Structure of an extended β₃ integrin

Integrins are a family of 24 α/β heterodimeric adhesion receptors that transmit bidirectional signals across the cell membrane.¹  Each integrin subunit contains a large extracellular domain composed of a headpiece and leg piece, a transmembrane domain, and usually a short cytoplasmic tail (supplemental Figure 1, available on the Blood Web site).²  The α headpiece consists of β-propeller and thigh domains and an inserted αI domain on top of the β-propeller in a subset of α subunits, such as leukocyte-specific α_L. The α leg piece includes calf-1 and ‒2 domains. The β headpiece contains βI, hybrid, PSI, and I-EGF1 domains (Figure 1A). The β leg piece contains I-EGF2 to -4 and β-tail domains (Figure 1A). Structural studies in the last 17 years, largely based on β₃ integrins, including α_IIbβ₃ and α_Vβ₃,^3-11  have greatly advanced our understanding of integrin conformational regulation upon activation. Platelet-specific α_IIbβ₃ plays an essential role in hemostasis and thrombosis,¹²  and α_Vβ₃ is important for the growth and survival of many cell types.^13,14  In the resting state, integrin adopts a bent conformation, with the headpiece folded onto the leg piece (supplemental Figure 1A). Crystallographic and electron microscopy (EM) studies of α_IIbβ₃ and α_Vβ₃ have revealed 2 major conformational rearrangements: the swing-out motion of the hybrid domain, resulting in headpiece opening, and the switchblade-like movement of the headpiece from the leg domains, leading to integrin extension and leg separation (supplemental Figure 1B-C).^2,7  Despite the visualization of β₃ extension by EM at low resolution (supplemental Figure 1B-C), a high-resolution extended β₃ structure is not available.

A Leu33Pro polymorphism at the β₃ PSI domain defines the human platelet-specific alloantigen 1 (HPA-1) system.¹⁵  Mismatched HPA-1a/b (Leu33/Pro33) is a major cause of 2 severe bleeding disorders: fetal/neonatal alloimmune thrombocytopenia and posttransfusion purpura, which result from the generation of anti–HPA-1a alloantibodies.^16,17  In addition, whether the Pro33 variant enhances β₃ activity, as a risk factor for thrombosis, is controversial.^18-20  It is unknown whether Leu33Pro substitution affects the conformation of the PSI domain, which undergoes outward movement along with headpiece opening and extension.

Crystallographic studies of a full-length extended integrin have not been successful. This is largely due to the flexibility of the β-leg domains in the extended conformation as indicated by EM studies.²  In our study, to obtain the high-resolution crystal structure of an extended β₃ integrin, we engineered a cβ₃ fragment containing the headpiece and β-knee region at where β₃ becomes extended. cβ₃ could be expressed without an α subunit. We solved the crystal structures of the cβ₃ fragment in both Pro33 and Leu33 forms. The structures revealed intermediate and fully extended β₃ conformations at the β knee, providing high-resolution information on how β₃ becomes extended. The structures of full-length β₃ integrin were built in both intermediate and fully extended conformations, agreeing well with the corresponding conformations observed by EM. Our structural and functional studies provide an atomic-level understanding of integrin conformational regulation on the cell surface.

The α_IIb, α_V, α_L, α₅, β₁, β₂, β₃, and EGFP-talin head constructs were as described previously.^21-23  Mutations were generated by the QuikChange XL Site-Directed Mutagenesis Kit (Agilent Technologies). The cβ₃ construct was designed by replacing the β₃ βI domain (residues Tyr110-Gly349) with the α_L αI domain (residues Gly128-Lys304). The full-length cβ₃ DNA construct was generated by overlapping polymerase chain reaction and cloned into the pcDNA3.1/Myc-His⁺ A vector using HindIII and XbaI restriction sites. The DNA fragment of the cβ₃ headpiece containing the PSI with either Pro33 or Leu33, hybrid, αI, I-EGF1, and I-EGF2 domains was cloned into the XhoI and MluI sites of the ExpressTag 1 vector, which was based on the pIRES2-EGFP vector.²⁴  The construct was C-terminal tagged with an HRV 3C protease cleavage site, protein C epitope, hexahistidine, and streptavidin binding protein affinity tag. All constructs were validated by DNA sequencing.

