Structural Basis of Regulation of von Willebrand Factor Binding to Glycoprotein Ib*

Background: Force in fluid flow regulates von Willebrand factor (VWF) A1 domain binding to glycoprotein Ibα (GPIbα). Results: X-ray crystal structures of high affinity A1-GPIbα complexes and mutations reveal interactions involving central leucine-rich repeats of GPIbα. Conclusion: Structural changes are on a pathway to a force-induced super high affinity state. Significance: A1-GPIbα complexes provide insight into mechanochemistry of bleeding disorders. Activation by elongational flow of von Willebrand factor (VWF) is critical for primary hemostasis. Mutations causing type 2B von Willebrand disease (VWD), platelet-type VWD (PT-VWD), and tensile force each increase affinity of the VWF A1 domain and platelet glycoprotein Ibα (GPIbα) for one another; however, the structural basis for these observations remains elusive. Directed evolution was used to discover a further gain-of-function mutation in A1 that shifts the long range disulfide bond by one residue. We solved multiple crystal structures of this mutant A1 and A1 containing two VWD mutations complexed with GPIbα containing two PT-VWD mutations. We observed a gained interaction between A1 and the central leucine-rich repeats (LRRs) of GPIbα, previously shown to be important at high shear stress, and verified its importance mutationally. These findings suggest that structural changes, including central GPIbα LRR-A1 contact, contribute to VWF affinity regulation. Among the mutant complexes, variation in contacts and poor complementarity between the GPIbα β-finger and the region of A1 harboring VWD mutations lead us to hypothesize that the structures are on a pathway to, but have not yet reached, a force-induced super high affinity state.

von Willebrand factor (VWF) 3 is a plasma glycoprotein with monomers that are 2,050 residues in length and are linked head-to-head and tail-to-tail into ultralong concatemers. VWF senses exposure of the subendothelium and changes in flow at sites of bleeding and is required for hemostasis in the rapid flow of the arteriolar circulation (1)(2)(3)(4)(5). VWF binds through its A1 domain to the GPIb␣ subunit of the GPIb-GPIX complex on platelets. No interaction between VWF and GPIb␣ is seen in stasis or in low flow; however, above a threshold flow, they bind and mediate in vivo platelet plug formation and in vitro aggregation of platelets in stirred cuvettes and rolling of platelets on VWF immobilized on the walls of flow chambers (6). Multiple mechanisms are thought to contribute to enhancement of VWF adhesiveness by flow. One mechanism by which flow can activate VWF is induction of transition from a bird's nest to an elongated conformation, leading to better exposure of the VWF monomers for multimeric binding to platelets (7).
Conformational change within the A1 domain of VWF and platelet glycoprotein Ib␣ is the second postulated mechanism for regulating their interaction (8). In flow, the hydrodynamic force exerted on VWF multimers results in tensile force exerted throughout the multimer at the connections between neighboring domains (e.g. on the N and C termini of the A1 domain). The force exerted near the middle of a long VWF multimer free in shear flow has been estimated to reach 10 pN, and it would be much higher for VWF in elongational flow or bound to the vessel wall at a site of hemorrhage (5,9). Named after the three A domains in VWF, VWA domains (or VWA folds) are widely distributed and also appear in integrins as the inserted domains in their ␣ and ␤ subunits (i.e. the integrin ␣I and ␤I domains). As a VWA domain, A1 has largely alternating ␤-strands and ␣-helices that assemble into a Rossman-like fold with a central ␤-sheet surrounded by amphipathic ␣-helices (10). In integrin VWA (␣I and ␤I) domains, a 10-Å axial displacement of the C-terminal ␣-helix is linked to communication of allostery to a neighboring domain and to a change in conformation at the metal ion-dependent adhesion site where ligand is bound (11). However, the A domains in VWF contain no bound metal ion, and GPIb␣ binds at a completely different site on A1 to the edge of the central ␤-sheet rather than to the turns between ␤-strands and ␣-helices at the C-terminal end of the ␤-sheet (12,13).
Mutations in VWF are the most frequent cause of heritable bleeding disorders and provide important insights into function. Interesting gain-of-function mutations in von Willebrand disease (VWD) type 2B localize to the region of A1 in or adjacent to the ordered N-and C-terminal segments, distal to the binding site for platelet GPIb␣. Therefore, it has been suggested that these mutations might release A1 from a lower affinity conformation (10,12). Notably, the sites of these mutations are near the termini where elongational force in VWF multimers is exerted on A1.
Gain-of-function mutations in platelet type VWD (PT-VWD) locate in GPIb␣ and have a phenotype similar to VWD type 2B mutations in VWF A1. GPIb␣ has an N-terminal leucine-rich repeat (LRR) domain with eight LRRs and N-and C-terminal segments that cap the LRRs (10). A partially ordered ␤-switch region in the C-terminal cap of GPIb␣ becomes ordered upon binding to A1 and forms a ␤-hairpin with two ␤-strands that add onto the central VWA ␤-sheet (12). GPIb␣ VWD mutations all map to the ␤-switch and are thought to favor the ␤-strand conformation found in the bound state over a disordered or ␣-helical state found in the unbound state (12)(13)(14).
Crystal structures of the A1 and GPIb␣ complex show two discontinuous binding interfaces, whether the components are wild-type or contain gain-of-function A1-R1306Q and GPIb-M239V mutations (12,13). Differences were seen between wild-type and gain-of-function complexes in the N and C termini and ␣1-␤2 loop of A1. However, no compelling gained interactions were seen in the gain-of-function mutant complex that could explain its higher affinity, and it has remained unclear how the affinity of A1 and GPIb␣ for one another could be regulated by hydrodynamic flow. Thus, in contrast to integrin I domains, an allosteric mechanism for regulating affinity of A1 and GPIb␣ for one another remains unclear.
