Annular Anionic Lipids Stabilize the Integrin αIIbβ3 Transmembrane Complex*

Background: Anionic lipids compete for electrostatic interaction in membrane proteins. Results: Despite competition, anionic lipids stabilize the integrin αIIbβ3 transmembrane complex. Conclusion: Stabilizing anionic lipid-protein interactions exist and supersede destabilizing effects. Significance: Anionic lipid-mediated stabilization of membrane proteins may be of a general nature. Cationic membrane-proximal amino acids determine the topology of membrane proteins by interacting with anionic lipids that are restricted to the intracellular membrane leaflet. This mechanism implies that anionic lipids interfere with electrostatic interactions of membrane proteins. The integrin αIIbβ3 transmembrane (TM) complex is stabilized by a membrane-proximal αIIb(Arg995)-β3(Asp723) interaction; here, we examine the influence of anionic lipids on this complex. Anionic lipids compete for αIIb(Arg995) contacts with β3(Asp723) but paradoxically do not diminish the contribution of αIIb(Arg995)-β3(Asp723) to TM complex stability. Overall, anionic lipids in annular positions stabilize the αIIbβ3 TM complex by up to 0.50 ± 0.02 kcal/mol relative to zwitterionic lipids in a headgroup structure-dependent manner. Comparatively, integrin receptor activation requires TM complex destabilization of 1.5 ± 0.2 kcal/mol, revealing a sizeable influence of lipid composition on TM complex stability. We implicate changes in lipid headgroup accessibility to small molecules (physical membrane characteristics) and specific but dynamic protein-lipid contacts in this TM helix-helix stabilization. Thus, anionic lipids in ubiquitous annular positions can benefit the stability of membrane proteins while leaving membrane-proximal electrostatic interactions intact.

Membrane proteins adopt pivotal physiological roles through mediating the exchange of matter, energy, and information between cells. Indispensable to the structural integrity of a membrane protein, membrane lipids engage membrane proteins in interactions that range from dynamic solvation to highly specific association, mediated by annular lipids and nonannular lipids, respectively (1)(2)(3). Protein-lipid interactions also play a critical role in defining the membrane topology of proteins. Membrane proteins are arranged such that positively charged residues localize to the intracellular membrane face (4). The structural basis of this phenomenon is the interaction of anionic lipids, which are concentrated to the intracellular membrane leaflet, with cationic protein residues (5,6). This fundamental mechanism implies that anionic lipids also compete with electrostatic interactions between transmembrane (TM) 3 helices within the intracellular lipid headgroup region. Anionic lipid competition with electrostatic interactions between TM helices would have profound consequences for the structure and stability of membrane proteins. To explore this scenario, we examine here the impact of anionic lipids on the thermodynamic stability of the integrin ␣IIb␤3 TM complex and the interaction of ␣IIb␤3 TM segments with anionic versus zwitterionic lipids.
K XY Measurements-The ␣IIb ϩ ␤3^␣IIb␤3 equilibrium constant on the mole fraction scale (K XY ) was quantified by NMR and ITC. For NMR measurements, fresh samples were prepared at the desired peptide concentrations for each titration point by mixing peptides in 67% CH 3 CN, 33% H 2 O. Subsequent to freeze-drying, peptides were taken up in 320 l of 25 mM HEPES⅐NaOH, pH 7.4, 6% D 2 O, 0.02% (w/v) NaN 3 , 400 mM DHPC, and 120 mM of long chain lipids (see Table 1). H N -N correlation spectra were acquired on a Bruker Avance 700 spectrometer equipped with a cryoprobe at 28°C, using the TROSY-HSQC pulse scheme (13) with acquisition times of 85.4 ms for 1 H N and 71.7 ms for 15 N. The volume of a monomeric, isotope-labeled ␤3 resonance at total peptide concentrations [␤3] and [␣IIb] was extracted (V M ) and expressed as 1 Ϫ V M /V M,0 , where V M,0 is the resonance volume in the absence of ␣IIb. For isotope labeled ␣IIb and unlabeled ␤3, analyte and titrant interchange. K XY was obtained by nonlinear curve fitting of 1 Ϫ V M /V M,0 as a function of unlabeled peptide concentration (see Fig. 2, A and B) as described in detail previously (14). ⌬G o is obtained as ϪRT ln K XY , where R denotes the gas constant, and T denotes the absolute temperature.