The cβ₃ headpiece construct was transfected into HEK293S GnTI⁻ cells. The transfected cells were cultured in a medium containing 1 mg/mL of G418 for at least 10 days before a pool of EGFP⁺ cells were sorted. The EGFP⁺ cells were cultured in roller bottles in the selection medium. The cell-culture supernatant was collected every 3 to 5 days. The cβ₃ headpiece proteins were purified from the supernatant using Strep-Tactin Superflow Plus beads following the manufacturer’s protocol (Qiagen). The proteins were treated with HRV 3C protease to remove the tags and then with Endoglycosidase H to remove the high-mannose type of N-linked glycans. The untagged proteins were finally purified with a Superdex 200 column (GE Healthcare) equilibrated with a buffer containing 20 mM of Tris-hydrochloride (potential of hydrogen [pH], 7.5), 150 mM of sodium chloride (NaCl), 1 mM of calcium chloride, and 1 mM of magnesium chloride and concentrated to 10 mg/mL for protein crystallization.

Crystallization was performed at 4°C by hanging-drop vapor diffusion method with 1:1 volume/volume ratio of protein and cocktail solutions. Both Pro33 and Leu33 forms of cβ₃ headpiece proteins were crystallized in either 20% PEG 6000 plus 0.1 M of Tris-hydrochloride (pH, 8.5) or 12% PEG 8000 plus 0.2 M of NaCl and 0.1 M of N-2-hydroxyethylpiperazine-N′-2-ethanesulfonic acid (pH, 7.5). The crystals were cryoprotected with additional glycerol in the 18% PEG 8000 solution in 5% increments up to 20% final concentration. For dehydration, crystals were soaked in the 18% PEG 8000 solution containing 2 M of NaCl and 20% glycerol at 4°C for >3 days before being flash frozen in a pool of liquid nitrogen.

Crystal X-ray diffraction data were collected at the APS LS-CAT beamline 21-ID-F and processed with iMosflm.²⁵  The structure was solved by molecular replacement with PHASER²⁶  using the α_L αI domain crystal structure (PDB 3F74) as a search model. The structures were manually built in Coot²⁷  and refined using PHENIX.REFINE.²⁸  The atomic coordinates and structural factors have been deposited in the protein data bank under accession numbers 6BXF (for β₃-Leu33), 6BXB (for β₃-Pro33 after dehydration), and 6CKB (for β₃-Pro33 before dehydration).

The ligand mimetic PAC-1 (BD Biosciences), human fibronectin (Sigma-Aldrich) labeled with Alexa Fluor-647, and human ICAM-1 with a C-terminal Fc tag (expressed as a secreted protein in HEK293FT cells) were used as ligands for α_IIbβ₃, α_Vβ₃, and α_Lβ₂, respectively. 9EG7 (BD Biosciences) is a rat anti-β₁ mAb that reports β₁ integrin extension.^29,30  Both Alexa Fluor-647–labeled fibronectin and 9EG7 were used for measuring the activation of α₅β₁ integrin. The ligand or monoclonal antibody (mAb) binding assay using integrin-transfected HEK293FT cells was performed as described previously in the presence of integrin inhibitor Ca²⁺/Mg²⁺ (1 mM) or Ca²⁺ (0.2 mM) plus Mn²⁺ (2 mM).^21-23  The binding of ligand or conformation-specific mAb was presented as the mean fluorescence intensity normalized to the mean fluorescence intensity of integrin expression reported by the mAbs AP3, TS2/4, and MAR4 for β₃, α_Lβ₂, and α₅β₁, respectively.

The βI and αI domains are structural homologs.³¹  The αI but not βI domains can be expressed autonomously.^32-34  By replacing the βI with the αI domain of α_L integrin, the cβ subunit could be expressed as a single-chain construct independent of an α subunit (A.M.M.T., Z.W., D.Z., Y.Z., Brian R. Curtis, and J.Z., manuscript submitted April 2018). We solved the crystal structure of the single-chain cβ₃ ectodomain that adopts a bent conformation resembling β₃ in the α_IIbβ₃ or α_Vβ₃ complex (A.M.M.T., Z.W., D.Z., Y.Z., Brian R. Curtis, and J.Z., manuscript submitted April 2018). In the bent conformation of the β₃ subunit, the headpiece folds onto the lower leg domains, forming close contacts between the hybrid and I-EGF3 to -4 domains (supplemental Figure 1A).^4,5  Such interactions help maintain β₃ in the bent conformation even in the absence of an α subunit (A.M.M.T., Z.W., D.Z., Y.Z., Brian R. Curtis, and J.Z., manuscript submitted April 2018) and upon disruption may facilitate integrin extension. The β integrin extension occurs at the I-EGF1/2 junction, denoted as β knee (supplemental Figure 1B-C; Figure 1A).²  We generated a cβ₃ headpiece construct containing the PSI, hybrid, αI, I-EGF1, and I-EGF2 domains, namely cβ₃-E2 (Figure 1A). In the absence of the restrains of the lower leg domains, we expected cβ₃-E2 might adopt an extended conformation. To investigate whether the Leu33Pro polymorphism affects the β₃ conformation, both Leu33 and Pro33 forms of cβ₃-E2 were prepared.

