α_Vβ₃ Integrin Crystal Structures and their Functional Implications

Abstract

Many questions remain about the significance of structural features of integrin α_Vβ₃ for its mechanism of activation. We have determined and re-refined, respectively, crystal structures of the α_Vβ₃ ectodomain linked to C-terminal coiled-coils (α_Vβ₃-AB) or four transmembrane (TM) residues in each subunit (α_Vβ₃-1TM). The α_V and β₃ subunits with four and eight extracellular domains, respectively, are bent at knees between the integrin headpiece and lower legs, and the headpiece has the closed, low-affinity conformation. The structures differ in occupancy of three metal binding sites in the βI domain. Occupancy appears related to the pH of crystallization, rather than to physiologic regulation of ligand binding at the central, metal-ion dependent adhesion site (MIDAS). No electron density was observed for TM residues and much of the α_V linker. α_Vβ₃-AB and α_Vβ₃-1TM demonstrate flexibility in the linker between their extracellular and TM domains, rather than previously proposed rigid linkage. A previously postulated interface between the α_V and β₃ subunits at their knees was also not supported, because it lacks high quality density, required rebuilding in α_Vβ₃-1TM, and differed markedly between α_Vβ₃-1TM and α_Vβ₃-AB. Together with variation in domain-domain orientation within their bent ectodomains between α_Vβ₃-AB and α_Vβ₃-1TM, the structures are compatible with the requirement of large structural changes, such as extension at the knees and headpiece opening, in conveying activation signals between the extracellular ligand-binding site and the cytoplasm.

Introduction

Integrins have noncovalently associated α and β subunits, each with an ectodomain (~ 1,000 residues in α and ~ 700 in β), a single transmembrane domain, and a typically short cytoplasmic domain. Integrin ectodomains have four domains in α and eight in β, of which one in each associate tightly to form a ligand-binding head (Fig. 1). The remaining domains connect the heads through upper legs, knees, and lower legs, to the plasma membrane. In the bent conformation, the knees are bent, and the headpiece (head and upper legs) folds over the lower legs, to form a 2,000 Ų interface (Fig. 1A). In two extended conformations, the legs are extended at the knees, and the headpiece may either be closed, as in the bent conformation, or open (Fig. 1B and C). Headpiece opening involves a large conformational change at the hybrid domain interface where the βI domain is inserted, i.e. attached at both its N- and C-termini. Headpiece opening is conveyed by α-helix connecting rod-like movements across the βI domain to its interface with the α-subunit where the ligand is bound, and increases affinity for ligand many orders of magnitude (Fig. 1C). Most integrins interact with the actin cytoskeleton, and transmit signals in both directions across the membrane to regulate binding to extracellular ligands, and traction through the cytoskeleton for cell migration and signaling (1).

Figure 1.

The three major integrin conformational states. A. Bent (with closed headpiece). B. Extended with closed headpiece. C. Extended with open headpiece. The dashed lower β-legs show that because of β-leg flexibility in the extended conformation, the closed and open headpieces are each compatible with TM domain close association and separation. However, a lateral force exerted by the cytoskeleton on the β-subunit can enforce TM domain separation and the open headpiece conformation (1).

The key question in the integrin field is how signals are transmitted from integrin cytoplasmic and transmembrane domains to the ligand-binding site in the ectodomain to regulate affinity for ligand. We and others have presented extensive evidence for two key types of conformational changes: 1) integrin extension to give the extended-closed state (Fig. 1B) and 2) integrin headpiece opening to give the extended-open state (Fig. 1C) (1). The extended-closed conformation has low affinity for ligand, and only the extended-open conformation has high affinity (2-9). Thus integrin activation by inside-out signals increases the proportion of integrins with the extended-open conformation. These conclusions are supported by a large number of electron microscopy (EM), small angle X-ray scattering (SAXS), and crystallographic studies as well as affinity measurements on cell surfaces. The studies include work on diverse integrins including α_Vβ₃, α_IIbβ₃, α₅β₁, α_Vβ₆, α_Lβ₂, and α_Xβ₂ (1). To communicate allostery over unusually large distances through the flexible lower leg of the β-subunit in the extended conformation, it has been emphasized that large separations between the integrin transmembrane and cytoplasmic domains would be required, and evidence for such separation has been obtained by transmembrane domain cross-linking and fluorescence resonance energy transfer (FRET) experiments (10-12).

On the other hand, another model proposes that integrin activation requires neither extension nor headpiece opening (13, 14). The body of evidence supporting this alternative model is smaller, and is based on studies of integrin α_Vβ₃ (13, 14). Crucial in the latest support for this hypothesis was a crystal structure of integrin α_Vβ₃ (14). Excellent density was reported for the ectodomain, as well as the first three residues of both the α_V and β₃ TM residues. It was unexpected that the α_V and β₃ TM residues were far from one another in the crystal structure (25 Å), because structures for the closely related α_IIb and β₃ TM domains show the same residues in close contact (15-17). Nonetheless, a model was made by bending the ectodomain-TM linkers in both α_V and β₃ to bring the TM domains close, and it was suggested that some rigidity in the linker region is essential to maintain the integrin in its inactive conformation (14). Linker rigidity sets α_Vβ₃ apart from integrins α_IIbβ₃ and α_Xβ₂. Ectodomain-TM domain linkers were disordered in α_IIbβ₃ and α_Xβ₂ crystal structures, suggesting flexibility (18, 19). Furthermore, disulfide cross-linking of the α_IIb and β₃ linkers on intact α_IIbβ₃ on cell surfaces demonstrated lack of a specific interface (15).

Several other features of α_Vβ₃ crystal structures appear unusual. The βI MIDAS locates in between two other divalent metal-binding sites, the adjacent to MIDAS (ADMIDAS) and synergistic metal-binding site (SyMBS). α_Vβ₃ was found to lack metal ions at the MIDAS and SyMBS, but to contain them when a ligand was soaked in, and metal ions were suggested to regulate ligand binding (14, 20). In contrast, α_IIbβ₃ with an identical β-subunit, binds metal ions at all three sites in absence of ligand (18, 21). α_Vβ₃ has been reported to have clear density at its β-knee (14), in contrast to α_IIbβ₃ (18). A specific, stabilizing interface between the α_V and β₃ knees was reported that is unusual for containing many hydrogen bonds between highly solvent-accessible carbonyl and carboxyl groups (14). Such carboxyls are considered to be deprotonated at physiologic pH, and interactions between oxygen lone pair electrons are highly unfavorable. A specificity-determining loop (SDL) in the βI domain of the β₃ subunit near the ligand-binding interface with the α-subunit was built markedly differently in a 2004 α_IIbβ₃ crystal structure (4) than in the 2001 α_Vβ₃ crystal structure (22). The conformation in α_Vβ₃ was retained in the 2009 crystal structure (14), perhaps implying an influence of the α-subunit on the conformation of the nearby SDL in β₃.

To obtain an independent view of integrin α_Vβ₃, we have determined a structure of its ectodomain fused to coiled-coil peptides to mimic associating TM α-helices (α_Vβ₃-AB). Furthermore, to better address special issues raised by the α_Vβ₃-1TM structure, we carefully examined this structure and its underlying electron density. After doing so, we decided that it would benefit from re-refinement. Therefore, we report here two different α_Vβ₃ crystal structures. Using these structures and their objective electron density, we report several novel structural features, and come to novel conclusions on the bent-closed structure of α_Vβ₃ and its implications for activation.

