β-Subunit Binding Is Sufficient for Ligands to Open the Integrin αIIbβ3 Headpiece*

The platelet integrin αIIbβ3 binds to a KQAGDV motif at the fibrinogen γ-chain C terminus and to RGD motifs present in loops in many extracellular matrix proteins. These ligands bind in a groove between the integrin α and β-subunits; the basic Lys or Arg side chain hydrogen bonds to the αIIb-subunit, and the acidic Asp side chain coordinates to a metal ion held by the β3-subunit. Ligand binding induces headpiece opening, with conformational change in the β-subunit. During this opening, RGD slides in the ligand-binding pocket toward αIIb, with movement of the βI-domain β1-α1 loop toward αIIb, enabling formation of direct, charged hydrogen bonds between the Arg side chain and αIIb. Here we test whether ligand interactions with β3 suffice for stable ligand binding and headpiece opening. We find that the AGDV tetrapeptide from KQAGDV binds to the αIIbβ3 headpiece with affinity comparable with the RGDSP peptide from fibronectin. AGDV induced complete headpiece opening in solution as shown by increase in hydrodynamic radius. Soaking of AGDV into closed αIIbβ3 headpiece crystals induced intermediate states similarly to RGDSP. AGDV has very little contact with the α-subunit. Furthermore, as measured by epitope exposure, AGDV, like the fibrinogen γ C-terminal peptide and RGD, caused integrin extension on the cell surface. Thus, pushing by the β3-subunit on Asp is sufficient for headpiece opening and ligand sliding, and no pulling by the αIIb subunit on Arg is required.

The platelet integrin ␣ IIb ␤ 3 binds to a KQAGDV motif at the fibrinogen ␥-chain C terminus and to RGD motifs present in loops in many extracellular matrix proteins. These ligands bind in a groove between the integrin ␣ and ␤-subunits; the basic Lys or Arg side chain hydrogen bonds to the ␣ IIb -subunit, and the acidic Asp side chain coordinates to a metal ion held by the ␤ 3 -subunit. Ligand binding induces headpiece opening, with conformational change in the ␤-subunit. During this opening, RGD slides in the ligand-binding pocket toward ␣ IIb , with movement of the ␤I-domain ␤1-␣1 loop toward ␣ IIb , enabling formation of direct, charged hydrogen bonds between the Arg side chain and ␣ IIb . Here we test whether ligand interactions with ␤ 3 suffice for stable ligand binding and headpiece opening. We find that the AGDV tetrapeptide from KQAGDV binds to the ␣ IIb ␤ 3 headpiece with affinity comparable with the RGDSP peptide from fibronectin. AGDV induced complete headpiece opening in solution as shown by increase in hydrodynamic radius. Soaking of AGDV into closed ␣ IIb ␤ 3 headpiece crystals induced intermediate states similarly to RGDSP. AGDV has very little contact with the ␣-subunit. Furthermore, as measured by epitope exposure, AGDV, like the fibrinogen ␥ C-terminal peptide and RGD, caused integrin extension on the cell surface. Thus, pushing by the ␤ 3 -subunit on Asp is sufficient for headpiece opening and ligand sliding, and no pulling by the ␣ IIb subunit on Arg is required.
Integrins are heterodimeric adhesion receptors that transmit outside-in as well as inside-out signals across the cell membrane. Ligands of ␣I-less integrins such as ␣ IIb ␤ 3 bind to a groove located in the interface between the ␣and ␤-subunits ( Fig. 1) (1). When activated, the ectodomain of ␣ IIb ␤ 3 undergoes large-scale structural reshaping to an extended conformation with an open headpiece ( Fig. 1) and binds with high affinity to the two distal ends of a fibrinogen dimer, causing platelets to form tight aggregates (2,3). Arg-Gly-Asp (RGD)-binding integrins, including ␣ IIb ␤ 3 , ␣ v ␤ 3 , and ␣ 5 ␤ 1 , bind the Arg side chain to the ␤-propeller domain of the ␣-subunit via charged hydrogen bond(s). In contrast, the Asp side chain of RGD coordinates to Mg 2ϩ held in the metal ion-dependent adhesion site (MIDAS) 3 and forms multiple hydrogen bonds to NH backbone amide groups, including two in the ␤1-␣1 loop of the ␤-subunit ␤I domain (3)(4)(5).
