Metal ion and ligand binding of integrin α₅β₁

Significance

Three metal ions are at the heart of the ligand binding site in integrin β-subunits. Furthermore, motions at many domain–domain junctions in integrins regulate ligand binding affinity. High-resolution crystal structures and highly accurate solution ligand binding assays are required to understand the mechanisms by which integrins transmit conformational change in their complex dodecadomain ectodomains. Here we provide information fundamental to understanding these processes in the integrin that assembles fibronectin into fibrils in matrices, α₅β₁.

Abstract

Integrin α₅β₁ binds to an Arg–Gly–Asp (RGD) motif in its ligand fibronectin. We report high-resolution crystal structures of a four-domain α₅β₁ headpiece fragment, alone or with RGD peptides soaked into crystals, and RGD peptide affinity measurements. The headpiece crystallizes in a closed conformation essentially identical to that seen previously for α₅β₁ complexed with a Fab that allosterically inhibits ligand binding by stabilizing the closed conformation. Soaking experiments show that binding of cyclic RGD peptide with 20-fold higher affinity than a linear RGD peptide induces conformational change in the β₁-subunit βI domain to a state that is intermediate between closed (low affinity) and open (high affinity). In contrast, binding of a linear RGD peptide induces no shape shifting. However, linear peptide binding induces shape shifting when Ca²⁺ is depleted during soaking. Ca²⁺ bound to the adjacent to metal ion-dependent adhesion site (ADMIDAS), at the locus of shape shifting, moves and decreases in occupancy, correlating with an increase in affinity for RGD measured when Ca²⁺ is depleted. The results directly demonstrate that Ca²⁺ binding to the ADMIDAS stabilizes integrins in the low-affinity, closed conformation. Comparisons in affinity between four-domain and six-domain headpiece constructs suggest that flexible integrin leg domains contribute to conformational equilibria. High-resolution views of the hybrid domain interface with the plexin–semaphorin–integrin (PSI) domain in different orientations show a ball-and-socket joint with a hybrid domain Arg side chain that rocks in a PSI domain socket lined with carbonyl oxygens.

Integrins are adhesion receptors that transmit bidirectional signals across the plasma membrane. The α₅β₁ integrin and its primary extracellular matrix ligand fibronectin (Fn) are of great biological importance (1). Studies of this prototypic integrin–ligand pair led to the remarkable observation that much of the specific recognition by α₅β₁ of the ∼2,000-residue fibronectin molecule was encoded by a 3-residue sequence, Arg–Gly–Asp (RGD), in Fn3 domain 10 (2). Eight of 24 integrin αβ heterodimers are now known to recognize RGD motifs within their ligands. Higher-affinity and more α₅β₁-selective cyclic RGD peptide antagonists have been discovered (3) and mutationally characterized (4) but remain uncharacterized structurally.

Integrin αI and βI domains coordinate acidic residues in their ligands, such as the Asp side chain of RGD, through an Mg²⁺ ion held in a metal ion-dependent adhesion site (MIDAS) (5–8). βI domains also contain flanking Ca²⁺ ions coordinated by residues in the adjacent to MIDAS (ADMIDAS) and synergistic metal ion binding site (SyMBS). The ligand-binding headpiece of integrins has two states: a high-affinity, open conformation with the β-subunit hybrid domain swung out at its interface with the βI domain, and a low-affinity, closed conformation with the hybrid domain swung in (8). The altered conformation at the βI–hybrid domain interface is transmitted to an ∼3-Å rearrangement in the MIDAS- and ADMIDAS-coordinating βI domain β1-α1 loop, which alters affinity by ∼1,000-fold (9). The SyMBS and ADMIDAS have important roles in regulating ligand binding affinity (10, 11). The ADMIDAS is a negative regulatory site responsible for integrin inhibition by high concentration of Ca²⁺ and for activation by Mn²⁺, which competes with Ca²⁺ for binding to the ADMIDAS (10, 11). However, the mechanism by which metals at the ADMIDAS regulate integrin affinity remains mysterious.

A recent structural study (12) revisited the role of Ca²⁺ at the ADMIDAS. The α₅β₁ headpiece was characterized at 2.9-Å resolution bound to the SG/19 Fab that stabilizes the closed conformation of β₁ integrins by binding to the βI–hybrid domain interface. Soaking with RGD peptide induced movement of the βI domain β1-α1 loop and α1-helix, resulting in a conformational state intermediate between closed and open. A decrease in electron density for the ADMIDAS Ca²⁺ ion was also observed (12).

Here we report high-resolution crystal structures of the α₅β₁ headpiece, alone or with soaked-in linear or cyclic peptides. Our results include important advances in understanding the role of Ca²⁺ binding to the ADMIDAS in ligand binding by β₁ integrins, how integrin α₅β₁ binds cyclic peptides with high affinity, and the observation that Ca²⁺ at the ADMIDAS inhibits RGD-induced β1-α1 loop and α1-helix movement.