Two types of evidence for force-induced conformational change in the A1-GPIb␣ complex motivated the current study. Chimeras between human and canine GPIb␣ showed that the LRRs between the two discontinuous contacts with A1 were unimportant at low shear but became important at high shear (15,16). Measurements on a receptor and ligand in a single molecule (ReaLiSM) construct in which A1 was connected at its C terminus through a polypeptide linker to the N terminus of GPIb␣ revealed reversible unbinding and rebinding (8). Two distinct pathways of force-dependent dissociation were found. The kinetics of force-dependent dissociation (k off ) of the two pathways were well fit by biophysical models that relate k off to k off in the absence of force (k off 0 ) and the exponential of tensile (elongational) force. A complex with faster k off 0 predominated below a force of 10 pN, and a complex with a slower k off 0 and lesser exponentiation by force predominated above 10 pN, suggesting two conformational states of the receptor-ligand complex (8).
Here, we use crystallographic structures and mutations to address the principles that regulate affinity of A1 and GPIb␣ for one another. Moving beyond the use of one gain-of-function mutation in A1 and one in GPIb␣ (12), we use a total of four VWD gain-of-function mutations, two each in A1 and GPIb␣. Using directed evolution in yeast, we discover further gain-offunction mutations in A1. Shifting the long range disulfide bond in A1 by one residue in sequence causes a large increase in affinity. We describe the structure of this mutant alone and complexed with GPIb␣ containing two VWD mutations. In our high affinity structures, a gained interaction is found in the region implicated in GPIb␣ chimeras as important at high shear stress (15) and is found to be important in mutational experiments. Structural analysis of the region harboring gain-of-function mutations in A1 suggests clashing rather than complementary interactions with the GPIb␣ ␤-finger and leads us to suggest that our structures are on a pathway to, but have not yet reached, a hypothetical force-induced super high affinity state.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The cDNA sequence encoding VWF A1 (Glu 1264 -Leu 1469 ) was cloned into the pET21a vector with an initiation Met codon and expressed in BL21(DE3) cells (17). Gain-of-function mutations were introduced using QuikChange (Stratagene). Refolding was performed essentially as described (17). Refolded A1 was purified with nickel-nitrilotriacetic acid beads (Qiagen), followed by a heparin-Sepharose affinity column (GE Healthcare), and a S75 gel filtration column (GE Healthcare). The cDNA for human platelet GPIb␣ (His 1 -Arg 290 ) was mutated using QuikChange (Stratagene). For crystallization, two N-linked glycosylation sites were removed with N21R and N159R mutations. GPIb␣ mutants were cloned into an ET-6 vector, which resembles ET-1 (18), except it contains only a C-terminal His 6 tag. Stable transfectants were established in HEK293T cells. Protein was expressed in 293 Freestyle medium (Invitrogen) and initially purified using an anti-GPIb␣ IgG-Sepharose column. Monoclonal antibody 12E4 (19) was coupled to cyanogen bromideactivated Sepharose 4B FF (GE Healthcare) using the manufacturer's protocols. Culture medium supernatant was flowed over 12E4-Sepharose and eluted with 0.1 M glycine, pH 2.4, and immediately neutralized with one-quarter volume of 1 M Tris, pH 8.5. GPIb␣ was further purified on a S200 gel filtration column (GE Healthcare).
Yeast Surface Display-The cDNA sequence encoding the above A1 fragment was cloned into the pCTCON2 yeast surface display plasmid in frame with Aga2-HA tag and with the C-terminal c-Myc tag. A1 random libraries were created using nucleotide analog PCR (20) and transformed into Saccharomyces cerevisiae strain EBY100 by electroporation (21). Library complexity was greater than 10 7 unique clones. Expression was induced using synthetic galactose medium with casamino acids (SGCAA), and 10 9 yeasts were sorted by magnetic activated cell sorting (LS Columns, Miltenyi Biotech) for functional expression using 200 nM biotinylated GPIb␣ prepared with EZ-Link NHS-PEG4-Biotin (Pierce) and coated onto anti-biotin magnetic beads (Miltenyi Biotech). The functional library was then regrown and sorted twice by fluorescence-activated cell sorting (FACS) with 50 nM biotinylated GPIb␣ detected with streptavadin-phycoerythrin (Invitrogen); and with a 1:100 dilution of 9E10 anti-c-Myc antibody (Covance) detected with FITC-labeled anti-mouse IgG (Invitrogen). The sorting gate was selected for improved apparent affinity. Last, the library was sorted for decreased off-rate by competition with 10 M nonlabeled GPIb␣ for 4 h before sorting for cells retaining fluorescence. Fourteen recovered clones were sequenced.
Flow cytometry of yeast was performed on a FACScan (BD Biosciences). 10 5 yeasts were harvested, washed three times in room temperature PBS ϩ 1% BSA (PBSB), and incubated with 50 nM biotinylated GPIb␣ and anti-c-Myc 9E10 IgG (1:200) at room temperature for 1 h. Cells were then centrifuged and washed three times in ice-cold PBSB before incubation with secondary reagents (1:200 dilution of streptavidin-phycoerythrin and 1:100 dilution of FITC-labeled goat anti-mouse IgG) for 20 min at 4°C. 10 4 yeasts were analyzed for mean fluorescence intensity for both GPIb␣ and c-Myc. The specific fluorescence intensity was calculated by subtracting autofluorescence from uninduced controls. Because the fluorescence signal is affected by both affinity and surface expression, we calculate the adjusted specific fluorescence intensity by dividing the specific fluorescence intensity from the GPIb␣ signal by the specific fluorescence intensity from c-Myc and normalizing to WT (WT adjusted specific fluorescence intensity ϭ 1). Single mutations of A1 expressed using yeast display were generated using QuikChange mutagenesis.
