Structure and dynamics of the platelet integrin-binding C4 domain of von Willebrand factor

Von Willebrand factor (VWF) is essential to the regulation of hemostasis and thrombosis. At sites of vascular injury, VWF binds to platelets and subsequently promotes their aggregation, thereby mediating adhesion to the damaged vessel wall.¹  In addition, VWF serves as a carrier and stabilizer for coagulation factor VIII in the blood circulation.^1,2 

VWF is a large mosaic protein, composed of 3 type A, 6 type C (VWC), and 4 type D domains, as well as a C-terminal cystine knot domain³  (Figure 1A). These domains are connected with one another like beads on a string, forming a long, linear, rodlike molecule of ∼60 nm in length.⁴  The monomeric units are organized into multimers ranging from dimers to gigantic concatamers of >20 000 kDa,^5,6  via disulfide bonding head to head at the N-termini⁷  and tail-to-tail at the C-termini.⁸  Furthermore, its great length and multidomain composition facilitate VWF to sense and respond appropriately to hydrodynamic forces of the blood flow, as well as the physiochemical environment.^6,9-11  Dimeric VWF, the smallest repeating unit, can adopt both open and flexible, and zipped-up and closed conformations, depending on factors including pH, temperature, presence of divalent ions, and shear force.^10-12  Thus, the conformational states of the minimal dimeric units also regulate the overall length of the concatamers, thereby tuning the hemostatic function of VWF.¹¹ 

Defects or deficiency of VWF can lead to von Willebrand disease, which is the most common inherited bleeding disorder.^1,13  Type IIB von Willebrand disease, although classed as gain-of-function (GOF) due to increased affinity for platelet glycoprotein Ibα (GPIbα), still lead to prolonged bleeding.¹⁴  New GOF variants have recently been identified to be associated with thromboembolism.¹⁵  One such GOF variant is the substitution of F2561 to a tyrosine located in the C4 domain (Y2561).

Despite the well-characterized functions, our current understanding of the molecular structure and thus mechanisms of function of VWF remains poor. The general architecture has been shown at low resolution by electron microscopy.¹⁰  Atomic-resolution structures, however, are only available for the 3 A domains,^16-18  the C-terminal cystine knot domain domain,¹⁹  and the small D′ segment of the D′D3 domain (TIL′E′).²⁰  The entire “stem” region consisting of 6 C domains has not yet been characterized structurally.

The VWF C domain presents one of the most common motifs in extracellular proteins, with 201 VWC domains found in 70 human proteins (http://smart.embl-heidelberg.de/).²¹  The VWC domain is a small, ∼10-kDa domain, typically of 75 to 100 residues in length. Available structures include the VWC domain of Collagen IIA (Col2a),^22,23  CCN3 growth factor,²³  and Crossveinless (CV-2),^24,25  an inhibitor for Bone Morphogenetic Protein signaling. A subdivision of an N- and C-terminal half is common among all 3 VWC structures, generally associated with an extended, flexible orientation. The structures of these domains have revealed a conserved pattern of 5 disulfide bonds with identical connectivity. Based on these findings, and in the absence of any experimental structural data on the C-terminal stem of the VWF protein, predictions of disulfide linkage in the VWF C domains have previously been made by sequence homology.³ 

In this study, the first high-resolution solution structure of a VWF C domain is presented. For structure determination, we have selected the C4 domain, which has a unique ability to bind to platelet integrin α_IIbβ₃ (also known as GPIIb/IIIa).^1,26  The C4 structure exhibits a distinct disulfide pattern, different from other known C domains, and reveals a subdomain arrangement that is unusually kinked. In the structure, the platelet integrin-binding loop is, nevertheless, highly exposed, and qualitative binding data to platelet integrin α_IIbβ₃ confirm its role in the interaction. Our structural findings on the C4 domain suggest that similarly kinked subdomain arrangements from neighboring VWF C domains may have an amplifying effect on the overall VWF C-stem architecture and may explain how this affects binding to platelet integrin.