Within fibrinogen, ␣ IIb ␤ 3 binds not to RGD, but to a 400 HHLGGAKQAGDV 411 sequence at the C terminus of the ␥ subunit (3,6,7). Lys-406 of ␥C forms charged hydrogen bonds with the ␣ IIb subunit; that is, Lys-406 of ␥C is functionally equivalent to the Arg of RGD. The side chain carboxyl group of Asp-410 directly coordinates to the ␤ 3 -subunit MIDAS Mg 2ϩ and forms hydrogen bonds to the ␤I ␤1-␣1 loop backbone, equivalently to the Asp of RGD (3). Additionally, the C-terminal ␣-carboxyl group of ␥C Val-411 forms a water-mediated, indirect coordination to the Ca 2ϩ held in the adjacent to MIDAS (ADMIDAS) (3).
We recently soaked different concentrations of RGD peptide into crystals containing the ␣ IIb ␤ 3 headpiece in the closed conformation (assigned as state 1) and resolved six intermediate states (states 2-7) between the low-affinity closed headpiece and the high-affinity open headpiece (state 8) (4). Between states 1 and 8, the ␤1-␣1 loop, which supplies three of the side chains that coordinate the MIDAS Mg 2ϩ ion, moved toward the Asp side chain, enabling the Asp side chain to form hydrogen bonds to ␤1-␣1 loop backbone NH groups. RGD slid in its binding groove, enabling its Arg side chain to closely approach ␣ IIb and to eventually form a charged hydrogen bond to ␣ IIb Asp-224. However, not until state 7 did a singular, well resolved electron density map appear for the Arg side chain. In states 1-6, the Arg side chain showed either weak density, reflecting low occupancy, or multiple conformations, including watermediated hydrogen bonds to Asp-224. In contrast, the Asp of RGD always had good density and always directly coordinated with the MIDAS Mg 2ϩ ion, suggesting that Asp binding to the ␤-subunit might be the main driver of headpiece opening. On the other hand, it was also possible that the attraction of the Arg side chain to the oppositely charged Asp-224 in ␣ IIb was responsible for opening by pulling RGD, and with it the ␤I domain ␤1-␣1 loop, toward ␣ IIb . Headpiece opening, rather than extension, is what greatly increases (by Ͼ100-fold) integrin affinity for ligands (4,8). Headpiece opening is communicated across the ␤I domain by ␣7-helix pistoning to swing out of the hybrid domain ( Fig. 1), which is a large-scale conformational change capable of being transmitted through the long integrin ectodomain legs to the cytoplasmic domains (1). Therefore, the question of whether headpiece opening is intrinsic to the ␤-subunit, or requires a pull by the ␣-subunit, is key to understanding both the biology and the mechanochemistry of integrins.
To resolve this issue, we have examined binding to ␣ IIb ␤ 3 of truncated fibrinogen ␥C AGDV peptides that lack the Lys of KQAGDV, and thus cannot bind to the ␣ IIb subunit. Previous studies have shown that the QAGDV pentapeptide blocks binding of full-length fibrinogen to human platelets (9). Furthermore, CHO K1 cells transfected with ␣ IIb ␤ 3 could adhere to monolayers functionalized with AGD or RGD peptides equally well, and adhesion to AGDVC monolayers was inhibited by similar concentrations of RGDS and AGDV peptides (10). Our studies address multiple questions beyond the minimal requirements for integrin headpiece opening. These include how AGDV peptide can bind and activate, and the specific role of its C-terminal carboxyl group in interaction with the ADMIDAS.
Remarkably, we find comparable affinities of AGDV peptide, fibrinogen ␥C dodecapeptide, and RGD peptide for the ␣ IIb ␤ 3 integrin headpiece. AGDV can induce complete headpiece opening in solution, and crystals soaked with AGDV for varying durations defined the structural basis for AGDV-induced headpiece opening. Moreover, the C-terminal carboxyl of AGDV contributes to affinity by a highly geometrically constrained and hence highly specific water-mediated interaction with the ADMIDAS metal ion.

Experimental Procedures
Ligands-FITC-aminohexanoyl-HHLGGAKQAGDV (FITCdodecapeptide) was synthesized by GenScript (Piscataway, NJ). Unlabeled fibrinogen ␥C dodecapeptide was from American Peptide Company (Sunnyvale, CA). All other peptides were synthesized by GenScript with Ͼ95% purity. Unless otherwise specified, peptides were not modified at either terminus. Ro-435054 was a generous gift from Dr. Paul Gillespie at Roche.