Radioligand Binding Experiments-Purified high affinity GPIb␣ containing PT-VWD mutations G233V and M239V (GPIb␣/VWD2) was radioiodinated using the IODO-GEN method (Pierce). 250 g of purified protein was iodinated with 1 mCi of Na 125 I radionuclide (PerkinElmer Life Sciences) for 20 min at room temperature. The reaction was quenched with 2 mg/ml L-tyrosine for 5 min. After the addition of 5 mM potassium iodide, the radiolabeled GPIb␣/VWD2 was purified on a PD-10 column (GE Healthcare) equilibrated with PBS. Radioligand concentration was determined with bicinchoninic acid assay (Pierce) with BSA as a standard. Iodination was Ͼ80% efficient. 10 5 yeast cells displaying A1 were incubated with increasing concentrations of iodinated GPIb␣/VWD2 in PBS with 1% BSA for 2 h. Samples were then centrifuged through oil (67% n-octyl-phthalate, 33% n-butyl-phthalate) (22). The samples were frozen on dry ice, and the cell pellet at the tip was cut and analyzed in a ␥ counter (PerkinElmer Life Sciences). Similar procedures were used for competition assays with 5-10 nM 125 I-GPIb␣/VWD2 as the probe.
Binding Analysis-Binding data were analyzed by non-linear regression using software Prism version 5.03 (GraphPad). Saturation binding data were collected alongside a negative control to determine the nonspecific binding. Both sets of data were globally fit to a one-site binding model extended by a nonspecific term using Equation 1, Specific competition binding data were fit to a one-site model with Equations 2 and 3, where K D,probe was the affinity of 125 I-GPIb␣/VWD2 for that particular A1 mutant, [probe] ϭ 5 or 10 nM depending on the experiment, and B 0 is the specific binding of the probe in the absence of the nonlabeled GPIb␣.
Structure Determination and Refinement-All data were processed with HKL-2000 or XDS. The structures were solved by molecular replacement using MOLREP (23). 1SQ0 served as the model for A1/VWD2-GPIb␣/VWD2; A1/VWD2-GPIb␣/ VWD2 served as the model for A1/SS-GPIb␣/VWD2; and 1AUQ served as the search model for A1/SS. Structure refinement was performed using COOT (24) and PHENIX (25). The resolution for A1/SS-GPIb␣/VWD2 was found to extend to 2.8 Å using cross-correlation (26); we could not similarly extend resolution of the data sets for the other two crystals because their higher resolution reflections went beyond the edge of the detector. Validation and Ramachandran statistics used MOLPROBITY (27). APBS was used to calculate electrostatic surfaces (28).

RESULTS
Directed Evolution of A1 Affinity-The A1 domain of VWF was fused to the Aga2 protein in S. cerevisiae and expressed on the yeast surface (29). A library of greater than 10 7 random mutants was selected for high affinity for GPIb␣ using several steps of isolation with magnetic activated cell sorting and FACS. Fourteen mutants were selected, most of which contained more than one mutation (Table 1). Remarkably, about half contain mutations of residues previously found to be mutated in VWD (Table 1).
The highest affinity single mutant found was Y1271C ( Table  2). Mutant Y1271S was markedly lower in affinity although still higher than wild type. This finding, together with the observation that the Y1271C mutation introduces a cysteine adjacent to Cys 1272 , which forms a long range disulfide bond to Cys 1458 , suggested that a novel disulfide bond introduced by the mutation might be responsible for the increase in affinity. To test this possibility, we prepared Y1271C/C1272R, which could form an alternative, long range Cys 1271 disulfide bond to Cys 1458 , and Y1271C/C1458R, which could form a vicinal disulfide between Cys 1271 and Cys 1272 . The latter showed GPIb␣ binding similar to wild type ( Table 2). In contrast, Y1271C/C1272R showed a 10-fold increase in fluorescent GPIb␣ binding compared with the 5-fold increase in Y1271C (Table 2). These results are consistent with a mixture of Cys 1271 -Cys 1458 and Cys 1272 -Cys 1458 long range disulfide bonds in Y1271C and only a Cys 1271 -Cys 1458 long range disulfide in Y1271C/C1272R. Formation of a disulfide bond in Y1271C/C1272R was confirmed by its differing migration in reducing and non-reducing SDS-PAGE (data not shown). We conclude that moving the long range disulfide bond in A1 so that it connects to residue 1271 rather than 1272 markedly increases affinity.
The primary interaction between A1 and GPIb␣ occurs between the ␤-switch region of GPIb␣ and the central ␤-sheet of the A1 domain (Fig. 1A). LRR proteins contain caps over their N-and C-terminal LRR (32). The ␤-switch of GPIb␣ contains two ␤-strands and extends from the C-cap, which also contains ␣-helices and two disulfide bonds (Fig. 1, A and D). More C-terminal is an anionic region containing sulfated tyrosines that is not ordered in any complex crystal structure to date (Fig. 1D). The ␤-switch joins the central A1 ␤-sheet to form an extended ␤-sheet (Fig. 1A). PT-VWD mutations locate in the ␤-switch (Fig. 1A). The green spheres in Fig  Many of the isolates contained more than one mutation. Mutations of residues known to be mutated in VWD are in boldface type, and those with the exact substitutions in VWD are additionally underlined.