ITC measurements were carried on a Microcal VP-ITC calorimeter. 10 M of ␤3 peptide in the 1.425-ml sample cell was titrated with ␣IIb peptide by injecting 9-l aliquots over a period of 10 s each. Unless otherwise specified, measurements were carried out at 28°C in 25 mM NaH 2 PO 4 /Na 2 HPO 4 , pH 7.4, 43 mM DHPC, and 17 mM of long chain lipid (see Table 1). Prior to data analysis, the measurements were corrected for the heat of dilutions of the ␣IIb and ␤3 peptides. The ␣IIb␤3 complex stoichiometry was fixed at 1:1 (14), and the reaction enthalpy (⌬H o ) and K XY were calculated from the measured heat changes, ␦H i , as described previously (14). The entropy change, ⌬S°, is obtained as (⌬H°Ϫ ⌬G°)/T. For the relatively weak ␣IIb(R995A)␤3 complex, K XY and ⌬H o values were measured by competitive binding experiments (15,16); specifically, the binding of ␣IIb to ␤3 was evaluated in the presence of competing ␣IIb(R995A).
We explicitly note that ITC experiments are often performed with the factor c ϭ K XY ϫ [peptide] 0 /[lipid] maintained above 1 when titrating to a 2:1 molar ratio of ligand-to-protein, which yields a sigmoidal titration curve (17). However, when titrating to binding saturation and when the complex stoichiometry is fixed, there is less restriction on c in terms of the accuracy of K XY (18,19). To compare ⌬G o values between different bicelle compositions, ideal solvent behavior of the hydrophobic phase (20) is required. Such behavior manifests in the independence of ⌬G o from the employed peptide to lipid ratio (20) and requires relatively high peptide to lipid ratios for integrin ␣IIb␤3 in 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC) bicelles (14). In POPC and POPG bicelles, we have experimentally verified ideal solvent behavior for the studied peptide to lipid range. Specifically, in 42 5A) yielding a final peptide concentration of 0.2 mM. HSQC-TROSY spectra were acquired at 28°C and 700 MHz. The detection of arginine -NH 2 resonances was attempted by HSQC experiments (22). Solvent magnetization was maintained at ϩI z throughout experiments whenever possible (23). Applying those same sample conditions, protection from paramagnetic Mn 2ϩ EDDA 2Ϫ was evaluated at 35°C. The ratio of H-N cross-peak intensities in the presence and absence of 1 mM Mn 2ϩ EDDA 2Ϫ , I/I 0 , was calculated. To correct for small differences in sample conditions, I/I 0 was uniformly scaled to 1 for the residues in the membrane center.
Assay of Integrin ␣IIb␤3 Receptor Activity in Lipid Nanodiscs-The receptor was purified from outdated human platelets based on a protocol modified from Ye et al. (24,25). In brief, outdated platelets were centrifuged at 300 ϫ g to remove erythrocytes and leukocytes, followed by centrifugation at 1800 ϫ g to pellet platelets. The platelets were washed twice with Tris-buffered saline and membrane proteins extracted by incubating overnight in 20 mM Tris, pH 7.4, 150 mM NaCl, 1% Triton X-100, 5 mM PMSF, 0.5 mM CaCl 2 , 10 M leupeptin, 10 M protease inhibitor E64 (Sigma), and 2.76 M calpeptin buffer. Integrin ␣IIb␤3 was purified on a concanavalin A column and passed through a heparin column to remove thrombospondin 1. Subsequently, the receptor was purified by gel filtration chromatography, and the inactive fraction was isolated as the flowthrough of an immobilized KYGRGDS affinity matrix, as described by Steiner and co-workers (26). Purified integrins were stored in 20 mM Tris, pH 7.4, 150 mM NaCl, 0.1% Triton X-100, 1 mM MgCl 2 , and 1 mM CaCl 2 buffer at Ϫ80°C.
Integrin nanodiscs were assembled based on a protocol adapted from a previous report (27,28). POPC, POPS, and POPG were solubilized in chloroform, mixed thoroughly, and dried onto a glass tube under a stream of nitrogen gas. The homogeneous lipid mixture was then solubilized in 100 mM cholate, 10 mM Tris, pH 7.4, and 100 mM NaCl, resulting in a lipid concentration of 50 mM. 72 l of the lipid solution was then mixed with 200 l of 200 M membrane scaffold protein in H 2 O and 200 l of 5 M purified inactive integrin (see above). This resulted in a final lipids:membrane scaffold protein:protein ratio of 90:1:0.025 in a total volume of 472 l. The integrin nanodiscs were assembled by removing the detergents with SM-2 biobeads overnight at room temperature. Finally, the assembled integrin nanodiscs were purified by gel filtration using a Superdex 200 column in 20 mM Tris, pH 7.4, 150 mM NaCl, 0.5 mM CaCl 2 solution. Successful nanodisc assembly was verified by SDS-PAGE.
