Ligand Binding Promotes the Entropy-driven Oligomerization of Integrin αIIbβ3 *

Integrin αIIbβ3clusters on the platelet surface after binding adhesive proteins in a process that regulates signal transduction. However, the intermolecular forces driving integrin self-association are poorly understood. This work provides new insights into integrin clustering mechanisms by demonstrating how temperature and ligand binding interact to affect the oligomeric state of αIIbβ3. The ligand-free receptor, solubilized in thermostable octyl glucoside micelles, exhibited a cooperative transition at ∼43 °C, monitored by changes in intrinsic fluorescence and circular dichroism. Both signals changed in a direction opposite to that for global unfolding, and both were diminished upon binding the fibrinogen γ-chain ligand-mimetic peptide cHArGD. Free and bound receptors also exhibited differential sensitivity to temperature-enhanced oligomerization, as measured by dynamic light scattering, sedimentation velocity, and sedimentation equilibrium. Van't Hoff analyses of dimerization constants for αIIbβ3 complexed with cHArGD, cRGD, or eptifibatide yielded large, favorable entropy changes partly offset by unfavorable enthalpy changes. Transmission electron microscopy showed that ligand binding and 37 °C incubation enhanced assembly of integrin dimers and larger oligomers linked by tail-to-tail contacts. Interpretation of these images was aided by threading models for αIIbβ3 protomers and dimers based on the ectodomain structure of αvβ3. We propose that entropy-favorable nonpolar interactions drive ligand-induced integrin clustering and outside-in signaling.

Integrins are a widely distributed family of heterodimeric transmembrane receptors that anchor cells to extracellular matrix proteins and mediate two-way communication between the exterior and interior of a cell (1,2). The ␣ IIb ␤ 3 complex is the classic example of a regulated integrin, a receptor whose affinity for adhesive macromolecules is modulated by changes in conformation and clustering (3)(4)(5). The receptor is maintained in a default inactive state on circulating human blood platelets, possibly by interactions between the short cytoplasmic domains of the ␣ IIb and ␤ 3 subunits or with integrinassociated proteins (6 -9). According to this model, proteins like fibrinogen or fibronectin are prevented from binding to extracellular region of ␣ IIb ␤ 3 until a platelet stimulus, such as thrombin, binds to a G-protein-coupled receptor (3,10,11). A rapid cascade of intracellular events releases an inhibitory lock, sending a signal some 15 nm outwards to the ectodomain of ␣ IIb ␤ 3 , where a rearrangement of intersubunit contacts leads to an "open" receptor with a functional binding site (12)(13)(14). This process of converting ␣ IIb ␤ 3 to an active form is referred to as "inside-out" signaling (3).
Receptor occupancy then sends an "outside-in" signal, leading to integrin clustering and downstream activation of kinases, especially focal adhesion kinase, that stabilize the integrin-mediated links between extracellular matrix proteins and actin filaments (15)(16)(17). Clustering of bound receptors may facilitate these processes by increasing the local concentration of integrin-associated proteins, especially those that bind to the cytoplasmic tail of ␤ 3 (4,18,19). A positive feedback mechanism may also be at work on the cell surface, in that integrin oligomers could be especially efficient at capturing multimeric adhesive proteins (15,20).
Our understanding of the molecular basis for integrin activation has been considerably enhanced by the recent publication of crystal structures for the extracellular domain of the ␣ v ␤ 3 integrin, in the absence and presence of ligand (21,22). However, many questions remain about the mechanistic details of the biomechanical coupling between the distant intraand extracellular domains of ␣ IIb ␤ 3 . Previous experiments have shown that truncating either the ␣ IIb or ␤ 3 cytoplasmic regions yields a receptor that is constitutively active for inside-out signaling, indicating that communication between the cytoplasmic domains maintains the inactive state of the integrin (23,24). We have recently shown that a truncation mutant, lacking the ␣ IIb cytoplasmic domain, underwent further activation in the presence of a high affinity fibrinogen mimetic peptide resulting in the formation of oligomers (25). These observations reinforce the concept that clustering plays a major role in the regulation of integrin affinity through outside-in signaling (13,15,26).
Concerning the forces that drive integrin clustering, we and others have demonstrated that the ability of RGDX peptides to block the function of ␣ IIb ␤ 3 (27) as well as to perturb the conformation of its ectodomain and to promote "tail to tail" oligomerization all increased with the hydrophobicity of the residue in the X-position (28,29). Likewise, we found that eptifibatide, a cyclized fibrinogen-mimetic peptide with a tryptophan residue in its integrin-targeting sequence (30), was . The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. especially effective at promoting ␣ IIb ␤ 3 oligomerization, as monitored by sedimentation equilibrium and electron microscopy (31). These observations have led to the concept that nonpolar receptor-ligand interactions contribute to both integrin activation and self-association (28,31). This article tests that hypothesis by investigating the effects of both temperature and ligand binding on ␣ IIb ␤ 3 structure/stability.
If hydrophobic effects are important in these interactions, thermodynamic principles (32)(33)(34) and experience with other self-assembling systems (35)(36)(37) predict that the extent of ␣ IIb ␤ 3 oligomerization should increase with increasing temperature. However, extracting thermodynamic data on a transmembrane protein such as the ␣ IIb ␤ 3 integrin requires pure protein isolated in a thermally stable environment amenable to spectroscopic studies. The neutral, nondenaturing detergent octyl glucoside (OG) 1 (38,39) is especially well suited for biophysical characterizations of integral membrane proteins because it exhibits a critical micellar concentration and aggregation number that are nearly invariant from 20 to 60°C (40). Octyl glucoside has been used to study the structure and dynamics of bacteriorhodopsin (41), mammalian rhodopsin (42), OmpF porin from Escherichia coli (43), and the E. coli outer membrane ferrichrome transporter FhuA (44). Extensive biophysical characterizations have shown that octyl glucoside, even at a concentration well above its critical micellar concentration, does not perturb the quaternary structure of the soluble lens protein ␣-crystallin (45,46). Furthermore, oligomerization studies of subunits B777 (47) and B820 (35) of the light-harvesting complex from Rhodobacter sphearoides have been performed in octyl glucoside micelles.