Clone
A1 mutation(s) A1 domain are not observed for any structure. Structural differences in A1 are restricted to the N and C termini and to the ␣1-␤2 loop (Fig. 1, B and E). Differences in the orientation of GPIb␣ relative to A1 are visualized by rotating the superimposed structures shown in Fig. 1B so that A1 faces the reader and then removing A1 to show the concave A1-binding face of GPIb␣ (Fig. 1C). In the WT complex, GPIb␣ is shifted laterally relative to complexes with high affinity mutants (i.e. to the right in Fig. 1C). Further differences between the structures are noted in the ␤-finger of GPIb␣ (Fig. 1, B and C). In contrast, there is little difference in position of the ␤-switch when superposition is performed on A1, in agreement with the constraints imposed by forming hydrogen bonds between the ␤-switch and the ␤-sheet in A1, and forming a large interface.
The ␤-Switch-The ␤-switch is the major binding interface between A1 and GPIb␣ and also the site of PT-VWD mutations. In complex with A1, the ␤-switch consists of two antiparallel ␤-strands linked by a ␤-hairpin turn (12, 13) (Figs. 1A and 2A). The ␤-hairpin joins the central A1 ␤-sheet, with switch residues 237-239 forming antiparallel hydrogen bonds to A1 ␤-strand 3 (Fig. 2, A-C). WT and all mutant structures show identical backbone hydrogen bonds between A1 and ␤-switch and between the two ␤-switch ␤-strands (Fig. 2, A-C) (data not shown). The GPIb␣/VWD1 (M239V) and GPIb␣/WT com- Structural Basis of VWF Binding to GPIb␣ FEBRUARY 28, 2014 • VOLUME 289 • NUMBER 9 plexes show different ␤-hairpin turn geometries near turn residue Val 234 (Fig. 2, B and C) (13); however, it is difficult to relate this to the Met to Val substitution at residue 239, which is four positions away in the ␤-ladder from the ␤-turn. In contrast, the G233V substitution in GPIb␣/VWD2 is adjacent to the ␤-turn ( Fig. 2A). To make room for the Val 233 side chain adjacent to the Val 236 side chain, the backbone at Gln 232 and Val 233 rotates and, together with a hydrogen bond between the Gln 232 side chain and Val 234 backbone, causes a markedly different ␤-turn conformation (Fig. 2, A-C).
The Val side chain added by the G233V substitution does not interact with the A1 domain ( Fig. 2A). Although the M239V substitution lies adjacent to A1, it does not markedly change the buried interface (12, 13) (Fig. 2, B and C). Residues 233 and 239 each lie in regions of the ␤-switch that form ␤-strands in A1-bound but not unliganded GPIb␣ structures (Fig. 2D). Val is a ␤-branched side chain and favors ␤-strand conformation (12); however, PT-VWD mutations may also act by destabilizing the unbound state of GPIb␣ (14). The finding that Val 233 does not interact with A1 supports the concept that substitutions of ␤-switch residues may raise affinity by shifting the conformation of the ␤-switch away from that present in the unbound state and toward that present in the bound state (12,14).
We point out here that in the unbound state of GPIb␣, Trp 230 and preceding residues in the ␤-switch adopt a ␤-conformation in common with the bound state (Fig. 2D). A remarkable -cation bond between Trp 230 and LRR5 residue Lys 132 is found in all GPIb␣ structures (Fig. 2, A-D) and, together with nearby aro-

Structural Basis of VWF Binding to GPIb␣
matic and hydrophobic residues (Fig. 2D), stabilizes this region in ␤-conformation. Because GPIb␣ residues Val 229 and Trp 230 are in ␤-conformation prior to A1 binding, they can act as a template for forming ␤-sheet hydrogen bonds to switch residues 240 -238, followed by joining to the ␤-sheet in A1 and zipping up the remainder of the ␤-switch as a ␤-ribbon. Aromatic residues are abundant on both the A1 and GPIb␣ sides of their ␤-sheet interface and contribute to the stability of this interface (Fig. 2, A-D). The ␣1-␤2 Loop-As already noted, when complex structures are superimposed on the A1 domain (Fig. 3A), the most notable differences in A1 are in the N and C termini and the ␣1-␤2 loop (Figs. 1 (A and B) and 3 (A-C); boxed in Fig. 3A). Previous comparison of single examples in crystals of A1/WT-GPIb␣/WT and A1/VWD1-GPIb␣/VWD1 complexes demonstrated differences in the conformation of the A1 ␣1-␤2 loop and suggested that this correlated with and was responsible for affinity differences (13). However, the larger number of high affinity complexes made available by this work does not support a simple correlation between ␣1-␤2 loop conformation and affinity (a more complex relationship is described under "Dis-  Fig. 3B). In contrast to the A1/SS-GPIb␣/VWD2 complex (and the A1/WT-GPIb␣/WT complex with botrocetin, which shows "sliding" of the A1-GPIb␣ interface (33)), all other complexes show shifts of the ␣1-␤2 loop from uncomplexed A1 WT (Fig. 3C). The ␣1-␤2 loop conformational change is shown in Fig. 3, D-I, by a solid line connecting Gln 1311 (C␣ sphere) to its position in uncomplexed A1WT. In A1/VWD1-GPIb␣/VWD1, the ␣1-␤2 loop is shifted slightly away from the GPIb␣ interface (Fig. 3, C and G); in the A1/VWD2-GPIb␣/VWD2 complex, the loop is shifted further (Fig. 3, C and H); and in the A1/WT-GPIb␣/WT complex, the loop is shifted most and in a different direction toward the A1 termini (Fig. 3, C and I). The greater shift of the ␣1-␤2   loop in the WT complex relates to the different overall orientation between A1 and GPIb␣ (Fig. 1, B and C). Variation in ␣1-␤2 loop conformation also relates to variation in GPIb␣ ␤-finger orientation among complexes (Figs. 1 (B and C) and 3 (C and F-I)). This variation appears to result from avoidance of clashes rather than from maintenance of specific interactions because contacts between A1 and GPIb␣ ␤-finger residues vary markedly among the complexes (Fig. 3, F-I).
