Closed headpiece of integrin α_IIbβ₃ and its complex with an α_IIbβ₃-specific antagonist that does not induce opening

Integrins are cell surface–adhesion receptors that contain noncovalently associated α and β subunits, each with a single-pass transmembrane domain and a typically short cytoplasmic domain.^1,2  Ectodomains are structurally complex, with 4 or 5 domains in α and 8 in β. The α and β subunits come together in the integrin head to form the binding site for ligands. Crystal structures of full-length ectodomains of integrins α_Vβ₃, α_IIbβ₃, and α_Xβ₂ have all revealed a compact, bent conformation.^3-6  The α and β subunits are each acutely bent at their knees, between the upper- and lower-leg domains (Figure 1A). The bend brings head and upper-leg domains (collectively termed “the headpiece”) into intimate contact with the lower-leg domains. An extensive, largely hydrophilic surface of > 4000 Ų  between the headpiece and lower legs, and between the α and β legs, becomes solvent exposed on integrin extension (Figure 1A). A large body of crystallographic, electron microscopic (EM), and mutational evidence shows that the bent conformation represents the low affinity, resting state of integrins,⁷  although some studies have nonetheless reached different conclusions.^3,8,9 

Two types of fundamental conformational changes have been found in integrins based on EM and crystallographic data: (1) integrin extension at the knees and (2) swing out at the interface between the β subunit βI and hybrid domains, converting the headpiece from the closed to the open conformation (Figure 1A). The C-terminal α7-helix of the βI domain links conformational changes at the interface with the hybrid domain to the opposite end of the βI domain, changing conformation and affinity at the ligand binding site at the interface with the α subunit β-propeller domain. Integrin affinity up-regulation is intimately associated with the transition from the closed to open headpiece conformation.⁷ 

Previously, complete ectodomain and headpiece integrin crystal structures have revealed different headpiece conformations. In the complete ectodomain crystal structures, integrin headpieces adopt a closed conformation.^3-6  Before the current study, the only crystal structures of an integrin headpiece fragment, those of α_IIbβ₃, were found in open conformation. All of these structures contained an α_IIbβ₃ ligand, ligand mimetic, or pseudoligand that coordinated the metal ion–dependent adhesion site (MIDAS) Mg²⁺, supporting the hypothesis that ligand binding induces and/or stabilizes the open/swung-out conformation.^10,11  The possibility has been raised, however, that the open headpiece conformation was not the result of ligand binding, but rather a loss of the lower legs and associated headpiece-tailpiece interactions.¹²  We considered this hypothesis unlikely, however, because ligand and ligand-mimetic drugs induce the open headpiece conformation in both headpiece fragments and the entire ectodomain, as demonstrated by small angle x-ray scattering and EM.^1,7,13-16  In fact, EM studies show that when the entire integrin ectodomain extends, individual molecules can have either a closed or open headpiece.^15,17  Nonetheless, a crystal structure of a headpiece fragment in closed conformation would exclude the possibility that the open conformation was an artifact of truncation. Furthermore, because the hybrid domain has extensive contacts with lower-leg domains in the bent conformation^4-6  (Figure 1A), in the absence of such restraints in a headpiece fragment, it would be interesting to know if atomic details at the ligand binding site and the βI/hybrid domain interface would differ from those seen in the bent conformation. Moreover, if such a closed conformation lacked a ligand or pseudoligand, it would further add to the correlation between ligand binding and headpiece opening. In addition, because controversy remains as to whether integrins bind metal ions before or after ligand binding,^5,6,18  the occupancy of metal binding sites would also be of interest.

Highly expressed on platelets, integrin α_IIbβ₃ plays an essential role in the maintenance of hemostasis and the formation of pathologic platelet-mediated thrombi.¹⁹  α_IIbβ₃ recognizes an Arg-Gly-Asp (RGD) sequence in ligands as well as the Lys-Gln-Ala-Gly-Asp-Val (KQAGDV) sequence of the fibrinogen γC module.^10,20  Two FDA-approved RGD-mimetic α_IIbβ₃ antagonists, tirofiban and eptifibatide, are currently in use for the treatment of select thrombotic complications of cardiovascular disease.^21,22  Cocrystallization of these drugs with the integrin α_IIbβ₃ headpiece revealed that they bind in the same way as peptides containing the RGD sequence.^10,11  The basic Arg side chain or nitrogen-containing drug moiety binds to Asp-224 in a pocket of the α_IIb β-propeller domain whereas the acidic Asp side chain or a carboxyl moiety in the drug directly coordinates the MIDAS Mg²⁺ ion in the β₃ βI domain.^11,23  Ligand binding stabilizes the open headpiece conformation, which is characterized by a coordinated series of movements, including: (1) movement of the βI domain β1-α1 loop toward the ligand, (2) movement of the β6-α7 loop “downward,” and (3) piston-like movement of the α7-helix downward, causing a large swing-out movement of the hybrid domain, which is visible as the open headpiece conformation at EM and small angle x-ray scattering resolution.^14,15,17 

