An internal ligand-bound, metastable state of a leukocyte integrin, α_Xβ₂

Integrin α and β subunits contain large ectodomains, one transmembrane domain each, and typically short cytoplasmic domains (Fig. 1). The α and β subunits come together to form the integrin head, which connects through upper and lower legs in each subunit to the cell membrane. Of 18 integrin α subunits, nine lack and nine contain an αI domain. In αI-less integrins, ligand binds at the interface between the α subunit β propeller and β subunit βI domains that form the head. αI integrins bind ligand to the αI domain, which is inserted in the β propeller near its interface with the βI domain (Luo et al., 2007; Springer and Dustin, 2012).

αI and βI domains are structurally homologous. They have a metal ion-dependent adhesion site (MIDAS) with a Mg²⁺ ion that binds an acidic residue in ligands and two states, closed and open. In the open state, a change in configuration in loops around the MIDAS is conveyed to the opposite end of the I domain where it is inserted at its N- and C-terminal ends in a neighboring domain, the α subunit β propeller domain for the αI domain and the hybrid domain for the βI domain (Fig. 1). Allostery is conveyed by the I domain C-terminal α7 helix, which pistons along its axis ∼10 Å in the C-terminal direction between the closed and open states (Fig. 1, C–F). Closed and open states have been defined by crystal structures of isolated αI domains (Lee et al., 1995; Luo et al., 2007). A crystal structure of complement receptor type 4 (CR4; integrin α_xβ₂) revealed a closed αI domain in the context of an intact α_Xβ₂ ectodomain, which was bent with a closed headpiece and closed βI domain (Fig. 1 A; Xie et al., 2010). The αI domain was surprisingly flexible and was visualized in only two of ten different lattice environments, where αI domain orientation was stabilized by specific lattice contacts.

Integrins exist in three overall conformational states (Luo et al., 2007; Springer and Dustin, 2012): bent with a closed headpiece (Fig. 1 A), extended with a closed headpiece (Fig. 1 C), and extended with an open headpiece (Fig. 1 F). In extension, the knees in the α and β subunits assume more obtuse angles, interfaces between the headpiece and lower legs become solvent exposed, and the ligand-binding headpiece extends further above the cell surface and assumes a better orientation for binding ligand on other surfaces (Fig. 1 C). In headpiece opening, reshaping of the βI domain MIDAS near its interface with the α subunit β propeller domain is transmitted by βI domain α7 helix pistoning to the interface with the hybrid domain, causing the hybrid domain to pivot about its other connection to the βI domain and swing out away from the integrin α subunit by 75 Å at the knees (Fig. 1, D and F).

Under resting conditions, the α_xβ₂ ectodomain is predominantly bent, and in this state the headpiece is closed. The bent and extended, closed conformations have low affinity, whereas the extended, open conformation has high affinity, as demonstrated for both α_Xβ₂ and α_Lβ₂ (Chen et al., 2010; Schürpf and Springer, 2011).

Thus, opening of the headpieces of α_Lβ₂ and α_Xβ₂ integrins transmits a signal that increases affinities of their αI domains for ICAM-1 and iC3b, respectively. It has been proposed that in relay the C-terminal portion of the αI domain α7 helix and its following C linker, including an invariant Glu in αI domains, binds to an interface between the β propeller and βI domains similar to the ligand-binding interface in αI-less integrins (Alonso et al., 2002; Shimaoka et al., 2002). A bell rope–like pull on the αI α7 helix appears to activate αI integrins (Yang et al., 2004; Weitz-Schmidt et al., 2011). However, the crystal structure of bent α_Xβ₂ with a closed αI domain showed that the invariant Glu, α_X Glu-318, was quite distant from the βI MIDAS, ∼16–19 Å away (Xie et al., 2010). How relay is accomplished remains unknown and is currently the most outstanding unanswered question about allostery transmission in integrin extracellular domains.

On cell surfaces, integrins likely exist in dynamic equilibrium between all three states (Fig. 1, A–F), with cellular activation and association with the cytoskeleton and ligand regulating this equilibrium temporally and spatially (Springer and Dustin, 2012). How is transition between the states achieved on a timescale rapid enough to regulate leukocyte adhesiveness in the vasculature (Shamri et al., 2005)? Here we describe internal ligand-bound and cocked conformations of β₂ integrins that may function to accelerate equilibration between the bent-closed and extended-open conformational states and reveal how allostery is relayed between αI and βI domains. Furthermore, we discover unprecedented extensional and rotational flexibility in an allosteric machine and specialization among integrin β subunits in the intermediate states they can assume between closed and open.

Previous α_Xβ₂ crystals at 3.5–3.95 Å resolution with metal-chelating salts as precipitants (Xie et al., 2010) lacked metals at the MIDAS and synergistic metal binding site in the βI domain (Fig. 2 A). We deleted seven N-linked glycosylation sites and added a disulfide between the membrane-proximal α_X calf-2 and β₂ β-tail domains to obtain greater conformational homogeneity of α_Xβ_2. Crystals in PEG 8000 with 0.2 M magnesium acetate and 0.1 M sodium cacodylate, pH 7.2, revealed metals at all three βI domain sites and diffracted at 2.75 or 2.9 Å resolution (Table 1 and Figs. S1 and S2).

The α_Xβ₂ ectodomain in the new crystal form was bent; surprisingly, the αI domain was open (Fig. 2, B and E). This open conformation of the αI domain was stabilized by the crystal lattice (Fig. 3). In opening, the β6-α7 loop reshapes as it moves with the pistoning α7 helix (Lee et al., 1995). A lattice contact with the αI domain β6-α7 loop selects the open conformation of this loop; in the closed conformation, the β6-α7 loop would extend upward in the orientation shown in Fig. 3 and severely clash with the lattice. Lattice contacts with the opposite side of the αI domain hold the αI domain in a specific position in the crystal lattice (Fig. 3). Meanwhile, other lattice contacts surround the headpiece, upper legs, and lower legs of the α_Xβ₂ ectodomain and stabilize its bent conformation (Fig. 3).

