Mechanistic Insights from a Refined Three-dimensional Model of Integrin αIIbβ3

The integrin αIIbβ3 plays an important role in platelet function, and abnormalities of this protein result in a serious bleeding disorder, known as Glanzmann thrombasthenia. Although crystallographic data exist for the related integrin αVβ3, to date, there are no high resolution structures of integrin αIIbβ3 available in the literature. Therefore, it is still unclear how specific elements of the αIIb subunit contribute to integrin αIIbβ3 function. Here we describe a refined model of the αIIb N-terminal portion of integrin αIIbβ3 obtained by using the αVβ3 template combined with a new method for predicting the conformations of the unique αIIb loop regions comprising residues 71-85, 114-125, and 148-164. The refined model was probed based on a structural prediction that differentiates it from standard homology models: specifically, that Lys-118 of αIIb contacts Glu-171 of β3. To test this hypothesis experimentally, the mutant integrin chains αIIb K118C and β3 E171C were cotransfected into HEK 293 cells. We show that the cells expressed the mutants αIIbβ3 on their surface as a disulfide-linked dimer, supporting the close proximity between αIIb Lys-118 and β3 Glu-171 predicted from the refined model. This validated model provides a specific structural context for the analysis and interpretation of structure-function relations of integrin αIIbβ3. In addition, it suggests mechanistic hypotheses pertaining to both naturally occurring mutations responsible for Glanzmann thrombasthenia and to point mutations that affect ligand binding.

Integrin ␣ IIb ␤ 3 or glycoprotein IIb/IIIa is the megakaryocyteand platelet-specific receptor that mediates the primary interactions of platelets with fibrinogen and other ligands, leading to platelet aggregation. Mutations in either the ␣ IIb or the ␤ 3 subunit result in Glanzmann thrombasthenia, an autosomal recessive bleeding disorder affecting platelet function (see Refs. 1-3 for recent reviews). Although a large number of point mutations causing this disorder have been identified experimentally in the integrin ␣ IIb ␤-propeller (sinaicentral.mssm.edu/intranet/research/glanzmann/menu), the mechanistic details of how these mutations affect receptor biogenesis, protein structure, and receptor function are not well understood.
To date, no high resolution structures of ␣ IIb ␤ 3 are available to provide a structural context for understanding the mechanism of action of this complex biological system. A three-dimensional model of this protein was first proposed from homology modeling using the ␤ subunit of the G-protein transducin as a template (4). Subsequently, the crystal structures of a cognate template (␣ V ␤ 3 ) (5, 6) with higher identity in sequence between ␣ V and ␣ IIb (ϳ40%) as compared with transducin (ϳ19%) enabled the construction of more accurate overall three-dimensional models of ␣ IIb ␤ 3 (7,8) using standard homology-based modeling techniques. The ␣ V ␤ 3 crystallographic data support the folding of the N terminus of ␣ IIb into a seven-blade ␤-propeller, as was originally predicted by Springer (9), and identify the propeller as the major site of interaction between ␣ IIb and ␤ 3 . However, ␣ IIb differs from ␣ V in loop regions that are strategically located with regard to interaction with ␤ 3 and ligand binding, and standard homology-based modeling techniques do not usually provide accurate prediction of the structures of non-conserved loops. In particular, the ␣ IIb N-terminal region contains three interacting non-conserved loop regions (residues 71-85, 114 -125, and 148 -164) that deserve special attention for their position in the integrin complex. Loops 114 -125 and 148 -164 are located at the interface between the ␣ IIb and ␤ 3 subunits, and loop 71-85 is in proximity to, and thus potentially influences, loop 114 -125.
In general, segments (or loop regions) connecting elements of defined secondary structure in proteins are likely to be the more flexible portions of the structure. This flexibility can impart significant functional roles to these loops, allowing their structural adaptation during interaction with other regions of the protein, with ligands, or with other molecules. For this same reason, however, these segments are also more difficult to predict by homology and, in many cases, are difficult to resolve by high resolution crystallographic techniques due to their large thermal fluctuations or crystal disorder (10). Thus, the three ␣ IIb loops we have chosen for structural refinement are likely to contribute significantly to the unique features of ␣ IIb ␤ 3 , and their conformational preferences may be difficult to establish even with the benefit of crystallographic data.