Fluorescence Anisotropy Binding Assay-Binding affinities for different peptide ligands were measured using fluorescence anisotropy as described previously (12). Anisotropy is defined as (Fʈ Ϫ FЌ)/(Fʈ ϩ 2FЌ), where Fʈ and FЌare the fluorescence intensities parallel and perpendicular to the excitation plane, respectively. mA units (shown in figures) correspond to 1,000 ϫ anisotropy. The affinity of the FITC-dodecapeptide probe (used at 3 nM) for the ␣ IIb ␤ 3 headpiece (K D F* ) was measured by increasing the concentration of purified recombinant ␣ IIb ␤ 3 headpiece to saturation. K D F* was fitted using one-site, specific binding; i.e. A ϭ A max ϫ C itg /(K D F* ϩ C itg ), where A represents the anisotropy signal from specific binding, C itg represents the concentration of integrin ␣ IIb ␤ 3 headpiece, and A max is the maximum anisotropy signal. Background anisotropy obtained by adding an excess of 10 M tirofiban was subtracted from anisotropy.
After determining K D F* , unlabeled peptides were used to compete FITC-dodecapeptide probe (3 nM) binding to 300 nM ␣ IIb ␤ 3 headpiece. I 50 values for unlabeled peptides were obtained by least squares fitting. I 50 values were converted to affinity values (K D p ) for each peptide of interest (12) using where K D F* is the FITC-dodecapeptide binding affinity to ␣ IIb ␤ 3 headpiece (643 nM), F* T is the total concentration of FITCdodecapeptide (3 nM), ␣ T is the total concentration of ␣ IIb ␤ 3 headpiece (300 nM), IC 50 and I 50 are the free and total concentrations of the peptide of interest causing the displacement of 50% of specifically bound FITC-dodecapeptide (i.e. IC 50 ϭ I 50 Ϫ 0.5 ϫ 300 nM), and [F*␣] 50 is the concentration of FITCdodecapeptide bound to ␣ IIb ␤ 3 headpiece at I 50 . All measurements were in 20 mM HEPES, pH 7.5, 150 mM NaCl, 2 mM MnCl 2 , 0.1 mM CaCl 2 . The Prism GraphPad program (La Jolla, CA) was used to fit the saturation binding curve, inhibition curves, and I 50 values.
Size-exclusion Chromatography and Dynamic Light Scattering (DLS)-The ␣ IIb ␤ 3 headpiece with coiled-coils was treated with chymotrypsin (chymotrypsin/integrin mass ratio ϭ 1/200) at room temperature (13) for ϳ16 h to remove C-terminal ␣-helical coiled-coils and to leave intact the thigh domain in the ␣-subunit. Digested headpiece was purified by passage through a nickel-nitrilotriacetic acid column (13). For Stokes radius measurements, the re-purified headpiece at 10 M (1 mg/ml by A 280 ) was incubated with 1 mM peptide ligand in 20 mM TBS with 1 mM Mg 2ϩ and 1 mM Ca 2ϩ at 25°C for 30 min and then applied to a Superdex 200 column (GE Healthcare Life Sciences) pre-equilibrated with the same buffer containing 1 mM peptide. Calibration and conversion of elution volume to Stokes radius was as described (11). For DLS, 20 M chymotrypsintreated, re-purified headpiece was filtered through a 100 nm cutoff membrane (Millipore Ultrafree centrifugal filter), incu-  ␤-Subunit Binding and ␣ IIb ␤ 3 Headpiece Opening FEBRUARY 26, 2016 • VOLUME 291 • NUMBER 9 bated with 1 mM peptide in TBS plus 1 mM Mg 2ϩ /Ca 2ϩ for 10 min at 25°C in the cuvette, and read in the Viscotek 802 DLS (Viscotek Corp.). The DLS-derived hydration radii (R h ) were obtained by intensity-based fitting using the OmniSize program for 2-3 independent measurements using the same batch of integrin headpiece sample, but measured in different days. Each measurement was an average of at least fifteen 10-s reads.
