Structure of a complete integrin ectodomain in a physiologic resting state and activation and deactivation by applied forces

Abstract

The complete ectodomain of integrin α_IIbβ₃ reveals a bent, closed, low-affinity conformation, the β-knee, and a mechanism for linking cytoskeleton attachment to high affinity for ligand. Ca and Mg ions in the recognition site, including the synergistic metal ion binding site (SyMBS), are loaded prior to ligand binding. Electrophilicity of the ligand-binding Mg ion is increased in the open conformation. The β₃ knee passes between the β₃-PSI and α_IIb-knob to bury the lower β-leg in a cleft, from which it is released for extension. Different integrin molecules in crystals and EM reveal breathing that appears on pathway to extension. Tensile force applied to the extended ligand-receptor complex stabilizes the closed, low-affinity conformation. By contrast, an additional lateral force applied to the β subunit to mimic attachment to moving actin filaments stabilizes the open, high-affinity conformation. This mechanism propagates allostery over long distances and couples cytoskeleton attachment of integrins to their high affinity state.

Introduction

Integrins are cell adhesion receptors that transmit bidirectional signals across the plasma membrane and link the extracellular environment to the actin cytoskeleton. The conformation of the integrin extracellular domain and its affinity for ligand are dynamically regulated by a process termed inside-out signaling. By coupling to the actin cytoskeleton, integrins promote firm adhesion and provide traction for lamellipodium protrusion and locomotion. In migrating cells the adhesiveness of integrins is spatially and temporally regulated so that integrins are activated near the leading edge to support lamellipod extension and deactivated near the trailing edge to facilitate uropod retraction and internalization (Alon and Dustin, 2007; Arnaout et al., 2005; Broussard et al., 2008; Evans and Calderwood, 2007; Luo et al., 2007).

Integrin α_IIb β₃, the most abundant receptor on platelets, binds to fibrinogen and von Willebrand factor, and mediates platelet aggregation and association with injured vessel walls. Inherited mutations in the α_IIb or β₃ subunits result in the bleeding disorder Glanzmann’s thrombasthenia. Antagonists to α_IIb β₃ are prescribed for the prevention of thrombosis (see (Springer et al., 2008; Xiao et al., 2004) and references therein).

The integrin α and β subunits have large N-terminal extracellular domains, single-pass transmembrane domains, and usually short C-terminal cytoplasmic domains. The entire ectodomain of α_V β₃ crystallized in a bent conformation revealed 10 of 12 domains (Xiong et al., 2001; Xiong et al., 2004; Xiong et al., 2002). A ligand-binding head formed by both subunits is followed in each subunit by legs that connect to transmembrane domains. The knees between the upper and lower legs are extremely bent. Integrin epidermal growth factor-like (I-EGF) domains 1 and 2 at the β-knee were disordered in the previous α_V β₃ structure. Crystals of β₂ leg fragments containing I-EGF domains 1 and 2 have been solved in two different orientations (Shi et al., 2007), but the conformation of these domains in the bent integrin conformation remains unknown.

Subsequent to the α_V β₃ crystal structure, mutational studies on cell surface integrins and EM studies on α_V β₃, α_L β₂, and α_X β₂ integrins demonstrated that the bent conformation is the physiologically relevant, low affinity integrin conformation (Nishida et al., 2006; Takagi et al., 2002). Nonetheless, a cryo EM study on α_IIb β₃ revealed a different, less compact conformation with a different arrangement of leg domains (Adair and Yeager, 2002). Furthermore, two recent studies have revealed extended conformations of α_IIb β₃ but failed to find a bent conformation (Rocco et al., 2008; Ye et al., 2008). Crystal structure studies on α_IIbβ₃ are important to resolve these controversies. Revealing the structure within a complete ectodomain of the bent β-knee is important for understanding the mechanism of integrin extension. Moreover, no bent integrin crystal structure to date has been described in light of current knowledge that this conformation occurs on cell surfaces, and corresponds to the low-affinity state.

Most studies find that upon activation, integrins extend (Luo et al., 2007). Upon extension, the headpiece can remain in the closed conformation, as when bent, or transition to an open conformation with high affinity for ligand, as shown in crystals of the α_IIb β₃ headpiece bound to ligands (Xiao et al., 2004). In contrast, a “deadbolt model” posits that activation can occur in the absence of extension (Arnaout et al., 2005). Binding of cytoskeletal proteins such as talin and kindlins to the integrin β cytoplasmic domain appears to interfere with α/β cytoplasmic domain association, and induce integrin extension (Wegener and Campbell, 2008). However, there is currently no known feature of integrin structure that would enable cytoskeleton binding to couple to the extended, open conformation with high affinity for ligand. This would appear to be important to fulfill the key role of integrins in integrating the extracellular and intracellular environments.

Three closely linked metal ion binding sites in the β I domain are especially important for ligand binding. Mg²⁺ at the central, metal ion-dependent adhesion site (MIDAS) directly coordinates the acidic sidechain shared by all integrin ligands. In previous unliganded, bent α_V β₃ structures, the MIDAS and one adjacent site were unoccupied, and it was proposed that metal binding was either caused by integrin activation or induced by ligand binding (Xiong et al., 2002); however, crystals have not been reported with a combination of the two metal ions important for integrin ligand binding, Mg²⁺ and Ca²⁺. Therefore, in current comparisons between low and high affinity β I domain conformations, the changes associated with ligand binding and metal binding cannot be deconvoluted.

Here, we describe the structure of platelet integrin α_IIb β₃ in the bent conformation. Crystals with Ca²⁺ and Mg²⁺ show that physiologically in the low affinity state the metal binding sites in the β I domain are fully occupied. Furthermore, the conformation is revealed of I-EGF domains 1 and 2 at the β-knee, at the epicenter of conformational change. The arrangement of the legs within the bent structure and variation among structures in interdomain orientation have profound implications for the mechanism of integrin activation. Use of this information in models of extended integrins experiencing forces at sites of cell adhesion reveals how integrin affinity is regulated by force exerted parallel to the membrane by a motile actin cytoskeleton. Integrin structure and mechanochemistry provide a natural mechanism for increasing integrin affinity upon cytoskeleton attachment and decreasing it upon cytoskeleton disassembly.

Results and Discussion

α_IIb β₃ crystal structure and negative stain EM

A 2.55Å resolution crystal structure of the complete α_IIb β₃ ectodomain with Ca²⁺ and Mg²⁺ has been refined to an R_free of 26.8% (Fig. 1A, Table 1). In comparisons to α_V β₃ below, differences in resolution and refinement should be kept in mind. The 3.1Å α_V β₃ structure is refined to an R_free of 36.7% (Xiong et al., 2004). α_IIb β₃ has 95% and 0.4% residues in favored and outlier Ramachandran regions, respectively, and geometry in the 98th percentile (where 100 is the best); whereas α_V β₃ has 76% and 6.7% residues in favored and outlier regions, respectively, and geometry in the 21st percentile; as reported by MOLPROBITY (Davis et al., 2007). Waters, which are important in hydrogen bonding and metal coordination, are in the α_IIb β₃ but not the α_V β₃ structure, as appropriate for their respective resolutions. No cis-prolines are present in the α_V β₃ structure, whereas 6 are present in α_IIb β₃. Two cis-prolines, Pro-163 and Pro-169, are in the ligand-binding β₃ I domain. The region around cis-Pro-169 has electron density typical for the α_IIb β₃ structure (Fig. S1). There is a shift in the sequence-to-structure register between α_V β₃ and α_IIb β₃ at β₃ 167–176, in the specificity-determining loop that forms the outer rim of the ligand-binding pocket in the β₃ I domain. Thus, the α_IIb β₃ structure provides details about backbone conformation, hydrogen bonding, and sidechain packing that are important for understanding ligand and metal binding; and for accurate molecular dynamics simulations and structure-guided mutagenesis. Furthermore, for the first time, the structure factors for an integrin ectodomain have been deposited, opening access to the experimental electron density upon which the atomic models are based.