Results

Overall α₅β₁ Headpiece Structure.

We solved crystal structures at 1.78–2.5 Å (Table S1) of a four-domain fragment of the α₅β₁ headpiece containing the α₅-subunit β-propeller domain and the β₁-subunit (Fig. S1) plexin–semaphorin–integrin (PSI), hybrid, and βI domains in a closed conformation (Fig. 1A). The orientation between the β₁-subunit βI and hybrid domains is very similar to that in the five-domain α₅β₁ headpiece–SG/19 Fab complex (12) and in the closed α_IIbβ₃ conformation in the absence of ligand (13, 14), but is distinct from that in the open α_IIbβ₃ conformation found with crystallization in the presence of ligand (7) (Fig. 1B). The overall closed βI–hybrid domain orientation is supported by lattice contacts (Fig. S2), and is maintained in three RGD-soaked crystal structures.

Fig. 1.

Crystal structure of the α₅β₁ integrin headpiece. (A) Ribbon diagram of α₅β₁. The β-propeller, βI, hybrid, and PSI domains and RGD are in green, yellow orange, cyan, light blue, and gray, respectively. Ca²⁺ (silver) and Mg²⁺ (gold) are shown as spheres. Disulfide bonds (orange) and RGD are shown in stick representation. (B) Integrin headpiece comparisons. Four-domain α₅β₁, color-coded as in A, five-domain α₅β₁ (magenta) (12), and open α_IIbβ₃ (silver) (31) are superimposed and shown in Cα ribbon; for clarity, only β₃ of α_IIbβ₃ is shown. SG/19 Fab bound to five-domain α₅β₁ (12) is shown as a surface. (C) Binding of SG/19 to the βI domain α1′-helix. Structures are colored and shown as in B, with the Arg-154 and -155 side chains also shown and the view rotated ∼180° around a vertical axis relative to B.

SG/19 binds to the βI–hybrid domain interface and inhibits ligand binding allosterically by stabilizing this interface in its closed, low-affinity conformation (12, 15). SG/19 Fab buries 350 Ų of solvent-accessible surface on the βI domain (260 Ų of which is on the α1′-helix) and 790 Ų on the hybrid domain. The identical βI–hybrid domain orientation in the presence and absence of SG/19 Fab (Fig. 1B) agrees with the previous EM observation (15) that SG/19 does not enforce an artificial orientation between the βI and hybrid domains but binds to the same overall closed hybrid–βI domain orientation as seen in the absence of Fab.

The closed conformation at the ligand binding site (Fig. 2), the α1-helix, and the N-terminal end of the α1′-helix are identical in the presence and absence of SG1/19 (Fig. 1C). However, SG/19 induces small conformational changes in its epitope. The hybrid domain βW-βA loop is flexible in the absence of SG/19: Our four-domain α₅β₁ structures lack electron density for 5–8 residues at the tip of this loop. The loop becomes ordered when bound to SG/19 (Fig. 1B), and forms a substantial part of the epitope. SG/19 binds to the C-terminal end of the α1′-helix, a C-terminal continuation of the α1-helix (12). SG/19 slightly pushes α1′ inward to enable the α1′ Arg-154 and Arg-155 side chains to reorient and form part of the epitope (Fig. 1C).

Fig. 2.

βI domain metal binding sites. (A–H) The indicated structures are in identical orientations. Backbones and side chains of metal-coordinating residues are colored wheat with red oxygens and blue nitrogens. Waters are shown as small red spheres. Ca²⁺ (silver) and Mg²⁺ (gold) are shown as large spheres. Putative ADMIDAS Ca²⁺ ions not included in molecular models in G and H are shown as small spheres. Simulated-annealing omit map F_o − F_c electron density contoured at 2.5σ is shown as mesh around metal ions.

Metal Ion and RGD Peptide Binding.

α₅β₁ crystals, formed in the presence of 1 mM Mg²⁺ and Ca²⁺ at pH 7.2, have excellent simulated-annealing omit map electron density for all three βI domain metal ions (Fig. 2A), as also seen with the α₅β₁ complex with SG/19 crystallized at pH 6.5 (Fig. 2 E and F) (12).