Human VWF C4 (residues 2497-2577, UniProt P04275, sequence denoted as “wild-type” [wt], F2561), plus an N-terminal extension of GSMA from the thrombin cleavage site, was recombinantly expressed in BL21(DE3) Escherichia coli using the pHAT3 vector,²⁷  which contains an N-terminal His₆-tag. The transformed cells were cultured in isotopically labeled M9 minimal medium at 37°C, induced with 400 μM isopropyl-β-d-thiogalactoside at optical density at 600 nm wavelength of 0.6 to 0.8 for 4 hours. VWF C4 was found in inclusion bodies, which were separated from the soluble fractions by centrifugation upon lysis of the cells. The protein was denatured in 6 M urea and guanidine hydrochloride before refolding by rapid 1:10 dilution into 50 mM sodium phosphate pH 7.5, 0.33 mM 3-[(3-cholamidopropyl)dimethylammonio]-1-propanesulfonate hydrate, 2 mM reduced glutathione, and 0.1 mM oxidized glutathione. Refolded protein was purified first by nickel-nitrilotriacetic acid affinity chromatography, followed by cleavage of the His-tag by thrombin protease, and by reversed phase chromatography (ACE 5 C18-300), before lyophilization and resuspension in 20 mM sodium phosphate pH 6.5. The C4 Y2561 variant gene was generated by a 2-step polymerase chain reaction protocol using overlapping oligonucleotides across the mutated regions, and the respective protein was produced as for the wt C4 domain, at a similar level of expression and thermal stability (supplemental Figure 2A, available on the Blood Web site).

All nuclear magnetic resonance (NMR) experiments were carried out at 298 K (except where stated otherwise) on a Bruker Avance III NMR spectrometer equipped with a cryogenic triple resonance gradient probe head at a magnetic field strength corresponding to a proton Larmor frequency of 600 MHz. All spectra were processed with NMRPipe²⁸  and analyzed using Sparky,²⁹  CARA (http://cara.nmr.ch/doku.php), and CCPNMR.³⁰  For backbone assignment, ¹⁵N-HSQC, HNCACB, CBCA(CO)NH, HNCA, and HNCO spectra were recorded.³¹  Side-chain assignments were done using ¹³C-HSQC, HBHACONH, HCCH-TOCSY, HCCCONH, and CCCONH spectra.³¹  Distance restraints were obtained from ¹⁵N- and ¹³C-edited 3-dimensional NOESY-HSQC, and ¹³C-edited HMQC-NOESY experiments.³¹  Assignment of amide proton/nitrogen backbone resonances of the Y2561 variant were transferred from the wt and confirmed by a ¹⁵N-NOESY-HSQC experiment using sequential H_N-H_α-1 resonances.

The solution NMR structure of the VWF C4 domain was determined by automated nuclear Overhauser enhancement (NOE) crosspeak assignment and torsion angle dynamics, using CYANA 3.0.³²  Distance restraints from the CYANA calculation and TALOS⁺³³  derived dihedral angle restraints were used in a water refinement calculation using ARIA1.2.³⁴  Structure validation of the final ensemble of 20 structures with lowest energies was done using PROCHECK³⁶  and WHATCHECK.³⁵  The structural statistics are shown in Table 1. Structure-based sequence alignment was performed using PDBeFold (http://www.ebi.ac.uk/msd-srv/ssm),³⁷  and geometric analysis using DynDom (http://fizz.cmp.uea.ac.uk/dyndom/).³⁸ 

Measurements of R₁, R₂, and ¹H-¹⁵N heteronuclear NOE experiments for VWF C4 wt and the Y2561 variant were acquired at static magnetic fields corresponding to proton Larmor frequencies of 500, 600, 700, and 800 MHz at 298 K using standard pulse sequences.^39,40  Experiments at 283 K and 310 K were recorded only at 600 MHz. For all R₁ experiments, relaxation delays of 20, 50, 100, 150, 250, 400, 500, 650, 800, 1000, 1300, and 1600 milliseconds were used. Duplicate data points for estimation of uncertainties in peak volumes were recorded for the 20-millisecond relaxation delay. All R₂ experiments were recorded with relaxation delays of 16, 32, 48, 64, 80, 96, 112, 128, 160, 192, 224, and 256 milliseconds. Duplicate data points were recorded for the 16-milliseconds relaxation delay.

To assess the time scale of dynamics for certain residues Carr-Purcell-Meiboom-Gill (CPMG),^41,42  relaxation dispersion experiments at 298 K were recorded at 600 and 800 MHz using a previously published pulse sequence.⁴³  Effective fields, defined as ν_CPMG = 1/(4τ_CP), where 2τ_CP is the interval between refocusing pulses of 33, 66, 100, 133, 200, 266, 333, 400, 466, 533, 600, 666, 800, and 933 Hz, were used. Peak intensities were transformed to effective transverse relaxation rates according to R₂,_eff = ln(I₀)/I(ν_CPMG)/T, where I(ν_CPMG) is the intensity for different values of ν_CPMG, I₀ is the intensity without the CPMG sequence block, and T is the duration of the constant time relaxation delay (30 milliseconds). PINT⁴⁴  was used for peak integration and fits of spin relaxation and CPMG data.