Based on our experience (28,31,49) and that of others (50) with biophysical characterizations of the ␣ IIb ␤ 3 integrin in octyl glucoside, we recognize the limitations of extrapolation from a detergent-solubilized protein to the physiological situation. However, the responses of solubilized integrins to ligand binding, such as conformational changes and oligomerization (28,31), have been remarkably similar to the effects seen in cellular membranes (20,51). In addition, Litvinov et al. (52) recently demonstrated that both purified ␣ IIb ␤ 3 and ␣ IIb ␤ 3 on the platelet surface displayed comparable force histograms for fibrinogen binding. These observations probably reflect the fact that the ectodomain of ␣ IIb ␤ 3 contains Ͼ90% of the integrin residues and only one narrow nonpolar segment on each subunit traverses the plasma membrane (28). Thus the model of a circumferential transmembrane belt comprised of a limited number of protein-bound detergent (43,53) or lipid molecules (54,55) established from high resolution crystal structures of integral membrane proteins supports the validity of the studies of ␣ IIb ␤ 3 in octyl glucoside micelles described here.
Platelet Isolation and Characterization Procedures-Platelet-rich plasma and gel-filtered platelets were isolated from blood obtained by venipuncture from healthy, adult, volunteer donors as previously described (58). Platelet counts were determined with a Coulter MDII Cell Counter (Beckman Instruments, Miami, FL). Platelet ␣ IIb ␤ 3 occupancy was determined with flow cytometric analysis using a Biocytex kit BX7001 (Marseilles, France) (25,59). Platelet aggregation profiles were obtained in a Chrono-Log model 500 aggregometer. Platelet adhesion to fibrin was determined in a microtiter plate adhesion assay with colorimetric read-out as previously described (25).
Biophysical Measurements-UV absorbance measurements were performed as a function of wavelength on a Beckman diode array spectrophotometer with ␣ IIb ␤ 3 samples contained in 1-cm path length, 0.1-ml volume quartz cuvettes. Circular dichroic spectroscopy was performed as a function of temperature from 20 to 60°C in a Jasco model 720 spectropolarimeter (Japan Spectroscopic Co., Tokyo) with samples in a thermostatted 0.05-cm path-length cuvette; data are expressed as molar ellipticity, [⌰], versus wavelength (25). In computing [⌰], each data set was corrected for the signal from the HSC-OG buffer or buffered octyl glucoside containing the ligand-mimetic peptide cHArGD at the same temperature. However, these background measurements changed by only Ϯ5% over the range 20 -60°C.
Fluorescence emission spectra were obtained over the same temperature range with an Aminco-Bowman series 2 luminescence spectrometer (SLM-Aminco, Rochester, NY) with samples contained in thermostatted quartz microcuvettes (Hellma Cells, Inc., White Plains, NY); an excitation wavelength at 278 nm was used (31). These data were also corrected for background fluorescence; the signal from HSC-OG buffer was found to vary by Ϯ6% over the range 20 -60°C. The small fluorescence intensity from the tryptophan-containing ligand-mimetic peptide cHArGD in HSC-OG exhibited an ϳ2-fold hyperbolic decrease with increasing temperature, as expected for intrinsic fluorescence measurements (60).
Static and dynamic light scattering measurements were performed in a Brookhaven Instruments BI-2030 AT correlator operated in conjunction with a BI-200 SM light scattering photometer/photon counting detector and a Spectra Physics 127 He-Ne laser (28); samples in quartz microcuvettes were contained in a thermostatted refractive index matching vat at temperatures in the range 20 -40°C. For translational diffusion coefficient determinations, each intensity-normalized photon count autocorrelation function obtained for the ␣ IIb ␤ 3 complex was first corrected for the contributions of octyl glucoside micelles. Following procedures described in our earlier work (49), the autocorrelation function was treated as an intensity-weighted sum of two exponential decay components, one corresponding to the macromolecule and the other to the faster moving detergent micelles. The contribution of the detergent was determined from separate measurements of buffered octyl glucoside, obtained under the same instrumental conditions. The resultant signal was weighted by its fractional intensity, then subtracted from the corresponding autocorrelation function for the integrin:detergent mixture to yield a corrected autocorrelation function. Each corrected autocorrelation function was then analyzed by the method of cumulants to obtain a z-average translational diffusion coefficient for the ␣ IIb ␤ 3 complex. Size distribution information was also obtained from these corrected autocorrelation functions with the CONTIN algorithm (61). We note that CONTIN analysis did not consistently identify a peak corresponding to the ϳ5-nm diameter OG micelles in the uncorrected data, probably because we worked under conditions where the deter-gent contributed less than 20% of the total scattering intensity. All translational diffusion coefficients reported here have been corrected for solvent viscosity to obtain D 20,w and Stokes radius (R s ) values.
Data collected from buffered HSC-OG indicated that the octyl glucoside micelle size distribution did not change appreciably from 20 to 40°C in that the scattering intensity varied by Ϯ13% and the mean particle diameter changed by Ϯ12%; no temperature-dependent trends were observed in these data. Our results obtained with 30 mM OG are consistent with the report of Aoudia and Zana (40), who found that both critical micellar concentration for octyl glucoside (23-25 mM) and aggregation number (107-92) exhibited minimal temperature dependence from 20 to 60°C.