Conformational changes in the α_IIbβ₃ receptor on the platelet surface in patients treated with RGD-mimetics have potentially important implications for therapy.^24-31  Preexisting or neoantigen-induced drug-dependent antibodies to α_IIbβ₃ may be responsible, at least in part, for the thrombocytopenia observed in as many as 5% of treated patients.^29-31  Furthermore, a paradoxical increase in the risk of thrombosis was observed in association with treatment using some orally active, RGD-mimetic α_IIbβ₃ antagonists.³²  In some cases, autoantibodies that can bind to ligand-induced binding site (LIBS) epitopes can activate patient platelets by simultaneously engaging the platelet Fc receptor.³⁰  In the majority of cases, however, the etiology of thrombosis is unclear. One possible explanation is that RGD-mimetic antagonists induce the high-affinity state of the receptor, thus priming it to bind ligand after drug dissociation from the receptor.^32-34  Ligand binding could then support platelet aggregation and potentially contribute to an increase in cardiovascular events.^32,33 

A novel small molecule (M_r 265) α_IIbβ₃ antagonist termed RUC-1 is effective in inhibiting platelet aggregation in vitro as well as thrombus formation in experimental models in vivo.^35,36  RUC-1 is specific for α_IIbβ₃ relative to α_Vβ₃. It demonstrates little or no α_IIbβ₃-priming activity and induces little or no exposure of the β₃ LIBS epitope recognized by monoclonal antibody AP5. Preliminary molecular docking studies suggested that, in contrast to RGD-mimetics, which bind to both the α_IIb and the β₃ subunits, RUC-1 binds exclusively to the α_IIb subunit.

We report the first crystal structure for an integrin headpiece fragment in the absence of a peptide ligand, ligand mimetic, or pseudoligand. Despite the absence of leg domains, the headpiece is in closed conformation. The crystal structure in the presence of RUC-1 confirmed that RUC-1 binds exclusively to α_IIb and provides the atomic details of its binding site. RUC-1 binding is not associated with remodeling to the high-affinity state, as further confirmed by gel filtration and dynamic light scattering (DLS). Molecular dynamics (MD) simulations and mutagenesis provide additional support for the proposed RUC-1 binding mechanism.

The soluble α_IIbβ₃ headpiece construct contains residues 1-621 of α_IIb and 1-472 of β₃ (Figure 1B).¹¹  The α_IIbβ₃ headpiece was stably expressed in CHO Lec.3.2.8.1 cells and purified with a Ni-NTA column (Figure 1B). After chymotrypsin digestion and a second Ni-NTA column purification, it was further purified by gel filtration (Figure 1C). The thigh domain of the α_IIb subunit remained after chymotrypsin digestion based on the molecular weight (∼67 kDa) deduced by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (Figure 1C). This untagged α_IIbβ₃ headpiece was used for gel filtration and DLS studies, as described below. The untagged α_IIbβ₃ headpiece-10E5 Fab complex was purified by gel filtration and treated with carboxypeptidase A and B in the presence of 1mM Zn²⁺, and purified by gel filtration in Tris-buffered saline with 1mM Ca²⁺/Mg²⁺. Carboxypeptidase treatment removed the α_IIb-thigh domain, decreasing the molecular weight to ∼ 50 kDa (Figure 1D). There was no molecular weight change for the β₃ subunit after carboxypeptidase treatment (Figure 1C-D). The α_IIbβ₃/Fab complex was concentrated in Tris-buffered saline with 1mM Ca²⁺/Mg²⁺ to 10 mg/mL and crystallized in 11% PEG 8000, 0.2M ammonium sulfate, 0.1M Tris-HCl, pH 8.9 at 4°C. RUC-1 was soaked into the α_IIbβ₃/Fab crystals at 500μM in the well solution containing 1mM Ca²⁺/Mg²⁺ for 2 days. Crystals were harvested in 15% PEG 8000, 0.2M ammonium sulfate, 0.1M Tris-HCl, pH 8.9 plus 1mM Ca²⁺/Mg²⁺ with glycerol as a cryoprotectant in 5% increments up to a 20% final concentration, and then flash-frozen in liquid nitrogen. We also soaked crystals with 200μM quinine in 15% PEG 8000, 0.2M ammonium sulfate, 0.1M Tris-HCl, pH 8.9, with 1mM Ca²⁺, 5mM Mg²⁺, and 20% glycerol for 3 days before freezing in liquid nitrogen. Because no electron density of quinine was observed, the crystal was considered a native crystal at a higher Mg²⁺ concentration. Diffraction data collected at ID-23 of APS was solved using molecular replacement. Final refinement with REFMAC5 used translation, libration, screw motion as well as noncrystallographic symmetry.

