Identification of the Collagen-binding Site of the von Willebrand Factor A3-domain*

Von Willebrand factor (vWF) is a multimeric glycoprotein that mediates platelet adhesion and thrombus formation at sites of vascular injury. vWF functions as a molecular bridge between collagen and platelet receptor glycoprotein Ib. The major collagen-binding site of vWF is contained within the A3 domain, but its precise location is unknown. To localize the collagen-binding site, we determined the crystal structure of A3 in complex with an Fab fragment of antibody RU5 that inhibits collagen binding. The structure shows that RU5 recog-nizes a nonlinear epitope consisting of residues 962–966, 981–997, and 1022–1026. Alanine mutants were constructed of residues Arg 963 , Glu 987 , His 990 , Arg 1016 , and His 1023 , located in or close to the epitope. Mutants were expressed as fully processed multimeric vWF. Mutation of His 1023 abolished collagen binding, whereas mutation of Arg 963 and Arg 1016 reduced collagen binding by 25– 35%. These residues are part of loops a 3

Platelet adhesion to damaged vessel walls is the first step in the formation of an occluding platelet plug, which leads to the arrest of bleeding during normal hemostasis. Platelet adhesion can also cause thrombotic complications such as the occlusion of atherosclerotic arteries (1). The multimeric glycoprotein von Willebrand factor (vWF) 1 plays an essential role in platelet adhesion under conditions of high shear stress (2,3). In this process vWF serves as a molecular bridge that links collagen exposed by the damaged vessel wall to glycoprotein Ib located on the platelet surface. Collagens that act as binding sites for vWF include types I and III in perivascular connective tissue and type VI in the subendothelial matrix (3,4).
Mature vWF consists of a 2050-residue monomer that contains multiple copies of so-called A, B, C, and D type domains and one CK (cystine knot) domain arranged in the order DЈ-D3-A1-A2-A3-D4-B1-B2-B3-C1-C2-CK (1,3). Disulfide bond formation between N-terminal D3 domains and between Cterminal CK domains generates vWF multimers that consist of up to 80 monomers. The A1 domain contains the binding site for glycoprotein Ib (5). The A3 domain (residues 920 -1111) contains the major binding site for collagen types I and III (6). The multimeric structure of vWF is essential for high affinity collagen binding (7). Multimeric vWF binds collagen with an apparent K d of 1-7 nM (8), while a recombinant A3 domain has a much higher K d of 2 M (9). Deletion of the A2 and D4 domains, which flank the A3 domain, or deletion of the A1 domain do not decrease collagen binding of multimeric vWF (6,8). These data show that a monomeric A3 domain contains a fully active collagen-binding site, the only requirement for tight binding to collagen being the presence of multiple A3 domains within one vWF multimer.
Integrin I-type domains are homologous to vWF A-type domains (10,11). I domains of integrin ␣-chains ␣ 1 , ␣ 2 , ␣ 10 , and ␣ 11 all possess collagen-binding sites. A crystal structure of the ␣ 2 -I domain reveals binding of a collagen-like peptide to a groove in the surface of the "top" face of the domain (12). This groove contains a so-called metal ion-dependent adhesion site (MIDAS) (13,14), which engages a glutamate residue of collagen.
The location of the collagen-binding site in the vWF-A3 domain is not known. Crystal structures of A3 do not display a collagen-binding groove in the top face, instead, the surface of A3 is rather smooth (15,16). Although the MIDAS motif is partly conserved, binding of A3 to collagen does not require a metal ion (17,18), and no metal ion is observed in crystal structures of A3. Moreover, point mutations in the MIDAS motif of A3 do not disrupt collagen binding (8,16) showing that the motif is not involved in collagen binding, at all. Site-directed mutagenesis studies of other residues in the top face of A3 have yielded conflicting results. Cruz et al. (19) reported in abstract form that amino acid substitutions D1069R, R1074D, R1090D, and E1092R resulted in a 50% reduction in binding of monomeric A3 to collagen. Van der Plas et al. (8), however, observed normal collagen binding of fully processed multimeric vWF containing mutations D1069R, D1069A, or R1074A. In the same study, mutations V1040A/V1042A, D1046A, and D1066A also displayed normal collagen binding, suggesting that the collagen-binding site of vWF-A3 is not located in its top face.