Materials and Methods

Clasped α_Vβ₃-AB ectodomain structure

The α_V ectodomain sequence ending at M960 was fused to sequences in the order GSGGEN, AQCEKELQALEKENAQLEWELQALEKELAQ, corresponding to a 6-residue vector-derived sequence and the ACID-p1 sequence of peptide Velcro (23) containing a Cys at the “d” position of the heptad repeat (24), respectively. The β₃ ectodomain sequence ending at P691 was fused to sequences in the order ESM, LENLYFQ, GGKN, AQCKKKLQALKKKNAQLKWKLQALKKKLAQ, TG, and HHHHHH, corresponding to a three-residue linker, a tobacco etch virus (TEV) protease site, a four-residue linker, the BASE-p1 sequence of peptide Velcro (23) containing a Cys at the “d” position of the heptad repeat (24), a two-residue linker, and a His6 tag, respectively (2). Protein was expressed in CHO Lec 3.2.8.1 cells (2) and purified from culture supernatant using Ni-NTA agarose (QIAGEN) followed by gel filtration (Superdex 200 HR) chromatography in 0.01 M Tris HCl pH 7.5, 0.14 M NaCl (TBS), and 1 mM Ca²⁺, and was concentrated to about 5.5 mg/ml.

Crystals from hanging-drop vapor diffusion at room temperature were in 2 M ammonium sulfate, 80 mM sodium cacodylate pH 5.8, 2 mM CaCl₂ and 6 mM MgCl₂. Coverslips with their hanging drops were transferred to wells containing saturated ammonium sulfate for dehydration over 16 h. Crystals were cryoprotected in 95% saturated ammonium sulfate and 5% glycerol before plunge freezing in liquid nitrogen.

Diffraction data from GM/CA-CAT beamline 23-ID of Advanced Photon Source (APS) at Argonne National Laboratory were processed using XDS (25). The structure was solved using molecular replacement with PHASER with both 3IJE and 4G1M α_Vβ₃-1TM structures (described below). The rotation function and translation function Z scores, respectively, were 15.0 and 26.5 for the 3IJE model, and 16.2 and 38.1 for the 4G1M model. Initial structure refinement with PHENIX (26) used rigid bodies for individual domains, individual sites, TLS, and individual B factor refinement. Using the same parameters, we obtained R_work and R_free of 27.4% and 33.7% with 3IJE and 24.4% and 31.4% with 4G1M models, respectively. Like the better molecular replacement scores with the 4G1M model, these refinement results showed that the 4G1M α_Vβ₃-1TM model fit better the α_Vβ₃-AB dataset than the 3IJE α_Vβ₃-1TM model. The initial α_Vβ₃-AB model obtained using replacement with the 4G1M α_Vβ₃-1TM model was therefore used as the starting model for subsequent refinement.

After several rounds of manual model rebuilding with Coot (27) and refinement with PHENIX, simulated annealing was used in refinement instead of rigid body refinement. Simulated-annealing composite omit maps were also calculated and used to guide manual rebuilding to decrease model bias. During iterative rebuilding and refinement, an 8% twin fraction was observed in the dataset with the twin law H, K, -L. After applying the twin law, both R_work and R_free dropped around 1.5%. The ACID and BASE α-helical coiled-coil (leucine zipper) peptides, but not the linkers to the C-termini of the α_V- and β₃-subunits, were present in electron density and were added to the model.

Refinement was first completed at 3.0 Å resolution. Using a recent cross-correlation method (28), we found that useful data extended to 2.85 Å. Density improved, and refinement was extended to this resolution.

Refinement of 3IJE

We calculated simulated-annealing (SA) composite omit maps using the α_Vβ₃-1TM (PDB code 3IJE) coordinates and structure factor amplitudes with both the programs CNS (29) and PHENIX (26). Neither program's SA omit maps showed 1σ electron density for a substantial portion of the β₃ I-EGF1 domain, portions of the I-EGF2 and PSI, much of the α-subunit linker, or the TM segments of the α- and β-subunits.

Based on these results, we started with a model lacking the PSI, I-EGF1, and I-EGF2 domains, and systematically rebuilt the α_Vβ₃ structure in 16 iterative steps of manual rebuilding with Coot (27), attention to outliers and clashes identified with MolProbity (30), and TLS and maximum-likelihood refinement with REFMAC (31). PyMol scripts were used to superimpose individual domains from high resolution α_IIbβ₃ structures to provide hints for rebuilding. At each step, we computed PHENIX simulated annealing composite omit maps (26). The PSI and most of the I-EGF1 and I-EGF2 domains appeared in the 2Fo-Fc map during refinement and were added back. R_free dropped from 28.5% in 3IJE to 26.9%. Another SA omit map calculated with CNS showed most of I-EGF1.

Many further cycles of rebuilding in Coot utilized PHENIX refinement with TLS, maximum likelihood, and simulated annealing (26). R_free dropped to 24.2%. Subsequent refinement cycles without simulated annealing resulted in a final R_free of 23.4%. In the last cycle, EGF-1 was carefully examined. Only short segments were missing in the 1 σ 2Fo-Fc map; however, the overall quality of the map remained lower than for other domains, suggesting flexibility. A low-resolution map computed from data between ~ 65 and 3.7 Å was used to check this less-ordered domain. This map clearly showed the N-acetyl glucosamine residue attached to β₃ Asn-452, which was added back. The C-termini of α_V and β₃ were each extended about two residues past the last residue in the 1 σ map, to the last peptide moiety in ~ 0.5 σ density, beyond which there could be no confidence in the path of the polypeptide chain.

α_Vβ₃ furin site mutation and function

Full-length human α_V and β₃ plasmids were subjected to site-directed mutagenesis and co-transfected into HEK293T cells as previously described (2). Cell surface expression measured with non-functional Cy3-conjugated β₃ mAb AP3 and binding of Alexa-488 conjugated fibrinogen and fibronectin were determined by fluorescent flow cytometry (8). Briefly, transfected cells were detached, re-suspended in 20 mM HEPES-buffered saline, pH 7.4 (HBS), supplemented with 5.5 mM glucose and 1% bovine serum albumin and incubated in room temperature for 30 min with human fibrinogen or fibronectin in the presence of either 5 mM EDTA, 5 mM Ca²⁺, or 1 mM Mn²⁺. Cells were then stained with Cy3-conjugated AP3 on ice for 30 minutes.

Transiently transfected cells were metabolically labeled with [³⁵S] cysteine/methionine and subjected to immunoprecipitation as described (11). Lysates in 20 mM Tris-buffered saline, pH 7.4 (TBS), supplemented with 1 mM Ca²⁺, 1% Triton X-100, 0.1% Nonidet P-40 and 1:100 v/v protease inhibitor cocktail (Sigma-Aldrich, St. Louis, MO) were immunoprecipitated with 1 μg of α_Vβ₃ mAb LM609 (EMD Millipore Corporation, Billerica, MA) and protein G-Sepharose at 4 °C for 1 h.