We present here a refined model of an initial ␣ IIb structure obtained from a homology modeling protocol based on the ␣ V ␤ 3 crystal structure (5,6). Specifically, an ab initio molecular mechanics structure prediction method recently developed (10) and evaluated (11,12) in our laboratory was employed to explore the conformational space of the three flexible loop regions of ␣ IIb described above. The method uses the CHARMM PAR22 force field in conjunction with the screened coulomb potentialimplicit solvent model (SCP-ISM) 1 for electrostatics. The refined model of the N-terminal portion of ␣ IIb obtained by using this computational method was then validated experimentally. In particular, the intermolecular contact between Lys-118 of ␣ IIb and Glu-171 of ␤ 3 , predicted by our model but not a standard homology model, was supported by cross-linking of the corresponding cysteine mutants. In the absence of a high resolution experimental structure of the ␣ IIb ␤ 3 complex, we propose the refined model presented here as a structural context to generate experimentally testable hypotheses and guide the design of future experiments to explore the functional roles of key structural elements.
Stable Cell Lines Generation-HEK 293 cells were transfected with either normal or mutant cDNA using PerFectin (Gene Therapy System, San Diego, CA) according to the manufacturer's instructions. 48 h after transfection, cells were selected in media containing 500 g/ml G418 for 2 weeks. To obtain a population of cells expressing higher levels of human and chimeric receptors, cells were labeled with the mAb 10E5 (␣ IIb ␤ 3 -specific) (13) and sorted using a FACScalibur cell sorter (BD Biosciences).
Biochemical Analysis of the Receptors-To vectorially label surface molecules, 10 6 resuspended cells/ml were incubated on ice for 30 min with sulfo-N-hydroxyl-succinimido biotin (1 mg/ml in phosphate-buffered saline) (Pierce). The reaction was stopped by adding glycine (5 mM final concentration), and then the cells were lysed with buffer (150 mM NaCl, 50 mM Tris-HCl, pH 7.5) containing 1% Triton X-100 and protease inhibitors (protease inhibitors mixture set III; Calbiochem). Lysates were centrifuged at 4°C at 12,000 ϫ g for 30 min, and supernatants were precleared with protein A-or protein G-Sepharose (Amersham Biosciences) at 4°C for 30 min. Immunoprecipitations were performed using mAb 10E5 (8 g/ml). Protein G-Sepharose (5%) was added and incubated for 1 h at 4°C. The beads were washed twice with lysis buffer containing 500 mM NaCl, and then the bound protein was eluted with SDS-PAGE sample buffer at 100°C. Samples (reduced and non-reduced) were separated by SDS-PAGE (7.5% gel) and electro-transferred to polyvinylidene difluoride membranes. Membranes were blocked in 5% nonfat dry milk for 1 h at room temperature and washed. To identify biotin-labeled cell surface proteins immunoprecipitated by mAb 10E5, the membranes were hybridized with horseradish peroxidase-conjugated avidin (Amersham Biosciences) for 1 h at room temperature. Membranes were washed and developed using chemiluminescence, as per the manufacturer's instructions (Amersham Biosciences).

Computational Approach
The refined three-dimensional model of the N-terminal portion of the ␣ IIb subunit of integrin ␣ IIb ␤ 3 proposed here was built in two steps. First, a standard homology modeling-based approach using the crystal structure of the cognate template ␣ V ␤ 3 was employed to obtain an initial three-dimensional model of the ␣ IIb ␤ 3 ␤-propeller domain. Second, an ab initio molecular mechanics structure prediction method was used to refine the three non-conserved loop regions (residues 71-85, 114 -125, and 148 -164 of human ␣ IIb ) of the N-terminal portion of the ␣ IIb subunit discussed above.
Construction of an Initial Three-dimensional Model of the ␤-Propeller Domain of Integrin ␣ IIb ␤ 3 -The initial three-dimensional model of the ␣ IIb N-terminal portion (residues 1-451) of ␣ IIb ␤ 3 was based on homology with the ␣ V subunit (residues 1-438) of integrin ␣ V ␤ 3 (sequence identity of 41%), for which atomic coordinates are available from x-ray crystallography (5, 6), using MODELLER Release 6 (salilab.org/modeller/modeller.html) (14,15) and the sequence alignment shown in Fig.  1. The resulting model of ␣ IIb -(1-451) was then used to replace the corresponding segment of ␣ V within the known structure of ␣ V ␤ 3 in complex with a cyclic peptide containing the prototypical RGD sequence found in a number of integrin ligands (7) to obtain a three-dimensional model of the ligand-bound N-terminal portion of integrin ␣ IIb ␤ 3 , including ␣ IIb -(1-451) and ␤ 3 -(109 -352).