Sedimentation velocity and equilibrium measurements were performed in a Beckman Optima XL-A analytical ultracentrifuge (Beckman) equipped with absorbance optics and an An60 Ti rotor (28,31). By always including HSC-OG buffer or buffer ϩ ligand-mimetic peptide in the reference compartment, the resultant optical signals were corrected for any temperature-induced effects on the solvent, although data previously described demonstrated these effects to be minimal. Sedimentation velocity data obtained at 20 and 40°C were analyzed using both SVEDBERG (version 1.04) and DCDTϩ (version 6.31) software (J. Philo, Thousand Oaks, CA) to obtain the weight average sedimentation coefficient (s w ) and distribution of sedimenting species, g(s*), respectively (62). All sedimentation coefficients reported here have been corrected for solvent density and viscosity to obtain s 20,w values.
The absorbance versus radial distance data obtained by sedimentation equilibrium were analyzed by nonlinear regression with WinNON-LIN3 (63) to obtain weight average molecular weights (M w ) for the ␣ IIb ␤ 3 complex alone and in the presence of ligand-mimetic peptide. M w data were analyzed for their dependence on rotor speed (at 6000 and 8000 rpm), a diagnostic for irreversible aggregation (64). In the absence of ligands, data obtained with ␣ IIb ␤ 3 yielded a ratio M w (6000 rpm)/M w (8000 rpm) of 0.905 at 30°C, 0.839 at 37°C, and 0.737 at 40°C. In addition, global fits to data obtained at 37 and 40°C yielded a variance of ϳ3 ϫ 10 Ϫ3 , a 5-fold decrease in quality compared with fits obtained with data at lower temperatures. These observations are indicative of thermal aggregation for the ligand-free integrin (64).
Because depletion of larger species can also influence the resultant M w data, the quantity of absorbing material present in cells containing ␣ IIb ␤ 3 at equilibrium was computed and compared with that initially added. This was done by numerical integration of the absorbance versus radial distance profiles obtained at 30 and 40°C, as well as the initial, pre-equilibrium scans obtained at 3000 rpm. Whereas ϳ100% recovery was achieved at 6000 rpm/30°C, this parameter fell to ϳ68% at 8000 rpm/30°C; because only absorbance values Ͻ3.0 were included, there is some uncertainty associated with these estimates. In addition, only data at absorbence Ͻ1.5 were included in the subsequent analyses to ensure compliance with Beer's law, thus 47% of the integrated area was used for fitting at 6000 rpm and 42% at 8000 rpm. These effects became more pronounced at 40°C where recoveries of 42 and 28% were obtained at 6000 and 8000 rpm, respectively; 33% of the 6000 rpm data and 25% of the 8000 rpm data were used for fitting.
In contrast, M w data obtained for the ligand-bound integrins showed a somewhat less pronounced dependence on rotor speed and temperature: for example, the ratio M w (6000 rpm)/M w (8000 rpm) was 0.907 at 30°C, 0.851 at 37°C, and 0.765 at 40°C for the ␣ IIb ␤ 3 ⅐cHArGD complex. Global fits of data obtained in this temperature range, at two rotor speeds, and receptor concentrations in the range 1.4 to 4.0 M, yielded a variance of ϳ7 ϫ 10 Ϫ4 , indicating the data obtained for the ligandbound integrin were more consistent with oligomerization, rather than the aggregation behavior observed for the free receptor. Sample depletion was also less pronounced with the ligand-bound integrin samples, as recoveries of 94 Ϯ 10% and fitted ranges of 42 Ϯ 5% resulted (average of data at 6000 and 8000 rpm, 30 and 40°C). Therefore, data obtained with the integrin-ligand complexes were subjected to additional analyses with WinNONLIN3 to obtain a set of temperature-dependent selfassociation constants (64). WinNONLIN3 has been provided by Dr. David Yphantis and the staff at the National Analytical Ultracentrifugation Facility, Storrs, CT.
Transmission Electron Microscopy-␣ IIb ␤ 3 samples at ϳ1 mg/ml in HSC-OG (in the presence/absence of 20 mol/liter cHArGD) were incubated for 3 h at either 20 or 37°C, then diluted to ϳ20 -25 g/ml in a buffer containing 0.05 mol/liter ammonium formate at pH 7.4, 30 mmol/ liter octyl glucoside, and 15% (v/v) glycerol. Samples were then sprayed onto freshly cleaved mica and shadowed with tungsten in a vacuum evaporator (Denton Vacuum Co., Cherry Hill, NJ) (65-67). These samples were examined in a Philips 400 electron microscope (FEI Co., Hillsboro, OR) operating at 60 kV and a magnification of ϫ60,000. Counts of molecules with different conformations or different amounts of oligomers were made from prints of the micrographs, using images from many different areas of several different preparations to get a random sample.
Threading/Homology Modeling-Models of the ␣ IIb ␤ 3 integrin were constructed using the crystal structure of the extracellular domain of homologous receptor ␣ v ␤ 3 (21) as a template. The resources at SwissPdB Viewer (68) (www.expasy.ch.spdbv/) were used to thread the sequence of the ␣ IIb subunit (P08514) into the three-dimensional structure of ␣ v (Protein Database file 1JVA) (21). Initially, a model was developed using the "bent" structure observed crystallographically (21). Because regions of the ␤ 3 subunit corresponding to epidermal growth factor modules 1 and 2 were not well defined in the crystal structure (21), these segments were replaced with homology models (68) developed from NMR structural data reported by Beglova et al. (69). Subsequently, an "extended" ␣ v ␤ 3 template was generated by rotating and translating each subunit about its flexible pivot as outlined by Xiong et al. (21). Thus, we obtained an extended ␣ IIb model that exhibited a C␣ root mean square deviation Ͻ1 Å with the ␣ v structure; however, sequence gaps and misalignments that could not be resolved precluded the development of a complete homology model. Replacing the coordinates of the ␣ v with those of our threaded subunit yielded a model for the extracellular domain of the ␣ IIb ␤ 3 integrin.