The A1 Termini-Superposition of the uncomplexed A1 domains listed in Table 4 reveals a structurally conserved pattern of hydrogen bonds that stabilize the conformation of their termini (Fig. 4A). The N-terminal end is stabilized at Cys 1272 , both through its long range disulfide bond and a backbone hydrogen bond to residue 1308 (Fig. 4A). The C-terminal end is stabilized by a network of backbone-backbone and backboneside chain hydrogen bonds from Cys 1458 to Pro 1462 (Fig. 4A).
VWD type 2B mutations occur in regions that neighbor the N and C termini and, with the exception of Trp 1313 , are distal from GPIb␣ (pink side chains; Fig. 4, A-F). The C-terminal network of hydrogen bonds in uncomplexed A1/WT (Fig. 4A) is conserved in the A1/WT-GPIb␣/WT complex, except for minor alterations due to the Arg 1274 rotamer (Fig. 4B). Many of these hydrogen bonds and consistently the Pro 1462 -Leu 1275 backbone hydrogen bond are also preserved in the mutant, high affinity A1-GPIb␣ complexes (Fig. 4, A-E), and uncomplexed A1/SS (Fig. 4F).
It is difficult to deconvolute those A1 N terminus-proximal and GPIb␣-proximal changes that are due to GPIb␣ binding and those that are due to affinity-enhancing mutations; nonetheless, it is possible to deduce from the structures the effects of the mutations on their immediate A1 environment. In A1/WT, the Arg 1306 side chain hydrogen-bonds to the backbone of residue 1302 (Fig. 4A). In the complexes with A1/VWD1 and A1/VWD2, the mutated Gln 1306 side chain points in a different direction than the WT Arg 1306 side chain (Fig. 4, C and E). At least in the higher resolution VWD2 complex, Gln 1306 forms two hydrogen bonds to the Asp 1451 and Arg 1450 side chains (Fig.  4C); Arg 1450 is a known site of mutation in VWD.
The I1309V mutation present in A1/VWD2 complexes is one of the most subtle VWD mutations; previous uncomplexed A1/I1309V structures showed little change in backbone or even side chain rotamer compared with wild type (34,35). The I1309V mutation, along with R1306Q, is in the A1 ␣1-␤2 loop. The A1/VWD2-GPIb␣/VWD2 complex reveals the behavior of the I1309V mutation in a complex (Fig. 3H). The smaller size of the Val 1309 side chain compared with Ile allows it to dive deeper into the hydrophobic core and hence closer to Met 1304 in the ␣1-helix and Val 1316 in the ␤2-strand (Fig. 3H), compared with Ile 1309 in the A1/VWD1-GPIb␣/VWD1 and A1/SS-GPIb␣/VWD2 complexes (Fig. 3, F and G). However, the markedly different shape of the ␣1-␤2 loop in the A1/WT-GPIb␣/WT complex allows its Ile 1309 to dive almost as deep as Val 1309 in the A1/VWD2-GPIb␣/VWD2 complex (Fig. 3I). Lack of a clear structural phenotype for Val 1309 in a complex structure is consistent with hypothesized further structural change to a super high affinity state (see "Discussion").
Shifting the long range disulfide in the Y1271C/C1272R A1/SS mutant imparts greater freedom to the N terminus of A1. Cys 1271 takes the place of Cys 1272 , with no significant change in the backbone and C␤ position of Cys 1458 and only 1-Å differences in positions of the C␣ and C␤ atoms of Cys 1271 compared with Cys 1272 (Fig. 3, D and E).
The comparisons to A1/SS reveal the Cys 1272 /Cys 1458 disulfide as an important transducer of A1 conformational change upon complex formation with GPIb␣. Upon complex formation, there is no significant change in the backbone position of Cys 1458 in A1/SS (Fig. 3, E and F), whereas a 2-Å shift occurs in A1/WT, A1/VWD1, and A1/VWD2 (Fig. 3, G-I). The shift at Cys 1458 is enforced by a shift in a similar direction at Cys 1272 of Ͼ2 Å (Fig. 3, G-I). In turn, the shift at Cys 1272 appears to be enforced by a subtle adjustment in the ␣1-␤2 loop backbone at residues Leu 1307 and Arg 1308 ; the Leu 1307 side chain would clash with the Cys 1272 side chain unless Cys 1272 moved (Fig. 3, G-I). In the A1/SS-GPIb␣/VWD2 complex, Cys 1271 moves away from Leu 1307 similarly to Cys 1272 ; however, this movement is accommodated by the extra backbone residue at 1272 in the A1/SS-GPIb␣/VWD complex and does not require movement of Cys 1458 (Fig. 3F).