To assay the activation state of integrin ␣IIb␤3 obtained in nanodiscs as a function of lipid composition, the binding of activity state-dependent antibody PAC1 was quantified by ELISA as described previously (24). Briefly, ELISA plates were coated with 5 g/ml AP3 antibody overnight at 4°C and then blocked with BSA for 1 h at 37°C. After washing the plate, integrin nanodiscs were added and incubated for 2 h at room temperature. The PAC1 antibody was used to detect active integrin receptors. PAC1 binding in the presence of activating antibody anti-LIBS6 was used as control for full activation. As negative control, PAC1 binding in the presence of 20 M eptifibatide was evaluated. After 2 h of incubation with PAC1 antibody, the wells were washed again, and HRP-conjugated anti-mouse IgM was added for 1 h more of incubation. Subsequent to the final wash, luminescence of the added ECL reagent was read using a VICTOR2 plate reader. An activation index was calculated as (L Ϫ L 0 )/(L max Ϫ L 0 ), where L denotes the luminescence intensity, L 0 denotes the luminescence in the presence of 20 M eptifibatide, and L max denotes the luminescence in the presence of anti-LIBS6 antibody.
Molecular Dynamics Simulations-Using the programs VMD 1.8.7 (29) and SOLVATE 1.0, ␣IIb(Glu 961 -Pro 998 ) and ␤3(Pro 688 -Glu 726 ) were immersed in lipid bilayers and solvated with TIP3P model water molecules (30). Na ϩ and Cl Ϫ ions were added to a concentration of 100 mM each and to neutralize the system. Monomeric ␣IIb and ␤3 TM and heterodimeric ␣IIb␤3 TM complex structures (Protein Data Bank codes 2k1a, 2rmz, and 2k9j) constituted starting structures. The energy-minimized average ␣IIb␤3 TM complex structure without ␣IIb-(Arg 995 )-␤3(Asp 723 ) salt bridge was calculated as described previously (8) in the absence of an explicit ␣IIb(Arg 995 )-␤3(Asp 723 ) distance restraint (31). Initial membrane embedding oriented the monomeric ␣IIb helix axis (Ile 966 -Lys 989 ) along the bilayer normal and centered it to the bilayer center. The ␣IIb␤3 TM complex was aligned on this ␣IIb orientation, and the embedding of the monomeric ␤3 helix (Ile 693 -Ile 721 ) replicated its embedding in the ␣IIb␤3 TM complex. To assess the dependence of protein-lipid contacts on starting conditions, two additional starting orientations were studied for each monomeric ␣IIb and ␤3 TM segment. These starting orientations were related to the original orientations by the following transformations: ␣IIb rotated relative to the z axis (membrane normal) by 15°and translated relative to the y axis (membrane surface) by Ϫ2 Å. Alternatively, ␣IIb translated relative to the x axis (membrane surface) by 2 Å and relative to the z axis by Ϫ2 Å. ␤3 rotated relative to the z axis by 10°and translated relative to the x axis by 2 Å. Alternatively, ␤3 translated relative to the x axis by 2 Å and relative to the z axis by Ϫ2 Å.
Dimer/monomer simulations in POPC typically consisted of 115/98 lipids, 6661/5851 water molecules, 10/9 Na ϩ atoms, and 15/13 Cl Ϫ ions. Analogous simulations in POPS typically contained 115/98 lipids, 7089/6145 water molecules, 125/107 Na ϩ atoms, and 15/13 Cl Ϫ ions. All-atom molecular dynamics (MD) simulations were carried out using CHARMM22 and CHARMM27 force fields for proteins and lipids, respectively, in the context of the program NAMD 2.9 (32)(33)(34). A uniform integration time step of 1 fs and periodic boundary conditions were employed. Electrostatic interactions were calculated using the particle mesh Ewald algorithm with a grid size of Ͻ1 Å (35). Only water molecules were treated rigidly using the SETTLE algorithm (36). The melting of lipid tails with all other atoms fixed was followed by minimization and equilibration with protein constrained and equilibration with protein released, each for a period of 0.5 ns. During simulations of 30 or 100 ns duration, the area per lipid was kept constant, and the cell dimension was variable. Simulations were carried out at 301.2 K and at constant pressure of 1 atm. Constant temperature was maintained using Langevin dynamics (37), with a damping coefficient of 1.0 ps Ϫ1 . Constant pressure was enforced using a Nosé-Hoover Langevin piston with a period of 200 fs and a time constant of 50 fs.