Results

α_Vβ₃-AB ectodomain structure

Previously described α_Vβ₃-ΔTM (22) and α_Vβ₃-1TM proteins (14) were each expressed in insect cells, differ in length from one another only by four TM residues in each subunit, and crystallize with 10 to 12% PEG as precipitant in nearly identical lattices. The spacegroup is P3₂21, and the unit cell dimensions are a = b = 130.0 Å (α_Vβ₃-ΔTM) or 130.26 Å (α_Vβ₃-1TM) and c = 307.3 Å (α_Vβ₃-ΔTM ) or 305.98 Å (α_Vβ₃-1TM). Such small differences in lattice dimensions are typical for different crystals of the same protein.

To obtain an independent view of α_Vβ₃ structure, we expressed in mammalian CHO-Lec cells the complete α_Vβ₃ ectodomain, connected through 6-residue (α_V) and 14-residue (β₃) linkers to the ACID and BASE leucine zipper Velcro peptides (2, 23) (α_Vβ₃-AB). Crystals formed with a different class of precipitant, 2 M ammonium sulfate. The spacegroup was still P3₂21, and the a and b dimensions of 128.46 Å were similar to those of 130.26 Å for α_Vβ₃-1TM; however, at 352.49 Å the c dimension was markedly larger than 305.98 Å for the previous crystals (Table 1). The structure includes the ACID-BASE leucine zipper; however, it has little if any effect on α_Vβ₃ because the long linkers between the α_V and β₃ subunits and the ACID and BASE peptides are disordered. The 2.85 Å α_Vβ₃-AB structure (Fig. 2A) is well refined, with R_free of 26.2%, excellent Ramachandran and geometry statistics, and high MolProbity (30) percentiles (Table 1).

Table 1.

Statistics of X-ray diffraction and structure refinement

Protein	αVβ3-AB	αVβ3-1TM
Data collection statistics
Space group	P3221	P3221
α, β, γ,°	90, 90, 120	90, 90, 120
Unit cell (a = b, c), Å	128.46, 352.49	130.26, 305.98
Resolution range (Å)	50.0-2.85(2.85-2.92)	65-2.9
Completeness (%)	99.9 (100.0)	99.2 (98.5)
Number of unique reflections	79,656 (5,806)	66,702 (6,515)
Redundancy	7.4 (6.6)	2.9 (2.2)
Rsym (%)	17.5 (457.6)	8.8 (100)
I/σ	10.2 (0.6)	10.5 (100)‡
CC½ (%)e	99.7(13.6)	NA
Wavelength (Å)	1.034	1.0332
Refinement statistics	4G1E	3IJE	4G1M
Resolution range (Å)	50.0-2.85	20-2.9	65-2.9
Rwork (%)	24.1	24.4	18.0
Rfree (%)	26.1	28.7	23.4
Model statistics
Bond RMSD (Å)	0.005	0.006	0.005
Angle RMSD (°)	0.6	1.086	0.735
Ramachandran plotd (Favored/allowed/outlier)	95.8/4.0/0.2	88.4/8.5/3.1	93.2/6.7/0.1
MolProbity percentiled (Clashscore/Geometry)	97/100	91/66	97/99
Residue range
α V	1-619, 621-838, 841-848, 867-959	1-838, 868-967	1-837, 868-959
β 3	1-33, 35-478, 483-691	1-695	1-692
Number of metals	6	6	7
Number of carbohydrates	48	34	37
Number of cis-Proline	7	0	7
Number of waters	104	0	108

^‡The number in parentheses cannot be correct, but cannot be recalculated by us, and is shown as originally reported.

^aThe numbers in parentheses refer to the highest resolution shell.

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

^cR_factor = 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/σ cutoff was applied.

^dCalculated with the described server (30).

^ePearson's correlation coefficient between average intensities of random half-datasets for unique reflection (28).

Figure 2.

Overall structures of two α_Vβ₃ crystal forms and comparison to α_IIbβ₃. A-C. Cartoons of α_Vβ₃-AB (A), α_Vβ₃-1TM (4G1M structure, B), and α_IIbβ₃ (C). Ca²⁺ and Mg²⁺ ions are shown as silver spheres. Disulfides are shown as gold sticks, and glycans are displayed with gray carbons.

α_Vβ₃-1TM ectodomain structure

The 2.9 Å α_Vβ₃-1TM structure was solved by molecular replacement (14) and is therefore subject to model bias. The standard method for obtaining a less model-biased map is to compute a simulated-annealing (SA) composite omit map (32). Independent SA composite omit maps were calculated with CNS and PHENIX using the deposited α_Vβ₃-1TM (PDB code 3IJE) data. The maps showed absent or poor electron density for much of the PSI, I-EGF1, and I-EGF2 domains, the linker between the ectodomain and TM domains, and TM domain residues. Furthermore, difference map peaks showed a substantial number of places where the 3IJE model and the electron density did not correlate (Fig. 3A, red and green mesh).

Figure 3.

Difference maps. Difference maps of 3IJE (A) and 4G1M (B) α_Vβ₃-1TM structures. Fo-Fc difference density is shown in green (+3 σ) and red (-3 σ). The head and upper legs of α_V and β₃ are shown in light blue and wheat, respectively, and the lower α_V and β₃ legs are shown in yellow and black, respectively. C-termini are labeled for α_V and β₃.

Therefore, we rebuilt the α_Vβ₃-1TM molecular model using many iterative steps of rebuilding with Coot and refinement that included use of REFMAC, PHENIX, and simulated annealing. We initially omitted regions with poor density. As other domains improved, 1σ density for PSI, I-EGF1, and I-EGF2 reappeared, and these domains were added back. We re-calculated SA composite omit maps after each refinement step, and used MolProbity to find clashes, bad rotamers, and Ramachandran outliers after each refinement. Density for C-terminal portions of α_V and β₃ did not reappear and they are thus absent in the final model.

The new and previous α_Vβ₃-1TM molecular models are referred to here by their PDB accession codes, 4G1M and 3IJE, respectively. If the PDB code is not specified, α_Vβ₃-1TM refers to the 4G1M structure. The 4G1M structure has many fewer regions of disagreement with electron density (Fig. 3B) than 3IJE (Fig. 3A). Furthermore, 4G1M is refined against data from 2.9 - 65 Å, as opposed to 2.9 - 20 Å for 3IJE (Table 1). The greater amount of low-resolution data used in 4G1M allows better definition of less ordered regions, i.e. the PSI, I-EGF1 and I-EGF2 domains (Fig. 2B). Differences in the β₃ PSI, I-EGF1, and I-EGF2 domains (Cα RMSD of 1.2 - 1.4 Å, Table 2) occur throughout their length, whereas differences in the larger integrin domains result primarily from rebuilding of loops. The rebuilding process resulted in a drop of R_free from 28.5 to 23.4%, a decrease in Ramachandran outliers from 3.1% to 0.1%, and a decrease in poor rotamers from 7.9% to 0.6%, as reported by MolProbity (30) (Table 1).

Table 2.