ROTDIF⁴⁵  was employed to determine the diffusion tensor from the R₂/R₁ ratios. The data were fitted better to the axially symmetric model than the isotropic model. Using an anisotropic model did not improve the fit. For calculation of the rotational correlation time, residues with heteronuclear NOE values <0.65 were excluded from the analysis.

Simulations employed a previously published protocol⁴⁶  using GROMACS 5.1.x⁴⁷  and the CHARMM22* all-atom force field.⁴⁸  The Verlet scheme was used with standard twin-range 10- to 12-Å cutoffs, utilizing Particle-Mesh Ewald electrostatics and Force-switching van der Waals interactions. Time steps were set at 4 fs using virtual site construction. For the wt C4 domain, each of the 20 NMR conformations was subjected to the following equilibration and production protocols: solubilization in an 8-nm dodecahedron box with periodic boundaries containing 150 mM NaCl, 2500-step energy-minimization by steepest descent, equilibration over 2 nanoseconds using gradually relaxing position restraints, first over the side chains and then over the backbone, and finally 100-nanosecond production simulations. All simulations were maintained under constant number, pressure, and temperature conditions, using velocity-rescaling thermostats (τ_t = 2.5 ps) and Berendsen barostats (τ_p = 5.0 ps) during equilibration and then switching to Nosé-Hoover thermostats and Parinello-Rahman barostats during production. The Y2561 variant was derived from the 20 wt structures by manual mutation in visual molecular dynamics (MD).⁴⁹  Each mutant was equilibrated identically to the wt, followed by 200-nanosecond production simulations, with the first 100 nanoseconds discarded. The computation of NMR spin-relaxation from the 20 × 100 nanoseconds ensembles follows procedures described in Chen et al,⁵⁰  except where the global tumbling of ROTDIF was used to replace simulated values before fitting to experiment.

Human α_IIbβ₃ was immobilized, and C4 domain variants were in solution. Details of experiments are described in supplemental Methods.

The C4 domain structure, covering residues S2497 to E2577, of the most frequently occurring human VWF sequence variant (F2561), was determined by solution NMR spectroscopy (Figure 1C-D;,Table 1; supplemental Figure 1). The backbone resonances of all residues were assigned except for S2568, A2569, and C2570, which indicate local motions in the intermediate exchange regime. This leads to a total assignment of 96.8% of all nuclei including side chains.

The overall C4 fold is divided into 2 subdomains, SD1 (S2497-N2546) and SD2 (S2548-E2577), marking V2547 as hinge residue (Figure 1C). SD1 is composed of 2 antiparallel β-sheets. The first one comprises 2 short β-strands I′-II′ connected by a long 10-residue hairpin loop, with the α_IIbβ₃ integrin binding arginine-glycine-aspartatic acid (RGD) motif (residues 2507-2509) at its tip. The absence of medium- and long-range NOEs at this motif and thus larger conformational heterogeneity between the 20 lowest energy structures of the NMR ensemble suggests high mobility (Figure 1D left). This has been further confirmed by ¹⁵N relaxation experiments, showing that the RGD loop region exhibits very fast dynamics in the nanosecond-picosecond time scale indicated by decreasing R₂/R₁ and heteronuclear NOE values (Figure 2). The second SD1 β-sheet comprises 3 antiparallel strands I-II-III. Together, the 2 SD1 β-sheets form a 2-layered structure. Interestingly, the interface of the 2 β-sheets is devoid of any specific hydrogen-bond interactions. The most prominent hydrophobic interaction is established between V2501 from β-strand I′ of the first sheet and W2521 from β-strand I of the second sheet. The C-terminal SD2 comprises one 2-stranded antiparallel β-sheet I-II. The sheet is preceded by a long sequence stretch without any regular secondary structure, connecting the C-terminus of SD1 with the first strand of the SD2 β-sheet.