The αI domain in bent, internal ligand-bound α_Xβ₂ has all canonical features of open αI domains. Compared with closed αI domains, the Mg²⁺ ion at the αI MIDAS moves 2.5 Å, breaking and gaining direct coordinations to Asp-240 and Thr-207, respectively (Fig. 4, A and B). Density around the αI MIDAS Mg²⁺ is strong and includes an additional feature that is modeled as a Cl⁻ ion (Fig. 4 A). The β1-α1 loop bearing the MIDAS-coordinating DXSXS motif and upper part of the α1 helix shift and the backbone of the β4-α5 loop flips (Fig. 4, A and B). Furthermore, the β6 strand tilts relative to β6 in closed αI domains, and the β6-α7 loop moves toward the C terminus with the pistoning α7 helix (Fig. 4 C).

In contrast, the conformation of the α7 helix of the open αI domain in an intact ectodomain differs markedly from that seen in isolated open αI domains (Fig. 4 E). In isolated αI domains, the α7 helix moves as a unit 10 Å in the C-terminal, axial direction between the closed and open conformations (Fig. 4 D). In internally liganded α_Xβ₂, the αI domain α7 helix retains few helical hydrogen bonds and its C-terminal portion is completely unwound (Fig. 4 C). Stretching is spring-like. The distance moved between the closed and internally liganded α_Xβ₂ structures progressively increases from 11 Å at residue 306 to 26 Å at Glu-318 (Fig. 4 C), and the α7 helix elongates to take an overall straight course between the αI domain and the binding interface that includes the βI MIDAS (Fig. 4 E). The open conformation of the αI domain and the binding of the internal ligand to the βI domain are associated with a 70° reorientation of the αI domain relative to the closed-bent conformation (Fig. 2, A and B; and Fig. S2).

αI domain residues following the β6-α7 loop in the internally liganded α_Xβ₂ ectodomain have a similar local conformation as in the open isolated α_M αI domain, but differ in position by a rigid body movement (Fig. 4, G and H). These segments are Asn-301 to Leu-308 in α_M and Asp-299 to Leu-306 in α_X. Side chains of α_M Asn-301 and α_X Asp-299 hydrogen bond to the backbone of the n + 2 residue to form an “N-cap” motif that caps the N-terminal end of the α7 helix (Fig. 4, G and H). Previous α_M-isolated αI structures showed Phe-302 was buried in the closed conformation and surprisingly exposed (156-Ų solvent-accessible surface area) in the open conformation (Fig. 4, D and H). In the open αI domain in intact α_Xβ₂, the equivalent Phe-300 packs against the side of the domain (Fig. 4, C and G) and has markedly less exposure (65 Ų). In the isolated open α_M αI domain, Leu-305 is partially exposed (15 Ų; Fig. 4 H); the relative rotation and tilt of the remnant α7 helix in intact α_Xβ₂ completely bury the equivalent α_X residue, Leu-303 (0 Ų exposure; Fig. 4 G). As a result, α_X residue Leu-303 occupies the position of a nonequivalent open α_M residue, Ile-308 (Fig. 4, G and H).

The remarkable elongation of the αI domain α7 helix is enforced by internal ligand binding. Residues in the C-terminal portion of the αI α7 helix lose all contact with the αI domain and completely reshape to form an internal ligand (Fig. 2, C–E; and Fig. S3, A and B). At the tip of a loop formed in the internal ligand, the invariant and mutationally important α_X Glu-318 side chain (Huth et al., 2000; Alonso et al., 2002; Yang et al., 2004) coordinates the βI MIDAS Mg²⁺ ion and hydrogen bonds to the backbone of the βI β1-α1 loop (Figs. 2 C and 5 A). The strength of these polar interactions is increased by their burial by hydrophobic internal ligand residues α_X Ile-314, Phe-315, and Ile-317 (Fig. 2 C). Furthermore, internal ligand residue Lys-313 helps bury Phe-315 and forms a salt bridge to β propeller residue Glu-331 (Fig. 5 D). Hydrogen bonds involving Thr-320 secure a type II β turn in the internal ligand with Glu-318 at its tip (Fig. 5 E). Another network of hydrogen bonds around Lys-313 and Glu-331 secures the conformation of the segment following the internal ligand, i.e., the C linker, and residues that form the binding pocket in the β propeller domain (Fig. 5 D). The position of the C linker completely reshapes during allostery relay (Fig. 2, D and E) and helps to bury the internal ligand (Fig. 2 C). The internal ligand has low B factors and excellent electron density (Fig. S3 A). The dehydrated structure has an identical internal ligand structure and hydrogen bonds, whereas the C linker differs in conformation at residues 322–324 (Figs. S3 B and S4).

The binding pocket for the internal ligand is formed by hydrophobic residues in the α_X β propeller and β₂ βI domains (Figs. 2 C and 5 A). The βI domain contributes MIDAS loop residue Tyr-115, synergistic metal ion binding site residue Leu-208, and Pro-168 and Pro-170 of the disulfide-bonded specificity-determining loop (Figs. 2 C and 5 A). The β propeller contributes hydrophobic residues Phe-328 and Met-332 (Figs. 2 C and 5 A).

The binding pocket for the internal ligand (Fig. 5 A) is not unlike that for external ligands in αI-less integrins (Fig. 5, B and C). The closest resemblance is to α₄β₇. β₇ Leu-236 is in the same position as β₂ Leu-208, and β₇ Leu-236 interacts with the phenylalanine-like moiety of an antagonist similarly to how β₂ Leu-208 interacts with internal ligand residue α_X I317 (Fig. 5, A and B, arrows). Interestingly, similar phenylalanine-based antagonists of β₂ integrins block relay between the αI and βI domains, and binding of these antagonists to β₂ integrins requires the βI MIDAS and not the αI domain (Shimaoka et al., 2003a).