The chymotrypsin-treated and purified α_IIbβ₃ headpiece at 5.6μM was incubated with RUC-1, eptifibatide, or tirofiban at 600 μM, 120μM, and 56μM, respectively, at 25°C for 1 hour and analyzed by chromatography (Superdex 200; GE Healthcare Life Sciences) in Tris-buffered saline with 1mM Ca²⁺/Mg²⁺.

Stokes radii of the chymotrypsin-treated and purified α_IIbβ₃ headpiece alone at 22μM or after mixing with RUC-1, eptifibatide, or tirofiban at 500μM, 240μM, and 56μM, respectively, were measured at 25°C by Viscotek 802 DLS (Viscotek Corporation) in Tris-buffered saline with 1mM Ca²⁺/Mg²⁺.

MD simulations of RUC-1 alone in a water environment or bound to normal human α_IIbβ₃ were carried out using Amber 10 suite of programs (http://ambermd.org), as described in the supplemental data (available on the Blood Web site; see the Supplemental Materials link at the top of the online article). The well-tempered metadynamics algorithm^37-39  was used to enhance conformational sampling of RUC-1 in the binding pockets of normal or mutant human α_IIbβ₃.

Human full-length α_IIb Y190F and D232H mutant cDNAs were generated by site-directed mutagenesis (QuickChange XL Site-Directed Mutagenesis Kit; Stratagene). HEK 293 transfectants were selected in 800 μg/mL G418. The cells expressing high levels of α_IIbβ₃ were sorted based on binding of Alexa 488-conjugated monoclonal antibody (mAb) 10E5⁴⁰  (BD FACSCalibur; BD Biosciences) and expression of α_IIbβ₃ was assessed by the binding of the same antibody on each day of experimentation.

For the Alexa 488-fibrinogen binding assay, HEK293 cells were incubated in HEPES buffer-modified Tyrode solution with or without 15 μg/mL activating antibody PT25-2⁴¹  for 5 minutes at 25°C and then Alexa 488 fibrinogen (200 μg/mL; Invitrogen) was added and incubated at 37°C for 30 minutes. Samples were then washed once and resuspended in 0.5 mL of HEPES-buffered modified Tyrode solution containing Ca²⁺/Mg²⁺ and analyzed by flow cytometry. Net normalized fluorescence intensity (NNFI) was calculated from the geometric mean fluorescence intensity of the cells after subtracting background fluorescence and dividing by the α_IIbβ₃ receptor expression level determined by the binding of mAb 10E5. The IC₅₀ of RUC-1 inhibition of fibrinogen binding to normal or mutant α_IIbβ₃ was determined after log transformation and removal of outlier values. Each NNFI value was divided by the NNFI of the PT25-2–activated sample in the absence of RUC-1. The latter sample and the maximally inhibited sample in the presence of the blocking mAb 10E5 were used to define the range of RUC-1 concentrations (0 and infinity). Mean (SD) was computed at each concentration level for normal α_IIbβ₃, α_IIb Y190F mutant and α_IIb D232H mutant. For the data with each receptor, a sigmoidal model was fitted on mean (SD) with log transformation on RUC-1 concentration to estimate the IC₅₀ for fibrinogen binding (BioDataFit 1.02; Chang Bioscience Inc).

Soluble headpiece protein was purified by Ni-NTA–affinity chromatography (Figure 1B right panel) and treated with chymotrypsin to remove C-terminal tags (Figure 1C). The untagged protein was digested by carboxypeptidases A and B in the presence of mAb 10E5 Fab and finally purified by gel filtration (Figure 1D). Carboxypeptidase, together with trace amounts of chymotrypsin (data not shown), removes the α_IIb thigh domain (note the change in α_IIb size in sodium dodecyl sulfate–polyacrylamide gel electrophoresis between Figure 1C-D).