RMSDs between 3IJE and 4G1M α_Vβ₃-1TM structures.^a

Domain	Residues	3IJE/4G1M non-H atoms (Å)	3IJE/4G1M Cα atoms (Å)
PSI	1-56+434-435	1.9	1.4
Hybrid	57-108+353-433	1.4	0.5
βI	109-352	1.3	0.7
I-EGF1	436-472	1.9	1.3
I-EGF2	473-522	2.9	1.4
I-EGF3	523-559	1.2	0.6
I-EGF4	560-600	1.5	0.8
Ankle	601-605	0.4	0.2
β-tail	606-690	1.3	0.7
linker	691-692	1.5	2.1
β-propeller	1-438	0.8	0.3
Thigh	439-594	1.3	0.6
Calf-1	595-737	1.8	0.9
Calf-2	738-954	1.5	1.1
Linker	955-959	2.0	1.6
Overall		1.4	0.8

^aRMSD was calculated using rms_cur command of PyMol.

Overall structures

Both α_Vβ₃ structures reveal an overall bent conformation (Fig. 2A and B). The α_V β-propeller and β₃ βI domains interact over a large interface to form the integrin head and its ligand binding site. Acute bends define the knees, at the junction between the upper and lower legs. The α_V leg is bent between the thigh and calf-1 domains, at the disulfide-bonded, Ca²⁺ binding genu loop, or α-knee. The bend in β occurs at the β-genu between I-EGF domains 1 and 2. Thus the thigh domain in α_V and the PSI, hybrid, and I-EGF1 domains in β₃ form the upper legs. The calf-1 and calf-2 domains in α_V and the I-EGF2, I-EGF3, I-EGF4 domains, β-ankle loop, and β-tail domains in β₃ form the lower legs (Fig. 1). Large interfaces are buried between the head and the lower legs, and between the α_V and β₃ legs, that stabilize the bent conformation (2). Since structures of α_Vβ₃ (14, 20, 22, 33) and α_IIbβ₃ (4, 18, 21, 34, 35) have previously been described, we focus our attention on the most notable differences and the structural features that are relevant for models of integrin ligand-binding and activation.

Lack of difference between the SDL of α_Vβ₃ and α_IIbβ₃

The specificity-determining loop (SDL) extends from the βI domain at its interface with the β-propeller domain. It locates just outside the binding footprint for small peptide ligands, and based on exchange between different integrin β-subunits, contributes to binding of macromolecular ligands (36). All previous α_Vβ₃ structures (Fig. 4A) differ from all α_IIbβ₃ structures (Fig. 4D) over β₃ SDL residues 168-176 in structure to sequence register, and by the absence and presence, respectively, of two cis-peptides in the SDL at Ser-162/Pro-163 and Ser-168/Pro-169 (Fig. 4A and D). We wondered if these differences were induced by differences between the α_IIb and α_V β-propeller domains with which the SDL associates. However, inspection of the electron density for the previous α_Vβ₃-1TM structure (14) reveals clear signatures for both cis-Pro, and a sequence-to-structure shift (Fig. 4E). For example, negative density on the carbonyl group of Pro-169 when it is trans (Fig. 4E), is eliminated when it is cis (Fig. 4F), and positive density between the sidechains of Pro-169 and Pro-170 (Fig 4E) is satisfied by the sidechain of Pro-169 when it is cis (Fig. 4F). Furthermore, positive densities near Glu-174, Asn-175, and Pro-176, and negative density at Leu-173 (Fig. 4E) are corrected when the sequence-to-structure register is shifted and the backbone is flipped at Pro-176/Cys-177 (Fig. 4F). Rebuilding to follow the electron density resulted in SDL backbones in α_Vβ₃-1TM (Fig. 4B) and α_Vβ₃-AB (Fig. 4C) similar to that in α_IIbβ₃ (Fig. 4D).

Figure 4.

The conformation of the specificity-determining loop in the βI domain. The SDL and surrounding area (A-D) or a portion of the SDL loop (E-G) are shown for the 3IJE (A and E) or 4G1M (B and F) α_Vβ₃-1TM structures, α_Vβ₃-AB (C and G), and α_IIbβ₃ (D). β- and α-subunits are wheat and light blue, respectively. Mesh in E-G shows 2 Fo-Fc density at 1 σ (grey) or Fo-Fc difference density at 3 σ (green) and -3 σ (red).

A metal ion at the SyMBS or MIDAS in unliganded α_Vβ₃

The α_Vβ₃-AB structure has no metal at the MIDAS or ADMIDAS, but has density for Ca²⁺ at the SyMBS (Fig. 5A and D). The α_Vβ₃-1TM structure has only weak density at the SyMBS, modeled as a water in 4G1M (Fig. 5B). Positive density at the MIDAS in the 3IJE dataset (Fig. 5E) has been satisfied by placing a Ca²⁺ ion at the MIDAS in 4G1M (Fig. 5F). Thus α_Vβ₃-1TM has metal ions at both the MIDAS and ADMIDAS (Fig. 5B). The Ca²⁺ at the MIDAS has higher B factors than the surrounding coordinating residues, and therefore may have partial occupancy; furthermore, a water or Mg²⁺ ion (which is lacking in buffers but could have been bound to α_Vβ₃) could also be built at the MIDAS. The most reasonable interpretation is Ca²⁺ or Mg²⁺, both because a metal ion is normally bound here, and because of the close proximity of three negatively charged residues. The position of the metal ion differs from that in α_IIbβ₃, which appears related to the lack of metal ion binding to SyMBS in α_Vβ₃-1TM and the consequent shift in position of the dual SyMBS and MIDAS-coordinating residue Glu-220 (Fig. 5B and C). Remarkably, the occupancy of the three βI metal ion binding sites is completely reversed between α_Vβ₃-AB and α_Vβ₃-1TM (Fig. 5A and B).

Figure 5.

The metal ion binding sites in the βI domain. A-C. Differences in metal binding to the βI domain among α_Vβ₃-AB (A), α_Vβ₃-1TM (B), and α_IIbβ₃ (C). D-F. Density at the metal-binding sites. D. The SyMBS in α_Vβ₃-AB. E and F. The MIDAS in 3IJE (E) and 4G1M (F) α_Vβ₃-1TM structures. Black mesh shows 2Fo-Fc density at 1 σ, and green and red mesh show Fo-Fc difference density at 3 σ and -3 σ, respectively. Ca²⁺ and Mg²⁺ ions are shown as gold and silver spheres, respectively. Oxygens are red and waters are small red spheres. Dashed lines in A-C show metal coordination.

Overall flexibility in the ectodomain

Crystallographic B-factors reflect atomic displacements in the crystal lattice and are hence related to protein flexibility. In the α_Vβ₃-1TM backbone, B-factors are much higher than the overall average in I-EGF1 and in the N-terminal portion of I-EGF2, which connects to I-EGF1 and is near to the thigh domain (Fig. 6A). In the α_Vβ₃-AB backbone, B-factors in the β₃ subunit are high in a long PSI domain loop, in the last disulfide-bonded loop of I-EGF1, and in the first loop of I-EGF2, which has a gap in continuous density (Fig. 6C). In α_IIbβ₃, the first loop in I-EGF2 also has high B factors, and is partially missing in density (Fig. 6E). The highest B-factors in α_IIbβ₃ occur in the β-tail domain, which is poorly ordered (Fig. 6F).

Figure 6.