The ␣ IIb and ␤ 3 transmembrane domains were modeled as helical segments, based on packing principles for membrane-bound polypeptide chains. The helical segment for ␣ IIb was extended toward its membrane-proximal region, based on NMR data, which also defined the conformation of its cytoplasmic domain (6). Secondary structure prediction algorithms identified a short helical segment in the cytoplasmic domain of ␤ 3 (71), a concept reinforced by the NMR study of Ulmer et al. (72). These segments were joined to the ␣ IIb ␤ 3 ectodomain to obtain a model of the complete integrin. Integrin dimers were modeled in a tail to tail configuration based on our electron microscopy observations (Refs. 28 and 31, and this work).

Biological Activity of Integrin Ligands
The biological activities of the ␣ IIb ␤ 3 ligands used in this study are summarized in Table I. cHArGD and cRGD have been described as high affinity analogs of the KQAGDV and RGD integrin-recognition sites of fibrinogen, respectively (73); eptifibatide is an FDA approved drug, an integrin antagonist used for the treatment of cardiovascular disease (57). Whereas the integrin-targeting sequence of each ligand is contained within chemically similar, seven-membered, cyclic rings (57,73), they exhibited different activity profiles. For example, cHArGD and eptifibatide, which share a common homoarginine residue, bound to platelet ␣ IIb ␤ 3 receptors with comparable EC 50 values, whereas cRGD bound an order of magnitude more weakly. This pattern extended to the ability of each ligand to block platelet aggregation, a fibrinogen-dependent function, as well as to their effects on platelet:fibrin adhesion. Based on its affinity and fibrin selectivity, cHArGD was used for most subsequent ␣ IIb ␤ 3 ligand binding studies; a minimum 5-fold molar excess of ligand was used to achieve Ͼ95% receptor saturation. a Ligand concentration yielding half-maximal receptor binding in a flow cytometric assay (25).
b Ligand concentration yielding half-maximal inhibition of platelet aggregation in an optical assay (28).
c Ligand concentration yielding half-maximal inhibition of platelet adhesion to fibrin in a colorimetric assay (25). d From Hantgan et al. (31).

Ligand Binding Effects on Global Stability of the ␣ IIb ␤ 3
Tryptophan fluorescence emission spectra were recorded as a function of temperature from 20 to 60°C for both free ␣ IIb ␤ 3 and the ␣ IIb ␤ 3 ⅐cHArGD complex to discern the effects of ligand binding on the thermal stability of the receptor (31,50). During heating, max shifted from 342 Ϯ 2 nm at 20°C to 337 Ϯ 2 nm at 60°C for the ligand-free integrin; in the presence of cHArGD, a shift from 340 Ϯ 3 to 338 Ϯ 2 nm was observed (n ϭ 3). These data were expressed as the ratio of the backgroundsubtracted emission at 350 nm to that at 320 nm, F 350 /F 320, thus correcting for the inherent temperature-dependent decrease in protein fluorescence emission (60).
As illustrated in Fig. 1 (open circles), in the absence of ligand, the F 350 /F 320 ratio for the ␣ IIb ␤ 3 complex exhibited a sigmoidal temperature dependence, decreasing by 25% from 20 to 60°C with a midpoint, T m ϭ 44°C. Data obtained in three such experiments yielded a 22 Ϯ 1% decrease and T m ϭ 41 Ϯ 3°C. In contrast, an approximately linear 8 Ϯ 3% decrease in the F 350 /F 320 ratio resulted for the ligand-bound receptor (solid circles). These data indicate that ligand binding alters the response of the integrin to increased temperature. However, the temperature-induced changes in emission ratio determined with both free and bound receptors were not readily reversible as indicated by data obtained during cooling cycles (light gray triangles, ␣ IIb ␤ 3 ; dark gray triangles, ␣ IIb ␤ 3 :cHArGD).
It is important to emphasize that our data ("Experimental Procedures") and that of Aoudia and Zana (40) demonstrate that the physical properties of octyl glucoside micelles are nearly invariant from 20 to 60°C, so that the effects we observed are not artifacts related to changes in micelle structure, bur rather reflect temperature-and ligand-induced changes in the physical state of the ␣ IIb ␤ 3 integrin. However, because blue-shifted fluorescence emission spectra are not expected for thermal denaturation (50), we investigated the basis for this thermal transition by additional biophysical techniques.

Ligand Binding Effects on the Secondary
Structure of ␣ IIb ␤ 3 Circular dichroic spectra were obtained for ␣ IIb ␤ 3 to determine the extent of thermal unfolding over the interval 20 -60°C. At 20°C both free receptor and the ␣ IIb ␤ 3 ⅐cHArGD complex exhibited similar minimum molar ellipticities of approximately Ϫ6000 to Ϫ7000 degree/mol at ϳ208 -211 nm, followed by a sharply increasing positive signal at lower wavelengths. However, because of the high absorbance of the octyl glucoside required for receptor solubilization, reliable data were not obtained below 205 nm, thus precluding deconvolution of the spectra to obtain estimates of secondary structure (75). Therefore, an empirical approach was followed, based on the observation that the difference signal between native ␣ IIb ␤ 3 and that unfolded in guanidinium chloride was greatest at 222 nm (29). As shown in Fig. 2 (open circles), the [⌰] 222 for ␣ IIb ␤ 3 exhibited a sigmoidal decline with increasing temperature. The 47°C midpoint of this transition was similar to that observed by fluorescence. Results from three experiments yielded T m ϭ 45 Ϯ 2°C and an 18 Ϯ 1% signal change. A linear decrease in [⌰] 222 of 9 Ϯ 1% was observed for the ␣ IIb ␤ 3 ⅐cHArGD complex, similar to the results of the fluorescence experiments in Fig. 1. These thermally induced molar ellipticity changes were not fully reversible, as the data in Fig. 2 also indicate (gray triangle, ␣ IIb ␤ 3 ; dark gray triangle, ␣ IIb ␤ 3 ⅐cHArGD). As was the case for the fluorescence data, the signal changes obtained by circular dichroism are opposite in direction to those expected for unfolding (29). Thus these data indicate that ␣ IIb ␤ 3 undergoes temperature-driven conformational changes that do not reflect global unfolding. Furthermore, ligand binding changes the conformational response of ␣ IIb ␤ 3 to increased temperature.