The purified headpiece/Fab complex in Tris-buffered saline containing 1mM Mg²⁺ and 1mM Ca²⁺ was crystallized in 11% PEG 8000, 0.2M ammonium sulfate, and 0.1M Tris-HCl, pH 8.9 at 4°C, which gave the 2.3 Å resolution structure (native-1; Table 1). Crystals were also soaked in 5mM Mg²⁺ and 1mM Ca²⁺, which gave the 2.25 Å resolution structure (native-2; Table 1). Both structures were refined to an R_free of 21.3% (Table 1). Two independent α_IIbβ₃ headpiece/Fab complexes were present per asymmetric unit (Table 1).

The structures contain the α_IIb β-propeller domain, the β₃ plexin/semaphorin/integrin (PSI), hybrid, βI, and I-EGF1 domains, and the 10E5 Fab bound to the β-propeller cap subdomain (Figure 2A). Our previous crystal structures of the α_IIbβ₃ headpiece were obtained with a similar protein preparation in complex with ligand-mimetics or a pseudoligand, with or without 10E5 Fab.^10,11  Four molecules in unique lattice environments, 3 per asymmetric unit in the absence of Fab, and one per asymmetric unit in the presence of Fab, all adopted similar open headpiece conformations (Figure 2B). In our current headpiece structures, density for a pseudoligand was not found around the ligand-binding site; only waters surround the MIDAS Mg²⁺ (Figure 2D). The conformations of the β-propeller and the 10E5 Fab are essentially identical to those in the previous structures (Figure 2B). However, in contrast to the open conformation adopted by the liganded headpiece fragments, the unliganded α_IIbβ₃ headpiece fragments adopt the closed conformation (Figure 2B). The orientation between the βI and hybrid domains is acute, as seen in bent ectodomain structure rather than obtuse, as seen in liganded, open headpiece structures (Figure 2B).

In our closed headpiece structures, the conformation of the βI domain, including the metal ion binding sites, the β1-α1 loop, the α1-helix, the β6-α7 loop and the α7-helix, is essentially identical to the conformation seen in the bent ectodomain structures (Figure 2C). Very clear densities are seen for the Ca²⁺ in the SyMBS, the Mg²⁺ in the MIDAS, the Ca²⁺ at the ADMIDAS, and their coordinating waters in the native-2 structure in 5mM Mg²⁺ and 1 mM Ca²⁺ (Figure 2D). Similar results were obtained in the native-1 structure in 1mM Mg²⁺ and 1mM Ca²⁺, except the densities for the MIDAS Mg²⁺ and its coordinating waters are poorly defined.

Some variation in βI/hybrid domain orientation is seen among α_IIbβ₃ and α_Vβ₃ structures with closed headpieces, and among structures with open headpieces (Table 2). The angle between the βI and hybrid domains was ∼ 12° wider in closed headpiece structures than in bent ectodomain structures (Table 2). This result was because of slight libration of the last C-terminal turn (Gly-349-Ser-353) of the α7-helix and its connecting loop to the β-strand of the hybrid domain (Figure 2C). The last turn of α-helices hydrogen bond only to preceding, and not succeeding, α-helical residues. Within the last turn of the α7-helix, beginning with Gly-349, small variations in the hydrogen bond pattern within this helical turn, and the subsequent loop permit slight rocking at the βI/hybrid interface. Similar variations (∼ 12°) in βI/hybrid domain orientation were observed among the open headpiece structures in different crystal forms (Table 2), and these also correlated with libration of the last turn of the α7-helix and the following loop to the β hybrid domain. This variation is much less than that between the open and closed headpiece (Figure 2B and Table 2).

The slightly swung-out position of the hybrid domain in the closed headpiece fragment structure with respect to that in the entire ectodomain structure was associated with some remodeling of the hydrogen-bonding network at the β₃ β I/hybrid domain interface (Figure 2E-F). In particular, the hydrogen bonds between the Ser-300, Arg-360, and Asp-358 side chains, and between the Lys-417 and Asn-303 side chains, were broken (Figure 2E-F). The Arg-352 side chain, which is central in the interface, adopted a different orientation, forming new hydrogen bonds with side-chain oxygens of Tyr-348 and Gln-106 (Figure 2E-F). In addition, the Lys-422 side chain formed a hydrogen bond with the Asp-241 side chain. The hydrogen bonds between the Asp-109 side chain and the Ser-147 side chain and backbone, and between the Ser-353 backbone oxygen and the Gly-388 and Leu-389 backbone nitrogens, were maintained (Figure 2E-F). The positions of Tyr-348 and Tyr-110, which are central in the interface, were not changed (Figure 2E-F).