Crystals soaked overnight with crystallization well solution containing 5 mM Mg²⁺, 1 mM Ca²⁺, and 10 mM GRGDSP peptide (corresponding to fibronectin residues 1523–1528) diffracted to 1.78-Å resolution (Table S1). RGD appeared in the ligand binding site with strong electron density. Complete occupancy (saturation binding) was confirmed by real space correlation coefficients (RSCCs) close to 1.0 (Table S2). The RGD Arg guanido group formed strong, bidentate, charged hydrogen bonds to α₅ Asp-227; in contrast, hydrogen bonding to α₅ Gln-221 was through two intermediate water molecules (Fig. 3A). The RGD Asp side chain coordinated the MIDAS Mg²⁺ ion using one oxygen and formed a weak, 3.4-Å hydrogen bond to the βI domain β1-α1 loop backbone through the other oxygen (Fig. 3A). RGD binding induced no movement in the β1-α1 loop or α1-helix, which were completely superimposable on the native (apo) structure (Fig. 4A), namely in closed “state 1” as defined by α_IIbβ₃ structures (9). Moreover, there was no loss of Ca²⁺ at the ADMIDAS; Ca²⁺ electron density after soaking with RGD peptide was as strong (Fig. 2B) as before (Fig. 2A) (Table S2).

Fig. 3.

RGD-binding pocket. (A–D) Vertical alignment in identical orientations with a dashed line for comparison. Residues of α₅ (light blue), β₁ (wheat), and ligand (RGD and two more C-terminal residues if built in the model) (orange) are shown in stick representation. Ca²⁺ (silver; ADMIDAS) and Mg²⁺ (gold; MIDAS) are shown as spheres. Waters are shown as small red spheres. Selected distances are shown in Å. (E) Binding of the cyclic peptide. The color scheme is as in A–D. All cyclic peptide residues visible in electron density are shown. 2F_o − F_c density at 1σ is shown as mesh for the cyclic peptide and α₅ Trp-157.

Fig. 4.

RGD-induced conformational rearrangements in the βI domain. (A–D) The indicated unliganded (wheat) and RGD-soaked (light blue) structures are shown in identical orientations. Arg and Asp residues and disulfide bonds are shown in stick representation. Metals at MIDAS and ADMIDAS are shown as spheres.

To test the effect of omitting Ca²⁺, we soaked crystals in crystallization solution containing 10 mM peptide, 5 mM Mg²⁺, and no Ca²⁺. The crystals diffracted to 2.5 Å and had excellent density for RGD (Fig. S3). In absence of added Ca²⁺, RGD induced substantial movement of the β1-α1 loop and α1-helix (Fig. 4B), enabling the non–MIDAS-binding Asp carboxyl oxygen to form a stronger, 3.0-Å hydrogen bond to the β1-α1 loop backbone (Fig. 3B). Accompanying the movement of the β1-α1 loop, the side chain of Ser-134 moved to directly coordinate the MIDAS Mg²⁺ ion (Fig. 2C), displacing a water at the MIDAS seen with RGD in the presence of Ca²⁺ (Fig. 2B). Slight differences were also seen at the RGD Arg guanido group, which now formed a direct hydrogen bond to α₅ Gln-221 (Fig. 3B). Moreover, density for Ca²⁺ at the ADMIDAS was diminished, whereas there was little change in Ca²⁺ density at the SyMBS or Mg²⁺ density at the MIDAS (Fig. 2C compared with Fig. 2 A and B) (Table S2). These differences in the presence and absence of Ca²⁺ directly demonstrate that Ca²⁺ binding to the ADMIDAS inhibits RGD-induced shape shifting away from the closed conformation.

The intermediate state and diminishment of ADMIDAS Ca²⁺ electron density we observed when soaking with RGD peptide in the absence of Ca²⁺ were essentially identical to that obtained by Nagae et al. (12) when both Mg²⁺ and Ca²⁺ were omitted from the RGD soaking solution (Fig. 2C compared with Fig. 2 G and H, Fig. 3B compared with Fig. 3D, and Fig. 4B compared with Fig. 4D).

Complex with a Cyclic Peptide.

The disulfide cyclized peptide ACRGDGWCG binds α₅β₁ strongly and specifically (3, 4). A crystal soaked with 1 mM cyclic RGD peptide, 1 mM Mg²⁺, and 1 mM Ca²⁺ in crystallization solution diffracted to 2.5 Å and showed strong density for the bound peptide (Fig. S3). Quite remarkably, binding of the cyclic peptide in the presence of Ca²⁺ induced shape shifting of the β1-α1 loop and α1 helix (Fig. 4C) similar to the linear peptide in the absence of Ca²⁺ (Fig. 4B). Furthermore, the cyclic peptide bound to α₅β₁ (Fig. 3C) similar to the linear peptide in the absence of Ca²⁺ (Fig. 3B). The cyclic peptide also induced movement of the Ser-134 side chain into direct coordination with the MIDAS Mg²⁺ (Fig. 2D). Importantly, the ADMIDAS Ca²⁺ ion moved with the β1-α1 loop backbone (Fig. 4C) and with its Ser-134 carbonyl oxygen, which forms one ADMIDAS coordination (Fig. 2D). Also moving with the ADMIDAS Ca²⁺ ion was the α1-helix (Fig. 4C) with its Asp-137 and Asp-138 side chains, which form two further ADMIDAS coordinations (Fig. 2D). Moreover, electron density for the moved ADMIDAS Ca²⁺ was good (Fig. 2D), although weaker than the SyMBS Ca²⁺ electron density (Table S2). Shape shifting with (i) the cyclic peptide and Ca²⁺, (ii) the linear peptide and Ca²⁺ depletion, and (iii) RGD in the SG/19 complex (12) was similar in extent to one another (Fig. 4 B–D), and approximated intermediate state 4 of α_IIbβ₃ (9).