Figure 1.

The α_IIb β₃ crystal structure. A. Cartoon diagram of molecule 1 in α_IIb β₃ crystals. Ca and Mg ions are shown as gold and silver spheres, respectively. Disulfides are shown as gold sticks and glycans are displayed as thinner sticks with grey carbons. C and N-termini are shown as small spheres. Loops with missing density are shown as dashes. B. A model of α_IIb β₃ extended by torsion at the α and β-knees. C-E. Superpositions of molecules 1 and 2 of α_IIb β₃ and α_V β₃ (Xiong et al., 2004) showing breathing. C. A view showing variation in the distance of the lower α-leg from the lower β-leg, opening its cleft, and variation in the lower β-leg: α_IIb β₃ molecule 1 (α_IIb grey, β₃ cyan) and α_V β₃ (α_V yellow, β₃ magenta). D. A view of the α-subunit only, rotated about 90° from the view in C, showing variation in the distance of the lower α-leg from the upper α-headpiece: α_IIb β₃ molecule 1 (yellow) and molecule 2 (cyan); α_V β₃ (magenta). E. The headpieces of α_IIb β₃ molecule 1 (cyan) and α_Vβ₃ (magenta), showing breathing at the β I/hybrid domain interface. All figures are made with PYMOL. F-H. Negatively stained α_IIb β₃ EM projection averages. F. α_IIb β₃ with a C-terminal coiled-coil clasp. G. α_IIb β₃ with the clasp removed. H. α_IIb β₃ disulfide-bonded near the C-termini of the β-tail and calf-2 domains. Panels 1–4 show representative class averages. Panel 5 shows the 20 Å resolution-filtered α_IIb β₃ crystal structure projection that best cross-correlates with panel 1. Panel 6 in F and G shows the masked headpiece region from panel 4, and panel 7 shows the corresponding best-correlated α_IIb β₃ headpiece crystal structure projection. Ribbon diagrams in panels 6 and 8 are in the same orientation (although enlarged) as the projections to their left. Numbers in panels 5 and 7 are normalized cross-correlation coefficients. White and yellow scale bars are 100 and 50 Å, respectively.

Table 1.

X-ray diffraction data and refinement

Protein	αIIb β3 ectodomain	αIIb β3 headpiece
Spacegroup	P41	P62
Unit cell (a, b, c) (Å)	81.3, 81.3, 654.6	332.1, 332.1, 88.3
(α, β, γ) (°)	90, 90, 90	90, 90, 120
Wavelength (Å)	0.97934	0.9760
Resolution (Å)	50-2.55	45-2.90
Number of reflections (total/unique)	614,293/135,066	1,251,268/122,126
Completeness (%)	98.6/93.9*	98.3/93.9*
I/σ(I)	12.2/2.1*	17.4/3.0*
Rmerge (%)¶	7.1/56.6*	9.7/60.2*
Rwork¶¶/Rfree‡‡	0.233/0.268	0.174/0.196
RMSD: Bond (Å)	0.003	0.006
Angle (°)	0.736	0.659
Ramachandran plot**	95.0%/4.6%/0.4%	96.9%/2.9%/0.2%
PDB code		(prev. 1TYE)

^*Asterisked numbers correspond to the last resolution shell.

^¶R_merge = Σ_h Σ_i |I_i(h) −<I(h)> |/Σ_h Σ_i I_i(h), where I_i(h) and <I(h)> are the ith and mean measurement of the intensity of reflection h.

^¶¶R_work = Σ_h||F_obs (h)|−|F_calc (h)||/Σ_h|F_obs (h)|, where F_obs (h) and F _calc (h) are the observed and calculated structure factors, respectively. No I/cutoff was applied.

^‡‡R_free is the R value obtained for a test set of reflection consisting of a randomly selected 1.3% subset of the data set excluded from refinement.

^**Residues in favorable, allowed, and outlier regions of the Ramachandran plot as reported by MOLPROBITY (Davis et al., 2007).

Overall bent structure

The overall arrangement of domains in the two independent α_IIb β₃ molecules in the asymmetric unit is similar to that in α_V β₃ crystals (Fig. 1C), except for differences in angles between domains (Table 2) that give insights into breathing. A similar bent conformation in solution with physiologic divalent cations is seen for three distinct α_IIb β₃ constructs in negative stain EM with class averaging (Fig. 1F–H and S2). The bent integrins from the three types of constructs are indistinguishable from one another (Fig. 1F panels 1–3, G panels 1–2, H panels 1–4) and show excellent cross-correlation with the α_IIb β₃ crystal structure (Fig. 1F panels 1 and 5, G panels 1 and 5, and H, panels 1, 5 and 6). One construct was clasped by appending to the α and β ectodomain C-termini 15-residue linkers containing TEV protease sites, followed by an α-helical coiled-coil (Nishida et al., 2006). Association near the C-termini of the α and β subunit ectodomains provided in vivo by association between the α_IIb and β₃ transmembrane domains (Luo et al., 2004) is mimicked by the clasp (Takagi et al., 2002). The clasped α_IIb β₃ particles were 64% bent and 32% extended (with 4% unclassified) (Fig. 1F). Unclasped particles, in which the clasp was removed with TEV protease, were 44% bent and 52% extended (Fig. 1G). A third construct, which was identical to that used in crystallization, contained cysteines introduced in C-terminal portions of the α_IIb and β₃ subunits in positions that resulted in efficient disulfide bond formation in cell surface integrins (Fig. S3). The disulfide-bonded construct was 100% bent (Fig. 1H).

Table 2.

Variation in inter-domain angles in integrins^a.