The C4 structure comprises 5 disulfide bridges (Figure 1C; Table 2). Chemical shift values of C_β resonances clearly indicated that all cysteines are involved in disulfide bonds⁵¹  (supplemental Table 1). Three disulfides are formed by cysteines within 1 of the 2 subdomains SD1 (bridge 1) and SD2 (bridges 5 and C4) and thus are thought to be fold determinants of these subdomains. The latter disulfide bridge (C2565-C2574) is specific to the VWF C4 domain, hence named “C4” bridge (further details below). The 2 remaining disulfide bridges connect the 2 subdomains (bridges 2 and 4) and thus appear to be determinants of the observed SD1/SD2 arrangement.

Comparison of the 20 lowest energy conformers unambiguously identifies a short residue segment centered on V2547 as a flexible hinge, allowing limited flexibility in the orientation between SD1 and SD2. In order to confirm these observations experimentally, ¹⁵N relaxation experiments were performed to obtain a residue-wise quantitative picture (Figure 2; supplemental Figure 3). The most prominent peak in the relaxation profile is found for V2547, which confirms that this residue undergoes slow conformational exchange in the microsecond to millisecond time scale. The exchange contribution R_ex to R₂ is considerable and ranges from 7.7 s⁻¹ to 19.7 s⁻¹ at 500 to 800 MHz, respectively. To assess the time scale of these exchanges, CPMG experiments at 600 and 800 MHz were performed (supplemental Figure 4). As the dispersion profile for residue V2547 and neighboring residues is flat, but retains high R₂ values throughout the CPMG frequencies used, we conclude that the conformational change is in the low microsecond range.

The consistently lower R₂/R₁ values for SD2 indicate higher global mobility compared with SD1 (Figure 2A-B). Due to the flexible hinge, SD2 presumably moves with respect to SD1 and samples different orientations. The global rotational correlation time τ_c of the C4 domain calculated from R₂/R₁ values is 7.5 nanoseconds, which agrees well with the value predicted from the NMR ensemble by ROTDIF (7.9 nanoseconds; Figure 2C). Interestingly, the dynamics of the hinge region and especially residue V2547 vary considerably upon temperature changes (supplemental Figure 5). At an elevated, physiological temperature (37°C), the flexibility is increased, and there is no exchange contribution to R₂. Dynamics are even slower at 10°C compared with 25°C, indicated by the broadened peak of V2547, which vanishes into the noise. In summary, these NMR analyses demonstrate microsecond-level flexibility between SD1 and SD2 domains that is highly sensitive to temperature variations.

To assist in interpreting the NMR dynamics, multiple MD simulations were conducted to examine possible dependence of domain dynamics upon mutation (Figure 3). Two angular parameters were formulated to define the relative SD1-SD2 arrangements observed. The dihedral twist angle φ is defined by the centers of mass of SD1 and SD2, plus the centers of mass of V2547 and C2528. The complementary tilt angle θ is defined by θ = 180° + θ₁ − θ₂, where θ₁ is defined by the angle (center of gravity SD1)-C2547-C2528 and θ₂ C2547-C2528-(center of gravity SD2) (Figure 3A inset). When both angles φ and θ are 180°, the SD1/SD2 arrangement is fully extended. The more both angles deviate from 180°, the more the arrangement becomes twisted and tilted. As a result of the MD simulation, we found a skewed tilt distribution centered at θ ∼190° with a θ range of 140° to 230° (Figure 3A left). The twist angle distribution is more bimodal, where the more populated one peaks at φ ∼170° (cluster I) and a less populated one (cluster II) in a φ angle range of 250° to 290°. In terms of an overall SD1/SD2 arrangement, the second cluster II is significantly more bent than the first cluster I (Figure 3B).

The most pronounced function of the C4 domain, contrary to all other VWF C domains, is to bind platelet integrin α_IIbβ₃.¹  To confirm that the C4 domain construct we used for structure determination and MD simulations is functional, we used a solid-phase binding assay to measure its interaction with α_IIbβ₃ (Figure 1B). The wt C4 domain binds to integrin α_IIbβ₃ with a half-maximal binding of 200 nM. As expected, C4 domain mutants, in which the 2 integrin binding residues of the RGD motif were substituted either by alanines or glutamates, lost the ability to bind integrin α_IIbβ₃.