The residues between the two disulfide-bonded cysteines in ACRGDGWCG were well-defined and show the basis for high-affinity binding (Fig. 3E). Cyclization decreases the conformational space accessible to a solvated peptide and thus lowers the entropic cost of becoming ordered when binding to a receptor. Moreover, the cyclic peptide conformation was stabilized by three backbone hydrogen bonds, including a 2.8-Å hydrogen bond with excellent geometry between the Arg-3 carboxyl oxygen and Trp-7 amide nitrogen. The cyclic peptide RGD moiety assumed a bound conformation very similar to that in linear RGD peptides (Fig. 3C compared with Fig. 3 A, B, and D). Importantly, the face of the cyclic peptide Trp side chain formed a T-shaped interaction with the edge of the α₅ Trp-157 side chain (Fig. 3E and Movie S1). T-shaped interactions are energetically favored by electrostatic interactions between the electron-rich edges and electron-poor centers of aromatic rings (16).

RGD Peptide Binding Affinity.

Taking advantage of the relatively high affinity of the cyclic peptide, we labeled it fluorescently and measured affinity using the increase in fluorescence anisotropy upon binding to α₅β₁ (17). The six-domain α₅β₁ integrin headpiece fragment, containing thigh and EGF1 domains, bound with a K_D of 17.6 nM to the fluorescent cyclic peptide in 1 mM Mn²⁺/0.1 mM Ca²⁺ (Fig. S4). Competitive binding showed a similar K_D of 22.6 nM for the unlabeled cyclic peptide (Fig. 5A). The linear GRGDSP peptide bound the six-domain headpiece with a markedly lower affinity, a K_D of 396 nM (Fig. 5A). Both cyclic and linear peptides exhibited twofold lower binding affinities to the four-domain integrin headpiece fragment lacking thigh and EGF1, with K_Ds of 43.7 nM and 862 nM, respectively (Fig. 5B). The affinity of cyclic RGD in 1 mM Mg²⁺/1 mM Ca²⁺ was too low for accurate measurement using our method; however, the affinity for the six-domain α₅β₁ headpiece in 5 mM Mg²⁺/1 mM Ca²⁺, the same conditions as used for GRGDSP soaking, was 1 μM (Fig. 5C). In contrast, the affinity in 5 mM Mg²⁺, in the absence of Ca²⁺, as also used in soaking, was 115 nM, and was 85 nM in 5 mM Mg²⁺/2 mM EGTA. Thus, the β1-α1 loop and α1-helix movement seen with Ca²⁺ depletion correlated with about a 10-fold increase in α₅β₁ affinity for ligand.

Fig. 5.

RGD binding affinity. (A and B) Binding of six-domain α₅β₁ (A) and four-domain α₅β₁ (B) headpieces (200 nM) to linear and cyclic RGD peptides measured by competition with 2 nM fluorescent cyclic peptide using fluorescence anisotropy in 1 mM Mn²⁺ and 0.1 mM Ca²⁺. (C) Binding of the six-domain α₅β₁ headpiece to 2 nM fluorescent cyclic RGD peptide in different divalent cations. Data show mean ± SD of triplicate samples.

The Flexible Hybrid–PSI Interface.

Among α₅β₁ headpiece crystal structures, the orientation between hybrid and PSI domains varies by up to 15° (Fig. 1B and Table S3). Our two highest-resolution structures reveal the basis for flexibility at this interface (Fig. 6). The side chain of hybrid domain residue Arg-104 can rock and change rotamer while maintaining six or seven hydrogen bonds between its three guanido nitrogens and a socket in the PSI domain formed by carbonyl oxygens from the backbones of residues 19, 22, 23, 25, and 63 and the side chain of Asn-62 (Fig. 6).

Fig. 6.

Rocking at the hybrid–PSI interface. The change in PSI orientation is shown after superimposition on the hybrid domain. The rocking Arg-104 side chain in the hybrid domain shown in stick representation binds to a socket in PSI lined with carbonyl groups. Dashed lines indicate hydrogen bonds. (A) Combined view. (B and C) Individual detailed views, with 2F_o − F_c density for Arg-104 contoured at 1σ shown in mesh. Unliganded and linear RGD peptide + Ca²⁺ α₅β₁ structures have carbons in light blue and wheat, respectively, and disulfide bonds in yellow.