Refinement of the Structure of ␣ IIb -(1-451)-The three non-conserved loops 71-85, 114 -125, and 148 -164 of the N-terminal portion of integrin ␣ IIb were refined using the molecular mechanics ab initio computational method that we have recently developed to predict the structures of flexible loop regions in proteins (10). Details of this method and its application are reported elsewhere (10).
Briefly, the method employs simulated annealing Monte Carlo (SA-MC) simulations combined with biased scaled collective variables (SCV) Monte Carlo techniques (SCV-MC), which were specially designed to model segments with 8 -13 amino acid residues that include parts of the defined secondary structure motifs to which they are attached. The higher yield isotropic trial moves performed in the space of the scaled collective variables correspond to anisotropic trial moves in the space of the (real) dihedral angles (for details of the methodology, see Refs. 10 and 11)). To incorporate solvent effects, the method uses a continuum electrostatic model based on screened coulomb potentials (SCP-ISM) (12). The screening function used in the SCP-ISM and the theoretical and experimental basis of its form have been discussed and illustrated in detail (12,16,17). In this ISM approach, the electrostatic representation of the protein in the solvent is determined by following a standard thermodynamic path in which the protein particles are first solvated individually and subsequently brought together to form the solvated protein. General expressions for the electrostatic, E SCP and the polar component of the solvation free energies for macromolecules, ⌬GV s pol , are derived from this model. The work done in each step is evaluated, and the total electrostatic energy E SCP of the macromolecule (or "effective energy" in the nomenclature of Ref. 18) is calculated (non-electrostatic effects are treated separately). The electrostatic energy is calculated from the equation, where q i are the partial charges of the particle i, n is the total number of atoms in the protein, D(r) are the screening functions of the electrostatic interactions within the protein in the solvent, and R i,B are the effective Born radii of the atom i in the protein, estimated as described in Refs. 12, 16, 17, and 19. Since a special treatment of hydrogen bonds has been shown to be necessary when continuum models are used (12,16), the short range donor-acceptor interactions were stabilized in the SCP-ISM independently of the strength of the long range electrostatic interactions that are controlled by the macroscopic screening D(r). The algorithm developed in the SCP-ISM stabilizes each single donor-acceptor partner available in the protein according to experimental hydrogen bond values, and a geometrical dependence of these interactions is introduced based on the state of hybridization of both acceptor and donor groups (16). Hydrophobic interactions (including entropic effects of the solvent and energy of cavity formation) are treated as a term proportional to the solvent-accessible surface area (SASA) of the molecule, i.e. E hp ϭ ␣ ϩ ␤SASA, where ␣ and ␤ are parameters usually obtained by fitting to solvation energies of hydrocarbons.
To refine the three loop regions 71-85, 114 -125, and 148 -164 of the N-terminal portion of integrin ␣ IIb , the procedure consisted of a first SA-MC phase followed by three distinct SCV-MC simulations. The first SA-MC phase of the method explored the conformational space for each fully solvated loop, tethered at the N terminus to a rigid protein segment of similar length. The internal structural preferences of each loop, including the effect on local folding of the region of the protein that is attached to the N terminus, were identified in this phase. The starting temperature in the SA-MC simulations performed was T ϭ 3000 K, and the final temperature was T ϭ 300 K. The three loop regions were initially placed in their extended conformations. A logarithmic schedule was used for the annealing in which the temperature was decreased in 12 steps. Trial moves of 29,000, 22,500, and 30,000 per temperature for loops 71-85, 114 -125, and 148 -164, respectively, were performed by selecting at random two dihedral angles (backbone or side chain) of the segment. 10 independent runs were performed for each segment, and structures were collected at the final temperature of 300 K. Specifically, 1148 different conformations were identified for loop 148 -164, 768 were identified for loop 114 -125, and 1106 were identified for loop 71-85. 100 representative conformations were selected from each of these three sets based on root mean square deviation values calculated among all conformations within a specific set (for details of the procedure, see Ref. 10). These representative conformations were considered as potential candidates to fold into the native structure of the segment in the following steps of the protocol.
