Joel S. Bennett, University of Pennsylvania School of Medicine, 914 BRB II/III, 421 Curie Blvd., Philadelphia, PA 19104, USA. Tel.: +1 215 573 3280; fax: +1 215 573 7039. E-mail: firstname.lastname@example.org
Summary. Integrins are a ubiquitous family of non-covalently associated α/β transmembrane heterodimers linking extracellular ligands to intracellular signaling pathways  [Cell, 2002; 110: 673]. Platelets contain five integrins, three β1 integrins that mediate platelet adhesion to the matrix proteins collagen, fibronectin and laminin, and the β3 integrins αvβ3 and αIIbβ3  [J Clin Invest, 2005; 115: 3363]. While there are only several hundred αvβ3 molecules per platelet, αvβ3 mediates platelet adhesion to osteopontin and vitronectin in vitro  [J Biol Chem, 1997; 272: 8137]; whether this occurs in vivo remains unknown. By contrast, the 80 000 αIIbβ3 molecules on agonist-stimulated platelets bind fibrinogen, von Willebrand factor, and fibronectin, mediating platelet aggregation when the bound proteins crosslink adjacent platelets  [J Clin Invest, 2005; 115: 3363]. Although platelet integrins are poised to shift from resting to active conformations, tight regulation of their activity is essential to prevent the formation of intravascular thrombi. This review focuses on the structure and function of the intensively studied β3 integrins, in particular αIIbβ3, but reference will be made to other integrins where relevant.
Electron microscopy (EM) of αIIbβ3 isolated from platelets revealed an 8 × 12 nm nodular headpiece containing its ligand binding site and two 18 nm flexible legs containing transmembrane (TM) and cytoplasmic domains extending from one side . Further, crystal structures for the extracellular portions of αvβ3 and αIIbβ3 have revealed a complex domain structure that rearranges as the integrins switch from inactive resting to active ligand-binding conformations (Fig. 1) [5–8].
The extracellular portions of αIIb and αv are similar, consisting of an amino-terminal β-propeller domain followed by a ‘thigh’ and two ‘calf’ domains, whereas β3 is substantially more complicated [5,7]. Its amino-terminal portion consists of two tandem nested domains: a βA domain whose fold resembles that of integrin α subunit I-domains, which is inserted into a ‘hybrid’ domain whose fold is similar to an I-set Ig domain; the hybrid domain in turn is inserted into a PSI (plexin, semaphorin, integrin) domain that contains the N-terminus of β3 [7,9]. The C-terminus of the PSI domain is continuous with four tandem EGF-like repeats that make up the ‘leg’ of β3, as well as a unique carboxyl-terminal βTD domain . αv and αIIb interact non-covalently with β3 via an interface between the α subunit β-propeller and the β3 βA domains that resembles the interface between the Gα and Gβ subunits of G proteins and forms the surface for ligand binding [5,7].
A surprising finding of the crystal structure of the extracellular portion of αvβ3 was a severe bend at ‘knees’ or ‘genus’ located between the first and second EGF-like repeats of β3 and the thigh and first calf domains of αv [5,10]. Recent high resolution crystal structures of the complete extracellular domain of αIIbβ3 also revealed that the inactive molecule is bent, with the 3rd and 4th EGF-like β3 domains inserted into a crevice formed by the upper β3 leg on one side, and the αIIb leg on the other side and with the αIIb β-propeller and β3 βA domains helping to form the crevice’s back . Negatively-stained EM images of purified ostensibly inactive αvβ3 and αIIbβ3 are consistent with these crystal structures, whereas EM images of active αvβ3 have revealed extended molecules, similar to previous rotary-shadowed EM images of α5β1 and αIIbβ3 [4,8,11,12]. This suggests that extension is an essential feature of integrin activation [13,14]. Nonetheless, despite unequivocal evidence that the extracellular portions of αvβ3, αIIbβ3, and other integrins can assume bent configurations, the significance of this observation for integrin regulation remains controversial. The ability to engineer disulfide bonds between N-terminal and C-terminal portions of αvβ3 and αIIbβ3 in transfected cells  and to show decreased fluorescence resonance energy transfer (FRET) efficiency between α4β1-bound and membrane-bound fluorophores following α4β1 activation [15,16] argues strongly that both bent and extended integrin conformations occur on cell surfaces. Conversely, reconstructed electron cryomicroscopy images of low affinity αIIbβ3 , cryoelectron tomography of isolated αIIbβ3 incorporated into liposomes , intramolecular FRET of αIIbβ3 before and after platelet stimulation , and hydrodynamic measurements  suggest that inactive αIIbβ3 has either an extended or only a slightly bent conformation. The basis for these discrepant observations remains unclear. In part, it may be because of the inherent flexibility of integrin legs, the loss of conformational constraints when integrins are extracted from membranes, and the difficulty in determining detailed molecular structures at the resolution of EM.