Using the wt C4 domain (F2561) as reference, we set out to probe possible structural effects of the Y2561 variant. This VWF variant leads to a significant increase of myocardial infarction risk.¹⁵  The corresponding residue is located at the distal face of SD2, close to the C-terminus of the C4 domain, with its side chain presented on the surface (Figure 1C). The overall structure of this variant is not perturbed, as only minimal chemical shift differences of ¹H-¹⁵N resonances can be observed (supplemental Figure 1), whereas the ¹⁵N relaxation profile also exhibits no significant changes (Figure 2A). Nevertheless, a slight increase of the global rotational correlation time can be observed (Figure 2A,C). This could be attributed to a higher oligomerization tendency for the Y2561 variant, which also explains the dependence of τ_c upon increasing concentrations. The change of F2561 to tyrosine also affects the R_ex of V2547, indicated by altered magnetic field dependence (Figure 2C-D).

Using the same MD approach employed for the wt C4 domain, we show that the 2-microsecond ensembles of the Y2561 C4 variant are broadly similar: although we observed only minor changes in terms of SD1/SD2 tilt, we found a third minor population in the twist analysis peaked at φ ∼115°, denoted as cluster III (Y2561) (Figure 3A-B right). This cluster has a significantly bent SD1/SD2 arrangement similar to cluster II, which has been found in both the wt C4 and Y2561 C4 variant domains. Calculated relaxation profiles also confirmed that the mutation does not appear to affect the picosecond-nanosecond behavior of the 2 C4 subdomains (Figure 2B). Thus, aside from potential contributions from minor conformations likely associated with hidden conformational changes suggested by the R_ex of V2547, we found no direct influence of the Y2561 variant on C4 dynamics itself. The Y2561 variant also binds to integrin α_IIbβ₃ similarly to wt C4 domain (Figure 1B), demonstrating that it does not impair the RGD platelet integrin binding site.

In this work, we present the first structure of a VWC domain of the VWF protein, from which the name of this domain was derived.⁵²  Key findings, different from previous structures, are its peculiar, in part noncanonical disulfide pattern, and the presence of an exposed and flexible loop region containing the RGD platelet integrin-binding motif, accessible for binding to integrin (supplemental Figure 2C). The structure of the C4 domain also indicates that the GOF Y2561 variant¹⁵  is unlikely to be associated with an alteration of platelet integrin binding. The structural position of this variant is close to the C-terminus of the C4 domain (Figure 1C), suggesting that instead it could play a role in the interface with the succeeding neighboring domain C5 thereby affecting shear-induced VWF self-association. Proof of such model will require additional structural data on larger VWF fragments covering both C4 and C5 domains.

To determine unique features of the VWF C4 domain in relation to other VWC domains, we compared the arrangement of its 2 subdomains SD1 and SD2 with those of known VWC structures: Collagen IIA (Col2a),^22,23  CCN3,²³  and CV-2.^24,25  As the basis for structural comparison, we used a total of 24 residues (Figure 4A boxed segments) covering the 3 β-strands I-II-III of the second larger SD1 β-sheet for superimposition with that of the crystal structures of Col2a, CCN3, and CV-2. Only these 3 segments are sufficiently structurally conserved to provide a reliable structure-based sequence alignment (Figure 4A,C). There are 2 conserved cysteines within the second boxed segment that are both engaged in disulfide bridges, indicating a conserved role in VWC domain fold determination. In contrast, such structure-based alignment is not possible for SD2 due to the lack of sufficiently conserved structural elements and sequence.

The long hairpin loop comprising the RGD platelet integrin-binding motif is unique in the C4 domain (Figure 4A,C right). In contrast, the β-strands of the first sheet I′-II′ in Col2a and CCN3 are each connected by a short 2-residue hairpin only. The difference between the C4 domain and the structures of Col2a and CCN3 is further reflected by how the 2 SD1 β-sheets are packed with each other. While in Col2a and CCN3, there is a π/π-interaction from 2 aromatic residues in the 2 SD1 β-sheets²³ ; in C4, the equivalent interaction is formed by V2501 (βI′) and W2521 (βI) (Figure 1C; turquoise highlights in Figure 4A). We note that in the CV-2 structure, the N-terminal part of SD1 is unrelated and partially unfolded.

A key signature of the VWC domains lies within their partially conserved pattern of disulfide bridges (Figure 4A-B; Table 2). Although the first of the 2 canonical intra-SD1 disulfide bridges is present in all 4 known VWC structures, the second one (number 3) is missing in the VWF C4 domain. Interestingly, the 2 β-strands without the canonical disulfide bridge are connected by an extended, charged loop in the VWC C4 domain (residues 2536-2539, indicated by an arrow in Figures 1C and 4C) that is missing in the other 3 structures of Col2a, CCN3, and CV-2. We notice that this loop is on the opposite side of the C4 platelet integrin-binding site (Figure 1C). Considering a zipped stem arrangement of the C-terminal VWF segment as inferred from available low-resolution data¹⁰  (Figure 1A), this loop may be in an attractive position to contribute to zipped VWC dimerization. Along with the additional intra-SD2 disulfide bond (number C4), these unique structural features in the C4 domain may impact the conformational stability of both SD1 and SD2, which merit further investigation in expanded structural studies.

