The von Willebrand Factor A3 Domain Does Not Contain a Metal Ion-dependent Adhesion Site Motif*

von Willebrand factor (vWF) is a multimeric plasma protein that mediates platelet adhesion to exposed subendothelium at sites of vascular injury. The A3 domain of vWF (vWF-A3) forms the principal binding site for collagens type I and III. We report here the crystal structure of the vWF-A3 domain at 2.2-Å resolution. As expected, the structure is similar to the integrin I domain but with several novel features. Sequence alignments had suggested that the domain contained an integrin metal ion-dependent adhesion site (MIDAS) motif, but the crystal structure shows that the motif is modified and that no metal ion is bound. We have introduced mutations into the vestigial MIDAS motif and report that, unlike the I domain of integrin α2β1, vWF-A3 continues to bind collagen after disruption of the motif. We conclude that collagen recognition by vWF-A3 occurs by a mechanism different from that of the integrin α2β1.

von Willebrand factor (vWF) 1 is a multimeric plasma protein that mediates platelet adhesion to exposed subendothelium at sites of vascular injury and also transports the procoagulant factor VIII (1). Abnormalities in vWF lead to von Willebrand disease, the most common congenital bleeding disorder. Clinical bleeding occurs when the plasma vWF level falls below the normal level of 5-10 g/ml (type I disease) or when dysfunctional molecules are produced (type II disease) (2).
Initial platelet adhesion is mediated by sequences within the A1 and A3 domains of multimeric vWF. The A1 domain (residues 479 -717 of mature vWF) binds to platelet glycoprotein Ib/IX, heparin, and cell surface sulfatides. Although both the A1 and A3 domains can bind to collagen, the A3 domain contains the principal collagen-binding site (8). The adhesive properties of vWF are tightly regulated so that plasma vWF does not normally interact with circulating platelets. vWF, however, binds to subendothelial collagen that has been exposed following vascular injury, leading to a conformational change in vWF that allows the A1 domain to bind to platelet glycoprotein Ib/IX. A large number of proteins contain vWF A-type domains, including the ␣-subunits of leukocyte integrins (9), where the domains are generally called I domains. Crystal structures have previously been reported for the I domains of integrins ␣M␤2 and ␣L␤2 (10,11). The I domain from the integrin ␣2␤1 is its principal collagen recognition domain (12), and its crystal structure has also been solved recently. 2 In all cases, the integrin I-domain adopts the classic "dinucleotide binding" fold with a central hydrophobic parallel ␤-sheet, flanked on two sides by amphipathic helices (10). The domain also contains a metal-binding site at the top of the ␤-sheet which is critical for its adhesive function (14). It has been called the MIDAS (Metal Ion Dependent Adhesion Site) motif (10), and recent mutagenesis studies support the proposal that the upper surface of the I domain forms the integrin-ligand interface (15,16).
We report here the crystal structure of the vWF-A3 domain, which we determined as a first step in understanding the molecular basis of vWF-ligand interactions. Since the integrin ␣2-I domain also binds collagen, it had been suggested that the two domains would have similar binding motifs. However, our results show that the vWF-A3 domain, unlike the ␣2-I domain, does not bind metal and that the collagen binding properties of vWF-A3 persist even after the disruption of its MIDAS-like motif by site-specific mutagenesis.

EXPERIMENTAL PROCEDURES
Crystal Structure Determination-Recombinant vWF-A3 containing residues 908 -1111 of mature vWF was expressed in Escherichia coli using the pQE9 expression vector and purified as described previously (8). Crystals were grown by hanging drop vapor diffusion from 24 to 32% PEG 8000, 100 mM Tris, pH 8.5, and 200 mM MgCl 2 at 22°C. Crystals belong to space group P2 1 with cell dimensions a ϭ 42.5 Å, b ϭ 66.8 Å, c ϭ 57.8 Å, and ␤ ϭ 101.9°. The asymmetric unit contains two * This work was supported by Grant G9607237MB from the Medical Research Council (UK) (to R. L.) and by Grant R01 HL54876 from the National Institutes of Health (to R. H.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The atomic coordinates and structure factors (accession codes 1ao3 and r1ao3sf) have been deposited in the Protein Data Bank, Brookhaven National Laboratory, Upton, NY.