Although the preponderance of evidence suggests integrin activation is accompanied by a shift from bent to extended conformation, extended integrins are not necessarily capable of binding ligands . Ligand binding also requires that integrin headpieces shift from closed to open high affinity conformations, a process occurring when the α7 helix of the βA domain is displaced downward relative to the hybrid and PSI domains [11,21,22]. This allows the hybrid and rigidly connected PSI domains to ‘swing out’, thereby ‘opening up’ the ligand binding surface. It has been proposed that headpiece opening is the end result of a concerted ‘switchblade-like’ rearrangement of the entire heterodimer, beginning with separation of the cytoplasmic tails and TM domains, straightening and separation of the α and β subunit legs at their knees, and movement of the ligand-binding headpiece away from the membrane [8,11,23]. However, other results suggest that extension is not required for headpiece opening and that bent integrins are fully capable of binding ligands. In this case, transition from a closed to an open headpiece would be regulated by a ‘deadbolt’ composed of a loop from the β subunit βTD domain that binds to and locks the βA domain in its inactive conformation [24,25]. Allosteric re-positioning of this loop would then be sufficient to permit headpiece opening, enabling ligand binding to the bent conformation. The evidence supporting either mechanism of integrin activation is contradictory. Thus, while negatively stained EM images have been interpreted as showing fragments of fibronectin (FN) bound to bent αvβ3 extracellular domains , other EM images showed FN fragments bound to the top of α5β1 headpieces, as predicted if the integrin were extended . Further, mutating the αMβ2 βTD domain caused constitutive αMβ2 activation , but mutating the βTD domains of αIIbβ3 or αvβ3 did not . Nevertheless, because a deadbolt mechanism, in the absence of integrin extension, orients the ligand binding site on the headpiece toward the cell surface, rather than toward extracellular ligands, it is difficult to reconcile the geometry of a bent integrin and accessibility of its ligand binding site.
Transmembrane domain interactions regulate integrin function
Although rearrangement of the extracellular portion of integrins is required for ligand binding, the signals initiating this rearrangement originate in the cytosol and must be transmitted across the plasma membrane via TM helices. Truncated integrins lacking TM and cytoplasmic domains are constitutively active, implying that the deleted domains constrain integrin activity [29–31]. Moreover, ‘scissor-like’ movements of integrin legs appear to regulate integrin activity; integrins are inactive when their legs are in proximity and active when their legs separate [23,32–34].
TM helices are generally believed to serve merely as hydrophobic anchors. Thus, the observation that TM helices actively participate in the ‘scissor-like’ movements that regulate integrin function was unexpected . However, TM helix-mediated protein oligomerization is a common mechanism for assembly of membrane proteins and regulation of protein function and is mediated by the presence of specific oligomerization motifs . For example, glycine residues spaced four residues apart (GxxxG) in the glycophorin A TM helix (i.e., glycines separated by one turn of the TM helix) mediate the association of glycophorin A homodimers in red cell membranes by forming a tightly-packed surface of complementary ridges and grooves [37–39]. It is also noteworthy that GxxxG is the most over-represented sequence motif in databases of TM helices .