Domain interface	bent αIIb β3b	bent αIIb β3c	bent αIIb β3d	open αIIb β3e	bent β3f	frag β2g
	bent αIIb β3	bent αV β3	open αIIb β3	open αIIb β3	frag β2	frag β2
α β-propeller -α thigh	0.5°	9.7–9.9°	-	-	-	-
α thigh - α calf1	1.3°	19–20°	-	-	-	-
α calf1 - α calf2	3.4°	14–17°	-	-	-	-
β I -β hybrid	0.2°	6.7–6.7°	58–70°	1.2–12°	-	-
β hybrid - β PSI	0.4°	7.2–7.8°	2.0–11°	1.8–9.2°	18–27°	5.0–8.3°
β PSI - β I-EGF1	0.8°	-	3.4–13°	5.7–18°	5.5–40°	5.5–41°
β hybrid - β I-EGF1	0.5°	-	6.8–23°	11–26°	29–51°	3.7–46°
β I-EGF1 - β I-EGF2	0.4°	-	-	-	140–170°	67°
β I-EGF2 - β I-EGF3	1.2°	-	-	-	8.4–8.5°	-
β I-EGF3 - β I-EGF4	2.3°	5.4–7.2°	-	-	-	-
β I-EGF4 - β ankle	1.2°	10–11°	-	-	-	-
β I-ankle - β TD	46°h	18°I	-	-	-	-
α β-propeller - β I	0.2°	3.0–3.3°	1.7–2.8°	0.6–1.1°	-	-

^aEach pair of domains from two molecules were superposed using the first domain, and the change in angle upon superimposing the second domain was calculated. Dashes indicate where no comparison is possible, because only one or no domain pairs are available.

^bTwo molecules in current structure (1 × 1).

^cTwo molecules in current structure versus PDB code IU8C (2 × 1).

^dTwo molecules in current structure versus PDB 2VDR and three molecules in PDB ITYE (2 × 4).

^eComparisons among 2VDR and three molecules in ITYE (3 × 4/2).

^fTwo molecules in current structure, and PDB 1U8C versus PDB 1YUK, PDB 2P26, and PDB 2P28 (3 × 3 to 3 × 1, depending on fragment length).

^gcomparisons among PDB 1YUK, 2P26, and 2P28 (3 to 1 comparisons depending on fragment length).

^hResidues common to molecules 1 and 2 in TD are used, 606–612

ⁱα_IIb β₃ molecule 1 compared to α_V β₃

The differing proportion of bent particles in the three preparations shows that tighter association near the C-termini correlates with maintenance of the bent conformation, and also, with resistance to activation on the cell surface (Fig. S3 and Suppl. Mat). This is in agreement with work on other soluble integrin preparations, and a large body of work on cell surface integrins, which has shown that association of the α and β subunit transmembrane and cytoplasmic domains stabilizes integrins in the low-affinity state and in the bent conformation (reviewed in (Luo et al., 2007)).

Significance of the bent conformation

Similar bent conformations have previously been described in EM studies of the resting states of α_V β₃, α_X β₂, and α_L β₂ (Nishida et al., 2006; Takagi et al., 2002). Furthermore, extensive studies using mutations and antibodies to ligand-induced binding sites show that α_IIb β₃ is compact on the cell surface when resting, and extended when activated (Honda et al., 1995; Luo et al., 2007). The similarity in packing of two independent examples of α_IIb β₃ and of α_Vβ₃ in crystal lattices and similar appearance of multiple soluble integrins in EM, together with the work cited above, strongly suggest that the bent crystal structure determined here is representative of the resting state of most, if not all, integrins. However, three cryo EM, EM, and hydrodynamic studies of detergent soluble α_IIbβ₃ from platelets have reached conclusions that are incompatible with one another, and with the domain arrangement seen here (Adair and Yeager, 2002; Rocco et al., 2008; Ye et al., 2008). The difficulty in obtaining a consensus view on α_IIbβ₃ structure may reflect the delicate equilibrium between bent and extended structures (Fig. 1G–H), averaging over ensembles of bent and extended conformations, the poor association of the α_IIb and β₃ transmembrane domains in detergent (Wegener and Campbell, 2008), and dissociation of the heterodimer into the α_IIb and β₃ subunits (Carrell et al., 1985).

Conceptual advances since the previously described α_V β₃ crystal structures allow us to describe the bent α_IIb β₃ crystal structure in light of its physiological relevance as the low affinity integrin state, and as the starting point for integrin extension. Furthermore, the α_IIb β₃ structure reveals I-EGF domains 1 and 2, and a highly acute bend between them in the bent conformation (Fig. 1A). In contrast, I-EGF domains 2, 3, and 4 extend in an almost straight orientation, with an approximate 90 left-handed twist between successive domains, to cover most of the length of the lower β-leg (Fig. 1A, 2A). The β-knee, at the junction between I-EGF1 and I-EGF2, is flanked on one side by the PSI domain and on the other by a knob-like projection in the thigh domain (Fig. 1A). The PSI and the knob are like goalpost uprights, which define the passage for I-EGF1 and I-EGF2. The importance of the knob is emphasized by its structural conservation between α_IIb and α_V, in contrast to the flexibility of loops at the opposite end of the thigh domain, adjoining the β-propeller domain (Fig. 2D).

Figure 2.

Integrin leg domains. A. The knee and lower β-leg of α_IIb β₃. Dashes mark gimbal flexion positions. B. Superposition using SSM (Krissinel and Henrick, 2004) of α_IIb β₃ I-EGF domains 1–4 in the same color scheme as in A. C. Superposition using I-EGF2 of I-EGF 1 and 2 module pairs. Domains from α_IIb β₃ are in red, and those from β₂ fragments (Shi et al., 2007) are in cyan and grey. D-F. Superposition of α_IIb (yellow) and α_V (magenta) (Xiong et al., 2004) thigh (D), calf-1 (E), and calf-2 (F) domains.

The I-EGF domains of the lower β-leg are deeply buried in a narrow crevice, between the upper α-leg on one side and the upper and lower α-leg on the other, with the β I and β-propeller domains helping to form the back of the crevice (Fig. 1A). The flexible calf-1 DX loop extends into the cleft and partially shields the lower β-leg (Fig. 1A, 2E). Exit of the lower β-leg from the crevice appears to be the key step in integrin extension.

Overall extended structure

In the extended conformation, the α and β-legs straighten at the knees, and extend away from the headpiece (Fig. 1F and G, panel 4). The closed headpiece from the crystal structure cross-correlates excellently with the headpiece in EM (Fig. 1F and G, panels 6–8) showing that with Ca²⁺ and Mg²⁺, extended α_IIb β₃ predominantly assumes the closed headpiece, rather than the open headpiece conformation (Fig. 3A). Most extended class averages, whether with clasped or unclasped α_IIb β₃, show the α-leg crossing over or under the β-leg (Fig. 1F, G, panel 4). Leg crossing appears to be a consequence of upper leg configuration in the bent conformation with the long axis of I-EGF1 pointing toward the α-knee (Fig. 1A). When the bent crystal structure is extended at the α and β-knees, leg crossing results (Fig. S5A). However, the legs are highly flexible (see below), and for clarity are shown side-by-side in Fig. 1B. Extended integrins with crossed and uncrossed legs have also been seen for activated α_V β₃, α_X β₂, and detergent soluble α_IIb β₃ integrins (Iwasaki et al., 2005; Nishida et al., 2006; Takagi et al., 2002).

Figure 3.