RESULTS AND DISCUSSION
Anionic Lipids Stabilize the ␣IIb␤3 TM Complex Relative to Electroneutral Lipids-To explore the influence of anionic lipids on ␣IIb␤3 complex stability, we quantified the ␣IIb ϩ ␤3␣ IIb␤3 equilibrium constant on the mole fraction scale, termed K XY , as a function of lipid type. Phospholipid bicelles, which consist of a long chain lipid bilayer disc that is stabilized by a rim of short chain lipids (Fig. 1B), served as proven membrane mimics for integrin ␣IIb␤3 (12,38). The symmetrical lipid bilayer of bicelles places anionic lipids in both membrane leaflets. Within the headgroup region of extracellular lipids, ␣IIb␤3 residues are dynamically unstructured without defined intersubunit contacts (Fig. 1, A and D). Consequently, contributions to ␣IIb␤3 TM stability from specific interactions with extracellular lipid headgroups are unlikely. As long chain lipids, zwitterionic POPC, anionic POPG, and anionic POPS were tested in combination with the short chain lipid DHPC. The long chain lipids differ only in their headgroup structure and charge (Fig.  1C), which have no significant effect on the backbone conformation of either the ␣IIb or ␤3 TM segments (11,12). Moreover, the bicelles served as a thermodynamically ideal solvent for ␣IIb␤3 at the employed peptide to lipid ratio (see "Experimental Procedures"). Determination of K XY was facilitated by the documented absence of ␣IIb or ␤3 homodimerization (14,39). To provide independent measurements of K XY and the associated change in free energy, termed ⌬G o , complex formation was evaluated by both NMR spectroscopy and ITC. ITC additionally provides the enthalpy and entropy changes upon ␣IIb␤3 TM complex formation, denoted by ⌬H o and ⌬S o , respectively.
In NMR experiments, integrin ␣IIb␤3 monomers and heterodimer are observable independently at 28°C (8,39). This exchange behavior permits the direct extraction of the fraction of ␣IIb␤3 dimer by quantifying the decline of peak volumes of monomer residues. We labeled either ␣IIb or ␤3 with 2 H/ 15 N isotopes and quantified the peak volumes of labeled monomer as a function of concentration of unlabeled partner peptide. Nonlinear curve fitting of the obtained ␣IIb␤3 complex fractions yielded K XY (Fig. 2, A and B, and Table 1). The results were similar for isotope labeling of either ␣IIb or ␤3 and, perhaps surprisingly, revealed a POPS-mediated ␣IIb␤3 complex stabilization. For instance, at a POPC:POPS ratio of 2:1, which approximates the ratio of anionic lipids in the intracellular membrane leaflet (40), the ⌻⌴ complex is stabilized by Ն0.19 Ϯ 0.02 kcal/mol compared with POPC only. At POPC/POPS ϭ 2:1, POPS is at least in 50-fold excess of ␣IIb. However, TM complex stabilization still increased at higher POPS ratios ( Fig. 2C and Table 1), showing that ␣IIb␤3 was unable to attract anionic lipids with high affinity. This inability demonstrates that annular lipids were responsible for the observed stabilizing effect.
ITC measurements detected the heat changes upon titrating ␤3 with ␣IIb to binding saturation and yielded K XY and ⌬H o from nonlinear curve fitting (Fig. 2D). We note that nonsigmoidal titration curves were chosen to ensure ideal solvent behavior of the hydrophobic phase (see "Experimental Procedures"). To control for the size of the long chain lipid bilayer area (Fig.  1B) and for any lipid headgroup-dependent difference in short and long chain lipid mixing (41), larger bicelles were employed in ITC than in NMR experiments (q eff ϭ 0.5 versus 0.31), which merely offset ⌬G o to lower values (Table 1) because of a more favorable binding entropy (14). For pure POPS, complex stabilization relative to POPC was 0.50 Ϯ 0.02 kcal/mol, and the corresponding value for pure POPG was 0.36 Ϯ 0.02 kcal/mol (Table 1). Relative to ⌬G o of Ϫ4.84 Ϯ 0.01 kcal/mol in POPC (Table 1), lipid composition modulates significantly TM com-plex stability. In sum, NMR and ITC measurements show that annular anionic lipids stabilize the integrin ␣IIb␤3 TM complex in a lipid headgroup structure-dependent manner.