Polypeptides corresponding to the αIIb and β3 TM and cytoplasmic domains readily associate homomerically in phospholipid micelles . Subsequently, Li et al.  scanned the β3 TM helix with polar Asn residues and found that replacing G708 or M701 with Asn induced spontaneous fibrinogen binding to αIIbβ3 expressed in CHO cells. The Asn replacements also induced constitutive phosphorylation of focal adhesion kinase (FAK) and the formation of αIIbβ3 patches on the cell surface, suggesting they caused αIIbβ3 clustering as well. Similarly, Vararattanavech et al.  found that an L686F mutation in the β2 TM domain caused constitutive αLβ2 activation and the formation of αLβ2 microclusters on the surface of transfected K562 cells. There are two plausible mechanisms to explain the effect of the Asn replacements on αIIbβ3 function. First, by enhancing the propensity of the β3 helix to oligomerize homomerically , the replacements could cause separation of the αIIb and β3 legs. The ability of Asn residues to enhance the tendency of β3 TM peptides to interact homomerically in vitro supports this possibility [35,44]. Second, it is possible that introducing a bulky Asn residue into a closely packed αIIb/β3 TM heterodimer could physically cause separation of the αIIb and β3 legs. In support of this possibility, replacing αIIb residues G972 or G976, the first and last residues of its GxxxG motif, with Asn induces constitutive αIIbβ3 activation, but disrupts homomeric αIIb TM helix interactions [44,45]. Taken together, these observations suggest a ‘push–pull’ mechanism for regulating αIIbβ3 activity . Processes destabilizing the association of the αIIb and β3 TM helices would be expected to ‘push’αIIbβ3 to its activated conformation. Conversely, intermolecular interactions requiring separation of the αIIb and β3 TM helices, such as homo-oligomerization of the TM helices, or are more favorable when they are separated, such as ligand-induced αIIbβ3 clustering , would be expected to ‘pull’αIIbβ3 toward the activated state.
The ability of homomeric TM helix interactions to induce αIIbβ3 activation in cells remains to be proven [47,48], but the evidence that heterodimeric TM interactions constrain αIIbβ3 in a low affinity state is compelling. Thus, scanning the TM helices of full-length recombinant αIIbβ3 with cysteine residues results in the formation of disulfide bonds with a helical periodicity, consistent with the presence of a unique αIIb/β3 TM heterodimer . Moreover, locking the heterodimer with a disulfide bond near its N-terminus prevents cell spreading on fibrinogen and adhesion-promoted tyrosine phosphorylation . Conversely, peptides designed to bind tightly to the αIIb TM helix cause constitutive αIIbβ3 activation by disrupting the αIIb/β3 TM heterodimer [51,52]. Mutations along the β3 TM helix that destabilize the heterodimer have the same effect .
Alanine and leucine scanning mutagenesis of the αIIb TM helix revealed that the same interface mediates its homomeric and heteromeric interactions and consists of the GxxxG-containing sequence VGxxGGxxxL . The β3 TM helix contains a variant of the GxxxG motif, SxxxA, but the interface mediating its homomeric and heteromeric interactions consists of the sequence LxxxVxxxIxxxG instead . This observation is noteworthy because it places SxxxA on the side of the β3 helix facing away from αIIb where it could mediate the association of β3 with other proteins. Moreover, it suggests that juxtaposing the VxxxIxxxG motif of β3 and the GxxxGxxxL motif of αIIb results in a novel, closely packed, and energetically-favorable zipper-like interface (Fig. 2).
The structure of integrin cytosolic tails
Disrupting TM domain interactions alone under experimental circumstances is sufficient to cause integrin activation, but integrin activation under physiologic circumstances occurs when intracellular signals interact with integrin cytoplasmic tails. αIIbβ3 can be activated by truncating the αIIb or β3 cytoplasmic tails just after the conserved membrane proximal sequences GFFKR or LLITIHD, respectively . Moreover, substituting Ala for Arg in GFFKR or Asp in LLITIHD has the same effect, whereas interchanging the residues does not . While indirect, these observations suggest that the membrane-proximal portions of the cytoplasmic tails interact to form an activation-constraining ‘clasp’, a feature of which is a salt-bridge between R995 in αIIb and D723 in β3 . Physiologic signals that activate αIIbβ3 might do so by disrupting the clasp. Nevertheless, neither Ulmer et al.  who used NMR to study the structure of the αIIb and β3 cytoplasmic tails tethered by a coiled-coil nor Li et al.  who analyzed the interaction of αIIb and β3 TM and cytoplasmic tail polypeptides in phospholipid micelles were able to detect the heteromeric association of these proteins.