A crystal soaked with RUC-1 diffracted to 2.4 Å (Table 1) with clear density for RUC-1 in each of the 2 molecules of the asymmetric unit (Figure 3A-B). RUC-1 binds exclusively to the ligand-binding pocket formed by the α_IIbβ-propeller domain (Figure 3A-B). In sharp contrast, the RGD-mimetic drug, tirofiban, binds to both the α_IIb β-propeller and β₃ βI domain^10,11  (Figure 3C). Compared with the native structure, RUC-1 did not induce any local conformational changes, such as in the β1-α1 and β6-α7 loops or α7-helix (supplemental Figure 1A).

RUC-1 fits into a hydrophobic pocket lined by α_IIb residues Phe-160, Tyr-190, Leu-192, and Phe-231 (Figure 3D). In the largest hydrophobic contact, the fused ring of RUC-1 lies flat against the aromatic ring of Tyr-190 to form a π-π stacking interaction (Figure 3D). The orientation of the Tyr-190 side chain is stabilized by a hydrogen bond between the Tyr-190 hydroxyl group and the backbone carbonyl oxygen of β₃ Arg-216 (Figure 3D). The basic piperazinyl nitrogen of RUC-1 forms hydrogen bonds to the α_IIb Asp-224 side chain and to the backbone carbonyl oxygen of Ser-225 (Figure 3D). The side chain orientation of Asp-224 is also stabilized by a hydrogen bond with the backbone nitrogen of Trp-162 (Figure 3D). The carbonyl oxygen of RUC-1 forms hydrogen bonds to 2 water molecules that are held in place by the side chain oxygens of α_IIb Asp-232 (Figure 3D). RUC-1 does not have direct contact with β₃ residues. However, one of the waters that interacts with the carbonyl oxygen of RUC-1 hydrogen bonds with the β₃ Ala-218 backbone nitrogen, and one of the nitrogens of RUC-1's fused ring forms a hydrogen bond with a water molecule that is stabilized by hydrogen bonding with the carbonyl oxygen of β₃ Arg-216 (Figure 3D).

Similar to RGD-based anti-α_IIbβ₃ and anti-α_Vβ₃ drugs, RUC-1 has a basic piperazinyl group that mimics the Lys or Arg. However, it lacks a carboxyl group mimic of the Asp (Figure 3E). RUC-1 binds to the same pocket as occupied by the Lys-Gln-Ala (KQA) sequence in the fibrinogen γC peptide KQAGDV (Figure 3F) and by the Arg of the RGD motif (Figure 3G). RUC-1 mimics the Lys or Arg by forming a hydrogen bond with Asp-224. The KQAGD sequence and RGD sequence take “upper” and “lower” paths in a hydrophobic pocket to Asp-224. It is of interest that the 3 rings of RUC-1 occupy both of these routes—and the space in between—to fill the hydrophobic binding pocket (Figure 3F-G). In addition, the RUC-1 carbonyl group mimics the water-mediated hydrogen-bond interaction of the carbonyl oxygen of the Ala of KQAGDV and the Arg of RGDV (Figure 3F-G). Compared with these peptides, RUC-1 is expected lose less entropy on binding because of its fused ring structure.

The binding pose of RUC-1 in α_IIbβ₃ headpiece crystals was almost identical to that obtained in MD simulation (Figure 4A). Several important interactions were maintained during the entire simulation, including: (1) the hydrogen bond between either of the 2 oxygens of the α_IIb Asp-224 carboxyl side chain and the piperazinyl nitrogen of RUC-1 (Figure 4B), (2) the hydrogen bond between the other α_IIb Asp-224 side chain oxygen not engaging the piperazinyl nitrogen and the backbone nitrogen of α_IIb Trp-162 (Figure 4B), (3) the π-π stacking interaction between the RUC-1 fused ring and the α_IIb Tyr-190 (Figure 4C), (4) the polar interaction between the Tyr-190 hydroxyl group and the backbone carbonyl oxygen of the β₃ Arg-216 (Figure 4C), and (5) the water molecule-mediated polar interactions between Asp-232 and the carbonyl oxygen of RUC-1 (Figure 4D).