Influence of Lipid Composition on Integrin Receptor Activity-Before we provide a structural view on ⌬G o POPS Ͻ ⌬G o POPC , we note that the platelet membrane that harbors integrin ␣IIb␤3 undergoes large scale changes in anionic lipid membrane distribution. The activation of the blood clotting enzyme thrombin requires the emergence of PS lipids in the outer membrane leaflet of platelets (42). Its failure results in the bleeding disorder known as Scott syndrome (43), in which the phospholipid scramblase TMEM16F is compromised (44). Anionic lipid scrambling will reduce the concentration of these lipids in the intracellular membrane leaflet, which may affect ␣IIb␤3 TM complex stability and thus the adhesive state of the receptor. The substrate binding abilities of two integrins have been examined in liposomes of different, symmetric lipid compositions (45,46). The affinity of integrin ␣V␤3 to bind vitronectin is larger in phosphatidylethanolamine/phosphatidylcholine (PE/PC)-based liposomes than liposomes containing solely PC (46). PE/PC-based lipids are both net neutral. A further increase in substrate binding is detected when incorporating PS and phosphatidylinositol-based lipids and cholesterol (46), which affect many different membrane characteristics. For integrin ␣IIb␤3, it is reported that the anionic lipid phosphatidic acid increases fibrinogen binding (45). Accordingly, membrane lipid characteristics may influence integrin activity.
To deepen our understanding of integrin ␣IIb␤3 activity as a function of membrane lipid composition, we examined the adhesive state of integrin ␣IIb␤3 in lipid nanodiscs that incorporated POPC, POPG, or POPS. For this purpose, integrin ␣IIb␤3 was purified from human platelets, reconstituted in the nanodiscs, and assayed using conformation-specific antibodies. If the TM structure in bicelles and nanodiscs match, the free energy difference between receptors in POPC and POPS nanodiscs is 0.50 Ϯ 0.02 kcal/mol. Except for a slightly higher receptor activity in POPG, no significant differences in integrin activity were detected between the different lipid compositions (Fig. 3). To assess the veracity of this result, we examined the effect of the alanine substitution of ␣IIb(Arg 995 ) on TM complex stability in POPC bicelles by ITC. Disturbing the ␣IIb(Arg 995 )-␤3(Asp 723 ) interaction in mammalian cell membranes leads to integrin activation (7), and a ⌬G o loss of 1.5 Ϯ 0.2 kcal/mol was detected for ␣IIb(R995A) ( Table 1 and Fig. 4). The loss of 0.50 Ϯ 0.02 kcal/mol upon replacing all POPS with POPC is indeed too small to trigger integrin activation. Notwithstanding, the threshold of integrin activation must shift with anionic lipid content, and the physiological scrambling of anionic lipid upon platelet activation has the capacity to facilitate integrin activation.
Anionic Lipids Establish Headgroup Structure-dependent Contacts with ␣IIb and ␤3-The stabilization of the ␣IIb␤3 TM complex by anionic lipids raises the question of whether anionic lipids were at all competing for the ␣IIb(Arg 995 )-  (Table  1). D, ITC measurements of ␣IIb␤3 TM association in phospholipid bicelles consisting of 34 mM DHPC and 17 mM of the depicted long chain lipid (q eff ϭ 0.5) using a starting ␤3 concentration of 10 M.  ␤3(Asp 723 ) interaction. Unlike the PC headgroup, PS and PG headgroups form intra-and interlipid hydrogen bonds and engage their Na ϩ counterion in lipid-bridging interactions (47,48). We therefore evaluated whether ␣IIb(Arg 995 ) established contacts with lipid headgroups. Electrostatic interactions, notably hydrogen bonding, prolong the lifetime of labile 1 H nuclei such as the ⑀-NH and -NH 2 nuclei of arginine (Fig. 1C). Lipid headgroup contacts to these nuclei may therefore increase their peak intensities in NMR spectra. 180°flips around the N ⑀ -C partial double bond are often sufficient to prohibit the NMR detection of -NH 2 nuclei (49), and we were indeed unable to observe these nuclei at pH 7.4 and 28°C. However, the ⑀-NH nuclei of the three arginines of ␣IIb (residues 962, 995, and 997; Fig. 1D) and the one arginine of ␤3 (residue 724) were detectable. Initially, we compared the cumulative ⑀-NH signal intensities of ␣IIb and ␤3 arginines as a function of bicelle lipid composition. Relative to POPC, ⑀-NH nuclei were stabilized by POPS and POPG (Fig. 5A). Moreover, POPS surpassed POPG in stabilizing ⑀-NH. The electrostatic character of this stabilization was confirmed by observing a reduction in signal intensity in the presence of 120 and 240 mM NaCl (Fig.  5A). Thus, the zwitterionic nature of POPC disfavored PO 4 Ϫ electrostatic interactions external to the lipid. In contrast, this functional group was available in POPG and POPS, which additionally offers a COO Ϫ moiety that further aided electrostatic ⑀-NH-lipid contacts. The electrostatic interaction of lipids with arginine side chains therefore confirmed the competitive nature of anionic lipids and paralleled the order of lipid-mediated ␣IIb␤3 TM complex stabilization.