Structurally important hydrophobic internal ligand and binding pocket residues equivalent to α_X Ile-314, Phe-315, Ile-317, Phe-328, and Met-332 are invariantly conserved in all nine αI integrin subunits (Fig. 5 F). Also conserved are Gly-319 and Thr-320, which stabilize the β turn at Glu-318, and Lys-313 and Glu-331, which interact and nucleate many hydrogen bonds at the C linker (Fig. 5 F). These findings suggest that the relay mechanism visualized here in α_Xβ₂ will be general for all αI integrins.

Binding to the internal ligand induces “cocking” of the βI domain (Fig. 6 A). The adjacent to MIDAS (ADMIDAS) metal ion, β1-α1 loop, α1 helix, and α1′ helix of the βI domain all move toward the internal ligand Glu-318 side chain, enabling it to form multiple hydrogen bonds to the β1-α1 loop backbone (Figs. 5 A and 6 A). Electron density for the βI α1 helix and α1′ helix is excellent in both native and dehydrated crystal structures (Fig. S3, C and D).

Val-124, Leu-127, Leu-131, Leu-135, and I-138 are hydrophobic residues that anchor the α1 and α1′ helices in their groove; they all move toward the βI MIDAS in the cocked conformation (Fig. 6 A). Of these, Val-124 and Leu-127 alternately occupy a single hydrophobic ratchet pocket in the cocked conformation. Val-124 is buried in the ratchet pocket in closed βI and partially exposed in cocked βI (Fig. 6 A). In contrast, Leu-127 is substantially solvent exposed in the closed structure and becomes buried in the cocked structure in the ratchet pocket vacated by Val-124 (Fig. 6 A). Movement of Leu-127 into the ratchet pocket is also associated with nearby backbone movements in the α1-α1′ loop that shorten α1 and lengthen α1′.

The α1 and α1′ helix displacements in the cocked β₂ structure are greater than in β₁ and intermediate conformations 2–6 of β₃ integrins obtained after soaking Arg-Gly-Asp mimetics into crystals (Fig. 6 C and Table 2; Xiong et al., 2002; Nagae et al., 2012; Zhu et al., 2013). Furthermore, homologous α1 and α1′ helix hydrophobic residues are buried in closed conformations of β₁ and β₃ integrins and remain so in intermediate and open conformations (Fig. 6, C and D). As the α1 and α1′ helices join in a single helix in the open state of the β₃ subunit, β₃ residue Leu-134 moves laterally to occupy a ratchet pocket that is vacated by gatekeeper residue Val-340 in the β6-α7 loop (Fig. 6 D). The semiexposure of homologous β₂ residue Leu-127 in the closed state enables it to vault over gatekeeper residue Val-330, achieving substantial α1 and α1′ helix movement without β6-α7 loop movement in cocked α_Xβ₂ (Fig. 6 A).

Remarkably, dehydration of the α_Xβ₂ crystal reversed cocking. A lattice contact with the βI domain α7 helix inhibits its pistoning and with other lattice contacts prevents swing-out of the hybrid domain (Fig. 3). As the solvent content of crystals decreased from 69 to 64% during dehydration, the lattice contact with the α7 helix expanded to include the adjacent α1′ helix (Fig. 3 B). The dehydrated crystal lattice is incompatible with the position of the cocked α1′ helix in the native crystal. Therefore, it appears that the growth of lattice contacts caused the α1′ helix and in turn the α1 helix to move back toward their positions in the closed conformation and assume a position we term uncocked (Fig. 6 B). Notably, in their cocked positions the βI domain α1 and α1′ helices have no significant lattice contacts in the native crystal (Fig. 3 A); thus it appears that cocking in the native crystal is a closer approximation of changes that occur in the βI domain in response to internal ligand binding.

These results demonstrate that cocking is not necessary for internal ligand binding. The conformation of the internal ligand is essentially identical in the two crystal forms (Fig. S4), and neither the internal ligand nor the wound or unwound portions of the αI domain α7 helix have lattice contacts in the native crystal (Fig. 3 A). Therefore, the energetics of internal ligand binding are sufficient to drive α7 helix unwinding in the absence of cocking. Compared with its position in closed β₂, the βI domain β1-α1 loop moves significantly toward α_X Glu-318 in both the native and dehydrated structures, and the ADMIDAS metal ion moves even more in the dehydrated than native structure (Fig. 6, A and B). Thus, βI domain β1-α1 loop movement and ADMIDAS movement are closely associated with internal ligand binding. Cocking appears to be a normal consequence of, but is not required for, internal ligand binding.

We tested with mutations the functional relevance of the internal ligand and the hypothesis that the internally liganded state and the cocked state are metastable intermediates in the process of integrin activation on cell surfaces. CR4 function was measured using rosetting with iC3b-sensitized erythrocytes (Fig. 7 A), which has previously been shown to require the open headpiece conformation of α_Xβ₂ with hybrid domain swing-out (Chen et al., 2010). We also measured the effect of mutations on activation epitope exposure. Binding of KIM127 antibody to a β leg epitope has been shown to require extension (Nishida et al., 2006), and binding of m24 and MEM148 antibodies to βI and hybrid domain epitopes has been shown to measure headpiece opening (epitopes are marked in Fig. 1 F; Chen et al., 2010).