To confirm the presence of a membrane-proximal cytosolic clasp and determine its structure, peptides corresponding to the αIIb and β3 cytoplasmic tails have been studied in aqueous buffers, dissolved in phospholipids micelles, or anchored to micelles by myristoylation [58–63]. Surprisingly, the structures resulting from these studies are substantially different. For example, Weljie et al.  used NMR to determine the structure of a complex that formed at low ionic strength between an 11-residue αIIb peptide and a 25-residue β3 peptide. When studied alone, neither peptide had secondary structure, but when mixed, the αIIb peptide exhibited α helical structure and two β3 conformers were detected that differed in backbone conformation (one had an elongated structure, and the other was bent back to adopt a closed L shape). A stable salt bridge at αIIb R995/β3 D723 was not detected. Vinogradova et al.  also used NMR to study interactions between full-length αIIb and β3 cytoplasmic tails, either as native peptides or as fusions with maltose-binding protein. When mixed in aqueous buffer, small chemical shift perturbations were detected involving N-terminal residues of β3 and N-terminal residues and C-terminal residues of αIIb, implying that these regions interact. Calculated structures indicated that the N-termini of the αIIb and β3 peptides were helical and the C-terminus of αIIb folded back to interact with the N-terminus. Further, an N-terminal α/β interface was present containing both hydrophobic and electrostatic interactions, including an electrostatic interaction between the guanidyl of αIIb R995 and the carboxyl of β3 D723. Nonetheless, the relevance of these experiments to inactive αIIbβ3 in cells remains unclear. For example, the experiments were performed at stoichiometries considerably different from the 1 : 1 stoichiometry of αIIb and β3 in native αIIbβ3. Further, interactions between αIIb and β3 were transient, rather than stable as might be expected for an activation-inhibiting clasp. Finally, the interactions were disrupted by detergent micelles, an unexpected finding for interactions occurring in the aqueous cytosol. Vinogradova et al.  subsequently used NMR to determine the structure of each peptide in the presence of diphosphocholine (DPC) micelles to mimic a membranous environment. Under these conditions, the N-terminus of αIIb remained helical, but the C-terminus was unable to interact with the N-terminus because the latter was embedded in lipid. β3 residues 716–721 were also embedded in lipid and residues 741 and 747 interacted with lipid as well. As a final step in agonist-stimulated αIIbβ3 activation, the FERM domain of talin binds to the β3 cytosolic tail . Accordingly, Vinogradova et al. found that when the talin FERM domain and β3 peptides were mixed in aqueous buffer chemical shifts for β3 residues T720–D723 and A735–A750 were perturbed. This suggests that in vivo, talin might act by physically disrupting the membrane-proximal clasp at residue 723 and detach the mid-portion of the β3 tail from membranes as well.
Remarkably, considering its short length, the β3 tail interacts with a large number of cytosolic proteins . To identify structural features responsible for these interactions, Li et al.  used NMR to characterize a membrane protein encompassing the β3 TM and cytoplasmic domains. Under conditions where the protein was monomeric in DPC micelles, it was found to fold into three α helices followed by a short unstructured C-terminus. The first helix corresponded to the β3 TM helix and extended into the membrane-proximal region of the cytoplasmic domain. Because the length of this helix likely exceeds that of a typical membrane bilayer, it is either oriented in the membrane in a upright fashion with the membrane-proximal region extending into the cytoplasm, buried in the membrane but tilted at an angle to accommodate its length, or buried upright in the membrane with increased membrane thickness (Fig. 3). A recent NMR analysis of the β3 TM helix embedded in lipid bicelles supports the second scenario . The first helix is followed by a hinge at residues H722–D723, a second helix extending from residues K725–A735, and a turn-helix configuration characteristic of the immunoreceptor tyrosine-based ITAM motif at the C-terminal end. The NPLY motif (residues 744–747) that supports the initial interaction of the β3 tail with the talin FERM domain  is located in a type I β-turn just before the third helix, while the NITY motif (residues 756–759) that interacts with kindlin-3, a second FERM domain-containing protein required for αIIbβ3 activation , is located just after this helix. Finally, approximately 3% of the c-Src present in platelets constitutively interacts with the Arg–Ala–Thr (RGT) sequence present at the C-terminus of the β3 tail .
Integrins, exemplified by the platelet αIIbβ3, undergo complex structural rearrangements as they shift from inactive to active conformations. Central to this process is disrupting a series of constraints imposed by conserved elements of their cytoplasmic, TM, and extracellular domains. Although the nature of many of these constraints has been determined, often at atomic-level resolution, their relative importance in regulating integrin activation remains to be clarified. Nonetheless, it is tempting to speculate that the constraints represent novel targets for manipulating integrin function for clinical benefit.
Disclosure of Conflict of Interests
The authors state they have no conflicts of interest.