The Electrostatic Interface-The calculated pI values for GPIb␣ (residues 1-265) and A1 (residues 1267-1465) are 5.5 and 9.2, respectively. Furthermore, their interaction interface is highly electrostatic (12) (Fig. 5). As described above, the orientation differs between the WT complex (Fig. 5A) and the high affinity mutant complexes, exemplified by A1/VWD2-GPIb␣/ VWD2 (Fig. 5B). The change in orientation shifts a cluster of six basic residues in A1 more into an acidic concave surface in GPIb␣ in the high affinity complexes (note the change in position of the non-overlapping portion of the A1 backbone in Fig.  5) and increases charge complementarity. This shift is also observed for A1/SS-GPIb␣/VWD2 (Fig. 5C).
The curvature of LRR proteins varies depending on the structure and number of residues in the portion of each repeat on the convex surface (32). Furthermore, there is some inherent flexibility in a given LRR protein. The curvature of an LRR protein may be characterized by its radius of curvature or the number of degrees subtended by each LRR (32). Flexion in GPIb␣ is more conveniently measured as the distance between LRR1 and LRR8 (Table 5). LRR1-LRR8 distances of 31 and 30.3 Å in the two examples of unliganded GPIb␣ in the 1M0Z structure correspond to total rotations of 85.9 and 94.0°about a central axis, respectively (32). Among six high affinity complexes, this distance is 31.06 Ϯ 0.16 Å (mean Ϯ S.D.) ( Table 5). This distance is lower in the wild-type complex and the wild-type complex with bound botrocetin, at 30.48 Ϯ 0.02 Å (mean Ϯ difference from mean), showing that GPIb␣ in the two wild-type complexes is more concave. Among 11 independent examples of GPIb␣ in isolation or in complex with peptide or ␣-thrombin (Table 5), the distances show much greater variation of 30.48 Ϯ 0.48 Å. The tighter distributions of distances among the high affinity and among the wild-type complexes and the differences between the two distributions suggest that their binding modes constrain GPIb␣ curvature. The greater stretch (lesser concavity) of GPIb␣ in the high affinity complexes enables A1 to come closer to the concave, acidic surface of GPIb␣, increasing electrostatic attraction.
An Interaction with LRR4 and LRR5-The closer approach of A1 to the less concave surface of GPIb␣ in high affinity complexes compared with the low affinity wild-type complex brings the C␣ position of A1 Lys 1371 closer to acidic patch residues GPIb␣ Glu 128 and Glu 151 and nearby polar residues Thr 103 and Gln 127 (Fig. 5, D-G). No hydrogen bond across the interface of Lys 1371 is present in the wild-type complex (Fig. 5D). The density of the Lys 1371 side chain is weak in the previous A1/VWD1-GPIb␣/VWD1 3.1 Å complex crystal structure, and no hydrogen bond of Lys 1371 across the interface was modeled (Fig. 5E). However, Lys 1371 density is well defined in the 2.08 Å A1/VWD2-GPIb␣/VWD2 structure and demonstrates that the Lys 1371 side chain hydrogen-bonds across the interface to Thr 103 and Gln 127 in GPIb␣ LRR4 and LRR5, respectively (Fig.  5F). The 2.8 Å A1/SS-GPIb␣/VWD2 crystal provides four independent examples in the crystal lattice of complex formation. Although there is some variation in A1 Lys 1371 side chain orientation, all four examples show hydrogen bonding across the interface to one or more of GPIb␣ residues 81, 103, 127, and 128. Density is best defined in chains G and H, which show hydrogen bonding of A1 Lys 1371 to GPIb␣ residues 81 and 103 (Fig. 5G). This is the first reported A1 interaction within LRR3-LRR5, which separate the two previously reported discontinuous interfaces in GPIb␣ with A1.
To test the importance of the A1 Lys 1371 hydrogen bond to GPIb␣ Thr 103 and Gln 127 without altering overall charge, we mutated Thr 103 and Gln 127 to Ala. 125 I-GPIb␣/VWD2 binding to yeast displaying A1/SS (Fig. 6A) or A1/VWD2 (Fig. 6B) was competed with GPIb␣/WT or GPIb␣/VWD2 with or without additional alanine mutations. The T103A and Q127A mutations each substantially decreased affinity (Fig. 6, A and B). The Q127A mutation showed the greatest effect, consistent with the close proximity of Gln 127 to nearby acidic residues Glu 128 and Glu 151 .

DISCUSSION
Binding of the A1 domain of VWF to GPIb␣ on platelets is important in arteriolar hemostasis and has the most interesting property of being regulated by hydrodynamic flow. Naturally occurring mutations in the A1 domain in VWD and in GPIb␣ in PT-VWD have been identified that enhance A1 and GPIb␣ interaction and cause disease by gain of function (3)(4)(5). Here, we have used directed evolution in yeast to identify further gain-of-function mutations. Moving the long range disulfide bond to encompass one more residue with the Y1271C/C1272R  FEBRUARY 28, 2014 • VOLUME 289 • NUMBER 9 mutation (A1/SS) increased affinity for GPIb␣ by 10-fold, similarly to the combined effect of two independent VWD mutations, R1306Q and I1309V (A1/VWD2).

Structural Basis of VWF Binding to GPIb␣
We determined multiple crystal structures that extend the resolution of complexes containing VWD mutations from 3.1 to 2.08 Å and extend structural knowledge to complexes with different mutations and with markedly higher affinity. Furthermore, we now have a set of three high affinity mutant complexes (not counting independent lattice environments) in contrast to the single previous example for comparison with the wild-type complex. GPIb␣ with the double G233V/M239V mutation is much more active than either single mutation alone in binding to VWF (30,36). This is in agreement with measurement here of 25-fold higher affinity of G233V/M239V than WT and previous measurements of 3-fold higher affinity of M239V than WT (12). The R1306Q/I1309V double mutant (A1/ VWD2) showed 10-fold higher affinity than WT, as did the A1/SS mutant, whereas single R1306Q and I1309V substitutions increased affinity severalfold, in agreement with a report of a 2.5-fold increase for the R1306Q mutation (12). Altogether, our complexes of A1/SS or A1/VWD2 with GPIb␣/VWD2 show a 250-fold increase in affinity relative to WT, compared with the 7.5-fold increase in affinity of the A1/VWD1-GPIb␣/ VWD1 complex used previously for crystallization (12).