Internal ligand substitutions α_X K313I, F315E, and I317H, predicted to destabilize internal ligand binding and hence allostery relay, all abolished Mn²⁺-dependent rosetting (Fig. 7, A and B). Mutation of Glu-318 to Ala, Lys, or Met abolished rosetting and mutation to Asp greatly diminished rosetting (Fig. 7 B). Disfavoring the type II β turn at Gly-319 by mutation to Pro diminished rosetting (Fig. 7 B). Internal ligand substitutions were also inhibitory using α_Xβ₂ activated by replacement of the human β₂ subunit with chicken β₂ (Fig. 7 C; Bilsland et al., 1994). These results establish the physiological importance of internal ligand residues to iC3b rosetting. The same mutations abolished, or diminished in the case of G319P, Mn²⁺-dependent exposure of the KIM127 epitope (Fig. 7 D), demonstrating that on cell surfaces binding of the internal ligand to its pocket is required for and associated with extension of α_Xβ₂. Mutation of the same residues also diminished, but to a lesser extent, Mn²⁺-dependent exposure of m24 and MEM148 epitopes (Fig. 7, E and F). Thus, allostery relay is also associated with headpiece opening on cell surfaces.

If an open αI domain associated with a bent ectodomain is metastable, then substitutions designed to stabilize the internal ligand-bound, open αI conformation should stabilize extension and headpiece opening rather than the bent, closed headpiece conformation crystallized here. α_X αI domain residues F300, I306, and I314 are more buried in the β6-α7 loop and α7 helix in bent, closed α_Xβ₂ than in bent, cocked α_Xβ₂. Mutations of these residues induced iC3b rosetting (Fig. 7, A and B), α_Xβ₂ integrin extension (Fig. 7 D), and headpiece opening in Mg²⁺/Ca²⁺ (Fig. 7, E and F). These findings suggest that the open αI domain on cell surfaces is normally associated with the extended, open headpiece rather than the bent, closed conformation of α_Xβ₂. In summary, these mutations were designed to stabilize internally liganded, bent α_Xβ₂ with a closed headpiece and an open αI domain, yet stabilized extension and an open headpiece, demonstrating metastability.

The metastability of the cocked state of the βI domain was independently examined by mutating βI domain ratchet residues implicated in cocking. In cocking, Leu-127 displaces Val-124 from the ratchet pocket and Val-124 becomes partially solvent exposed; therefore, mutation of Val-124 to Ala should stabilize the cocked conformation. However, V124A mutant α_Xβ₂ did not remain bent because in Mg²⁺/Ca²⁺ the KIM127 epitope in the β₂ leg became exposed, demonstrating extension (Fig. 7 D). Furthermore, V124A mutation induced m24 and MEM148 epitope exposure (Fig. 7, E and F). MEM148 binds far from the α1 and α1′ helices, to the inner face of the βI–hybrid interface (Fig. 1 F), and demonstrates that the V124A mutation induces hybrid domain swing-out. Moreover, V124A mutation strongly activated ligand binding by α_Xβ₂ in Mg²⁺/Ca²⁺ (Fig. 7 B). These findings demonstrate that the cocked conformation of the βI domain is metastable; stabilizing cocking induces on cell surfaces the extended-open headpiece conformation rather than the bent, closed headpiece conformation seen here in crystals. The other ratchet residue, Leu-127, is buried in the pocket in the cocked conformation and partially exposed in the closed conformation; therefore, the L127A mutation should inhibit cocking. L127A mutation inhibited Mn²⁺-stimulated CR4 function and KIM127, m24, and MEM148 exposure (Fig. 7). These results are consistent with the cocked conformation of the βI domain resembling, and being on pathway to, the active, high affinity state of α_Xβ₂. Thus, although lattice contacts with the αI domain stabilize its open conformation, other lattice contacts with the body of the ectodomain stabilize the bent conformation with a swung-in hybrid domain.

We have described a novel bent, internally liganded conformation of integrin α_Xβ₂ with an open αI domain. Internal ligand binding induced a cocked conformation of the βI domain; however, internal ligand binding did not require cocking. The protein construct differed from one that previously yielded crystals of a bent α_Xβ₂ ectodomain with a closed αI domain (Xie et al., 2010). However, the removal of seven N-linked glycosylation sites and introduction of a C-terminal disulfide are unrelated to internal ligand binding and cocking; instead, the open conformation of the αI domain is stabilized by a crystal lattice contact with the αI domain β6-α7 loop.

The internal ligand-bound, cocked and internal ligand-bound, uncocked states seen here appear to be at local energy minima between the bent-closed and the extended-open conformations. However, they are metastable in the sense that they may be higher energy than either the bent-closed or extended-open conformations and thus not at a global energy minimum. A precedent for stabilization by a crystal lattice of two alternative conformational states comes from integrin α_Mβ₂, i.e., CR3. α_Mβ₂ and α_Xβ₂ are sister integrins with unusually high sequence identity among α subunits (60%; Yu et al., 2012). The isolated α_M αI domain was found to be in the closed conformation in one crystal lattice and in the open conformation in another (Lee et al., 1995). In the latter lattice, the α_M MIDAS Mg²⁺ ion coordinated with a pseudo-ligand, a Glu residue in another αI domain in the lattice. Although the previous isolated open α_M αI domain and current open α_x αI domain in an intact ectodomain are stabilized by lattice contacts, and these contacts are with different portions of the αI domain, these lattice-stabilized open αI domain conformations are essentially identical to one another and indistinguishable from those of open α₂ and α_L αI domains cocrystallized with biological ligands (Emsley et al., 2000; Shimaoka et al., 2003b). The similarities in open αI domain conformations stabilized by crystal lattices and biological ligands support the concept that crystal lattices select conformational states that are near local energy minima.

Although it is often convenient to think of conformational pathways as flowing in one direction, in general they are reversible. We may think of the crystal lattice as inducing the open αI domain state seen here. However, it is equally appropriate to think of the α_Xβ₂ ectodomain as having multiple conformational states in solution. We may therefore think of the bent, internally liganded, cocked conformation as one of the biological conformations of α_Xβ₂, which enabled crystallization in the particular lattice seen here.