RUC-1 is a less potent inhibitor of mouse and rat α_IIbβ₃ and does not inhibit α_Vβ₃.^35,36  There is no equivalent of residues Asp-224, Phe-160, and Phe-231 in α_V, consistent with lack of anti-α_V integrin activity. Human, murine, and rat α_IIb all have the Asp-224 that interacts with the basic piperazinyl nitrogen of RUC-1 as well as Phe-160 and Phe-231 that interact with the heterocyclic fused-ring structure of RUC-1 (Figure 3D). However, both mouse and rat have α_IIb Y190F substitutions. In addition, rat α_IIb has a D232H substitution. The Y190F substitution may change this residue's orientation because of the loss of the hydrogen bond to β₃ Arg-216, with the D232H substitution disrupting the water-mediated interaction between RUC-1 and α_IIb Asp-232 (Figure 3D).

Well-tempered metadynamic simulations³⁷  using a restraining potential to limit exploration of ligand phase space showed that RUC-1 formed the same type of interactions with the receptor as found by standard MD simulations and X-ray crystallography (supplemental Figure 2A). In contrast, this RUC-1–preferred binding mode did not correspond to the lowest-energy minima in simulations of the Y190F (supplemental Figure 2B) or D232H mutants (supplemental Figure 2C).

To test the importance of α_IIb Tyr-190 and Asp-232 for RUC-1 binding, we expressed Y190F and D232H mutants in HEK 293 cells. Fibrinogen bound equally well to human α_IIbβ₃ and both mutants in the presence of the activating mAb PT25-2 (Figure 5). RUC-1 blocked fibrinogen binding to HEK293 cells expressing normal human α_IIbβ₃ with an IC₅₀ of 8μM (95% confidence interval [CI] 6-12μM; Figure 5). In sharp contrast, the RUC-1 IC₅₀ was increased to 80μM (95% CI 45-138μM) for the α_IIb Y190F mutant and 1000μM (95% CI 886-1127μM) for the α_IIb D232H mutant.

RUC-1, unlike tirofiban and eptifibatide, does not induce conformational changes in β₃, as judged by the binding of the β₃ LIBS mAb AP5. Nor does it induce ligand binding, as judged by a priming assay.³⁵  The crystal structure data in the current study showed that RUC-1 failed to initiate movements in the β1-α1 and β6-α7 loops or the α7-helix. However, since lattice contacts could interfere with large structural rearrangements, such as hybrid domain swing out, we examined the effect of RUC-1 on the conformation of the α_IIbβ₃ headpiece in solution.

Induction of conformational change in α_IIbβ₃ was measured by an increase in Stokes radius in gel filtration or DLS. In these experiments, the chymotrypsin-treated, purified α_IIbβ₃ headpiece, including the α_IIb thigh domain (see “Methods” and Figure 1C) was analyzed before and after mixing with saturating concentrations of RUC-1, tirofiban, or eptifibatide. RUC-1 did not induce a shift in the elution volume of the α_IIbβ₃ headpiece in gel filtration (Figure 6A). In contrast, tirofiban and eptifibatide each substantially shifted elution volume, indicating a less compact structure (Figure 6A-B). When measured by DLS, the Stokes radii for the α_IIbβ₃ headpiece were 4.88 nm and 4.86 nm in the absence and presence of RUC-1, respectively (Figure 6B)—a result that is again consistent with the absence of a conformational rearrangement (Figure 6C). In contrast, tirofiban and eptifibatide both increased the Stokes radius of the α_IIbβ₃ headpiece, to 5.27 nm and 5.22 nm, respectively (Figure 6B)—corresponding with what would be expected from a swing-out motion of the β₃ hybrid domain (Figure 6D). Estimates of the Stokes radii with HYDROPRO⁴²  were 4.68 nm for the closed headpiece structure (Figure 6C) and 5.03 nm for the open headpiece structure (Figure 6D). The Stokes radii measurements demonstrate that the α_IIbβ₃ headpiece in solution remains in a compact closed conformation after treatment with RUC-1.