The crystallographic study presented here was conducted to provide new clues on the location of the collagen-binding site of the vWF-A3 domain. We determined the structure of the A3 domain in complex with a Fab fragment of monoclonal antibody RU5, which inhibits binding of vWF to collagen. Site-directed mutagenesis of residues located in the epitope region show that the collagen-binding site is located distant from the top face of A3.

EXPERIMENTAL PROCEDURES
Purification of vWF-A3 and RU5-Recombinant selenomethionine (Se-Met) A3, comprising residues 920 -1111 of human vWF, was expressed and purified as described before (15). For production of monoclonal antibody RU5 (IgG 2a, ) hybridoma cells were injected in mice, and ascites fluid was collected (Eurogentec, Seraing, Belgium). IgG was purified on a protein G-Sepharose column, and Fab fragments were generated using an ImmunoPure Fab kit (Pierce, Rockford, IL). RU5-Fab was further purified with A3-affinity chromatography. For that goal, 10 mg of A3 was irreversibly bound via its N-terminal histidine tag to cobalt(III)-iminodiacetate-chelating Sepharose (Amersham Pharmacia Biotech, Uppsala, Sweden), according to a procedure described by Hale (20). The RU5-Fab fragment was loaded onto the A3-affinity column, washed with PBS, and subsequently eluted with 50 mM triethylamine solution (pH 10.0). Some aggregates were removed on a Superdex 75 HR 10/30 gel filtration column (Amersham Pharmacia Biotech). Running buffer was 10 mM Tris-HCl, pH 8.0, 25 mM NaCl. Next, RU5-Fab was mixed with a 2-fold molar excess of (Se-Met) A3. The A3⅐RU5 complex was separated from excess A3 by gel filtration chromatography. Dynamic light scattering measurements on a Dynapro-801 DLS instrument (Protein Solutions, Charlottesville, VA) indicated the presence of 69-kDa particles in agreement with an expected molecular mass of about 72 kDa of the complex between A3 and RU5.
Determination of RU5-variable Domain Sequences-The amino acid sequence of the variable domain of the heavy chain (V H ) of RU5 was deduced from a cDNA nucleotide sequence. Cloning and sequencing was carried out using established procedures. Total RNA was extracted from 4 ϫ 10 6 RU5 hybridoma cells with RNAzol using the RNA isolation protocol of the supplier (Campro Scientific, Veenendaal, The Netherlands). First-strand cDNA was synthesized using Superscript II reverse transcriptase (Life Technologies, Inc., Rockville, MD) in the presence of 3.2 units/l RNase inhibitor RNaseOUT (Life Technologies, Inc.). cDNA, encoding the V H domain, was amplified by polymerase chain reaction with Pwo DNA polymerase (Roche Molecular Biochemicals, Mannheim, Germany). A backward variable region consensus primer (B4: CCA GGG GCC AGT GGA TAG ACA AGC TTG GGT GTC GTT TT) was used together with a forward primer (B3c: CGG ATG GCC AGG T(C/G)(A/C) AGC TGC AG(C/G) AGT C(A/T)G G) that hybridizes with a consensus sequence in the J H region (21). The polymerase chain reaction product was extended with a 3Ј-A overhang using Taq DNA polymerase (Promega, Madison, WI) and cloned into the pCR2.1-TOPO vector according to the manufacturer's protocol (Invitrogen, Leek, The Netherlands). Nucleotide sequences of three clones were determined (GenBank accession number AF286587) (Fig. 1A).
For determination of the sequence of the variable domain (V L ) of the RU5 -light chain, the same procedure was used as for the V H domain. For cloning of V L cDNA, one backward variable region consensus primer (PD1: GAT ATT GTG ATG ACC CAG TCT (C/G)T) and two forward variable-region consensus primers (PD3: CAG GAA ACA GCT ATG ACC GAG CTC GTG ATC ACC CAG TCT CC; PD4: TGT AAA ACG ACG GCC AGT TCT AGA TGG TGG GAA GAT GGA) were used. However, determination of the sequence was hampered by the abundance of mRNA encoding the light chain of the nonproducing myeloma fusion partner of the RU5 hybridoma cell, a known drawback of myeloma cell line P363Ag8.653 (22).