Metal ion rearrangements in β I domain activation. A. Superposition of headpieces from our unliganded-closed structure and liganded-open α_IIb β₃ (Springer et al., 2008). The β I and hybrid domains are yellow (open) and magenta (closed) while PSI and I-EGF1 domains are red and green, respectively. The α-headpieces are cyan (open) and grey (closed). B. Enlarged view of β I domains with major differences in yellow (open) and magenta (closed). C and D. β I domain metal coordination sites in unliganded-closed α_IIb β₃ (C) and liganded-open α_IIb β₃ (D). Ca (gold) and Mg (green) ions are large spheres; waters (red or pink) are smaller spheres. N atoms are blue and O atoms are red or pink. Metal coordination and hydrogen bonds are dashed. The loop bearing M335 moves far away in (D). E. Superposition at the β I MIDAS. F. Superposition at the α I MIDAS of unliganded-closed (PDB code 1LFA) and liganded open α_L (PDB code 1T0P), in the same orientation as the β I MIDAS in D. In C-F, carbons for unliganded-closed and liganded-open integrins and for ligands are wheat, grey, and cyan, respectively. G and H. Electrostatic potential surfaces at the unliganded (G) and liganded (H) binding sites.

After physiological activation of α_IIb β₃ on platelets or treatment with high concentrations of ligands, multiple ligand-induced binding site (LIBS) epitopes are exposed. These epitopes map to the lower β-leg, and to the PSI domain (Honda et al., 1995). The lower β-leg is buried in a cleft in the bent conformation (Fig. 1A), but will be exposed in the extended conformation (Fig. 1B and S5A). Similarly, the LIBS epitope in the PSI domain, mapped to residues 1–6 (Honda et al., 1995), is masked by I-EGF2 in the bent conformation (Fig. 1A). By contrast, this epitope is exposed after extension at the I-EGF1/I-EGF2 interface in the β-knee brings I-EGF2 away from the PSI domain (Fig. 1B). The previous functional studies, together with the location of these epitopes within the α_IIb β₃ structure, demonstrate that bent and extended α_IIb β₃ represent latent and activated integrins, respectively, contradict suggestions that α_IIb β₃ is extended in the resting state (Rocco et al., 2008; Ye et al., 2008), and agree with electron tomography of active, detergent soluble α_IIbβ₃ showing that it is extended (Iwasaki et al., 2005).

The ligand binding site is preloaded with metals

Integrins bind ligands at the interface between the α subunit β-propeller domain and β subunit I domain (Xiao et al., 2004; Xiong et al., 2002). These domains associate over an interface far larger than between other integrin domains (Table S1), to form the integrin head (Fig. 1B and 3A). Three metal binding sites formed by loops in the β I domain underpin the ligand binding site (Fig. 3A–C). Strong densities at all three sites reveal that they are occupied when physiologic divalent cations, Ca²⁺ and Mg²⁺, are present (Fig. 3C). Mg²⁺ at the central MIDAS site and Ca²⁺ at the two flanking sites are assigned by the coordination chemistry at these sites and the stronger electron densities at the two Ca²⁺ sites. In contrast, in previous α_V β₃ crystals in absence of ligand, only one divalent cation, either Mn²⁺ or Ca²⁺, was present, and neither the ligand-associated metal binding site (LIMBS) or MIDAS was occupied (Xiong et al., 2001; Xiong et al., 2002).

Since our results show that the LIMBS is not a ligand-associated or induced metal binding site, new nomenclature is required. We propose to rename this Ca²⁺-binding site the synergistic metal ion binding site (SyMBS). Low concentrations of Ca²⁺ synergize with low concentrations of Mg²⁺ for ligand binding to integrins (Marlin and Springer, 1987), and mutational studies show that the SyMBS has a positive regulatory effect on ligand binding, and that the SyMBS is the site responsible for Ca²⁺ synergy (Chen et al., 2003; Mould et al., 2003). The SyMBS designation also honors the finding with α_Vβ₃ crystals that ligand binding at the MIDAS synergizes with Mn²⁺ binding at the SyMBS when Ca²⁺ is absent (Xiong et al., 2002).

Fig. 3C reveals the structural basis for synergy between the SyMBS and MIDAS. Although coordination at the MIDAS is octahedral, with six ligands, four of these are waters. An unusually low number of two Mg²⁺ oxygen ligands come from protein, donated by the sidechains of β₃ Ser-121 and Glu-220. The sidechain of Glu-220 orients between the SyMBS and MIDAS, with one oxygen coordinating the SyMBS Ca²⁺ and the other coordinating the MIDAS Mg²⁺ (Fig. 3C). In the absence of either of these metals, the Glu-220 would likely reorient, and not form a proper coordination to the other site, explaining the basis for synergy.

It is conceptually attractive to find that the physiologically important metals Ca²⁺ and Mg²⁺ are pre-loaded prior to ligand binding. The Asp sidechain of integrin ligands such as Arg-Gly-Asp (RGD) directly coordinates the MIDAS Mg²⁺ (Fig. 3D). In the absence of Mg²⁺, approach of this Asp would be electrostatically repelled by Glu-220, Asp-119, and Asp-251 around the MIDAS. Furthermore, the SyMBS site is completely buried, and would be difficult to occupy after ligand binding.

The major differences between low and high affinity ligand binding site conformations are the movements of the β1- α1 loop and its bound Adjacent to MIDAS (AdMIDAS) Ca²⁺ toward the MIDAS Mg²⁺, as previously described (Xiao et al., 2004; Xiong et al., 2002) (Fig. 3C–E). In a similar movement of the β1- α1 loop in α I domains, the MIDAS metal ion moves 2 Å away from an Asp and towards a Thr in the high affinity state, and thus becomes more electrophilic for the acidic ligand residue (Arnaout et al., 2005; Luo et al., 2007) (Fig. 3F).

The current structure enables comparison for the first time of MIDAS metal ion and water positions between the low and high affinity states of β I domains. Our structure reveals that in contrast to α I domains, there is no lateral movement of the MIDAS metal ion across the ligand-binding pocket (Fig. 3E,F). In α I domains, the second Ser of the DXSXS motif (α_L Ser-141 in Fig. 3F) moves to the position of the Mg²⁺ ion, and pushes the Mg²⁺ toward Thr-206 in the high affinity conformation (Fig. 3F). In β I domains, the Mg²⁺ is already in a ligand-binding position in the low affinity state, and the second Ser of the DXSXS motif, Ser-123, moves toward the Mg²⁺ from secondary coordination and displaces a water to occupy the primary coordination sphere (Fig. 3C–E).

What then is the basis for the increase in affinity of β I domains in the open conformation? In part, this must be due to the movement toward the MIDAS of the β1- α1 loop bearing the DXSXS motif, enabling two of its backbone N atoms to form stabilizing hydrogen bonds to the two O atoms of the ligand Asp sidechain (Fig. 3D). Additionally, movement in the open state of the AdMIDAS Ca²⁺ toward the MIDAS brings it into primary coordination with Asp-251, enabling Asp-251 to polarize toward the AdMIDAS Ca²⁺ (Fig. 3D) rather than toward the MIDAS Mg²⁺ (Fig. 3C). Furthermore, there is a significant increase of 1 Å in separation between the Mg²⁺ and Asp-119 in the open state (Fig. 3E and Supplement). The backbone, AdMIDAS, and Asp-119 movements all increase the positive potential in the environment of the MIDAS, and promote greater electrophilicity of the Mg²⁺ for ligand in the high-affinity, open state. The overall increase in positive potential near the MIDAS in the high affinity state is apparent in electrostatic potential surfaces (Fig. 2G, H).