Arg 995 of ␣IIb is Positioned for Electrostatic Contacts with Lipid Headgroups-The ⑀-NH resonances of the three arginines of ␣IIb overlapped. To observe individual resonances, Arg to Lys substitutions were carried out. Individual arginine resonance intensities were quite different and decreased in the order of 962 Ͼ 997 Ͼ 995 (Fig. 5B). The ⑀-NH resonance of ␣IIb(Arg 995 ) was barely detectable. Among the three arginines, Arg 995 is closest to the membrane core (Fig. 1D), and perhaps it  was able to evade headgroup contacts either through its positioning or through shielding by surrounding cationic residues (Fig. 1, A and D). To seek an atomistic understanding of ␣IIb(Arg 995 )-lipid contacts, all atom MD simulations were carried out for ␣IIb and ␤3, which were immersed in a bilayer of either POPC or POPS. To consider bias from the starting configuration, three 30-ns simulations with different initial ␣IIb/␤3 bilayer immersions were performed for each lipid type. For all arginines, lipid contacts were observed and, in accordance with solvent exchange (Fig. 5A), the incidence of electrostatic ⑀-NH contacts to lipid headgroups was on average higher in POPS than in POPC bilayers (Fig. 5C), an observation that is likely of a general nature (50). Remarkably, ␣IIb(Arg 995 /⑀-NH) produced the highest incidence of lipid headgroup contacts in both POPC and POPS bilayers relative to the other arginines (Fig. 5C). Simultaneously, contacts to water were scarcest for ␣IIb(Arg 995 /⑀-NH) (data not shown). As a result, it appears that Arg 995 is suitably positioned to interact with lipid headgroups. As a consequence, the difficulty of observing the ␣IIb(Arg 995 /⑀-NH) resonance by NMR is not indicative of its inability to establish lipid contacts. Its low peak intensity could arise from a high intrinsic hydrogen exchange rate in the particular chemical environment of Arg 995 , which would imply that anionic lipids only marginally stabilize its lifetime. Alternatively, if the ⑀-NH nucleus exchanges between different chemical environments on an unfavorable time scale relative to the NMR chemical shift time scale, its resonance would severely broaden. Interestingly, the addition of ␤3 peptide did only slightly increase the low signal intensity of ␣IIb(Arg 995 /⑀-NH) (Fig. 5D). In light of the strong effects of ␣IIb(Arg 995 ) mutations on TM complex stability (Table 1), the low ␣IIb(Arg 995 /⑀-NH) signal intensity suggests that the ␣IIb(Arg 995 )-␤3(Asp 723 ) interaction is dynamic. In conclu-  (Fig. 1D). B, arginine ⑀-H N NMR cross-peak intensities of the depicted ␣IIb mutants reveal the relative contribution of each ␣IIb arginine to the cumulative signal intensity observed in A. Spectra were recorded at 28°C and 700 MHz using TM peptides reconstituted at concentrations of 0. sion, it is unlikely that ␣IIb(Arg 995 ) is able to evade lipid headgroup contacts, which means anionic lipids will compete for ␣IIb(Arg 995 ) with ␤3(Asp 723 ).