Crystallization and Data Collection-Crystals of the (Se-Met) A3⅐RU5 complex were grown by hanging-drop vapor diffusion at 4°C using a protein concentration of 15 mg/ml and a precipitant solution consisting of 13% (v/v) isopropanol, 22% (v/v) 2-methyl-2,4-pentanediol, and 100 mM cacodylic acid, pH 5.3. The crystal used for structure determination had a size of 0.1 ϫ 0.1 ϫ 0.2 mm 3 and was equilibrated in a cryoprotectant solution consisting of 8% isopropanol, 30% 2-methyl-2,4-pentanediol, 100 mM cacodylic acid, pH 5.3. X-ray diffraction data were collected on beam line BW7B of the synchrotron radiation facility of the EMBL outstation (Hamburg, Germany) with a Mar345 imaging plate (Mar, Evanston, IL). Data reduction, merging, and scaling were performed with DENZO and SCALEPACK (23). Diffraction data statistics can be found in Table I. Diffraction of the A3⅐RU5 crystal was anisotropic. For structure determination, anisotropic B-factors were applied with SCALEIT and SFTOOLS (24) to correct for the observed anisotropy.
Structure Determination and Refinement-A self-rotation function calculated with POLARRFN (24) and the unit-cell volume suggested the presence of three A3⅐RU5 complexes in the asymmetric unit (a.u.) with a V M of 3.3 Å 3 /Da and a solvent content of 63%. Initial attempts to solve the structure by molecular replacement failed. Therefore, a rather weak anomalous signal (͗⌬F ano /(F)͘ ϭ 1.18) arising from the presence of Se-Met in A3 was used to locate Se sites from an anomalous Patterson map. Six Se sites could be assigned to Met 947 , Met 998 , and Met 1097 of two A3 molecules on the basis of inter-atomic distances and the observed 2-fold non-crystallographic symmetry (n.c.s). The positioning of two A3 molecules in the a.u. allowed further molecular replacement to proceed in a straightforward manner. A Fab fragment of protein data bank entry 2MPA (25,26) was oriented using the program AmoRe (27) followed by Patterson correlation refinement using the program CNS (28,29). Cross-translation functions calculated with CNS identified the position of three Fab fragments. A third A3 molecule was placed on the basis of n.c.s. The a.u. finally contained three A3⅐RU5 complexes with an R-factor of 42.6% after rigid body refinement.
For model building, sequences of the constant domain of the light chain (C L ) and of the constant domain of the heavy chain (C H1 ) of RU5 were taken from IgG 2a, monoclonal antibody 4-4-20 (30,31). The sequence of residues 1-110 of the V L domain of RU5 ( Fig. 1B) was derived from the electron density aided by a consensus sequence based on an alignment of 110 Fab sequences. For model refinement, cycles of rebuilding using O (32) and positional and B-factor refinement using CNS were performed until convergence. Cross-validation was used throughout refinement using a 5% test set of reflections. Refinement used the maximum-likelihood algorithm (33). Bulk solvent correction and anisotropic scaling of diffraction data were applied. During the first cycles of refinement n.c.s. restraints were used. Based on the behavior of the free R-factor, n.c.s. restraints were omitted in later stages of refinement. Water molecules were placed in difference electron density peaks with a peak height of at least 2.8 , a distance of 2.5-3.4 Å to a hydrogen bond donor or acceptor, and a B-factor smaller than 65 Å 2 .