Effects of Ligand Binding and Temperature on Integrin Oligomerization
Dynamic Light Scattering-Recognizing the possibility that changes in integrin oligomerization could have contributed to the thermal transitions observed by fluorescence and circular dichroic spectroscopy, a series of dynamic light scattering experiments (28) was initiated to characterize the quaternary structure of the integrin at two temperatures, 20 and 40°C. As illustrated in Fig. 3, the free ␣ IIb ␤ 3 integrin (white bars) exhibited a stable, narrow size distribution centered at ϳ14 nm diameter. Following a 4-h incubation at 40°C, the peak shifted to ϳ32 nm (black bars). A small quantity of material, Ͻ1% by weight, was also present at diameters Ͼ500 nm (data not shown). Data obtained with buffered octyl glucoside yielded a peak diameter of 5 nm at 20°C and 5-6 nm at 40°C (data not shown), again indicating that changes in OG micelle size were unlikely to make a significant contribution to the these observations. We next turned our attention to the effects of ligand binding on the quaternary structure of ␣ IIb ␤ 3 . One hour after addition of excess ligand-mimetic peptide cHArGD to a sample of ␣ IIb ␤ 3 integrin at 20°C, the peak of the size distribution shifted to ϳ30 nm (light gray bars). An additional increase in peak diameter to ϳ47 nm was observed following a 4-h incubation of the ␣ IIb ␤ 3 ⅐cHArGD complex at 40°C (dark gray bars); ϳ1% of the material formed oligomers Ͼ500 nm in diameter (data not shown). These results are representative of the temperatureand ligand-induced changes observed by dynamic light scattering in two experiments with ␣ IIb ␤ 3 alone and three with ␣ IIb ␤ 3 ⅐cHArGD complexes.
The light scattering data indicate that integrin oligomerization increases at 40°C, and that ligand binding further enhances self-association above that induced by increased temperature alone. Therefore, integrin oligomerization probably contributed to the temperature-induced, ligand-sensitive conformational changes in ␣ IIb ␤ 3 we observed by fluorescence and circular dichroic spectroscopy. However, size distributions derived from dynamic light scattering data are inherently sensitive to small quantities of high molecular weight material because they return a z-average profile (76). Therefore, we investigated this issue with additional biophysical techniques.
Sedimentation Velocity-Complementary information about the effects of ligand binding and temperature on ␣ IIb ␤ 3 selfassociation was obtained by sedimentation velocity measurements that return a weight average size distribution (77). Time derivative analyses (62) showed that at 20°C the free integrin sedimented as a stable single species with s 20,w ϭ 8.5 S, as indicated by the open triangles in Fig. 4. Evidence of substantial oligomerization was obtained after a 3-h incubation at 40°C (black triangles). A weight average s 20,w ϭ 12.2 S was obtained, and the distribution of sedimenting species extended out to 30 S. These data are consistent with a mixture of protomers, dimers, trimers, and higher order species (28). The ␣ IIb ␤ 3 ⅐cHArGD complex sedimented as a single 8.0 S species at 20°C (light gray circles); this shift toward a slower sedimenting species upon ligand binding has been shown by us to reflect a ligand-induced conformational change to a more open integrin conformation (28,31). Following a 3-h incubation at 40°C, oligomers were also observed for the ␣ IIb ␤ 3 ⅐cHArGD complex at 40°C (dark gray circles), although the distribution (weight average s 20,w ϭ 9.7 S) was shifted somewhat toward smaller oligomers compared with the ligand-free integrin. Consideration of data obtained in replicate experiments yielded a temperatureinduced 43% increase in the weight average s 20,w for the free integrin but only a 22% increase for the ligand-bound form.
Whereas both sedimentation velocity and dynamic light scattering demonstrated that increased temperature and ligand binding promoted ␣ IIb ␤ 3 oligomerization, the size distributions obtained by these techniques differed in their quantitative details, as can be seen by comparing Figs. 3 and 4. For example, DLS detected increased oligomerization for the ␣ IIb ␤ 3 ⅐cHArGD complex at both 20 and 40°C compared with the free receptor. Conversely, no ␣ IIb ␤ 3 ⅐cHArGD oligomers were detected by sedimentation velocity at 20°C, and somewhat larger oligomers were observed for the free receptor compared with the bound at 40°C.
Because a substantial pressure gradient develops in a rapidly rotating ultracentrifuge cell (79), we considered the possibility that pressure-induced dissociation (79,80) could have influenced the distribution of integrin oligomers observed by sedimentation velocity. Following the treatment presented by Kegeles et al. (79), we estimated that the pressure reached 100 atmospheres near the cell bottom during a sedimentation velocity experiment at 35,000 rpm. Therefore, we addressed this issue by examining the distribution of sedimenting species as a function of sedimentation time, based on the reasoning that pressure-induced dissociation would become more pronounced later in the run. However, examination of g(S) profiles for the ␣ IIb ␤ 3 ⅐cHArGD complex at 20°C such as that presented in Fig.  4, revealed a set of single, symmetric profiles that exhibited the expected peak sharpening with time (62). We did not observe any significant accumulation of sedimenting material in the 5-6 S range expected for the dissociated ␣ IIb and ␤ 3 subunits. Whereas the patterns obtained at 40°C were more complex because of the increased presence of oligomeric species, evidence for pressure-induced dissociation was not observed there either.