In the subsequent phase of the method, a first application of SCV-MC drove the C terminus of each representative loop conformation toward its final attachment point by using an adjustable harmonic constraint protocol imposed on the dummy residue attached to the C terminus. Starting with a force constant k ϭ 0, an SCV-MC simulation at T ϭ 300 K was carried out for each of the representative conformations obtained previously to relax the peptide around the closest local minimum in the free energy surface. This first simulation allowed the initial conformation of the segment to adjust to the characteristics of the new environment while partially preserving its intrinsic structural conformation calculated in the first step. The eigenvalues of the Hessian were updated twice to maintain the acceptance rate in the range 0.3-0.5; a total of 7000 SCV-MC steps were performed. Next, the harmonic constant k was increased in successive steps to facilitate the shift of the C terminus toward the attachment point. The migration of the conformations of the segment from its intrinsic folding to its "closed" folding was simulated with the SCV-MC for each particular value of the harmonic constant k. For each k, the simulation protocol was the same as described above for k ϭ 0, with an acceptance rate maintained in the range 0.3-0.5 in all cases. In previous calculations, it was observed that a power schedule of the form k ϭ 10 mϪ6 kcal/mol/Å 2 , where m is an integer that represents the successive increment of the constant k (m ϭ 1, . . . , 10), was appropriate in practice to slowly close the segments, and the same protocol was followed here.
After performing 250 minimization steps to eliminate bad contacts that could remain after the SCV-MC simulation, the closed conformations of each loop were sorted by energy. The lowest energy conformations within a 5 kcal/mol range were then selected for each loop. Specifically, the procedure yielded three different conformations for loop 71-85, 5 for loop 114 -125 and six different conformations for loop 148 -164 within this energy range. Assembly of all of these conformations into the protein yielded 90 (ϭ 6 ϫ 5 ϫ 3) different conformations of the three interacting loops in the ␣ IIb -(1-451) subunit. A second application of SCV-MC, using the same protocol described above for the first SCV-MC simulation, allowed relaxation of the three interacting loops in the environment created by the native protein and the solvent, where all the loops could move and interact simultaneously. This step produced the optimized structure of the ␣ IIb -(1-451) subunit with the three refined loops, which was then used to replace the initial model of ␣ IIb in the N-terminal portion of the ␣ IIb ␤ 3 complex. Specifically, this optimized structure corresponded to the lowest energy conformation among the 90 structures that were relaxed using SCV-MC.
The three interacting loops of the ␣ IIb -(1-451) optimized structure were subjected to a final relaxation process in the presence of the   148 -164). The refined model of the N-terminal portion of ␣ IIb differs from a standard homology modeling-based structure in the conformational preferences of loops 71-85, 114 -125, and 148 -164. Interestingly, ␣ IIb residues that, in the refined model, are involved in intermolecular interactions with the "specificitydetermining loop" (SDL) of the ␤ 3 subunit (residues 159 -188 in human ␤ 3 ) were not predicted to be located at the interface between the subunits in a previous model obtained from standard homology modeling. However, the predictions of ␣-␤ intermolecular contacts are key to understanding the biogenesis and function of the receptor. Most importantly, the predictions are amenable to experimental verification that can test the validity of the refined ␣ IIb model proposed here.