Construction of vWF Point Mutants-Point mutations were introduced in the vWF-A3 domain using the QuikChange method (Stratagene, La Jolla, CA) and specific primers (Amersham Pharmacia Biotech, Roosendaal, The Netherlands). First, a 518-base pair NheI-Csp45I fragment corresponding to amino acid residues 940 -1113 of mature vWF was subcloned into pBluescript SKII (Stratagene, La Jolla, CA). To this end a unique Csp45I restriction site was introduced at position 5628 of expression vector pNUT-vWFcas (8) in the following way. The BamHI-EcoRV fragment of pNUT-vWFcas was ligated into pBluescript SKII. The Csp45I site was introduced using QuikChange and confirmed by sequencing. The NheI-EcoRV fragment containing the new Csp45I site was ligated into pNUT-vWFcas generating pNUT-vWFcas2. Next, the NheI-Csp45I fragment of pNUT-vWFcas2 was made blunt by filling in the 5Ј-overhangs using Pwo DNA polymerase. Ligation of this fragment into EcoRI-AccI-digested and Pwo-treated pBluescript SKII, produced mutagenesis plasmid pBSvWF-NC. This plasmid retains the NheI and Csp45I restriction sites. Point mutations R963A, E987A, H990A, R1016A, and H1023A were introduced in pBSvWF-NC. Mutation H1023A causes the disappearance of a NsiI restriction site. The NheI-Csp45I fragment of this mutant was ligated directly into pNUT-vWFcas2 and confirmed by restriction analysis. The NheI-Csp45I fragment of the other point mutants were ligated into pNUT-vWFcas2-H1023A, with reappearance of the NsiI site.
Expression, Purification, and Characterization of vWF-vWF was stably expressed in baby hamster kidney cells overexpressing furin, necessary for proper removal of vWF propeptide (6, 34). vWF was purified by immunoaffinity chromatography using monoclonal antibody RU8, which is directed against the D4 domain (6). vWF-containing fractions were pooled and stored in aliquots at Ϫ20°C until use.
The concentration of vWF was determined by a sandwich enzymelinked immunosorbent assay using polyclonal ␣-vWF and horseradish peroxidase-conjugated polyclonal ␣-vWF (Dako, Glostrup, Denmark) for immobilization and detection, respectively (6). Normal pooled plasma from 40 healthy donors was used as a reference.
The multimeric structure of vWF was analyzed by agarose gel electrophoresis followed by Western blotting as described by Lawrie et al. (35).
Collagen Binding Assay-vWF binding to fibrillar human placenta collagen type I (Sigma Chemical Co., St. Louis. MO, cat. no. C-7774) and collagen type III (Sigma, cat. no. C-4407) was studied in a solid-state binding assay according to Van der Plas et al. (8). Collagen was coated at 50 g/ml instead of 100 g/ml as used previously (8). A vWF concentration of 2.5 g/ml was used in the binding experiments.

A3⅐RU5 Structure Determination and Refinement-
The structure of the vWF (Se-Met) A3 domain in complex with an Fab fragment of RU5 was solved to 2.0-Å resolution. The structure was determined using the anomalous signal from selenium atoms in (Se-Met) A3 to position two A3 molecules and was completed by molecular replacement. The a.u. contains three A3⅐Fab complexes. The structure has been refined to an R-factor of 22.7% and a free R-factor of 26.4% (Table I).
The sequence of the RU5 V L domain could not be determined from cDNA and was therefore deduced from electron density aided by an alignment of 110 Fab V L sequences. The identity of 13 amino acid residues could not be determined uniquely (Fig.  1). The structure displays good model geometry with 88.8% of the residues in the most favored region of the Ramachandran plot and 10.8% in the additionally allowed region. Residue Thr 51 of RU5 V L domains occurs in the disallowed region of the Ramachandran plot, but its electron density is convincing.
The a.u. consists of three A3⅐RU5 complexes that are denoted A, B, and C. The overall structure of the (Se-Met) A3 molecule is the same as the structure of native A3 (15,16).

N-glycosylated. Electron density near Asn 56 accounted for a GlcNAc(-Fuc)-GlcNAc moiety in complexes A and B and only a
GlcNAc-Fuc moiety in complex C. The model contains one cacodylate ion from the crystallization solution.
Crystal Packing and Differences between Complexes-In the a.u. A3⅐RU5 complexes B and C form a tightly interacting anti-parallel dimer and are related by 2-fold n.c.s. A3⅐RU5 complex A forms a similar anti-parallel dimer with a crystallographically related complex AЈ. Complexes within dimers B-C and A-AЈ have large contact areas of 932 Å 2 and 1454 Å 2 , respectively (GRASP (37)). These large interaction surface areas suggest that the dimer of A3⅐RU5 complexes may also be stable in solution. Dynamic light-scattering measurements, however, clearly indicated that the A3⅐RU5 complex does not form dimers in solution. Therefore, dimers observed in the crystal are a result of crystal packing.