These findings are in keeping with the observation that the sedimentation coefficients calculated with the Svedberg equation (60) using weight average molecular weights obtained from our classical light scattering data and z-average translational diffusion coefficients obtained from our dynamic light scattering data are in approximate, although not quantitative, agreement with those obtained directly by sedimentation velocity (data not shown). Given the differential sensitivity of these two hydrodynamic techniques to macromolecular size distributions (81,82), we investigated this issue in more detail by a thermodynamic approach, sedimentation equilibrium (64).
Sedimentation Equilibrium-Changes in the molecular weight distribution of ␣ IIb ␤ 3 were determined by sedimentation equilibrium as a function of temperature and ligation state. Data presented in Fig. 5 depict the weight average molecular weight, M w , versus temperature for free ␣ IIb ␤ 3 (open triangles), as well as its complexes with three integrin ligands: cHArGD (solid circles), cRGD (gray squares), and eptifibatide (dark gray diamonds). In the absence of ligands, ␣ IIb ␤ 3 exhibited M w values ranging from 210,000 to 247,000 over the range 20 -35°C, consistent with the 232,000 expected for the integrin protomer (dashed line). However, substantial molecular weight increases, to ϳ400,000 and 370,000, were observed at 37 and 40°C, respectively, indicative of temperature-induced aggregation for the ligand-free integrin (64). In contrast, M w data obtained for the ligated integrins showed a more gradual 1.6fold increase over the range 20 to 40°C (solid line), indicating the data obtained for the ligated integrin were more consistent with oligomerization, rather than the aggregation behavior observed for the free receptor.

Evidence for Entropy-driven Oligomerization
Van't Hoff Analyses-Sedimentation equilibrium data obtained with the integrin-ligand complexes were analyzed further by nonlinear regression (63) to extract a set of dimerization constants for each complex as a function of temperature. The resultant parameters are plotted in Fig. 6 as ln K a versus 1/T. Following the Van't Hoff equation, d ln K a /d͑1/T͒ ϭ Ϫ⌬H/R (Eq. 1) the dimerization enthalpy was obtained from the linear regression slope of each plot, then the entropy change was obtained from the Gibbs relationship. The solid line was obtained with data for the ␣ IIb ␤ 3 ⅐cHArGD complex (solid circles); Van't Hoff analysis yielded ⌬H ϭ 42 Ϯ 9 kcal/mol and ⌬S ϭ 164 Ϯ 40 cal-degree/mol for dimerization of this receptor:ligand pair. Data obtained with the ␣ IIb ␤ 3 ⅐cRGD complex (gray squares) agree reasonably well with this fitted line, whereas data for the ␣ IIb ␤ 3 ⅐eptifibatide complex (dark gray diamonds) follow a somewhat shallower slope. As shown in Table II, a comparable set of positive enthalpy and entropy changes resulted for each integrin-ligand complex.
Electron Microscopy Studies of the Effects of Temperature and Ligand Binding on Oligomerization of ␣ IIb ␤ 3 -The preceding results are supported by examination of ␣ IIb ␤ 3 by electron microscopy in the presence and absence of cHArGD and at 20 or 37°C. Following a 3-h incubation in HSC-OG buffer, complexes of ␣ IIb ␤ 3 were diluted into 0.05 M ammonium formate at pH 7.4, 30 mM octyl glucoside, 15% glycerol to a final concentration of 20 -25 g/ml. The samples were examined by transmission electron microscopy following rotary shadowing with tungsten. Glycerol is necessary to prevent surface artifacts, but the concentration was lowered relative to that used in previous experiments, because we found that the size distribution of ␣ IIb ␤ 3ligand complexes shifts toward larger oligomers in the presence of 30% glycerol (28,31).
Images obtained from samples prepared at 20°C in 15% glycerol showed almost entirely monomeric forms of ␣ IIb ␤ 3 , which are illustrated in previous papers (28,31). The monomers consist of a globular head region with two long tails extending from one side. The tails often appear to be connected to each other at their distal ends. In samples prepared at 37°C, there was a striking change in the complexes, in that many of them were clustered into oligomers. The oligomers have a specific structure, such that the complexes appear to be interacting with each other via their tails. Dimers are present as two complexes interacting via their tails, with their heads at opposite ends, such that the two complexes are related by a 180°r otation about an axis perpendicular to the page at the center of the complex (Fig. 7A). Larger oligomers are seen as clusters with their tails in the center and their heads at the periphery, a structure with the appearance of a "rosette" (Fig. 7A). Oligomers with similar structures were observed with ␣ IIb ␤ 3 samples incubated in the presence of excess cHArGD peptide at 37°C (Fig. 7B).
The effects of temperature and ligand binding were quantified by counting the number of integrin molecules present as ␣ IIb ␤ 3 monomers, dimers, and higher-order oligomers. As shown in Fig. 8, 87% of the ligand-free integrins were mono-FIG. 5. Effects of temperature and ligand binding on the distribution of integrin weight average molecular weights determined by sedimentation equilibrium. Data obtained in sedimentation equilibrium runs at 8000 rpm (and in selected cases also at 6000 rpm) were analyzed with Win NONLIN software to obtain M w for each condition of temperature and ligation state. Open triangles, ␣ IIb ␤ 3 ; solid circles, ␣ IIb ␤ 3 ⅐cHArGD; dark gray diamonds, ␣ IIb ␤ 3 -eptifibatide; light gray squares, ␣ IIb ␤ 3 ⅐cRGD. The dashed line corresponds to M w ϭ 232,000, the calculated protomer molecular weight. The solid line was obtained by linear regression with the ␣ IIb ␤ 3 ⅐cHArGD data.