Despite the markedly higher affinity of the two types of double mutant complexes crystallized here, all mutant complexes show remarkably similar orientations between A1 and GPIb␣. This high affinity orientation differs from that seen in the wildtype complex, both in terms of having less concavity in GPIb␣ and allowing LRR4 and LRR5 to approach closer to A1 and in allowing lateral sliding of A1 over the concave GPIb␣ surface with better electrostatic complementarity. A caveat is that we only have one example thus far of the WT complex in crystals.
The closer approach to A1 at LRR4 and LRR5 in the high affinity complexes is consistent with previous evidence implicating this region in shear-regulated increased adhesiveness of GPIb␣ for VWF (15). LRR3 to -5 were previously shown to be required for ristocetin-stimulated binding of VWF to platelets (16). Ristocetin is a glycopeptide antibiotic that adventitiously activates VWF binding and is used to measure VWF function in the diagnosis of VWD. VWD type 2B mutations, two of which were used here in the A1/VWD2 mutant, reduce the concentration of ristocetin required for VWF to bind platelets (37). It is widely believed that ristocetin mimics the effect of shear on VWF (38). Ristocetin has been shown to bind to residues in the N and C terminus of A1 (39). The observations on the importance of LRR3 to -5 in ristocetin-stimulated binding were made by substituting canine for human sequence in this region of GPIb␣, followed by expression in CHO cells co-transfected with the associating GPIb␤ and GPIX subunits (16). Subsequently, the same chimeras were examined for shear-dependent interaction of transfectants with VWF. Remarkably, as shear was increased, having human rather than canine LRR3 to -5 was increasingly important (15). The distance between LRR1 and LRR8 in GPIb␣ is measured as a surrogate for curvature (curvature increases as distance decreases) (32). All distances are given in Å.  For the first time, we observed a structural interaction in this region, between Thr 103 and Gln 127 of GPIb␣ and Lys 1371 of A1. Furthermore, we demonstrated by mutation that Thr 103 and Gln 127 are important for high affinity for A1. We examined conservation of these residues in humans, canines, mice, rats, pigs, cows, and chickens. Lys 1371 is conserved as Lys or Arg in VWF in all of these species. Gln 127 is similarly conserved as Gln or Glu in GPIb␣ in the same species; however, canine is the exception with a His residue, possibly explaining the poor performance of human-canine chimeric GPIb␣ in ristocetin-stimulated adhesion and in adhesion at high shear to VWF (15,16).

Structure
The previous functional assays using canine-human GPIb␣ chimeras, together with a closer approach in LRR of high affinity complexes, and our mutational studies suggest that the high affinity complexes are on a pathway toward the high affinity state induced by shear flow. An increase in shear flow to supraphysiologic levels, as well as transition of shear flow to elongational flow at sites of vasoconstriction in hemostasis, will induce an overall change in VWF concatemer shape from bird's nest to extended and therefore expose multiple A1 domains in VWF for multimeric binding to GPIb␣ on platelets (5,7). The A1 domain is flexibly connected to the neighboring DЈD3 assembly and to the A2 domain by highly flexible, O-glycosylated mucin-like linkers (40). After elongation of VWF, tensile force applied by hydrodynamic flow to VWF concatemers will be communicated through the flexible linkers between domains (9) to the structured portions of the A1 domain that lie N-and C-terminal of its long range disulfide bond. These are the regions in which many VWD mutations are found and in which rearrangements occur in A1 in both high affinity and WT complexes, relative to uncomplexed A1. The hydrodynamic force exerted on the A1 N and C termini could potentially have much larger effects on A1 structure than seen here as a consequence of mutation and complex formation. Moreover, this force would increase after platelet binding to the A1 domain in VWF, as a consequence of the additional hydrodynamic force on the platelet.
Such forces might induce large scale conformational rearrangements in A1, as are seen in the VWA folds present in integrin ␣I and ␤I domains. The presence of long range disulfide bonds does not prevent conformational change in VWA domains, as illustrated by the 15-Å movement of the long range disulfide in the VWA domain of a sporozoite adhesin (41). In integrin I domains, a force exerted in a similar C-terminal direction on the C-terminal ␣-helix induces axial pistoning of this helix that leads to a large increase in affinity for ligand (11). Integrin I domains and the A1 domain differ greatly in the location of their ligand-binding sites, and any hypothetical large scale rearrangements in A1 would necessarily affect affinity through a different mechanism than in integrins. Above, we have argued that high affinity mutant complexes studied here and previously are on-pathway to a super high affinity A1-GPIb␣ state induced by force; however, so as not to discourage further research in this area, it is important to operate under the hypothesis that the force-induced super high affinity state of the A1-GPIb␣ complex has not yet been seen by crystallography.