Validation of the Refined Model-Among the intermolecular interactions predicted using our refined model was the interaction between ␣ IIb Lys-118 and ␤ 3 Glu-171. As shown in Fig. 3, in our refined model of the N-terminal portion of ␣ IIb (in light blue), the location of Lys-118 allows it to form a salt bridge with ␤ 3 Glu-171, whereas in a standard homology modeling-based model built using MODELLER 6.0 (in magenta), Lys-118 is far away from the interface. To test the validity of this predicted intermolecular interaction, we reasoned that the proximity of these residues could be determined by substituting them with cysteines and testing their ability to form disulfide bonding between ␣ IIb and ␤ 3 . A similar strategy was recently employed by Takagi et al. (20) to demonstrate the proximity of ␣ IIb Arg-320 to ␤ 3 Arg-563, and this construct was used as a control. Fig. 4 demonstrates that when we mutated both ␣ IIb Lys-118 and ␤ 3 Glu-171 to cysteines, the resulting heterodimer was expressed on HEK 293 cells, as judged from mAb 10E5 binding. The level of 10E5 binding to the mutant construct was comparable with that of cells transfected with normal human ␣ IIb and ␤ 3 or cells transfected with ␣ IIb R320C and ␤ 3 R563C. When biotin-labeled surface proteins from cells transfected with normal human ␣ IIb ␤ 3 were immunoprecipitated with mAb 10E5 and then analyzed by SDS-PAGE, the ␣ IIb and ␤ 3 subunits demonstrated their typical patterns of migration under non-reducing conditions (␣ IIb ϳ140 kDa, ␤ 3 ϳ90 kDa) and reducing conditions (␣ IIb ϳ120 kDa, ␤ 3 ϳ110 kDa). In contrast, and consistent with the previous report by Takagi et al. (20), the ␣ IIb R320C-␤ 3 R563C heterodimer migrated under nonreducing conditions as a dimer of ϳ230 kDa; with reduction of disulfide bonds, the dimer disappeared, and the individual subunits migrated at approximately their expected positions (ϳ120 and 110 kDa, respectively). Therefore, the dimer obtained under non-reducing conditions likely reflects a disulfide bond between ␣ IIb R320C and ␤ 3 R563C. The results with ␣ IIb K118C and ␤ 3 E171C are nearly identical to those with ␣ IIb R320C and ␤ 3 R563C, with the presence of a dimer under non-reducing conditions and the individual chains under reducing conditions. To assess whether the cross-linking was actually occurring between the ␣ IIb Cys-118 and ␤ 3 Cys-171, we tested the ability of antibody 7E3 to immunoprecipitate the complex since we have recently localized the 7E3 epitope to an adjacent region in ␤ 3 including the Cys-177-Cys-184 loop and Trp-129 (21). 7E3 was able to precipitate ␣ IIb ␤ 3 from cells expressing normal ␣ IIb and ␤ 3 and cells expressing the combination of ␣ IIb R320C and ␤ 3 R563C mutants. It was not, however, able to precipitate the ␣ IIb Cys-118-␤ 3 Cys-171 mutant protein. Similarly, cells expressing the ␣ IIb Cys-118-␤ 3 Cys-171 mutants did not adhere to immobilized fibrinogen, whereas cells expressing native ␣ IIb ␤ 3 and mutant-expressing cells treated with 2 mM dithiothreitol adhered to immobilized fibrinogen (data not shown). These data support the conclusion that the disulfide cross-linking between ␣ IIb Cys-118-␤ 3 Cys-171 occurred between the two newly introduced cysteines and thus reinforce the prediction from our refined model, but not the standard homology model, that ␣ IIb Lys-118 and ␤ 3 Glu-171 are in close proximity in the normal, solvated molecule. DISCUSSION We present a refined model of the N-terminal portion of the ␣ IIb subunit of integrin ␣ IIb ␤ 3 that differs significantly from the homology models built on the ␣ V template (7,8). The refinement involves the structures of the three non-conserved loop regions (residues 71-85, 114 -125, and 148 -164 of human ␣ IIb ) that deserve special attention for their strategic positions in the integrin complex, where they are likely to play functional roles in both heterodimer formation with the ␤ 3 subunit and ligand binding. The refinement was obtained using an ab initio molecular mechanics structure prediction approach based on energy calculations and simulation protocols developed recently in our laboratory (10) and evaluated for various molecular systems (11,12).
Characteristics of the Loop Refinement Approach-The novel method we used for loop structure prediction reflects the key requirements for reliability and predictive power based on energy calculations: (i) the force field used for the calculation of energies must be able to discriminate among the many possible local minima corresponding to the different conformations of the segments in the context of the native protein and (ii) the sampling method must be able to reveal the existence of those minima despite the formidable challenges that occur as the length of the segment increases. The computational method used here to refine the N-terminal portion of ␣ IIb was originally designed to address both of these requirements to model segments that connect regions with defined secondary structures in proteins (10). The unique features of this method, which was shown to reproduce the proper folding at the two ends of the segments as well as the correct H-bond pattern in test case peptides (e.g. see Ref. 10), are outlined below.