In our integrin α_IIbβ₃ headpiece fragment preparation, the α_IIb subunit was truncated after the β-propeller domain; the β₃ subunit was truncated after the I-EGF1 domain (Figure 1D). Before this work, all crystal structures showing the closed headpiece were of intact ectodomains in the bent conformation,^3-5  whereas all crystal structures showing the open headpiece were of the same headpiece fragment used here.¹⁰  Open headpiece crystal structures have been obtained with or without 10E5 Fab, which binds a region of α_IIb with our current closed headpiece crystal structure, that does not shift in allostery.¹¹  Together with our current closed headpiece crystal structure, these data exclude the possibility that leg truncation alone is sufficient to induce hybrid domain swing out.¹² 

The presence of closed and open integrin headpieces in crystal structures, in the absence of constraints imposed by contacts with the lower legs, is a finding that is in agreement with EM studies of integrins α_Vβ₃, α_Lβ₂, and α_Xβ₂—all of which revealed closed and open headpiece conformations in extended integrins, with the closed headpiece conformation predominating under nonactivating conditions.^15,17,43  EM study of the α₅β₁ headpiece also revealed a closed conformation in the absence of ligand.¹⁶  Cocrystals of the α_IIbβ₃ headpiece with 3 ligand-mimetic antagonists all demonstrated the open conformation.¹¹  It is of interest that the cacodylate ion was found to act as a pseudo-ligand, coordinating the β₃ MIDAS and stabilizing the open conformation when the α_IIbβ₃ headpiece was crystallized in the absence of antagonists.¹¹  No electron density for such a pseudo-ligand was found in the closed headpiece crystal structure in the current study. Instead, we observed ordered water molecules around the MIDAS Mg²⁺ ion. Similarly, none of the closed headpiece, bent ectodomain crystal structures showed an electron density corresponding to a ligand or pseudo-ligand. We conclude that the presence of a closed or open headpiece correlates with the absence or presence of MIDAS ligation, respectively, and does not correlate with the absence or presence of leg domains. In vivo, two mechanisms are thought to stabilize the high affinity, open headpiece conformation. One is binding to ligand. The other is lateral force exerted parallel to the direction of hybrid domain swing out by actin cytoskeleton association with the integrin β-subunit cytoplasmic domain.⁵ 

The two independent α_IIbβ₃ headpiece molecules in our crystal asymmetric unit provide further information on the range of βI/hybrid domain orientations and interfaces accessible in the closed headpiece conformation. The conformation of the βI domain in our closed headpiece structure is essentially identical with that of the bent α_IIbβ₃ entire ectodomain structure, with Cα-RMSD of 0.48 Å for residues Asp-109-Ser-353. However, the orientation between the βI/hybrid domains varied between the closed headpiece and the bent ectodomain structures. In addition, some of the hydrogen-bond networks in the βI/hybrid interface were remodeled in our closed headpiece structure (Figure 2E,F). Interestingly, similar variations in βI/hybrid domain orientation were observed among the open headpiece structures in different crystal forms (Table 2). These variations are all the result of displacement at the last turn of the α7-helix and its linked loop to the β hybrid domain.

Similar to our previous structure of the complete α_IIbβ₃ ectodomain, we found that MIDAS, SyMBS, and ADMIDAS were preloaded with metal ions when the protein solution contained 1mM Mg²⁺ and 1mM Ca²⁺, even though the crystallization well solution did not contain cations. Metal ion occupancy of SyMBS and ADMIDAS with Ca²⁺ was complete under these conditions, whereas occupancy of MIDAS was partial. Occupancy of all sites was complete when the crystals were soaked in the well solution containing 5mM Mg²⁺ and 1mM Ca²⁺. Furthermore, excellent densities were observed for all of the water molecules in the inner, octahedral coordination shells of MIDAS Mg²⁺ and ADMIDAS Ca²⁺. The apparently looser binding of MIDAS Mg²⁺ compared with the ADMIDAS Ca²⁺ may be the result of its coordination by only 2 groups supplied by the protein; the other 4 inner coordinations are to water. The finding in an α_Vβ₃ ectodomain structure that there is a lack of metal ion at the SyMBS, and only partial occupancy at MIDAS,^3,6  is explicable based on the use of only Ca²⁺ in crystallization, the synergy between Ca²⁺ and Mg²⁺ in integrin ligand binding, and the coordination of SyMBS and MIDAS by Glu-220 (Zhu et al⁵  and references therein). Our results showing that metals are preloaded make the proposal highly unlikely that metal binding to SyMBS and MIDAS is induced by ligand binding.^3,23  Furthermore, MIDAS is buried by ligand binding, and SyMBS, which is completely surrounded by protein ligands, is already buried with no water accessibility before ligand binding and becomes buried more deeply afterward.