Superposition of C ␣ atoms of the three A3⅐RU5 complexes gives large root mean square coordinate differences ranging from 0.7 to 0.9 Å. When superpositions are limited to A3 molecules with bound V H and V L domains, root mean square coordinate differences of 0.5-0.6 Å are obtained. Visual inspection of these superpositions (Fig. 2) shows that the A3⅐RU5 interaction region is well conserved, whereas regions of A3 and the variable domains located further away from the interaction region superimpose less well. A more detailed inspection of the A3⅐RU5 binding region reveals that all side-chain conformations are very similar (data not shown). In conclusion, the A3⅐RU5 interaction region is well conserved among the three complexes in the a.u. The Epitope of RU5-A3 interacts with RU5 through a nonlinear epitope comprising residues in loop ␣1␤2, loop ␤3␣2 followed by helix ␣2, and loop ␣3␤4 (Table II). All three loops that contribute to the epitope are located in the bottom face of A3 (Fig. 3). The N and C termini of A3 that are also located in  the bottom face do not interact with RU5. This observation is in agreement with the fact that RU5 was raised against complete vWF in which the termini connect A3 to flanking A2 and D4 domains. Residues of RU5 that interact with loops ␣1␤2 and ␣3␤4 are located in CDR L1 and CDR L2 (Fig. 1). All six CDRs interact with residues in the contiguous segment formed by loop ␤3␣2 and helix ␣2. The A3⅐RU5 interactions involve five hydrogen bonds and one salt bridge. The buried surface area of A3 molecules in the complex is ϳ1200 Å 2 , which is about 7% of the total surface area of A3. Interestingly, the fucose moiety attached to residue Asn 56 in CDR H2 of complex A interacts with Lys 988 -Ala 989 of A3. The carbohydrate moieties attached to Asn 56 in complexes B and C do not interact with A3.
Conformational Changes in A3-To analyze whether binding of RU5 causes conformational changes in A3, we compared models of A3 in the A3⅐RU5 complex with two structures of free A3 2 (15,16). The two structures of free A3 were determined from different crystal forms with unrelated crystal-packing interactions. In both crystal forms, two molecules are present in the a.u. Because the A3⅐RU5 complex was obtained with (Se-Met) A3, we also included the structure of free (Se-Met) A3 in the comparison to detect possible structural differences caused by Se-Met. 3 Root mean square coordinate differences after pairwise superimposing C ␣ atoms of all eight models range from 0.24 to 0.81 Å. Large differences are restricted to three loops located within the RU5-binding site and to two loops that are located distant from the epitope (Fig. 3 and Table  II).
The three conformational diverse loops located in the RU5 epitope are ␣1␤2, ␤3␣2, and ␣3␤4 (Fig. 4A). Loop ␣1␤2 has a vWF immobilized in microtiter plate wells via polyclonal ␣DЈD3 was incubated with 5 g/ml RU5. Bound RU5 was detected by horseradish peroxidase-conjugated rabbit-anti-mouse antibody as described under "Experimental Procedures." Wild-type vWF and ⌬A3-vWF, which lacks the A3 domain (6), were used as positive and negative controls, respectively. Binding of RU5 to vWF mutants is shown as a percentage of wt-vWF. Each data point represents the mean Ϯ S.D. of two measurements in duplicate.
FIG. 6. Binding of vWF point mutants to collagen types I and III. Collagen type I (white bar) or III (black bar) was coated in microtiter-plate wells. Wells were incubated with vWF at a concentration of 2.5 g/ml, and bound vWF was detected as described under "Experimental Procedures." ⌬A3-vWF was used as a negative control. Bound vWF mutant is shown as a percentage of wt-vWF. Each data point represents the mean Ϯ S.D. of three independent experiments performed in duplicate. unique conformation in the A3⅐RU5 complex, indicating that this conformation is induced by RU5. Loops ␤3␣2 and ␣3␤4 have considerable conformational freedom. Conformations observed in the A3⅐RU5 complex are, however, not systematically different from the conformations observed in free A3, indicating that the conformations observed in the complex are not induced by RU5.