The highly elongated αI domain α7 helix seen here enables much greater flexibility of the αI domain when bound to external ligand than had previously been imagined. Previous studies had shown that β6-α7 loop conformation, but not α7 helix conformation, is canonical for the open conformation (Shimaoka et al., 2003b). Nonetheless, the amount of α7 helix extension seen here was surprising. It demonstrates that the interaction between the internal ligand and its binding pocket is quite strong and provides enough energy to break nine α-helical hydrogen bonds in the α7 helix. Notably, there are no lattice contacts with the α7 helix itself (Fig. 3), showing that its highly elongated conformation is only a function of connecting to the αI domain β6-α7 loop at one end and the internal ligand at the other end. Indeed, the largely unwound α7 helix takes a more or less direct, overall straight path between these points. In the dehydrated crystal form, the orientation between the αI domain and the body of the ectodomain differs, and the unwound α7 helix adjusts to again take a direct path between these points (Fig. 4 F and Fig. S2).

Flexibility between our two internally liganded α_Xβ₂ structures is quantitated in Fig. 4 (I and J). Flexibility in the N linker largely occurs between Cys-126, which is disulfide bonded to the β propeller domain, and Glu-130, which is integrated into the αI domain with a backbone hydrogen bond (Fig. 4 I). Flexibility in the unwound α7 helix occurs between invariant Leu-303, which binds to the hydrophobic pocket below the β6-α7 loop, and Lys-313, which is important in the internal ligand (Fig. 4 J).

Although the specific orientations between the αI domain and body of the α_Xβ₂ ectodomain seen here are trapped in crystals, there is reason to think that the orientation will be similar when α_Xβ₂ engages physiological ligands on cell surfaces. Adhesion and binding to iC3b are strongly associated with extension and require the open headpiece conformation of α_Xβ₂ (Chen et al., 2010, 2012). Our internally liganded, bent structures already reveal the open αI domain and how the internal ligand binds at the β propeller–βI interface. All that remains to make a model of extended-open α_Xβ₂ using bent, internally liganded α_Xβ₂ is to swing out the hybrid domain, using the open headpiece of α_IIbβ₃ as the model, and to extend the α and β subunits at their knees (Fig. 8). Integrins predominantly bind to the cytoskeleton through their β subunit cytoplasmic domains (Gahmberg et al., 2009). Elongational force exerted by the cytoskeleton and resisted by an opsonized particle bound to the α_Xβ₂ αI domain will tend to straighten all the intervening domain–domain orientations. The resulting orientation between the αI and βI domains, and the alignment of their axes, is very similar to that crystallized here (and distinct from that in bent-closed α_Xβ₂ [Fig. 2 A]); a line drawn through the αI and βI MIDAS Mg²⁺ ions also passes through the βI–hybrid interface and continues in a similar direction as the hybrid domain would assume when swung out (Figs. 1 F, 2 B, and 8 C). Furthermore, the high affinity state of the αI domain is more extended between the MIDAS and Glu-318 than the closed state, and the open headpiece is more extended between the βI MIDAS and the end of the hybrid domain than the closed state (Astrof et al., 2006). Force times the difference in distance between conformational states gives a difference in energy. Therefore, cytoskeletal force applied across the integrin–ligand complex will stabilize the open, high affinity state over the closed, low affinity state, and help to offset the tendency of force to rupture adhesive bonds.

Despite these constraints that favor parallel, aligned αI and βI domain axes, there is likely to be considerable latitude for variation in αI MIDAS–βI MIDAS distance and for rotation of the αI domain about the αI MIDAS–βI MIDAS axis. The unwound α7 helix and N linker each have highly extended conformations in the bent, internally liganded structures (Figs. 2 E, 3, and 4 [E and F]). A more compact conformation of the N linker, and less unwinding of the α7 helix, should enable shortening by ∼10 Å of the αI MIDAS–βI MIDAS distance. Lengthening also seems possible with untwisting (see next paragraph). Thus, substantially greater variation in αI MIDAS–βI MIDAS distance may occur than the increase from 63 Å in the native crystal to 65 Å in the dehydrated crystal seen here.

Looking down on the αI MIDAS, both connections of the αI domain to the ectodomain body are twisted ∼180° in the counterclockwise direction in bent, internally liganded α_Xβ₂ (Figs. 2 B and 3), whereas there is essentially no twisting in closed α_Xβ₂ (Fig. 2 A). Because the N linker and unwound α7 helix are each highly extended, further counterclockwise twisting, as well as clockwise untwisting, of internally liganded α_Xβ₂ is easily attainable without creating clashes between the αI domain and the remainder of the integrin headpiece.

Complexes of integrin α_Xβ₂ ectodomain with the C3c moiety of its ligand iC3b (Chen et al., 2012) are suggestive of such αI domain rotation (Fig. 8, D and E). Negative stain electron microscopy shows two classes of C3c α_Xβ₂ complexes with indistinguishable orientations of α_Xβ₂ but orientations of C3c that are flipped 180° (Fig. 8, D and E). The orientation of the integrin αI domain is not discernable. However, the orientation of the remainder of α_Xβ₂ and three bound Fab fragments is quite clear (Fig. 8, D and E, schematics). Similarly, the orientation of C3c, with its distinctive head-like C345C knob, is quite clear. The lack of significant interaction between the body of the open αI domain and the headpiece in the internal ligand-bound α_Xβ₂ crystal structures now allow us to interpret these images. It is unlikely that such a large rotation could occur at the ligand–receptor interface between C3c and the αI domain. In contrast, our structures suggest that rotation can readily occur between the αI domain and the headpiece, and the images in Fig. 8 (D and E) suggest that a rotation of at least 180° is readily obtainable in α_Xβ₂.