§ Performed much of this work while at the Dana-Farber Cancer Institute, Boston.
vWF-A3 domains and has a low solvent content (34%). Magnesium is not required for crystal growth but affects the growth rate. Crystallization from 0, 10, 100, and 200 mM MgCl 2 took 2 months, 1 month, 7 days, and 2 days, respectively, to produce crystals suitable for diffraction. A second crystal form was also observed under the same conditions in space group P2 1 with unit cell parameters a ϭ 58.0 Å, b ϭ 65.1 Å, c ϭ 84.5 Å, and ␤ ϭ 101.8°but has not been further studied. A mercury derivative was obtained by soaking a crystal in 1 mM CH 3 HgNO 3 for 4 h, and a gadolinium derivative was obtained by soaking a crystal in 10 mM Gd(NO 3 ) 3 for 3 days. Data from native and derivative crystals were collected on a MarResearch Imaging Plate and a Rigaku rotating anode generator with focusing mirrors at room temperature. Data were reduced with DENZO and scaled with SCALEPACK (17). The positions of six gadolinium atoms and two mercury atoms were readily determined by inspection of difference Pattersons and Fouriers using phases calculated by HEAVY (18). Further heavy atom refinement and phasing was performed using the CCP4 suite (19). Phases were refined using MLPHARE with the anomalous signal included. An electron density map calculated at 3-Å resolution could be partly interpreted by manually steering a model of an integrin A domain into the electron density. The first round of refinement used a polyalanine model of the central ␤-sheet and helices 1-5. This partial model was refined using XPLOR (20), and the model phases were recombined with MIR phases using SIGMAA to produce a new map from which the loops, helix 6 and side chains were constructed. The R factor at this stage was 31%. After additional rounds of refinement, model building, and extension of the data to 2.2-Å resolution, weak density emerged for the C-terminal helix, ␣8 (residues 178 -197), and the disulfide bridge. A round of non-crystallographic symmetry averaging was performed to resolve the ambiguities in this region. The final R factor was 16% for all data between 8 and 2.2 Å (R free ϭ 21.5%) with restrained individual atomic B factors. The final model contains two molecules of vWFA3 and 169 water molecules. The geometry is excellent with r.m.s. difference from ideal bond lengths of 0.010 Å and 1.58°for bond angles. All main chain torsion angles lie within most favored (92.5% of residues) or additional allowed regions (7.5% of residues) of the Ramachandran plot as defined in PROCHECK.
Mutagenesis-Mutations were introduced into vWF-A3 cDNA using two rounds of polymerase chain reaction amplification. First, each reverse oligonucleotide primer was combined with the 5Ј-end primer and each forward primer was combined with the 3Ј-end primer and amplified. Second, the resultant DNA fragment was then combined with the opposite end primer to produce the cDNA fragment. The outside primers introduced BamHI and HindIII restriction sites for cloning. The mutant cDNA fragments were isolated by digestion with
Collagen Binding Assay-A final concentration of 2 mg/ml homogenized insoluble bovine type I collagen (Sigma) or 1% bovine serum albumin was added to microtiter wells in 65 mM sodium phosphate buffer, pH 7.2, for 90 min at 37°C. After washing three times with Tris-buffered saline, pH 7.4, to remove unadsorbed collagen, residual binding sites were blocked by the addition of 3% bovine serum albumin, 0.1% Tween 20 in Tris-buffered saline for 60 min. Concentrations varying from 1 to 4 M of either wild type or each of the vWF-A3 mutant proteins was added to the wells and incubated for 60 min at room temperature. Wells were then washed with Tris-buffered saline and the remaining bound vWF-A3 detected with an enzyme-linked immunosorbent assay. Wells were incubated with a 1:2000 dilution of monoclonal anti-His antibody (Qiagen, La Jolla, CA) for 1 h at 37°C. After washing three times, a secondary anti-mouse antibody that was conjugated with horseradish peroxidase was added for 1 h and bound antibody visualized by the addition of o-phenylenediamine. Nonspecific binding was determined by subtracting peroxidase-conjugated antibody bound to wells coated only with bovine serum albumin and subtracted from total binding.