Flexibility among α_Vβ₃ crystal forms and α_IIbβ₃ assessed by B factor distributions. The orientations are similar to those in Fig. 2A-C, except in the lower panels the integrins are rotated about the axis vertical in the page. Integrins were superimposed as in Fig. 2, and separated horizontally on the page. Cα-traces show B factors in rainbow, set using the lowest and highest Cα B factors in each integrin as the blue and red ends of the rainbow, respectively. The B factors shown include contributions from TLS. The genu Ca²⁺ ions are shown as spheres and also colored by B factor. Disulfide bonds are silver. The names of the α_V, α_IIb, and β₃ subunits are shown next to their C-termini.

The overall position of the genu moves markedly between the two α_Vβ₃ structures (Fig. 2A and B), and the α-genu in α_Vβ₃-AB also has high B-factors (Fig. 6C). Genu movement is tied to changes in position of both the thigh and calf-1 domains between the two structures. The genu-proximal end of the calf-1 domain is 3.5 Å closer to I-EGF2 in α_Vβ₃-AB than in α_Vβ₃-1TM. Furthermore, the thigh domain rotates along its long axis. Rotation is not about the center of the domain, but is centered on β-strand-F, which faces inward toward the lower α- and β-legs. Far from the center of rotation, the long C-D loop in thigh moves 6 Å closer to I-EGF2 in α_Vβ₃-1TM than in α_Vβ₃-AB.

The other large movement between α_Vβ₃-1TM and α_Vβ₃-AB occurs in the PSI and I-EGF1 domains and in the PSI- and I-EGF1-proximal end of the hybrid domain. While there is little change in position of the βI-proximal end of the hybrid domain, the PSI-proximal end of the hybrid domain moves 2 to 2.5 Å. This libration of the hybrid domain is coupled to displacement of 2 to 3 Å of the PSI and I-EGF1 domains, in a direction that places PSI and I-EGF1 further away from α_V in α_Vβ₃-1TM than in α_Vβ₃-AB.

Flexibility in I-EGF1 and I-EGF2

Flexibility occurs within I-EGF1 and I-EGF2, as well as at their junctions with other domains. Cα-traces colored in rainbow by B factor show marked flexibility within the loops at the C-terminal end of I-EGF1 and at the N-terminal end of I-EGF2, near the junctions between these domains (Fig. 7). Just before the final Cys-521 of I-EGF2 (C8, Fig. 7E-G), Tyr-520 (Y in Fig. 7E-G) participates in hydrophobic interactions with core elements of the domain, and is shielded on one side by Lys-519 (K in Fig. 7E-G). Tyr-520 is highly conserved as an aromatic or hydrophobic residue in I-EGF2 of other integrin β-subunits. In contrast, I-EGF1 lacks elements that stabilize the conformation of its C7-C8 loop, and has Ser-469 and Gln-470 (S and Q in Fig. 7A-D, respectively) at the sequence positions equivalent to Lys-519 and Tyr-520 (Fig. 7E-G).

Figure 7.

The I-EGF domains at the β₃-knee. A-G. I-EGF1 and I-EGF2 domains from structures determined here, the α_IIbβ₃ ectodomain (18), or the closed α_IIbβ₃ headpiece (PDB ID 3T3P) (35) were superimposed, and are shown in identical orientations aligned vertically and horizontally on the page. Cα ribbons are rainbow colored from lowest (blue) to highest (red) Cα B-factor in each domain. Disulfide bonds are color-coded and cysteines numbered according to the same scheme as shown in panel I. Selected sidechains (S469 and Q470 in I-EGF1 and the equivalent K519 and Y520 in I-EGF2) are shown. The C-terminus of I-EGF1 is indicated by showing the Cα-C bond of C8 in silver. H. The markedly different orientation of the I-EGF1 domain C7-C8 loop in a closed headpiece crystal structure (blue) is shown by superimposing the I-EGF1 domain of α_Vβ₃-AB (red). Surrounding molecules in the closed headpiece crystal lattice are shown in different colors. I. Sequence alignment of the β₃-integrin I-EGF domains.

The flexibility of the C7-C8 loop in I-EGF1 is emphasized in α_IIbβ₃ headpiece structures in which the β₃ subunit terminates in I-EGF1. This loop is disordered in most such structures (4, 34). However, in recent closed headpiece structures with two headpieces per asymmetric unit (21, 35), the electron density is good enough to build most of the C7-C8 loop in one molecule and all of the C7-C8 loop in the other molecule (Fig. 7D). Remarkably, this loop adopts a markedly different orientation. Although there is room in the crystal lattice for the C7-C8 loop to adopt the same conformation as seen in the complete ectodomain, as shown by superposition (Fig. 7H), the C7-C8 loop instead moves to another position in the lattice, where the backbones of Pro-464 and Gly-465 participate in a network of water-mediated hydrogen bonds to the β-propeller domain of a neighboring molecule in the crystal lattice of the 2.2 Å structure.

Lack of a stable interface between the I-EGF2 and thigh domains

Substantial negative and positive Fo-Fc difference density is present in the interface between the β₃ I-EGF2 and α_V thigh domains in the original α_Vβ₃-1TM model (Fig. 3A) but not in the map calculated using improved coordinates (Fig. 3B). This is a region of high B-factors, particularly in I-EGF2, but also in the extended CD loop of the α_V thigh (Fig. 6A and C). Near this interface in the I-EGF1 domain, between the two α_Vβ₃-1TM models the Cα atom positions shift 2.3 Å for Leu-467 and 1.2 to 1.5 Å for Gly-463, Pro-464, and Gly-465. In the C1-C2 loop of the I-EGF2 domain, at the interface with the thigh domain, the Cα atoms of Glu-475, Glu-476, Asp-477 (Fig. 8A and B), Tyr-478, and Arg-479 shift 1.8, 1.8, 1.5, 1.6, and 3.5 Å, respectively, between the two models. Although these residues are more accurately located in our model, as shown by presence of backbone density in simulated annealing composite omit maps calculated with 4G1M but not 3IJE structures (Methods), the differences in position emphasize that these are among the regions with the poorest electron density, and the most difficult to build.

Figure 8.

Lack of stable interfaces between I-EGF2 and thigh domains. Interfaces are shown in identical orientations after superposition of the ectodomains using the super command of PyMol. Thigh (green) and I-EGF2 (yellow) are shown with red oxygen and blue nitrogen atoms. All inter-subunit hydrogen bonds found as polar contacts by PyMol are shown with dashed lines. Only 2 of the 11 hydrogen bonds shown in Fig. 4D of Xiong et al. (14) appear in (A), either because highly solvated carboxyl-carboxyl and carboxyl-carbonyl interactions are considered repulsive at neutral pH or the distances are too far; the orientations shown in Fig. 4D of reference (14) differ markedly from the deposited 3IJE coordinates. All sidechain or backbone atoms shown in Fig. 4D of reference (14) are shown in stick.

The C1-C2 loop is markedly longer in I-EGF2 than in other integrin I-EGF domains (18). Several residues in this loop were too disordered to build in α_Vβ₃-AB or α_IIbβ₃ (Fig. 7E and G). Missing density here and elsewhere in the molecular model is clearly not due to proteolytic cleavage, as demonstrated by migration of the α_V and β₃ subunits at the expected molecular weight in SDS-PAGE (Supp. Fig. S1).