The ability of the αI domain to bind ligand at a range of distances and rotations relative to the reminder of the integrin endows αI integrins with great versatility. Although αI-less integrins are present in all metazoans, from sponges to vertebrates, αI integrins are known only in chordates (Whittaker and Hynes, 2002). αI α subunits expanded in vertebrates to comprise half of all integrin α subunits. The flexibility of αI domains, both in the closed state as shown previously (Xie et al., 2010) and in the open state shown here, may give αI integrins an advantage over αI-less integrins for recognition of certain types of ligands. These may include ligands in less accessible environments, including collagens in bundles (integrins α₁β₁, α₂β₁, α₁₀β₁, and α₁₁β₁), adhesion molecules on cell surfaces (integrins α_Lβ₂ and α_Eβ₇), and foci of complement activation on pathogen surfaces (integrins α_Mβ₂ and α_Xβ₂).

In cocking, the βI domain α1′ helix moves more than in any previously described intermediate state of β₁ or β₃ integrins (Fig. 6 and Table 2; Xiong et al., 2002; Nagae et al., 2012; Zhu et al., 2013). We therefore examined β₂ for specializations that could support the cocked conformation. The α1 and α1′ helices lie in a groove formed by the α7 helix on one side and the α2 helix on the other (Fig. 9 A). Any features that facilitate helix sliding in this groove toward the MIDAS would promote formation of the cocked state. One such feature is a wider groove in β₂ that is lined with smaller side chains (Fig. 9 A).

In the most striking specialization of β₂ for cocking, residue Leu-127 in the βI domain α1′ helix is precocked in the closed conformation. Residues corresponding to β₂ Val-124, Leu-127, Leu-131, Leu-135, and Ile-138 are conserved as hydrophobic in all eight vertebrate integrin β subunits (Fig. 9). The corresponding residues in β₁ and β₃ are completely buried in the α1/α1′ groove (Fig. 9, B and D). However, ratchet residue Leu-127 in β₂ is exceptional; as described in Results, it is not buried (Fig. 9 C). Additionally, although α1′ helix residue Leu-131 is buried in β₂ as in other integrin β subunits, it adopts a different rotamer that places its center 3 Å closer to the β1-α1 loop (Fig. 9). The Leu-135 residue in β₂ also is more aligned with and forward in the groove than the equivalent Met residue in β₁ and β₃.

The residue equivalent to β₂ Val-330 in the β6-α7 loop points toward the α1′ helix and may be considered a gatekeeper for βI domain shape shifting (Fig. 9 A). The partially exposed, relatively high position of Leu-127 in the groove in the precocked state of β₂ enables it during cocking to pass Val-330 by vaulting over the gate.

The order of secondary structure movements thus differs in intermediate states of β₂ and β₃ integrins. In β₂, large movements of the α1 and α1′ helices occur with no movement of gatekeeper residue Val-330 in the cocked conformation. In β₃, movement of the gatekeeper Val-340 residue precedes α1′ helix sliding (Zhu et al., 2013).

Do other integrin β subunits have a precocked configuration of βI domain α1 and α1′ helix hydrophobic residues, suggesting they might also undergo cocking? The β₁ and β₇ subunits are particularly interesting because they can associate with both αI and αI-less integrin α subunits (β₂ is unique in only associating with αI α subunits). In β1, the first four hydrophobic α1 and α1′ helix residues are chemically identical to those in β₃, and their side chains are in essentially identical positions. In contrast, in β₇, Leu-155 has a solvent-exposed rotamer identical to that of key β₂ ratchet residue Leu-127 (Fig. 9 E). β₇ like β₂ also has Leu in the fourth hydrophobic position, whereas β₁, β₃, and β₆ all have a Met residue at this position (Fig. 9 F). These comparisons suggest that a cocked conformation might also be found for the leukocyte integrins α₄β₇ and α_Eβ₇.

Cocking thus has been found for β₂ integrins and is predicted for β₇ integrins, which are unique among integrins in only being expressed on leukocytes. β₂ has higher sequence identity to β₇ than to any other integrin β subunit: 46% in the ectodomain and 67% in the βI domain, whereas the values range from 31 to 42% and 43 to 54%, respectively, for the other six integrin β subunits (Fig. S5). This level of sequence identity is consistent with similar regulatory specializations in β₂ and β₇ integrins.

Previously, the process of αI integrin activation has been presumed to require transitions between five states, a bent conformation with separated lower legs (Fig. 1 B), an extended conformation with a closed βI domain (Fig. 1 C), an extended conformation with an open βI domain (Fig. 1 D), and an extended conformation with open βI and αI domains (Fig. 1 F). With the requirement for four transition states, each with an energy barrier that must be traversed before the next state is reached, activation could be slow. The metastable internal ligand-bound and cocked states defined here have several important implications for speeding conformational transitions in β₂ integrins and possibly in β₇ integrins. They directly connect the bent conformation to the extended conformation with open βI and αI domains (Fig. 1 E). Hybrid domain swing-out is unimpeded in the bent conformation (Takagi et al., 2002) and would bring with it the lower β leg, thus inducing lower leg separation and extension as well as headpiece opening in one fell swoop (Fig. 1, E and F). Therefore, all key conformational transitions could be accomplished with a single intermediate, internal ligand-bound state. The internal ligand-bound state of the βI domain found here lowers the energy barrier of the transition state for conformational change by enabling the βI domain to activate the αI domain before βI opening and integrin extension. αI domain opening may thus occur before βI domain opening (Fig. 1 E), instead of in the reverse order (Fig. 1 D). The cocked conformation of the βI domain must lower the energy of the internal ligand-bound state even further because it is found in the absence of countervailing lattice interactions. In terms of α1 and α1′ helix movement, the cocked conformation is more than halfway to the open conformation (Table 2). The internal ligand-bound, cocked, bent conformation is thus clearly on the pathway to the extended conformation with open αI and βI domains (Fig. 1, E and F).