The flexible β-knee and β-leg

The structure shows that all 56 cysteines in the integrin β₃ subunit are disulfide-bonded and the disulfides exhibit no rearrangements with respect to integrin fragment structures (Beglova et al., 2002; Shi et al., 2007). As shown below, the disulfides are compatible with large inter-domain rearrangements, and there is no need for disulfide reduction for integrin activation as previously suggested (Yan and Smith, 2001).

Integrin EGF domains

Eight cysteines are disulfide bonded in a C1-C5, C2-C4, C3-C6, and C7-C8 pattern in the small, 37–50 residue I-EGF domains (Fig. 2B). The exception is I-EGF1, which lacks the C2-C4 disulfide. This enables the C1-C3 loop to occupy a position different than in other I-EGF domains (Fig. 2B) and prevents clashes and enables greater flexibility at interfaces with the PSI and I-EGF2 domains.

The connection between tandem I-EGF domains is gimbal-like, since two flexion points are revealed at the highly acute bend between I-EGF1 and I-EGF2 at the β-knee (dashed lines, Fig. 2A). Between tandem I-EGF domains, and also at the PSI/hybrid and hybrid/I-EGF1 junctions, only one residue intervenes between the last Cys of one domain and the first Cys of the next, limiting flexion at this C-X-C junction (dashed line i., Fig. 2A). However, the disulfide unique to I-EGF domains between C1 at the N-terminus, and C5 in the β-hairpin turn between the two β-strands (Fig. 2B,C), is surprisingly flexible. Most of the movement at the β-knee occurs at a second point of flexion within the tip of I-EGF2, in the C1-C5 disulfide and in the disordered loop connecting C1 and C2 (dashed line ii., Fig. 2A). This is evident from superpositions of β3 I-EGF domains 1–4 (Fig. 2B), and β2 leg fragments (Fig. 2C). The position of C1 is highly variable in these superpositions, demonstrating flexiblity of the N-terminal tip of the I-EGF domain, particularly in I-EGF2. Although previous examples of I-EGF1/I-EGF2 interfaces in β2 leg fragments had been termed bent and extended (Shi et al., 2007), the two fragments differ in angle by 140 to 170 from I-EGF1/I-EGF2 in α_IIb β₃, and both are extended compared to the highly bent conformation revealed here (Table 2 and Fig. 2C). Thus the gimbal-like connection between I-EGF domains is permissive of extreme rotations, and a wide range of inter-domain orientations.

A corollary to the gimbal-like junction at I-EGF domains is that flexibility should be related to the length and disorder of the polypeptide chain between C1 and C2 (or between C1 and C3 in I-EGF1 which lacks the C2-C4 disulfide). As fitting for its role in extension of the β-leg, the C1-C2 loop of I-EGF2 is the only disordered I-EGF loop in our structure, and is located at the apex of the β-knee (Fig. 1A and 2A). Furthermore, alignment of integrin β subunits β1 to β8 in diverse vertebrates shows that I-EGF2 has the longest C1-C2 loop, with 9 to 13 residues. By contrast, the C1-C2 loop has 4 residues in I-EGF3, and 6 to 7 residues in I-EGF4. In I-EGF1 the 8–10 residue length of the C1-C3 loop is consistent with variations of up to 41° at the PSI/I-EGF-1 junction in comparisons to α_IIb β₃ headpiece and β₂ leg fragments (Table 2, Fig. S7). Furthermore, disorder of I-EGF1 and I-EGF2 in the α_Vβ₃ crystal structure (Xiong et al., 2001) implies flexibility at the PSI/I-EGF1 and I-EGF2/3 interfaces.

The β-ankle

A disulfide bonded loop between I-EGF4 and the β-tail domain, previously defined as part of I-EGF4, has no equivalent in other EGF domains, and is termed here the β-ankle. It is not integrated by backbone hydrogen bonds into I-EGF4 or the β-tail, and is likely to be flexible in extended integrins.

The β-tail

Flexibility between the β-ankle and β-tail is substantial, with up to 46° variation (Table 2). Flexibility within the N-terminal α-helix of the β-tail is suggested by loss of density in molecule 2 midway through this helix, and by the weak density of the β-tail domain in molecule 1. The better order of the β-tail in α_V β₃ appears due to an unusually large lattice contact of 860 Ų, which is larger than any domain-domain junction within the β-knee or lower β-leg (400–550 Ų, Table S1).

A contact between the CD loop of the β-tail domain (termed the “deadbolt”) and the 7-helix of the β I domain has been proposed to inhibit integrin activation (Arnaout et al., 2005); although the size of this interface at 60 Ų is too small to be significant (Janin, 1997). There is no such contact in α_IIb β₃, either in molecule 1 where the orientation of the β-tail differs, or in molecule 2 where this part of the β-tail is disordered. Since the β I domain in both molecules is in the inactive, closed conformation, the CD loop does not act as a deadbolt to restrain integrin activation. In agreement, mutation or deletion of the CD loop has no effect on activation of cell surface α_V β₃ or α_IIb β₃ integrins (Zhu et al., 2007a).

Integrin breathing and extension

Breathing in the bent conformation

In molecule 2 of α_IIb β₃ compared to molecule 1, the lower α-leg swings outward at the genu, away from both the upper α- and β-legs, thus widening the crevice in which the lower β-leg is buried (Fig. 1D). The I-EGF4 and β-tail domains also swing with the α-leg, away from the upper β-leg. In α_Vβ₃ crystals, the lower α-leg swings out further, and the I-EGF domains 3 and 4 in the lower β-leg move away from the hybrid domain in the upper β-leg (Fig. 1C, D). About 10% of α_V β₃ and α_IIb β₃ particles exhibit substantially more opening between the lower legs and headpiece than other particles (Fig. 1G, panel 3) (Takagi et al., 2002). This opening is similar in directionality, but greater in amplitude than in the crystal structure comparisons.

A second component of breathing motion is swinging of the hybrid domain relative to the β I domain. The hybrid domain is more swung out in α_V β₃ than in α_IIb β₃ (Fig. 1E) and has a smaller interface with the β I domain. Swinging opens the crevice in which the lower β-leg is buried, and has the same directionality as transit to the open headpiece.

A flexible β-ribbon extension of the β2- β3 loop of β-propeller blade 5 (Fig. 1A) also evidences breathing. It differs in position between α_V β₃ and α_IIb β₃ (Fig. 1E), and between α_IIb β₃ structures (Fig. 3A). An introduced disulfide bond between the β-ribbon and I-EGF4 stabilizes α_Vβ₃ and α_IIb β₃ in the bent conformation and prevents integrin activation on cell surfaces (Takagi et al., 2002). The residues mutated to cysteine are close in α_V β₃ and not in α_IIb β₃, demonstrating that the β-ribbon can differ significantly in position in bent α_IIb β₃ (Supplement).

The main components of motion, hinging of the legs and swinging of the hybrid domain, provide a plausible pathway for integrin extension. Each of these components is confirmed to be a low frequency and hence important normal mode (not shown). Both types of motions will allow release of the lower β-leg from its crevice between the upper β-leg and lower α-leg (Fig. 1A).