Unlike other loop closure techniques, the method used here is based on the assumption that segments connecting elements of secondary structure in proteins, in particular loops, have an intrinsic propensity for a particular set of conformations based on their amino acid sequence, but this intrinsic folding has to be disrupted for the best fit of the segment within the tertiary structure of the native protein. Thus, the final folding of the segments in the context of the protein is a compromise between two opposite effects: the intrinsic tendency to adopt a specific folding pattern dictated by the amino acid sequence and the partial unfolding that is imposed by the inclusion of the loop in the native conformation of the protein. The sampling methodology employed by our method follows these two processes in a rational and efficient way (see details in Ref. 10). Thus, it first uses SA-MC simulations to find conformations that are representative of the segment structure in solution, as encoded in the primary sequence, and subsequently forces a slow unfolding of the segment to fit the final protein conformation using an adjustable force constant scheme and MC simulations with a scaled collective variables technique. The scaled collective variables technique allows the MC simulation to improve the efficiency of the search. In fact, the strong anisotropy of the free energy surface of peptides and proteins makes a standard MC simulation with a Metropolis algorithm highly inefficient due to the high probability of rejecting trial moves, especially in the neighborhood of a local minimum. If plain MC simulation were performed for such cases, small changes in each trial conformation would have to be adopted to keep the acceptance rate high enough for statistical significance and correct convergence in the calculation of thermodynamic quantities. This requirement restricts the exploration of the conformational space, compromising the quality of the results. However, it was previously shown (22) that the sampling can be improved if the trial moves are chosen anisotropically in the conformational space, giving more preference to the movements along the soft directions (low energy barriers) and reducing the movement along the hard directions (rough surface due mainly to steric effects). The scaled collective variables technique makes this possible by allowing the MC simulation to sample the space in this anisotropic way, and therefore, it improves the efficiency of the search we perform. Finally, since an accurate force field for the study of peptide and protein conformational preferences must account for the hydrophobic and electrostatic effects of the solvent, our method uses a continuum electrostatic model based on screened coulomb potentials, which has been validated in a number of systems ranging in size from small molecules to large proteins (12,16,17,19).
Mechanistic Implications of Structural Details Revealed by the Refined Model-The refined model of the N-terminal portion of ␣ IIb obtained by using the computational method discussed above was supported experimentally by demonstrating that cysteine mutants of Lys-118 of ␣ IIb and Glu-171 of ␤ 3 that are predicted from our model, but not a standard homology model, to be in close proximity resulted in a disulfide-bonded dimer of ␣ IIb ␤ 3 . To provide further support for the location of the cross-link responsible for heterodimer formation, we took advantage of our recent localization of the antibody 7E3 epitope to the region near ␤ 3 Glu-171 (21) and demonstrated that the cross-link led to the selective loss of 7E3 binding. Although these data support our model, we recognize that they only a Indicated residues, when mutated to alanine, resulted in decreased binding of one or more ligand-mimetic antibodies (LMA) or one or more non-ligand mimetic antibodies (NLMA).
b The proposed interacting ␤ 3 partners (residues of ␤ 3 in which backbone and/or side chain atoms are within 5 Å of residues at the subunit interface) are reported for ␣ IIb residues at the interface between subunits FIG. 5. Residues that affect ligand binding (magenta and orange) in the context of our refined ␣ IIb -(1-451) model (in green) and the ␤ 3 subunit (in red). In particular, residues in orange belong to the loops refined here.
indicate that the cysteine-mutated residues Lys-118 of ␣ IIb and Glu-171 of ␤ 3 come close to each other sometime during biogenesis to form the disulfide bond.
In addition to the ␣ IIb Lys-118-␤ 3 Glu-171 interaction, our refined model differs from a standard homology-based model in a number of ways. Most importantly, the refined model predicts that ␣ IIb residues Leu-116, Lys-124, and Arg-153 are close to one or more residues of the ␤ 3 SDL region (residues 159 -188). Specifically, ␤ 3 Pro-170 is within 5 Å of ␣ IIb Leu-116; the ␤ 3 Ile-167, Ser-168, and Pro-169 are within 5 Å of ␣ IIb Lys-124; and ␤ 3 Tyr-166, Asp-179, and Met-180 are within 5 Å of ␣ IIb Arg-153. These interactions are likely to account in large part for the ligand specificity of ␣ IIb ␤ 3 relative to ␣ V ␤ 3 since the shorter corresponding loops of ␣ V do not interact with the ␤ 3 SDL. Interestingly, replacement of one of these residues, Leu-116, by a valine results in the Glanzmann thrombasthenia phenotype (23). Since in our model, but not the standard homology model, this residue is at the ␣ IIb ␤ 3 interface, and only 5 Å from the ␤ 3 SDL, our hypothesis is that a mutation at this position alters the ligand binding properties of the heterodimeric complex.