Discussion

The resolution of the structures here surpasses previous limits of 2.35 Å for an open α_IIbβ₃ headpiece–RGD–Fab complex in Mn²⁺/Ca²⁺ (9) and 2.2 Å for a closed α_IIbβ₃ headpiece–drug–Fab complex (18). Our 1.85-Å unliganded and 1.78- to 2.5-Å RGD-bound α₅β₁ structures also represent improvements of 1.1–0.4 Å over previous 2.9-Å α₅β₁ headpiece crystals (12). Deletion of the thigh domain in our four-domain α₅β₁ headpiece compared with the previous five-domain headpiece (12) not only removed a flexible domain but also enabled formation of more compact crystals with 46% (vol/vol) solvent content (four-domain headpiece) compared with 61% solvent content (five-domain headpiece–Fab complex).

Headpiece structures provide insights into the flexibility of knee-proximal thigh, PSI, and EGF1 domains in the extended conformation of integrins. These domains show some flexibility in crystal structures of intact integrin ectodomains (19–22), in which the headpiece is bent at α-subunit thigh–calf-1 and β-subunit EGF1–EGF2 junctions and is in contact with the lower legs. Upper leg–lower leg interfaces and the closeness of the α-subunit and β-subunit knees help to stabilize the orientation of knee-proximal domains in the bent conformation. However, these constraints are removed upon integrin extension. PSI and EGF1 extend from the same end of the hybrid domain (Fig. 1 A and B). Interaction between PSI and EGF1 in β₂ integrins maintains similar hybrid–PSI and hybrid–EGF1 orientations in different α_xβ₂ ectodomain crystal forms and β₂ leg fragments (20, 21, 23, 24). In contrast, in α₄β₇ headpiece 3.1-Å structures, both PSI and EGF1 domains were missing from electron density (25).

Electron density for the PSI domain is generally poorer than for other integrin domains in both bent ectodomain and headpiece structures. Although this is true in our structures as well, the overall higher-resolution (1.78 and 1.85 Å) α₅β₁ headpiece structures show clear density at the hybrid–PSI interface and reveal the basis for flexibility. When the PSI domain moves relative to the hybrid domain, the side chain of hybrid domain residue Arg-104 rocks and changes rotamer to maintain an extensive network of hydrogen bonds to a socket in the PSI domain that is lined with backbone carbonyl oxygens. The presence of the hybrid domain Arg side chain in the PSI interface was previously described (7, 26), but not its role in PSI–hybrid reorientation. The Arg functions as a ball in a ball-and-socket joint between two tandem domains. We propose to call this motif a rocking Arg in a carbonyl socket. Perhaps the closest analogous interface is found in selectins between the lectin and EGF domains, where the α-amino group of the N-terminal Trp of the lectin domain pivots in an EGF domain carbonyl cage in selectin allostery (27).

Flexible domains or polypeptide segments distal from ligand binding interfaces may alter allosteric equilibria (28). We consistently found twofold higher ligand binding affinity of the six-domain than the four-domain α₅β₁ headpiece for cyclic and linear RGD peptides. These differences cannot be accounted for by structural differences in the ligand-binding pocket. α₅β₁ headpiece structures with (12) and without (here) the α-subunit thigh domain show no significant difference in the adjacent α-subunit β-propeller domain (0.29-Å rmsd over 402 Cα atoms). The thigh domain in integrins shows considerable flexibility at its interface with the β-propeller domain (20, 21, 25, 29). The hybrid, PSI, and EGF1 domains come much closer to the thigh domain in the closed than open headpiece conformation (Fig. 1B) and thus limit the range of orientations available to the thigh domain. Therefore, the presence of the thigh domain will entropically favor the open headpiece conformation. Similarly, the presence of the EGF1 domain in our six-domain but not four-domain headpiece will limit thigh domain flexibility in the closed conformation and favor the open headpiece conformation.

The cyclic peptide ACRGDGWCG was previously found to more effectively mediate and inhibit integrin α₅β₁-dependent cell attachment to substrates than linear RGD peptides (3, 4). We found a 20-fold higher affinity of cyclic than linear RGD peptide for both four-domain and six-domain α₅β₁ headpieces. This increase in affinity may be because cyclization together with backbone hydrogen bonds stabilizes a favorable conformation for binding to α₅β₁ as shown in our structure. Alteration of the cyclic peptide Trp to Ala or mutation of α₅ Trp-157 to Ala greatly diminished cell attachment (4), successfully predicting the interaction of the peptide Trp with the α₅ Trp-157 visualized here. We see a T-shaped edge-to-center interaction known to stabilize aromatic interactions between the two Trp residues (16). Despite the importance of α₅ Trp-157 in binding the cyclic peptide, it is not important in binding fibronectin, which has a Pro residue in the equivalent position (4).