Across from I-EGF2, the electron density on the thigh side of the interface is better; nonetheless, substantial rebuilding was required to fit the density. Thus, in the interface the Cα atoms of thigh residues Gln-504, Lys-505, Gly-506, and Glu-547 shift 1.4, 2.1, 1.4, and 3.1 Å, respectively, between the two models (see E547 in Fig. 8A and B). There are no hydrogen bonds or significant hydrophobic interactions between I-EGF2 and thigh in our model of α_Vβ₃-1TM (Fig. 8B), consistent with the substantial shift in orientation at the I-EGF2/thigh interface between α_Vβ₃-1TM and α_Vβ₃-AB (Fig. 8B and C).

Lack of a stable interface between the β-tail and βI domains

The interface between the β-tail CD (deadbolt) loop and the βI domain is extremely small in both the 4G1M α_Vβ₃-1TM and α_Vβ₃-AB structures, at 40 Ų and 50 Ų, respectively. Furthermore, there are no significant van der Waals interactions or hydrogen bonds between these domains. Moreover, the β-tail CD loop has high B-factors, particularly in α_Vβ₃-1TM (Fig. 6B).

Interaction of an α_V glycopeptide moiety with β₃

Near the furin cleavage site in the calf-2 domain of α_Vβ₃-AB, we found a polypeptide-like electron density protruding from calf-2 toward β₃, into a cavity between the I-EGF4 and β-tail domains. A second large, branched density corresponding to an N-linked carbohydrate extends from the first density. This enabled us to identify the polypeptide density as α_V residues 841-848, and to build two N-acetylglucosamine and three mannose residues N-linked to Asn-844 of the consensus N-glycosylation sequence (Fig. 9A). Around the furin cleavage site at R860, residues 839-840 and 849-866 are disordered.

Figure 9.

Interaction of an α_V glycopeptide moiety near the furin cleavage site with β₃. A. Interaction of the glycopeptide with the β-subunit. The glycan N-linked to α_V-Asn-844 of the calf-2 domain (orange) inserts into the pocket formed by I-EGF4 domain (light blue), β-ankle (pink), β-tail domain (green), hybrid domain (yellow) and βI domain (red). Carbohydrate residues are displayed with gray carbons. α_V residues 838, 841, 848, and 867 are marked, which are adjacent to positions missing in density. B. Verification of furin cleavage site mutation by immunoprecipitation of [³⁵S]-labeled α_Vβ₃ followed by reducing SDS 7.5% PAGE and fluorography. C and D. Ligand binding to cells with wild-type or mutant α_Vβ₃ measured by fluorescent flow cytometry. Binding of fluorescently labeled fibrinogen or fibronectin to transfectants was tested in 5 mM EDTA, 5 mM Ca²⁺, or 1 mM Mn²⁺. The wedge mutant (8) was used as positive control. Results are expressed as mean fluorescent intensity of Alexa-488 conjugated ligand as a percentage of mean fluorescent intensity of Cy3-conjugated AP3 antibody to β₃.

The α_V 841-848 glycopeptide moiety inserts in a cavity between the β₃ headpiece and lower leg, and forms contacts with the βI, hybrid, I-EGF4, β-ankle, and β-tail domains (Fig. 9A). It is unlikely to be able to occupy the same position prior to furin cleavage. Therefore, we tested the effect on α_Vβ₃ function of an α_V-R860A mutation. This mutation eliminated the cleavage by furin of the α_V-subunit (Fig. 9B). However, it had little effect on binding of fibrinogen or fibronectin to α_Vβ₃ transfectants measured by fluorescent flow cytometry (Fig. 9C and D). In contrast, introduction of an N-linked glycan wedge into the βI/hybrid domain interface to favor the open headpiece conformation markedly increased binding in both Ca²⁺ and Mn²⁺ (Fig. 9C and D).

Linkage between the extracellular and transmembrane portions of α_Vβ₃

Improvement of the molecular model of α_Vβ₃-1TM included rebuilding of the C-terminal portion of the calf-2 domain. The backbones of L755 and G954 were rotated, and the backbone of W953 was flipped (Fig. 10A and B). These changes allowed these residues to fit better into density and enabled the strong hydrogen bond between the backbones of α_V L755 and G954 to fit into continuous 1 σ 2Fo-Fc electron density (Fig. 10B). We traced generously, i.e. using 0.5 σ 2Fo-Fc electron density, the linkers between the ectodomain and the membrane, which have high B factors (Fig. 6B). Our α_Vβ₃-1TM model thus extends to the A958-P959 peptide bond in α_V and the P691-D692 peptide bond in β₃ (Fig. 10B). We were unable to discern any trace of the α_V and β₃ polypeptide chains beyond these positions. Thus, we were unable to build residues α_V M960-I967 or β₃ I693-V695 that were included in the 3IJE model (Fig. 10A). In α_Vβ₃-AB, we were able to build up to α_V residue P959 and β₃ residue P691 (Fig. 10C).

Figure 10.

The structure of the linkers between the α_Vβ₃ ecto and transmembrane domains. A. 3IJE α_Vβ₃-1TM. B. 4G1M α_Vβ₃-1TM. C. α_Vβ₃-AB. Shown are the C-terminal α_V and β₃ extensions together with the C-terminal portions of calf-2 and β-tail domains as a Cα-ribbon, together with sidechains and backbones of selected regions. Hydrogen bonds near the end of the calf-2 and β-tail domains are shown as red dashed lines. Electron density is shown as mesh around α_V residues 755 and 953-967 and β₃ residues 663 and 685-695. Mesh for 1 σ 2Fo-Fc, -3 σ Fo-Fc, and 3 σ Fo-Fc electron density is shown in blue, red, and green, respectively. CCP4 format maps were opened and displayed with the isomesh command in PyMol using carve = 2.4 and 1.9 Å for 2Fo-Fc and Fo-Fc maps, respectively. Mesh was calculated in A using 3IJE data and in B using 4G1M data. In each of A and B, mesh was separately calculated (carved) around the positions of atoms in both the 3IJE and 4G1M models, and the combined mesh is displayed. The view is similar to that in Fig. 4A of (14).

The difficulty of tracing the α_Vβ₃-1TM α_V linker is emphasized by the markedly different position of the A958 sidechain in the two models (Fig. 10A and B). Lack of experimental support for the C-terminal portion of α_V in the 3IJE model is further shown by negative 3 σ Fo-Fc difference electron density on the backbone of α_V residues P957, P959, M960, P961, V962, P963, W965, and V966, as well as on the sidechain of W965 (Fig. 10A). Scattered 1 σ 2Fo-Fc electron density in this region of the 3IJE backbone (Fig. 10A) did not appear in electron density calculated using the 4G1M model (Fig. 10B), suggesting that this density was due to model bias. Absence of α_V residues 960-967 in our model eliminated negative Fo-Fc difference electron density, and positive electron density did not appear along the omitted backbone (Fig. 10B).

We were also unable to build β₃ TM residues I693-V695 (Fig. 10B). Continuous negative 3 σ Fo-Fc electron density is present along the backbone of β₃ residues 691-695 in the 3IJE dataset (Fig. 10A), providing evidence against the path built for these residues. In this region of the electron density map for the 4G1M α_Vβ₃-1TM model, no negative or positive difference electron density appears (Fig. 10B).