Crystallization, Ligand Soaking, and Structure Determination-The ␣ IIb ␤ 3 headpiece was crystallized in the closed conformation with the thigh domain removed (11). Briefly, chymotrypsin-and carboxypeptidase-treated ␣ IIb ␤ 3 -10E5 Fab complex was concentrated to 10 mg/ml in TBS (20 mM Tris, 150 mM NaCl, pH 7.4) with 1 mM Mg 2ϩ , 1 mM Ca 2ϩ and crystallized in 11% PEG 8000, 0.2 M ammonium sulfate, 0.1 M Tris-HCl, pH 8.9 at 4°C (11). Crystals were then stabilized with the same crystallization buffer but containing 15% PEG 8000, and glycerol concentration was incremented in 5% steps to 20% for cryoprotection. Finally, crystals were soaked in ligand in the same buffer with defined metal conditions for a specified duration at 4°C (see Table 1). Diffraction data were collected at beamline ID-23 of the Advanced Photon Source (Argonne, IL). Resolution limit was determined by cross-correlation (14). Refinements using Phenix (15) were as described previously (4).
Ligand-induced Binding Site (LIBS) Epitope Expression-CHO K1 cells stably transfected with full-length ␣ IIb ␤ 3 (16) were incubated at room temperature with serially diluted peptide in 20 mM HEPES, pH 7.4, 150 mM NaCl, 5.5 mM glucose, 1% BSA, 1 mM Mg 2ϩ /Ca 2ϩ , and 10 g/ml LIBS1 antibody for 30 min. Mixtures were washed twice in the same buffer minus LIBS1. The washed cells were incubated with FITC-conjugated anti-mouse antibody at room temperature for 30 min, washed again, chilled on ice, and subjected to flow cytometry. The doseresponse curves were fit, and EC 50 values were calculated by Prism GraphPad.

Results
Affinities of Peptides Lacking Arg or Lys for the ␣ IIb ␤ 3 Headpiece-Truncated fibrinogen ␥C peptides such as QAGDV and AGD have been shown to block fibrinogen binding to platelets with similar IC 50 values of ϳ100 M (9), but more precise affinity measurements are required to understand the contribution of Lys or Arg. To test whether the Arg of RGD or Lys of KQAGDV is required for integrin binding, we quantitated the binding affinity to purified ␣ IIb ␤ 3 headpiece in buffer with 2 mM Mn 2ϩ and 0.1 mM Ca 2ϩ using fluorescence polarization anisotropy. The fluorescent probe FITC-aminohexanoyl-HHLGGAKQAGDV bound saturably with a dissociation constant (K D ) ϭ 0.64 Ϯ 0.13 M (Fig. 2A). Competition with fluorescent probe at 3 nM and integrin at 300 nM was then used to determine the K D values of different peptides (Fig. 2, B-F) (see "Experimental Procedures"). The fibrinogen ␥C dodecapeptide and fibronectin hexapeptide GRGDSP showed K D of 1.27 Ϯ 0.08 and 0.41 Ϯ 0.26 M, respectively (Fig. 2, B and C). In contrast, QAGDV and AGDV bound more weakly, with K D of 7.85 Ϯ 2.53 and 9.03 Ϯ 2.75 M, respectively (Fig. 2, D and E). N-acetylated AGDV (Ac-AGDV) binds ϳ9 times more strongly Fig. 2F) than unmodified AGDV. Thus, Ac-AGDV has an affinity almost identical to the native ␥C dodecapeptide (Fig. 2, B-F).
AGDV Induces Complete Headpiece Opening in Solution-We determined whether headpiece opening can occur without the Arg of RGD or the Lys of fibrinogen ␥C peptide. AGDVinduced opening (i.e. hybrid domain swing-out, Fig. 1, B and C) of a six-domain ␣ IIb ␤ 3 headpiece fragment was measured as an increase in Stokes radius (Fig. 3) (11). The hydrodynamic radii of closed and open six-domain ␣ IIb ␤ 3 conformations were estimated from crystal structures (3,17) with HYDROPRO (18) as 4.68 and 5.03 nm, respectively. When incubated with 1 mM dodecapeptide and RGDF peptide (a high-affinity ligand of ␣ IIb ␤ 3 (19)), in the resting condition (TBS, pH 7.4, plus 1 mM Mg 2ϩ and Ca 2ϩ ), the Stokes radius of ␣ IIb ␤ 3 headpiece increased from 4.69 Ϯ 0.02 to 5.22 Ϯ 0.06 nm and 5.19 Ϯ 0.04 nm, respectively (Fig. 3A) as measured by DLS. A similar increase (to 5.24 Ϯ 0.03 nm) was observed with 1 mM AGDV peptide. The change in Stokes radius was also measured by size-exclusion chromatography (Fig. 3B). The elution volume shifted from 12.56 Ϯ 0.01 ml in buffer alone to 12.15 Ϯ 0.06 ml in the presence of 1 mM RGDF and to 12.19 Ϯ 0.05 ml in AGDV. The corresponding Stokes radius shifted from 4.78 nm in buffer to 5.23 nm in RGDF and to 5.17 nm in AGDV (Fig. 3A). The DLS and gel-filtration results (Fig. 3A) (20). Two copies of the ␣ IIb ␤ 3 headpiece complex (molecules 1 and 2) exist in one asymmetric unit (4), so we were able to obtain two different conformational states in a single crystal (Table 1). Based on the conformation and position of the moving parts of the ␤I domain, i.e. the ␤1-␣1 loop, ␣1-helix, ␤6-␣7 loop, and the ADMIDAS metal, each intermediate (Fig. 4) is assigned a state number between 1 (Fig. 4A) and 8 (Fig. 4B) corresponding to previous RGD-bound intermediates (4). AGDV binds in the groove formed between the ␣and ␤-subunits (Fig. 4, C, E, G, and I), but has limited contact with ␣, because it occupies only a portion (about two-thirds) of the groove occupied by RGD (Fig. 4B). The N-terminal amine of AGDV is Ͼ9.0 Å away from Asp-224 of ␣ IIb , which forms salt bridges to the Arg of RGD and the Lys of KQAGDV. Moreover, unlike the extended backbone conformation adopted by RGD (Fig. 4, A and B), the backbone of AGDV is curved (Fig. 4, C, E, G, and I).