Our structures provide insights into how Ca²⁺ at the ADMIDAS regulates integrin affinity. Refinement in PHENIX (Methods) with different metal ions in our 1.78- and 1.85-Å structures strongly supports the assumption that in the closed conformation the SyMBS binds Ca²⁺, the MIDAS binds Mg²⁺, and the ADMIDAS binds Ca²⁺. Importantly, when we compared the effect of soaking crystals with linear RGD peptide in the presence of Mg²⁺ and Ca²⁺ or in the presence of only Mg²⁺, we found that RGD only induced β1-α1 loop and α1-helix movement toward the open conformation when Ca²⁺ was depleted. Furthermore, density for Ca²⁺ was diminished at the ADMIDAS but little affected at the SyMBS or the four β-propeller Ca²⁺-binding hairpins (Table S2). These findings directly demonstrate that Ca²⁺ binding to the ADMIDAS stabilizes the closed conformation, and agree with our finding that Ca²⁺ depletion increases affinity for RGD peptide by 10-fold. We soaked crystals substantially longer (overnight) with RGD in the absence of Ca²⁺ than Nagae et al. (30 min) (12). However, the t_1/2 for diffusion of small ions such as Ca²⁺ into or out of ∼400-μm diameter protein crystals is only ∼20 s (30). In our RGD-soaked Ca²⁺-depleted structure, we have placed Ca²⁺ at the ADMIDAS in the model, whereas Nagae et al. did not. However, RSCCs suggest in both cases a Ca²⁺ occupancy of 0.5–0.7 (Table S2).

Why in Ca²⁺ did the cyclic RGD peptide induce β1-α1 loop and α1-helix movement toward the open conformation whereas the linear peptide did not? Occupancy by RGD is essentially complete in both cases (Table S2). Thus, in contrast to our use of different concentrations of RGD to achieve different occupancies and different extents of opening in α_IIbβ₃ crystals (9), differing occupancy cannot explain the differences here. Three effects may be in play. (i) Although the bound conformations of RGD appear similar, there may be a significant difference in binding enthalpy to drive the shift. (ii) RGD slides slightly in its groove from β₁ toward α₅ between closed state 1 and intermediate state 4, as described for α_IIbβ₃ (9). Sliding is greater for cyclic RGD than linear RGD in state 4, particularly at the RGD Gly and Asp, and may be enhanced by the peptide Trp interaction with α₅ Trp-157, consistent with greater movement in this direction in the cyclic peptide of the backbone following the peptide Asp (Fig. 3C compared with Fig. 3B). (iii) In the linear peptide, the C-terminal Ser and Pro are disordered in the presence of Ca²⁺, and have a spatial probability distribution that includes space that the side chain of Tyr-133 in the β1-α1 loop must occupy in intermediate state 4. This entropic barrier to β1-α1 loop shape shifting by the peptide C terminus is removed by cyclization. Notably, the peptide C terminus becomes ordered by the close approach of the β₁ Tyr-133 side chain in the Ca²⁺-deficient complex (Fig. 3B compared with Fig. 3A), despite lower resolution compared with the Ca²⁺-replete linear peptide complex.

Good density for the ADMIDAS Ca²⁺ in the intermediate state with the cyclic peptide shows that movement of ADMIDAS-coordinating residues does not abolish their interaction with Ca²⁺; indeed, in our Ca²⁺-depleted and -replete intermediate structures, as well as in models with Ca²⁺ added to the Nagae et al. structures, coordination to the Ser-134 backbone and Asp-137 and Asp-138 side chains is present in all cases (Fig. 2). However, we agree with the suggestion by Nagae et al. that breakage of the ADMIDAS coordination to the Ala-342 backbone in the β6-α7 loop, the first consequence of ADMIDAS movement, is associated with weaker ADMIDAS Ca²⁺ density. This is apparent from RSCCs with our cyclic RGD intermediate state, which shows an ADMIDAS Ca²⁺ occupancy of ∼0.6 (Table S2), and from the weaker density for the ADMIDAS metal ion in intermediate states 3–5 of α_IIbβ₃ (figure 2 in ref. 9).