Discussion

Differences among crystal structures

Two crystal structures of α_Vβ₃ have revealed unanticipated features in this well-studied integrin. The SDL is functionally important for binding macromolecular ligands (36). Although previous structures suggested major conformational differences between this β₃ loop in α_Vβ₃ and α_IIbβ₃, new α_Vβ₃ structures reported here reveal no significant differences. The SDL loop in β₃ is thus little influenced by the associating α-subunit.

Three metal binding-sites are present at the ligand binding site of the integrin βI domain. Density appropriate for a metal ion is present at the MIDAS in α_Vβ₃-1TM, when phases are calculated using 3IJE or 4G1M coordinates. This contrasts with the previous assertion that no metal was present at this site in α_Vβ₃-1TM (14); however, our interpretation of a metal ion at the MIDAS is consistent with report of a density feature at this site in the α_Vβ₃-ΔTM structure (22). In α_Vβ₃-AB only the SyMBS site is occupied, whereas in α_Vβ₃-1TM only the MIDAS and ADMIDAS are occupied (Fig. 5). In contrast, all three metal ion binding sites are occupied in unliganded α_IIbβ₃ crystal structures (18, 21). The SyMBS had previously been designated LIMBS, for ligand-induced or –associated metal binding site (20). However, our α_Vβ₃-AB structure clearly demonstrates that ligand binding is not required for occupancy of this site, supporting earlier observations with α_IIbβ₃ and the SyMBS designation (18, 21). Previous studies have shown that binding of Ca²⁺ to the SyMBS and Mg²⁺ to the MIDAS is synergistic (37); this synergism is explicable both by the close proximity of these two metal ion-binding sites, and bidentate coordination of the Glu-220 sidechain to both the SyMBS and ADMIDAS metal ions (Fig. 5C).

The striking differences between metal ion binding to identical β₃ subunits in α_Vβ₃ and α_IIbβ₃ may be due to crystallization conditions. The use of both Mg²⁺ and Ca²⁺ in α_IIbβ₃ (18, 21) and α_Vβ₃-AB crystals, and only Ca²⁺ or only Mn²⁺ in α_Vβ₃-ΔTM and α_Vβ₃-1TM crystals (14, 20, 22, 33) may be one factor. The use of PEG and ammonium sulfate as precipitants in crystallization of α_Vβ₃-1TM and α_Vβ₃-AB, respectively, may also contribute to differences. The different pH ranges at which α_Vβ₃ and α_IIbβ₃ crystallize may be the major factor. α_Vβ₃-ΔTM crystallized at pH 6.0 (22); α_Vβ₃-1TM was variously reported to crystallize at pH 5.5 in the publication or pH 4.8 in the deposited coordinates (14); and α_Vβ₃-AB crystallized at pH 5.8. Among α_IIbβ₃ structures in absence of ligand, the complete ectodomain crystallized at pH 7.0 (18) and the headpiece at pH 8.9 (21). Asp and Glu have pK values of 3.9 and 4.3, respectively. When these residues are close to other negatively charged residues, as they are at the SyMBS, MIDAS, and ADMIDAS, Coulomb's law dictates an increase in pK (38). This effect could easily result in substantial protonation of metal-coordinating residues in the pH range 4.8 to 6.0 used in crystallization of α_Vβ₃, and hinder occupation of metal-binding sites. In contrast, the same residues would be more ionized at the physiologic pH of 7.4, and at pH 7.0 and 8.9 as used in α_IIbβ₃ crystallization. In contrast to the βI domain sites, the β-propeller β-hairpin Ca²⁺-binding sites and genu are well occupied in α_Vβ₃ crystals. This difference is easily explicable by the close spatial relationship of the SyMBS, MIDAS, and ADMIDAS to one another (38). Furthermore, the MIDAS and ADMIDAS have more water coordinations than any of the other integrin metal ions.

These observations suggest that the lack of simultaneous metal ion-binding to all three βI domain sites in α_Vβ₃ is most simply explained as an artifact of crystallization conditions. An alternate explanation, that the sidechain of SyMBS residue Asp-217 orients differently in α_Vβ₃ than in α_IIbβ₃ (14) encounters the issues that 1) either orientation fits the electron density at 2.9 Å and we have built the same orientation in α_Vβ₃-1TM and α_Vβ₃-AB as in α_IIbβ₃; 2) the orientation we have built is required to coordinate the SyMBS metal ion and satisfy all six octahedral coordination positions; 3) Fig. 2B of (20) shows Asp-217 of α_Vβ₃-ΔTM in the same orientation as we have built in α_Vβ₃-1TM; and 4) the Asp-217 orientation built in the 3IJE model is disfavored by overlap of the electron orbitals of Oδ1 of β₃ Asp-217 and Oδ1 of α_V of Asp-219, which are only 2.8 Å apart.

The concept that Ca²⁺ and Mg²⁺ bind synergistically at the SyMBS and MIDAS (18), and that at low pH conditions with α_Vβ₃, metal binding and ligand binding are also synergistic, is more plausible than the concept that ligand binding is regulated by or induces metal ion binding (39). It would be difficult for the Asp of RGD to bind to the MIDAS in the absence of a metal ion and in the face of strong repulsion from negatively charged MIDAS residues. Furthermore, the SyMBS site is buried, and metal ion access to this site would be further limited after ligand binding.

The remarkable contact of a furin cleavage site-proximal α_V glycopeptide moiety with the β₃ subunit seen here might also be dependent on crystallization conditions. (NH₄)₂SO₄ was the α_Vβ₃-AB precipitant, whereas in the α_Vβ₃-1TM crystal structure with PEG precipitant, no density for a similar glycopeptide moiety is discernable. These differences are in agreement with the highly polar interactions of the α_V glycopeptide moiety with β₃, which are expected to be easily replaceable with water or PEG. We found no effect of furin cleavage on basal or Mn²⁺-stimulated α_Vβ₃ ligand binding, although effects might emerge in other assays. Mutational elimination of furin cleavage sites has little effect on ligand binding in α_IIbβ₃ and α₄β₁ (40, 41) but is required for inside-out stimulation of adhesion by α₆β₁ (42).

Implications for integrin activation

A deadbolt model proposed that removing β-tail domain constraints on the βI domain could induce integrin activation without extension (43). In the original proposal of this hypothesis, it was acknowledged that the β-tail domain CD or “deadbolt” loop had high temperature factors, buried a very small surface area on the βI domain, and was unlikely to contribute stabilizing energy in the conformation seen in the α_Vβ₃-ΔTM structure; therefore, it was suggested that minor rearrangements might result in a larger interface in intact α_Vβ₃ on the cell surface (43). However, deletion of the β-tail domain CD loop has no activating effect whatsoever on α_Vβ₃ and α_IIbβ₃ on cell surfaces (44). Moreover, despite the negligible contact between the β-tail CD loop and the βI domain in α_Vβ₃-ΔTM, there is even less in α_Vβ₃-1TM, none in α_Vβ₃-AB where the CD loop is disordered, and none in α_IIbβ₃ (18).