When compared with RGD (Fig. 4, A and B) and dodecapeptide structures in complex with the natively open headpiece in state 8 (Fig. 4J), the ⌽ angle of Gly of AGDV is rotated about 180° (Fig. 4, C, E, G, and I). This correlates with the markedly different backbone carbonyl oxygen orientation of the Arg in RGD (Fig. 4, A, B, and F) and Ala in KQAGDV (Fig. 4J) when compared with Ala in AGDV and AGDV-NH 2 (Fig. 4, C-E and  G-I). The peptide ␣-amino group, with a pK a estimated to be ϳ8, is weakly basic, and would be partially charged at the bind-ing assay pH of 7.4 and the crystallization pH of 8.9. To test for a weak -cation interaction between the N-terminal primary amine and Tyr-190 of ␣ IIb (plan distance ϳ4 Å), we abolished protonation by N-terminal acetylation. Acetylation increased peptide affinity (Fig. 2F) and therefore ruled out a contribution by a charged ␣-amino group to AGDV to binding.
Overall, the intermediate states caused by AGDV are similar to those induced by GRGDSPK (4) (Fig. 4). The exception is AGDV state 3, in which the ␤1-␣1 loop backbone is shifted similarly to that with RGD in state 3, but the ADMIDAS Ca 2ϩ ion is shifted more toward the MIDAS and has lost its coordinations to Asp-126 and Asp-127 (Fig. 4E). In contrast, the ADMIDAS Ca 2ϩ ion retained its Asp-126 and Asp-127 coordinations in RGD state 3 (Fig. 4F). However, electron density for the ADMIDAS Ca 2ϩ in AGDV state 3 is weak, which suggests an ill-defined position in shape-shifting. Importantly, during soaking for 1 week, AGDV induced the ␤ 3 -subunit in molecule 1 in crystals to shift to state 7, i.e. almost all the way to the open conformation, despite little interaction with the ␣ IIb subunit (Fig. 4I).