According to the law of mass action, Ca²⁺ at the ADMIDAS can only favor the closed conformation over the open conformation if it is bound with higher affinity in the closed than open conformation. We have obtained the first structural correlate of this prediction, to our knowledge, by demonstrating that Ca²⁺ removal destabilizes the closed conformation and enables RGD to induce integrin movement toward the open state. Higher resolution than obtained here for the closed state in Mg²⁺/Ca²⁺ (1.78 and 1.85 Å) or obtained for the open state in Mg²⁺/Ca²⁺ (2.4 Å) (31) or in Mn²⁺/Ca²⁺ (2.35 Å) (9) will be required for detailed analysis of metal ion coordination in the two states and to determine whether Mg²⁺ can replace Ca²⁺ at the ADMIDAS, as Mn²⁺ does (6, 9). At our current resolution it appears that the ADMIDAS Ca²⁺ coordination in the closed state is closest to pentagonal bipyramidal, with bidentate coordination in the closed state by the carboxyl oxygens of Asp-137 in the plane of the pentagon (Fig. 2 A and B). Whereas we see seven ADMIDAS Ca²⁺ coordinations in our 1.78- and 1.85-Å closed states (Fig. 2 A and B), we see only four or five ADMIDAS Ca²⁺ coordinations in the 2.5-Å intermediate-state structures (Fig. 2 C and D), correlating with lower Ca²⁺ occupancy. Furthermore, Asp-137 coordination is monodentate (Fig. 2 C and D). In open-conformation structures of α_IIbβ₃, coordination of Ca²⁺ or Mn²⁺ at the ADMIDAS is more octahedral, with monodentate coordination by the β₃ Asp corresponding to β₁ Asp-137 (9, 31). The open ADMIDAS coordination also differs from closed because ADMIDAS coordination by a β6-α7 loop backbone carbonyl is replaced by an Asp side chain that coordinates through a water to the MIDAS Mg²⁺ ion (7, 9). The preference of Ca²⁺ for pentagonal bipyramidal over octahedral coordination (32), together with changes in metal ligands between closed and open conformations, may explain why Ca²⁺ removal increases affinity for RGD and enables RGD to induce β1-α1 loop and α1-helix shape shifting. Conversely, Mg²⁺ and Mn²⁺ favor octahedral over pentagonal bipyramidal coordination (32), and their substitution for Ca²⁺ at the ADMIDAS may explain why Mn²⁺ and Mg²⁺/EGTA (EGTA selectively chelates Ca²⁺) are general activating agents for integrins (33, 34).

Our results are in excellent agreement with previous findings that Ca²⁺ exerts an inhibitory effect on β₁, β₂, and β₇ integrins by binding to the ADMIDAS (10, 11, 34, 35). However, the ideas expressed in the previous paragraph on how Mn²⁺ activates integrins are controversial, particularly for α₅β₁, for which a large body of work has suggested that Mn²⁺ activates by binding to the MIDAS (10, 36). It is possible that Mn²⁺ activates α₅β₁ by replacing both Mg²⁺ at the MIDAS and Ca²⁺ at the ADMIDAS; however, this fails to explain why, for most integrins, removal of Ca²⁺ by EGTA in the presence of excess Mg²⁺ activates as well as Mn²⁺. An early α_vβ₃ crystal structure showed an intermediate RGD-bound state similar to that seen here, with separation between the MIDAS and ADMIDAS metal ions, and reported that in contrast to integrin αI domains, βI domains did not convert between open and closed states (6). α₅β₁ mutational studies were therefore interpreted with the assumption that the MIDAS and ADMIDAS were independent of one another. However, we now know that integrin βI domains undergo opening to a high-affinity state, and that the ADMIDAS metal ion moves closer to the MIDAS metal ion. In the open state, the side chain of β₃ Asp-251, equivalent to β₁ Asp-259 (Fig. 2), forms a direct coordination to the ADMIDAS metal ion and an indirect coordination through a water to the MIDAS metal ion (7). Mutation of β₁ MIDAS residue Asp-130 abolishes the ability of Mn²⁺ to activate, but not the ability of Ca²⁺ to inhibit, binding of 12G10 antibody to an activation epitope on the βI domain α1′-helix involving residues Arg-154 and Arg-155 (36). In light of current structural knowledge, this suggests either that Mn²⁺ activates by binding to the MIDAS, or that Mn²⁺ activates at the ADMIDAS in a MIDAS-dependent manner, for example, that ADMIDAS movement into close proximity to the MIDAS with 12G10 epitope exposure requires a metal at the MIDAS. In further support for Mn²⁺ activation at the MIDAS, mutation of ADMIDAS residues Asp-138 and Asp-139 to Ala was found to still permit activation by Mn²⁺ and inhibition by Ca²⁺ (10). However, the D137A mutation lowered ligand binding affinity more than D138A, raising the possibility that D138A might not completely inhibit metal binding at the ADMIDAS; furthermore, ligand binding experiments showing stimulation by Mn²⁺ used the partially active D138A mutant and the stimulatory antibody 12G10, which might have acted like crutches to enable a crippled ADMIDAS to bind to Mn²⁺ when in its active, MIDAS-proximal conformation. Integrins are perversely complicated, and a definitive understanding of how Mn²⁺ and Mg²⁺/EGTA activate will require further work.