Extension

After release of the lower β-leg, the highly flexible I-EGF1/I-EGF2 interface could transit from its bent to extended conformation, while at the same time, extension occurs at the α-genu, resulting in a switchblade-like opening of the integrin. Flexibility described above at the PSI/I-EGF1, I-EGF1/I-EGF2, and β-ankle/β-tail interfaces is also important to enable the lower α and β-legs to extend without clashing near the knees, where the long axis of I-EGF1 points toward the α-knee (Fig. 1A). Reorientation of the ligand-binding head so that it points away from, rather than toward the cell surface, and extends further above it, will greatly facilitate ligand binding (Fig. 1B). Furthermore, extension frees the hybrid domain from extensive interfaces in the bent conformation (Table S1), making hybrid domain swing-out to the open conformation less energetically costly. However, it is sterically possible for hybrid domain swing-out to occur simultaneously with, or precede, extension.

Integrin ectodomain structure provides a mechanism for force-induced integrin activation and deactivation

Binding of the actin cytoskeleton-associated proteins talin and kindlins to specific residues in the integrin β subunit cytoplasmic domain is crucial for inside-out activation of integrins and bidirectional signal transmission (Moser et al., 2008; Wegener and Campbell, 2008). Talin and kindlins bind through their FERM domains to distinct NPX(Y/F) motifs in integrin β subunit cytoplasmic domains that are membrane proximal and distal, respectively. Talin binding to the β subunit cytoplasmic domain has been proposed to sterically interfere with α and β subunit cytoplasmic domain association. One model is that talin binding would alter the depth in the membrane of the β subunit transmembrane domain or its angle with respect to the α subunit transmembrane domain. Another model is that talin binding would cause dissociation of the α and β subunit transmembrane domains, i.e. separation in the membrane (Luo et al., 2004; Wegener and Campbell, 2008). However, steric interference with α and β subunit association is more difficult to envision for kindlins, which bind to a membrane-distal motif that is disordered in α β cytoplasmic complexes (Wegener and Campbell, 2008).

Crosslinking the integrin α and β subunit transmembrane domains, and fluorescence resonance between probes attached to the cytoplasmic domains, have each demonstrated that both inside-out and outside-in signaling require separation of the α and β subunit transmembrane domains, and are associated with separation of the cytoplasmic domains (Luo et al., 2007). Transmembrane and cytoplasmic domain separation would lead to separation of the lower α and β-legs, which in turn would destabilize their interfaces with the head and upper legs, and trigger integrin extension. However, extended integrins can have either the closed or open headpiece with low or high affinity for ligand, respectively. This begs the question of whether the open, high affinity headpiece conformation could be enforced by signals within the cell. The complete integrin ectodomain structure described here now allows an analysis of how force is transmitted between the ligand binding site and the transmembrane domains to regulate transition between low and high affinity states.

There is great interest in the concept that force is important in regulating the adhesiveness and conformation of integrins (Alon and Dustin, 2007; Astrof et al., 2006; Evans and Calderwood, 2007; Puklin-Faucher et al., 2006). However, a key factor that has not previously been considered is the lateral force exerted by actin cytoskeleton treadmilling or contraction. Talin and kindlins link integrins through a molecular clutch to the actin cytoskeleton. As a consequence, integrins move laterally on the cell surface at speeds up to 130 nm/s, in the same direction as actin filaments (Hu et al., 2007; Kaizuka et al., 2007). Thus, integrin binding to talin and kindlins is essentially synonymous with attachment to the cytoskeleton, and exertion of a lateral force on the β-subunit cytoplasmic domain.

Let us envision at the molecular level the consequences of dragging an integrin by its β-tail across the cell surface. Bent integrins have a cross-section of 80 × 100 Å near the plasma membrane that is unusually large for a cell surface glycoprotein. The dense packing of extracellular domains on the cell surface is one of the major barriers to diffusion (Sheetz, 1993). Thus, as an integrin is dragged across the cell surface, it would be buffeted by collisions with other cell surface glycoproteins. The thin, lower β-leg is shielded in its cleft by the robust α-subunit and upper β-leg and head (Fig. 1A), which will bear the brunt of buffeting. Thus the integrin will be forced through a gauntlet of other cell surface proteins, that will strip the α subunit and upper β-leg away from the lower β-leg that they shield. Furthermore, the calf-2 domain has an unusually broad base for a cell surface domain, and two long unstructured loops near the plasma membrane, one of which is cleaved during biosynthesis (Fig. 2F). Frictional forces due to interactions of these segments and the α subunit transmembrane domain with the plasma membrane will also pull the α-leg away from the β-leg.

These forces are on pathway with the breathing movements described above, and will shift the equilibrium toward integrin extension. Although the interfaces buried in the bent conformation are extensive (Table S1), they have low shape complementarity, are mainly hydrophilic, and are readily replaced by water, as shown by the shift in equilibrium toward extension upon C-terminal clasp cleavage (Fig. 1F–H) (Nishida et al., 2006; Takagi et al., 2002). Because of the central role of the lower β-leg in these interfaces (Table S1, Fig. 1A), its removal from the cleft will destabilize the bent conformation far more than clasp removal, and result in extension. Furthermore, extension places the bulky head of the integrin above the height of most cell surface glycoproteins, reducing frictional drag and favoring maintenance of extension.

Once the integrin is extended, lateral pulling on the β-tail will align the integrin so that β orients toward and α away from the pulling direction. Orientation will be enhanced by the frictional resistance of the three leg domains in α, which, at 140–170 residues and with extra β-strands compared to Ig domains, are unusually stout (Supplement) and will act as a sea anchor. In contrast, β-leg domains are at the lower size limit found on cell surfaces.

Extended and not bent integrins are competent for binding large biological ligands such as fibronectin and fibrinogen (Zhu et al., 2007a). Once a ligand in the extracellular matrix or on the surface of another cell is bound, resistance to lateral pulling by the β-subunit cytoplasmic domain will increase greatly, with a corresponding increase in the lateral force. Because the β-subunit hybrid domain extends laterally away from the ligand binding site in the open headpiece conformation (Fig. 3A), the lateral force should stabilize the open headpiece conformation.

To test how forces regulate the conformation of the integrin headpiece, we used the complete ectodomain structure and the liganded, open headpiece structure (Springer et al., 2008; Xiao et al., 2004) to build extended integrin models with the headpiece closed or open and bound to the specific peptide recognition motif in fibrinogen (Fig. 4A and D). The C-termini of the α and β subunit ectodomains were tethered to a plane to mimic the plasma membrane. To mimic a cell pulling away from ligand in matrix or on the surface of another cell, a tensile (pulling) force normal to the membrane was applied to the α and β ectodomain C-termini, and an equal and opposite resisting force was applied to ligand (Fig. 4C and F). This mimics forces experienced by the integrin and ligand during cell adhesion in the absence of attachment of the integrin to the cytoskeleton. Alternatively, the same membrane normal forces were applied, and an additional lateral force parallel to the membrane was applied to the C-terminus of the β-subunit ectodomain that was resisted by an opposite force applied to ligand (Fig. 4B, E). This mimics forces experienced during cell adhesion in the presence of attachment of the integrin β subunit to talin or kindlins and the actin cytoskeleton.