FIG. 6. Van't Hoff plots for temperature-enhanced oligomerization of ligand-bound ␣ IIb ␤ 3 . Data obtained by sedimentation equilibrium, as described in the text and legend to Fig. 5, were also analyzed by Win NONLIN to extract dimerization constants for ␣ IIb ␤ 3 complexed with cHArGD (dark circles), cRGD (light gray squares), and eptifibatide (dark gray diamonds) at temperatures ranging from 20 to 40°C. Linear regression analyses of the resultant ln K a versus 1/T data yielded the solid line for the ␣ IIb ␤ 3 ⅐cHArGD complex. The thermodynamic parameters cited in Table II were derived from this analysis, applied to each integrin:ligand pair. meric at 20°C compared with only 56% at 37°C (compare white and black bars). In the presence of the cHArGD peptide, 68% of the integrins were present as monomers at 20°C (Fig. 8, light gray bars). In contrast, with cHArGD peptide at 37°C, only 41% were monomers (Fig. 8, dark gray bars). Whereas these trends toward increased oligomerization associated with both increased temperature and ligand binding are striking, the results should be interpreted with caution because of the difficulties of counting individual integrins within these complex oligomers. Likewise, direct quantitative comparison of these size distributions to those obtained by light scattering and analytical ultracentrifugation are precluded by the necessity to include glycerol in the sample preparations for electron microscopy, a procedure we have shown to promote oligomerization (28,31). Despite these reservations, the electron microscopy results clearly reinforce the conclusion that both ligand binding and increased temperature promote ␣ IIb ␤ 3 oligomerization.

DISCUSSION
This investigation provides new insights into two questions related to the dynamic regulation of integrin function. How do elevated temperatures, especially those in the physiological range, affect the secondary, tertiary, and quaternary structures of ␣ IIb ␤ 3 ? What intermolecular forces drive integrin self-association?
Regarding the response of ␣ IIb ␤ 3 to increased temperature, our spectroscopic probes of protein secondary and tertiary structure have detected a transition at 43 Ϯ 3°C for the ligandfree receptor, although not to an unfolded state. In fact, this new thermostable conformer displays increased helicity, as monitored by a deeper 222-nm trough in the circular dichroic spectrum of ␣ IIb ␤ 3 , and a blue-shifted intrinsic fluorescence emission spectrum. The direction of both changes is opposite to that we and others (29,31,50) have described for guanidinium denaturation of the ␣ IIb ␤ 3 integrin. Our observations are also consistent with those of Makogoneko et al. (50) who reported that ␣ IIb ␤ 3 was stable in octyl glucoside up to at least 95°C, based on fluorescence emission and scanning calorimetry measurements. Combining this information with the temperature-induced changes in receptor quaternary structure observed here by dynamic light scattering, sedimentation velocity, sedimentation equilibrium, and electron microscopy, we propose that the ligand-free ␣ IIb ␤ 3 integrin undergoes conformational changes and forms aggregates at temperatures above 35°C.
Our integrated spectroscopic, hydrodynamic and thermodynamic approach also demonstrates that receptor occupancy by ligand mimetic peptides promotes oligomer formation at lower temperatures, and protects the bound receptor from the cooperative thermal transition seen with the free integrin. Complementary size distribution data obtained by electron microscopy reinforces the concept that ligand binding and increased temperature interact to promote integrin oligomer formation.
Concerning the intermolecular forces that drive integrin oligomerization, the thermodynamic parameters determined in this investigation support the concept that entropically favorable hydrophobic interactions play a major role in ␣ IIb ␤ 3 selfassociation. In particular, van't Hoff analyses of the temperaturedependent dimerization constants, measured with ␣ IIb ␤ 3 complexed with each of three ligand mimetics, yielded a pattern of large positive enthalpy changes offset by even larger positive entropy changes. As shown in Table II, we obtained comparable ⌬H and ⌬S parameters with two structurally similar cyclized ligands, cRGD and cHArGD, developed by Cierniewski et al. (73) as high affinity analogs of the integrintargeting regions found on the adhesive proteins, fibronectin and fibrinogen. A similar pattern of positive ⌬H and ⌬S dimerization values also resulted when the pharmaceutical integrin antagonist, eptifibatide, developed by Scarborough et al. (30) as a fibrinogen mimetic, was bound to ␣ IIb ␤ 3 . This pattern of entropy/enthalpy compensation has been observed for other self-assembling, entropy-driven systems (37,48,78).
We note that the uncertainty associated with the thermodynamic parameters cited in Table II may be because of several factors. First, integrin self-association driven by increased temperature and ligand binding has an irreversible component that becomes more pronounced at temperatures above 35°C, as considered under "Experimental Procedures." Second, whereas we have treated the process as a monomer-dimer equilibrium, oligomers larger than dimers are seen by electron microscopy. We have previously found that the sampling conditions required for microscopy, especially the presence of glycerol, enhances ␣ IIb ␤ 3 aggregation (28,31). Third, the effects of octyl glucoside micelles must be considered in interpreting our results. Our work and that of others (40) have demonstrated that the physical properties of octyl glucoside micelles are essentially constant across the temperature range explored, so it is unlikely that detergent effects dominate our thermodynamic data. This point is reinforced by the sensitivity of each of our conformational reporters to the ligation state of ␣ IIb ␤ 3 , again FIG. 7. A gallery of electron microscope images of rotary shadowed ␣ IIb ␤ 3 complexes at 37°C in the presence and absence of cHArGD. A, ␣ IIb ␤ 3 with no peptide. B, complexes of ␣ IIb ␤ 3 with cHArGD. In each row, the first two images on the left are dimers in which two complexes are joined tail to tail with the globular heads at opposite ends; sometimes there is a small nodule at the tail to tail junction (probably representing the detergent micelle). The other two images in each row are examples of larger oligomers in which the tails are interacting with each other in the center, and the heads are at the periphery, forming a rosette-shaped structure. Bar, 50 nm.