The Pathway of AGDV-induced Headpiece Opening-The Asp carboxylate of AGDV in state 2 (Fig. 4C) shows a different angle from later states (Fig. 4, E, G, and I), with one carboxyl oxygen coordinating the MIDAS, and the other oxygen pointing away from the binding pocket. In this orientation, the ligand Asp would clash with the Tyr-122 side chain in the following states. In state 3, the Asp carboxyl re-orientates to enable for Tyr-122 and Ser-123 in the ␤1-␣1 loop to move closer to the ligand and forms a hydrogen bond with the Tyr-122 backbone (Fig. 4E). Between states 2 and 3, the side chain hydroxyl of Ser-123 moves closer to the ligand and replaces a water molecule in the MIDAS to directly coordinate with the MIDAS metal ion (Fig. 4, C-E). The direct coordination between the Ser-123 side chain and the MIDAS and the hydrogen bond between the Asp of AGDV and the Tyr-122 backbone each remain in place through later states. However, in addition, between states 2 and 3, the valine side chain of AGDV switches to a different rotamer that is sustained through state 7, and the C-terminal carboxyl group rotates toward the ADMIDAS metal (compare Fig. 4, E, G, and I with Fig. 4C). In state 6, the ␤1-␣1 loop continues its movement toward the ligand (Fig. 4G). At this point, the C␤ carbon of Ser-123 is 2.4 Å away from its position in state 1. In state 6, the ADMIDAS metal ion is 3.4 Å away from its position in state 1 and is much closer to the C-terminal carboxyl of AGDV (4.8 Å). Importantly, starting from state 6, the C-terminal carboxyl of AGDV forms an indirect, water-mediated coordination with the ADMIDAS metal ion (Fig. 4G). Between states 6 and 7, the C␣ atoms of Asp-126 and Asp-127 on the ␣1-helix move 2.2 Å toward the ADMIDAS metal ion (and 3.5 Å from state 1) and form direct coordinations (Fig. 4I). In addition, the ␤6-␣7 loop flips away from the ADMIDAS to accommodate movement of the ␣1-helix toward the ligand (Fig. 4, G and I). In state 7, the conforma-tions of the ␤1-␣1 loop, ␣1-helix, and ADMIDAS metal of ␤I are very close to the natively open state 8 bound to fibrinogen ␥C dodecapeptide (compare Fig. 4I with Fig. 4J). The ␣-carboxyl of AGDV now forms a stronger 2.7 Å hydrogen bond with an ADMIDAS-coordinating water molecule (Fig. 4I).
The C-terminal Carboxyl Group Is Not Required for Headpiece Opening-We tested the role of the AGDV peptide ␣-carboxyl group in headpiece opening. We first measured the affinities of a series of C-terminally modified AGDV peptides, including AGDV-NH 2 (amidated), AGDV-OCH 3 (O-methylated), and AGD(v) (D-valine in place of L-valine). Both AGDV-NH 2 and AGDV-OCH 3 showed weaker binding than AGDV, with 5.6-and 7.5-fold reductions in affinity, respectively (Fig. 2, G and H), in agreement with results using amidated dodecapeptide (21). On the other hand, substituting L-valine with D-valine in AGD(v) resulted in a more pronounced, 24-fold decrease in binding affinity (Fig. 2I). These results clearly demonstrate that the contact between the ␣-carboxyl moiety of ␥C and the ADMIDAS metal ion makes an important contribution to affinity and is highly specific both for the carboxyl group and for the chirality of the Val residue.
A free C terminus was not required for headpiece conformational change. At 1 mM (20 ϫ K D ), amidated AGDV increased the hydrodynamic radius from 4.69 to 5.04 nm, whereas at 1 mM (100 ϫ K D ), AGDV increased the radius to 5.2 nm (Fig. 3A). Attempts to increase the concentration of AGDV-NH 2 to 2.5  ␤-Subunit Binding and ␣ IIb ␤ 3 Headpiece Opening and 5 mM resulted in headpiece aggregation. However, the use of AGDV at 0.14 mM (16 ϫ K D ) and 0.48 mM (50 ϫ K D ) induced similar shifts in Stokes radius as AGDV-NH 2 at 1 mM (20 ϫ K D ) (Fig. 3A), showing that when corrected for amount of binding, AGDV-NH 2 and AGDV are similar in ability to induce headpiece opening. When closed headpiece crystals were soaked with 50 mM AGDV-NH 2 for 2 h, the peptide was observed in both integrin molecules in the asymmetric unit, with molecule 1 in state 5 (Fig. 4H) and molecule 2 in state 3 (Fig. 4D). Similar soaking for 2 h with AGDV resulted in binding to headpiece molecule 1 in state 6 and molecule 2 in state 3 (Fig. 4, G and E). The slight reduction in shape-shifting induced by AGDV-NH 2 is compatible with its lower affinity. The conformation of AGDV-NH 2 in state 5 is similar to AGDV in states 6 and 7. The main difference is that the terminal amide oxygen of AGDV-NH 2 is 3.9 Å away from the ADMIDAS water (Fig. 4H), whereas the corresponding distance in AGDV states 6 and 7 are 3.0 and 2.7 Å, respectively (Fig. 4, G and I). The conformations of Val of AGDV-NH 2 and AGDV in states 5-8 are very similar (Fig. 4, G-J). In state 3, the ADMIDAS metal ion position with AGDV-NH 2 (Fig. 4D) resembled that with RGD ( Fig. 4F) rather than with AGDV (Fig. 4E).