Structure of the vWF-A3 Domain-
We crystallized a recombinant vWF-A3 domain and solved its structure at 2.2-Å resolution using two heavy atom derivatives (Table I). The final structure contains two molecules of vWF-A3 and 169 water molecules. All of the main chain is well defined, with the exception of residues at the N and C termini (923-925 and 1098 -1109 (␣8)) in both molecules, for which the electron density is weak. The disulfide bridge linking the N and C termini (Cys 923 -C 1109 ) is visible and delimits the ordered residues. The overall fold is, as expected, very similar to the integrin I domain, with a central hydrophobic parallel ␤-sheet flanked on two sides by amphipathic helices (Figs. 1 and 2). The domain is globular with a height of 40 Å and a width (across the ␤-sheet) of 30 Å. The main chain atoms of the central ␤-sheet overlap with the integrin ␣M I domain with a r.m.s. deviation of 0.5 Å. There are several surprising features of the structure that could not have been readily predicted from homology modeling using the integrin ␣M and ␣L I domains. First, the ␣2-helix is replaced by a loop and helix ␣3 is extended by 5 residues. Second, at the top of the domain, the ␤E-␣6 loop adopts a different conformation. Third, the ␣7-helix is only two turns long and is followed by an abrupt 90°turn before forming a new 8th helix, ending with the C-terminal cysteine Cys 1109 .
The Vestigial MIDAS Motif-The MIDAS motif in integrin I domains consists of three closely apposed loops which together form the metal-binding site. In vWF-A3, these three loops have a very similar conformation, and two of the loops have identical or analogous residues as follows: the ␤A-␣1 loop contains the DXSXS consensus sequence (residues Asp 934 -Ser 938 ), and the ␣3-␣4 loop contains a serine (Ser 1005 ) in place of the threonine in the integrin I domains. However, in the third loop, ␤D-␣5, the position of the MIDAS aspartic acid (Asp 242 in ␣M␤2) is taken by Thr 1038 . Sequence comparisons had suggested that the next residue, Asp 1039 , might be analogous to the MIDAS aspartic acid, but the crystal structure shows that the sequence alignment was incorrect.
Although crystallization occurred in the presence of 200 mM Mg 2ϩ , no metal is bound to the vestigial MIDAS motif (Fig. 3). There is a buried water molecule that hydrogen-bonds to the side chains of Asp 934 , Ser 1005 , and Tyr 972 ; it also hydrogenbonds to the backbone nitrogens of Asp 1039 and Thr 1038 , which would not be possible for a metal ion. The water molecule corresponds most closely in position to one of the metal-coordinating water molecules in the integrin I domain, not to the metal ion itself.
At the top of the ␤A-␣1 loop, a phenylalanine side chain (Phe 939 ) points out into solution, creating a pocket occupied by three water molecules (in the integrin I domains, the analogous residue is buried in the hydrophobic core). The buried MIDAS aspartate (Asp 934 ) lies at the base of this pocket. This waterfilled cavity is sealed at the other end by the side chain of Arg 1000 ; although 8 Å distant, the positive charge, together with the solvent channel, may help to stabilize the buried aspartate.
Dimer Contact-The two molecules in the asymmetric unit of the crystal form a dimer related by a 2-fold axis of symmetry. The dimer interface comprises parts of the ␤C-␣3 turn, and the ␣4-␤D turn is chiefly hydrophobic and buries 1220 Å 2 of surface area, more than twice the area of the next largest contact. The two molecules can be superimposed very closely; the overall r.m.s. difference for backbone atoms is 0.65 Å, 0.17 Å for the central ␤-sheet (Fig. 1b). Curiously, there is a breakdown of symmetry at the dimer interface where the ␤C-␣3 turn adopts a quite different conformation in the two molecules. In one conformation, Asn 983 and Val 985 are involved in dimer contact, whereas Val 984 packs against the backbone of the loop. In the second conformation, Asn 983 substitutes for Val 984 , and Val 984 and Val 985 form the dimer contact. At opposite ends of the interface there are two pairs of closely apposed histidine side chains. In one case, a pair of aspartates can form salt bridges with the histidines. Despite the intriguing crystal structure, we do not yet have independent evidence that the dimer has physiological relevance. The interaction is not strong, since recombinant A3 runs as a monomer on a gel filtration column at concentrations up to 50 M.