FIG. 8. Size distributions of integrin oligomers obtained by electron microscopy. 1000 integrin molecules were counted for each condition and the results expressed as the % present as ␣ IIb ␤ 3 monomers, dimers, trimers, and oligomers containing four or more integrins. Open bars, ␣ IIb ␤ 3 at 20°C; black bars, ␣ IIb ␤ 3 at 37°C; light gray, ␣ IIb ␤ 3 ⅐cHArGD at 20°C; dark gray bars, ␣ IIb ␤ 3 ⅐cHArGD at 37°C. an effect that cannot be readily explained by changes in detergent properties.
Returning to a consideration of the structural basis for integrin clustering, homology modeling (68), based on the recently published ectodomain crystal structure of the closely related ␣ v ␤ 3 integrin (21), indicates that 19 of the 24 tryptophan residues in the ␣ IIb ␤ 3 complex are present in the ectodomain of the receptor. Hence, conformational changes, such as ligationlinked perturbation of the ϳ1600 A 2 ␣/␤ subunit interface in the ␣ v ␤ 3 integrin (21, 22), could shield some of these ectodomain spectral reporters from solvent and contribute to the observed blue-shifted emission. Whereas activation of integrin ␣ 2 ␤ 1 involves ectodomain conformational changes that remove one helical turn and break a hydrophobic contact (74), neither of those effects would explain our results. Conversely, there are three more tryptophan residues in the transmembrane segment of ␣ IIb and one each on transmembrane and cytoplasmic regions of ␤ 3 . Thus we propose that thermally driven integrin oligomers are stabilized, in part, by formation of helix bundles that bury key tryptophan residues in transmembrane and cytoplasmic domains of ␣ IIb ␤ 3 . Electron microscopy reinforces this concept by showing how oligomers of ␣ IIb ␤ 3 are joined at their tails, an observation that may also explain their increased helical CD signal and blue-shifted fluorescence spectra.  Fig. 7. Transmembrane segments from each integrin subunit are shown inserted into an octyl glucoside micelle using a model developed by Bogusz et al. (39). Panel B depicts an integrin protomer and dimer with their transmembrane segments spanning a lipid bilayer (70). Individual integrin protomer models were developed by a threading approach using the (extended) ␣ v ␤ 3 ectodomain structure (21) as a template. Helical transmembrane segments and cytoplasmic domain models, with conformations based on NMR structural data and secondary structural prediction algorithms (6,71,72), were then fused to the ␣ IIb and ␤ 3 polypeptide chains.
Evidence supporting a helix-bundling, integrin-clustering mechanism also comes from recent biophysical characterizations of protein fragments corresponding to integrin transmembrane and/or cytoplasmic domains (6,26,72). The extent to which the integrin cytodomains alone form ordered structures remains controversial, as Vinogradova et al. (6) first reported CD and NMR evidence identifying a helical segment in the membrane-proximal region of the ␣ IIb subunit, when tethered to a lipid environment. Weljie et al. (9) recently reported NMR data indicative of helical dimer interactions for the membraneproximal regions of the ␣ IIb and ␤ 3 cytoplasmic domains. In contrast, Ulmer et al. (72) found both cytoplasmic regions to be flexible and unstructured in aqueous solution, although their NMR data did indicate a helical structure could form within the tail of ␤ 3 . Li et al. (26) detected homomeric dimers and trimers with constructs encompassing the ␣ IIb and ␤ 3 membrane-spanning and cytoplasmic domains and proposed a role for these interactions in integrin clustering.
Taken together, these observations support the postulate that integrin self-association via its transmembrane and cytoplasmic domains could be stabilized by new contacts between regions poised to interact in the resting integrin protomers. This concept is illustrated schematically in Fig. 9A with an image of an integrin dimer assembled from a threading model of the ␣ IIb ␤ 3 ectodomain, based on the crystal structure of the ␣ v ␤ 3 construct recently described by Xiong et al. (21). We have attached helical transmembrane segments to the resultant ectodomain model; the conformations of the cytodomain segments are based on NMR data and secondary structure predictions (6,72). We have chosen the extended version of the integrin model proposed by Xiong et al. (21), rather than the original bent structure that they observed for ␣ v ␤ 3 in both the presence and absence of ligand (21,22). Our previous hydrodynamic and electron microscopy data strongly indicate that the extended conformation is favored under our experimental conditions (28,31).
Whereas this schematic is designed to resemble the dimers seen by electron microscopy, we propose that additional ectodomain interactions may be favored with integrins that reside on a cell surface. Tail to tail oligomers may form after an initial side by side integrin self-association, followed by micelle fusion and rearrangement to a more stable conformation induced by the high curvature of the octyl glucoside micelle (31). We hypothesize that in vivo both ligand binding and increased temperature favor the assembly of integrin dimers that are stabilized by helix bundles formed from their transmembrane (and possibly cytoplasmic) domains as well as through additional contacts between their ectodomains (illustrated in Fig. 9B). These integrin clusters could enhance the local concentration of integrin-associated cytoplasmic proteins, thus insuring the rapid, efficient transmission of a signal reporting ligand binding at their distant ectodomains to the interior of the cell.