The Stability of the ␣IIb(Arg 995 )-␤3(Asp 723 ) Interaction Is Similar in POPC and POPS-To further show electrostatic competition for ␣IIb(Arg 995 )-␤3(Asp 723 ) by the headgroup of POPS, we examined the effects of the compound O-phospho-L-serine (OPS), which corresponds to the POPS headgroup (Fig.  1C). For equimolar amounts of POPC and OPS, the TM complex was destabilized by ⌬⌬G o OPS ϭ ⌬G o POPC/OPS Ϫ ⌬G o POPC ϭ 0.13 Ϯ 0.01 kcal/mol ( Table 1). As a reference, for 500 mM NaCl the corresponding ⌬⌬G o NaCl was 0.23 Ϯ 0.01 kcal/mol (Table 1). Electrostatic competition is thus apparent. However, as noted earlier, the ␣IIb(R995A) substitution destabilized the TM complex by 1.5 Ϯ 0.2 kcal/mol in POPC bicelles ( Table 1). The relatively weak competition observed suggests that OPS and NaCl concentrations near ␣IIb(Arg 995 ) were reduced by solute concentration gradients in lipid headgroup regions. The headgroup of POPS should accordingly be more destabilizing to ␣IIb(Arg 995 )-␤3(Asp 723 ) than OPS. If so, ␣IIb(Arg 995 )-␤3(Asp 723 ) will contribute less to overall TM complex stability in POPS than in POPC, and ⌬⌬G o ␣IIb(R995A) of 1.5 Ϯ 0.2 kcal/mol in POPC will diminish in POPS. Paradoxically, ␣IIb(R995A) destabilized the TM complex by 2.0 Ϯ 0.3 kcal/mol in POPS (Table 1 and Fig. 4), demonstrating that the stability of the ␣IIb(Arg 995 )-␤3(Asp 723 ) interaction in POPS was not diminished relative to POPC. For reference, the stability of the mutant ␣IIb(R995A)␤3 TM complex is similar to the TM complex stability of the fibroblast growth factor receptor 3 at Ϫ2.8 Ϯ 0.1 kcal/mol (51). By what means ␣IIb(Arg 995 )-␤3(Asp 723 ) was able to compensate for the competition for ␣IIb(Arg 995 ) from POPS?
Anionic Lipids Induce a Ternary ␣IIb(Arg 995 )⅐POPS⅐␤3 Complex-To gain insight into lipid-protein interactions in the context of the ␣IIb␤3 TM complex, ␣IIb(Arg 995 )-lipid and ␣IIb(Arg 995 )-␤3(Asp 723 ) contacts were compared in MD simulations of 100-ns duration. The suitability of MD simulations for studying the ␣IIb␤3 TM complex is well established (52). The TM complex was examined in the presence of POPC and POPS bilayers starting with engaged and disengaged ␣IIb(Arg 995 )-␤3(Asp 723 ) contact, respectively. When starting with an engaged salt bridge, it dissociated intermittently in both lipid environments (Fig. 6A). However, ␣IIb(Arg 995 )-␤3(Asp 723 ) distance fluctuations in POPS were significantly larger in amplitude and duration than in POPC, which is indicative of the competitive nature of POPS lipids. When starting with the salt bridge dissociated, lipid interactions of ␣IIb-(Arg 995 ) were substantially more stable in POPS as compared with POPC (Fig. 6B). Interestingly, the lipid that is interacting with ␣IIb(Arg 995 ) formed dynamic but persistent contacts with the ␤3 TM helix ( Fig. 6C and supplemental Movie S1). Relative to the lifetime of the ␣IIb␤3 TM complex of 0.5 s (14), the MD simulations afford only a brief glimpse into protein-lipid interactions. However, our observations agree with dynamic ␣IIb(Arg 995 )-␤3(Asp 723 ) interaction (Fig. 5D) and reflect the different lifetimes of Arg(⑀-NH)-lipid headgroup contacts in POPC and POPS (Fig. 5A). To explain the similar stability of ␣IIb(Arg 995 )-␤3(Asp 723 ) in POPC and POPS, we present the following hypothesis.
Structural View of Anionic Lipid-mediated ␣IIb␤3 TM Complex Stabilization-The net stabilization of the ␣IIb␤3 TM complex by anionic lipids raises the question of its structural basis. Next to lipid contacts involving ␣IIb(Arg 995 ), we consider differences in physical membrane characteristics and specific TM helix-lipid contacts to be relevant to differences in overall ␣IIb␤3 TM complex stability. Electroneutral and anionic lipids give rise to differences in physical membrane characteristics (47,48) that may directly modulate TM helix-helix interactions (1,54). Here, we examined whether differences in POPC and POPG headgroup characteristics (Fig. 1C) lead to differences in the protection of protein backbone H N nuclei from the net neutral paramagnetic agent Mn 2ϩ EDDA 2Ϫ (12). For the ␣IIb and ␤3 TM segments, two to four additional residues were protected by POPC compared with POPG at both the intra-and extracellular membrane faces (Fig. 7, A and B). This observation indicates that small molecules can penetrate the headgroup region of PG more deeply than of PC by a range of 3-6 Å. Accordingly the concentration gradients of solvent and ions in the PC and PG/PS headgroup region will differ. We anticipate PS to resemble PG based on their similar headgroup volumes (Fig. 1C) and molecular interactions (47,48). Independent of differences in Mn 2ϩ EDDA 2Ϫ protection, we noted that the "charge content" of the lipid headgroup region stemming from lipid charges, charged protein residues, and small ions affected the magnitude of enthalpy changes (⌬H o ) that accompanied ␣IIb␤3 TM complex association (Table 1). Thus, substantial differences in the structural context of integrin ␣IIb␤3 TM segments in zwitterionic and anionic lipids entail the capacity of contributing to ␣IIb␤3 TM complex stabilization by annular anionic lipids. For instance, it is conceivable that the hydrophobic effect in the lipid headgroup region increased in POPG/ POPS relative to POPC.
