115 , and Lys 117 (botrocetin residues are numbered according to Ref. 19). We propose that these positively charged patches observed in both venom proteins are the key to the activation of GPIb␣ binding to the A1 domain.
At the C terminus of the GPIb␣ fragment, there is an anionic region (residues 269 -279) characterized by three sulfated tyrosine residues, Tyr 276 , Tyr 278 , and Tyr 279 (32). The removal of these sulfate moieties, whether by mutation or by inhibition of sulfation, severely impairs the ability of botrocetin to induce VWF-dependent platelet aggregation (33,34) indicating that the anionic region of GPIb plays an important role in the high affinity binding of botrocetin-A1 complexes. The anionic region of GPIb␣ was disordered in the structure of the binary VWF A1-GPIb␣ complex (31). In the crystal structure of uncomplexed GPIb␣, there were two molecules in the asymmetric unit; in one molecule the anionic region was observed but in the other it was disordered (35). These observations suggest that the anionic region of GPIb␣ is flexible but could adopt a specific conformation upon binding to a nearby positively charged site on bitiscetin or on botrocetin complexed with VWF A1.
Based on their GPIb␣-VWF A1 structure, Huizinga et al. (31) proposed that the N and C termini of the A1 domain, which are located at the bottom face, shield the binding site from the ␤-finger of GPIb␣. Consequently, structural changes in domain A1 would be necessary for GPIb␣ binding, and such a requirement for conformational change is consistent with many biochemical and structural studies. However, the crystal structures of the botrocetin-A1 complex (21) and the bitiscetin-A1 complex (Fig. 1) show no significant structural changes in the GPIb␣-binding sites of the A1 domain compared with uncomplexed A1. Therefore, the possibility that botrocetin or bitiscetin allosterically activates the A1 domain seems unlikely. We propose that bitiscetin-or botrocetin-induced binding is not dependent on major structural changes of the A1 domain but instead depends mainly on electrostatic interactions between the anionic region of GPIb and a complementary electropositive site contributed by the venom protein in the VWF A1-botrocetin or VWF A1-bitiscetin complex (Fig. 5).
Some experimental data indirectly support this model for botrocetin-induced binding of VWF to GPIb, but the effects of removing sulfated tyrosines or deleting the anionic region of GPIb␣ have not been studied yet with bitiscetin. For either of these venom proteins, a definitive test of this activation model will require additional biochemical and crystallographic studies of the corresponding ternary complex.