A Dicarboxylate ␣ IIb ␤ 3 Antagonist-Some ␣ IIb ␤ 3 antagonists have two carboxyl groups, such as Ro-435054 (22) (see Fig. 7A). To test whether the second carboxyl group would orient similarly to the ␣-carboxyl group of KQAGDV and AGDV, we soaked Ro-435054 into crystals. When soaked in 1 mM Ro-435054 with 2 mM MnCl 2 and 0.1 mM CaCl 2 , crystals either dissolved, cracked, or diffracted poorly. However, crystals were preserved when we soaked them with 1 mM Ro-435054 for 2 h in Mg 2ϩ /Ca 2ϩ . The compound bound to both ␣ IIb ␤ 3 molecules in the asymmetric unit and induced shape-shifting to states 6 and 3 (Table 1). This contrasts with soaking for 24 h with 10 mM GRGDSP in Mg 2ϩ /Ca 2ϩ , which only shifted molecule 1 to state 2 and bound to molecule 2 without shifting it from state 1 (4). Density for Ro-435054 was well defined; however, electron densities for ADMIDAS Ca 2ϩ ions in both molecules were very weak, and ADMIDAS Ca 2ϩ ions were therefore omitted during refinement. Ro-435054 is a peptidomimetic. One of the nitrogens in its N-terminal benzamidine overlaps with the ⑀-amino group of the Lys side chain of KQAGDV. Most importantly, the side chain carboxyl of the Asp residue and the ␣-carboxyl groups of the Phe residue of Ro-435054 align very well with those of the Asp and Val residues of ␥C dodecapeptide and AGDV (see Fig. 7B).
AGDV Induces Integrin Extension on Cell Surfaces Similarly to RGD Peptide and Fibrinogen Dodecapeptide-Exposure of LIBS epitopes on integrin ␣ IIb ␤ 3 is widely used as a surrogate of conformational change; however, the type of conformational change (e.g. integrin extension or headpiece opening) that LIBS antibodies report on ␣ IIb ␤ 3 is poorly characterized. Detergentsoluble, intact ␣ IIb ␤ 3 particles purified from human platelets are 91% in the bent conformation (23). Incubation with LIBS1 Fab fragment had little effect on the proportion of bent particles (95%), and no binding of LIBS1 Fab was detected by EM (Fig.  5A). After incubation with the high-affinity RGD-mimetic L-739758 (24) at a final concentration of 10 nM, plus 1 mM Mn 2ϩ and 0.1 mM Ca 2ϩ , most particles are in extended conformations (23); LIBS1 bound to the lower ␤-leg of extended ␣ IIb ␤ 3 , near the EGF4 and ␤-tail domains (Fig. 5, B and C). These results show that LIBS1 exposure does not occur in bent ␣ IIb ␤ 3 and requires extension; because the RGD-mimetic also induces headpiece opening, it remains to be determined whether LIBS1 binding requires headpiece opening in addition to extension.

Discussion
We have shown that the AGDV peptide can bind to ␣ IIb ␤ 3 integrin with affinity comparable with GRGDSP peptide and fibrinogen ␥C dodecapeptide, and is capable of inducing complete integrin headpiece opening in solution. The structural basis for AGDV-induced headpiece opening was determined by soaking AGDV into crystals and observing a series of intermediate conformational states. AGDV peptides differing in C-terminal modifications showed that the indirect coordination between the ␣-carboxyl of AGDV and the ADMIDAS metal ion contributes to affinity for ␣ IIb ␤ 3 . Binding of AGDV peptide is also able to induce LIBS1 epitope exposure and hence full-length ␣ IIb ␤ 3 integrin extension on cell surfaces.
Crystal structures showed that AGDV has limited contact with the ␣ IIb subunit. Of the total solvent-accessible surface area (25) buried by AGDV on ␣ IIb ␤ 3 (ϳ340 Å 2 ), only 23% (80 Å 2 ) is contributed by the ␣ IIb subunit. The ␣ IIb subunit contributes no specific interactions. In contrast, the tetrapeptide has a much larger, 260-Å 2 interface with ␤, or 77% of the total buried solvent-accessible surface area. Relative to typical interfaces buried in antibody-protein antigen interactions of 800 Å 2 (26), the overall burial of AGDV is small. Thus, the highly specific contacts made by AGDV must be important. The interface between AGDV and the integrin ␤I domain includes multiple hydrogen bonds with the ␤1-␣1 loop, and most importantly, the coordination of Asp side chain carboxyl to the MIDAS metal, and the water-mediated contact with the ADMIDAS metal.