Figure 4.

Regulation of integrin conformation by tensile force. Molecular surfaces show α_IIb in yellow and β3 in magenta with hybrid domain in green. Fibrinogen peptide ligand is shown in cyan as Cα spheres. A and D are starting models. B and C are derived from A, and E and F from D, after applying tensile forces (arrows) in steered molecular dynamics simulations to α_IIb and β3 C-terminal and ligand N-terminal atoms shown as large spheres. Models in A-F are aligned by superposition on the β-propeller and β I domains. Numbers show the distance after superposition of C-terminal residue 433 of the hybrid domain from the closed (A) and open (D) conformations. The underlined distance shows the conformation that models most closely resemble.

In the presence of both the membrane-normal force exerted by ligand and the lateral force exerted by the cytoskeleton, the extended-open integrin remained open (Fig. 4E). In the case of the extended-closed integrin (Fig. 4B), the lateral force was transmitted through the lower β-leg domains to the hybrid domain, which swung out and assumed an orientation similar to that in the open headpiece (Fig. 4B). Thus, binding to the actin cytoskeleton provides an active mechanism for separating the integrin α and β legs and inducing the extended-open integrin conformation with high affinity for ligand at sites of actin polymerization and contraction.

In the presence of only the force pulling the integrin away from ligand, the α and β legs remained together, and were extended (Fig. 4C, F). The closed-extended integrin remained closed (Fig. 4C). In the case of the open-extended integrin, the greater leg extension induced by tensile force caused the hybrid domain to swing inward, towards the closed headpiece conformation (Fig. 4F). Thus, a tensile force exerted on the ligand-integrin complex in the absence of cytoskeleton engagement stabilizes the closed, low-affinity state of the headpiece, opposite to previous expectation (Alon and Dustin, 2007). Thus, force-induced stabilization of the low-affinity state provides an active mechanism for downregulating integrin adhesion in migrating cells at sites of actin cytoskeleton disassembly, including the uropod, and enables integrins and plasma membrane to be internalized for transport in intracellular vesicles toward the front of the cell (Broussard et al., 2008).

Implications for integrin-mediated cell adhesion and migration

The above results show that inherent in integrin structure is a mechanism for activating the ectodomain by attachment of the actin cytoskeleton to the β subunit cytoplasmic domain, explaining the structural basis for inside-out signaling (Fig. 5). New sites of attachment of integrins to matrix are formed in the lamellipodium, where actin polymerization and branching is rapid (Broussard et al., 2008; Choi et al., 2008). Actin filaments in this region actively move backward toward the lamella. Integrins that associate with the actin cytoskeleton move rapidly along the cell surface, as observed with nanobeads (Giannone et al., 2003) and would be extended by the frictional buffeting forces described above (Fig. 5B). Once the integrins bind ligand, the ligand will resist the pulling force, increase the lateral tensile force, and thereby stabilize the high affinity state (Fig. 5C). Moreover, there is a positive feedback loop, because ligand binding reinforces integrin attachment to the cytoskeleton, and is important in the maturation of nascent adhesions (Broussard et al., 2008; Giannone et al., 2003). Thus, firm adhesion to the matrix is established at sites in the lamellipodium where actin polymerization occurs and contact with ligand is made, further reinforcing attachment between integrins and the cytoskeleton. This provides traction near the lamellipodium to support further protrusion of the leading edge of the cell. As new membrane is added and the cell moves forward, the nascent adhesion sites will be left behind, and find themselves in the lamella. Here, double-headed myosin II is present, which contracts actin filaments, and nascent adhesions can mature to focal adhesions, or disassemble (Broussard et al., 2008; Choi et al., 2008).

Figure 5.

The integrin cycle. A. In the bent conformation, integrins have low affinity for ligand. B. At sites where actin filaments are formed, the integrin β subunit cytoplasmic domain binds through talin or kindlins. Lateral translocation on the cell surface and buffetting cause integrin extension. Both open and closed headpiece conformations are putatively present. C. Binding to an immoblized extracellular ligand greatly increases the lateral force, and markedly favors the high-affinity, open headpiece conformation. D. Disassembly of the actin cytoskeleton removes the lateral force. Tensile force between the ligand and the integrin cytoplasmic domains favors the closed headpiece conformation, and ligand dissociation. E. Ligand dissociates, further favoring the closed headpiece conformation. F. In the absence of ligand and tensile force, the bent conformation is favored, completing the cycle, and the integrin returns to the same state as shown in (A).

For cell migration, deadhesion at the rear of the cell is just as important as adhesion at the leading edge. When integrins are artificially locked in the high affinity state, or actin disassembly in the uropod is blocked with rho kinase inhibitors, uropod retraction is blocked, and cells become highly elongated and stop migrating (Smith et al., 2007). As cells migrate over a substrate, the integrins that remain bound to ligand on the substrate will eventually find themselves in the uropod. Cytoskeleton disassembly in the uropod is coordinated with destabilization of adhesions (Broussard et al., 2008). In the absence of a lateral force on the β subunit, the normal force exerted on the integrin as the uropod pulls away from ligand stabilizes the closed headpiece with low affinity for ligand (Fig. 5D). This favors ligand dissocation (Fig. 5E). The integrin can thus return to the low affinity, bent conformation, completing the integrin adhesion/deadhesion cycle (Fig. 5F).

Activation of the high affinity state by the lateral pull exerted by the actin cytoskeleton provides an elegant solution to the problem of coordinating ligand binding by integrins to attachment to the cytoskeleton. The importance of the cytoskeleton for integrin function is supported by many studies. For example, actin cytoskeleton-disrupting agents inhibit integrin adhesion stimulated by inside-out signaling, i.e., affinity regulation (although not avidity regulation) (Kim et al., 2004). Actin filament movement is highly coordinated with movement of ligand-bound integrins into the ring-shaped immunological synapse, and actin poisons rapidly disperse integrins but not other adhesion molecules from the synapse (Kaizuka et al., 2007).

Our simulations used comparable forces normal and lateral to the cell surface as might be found in the lamellipodium; however, in focal contacts where actin stress fibers terminate, the lateral force exerted on the β tail by the cytoskeleton may be much greater than the normal force. This would only increase the tendency of lateral force to stabilize the open headpiece conformation with high affinity for ligand. In geometries in which no force is applied to the α tail, applying tensile force to the β tail alone will stabilize the high affinity state, since in the open headpiece the β subunit is much more extended than in the closed headpiece, with distances between the ligand-binding MIDAS in the β I domain and C-terminus of the hybrid domain of 83 and 64 Å, respectively (Astrof et al., 2006). In typical representations integrins extend approximately normal to the cell surface (Fig. 5). However, the direction of integrin extension will be determined by the force vectors. When lateral force is stronger than normal force, integrins will extend more in the direction parallel than normal to the membrane, i.e. they will lean over. Such a change in integrin orientation is consistent with measurements of cell-substrate distances by interference reflection microscopy and EM (Verschueren, 1985). The height to which integrins extend above the surface of 200 to 250 Å in our simulations is similar to the distance of 300 Å between the plasma membrane and extracellular matrix at close contacts in the lamellipodium of migrating cells. However, a membrane-matrix distance of 100 to 150 Å is found at focal contacts, consistent with an extended, leaned-over integrin conformation in the presence of strong lateral force.