There are several reasons to think that a high affinity, forceextended state of A1-GPIb␣ has not been seen in the current study. The PT-VWD mutations in GPIb␣ used here are thought to increase the propensity for ␤-strand formation in the ␤-switch (12) and, like other PT-VWD mutations, may also destabilize the unliganded conformation of the ␤-switch relative to the ligand-bound conformation (14). We hypothesize that the ␤-switch has the same conformation in complexes in the presence and absence of force; thus, PT-VWD mutations would not shift the equilibrium between these types of complexes. VWD mutations in A1 locate in or proximal to regions where force would be exerted physiologically and where shape shifting might occur; these mutations are thus hypothesized to shift the equilibrium to the high affinity state. In integrins, cocrystallization with ligand is sufficient to induce the high affinity state, and this state is even sometimes glimpsed fortuitously in a crystal lattice-dependent manner (11). However, single molecule experiments suggest that A1-GPIb␣ binding is not sufficient to induce a high affinity, extended state and that this state requires a force over 10 pN to be exerted on the N terminus of A1 and the C terminus of GPIb␣ and across the receptorligand complex (8). It is currently not possible to mimic such a force in a crystallization experiment. Our VWD R1306Q and I1309V mutations may have lowered the energy to reach such a high affinity state; however, in contrast to one report (42), we find in single molecule experiments with R1306Q and M239V mutations that two states are still observed. 4 Ristocetin selectively strengthens the state seen in high force in single molecule experiments (8). The best-diffracting A1/VWD2-GPIb␣/ VWD2 crystal selected for refinement here was crystallized in the presence of 0.4 mg/ml ristocetin and was further soaked with 4 mg/ml ristocetin. However, we found no electron density for a bound ristocetin, suggesting either that the state we have crystallized is not that stabilized by ristocetin or that lattice contacts prevented ristocetin binding.
Overall, comparisons among complexes and with uncomplexed A1 suggest that strain in VWD mutation-proximal regions occurs in both wild-type and mutant A1-GPIb␣ complexes. The long range disulfide bond moves at the Cys 1272 or Cys 1271 backbone in all complexes and at the Cys 1458 backbone except in the Cys 1271 mutant A1/SS complex. The hydrophobic cavity between the wild-type disulfide and the core of the A1 domain is sealed by a backbone hydrogen bond between Arg 1308 and Cys 1272 in wild-type A1 (Fig. 3D). The cavity becomes exposed to water in the A1/SS mutant, in which a water replaces residue 1272 in hydrogen bonding to the Arg 1308 backbone (Fig. 3E). The movement of the disulfide upon binding GPIb␣ appears to be transmitted from N-terminal-proximal portions of A1 that contact GPIb␣ near its ␤-finger (Fig. 3,  F-I). The dislocation of the disulfide opens up a cavity at the position vacated by Cys 1272 . In the high resolution A1/VWD2-GPIb␣/VWD2 complex, water also gains access to the interface between the disulfide and A1 core (Fig. 3H). Similarly, a cavity formed by the I1309V mutation is occupied by a water (43). Burial of waters in hydrophobic cavities is energetically unfavorable and thus a symptom of strain.
Movement of the long range disulfide appears to result from movement of contacting residue Leu 1307 at the end of the ␣1-helix (Fig. 3, D-I). The ␣1-␤2 loop neighbors Leu 1307 , contains the R1306Q and I1309V VWD mutations studied here in addition to Trp 1313 , and has been extensively discussed previously with respect to a role in affinity regulation (44). The ␣1-␤2 loop occupies multiple conformations in the different structures. Compared with uncomplexed A1, the position of the C␣ atom of residue Q1311 in the ␣1-␤2 loop varies only slightly in A1/SS-GPIb␣/VWD (0.7 Å), more in A1/VWD1-GPIb␣/ VWD1 (1.7 Å), greatly in A1/VWD2-GPIb␣/VWD2 (4 Å), and severely with an overall different loop conformation in A1/WT-GPIb␣/WT (7.5 Å) (Fig. 3, D and F-I). The ␤-finger is variable in conformation in isolated GPIb␣ structures (Fig. 3, D and E) and is sometimes disordered (see structures listed in Table 4).
Remarkably, upon A1-GPIb␣ binding, Leu 1307 , Ile 1309 in the ␣1/␤2 loop, and Cys 1272 (or Cys 1271 ) in the long range disulfide in A1 move in a similar direction away from GPIb␣, despite the many different conformations of the contact sites in the ␣1-␤2 loop and ␤-finger (Fig. 3, G-I). These results are consistent with a clash, rather than a favorable binding interaction, in the ␣1-␤2 loop and ␤-finger region. Clashes can be avoided by different types of structural rearrangements, whereas favorable interactions generally result in a single type of binding interface. The overall binding interaction between A1 and GPIb␣ appears to be largely driven by ␤-sheet formation and burial of hydrophobic and aromatic groups around the ␤-switch and electrostatic complementarity at the A1 interface with the concave LRR surface of GPIb␣. None of the binding interfaces between the ␣1-␤2 loop and ␤-finger region show notable burial of hydrophobic residues or favorable hydrogen bond or electrostatic complementarity (Fig. 3, F-I).
In summary, the structures described here markedly advance our understanding of high affinity A1-GPIb␣ complexes. Strain appears to be induced in the ␣1-␤2 loop, long range disulfide, and VWD mutation-containing regions of A1 upon binding at the interface between the A1 ␣1-␤2 loop and GPIb␣ ␤-finger. We hypothesize that a large conformational change in these regions could dissipate this strain and reshape A1 to form a complementary interface possibly extending across the concave GPIb␣ surface. For these reasons, we wish to keep alive the hypothesis that in structures crystallized to date, the A1-GPIb␣ complex has not yet reached a hypothesized VWD mutationfacilitated, force-extended, high affinity conformation.