Methods

A six-domain α₅β₁ headpiece containing α₅ residues F1–L609 and β₁ residues extending to the end of the EGF1 domain did not yield crystals, and its thigh domain resisted removal by proteolysis. Removal of thigh from α_IIbβ₃ by chymotrypsin is important in its crystallization (7). Therefore, we used a chimeric α-subunit constructed with the signal peptide and mature β-propeller residues F1–P450 of α₅ and thigh residues V453–D601 of α_IIb. The β₁-subunit was expressed with EGF1 (mature residues Q1–E481) or without EGF1 (Q1–E445). α₅ or α₅/α_IIb chimeric cDNAs were inserted into the XbaI and AgeI sites of pcDNA3.1/Hygro(-) vector, which contained a C-terminal tobacco etch virus (TEV) protease site, an acidic coiled-coil with a cysteine for disulfide-bond formation, and a Strep II tag (37). β₁ cDNAs were inserted into the BamHI and XbaI sites of pEF1-puro with a C-terminal TEV site, a basic coiled-coil with a cysteine, and a hexahistidine tag (37). HEK 293S N-acetylglucosaminyl transferase I–negative (GnTI^−/−) cells were cotransfected with the α₅ and EGF1-containing β₁ plasmids or the α₅/α_IIb chimera and EGF1-lacking β₁ plasmids and selected with 100 μg/mL hygromycin B and 3 μg/mL puromycin. Clones with high expression were cultured in EX-CELL 293 serum-free medium (Sigma-Aldrich) using shaking square bottles (38). Culture supernatant supplemented with 50 mM Tris⋅HCl (pH 8.0) and 300 mM NaCl was mixed with 40 mL of Ni-NTA Sepharose per liter. A column was washed with 20 mM Tris⋅HCl, 0.5 M NaCl, 10 mM imidazole (pH 8) and eluted with the same buffer with 300 mM imidazole. Eluted fractions were concentrated to ∼1 mg/mL in 20 mM Tris⋅HCl (pH 8.0), 150 mM NaCl (TBS) with 1 mM Mn²⁺, 0.1 mM Ca²⁺, 0.1 mM GRGDSP peptide and treated with 0.01 mg/mL α-chymotrypsin (crystallized and lyophilized; Sigma; C-4129) at 37 °C for 16 h to remove the thigh domain, C-terminal coiled-coil, and tags (Fig. S1). Protein was further purified and separated from GRGDSP peptide by Superdex 200 size-exclusion chromatography in TBS with 1 mM Mg²⁺ and 1 mM Ca²⁺ to produce the four-domain (β-propeller, PSI, hybrid, and βI) construct. A six-domain α₅β₁ headpiece construct (β-propeller, thigh, PSI, hybrid, βI, and EGF1) was expressed and purified identically, except TEV protease was used to remove the C-terminal coiled-coil and tags.

The four-domain α₅β₁ integrin headpiece was concentrated to 5 mg/mL in TBS, 1 mM Mg²⁺, and 1 mM Ca²⁺ and initially crystalized at 20 °C in 0.1 M Hepes (pH 6.5), 20% PEG 6000 using hanging-drop vapor diffusion. Diffraction-quality single crystals were obtained in 0.1 M Hepes (pH 7.2), 16% PEG 6000 with microseeding. For soaking overnight, single crystals were transferred to well solution supplemented with the indicated concentrations of peptides and metal ions (Table S1). Finally, crystals were transferred to well solution supplemented with 15% ethylene glycol with about 30 s at each of three 5% steps and frozen in liquid nitrogen. Diffraction data were processed with XDS (39). Structures were solved by molecular replacement. The apo structure was solved with the α₅β₁ structure [Protein Data Bank (PDB) ID code 3VI3] and the refined apo structure was used as the search model for RGD complexes. Models were built with Coot (40) and refined with PHENIX (41). Simulated-annealing omit and composite omit maps and RSCCs were calculated with PHENIX.

Cyclic and linear RGD peptides (>95% pure) were synthesized by GenScript. The cyclic RGD peptide was made fluorescent by synthesizing it with fluorescein isothiocyanate and 6-aminohexanoic acid linked to its N terminus.

All saturation and competition fluorescence anisotropy was performed in 20 mM Tris buffer (pH 7.4), 150 mM NaCl supplemented with the indicated metal ions and 2 nM fluorescent cyclic RGD. Binding affinities were calculated as described (17). In saturation binding assays, anisotropy was measured with the six-domain or four-domain α₅β₁ (starting at 500 nM) serially diluted in twofold decrements. Competition assays contained 200 nM α₅β₁, 2 nM fluorescent probe, and twofold serially diluted competing peptide.