Distant from the RU5 epitope conformational variation is observed for loops ␤1␣1 and ␣5␤6. Loop ␤1␣1 is located in the top face of the domain and contains part of the vestigial MIDAS motif. Multiple conformations are observed for residues Ser 938 -Phe 939 -Pro 940 (Fig. 4B). Phe 939 is solvent-exposed in A3⅐RU5 complex C and all models of free A3. In complexes A and B Phe 939 is buried in the hydrophobic core of A3. The buried conformation is likely caused by crystal packing interactions involving A3 residues 940 -942. These packing interactions would not allow for the position of Pro 940 observed in models of A3 that have an exposed conformation of Phe 939 . Because both the exposed and buried conformations of Phe 939 are observed in the three A3⅐RU5 complexes, the conformation of this loop is certainly not determined by binding of RU5.
Loop ␣5␤6 located in a side face of A3 has an aberrant conformation in A3 domain B of the structure of Bienkowska et al. (16) (Fig. 4C). This aberrant conformation coincides with crystal contacts unique to molecule B. No crystal contacts are present in the other structures. The conformations observed in the A3⅐RU5 complex are not systematically different from the conformations observed in free A3. Thus, binding of RU5 does not cause conformational changes in regions of A3 that are distant from the epitope.
Selection and Characterization of vWF Mutants-To confirm the location of the collagen-binding site indicated by the A3⅐RU5 complex, we constructed charged-to-alanine mutations of five residues. These residues are located within 5 Å of the RU5 Fab fragment and are solvent-exposed in free A3. Residues selected were Arg 963 located in loop ␣1␤2, Glu 987 (␤3␣2), His 990 (␣2), Arg 1016 (␣3), and His 1023 (␣3␤4). Mutants were produced as multimeric vWF in stable baby hamster kidney cell lines. The multimer distribution of vWF mutants and wildtype vWF, as analyzed by agarose gel electrophoresis, were indistinguishable (data not shown). RU5 binding to purified point mutants was not significantly different from RU5 binding to wild-type vWF (Fig. 5).
Collagen Binding of vWF Mutants-The effect of point mutations on vWF binding to collagen types I and III was investigated by enzyme-linked immunosorbent assay (Fig. 6). Similar results were obtained for both types of collagen. Mutation H1023A almost completely abolished collagen binding. The level of residual binding observed for this mutant was similar to binding observed for ⌬A3-vWF, a deletion mutant that lacks the entire A3 domain (6). Collagen binding of mutants R963A and R1016A was reduced by 25-35%, whereas collagen binding of mutants E987A and H990A was normal.

DISCUSSION
The current study was aimed at locating the collagen-binding site of the vWF-A3 domain. For this purpose we solved the crystal structure of A3 in complex with a Fab fragment of RU5 that inhibits collagen binding. The structure of the complex shows that RU5 binds to residues within A3 sequences 962-966, 981-997, and 1022-1026. These residues are located in ␣-helix 2 and in loops ␣1␤2, ␤3␣2, and ␣3␤4 at the bottom of one of the side faces of the A3 domain (see Fig. 3). Comparison of structures of A3 shows that RU5 binding does not induce long range conformational changes. This excludes a mechanism in which RU5-induced conformational changes inhibit collagen binding. It seems likely, therefore, that RU5 inhibits collagen binding by steric hindrance, which implies that the collagenbinding site is located at or close to the RU5 epitope.
To confirm the location of the collagen-binding site, we constructed five charged-to-alanine mutations of residues located in or close to the RU5 epitope. The multimer distribution of these mutants was similar to that of wild-type vWF. Therefore, observed differences in collagen binding are not caused by the known dependence of collagen binding on vWF multimer size (7). All five mutants bound normally to RU5, which shows that none of the mutated residues plays a dominant role in the A3⅐RU5 interaction and, more importantly, that the conformation of A3 in the neighborhood of the epitope and the collagenbinding site is not disturbed.
Mutation H1023A abolished collagen binding almost completely, residual binding being similar to that observed for ⌬A3-vWF, a deletion mutant that lacks the entire A3 domain. Therefore, His 1023 plays a central role in A3-mediated collagen binding. His 1023 is located in loop ␣3␤4 and lies at the edge between the "front" face of the domain, formed by helices ␣2 and ␣3 and strand ␤3, and the bottom face, which is composed of several loops and contains the N and C termini (Fig. 7). A small reduction of collagen binding was observed for mutants R963A and R1016A located in the bottom and front face of the domain, respectively.