The importance of the bent, internal ligand-bound metastable state in integrin extension and headpiece opening postulated here is supported by previous work on LFA-1 (integrin α_Lβ₂; Salas et al., 2004). This work showed that α_L Glu-310 (equivalent to α_X Glu-318) is important for inducing LFA-1 extension, consistent with passage of β₂ integrins through an intermediate similar to that shown in Fig. 1 E on the pathway to extension. Furthermore, the experimental requirement of α_L Glu-310 for leukocyte rolling and Mn²⁺-induced β₂ integrin extension (Salas et al., 2004) is inconsistent with the pathway for extension shown in Fig. 1 (A–C). Like α_Lβ₂, α_Xβ₂ extension on cell surfaces is facilitated by the internal ligand, as shown here with inhibition by four different Glu-318 mutations of Mn²⁺-dependent KIM127 expression. Thus, equilibration between bent and extended β₂ integrin conformational states on cell surfaces may require brief flickering to metastable conformational states similar to the internally liganded and cocked conformations seen here.

In intact integrins on cell surfaces, the αI domain is likely to spontaneously equilibrate between closed and open states. After αI domain opening, the internal ligand would be exposed to solvent, but probably not yet in its bound conformation. As part of the same molecule, the internal ligand would have a very high local concentration near the internal ligand-binding pocket. Binding to the internal ligand-binding pocked should therefore occur rapidly and yield a structure similar to our internally liganded, uncocked structure. This metastable state would then transition to a lower energy, but still metastable, internally liganded, cocked structure. Opening of the βI domain could then follow with hybrid domain swing-out, destabilizing interfaces between the headpiece and lower legs and between the lower α and β legs, and result in integrin extension. Our two metastable crystal structures thus provide a structural pathway that explains the experimentally demonstrated importance of the Glu of the internal ligand in integrin extension on cell surfaces.

Kinetic rates are determined by the height of the energy barrier that needs to be traversed at the transition state. The metastable transition state defined here implicitly defines a mechanism for lowering the energy barrier for integrin conformational change and thus for increasing kinetics. Measuring these kinetics is an important topic for further research.

A remarkable finding in this study is the extensive interface buried upon binding of the internal ligand, and the extensive reshaping of the αI domain α7 helix and C linker to form the internal ligand. This interface’s size, excellent hydrogen bonding, and burial of the Glu-318 MIDAS interaction by hydrophobic internal ligand residues must contribute to its strength and its ability to pay the energetic cost of extensively unwinding the αI domain α7 helix. The results certainly support previous suggestions that the invariant Glu following αI domains has a function similar to the Asp of Arg-Gly-Asp (Alonso et al., 2002), of “exertion of a bell-rope–like pull on a segment within the C-terminal linker region” and “interaction of the C-terminal linker with the β propeller and/or the βI domain” at a site “equivalent to the ligand-binding site in integrins that lack αI domains,” and that “the αI, β propeller, and βI domains make a ternary interaction interface” (Shimaoka et al., 2002).

Yet the internally liganded and cocked states yield further unexpected discoveries. The surprising lift-off of the αI domain over the β propeller and βI domain platform enables a range of extensional and rotational motions that are without precedent in other allosteric machines. The metastable, internally liganded, cocked structure enables rapid equilibration between bent and extended-open integrin structures and an explicit mechanism for linking integrin extension and headpiece opening. Finally, the difference in order of secondary structure movements in opening of β₂ and β₃ βI domains and the cocked conformation special to β₂ demonstrate that integrin β subunits can be specialized not only to modulate ligand recognition but also to assume different intermediate states between closed and open, with important consequences for signal transmission in integrin ectodomains.

In previous α_Xβ₂ crystal structures (Xie et al., 2010) the most C-terminal residues visible in density, α_X K1082 and β₂ V674, are 15 to 18 Å apart, whereas α_X N920 and β₂ V674 are 7 to 10 Å apart (Cα distances). The α_X N920C and β₂ V674C mutations formed a disulfide and expressed well. α_X subunit cDNA encoding mature residues 1–1082 was fused to a C-terminal 53-residue sequence (5′-PAALQTLFQGPLGAQLEKELQALEKENAQLEWELQALEKELAQHHHHHHA-3′) and inserted in ET1 vector containing an internal initiation site for enhanced GFP (Mi et al., 2008). β₂ cDNA encoding the signal sequence and mature residues 1–674C was fused to C-terminal sequence 5′-GPAALQTLFQGPLGAQLKKKLQALKKKNAQLKWKLQALKKKLAQTTGWRGGHVVEGLAGELEQLRARLEHHPQGQREPHHHHHHA-3′ and inserted in pEF1-puro vector (Takagi et al., 2001). The two C-terminal sequences contained a 3C protease recognition site (bold), ACID or BASE coiled-coils (underlined; O’Shea et al., 1993), respectively, a streptavidin binding peptide sequence (in β₂ construct only; italics), and a hexahistidine tag.