Unlike RGD or ␥C dodecapeptide, AGDV peptides do not form salt bridges with the integrin ␣ IIb subunit. Despite lacking the salt bridges, Ac-AGDV shows an almost identical affinity with dodecapeptide, and AGDV peptide is only ϳ7 times less potent than the dodecapeptide in Mn 2ϩ /Ca 2ϩ . These observations show that the contributions in binding free energy of the salt bridge and other additional contacts with ␣ are relatively minor when compared with the interactions with ␤.
The C-terminal carboxyl group of AGDV contributes positively to binding based on the ϳ7-fold reduction in affinity with amidated AGDV (AGDV-NH 2 ). Nevertheless, the ␣-carboxyl group is not required for headpiece opening. In DLS experiments, similar saturation of the headpiece with AGDV-NH 2 and AGDV gave similar increases in hydrodynamic radius. Moreover, AGDV and AGDV-NH 2 induced similar headpiece opening when soaked into crystals.
In aggregate, our results show that Asp binding to the ␤ 3 -subunit contributes the majority of the energy for ligand binding and that interaction with the ␤ 3 -subunit is sufficient to induce headpiece opening and extension. These findings also suggest that ␤ 3 integrins may have broader ligand specificities than previously expected.
The native C-terminal carboxyl of AGDV peptide was more favorable than any other chemical modification we tested, including amidation, methylation, and replacing the C-terminal L-valine with a D-valine. The order of affinity was native carboxyl Ͼ amide Ͼ methyl ester Ͼ D-valine). The C-terminal carboxyl group forms a water-mediated coordination to the ADMIDAS metal (Fig. 4, G and I) (3). One explanation for the weaker affinity of AGDV-NH 2 is that the long-range ionic interaction (4.7 Å) between the ␣-carboxyl anion and ADMIDAS divalent metal cation is absent in AGDV-NH 2 . The other explanation is the different hydrogen bond strength among these derivatives. A charged carboxyl oxygen forms a stronger hydrogen bond to water than an uncharged carbonyl oxygen in an amide or ester. Interestingly, the chirality of Val of AGDV was very important, because substitution with D-valine resulted in a 24-fold reduction in affinity. These results are consistent with the identical orientation and rotamer of the Val side chain in AGDV and AGDV-NH 2 states 3-7 and in native state 8 of the dodecapeptide. The ␣-carboxyl group of D-valine cannot form polar contacts with the ADMIDAS, because the side chain of D-valine will clash with Tyr-122 of the ␤ 3 -subunit when its ␣-carboxyl group is oriented identically to that of L-valine. The dicarboxylate ␣ IIb ␤ 3 antagonist Ro-435054 binds the high-affinity state of ␣ IIb ␤ 3 ϳ100-fold better (K D ϭ 6 nM) than the low-affinity state (K D ϭ 580 nM) (27). Ro-435054 is a 4-residue peptidomimetic, with an N-terminal amidinobenzoyl, a ␤-alanine, an L-Asp, and a C-terminal L-Phe (Fig. 7A). Most interestingly, the C-terminal Phe residue of Ro-435054 adopts an identical orientation to the Val of KQAGDV and AGDV and thus may be considered a fibrinogen-mimetic. The high-affinity RGDF peptide (19) may be considered both fibrinogen-mimetic and RGD-mimetic. RGDF has the same two C-terminal amino acids as Ro-435054 and may be predicted to bind similarly.
Following fibrin formation in hemostasis, fibrin molecules are cross-linked by factor XIIIa. Factor XIIIa catalyzes the transglutamination reaction between Lys-406 of ␥C in one fibrin molecule and Gln-398 or Gln-399 in another to cross-link two neighboring dimers (28). Gln-399 immediately precedes the ␥C dodecapeptide, and in the dodecapeptide, Lys-406 is the Lys of KQAGDV that binds to ␣ IIb. This C-terminal portion of the fibrinogen ␥C-subunit is unstructured in the absence of binding to ␣ IIb ␤ 3. Although factor XIIIa covalently stabilizes fibrin, it at the same time makes Lys-406 unavailable for integrin binding. Because our results demonstrate that this Lys is not required for binding to ␣ IIb ␤ 3 , it is conceivable that the crosslinked fibrin may still interact with ␣ IIb ␤ 3 using the C-terminal AGDV sequence. Further investigations on ␣ IIb ␤ 3 binding to factor XIIIa-cross-linked fibrin are needed to understand the physiological relationship between fibrin cross-linking and binding to ␣ IIb ␤ 3 .