Interestingly, His 1023 and flanking residues display a large variety of conformations among eight models of A3 (Figs. 4 and  7). In some A3 structures His 1023 protrudes prominently from the surface of the domain, which may be a favorable position for interaction with collagen. Multiple conformations are also ob- FIG. 7. Residues of vWF-A3 involved in collagen binding. Stereographic representation of the vWF-A3 domain showing amino acid residues that have been mutated to alanine in ball-and-stick. Mutation of His 1023 (magenta) abolishes collagen binding almost completely. Mutation of Arg 963 and Arg 1016 (green) reduces collagen binding by 25-35%. Mutation of Glu 987 and His 990 (gray) does not have an effect on collagen binding. To illustrate the range of conformations of His 1023 observed in different crystal forms of A3, two conformations of this residue and the backbone trace of flanking residues are shown. Residues that have been mutated in previous studies (8,16,19) are located in the top face of A3 and are indicated by black spheres. The diagram was generated with MOLSCRIPT (43) and RASTER3D (44). served for loop ␤3␣2. Like His 1023 , loop ␤3␣2 is located at the edge between the front and bottom faces of A3. Because we did not mutate residues in loop ␤3␣2, its involvement in collagen binding remains to be established. The observed flexibility of His 1023 suggests that collagen binding may involve an inducedfit mechanism in which significant conformational changes occur in loop ␣3␤4 upon binding of A3 to collagen.
The amino acid sequence of collagen that is recognized by vWF-A3 has not yet been identified. We hypothesized previously that negatively charged residues in A3 could interact with basic residues on collagen (15). Residues now implicated in collagen binding are positively charged. Therefore, interaction of A3 with negatively charged residues on collagen appears more likely.
In contrast to binding sites of other collagen binding domains, like the ␣ 1 and ␣ 2 -I domains and the A domain of Staphylococcus aureus adhesin (12,38), the collagen-binding region of A3 does not have a groove or trench that could accommodate a collagen triple helix. The front face of the domain, harboring Arg 1016 , is rather flat. The bottom face, which contains Arg 963 is less smooth, but no groove is present. Docking of a collagen triple helix on the A3 domain is not straightforward. In particular, it is not obvious how His 1023 , Arg 963 , and Arg 1016 could simultaneously contact a triple helix in an extended conformation. To define the collagen-binding site more precisely, characterization of additional mutants will be necessary.
Previously, the collagen-binding site of A3 was proposed to be located at its top face (15,19) similar to the homologous I domains of integrins ␣ 1 ␤ 1 and ␣ 2 ␤ 1 (12, 39 -41). Point mutations introduced in the top face of an A3 monomer (19) and in multimeric vWF (8) gave conflicting results. Our results now show conclusively that the collagen-binding site is located close to the bottom face and not in the top face of A3.
Although our results rule out a role for the top face of the A3 domain in collagen binding, this side of the molecule may still be engaged in other interactions, such as binding of the A1 domain. Interaction between A1 and A3 has been suggested to play a role in activation of the A1 domain for binding to platelet receptor glycoprotein Ib (42). Interesting in this respect are the buried and solvent-exposed conformation observed for residue Phe 939 , which is located close to the vestigial MIDAS motif in the top face of the domain (Fig. 4B). Solvent exposure of Phe 939 has been proposed to stabilize the buried Asp 934 of the vestigial MIDAS motif in the absence of a bound metal ion (16). Our observation of a buried conformation shows that exposure of Phe 939 is not critical for structural stability. The two conformations of Phe 939 may, however, be relevant for the putative interaction between A1 and A3, because the shape and hydrophobicity of the upper surface of A3 differs significantly between the solvent-exposed and buried conformation.
In conclusion, the collagen-binding site of vWF-A3 is distinctly different from collagen-binding sites of I domains of integrins ␣ 1 ␤ 1 and ␣ 2 ␤ 1 . vWF-A3 residues involved in collagen binding are located close to the bottom face of the domain. His 1023 is essential for collagen binding, whereas Arg 963 and Arg 1016 play ancillary roles. Multiple conformations observed for His 1023 and adjacent residues suggest that binding of A3 to collagen involves an induced-fit mechanism.