Comparison of our unliganded closed α_IIbβ₃ headpiece structure with our previously described liganded open α_IIbβ₃ headpiece structure, along with the failure of RUC-1 to induce the open conformation of β₃, supports the hypothesis that engaging the MIDAS metal ion and the β1-α1 loop backbone by the ligand carboxyl group is the mechanism by which ligand binding triggers adoption of the open conformation. It could be argued that crystal contacts prevented α_IIbβ₃ from adopting the open conformation when RUC-1 was soaked into the crystal. However, soaking the RGD-based antagonist cilengitide into the α_Vβ₃ ectodomain crystal induced the movement of the β1-α1 loop, the breaking of the Met-335 carbonyl coordination of the ADMIDAS metal ion, and ADMIDAS metal ion movement toward MIDAS²³  (supplemental Figure 1B). Moreover, soaking tirofiban or eptifibatide into our closed headpiece crystals also induced similar local conformational changes (J.Z., J.Z., T.A.S., unpublished data, 2010).

Our biophysical studies demonstrated that RUC-1 does not alter the gel filtration or DLS properties of the α_IIbβ₃ headpiece, providing support for it not inducing a conformational change in β₃. In contrast, eptifibatide and tirofiban each induced an increase in Stokes radius of approximately 0.34 nm. This increase is in accord with the calculated increase of 0.35 nm in Stokes radius using the coordinates of the closed and open headpiece structures. Thus, these results provide strong support for the hypothesis that tirofiban and eptifibatide induce hybrid domain swing out in solution. The difference between RUC-1 and the traditional RGD antagonists in their ability to induce hybrid domain swing out provides additional support for the importance of the MIDAS-coordinating carboxyl group in inducing this conformational change. Ligand engagement of the β₃ I domain, and in particular coordination to the MIDAS metal ion and hydrogen bonds to the backbone of the β1-α1 loop, is therefore the most likely trigger for β₃ swing out. These interactions may be sufficient to induce opening because the cacodylate ion, acting as an α_IIbβ₃ pseudoligand and binding exclusively to the MIDAS metal ion and the backbone of the β1-α1 loop, was associated with the open conformation.¹¹ 

Previous attempts to target α_IIbβ₃ and other integrin receptors that bind RGD-containing peptides have focused on developing high-affinity congeners that mimic the RGD sequence in having both positive and negative charge groups separated by a distance similar to that in RGD peptides. The weight of evidence now suggests that such molecules may induce conformational changes that are undesirable in that they can induce the high-affinity ligand binding state of the receptor and can expose regions of the receptor that can serve as neoepitopes for preexisting or induced antibodies that may be responsible, in part, for the thrombocytopenia associated with these agents.⁴⁴  Our studies with RUC-1 indicate that it is possible to identify molecules that can block ligand binding without engaging the β₃ MIDAS metal ion or inducing either a high-affinity binding state or a major conformational change in the β₃ subunit. Moreover, RUC-1 demonstrates that it is possible to obtain high specificity for human α_IIbβ₃ relative to human α_Vβ₃—and even α_IIbβ₃ from other species—with a small molecule that binds exclusively to α_IIb. Although RUC-1 has considerably lower affinity for α_IIbβ₃ than eptifibatide or tirofiban, it is within the range of affinities of a sizable percentage of drugs. Thus, RUC-1 provides important proof of concept in designing a new class of integrin receptor antagonists that may have therapeutic advantages over those based on the RGD sequence.

We thank Jihong Li of Rockefeller University for site-directed mutagenesis and fibrinogen binding studies and Tingting Song of Rockefeller University for statistical analyses.

This work was supported in part by grants from the National Institutes of Health (HL19278, HL46875, and ULRR024143) and Stony Brook University. Computational resources were obtained from NSF TeraGrid (MRAC MCB080077).

National Institutes of Health

Contribution: Jieqing Zhu, Jianghai Zhu, A.N., and D.P. performed research, analyzed data, and wrote the paper; M.F. and B.S.C. designed some of the studies, analyzed data, and wrote the paper; T.A.S. supervised the research and data analysis and wrote the paper.

Conflict-of-interest disclosure: B.S.C. is an inventor of abciximab (Centocor) and the VerifyNow assay (Accumetrics Inc), and in accord with federal law and the policies of the Research Foundation of the State University of New York at Stony Brook and Mount Sinai School of Medicine, respectively, he shares in royalty payments. Rockefeller University has applied for a patent on RUC-1 and B.S.C. may ultimately share in royalties from RUC-1. B.S.C. serves as a consultant to Accumetrics Inc. The remaining authors declare no competing financial interests.

Correspondence: Timothy A. Springer, Immune Disease Institute, Children's Hospital Boston, and Department of Pathology, Harvard Medical School, Boston, MA 02115; e-mail: springer@idi.harvard.edu.