Collagen Binding-We have previously shown that the principal recognition site for types I and III collagen resides in the A3 domain of vWF (8). To explore the role of the MIDAS motif in vWF-A3 binding to collagen, we introduced a series of mutations into the MIDAS-like sequence, expressed the recombinant mutant proteins, and examined their ability to bind to collagen by two independent methods. The mutations were either to homologous residues in vWF-A1 (Ser 938 3 Arg and Ser 1005 3 Val) (since vWF-A1 binds only weakly to collagen), two alanine mutants (Asp 934 3 Ala and Asp 1039 3 Ala), and one loss of charge mutant (Asp 1039 3 Gln). All five of the mutant proteins were readily expressed in E. coli and purified to homogeneity. As shown in Fig. 4, none of the five substitutions perturbed collagen binding.
Comparison with the ␣ 2 ␤ 1 I Domain-The integrin ␣ 2 ␤ 1 binds collagen via the homologous I domain. This led to the suggestion that vWF and ␣ 2 ␤ 1 might bind collagen via similar motifs. The crystal structure of ␣ 2 ␤ 1 I domain has recently been determined 2 ; it contains, as expected, an authentic MIDAS motif with bound Mg 2ϩ . Two groups have shown that collagen binding to ␣ 2 ␤ 1 is metal ion-dependent, being supported by Mg 2ϩ and Mn 2ϩ , but not by Ca 2ϩ (21,22). In addition, point mutations in residues that comprise the MIDAS motif abolish collagen binding, implicating the upper surface of the domain (23). The two domains are different in other respects also: the ␣ 2 ␤ 1 I domain contains an insertion at the top of strand ␤E which forms a prominent ␣-helix that protrudes from the MIDAS face of the domain. vWF-A3 contains no such insertion (Fig. 2b). The surface charge distribution on the MIDAS face is also quite different; in vWF-A3 the upper surface is largely acidic, and in ␣2-I, the metal ion creates a region of positive charge (Fig. 5). One similar feature is the lack of helix ␣2 and the extension of helix ␣3 compared with the leukocyte (␣M and ␣L) I domains, which creates a flat hydrophobic surface at one edge of the ␤-sheet (Fig. 2b). DISCUSSION The crystal structure of the vWF-A3 domain shows that the domain does not bind metal even though the vestigial MIDAS motif contains only one non-conservative change (Asp to Thr). The lack of metal binding to the isolated vWF-A3 domain did not preclude the possibility that metal bound only to the A domain-collagen complex, providing a bridge between them (as suggested for integrin I-domain-ligand interactions (10)). However, our mutagenesis studies demonstrate that residues comprising the vestigial MIDAS motif are not critical for collagen binding, effectively ruling out a role for metal in vWF-A3 collagen interactions.
The lack of metal binding may be related to the unusual conformation of the phenylalanine residue that follows the DXSXS motif. Its side chain sticks out into solution, rather than packing into the hydrophobic core as the analogous residues do in the integrin I domains. This opens a water channel to the buried aspartic acid, Asp 934 . In the integrin I domains, the negative charge is neutralized by the metal ion. The water channel may thus be important for the stability of the folded conformation of vWF-A3, obviating the need for metal. In vWF-A1, one of the MIDAS serines is replaced by an arginine (Arg 524 ), which may play a role analogous to the metal ion in stabilizing the buried aspartic acid (Asp 520 ).
Comparison with the structure of the ␣ 2 ␤ 1 I domain shows that although they have similar three-dimensional folds, there are no obvious shared structural features that could define a common collagen binding motif. The metal ion dependence of ␣ 2 ␤ 1 -collagen binding and the existence of a conventional MIDAS motif make it seem very likely that the mode of collagen binding is different and therefore that the requirement for specific collagen sequences will be different.
The A/I domain fold is a member of a larger family called the dinucleotide binding fold, which is present in many intracellular phosphoryl transfer enzymes (24). In enzymes with this class of fold, the top of the central ␤-sheet (which contains the MIDAS motif in integrin I domains) always plays a functional role. Our results do not rule out a role for the upper surface of the vWF-A3 domain in collagen binding. Alternatively, vWF-A3 may employ this surface for another purpose, such as binding to the A1 or A2 domains, with collagen binding to a different surface. Our crystal structure will allow the design of experiments to test these hypotheses.