TM complex stability increased in POPS compared with POPG lipids (Table 1) in correlation with an increase of the lifetime of lipid-protein contacts in POPS relative to POPG (Fig. 5A). This apparent correlation, which relates to the wider charge distribution of the PS than the PG headgroup (Fig. 1C), suggests that specific lipid-protein contacts may contribute to anionic lipid-mediated complex stabilization. In the structured region of the TM complex, four cationic residues are found (Fig.  1, A and D). Next to ␣IIb(Arg 995 ) and the first positively charged residues on the intracellular side, ␣IIb(Lys 989 ) and ␤3(Lys 716 ), ␣IIb(Lys 994 ) is noteworthy for its high conservation among human integrin ␣ subunits at 89% (8). The Ala substitution of Lys 994 altered the protection profile of the ␣IIb TM segment from paramagnetic Mn 2ϩ EDDA 2Ϫ (Fig. 7, C and D). In both POPG and POPC, increased H N exposures of especially Gly 991 and Ala 994 were detectable relative to the wild-type peptide. These residues are structurally adjacent, and it is evident that lipids rearranged around residue 994 in support of altered lipid-protein contacts. If an electrostatic nature of such contacts were important to TM complex-stabilizing lipid-protein contacts, ␣IIb(K994A) would increase ⌬G o in POPS and not POPC. In POPC, ␣IIb(K994A) did not change ⌬G o relative to wild type, but in POPS ⌬G o was increased by 0.45 Ϯ 0.03 kcal/ mol (Table 1). Thus, ␣IIb(Lys 994 )-lipid contacts appear to be important for anionic lipid-mediated TM complex stabilization. The stabilizing effect of ␣IIb(Lys 994 )-anionic lipid interactions may arise from a topographical stabilization of the ␣IIb helix in the membrane (14) and/or by providing an "optimal" structural environment at residue 994.
Conclusions-The evolution of membrane proteins in a bilayer, where anionic lipids are concentrated in the inner leaflet, has resulted in the usage of electrostatic contacts between anionic lipids and cationic protein residues to establish membrane protein topology (4 -6). Although the headgroup of anionic lipids did compete for electrostatic interactions within the integrin ␣IIb␤3 TM complex, the net stability of its key ␣IIb(Arg 995 )-␤3(Asp 723 ) electrostatic interaction was not compromised, and an overall TM complex stabilization was induced. The observed complex stabilization of up to 0.5 kcal/ mol is a noteworthy contribution to the threshold of integrin ␣IIb␤3 receptor activation, which is reached at 1.5 Ϯ 0.2 kcal/ mol in POPC. Our work provides thermodynamic insight into principles of membrane protein-lipid interactions that is of general interest to the study of membrane proteins. For instance, compared with the affinity of other receptor TM complexes (14), the complex stability of integrin ␣IIb␤3 at Ϫ4.84 Ϯ 0.01 kcal/mol is relatively high, indicating the potential of lipids to significantly modulate the strength of physiological TM helix-helix interactions (58). We emphasize that integrin ␣IIb␤3 does not possess a high affinity anionic lipid-binding site, which means that solely annular lipids were responsible for the observed effects. Membrane proteins often use nonannular lipids, which conceptually resemble co-factors of soluble proteins, to stabilize their conformation (1)(2)(3). Moreover, high affinity binding sites for annular anionic lipids have been observed in membrane proteins (55)(56)(57). However, by stabilizing the TM helix-helix association of integrin ␣IIb␤3 in ubiquitous low affinity annular positions, our results point to a general role of anionic lipids in enhancing membrane protein stability.