Another interface, between the α_V thigh domain and β₃ I-EGF2 domains, was recently proposed to be an energetic barrier, that when overcome, could enable α_Vβ₃ activation in the absence of extension (14). However, electron density is poor at this interface. We found it necessary to substantially rebuild PSI, I-EGF1, I-EGF2, and loops in thigh adjacent to I-EGF2 in our α_Vβ₃-1TM model. Cα atom positions in loops relevant to a proposed interface are shifted substantially between the 3IJE and 4G1M α_Vβ₃-1TM models. Furthermore, the sidechain orientations shown in Fig. 4D in (14) are inconsistent with the associated 3IJE coordinate file, and suggest that different coordinate files were used for figure preparation and coordinate deposition. Thus, Fig. 4D in (14), but not the 3IJE coordinates (Fig. 8A) show hydrogen bonds between the α_V Glu-547 sidechain and β₃ Asp-477 backbone, and between the α_V Phe-548 backbone and β₃ Glu-476 sidechain. However, questionable hydrogen bonds between the sidechains of α_V Asp-550 and β₃ Glu-500, and between the backbone carbonyl of α_V Glu-547 and sidechain of β₃ Glu-476, remain in the 3IJE coordinate file (Fig. 8A). The 4G1M α_Vβ₃-1TM coordinate file lacks favorable, as well as unfavorable polar interactions in the same region (Fig. 8B). The lack of a stable interface here, and its irrelevance for regulation of integrin affinity state, is further supported by its marked reorientation in α_Vβ₃-AB (Fig. 8C), with no attendant reshaping of the βI domain ligand binding site. These results are consistent with extensive mutagenesis evidence that the loop between cysteines 1 and 2 of I-EGF2 acts as an entropic spring, and that its length, but not its sequence, is important in regulating the equilibrium between bent and extended integrin conformations (45).

In α_Vβ₃-1TM, we were unable to trace α_V linker residues Met-960 to Pro-963, α_V TM residues Val-964 to Ile-967, or β₃ TM residues Ile-693 to Val-695. Strong evidence against the placement of α_V residues 960-967 and β₃ residues 693-695 is shown by the presence of 3 σ negative difference electron density along this backbone and on the sidechain of α_V Trp-965 in 3IJE (Fig. 10A), and the lack of appearance of positive difference electron density in the corresponding regions of the 4G1M α_Vβ₃-1TM model (Fig. 10B). Flexibility in the α_V and β₃ linkers is further evidenced by differences in orientation between α_Vβ₃-1TM and α_Vβ₃-AB beginning with α_V residue Gln-956 and β₃ residue Lys-689 (Fig. 10B and C).

Flexibility of ectodomain linker residues α_V 956-963 and β₃ 689-692 is to be expected from the lack of stabilizing interactions in the linker structures (Fig. 10B and C). The last residue well supported by structural interactions in α_V is G954, which has a backbone hydrogen bond to L755 (Fig. 10B and C). Similarly, the last such residue in β₃ is C687, which is disulfide-bonded to C663 (Fig. 10B and C).

Our conclusion that the α_V and β₃ linkers are flexible is in excellent agreement with a similar lack of density for the corresponding regions in α_IIbβ₃ and α_Xβ₂ crystal structures (18, 19). Furthermore, the use of disulfide cross-linking restraints in the linker regions of intact α_IIbβ₃ on cell surfaces showed that the α_IIb and β₃ linkers are flexible, and even without any motions internal to the bent conformation of α_IIbβ₃, enable marked change in orientation of the ectodomain with respect to the membrane (15). However, there are limits to linker flexibility; coupling of conformational change between extracellular and TM domains is disrupted by insertions of 10 to 11 extra residues (15, 46). The finding that signals can be transmitted across membranes despite flexibility of extracellular linkers has also emerged in the epidermal growth factor receptor field, where spatial proximity between two monomers, rather than precise orientation, is important in transmembrane signaling by dimers (47).

We also reveal differences in positions of the structured domains in the two snapshots of the ectodomain in α_Vβ₃-1TM and α_Vβ₃-AB. A large number of leg domains, three in α and six in β, intervene between integrin ligand-binding domains and the ectodomain-TM domain linkers. As demonstrated in variation in orientation between these leg domains in α_Vβ₃-1TM and α_Vβ₃-AB (Table 2), changes in TM domain orientation could be dissipated by change in leg domain orientation. Any alternative theory of integrin activation within the bent conformation should propose a specific interdomain pathway for signal transmission, and a change in orientation between the α and β subunit TM domains sufficiently large in magnitude to prevent damping of transmission by ectodomain flexibility. The only specific pathway in the ectodomain described to date for transmitting conformational change to the ligand-binding site is through hybrid domain swing-out (4). Such a large conformational change, including a 70 Å increase in separation at the knees, is incompatible with maintenance of a compact, bent conformation.

Many mutational studies on α_V and β₃ support the importance of integrin extension, lower leg separation, and headpiece opening in integrin activation. Interfaces between the α_V and α_IIb calf-2 domain and β₃ I-EGF4, β-ankle, and β-tail domains are required for maintaining the low-affinity state; these interfaces prevent lower leg separation (48, 49). Interfaces between hybrid domain and β-tail domain (50), and between the α_V β-propeller domain and the upper β-leg (51) stabilize the low-affinity state by preventing integrin extension. Mutations that open up disulfide bonded loops in the I-EGF domains (52, 53) or introduce bulky residues (54) or N-glycosylation sites (55) into portions of the β-leg that are buried in the bent conformation activate α_Vβ₃ and α_IIbβ₃; these mutations destabilize the bent conformation. Mutational introduction of disulfides to prevent integrin extension (2, 56), headpiece opening (2, 56), or lower leg separation (56) abolish activation of α_Vβ₃. Stabilizing the open headpiece relative to the closed headpiece with glycan wedges or computationally predicted mutations stimulated α_Vβ₃ and α_IIbβ₃ integrin activation and LIBS epitope exposure (8, 57). Conversely, stabilizing the closed headpiece with a disulfide bond in the βI domain or computationally predicted mutations inhibited activation (57, 58).

In summary, extensive mutagenesis and structural studies of α_Vβ₃ and α_IIbβ₃ have demonstrated the importance of multiple interfaces between the integrin headpiece and lower legs, and between the integrin α- and β-legs, that are present in the bent conformation and restrain integrin activation. These interfaces involve large buried surface areas and well-ordered sidechains, and thus contrast with the interfaces discussed above between the β-tail and βI domains and between the α- and β-knees. Integrin extension, lower leg separation, and headpiece opening have each been shown to be required for integrin activation. The structures described here provide a sound foundation for better understanding of the complex molecular mechanisms that regulate integrin activation.

Abbreviations

MIDAS

Metal ion-dependent adhesion site

ADMIDAS

adjacent to MIDAS

SyMBS

synergistic metal binding site

TM

transmembrane

LIBS

ligand-induced binding site

EM

electron microscopy

FRET

fluorescence resonance energy transfer

SAXS

small angle X-ray scattering

PDB

Protein Data Bank

TEV

tobacco etch virus

APS

Advanced Photon Source

MR

molecular replacement

ML

maximum likelihood

NMR

nuclear magnetic resonance

SDL

specificity-determining loop