Previous studies (4) using site directed mutagenesis to substitute alanine for other amino acids identified residues in ␣ IIb that, when mutated to alanine, affect the binding of fibrinogen, ligand-mimetic antibodies, or non-ligand-mimetic antibodies to ␣ IIb ␤ 3 . Table I lists the residues contained in the loops we refined that were identified as being important in fibrinogen binding. These residues (Asp-74, Leu-84, Asn-114, Glu-117, Lys-124, Thr-125, Tyr-155, Phe-160, Asp-163, and Lys-164) are shown in orange in Fig. 5 (␣ IIb is shown in green, and ␤ 3 is shown in red). For each of these ␣ IIb residues, we used the atomic coordinates of our ␣ IIb -(1-451)-␤ 3 -(109 -352) model to calculate solvent accessibility values (24) and predict whether they are internal in the protein (and thus most likely important for proper folding of the protein), surface-exposed (and thus likely in a position to contact ligand directly), or both surfaceexposed and at the interface with ␤ 3 (and thus potentially involved in stabilizing ␣ IIb -␤ 3 interactions and/or directly contacting ligand). Table I reports the position (internal, external, and/or interface) of these residues in the protein, the effect of alanine substitution on mAb binding, and the proposed ␤ 3 partners (␤ 3 residues with backbone and/or side chain atoms located within 5 Å of ␣ IIb ) of residue Lys-124, which was identified as being at the interface between the subunits.
Among the residues reported in Table I, Asn-114, Thr-125, Asp-163, and Lys-164 were predicted to be internal in the ␣ IIb subunit. The D163A substitution (4) affected the binding of fibrinogen, ligand-mimetic, and non-ligand-mimetic antibodies, suggesting that this residue has a generalized effect on the ␣ IIb ␤ 3 structure. Since the other residues in Table I that are predicted to be internal in ␣ IIb had little or no effect on the binding of either category of mAb (4), presumably they cause protein folding changes that affect fibrinogen binding more than the binding of RGD-containing peptides. Notably, the ligand-mimetic antibodies contain RYD sequences in their binding sites (25) so that our observations may provide insights into the differences in binding of the fibrinogen ␥-chain peptide as compared with RGD-containing peptides.
Residues Asp-74 and Phe-160 are predicted by our model to be on the surface of ␣ IIb but not at the ␣ IIb -␤ 3 interface, and their mutations to alanine affected fibrinogen binding much more than the binding of ligand-mimetic or non-ligand-mimetic antibodies (4). Therefore, they too may provide insights into the unique features of fibrinogen binding.
The ␣ IIb residue Lys-124 is predicted to be at the interface of ␣ IIb and ␤ 3 with three potential partners within 5 Å. Alanine mutation affects fibrinogen binding but not the binding of the mAbs. Thus, interaction of Lys-124 with one or more ␤ 3 partners may influence the interface in a way that affects the binding of fibrinogen to a greater extent than the binding of mAbs (4).
The ␣ IIb residues Leu-84 and Tyr-155 are also predicted to be on the surface, but not at the interface between the subunits. When mutated to alanine, they have a negative effect on the binding of both fibrinogen and one of the ligand-mimetic antibodies but not the non-ligand-mimetic antibodies (4). Thus, these residues appear to be important in the binding of both RGD-containing peptides and fibrinogen, without causing marked disruption of the entire molecule. In contrast, alanine mutation of ␣ IIb Glu-117, which is also on the surface but not at the interface between the subunits in our model, affects the binding of both fibrinogen and one of the non-ligand-mimetic antibodies but has little effect on ligand-mimetic antibodies (4).
Concluding Remarks-The refined and experimentally validated three-dimensional model of the N-terminal portion of integrin ␣ IIb ␤ 3 proposed here provides a structural context for the interpretation of elements of the ␣ IIb subunit of integrin ␣ IIb ␤ 3 that are critical for biogenesis and function. In particular, analysis of the model offers testable mechanistic hypotheses for the role of residues within three refined ␣ IIb loops that are involved in ligand binding and subunit association; it also allows for analysis of naturally occurring mutations that produce Glanzmann thrombasthenia. Large-scale molecular dynamics simulations of the proposed model are currently ongoing in our laboratory to investigate the dynamic effects of the proposed intermolecular interactions on the integrin structure. This model has been deposited in the Protein Data Bank with the accession code 1RN0.