We believe that application of lateral force is the most physiologically relevant mechanism for activating integins, but not the only one. Separation of the transmembrane and cytoplasmic domains, in the absence of an applied force, appears sufficient to induce integrin extension. This should enable at least a small fraction of integrins to transition to the extended-open conformation, and may be sufficient for integrin adhesiveness in many commonly employed assays. Disruption of association between membrane-proximal cytoplasmic regions may be sufficient to induce extension, and may explain why deletion of either the integrin α or β subunit cytoplasmic domains is activating (Lu et al., 2001; Lub et al., 1997). However, in contrast to other activation mechanisms, association with the actin cytoskeleton uniquely enforces selection of the extended-open conformation over the bent and extended-closed conformations, and selectively enhances the adhesiveness of those integrins that are experiencing tensile forces as a consequence of simultaneous binding to ligand and the cytoskeleton (Fig. 5C). Thus, this mechanism exquisitely supports the function of integrins in integrating cell adhesion and cell migration.

The mechanism described here is mechanochemical, since force alters the chemical equilibrium between conformational states and drives integrins from a bent, closed-headpiece, low affinity state to an extended, open-headpiece, high affinity state. The mechanism is also allosteric, with force as the allosteric effector. Previously, it has not been thought possible to transmit signals through flexible protein domains, because allostery involves relative changes in orientation and position, and flexibility decouples the relative positions of sending and receiving domains. However, tensile force extends and imparts stiffness to flexible proteins, and thus enables signal transduction through otherwise flexible protein regions. The outward swing of the hybrid domain, and with it the PSI domain in the upper β-leg, transmitted through stiffened β-knee and β-leg domains, elegantly couples rearrangements at the ligand-binding site in the β I domain to lateral force exerted by the actin cytoskeleton.

The integrin mechanochemical mechanism works well whether talin or kindlins link integrins to actin microfilaments that are moving as a consequence of myosin contraction, extension/polymerization at the leading edge, or treadmilling. This mechanism would not work for reception of signals from soluble ligands, but is uniquely well suited for adhesion receptors. This structural mechanism for linking cytoskeleton binding to ligand binding appears to be at the heart of the integrating function that gives integrins their name, and enables them to provide the traction for cell migration. Outside-in signaling by integrins requires both α and β subunit transmembrane domain separation, and clustering (Miyamoto et al., 1995; Zhu et al., 2007b). Cytoskeleton association may contribute to both of these components, first by inducing transmembrane domain separation, and second by cooperating with binding to multivalent ligands in inducing clustering.

No doubt further levels of complexity are added by the many members of the integrin family, including the distinctive β4 subunit; the large number of proteins that interact with integrin β cytoplasmic tails, including many such as talin and kindlins that contain FERM or protein tyrosine-phosphate binding domains, and other proteins that bind to the α tail (Wegener and Campbell, 2008).

Experimental Procedures

Crystallography

α_IIb and β₃ ectodomains were fused to C-terminal segments containing a protease site, coiled-coils, and tags, with or without α_IIb-L959C and β₃-P688C mutations to introduce a disulfide bond, and expressed in CHO Lec 3.2.8.1 cells. Purified α_IIb-L959C/β₃- P688C with the C-terminal tag removed by protease in buffer containing 1mM CaCl₂ was crystallized in 10% PEG 3350, 50 mM magnesium acetate, and 0.1 M imidazole, pH 7.0. Diffraction data collected at 19-ID of APS was solved using molecular replacement in spacegroup P4₁. Final refinement with REFMAC5 utilized TLS and NCS. Crystals contain two molecules per asymmetric unit. Density is present for all ectodomain residues (α_IIb 1–959 and β₃ 1–690) except for five loops, and in one molecule, the C-terminal portion of the β-tail domain. Thirteen or 18 N-linked carbohydrate residues are visualized in each molecule. I-EGF1 from the complete α_IIb β₃ ectodomain was used to model density for this domain in re-refined α_IIb β₃ headpiece structures with (Springer et al., 2008) or without Fab (Table 1).

Negative stain EM

The clasped and unclasped α_IIb β₃ was purified on a Superdex 200 HR column in Tris saline, 1 mM Ca²⁺, 1 mM Mg²⁺. The peak fraction was adsorbed to glow discharged carbon-coated copper grids, stained with uranyl formate, and inspected with an FEI Tecnai 12 electron microscope operated at 120 kV. Images were acquired at a nominal magnification of 67,000 x. Imaging plates were scanned and digitized with a Ditabis micron imaging plate scanner (DITABIS Digital Biomedical Imaging System, AG, Pforzheim, Germany) using a step size of 15 μm and 2 × 2 pixels were averaged to yield a final pixel size of 4.46 Å at the specimen level. 2,000–5,000 particles were interactively collected, windowed into 75 × 75-pixel individual images, and subjected to ten cycles of multi-reference alignment and classification. Images were processed and cross-correlated using SPIDER (Frank et al., 1996) as described (Nishida et al., 2006).

Steered molecular dynamics

To build an extended model with closed headpiece, bent α_IIb β₃ was extended at the junction between thigh and calf-1 domains in α, and between I-EGF1 and I-EGF2 in β. For extended α_IIb β₃ with open headpiece, we substituted the headpiece bound to the 10-residue peptide from the C-terminus of fibrinogen γ subunit. The same peptide ligand was added to extended-closed α_IIb β₃. Smaller adjustments were made at the hybrid/I-EGF1, I-EGF2/3, and β-ankle/β tail interfaces, guided by angles found in other structures and the gimbal-like nature of I-EGF domain interfaces, to retain the calf-2/β-tail interface. To accelerate simulations with lateral force, the α- and β-legs were pre-separated by changing the angles between thigh and calf-1 domains and between I-EGF1 and I-EGF2 domains.

During simulations, the N-terminus of the ligand (chain C, residue 402) was kept fixed, and forces were applied to the C-termini of the integrin α_IIb β₃ ectodomain (α_IIb, residue 963 and β₃, residue 690). Models were aligned in the same coordinate system using the β-propeller and β-I domains. The coordinate system was chosen such that the C-termini are confined to the xy plane mimicking the cell membrane, and the centers of masses of β propeller and β I domains are aligned parallel to the y axis. To mimic talin binding, a lateral force of 15 pN was applied at the C-terminus of β in the -y direction, away from α, as would occur on cells because of frictional resistance to pulling by α. To mimic tensile force extending the ligand-integrin complex, models were pulled along the -z direction at the C-termini of the α_IIb and β₃ subunits, each with force of 15 pN. The protein structures were solvated in a 4 Å thick water shell and neutralized with Na ions. Molecular dynamics simulations were with NAMD (Phillips et al., 2005) using the CHARMM22 force fields. The supplement contains further details and rationale