N-linked sites in α_X (nine) and in β₂ (six) were individually tested for effect on transient expression of soluble heterodimer using a capture ELISA. All mutations were done with QuikChange XL (Agilent Technologies). Individual elimination of five α_X sites and four β₂ sites had no adverse effect. After testing combinations of multiple substitutions, we used five substitutions in α_X (N42D, N368S, N678T, N885S, and N920C) and two in β₂ (N190D and N232K), resulting in expression at 30% of wild-type levels. HEK293S N-acetylglucosaminyl transferase I-negative (GnTI^−/−) cells, which express high mannose but not complex N-glycans (Reeves et al., 2002), were transfected using calcium phosphate, isolated by sorting for single GFP⁺ cells, and selected with 10 µg/ml puromycin and 1 mg/ml G418. A clone with high expression was cultured in Ex-Cell 293 medium (Sigma-Aldrich) in square bottles (Muller et al., 2005). Culture supernatant supplemented with 20 mM Tris, pH 8.2, 200 mM NaCl, and 0.25 mM NiCl₂ was centrifuged at 3,000 g, concentrated fivefold using tangential flow with a 100,000 Mr cutoff, diluted twofold, concentrated twofold, and loaded on Ni-NTA (10 ml/l supernatant) equilibrated in 20 mM Tris, pH 8, 150 mM NaCl, 1 mM CaCl₂, 5 mM MgCl₂, and 13 mM imidazole (loading buffer). After washing with 10 column volumes of loading buffer, a Strep-tactin Sepharose (IBA GmbH; 2.5 ml/l supernatant) column was attached downstream and the NTA column was eluted with 10 column volume of loading buffer plus 250 mM imidazole. The Strep-tactin column was washed with 10 column volumes of 20 mM Hepes, pH 7, 150 mM NaCl, 1 mM CaCl₂, and 5 mM MgCl₂ and eluted in the same buffer containing 2.5 mM desthiobiotin. After tag and coiled-coil removal with 3C protease, a final step of size exclusion chromatography using a Superdex S200 column demonstrated that the material was largely monomeric (Fig. S1 A). The α_X and β₂ bands are unusually sharp in SDS-PAGE because of the small number of N-linked sites, and native PAGE showed a single band (Fig. S1, A and B). Final yield was ∼3.5 mg/l of culture supernatant.

Protein was concentrated to 9 mg/ml and screened for crystallization using sitting and hanging drop vapor diffusion at 4°C. Microseeding improved crystal quality. Rod-like crystals formed (Fig. S1 C) in an optimized solution of 6% PEG 8000, 0.2 M magnesium acetate, and 0.1 M sodium cacodylate, pH 7.2, and were cryoprotected by transfer in 5% increments over 10 min to mother liquor containing 30% vol/vol glycerol. One crystal was dehydrated (and cryoprotected) by transfer into 15% PEG 20000, 0.1 M sodium cacodylate, pH 7.2, and 0.2 M magnesium acetate and overnight vapor diffusion equilibration. Crystals were frozen in liquid nitrogen.

Diffraction data collected at APS beamline ID-23 was processed with XDS (Kabsch, 2001). The structure was solved by molecular replacement using the α_X αI domain (PDB ID no. 1N3Y) and β propeller and βI domains (PDB ID no. 3K6S) using PHASER (McCoy et al., 2007). Other domains from 3K6S were placed into the map manually, and an initial model was obtained by refining each domain as a rigid body. After torsion angle–simulated annealing, many rounds were performed of rebuilding with Coot (Emsley and Cowtan, 2004) and refinement with PHENIX (Adams et al., 2010). The model was further validated and improved, and Ramachandran statistics were calculated using MOLPROBITY (Davis et al., 2007). Simulated annealing composite omit 2Fo–Fc electron density was calculated with PHENIX with 128 omitted regions (Adams et al., 2010).

All superpositions were on Cα atoms of indicated domains. Superimposed structures are shown in identical orientations and are aligned vertically or horizontally on the page. Figures were prepared with PyMol. Detailed comparisons of the closed β₂ βI domain use a structure of the α_Lβ₂ headpiece, which has metal ions at all three sites in the βI domain, has better density for the βI domain, and is at higher resolution (2.5 Å) than the α_Xβ₂ ectodomain (Xie et al., 2010).

Rosetta-fixed backbone design (Kuhlman et al., 2003) was used to predict mutations that differentially destabilized the previous (Xie et al., 2010; PDB ID no. 3K6S) and current α_Xβ₂ structures. For each residue from α_X 296 to 326, the Rosetta energy function (score 12) was calculated for each of the 20 amino acids. Mutations were chosen that showed differential effects, destabilized one structure little, and after inspection by molecular graphics appeared unlikely to be changed in effect by minor backbone movements.

Sheep erythrocytes were sensitized with iC3b as follows. To prepare E-IgM-iC3b, erythrocytes were incubated with IgM anti-Forssman mAb M1/87, washed three times, and treated at 37°C for 1 h with human C5-deficient serum as described previously (Bilsland et al., 1994). E-IgM were prepared by the same method but omitting the final sensitization with C5-deficient serum. E-IgM-iC3b and E-IgM as control were assayed for binding to α_Xβ₂ HEK293T transfectants (10⁵ cells/well in cluster-24 plates; Zang and Springer, 2001). In brief, 250 µl of E-IgM-iC3b in HBS was added and incubated for 1.5 h at 37°C. After three washes, rosettes (>10 erythrocytes/HEK293T cell, over 100 HEK293T cells examined) were scored by microscopy.

Immunofluorescent flow cytometry (Lu et al., 2001) used fluorescein (FITC)-labeled CBR LFA1/7 and biotinylated KIM127, MEM148, or m24, referenced elsewhere (Chen et al., 2010), in the presence of 1 mM MgCl₂/CaCl₂ or 1 mM MnCl₂. HEK293T transfectants were stained with phycoerythrin-streptavidin to recognize the activation-dependent epitopes. A single FITC gate was set for all samples that included few nontransfected cells and used to collect PE fluorescence. Mutant phycoerythrin mean fluorescence intensity (MFI) was normalized by multiplying it by (FITC MFI of wild-type)/(FITC MFI of mutant). Results for both rosetting and immunofluorescence are averages ± SD of replicates from at least three independent transfections with three different replicates (transfectants seeded in three different wells) each.

Fig. S1 shows purification, PAGE, and crystals of α_Xβ₂. Fig. S2 shows differences in interdomain angles among different α_Xβ₂ structures. Fig. S3 shows simulated annealing composite omit map density for the internal ligand and the βI domain α1/α1′ helices. Fig. S4 compares the internal ligand and C linker in native and dehydrated structures. Fig. S5 shows sequence relationships among integrin β subunits.

We thank Jianghai Zhu for help with data processing and the staff of Advanced Photon Source beamline 23-ID (GM/CA-CAT).

This work was supported by National Institutes of Health grant NIAID AI072756 and a GlaxoSmithKline fellowship.