The 2 SD1/SD2 connecting disulfide bridges (numbers 2 and 4), although being in related sequence positions in the structure-based alignment, are in structurally quite divergent positions in the C4 domain when compared with the other VWC structures (Figure 4C zoomed-in inset). Taking the data together, our findings indicate that limited sequence and structural variability of an otherwise conserved disulfide pattern are main determinants of SD1/SD2 arrangement differences in the VWC structures. In the VWF C4 domain the focal point of this hinge can be localized to residue V2547, which is close to the disulfide bridge number 4 (Figure 1C). To determine whether this observation is part of a general mechanism, it would be interesting to confirm the existence of similar focal SD1/SD2 hinge residues in the related VWC structures by measuring their dynamics as well.

For a more quantitative comparison, equivalent SD1/SD2 hinge and complementary angles were computed the same way as for the C4 MD analysis (Figure 3), including the deposited lowest-energy NMR ensembles of Col2a and CV-2 (Figure 4E). In the absence of a subsequent MD analysis available for those NMR structures, we limit our further comparison with the 3 known crystal structures, keeping in mind that their arrangements may be artificially rigidified due to crystal lattices. Although the CCN3 crystal structure displays a twist angle of ∼225° leading to a right-handed kinked SD1/SD2 arrangement, the Col2a structures reveal an opposite left-handed bent with a twist angle of <150°. In contrast, the SD1/SD2 arrangement of CV-2 is almost entirely extended with a twist angle of close to 180°. Naturally, the angles observed in the different NMR ensembles are more diverse (Figure 4E), as long distances between SD1 and SD2 are less defined.

In our MD simulations, an extended SD1/SD2 arrangement of the C4 domain is most populated, with twist angles deviating <10° from a 180° angle (region I in Figure 3A-B). The presence of neighboring VWC domains (C1-C3 and C5-C6) may indeed impact the population of different SD1/SD2 arrangements, and hence, dimeric stem formation and platelet integrin accessibility.¹⁵ 

Ultimately, high-resolution structures of multidomain stretches of the VWF C-segment are needed for further elucidation of the molecular requirements for parallel dimerization, integrin binding, and the effect of external shear forces in these processes. Such studies could also unravel further insights into previously established differences in the predicted disulfide pattern of odd- (1, 3, 5) and even- (2, 4, 6) numbered VWF C domains,³  and how they arrange in possible tandems (1/2, 3/4, 5/6) with very short linkers in between.

The authors are grateful to Kai Horny, who assisted in calculations, and thank Remco Sprangers for sharing the CPMG pulse sequence.

E.-R.X., K.K., C.A.-S., F.G., R.S., and M.W. acknowledge the SHENC research grant (FOR1543) from the Deutsche Forschungsgemeinschaft. P.-C.C. is supported via the EI₃POD program by EMBL and the EU Marie Curie Action Cofund grant 664726.

Contribution: E.-R.X. conceived the study, produced the protein, analyzed the structure, and wrote the manuscript; S.v.B. and J.H. performed the NMR experiments, structure calculation, and dynamics analysis; P.-C.C. conducted MD simulations, NMR spin-relaxation computations, and contributed to manuscript writing; P.J.L. and C.V.D. performed solid-phase binding assays; K.K., C.A.-S., and F.G. conducted MD simulations and dihedral angles analysis; B.S. optimized and set up the CPMG pulse experiment and did water refinement; J.F. conducted HSQC experiments; T.O. provided the VWF gene; F.G., R.S., and M.W. initiated the study; and M.W. and J.H. designed and supported the study and wrote the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Matthias Wilmanns, European Molecular Biology Laboratory, Hamburg Unit, Notkestrß 85, 22607 Hamburg, Germany; e-mail: matthias.wilmanns@embl-hamburg.de; and Janosch Hennig, European Molecular Biology Laboratory, Meyerhofstrß 1, 69117 Heidelberg, Germany; e-mail: janosch.hennig@embl.de.