Both Leu33 and Pro33 forms of cβ₃-E2 crystals belong to the same space group (Table 1). The original crystals diffracted X-ray up to 2.8-Å resolution, which was improved by crystal dehydration. The unit cell was shortened in 1 dimension after dehydration (Table 1), as a result of more tightly packed crystal lattice (supplemental Figure 2A). The cβ₃-E2-Pro33 structure obtained after crystal dehydration was refined to a resolution of 2.39 Å, with an R_free of 0.267. The cβ₃-E2-Pro33 structure obtained without crystal dehydration was refined to a resolution of 2.80 Å, with an R_free of 0.296. The cβ₃-E2-Leu33 structure was obtained with the crystals without dehydration and was refined to a resolution of 3.2 Å, with an R_free of 0.335 (Table 1). Two cβ₃-E2 molecules are present per asymmetric unit (Figure 1B).

All the domains in the cβ₃-E2 structure are well resolved in both molecules of the asymmetric unit (Figure 1B). As seen in the isolated αI structures, either Mg²⁺ or Ca²⁺ occupies the metal-ion–dependent adhesion site (Figure 1B). The remaining residues of the C-terminal tag after 3C protease cleavage form an antiparallel β-sheet interaction with the I-EGF2 domain (Figure 1B). Structural superimposition shows that the 2 molecules share similar conformations at the αI, hybrid, and PSI domains but have remarkably distinct conformations at the junction of I-EGF1 and I-EGF2 domains (Figure 1C). One molecule adopts an extended confirmation, whereas the other shows a bent-like conformation (Figure 1C). When compared with the native β₃ headpiece structures, cβ₃-E2 resembles the closed rather than open headpiece conformation (Figure 1D). However, when superimposed onto the αI domain of the cβ₃ ectodomain crystal structure (A.M.M.T., Z.W., D.Z., Y.Z., Brian R. Curtis, and J.Z., manuscript submitted April 2018), a slight swing-out of the hybrid domain was observed (supplemental Figure 3A-B). This was associated with the downward displacement of the α7 helix C-terminus (supplemental Figure 3A-B). Nonetheless, the αI domain of cβ₃-E2 was in a low-affinity conformation, as seen in the cβ₃ ectodomain crystal structure (supplemental Figure 3A-B).

These structural features were observed in both cβ₃-E2-Leu33 and cβ₃-E2-Pro33 crystals. No significant structural differences were observed between these 2 isoforms of cβ₃-E2 (supplemental Figure 4A-D). Therefore, the Pro33 form was used to represent the cβ₃-E2 structure, because it gave a higher resolution than the Leu33 form (Table 1). Of note, the 2 cβ₃-E2 molecules in the asymmetric unit packed closer to each other after crystal dehydration (supplemental Figure 2B-C), but no significant conformational changes were observed after dehydration (supplemental Figure 2B-C).

As expected, we observed an extended conformation at the I-EGF1/I-EGF2 junction in the cβ₃-E2 structure. In the crystal lattice, the I-EGF2 domain of 1 molecule of the asymmetric unit was free of crystal contact and thus had enough room to adopt an extended conformation (supplemental Figure 2A). Remarkably, the I-EGF2 domain of the other molecule of the asymmetric unit was trapped in the bent-like conformation by the close lattice contacts with its neighboring symmetry mates (supplemental Figure 2A). When superimposed with the bent native β₃ structure based on either the I-EGF1 or I-EGF2 domain (Figure 2A), the bent-like conformer observed in the cβ₃-E2 structure actually exhibits an intermediate state in between the fully bent and fully extended conformations (Figure 2A). Therefore, we captured a snapshot of β₃ integrin on the pathway from the bent to extended conformational transition.

Each of the integrin I-EGF domains contains 8 cysteines that form disulfide bonds in the pattern of C1-C5, C2-C4, C3-C6, and C7-C8, except the I-EGF1 domain⁵  (Figure 2B). The C1-C5 disulfide bond and C1-C2 (C473-C486 of β₃) loop of the I-EGF2 domain are pivotal in the transition from the bent to extended conformation at the I-EGF1/2 junction (Figure 2B-D). The C1-C2 loop of I-EGF2 was incompletely defined in the bent α_IIbβ₃ and 1 of the bent α_Vβ₃ structures as a result of the lack of complete electron density, but a complete backbone conformation was relatively well defined in another bent α_Vβ₃ structure (supplemental Figure 5A-C).^4,5,35  In our intermediate cβ₃-E2 structure, the strong electron density allowed us to build a complete conformation for both the backbone and side chains of the C1-C2 loop in the I-EGF2 domain (supplemental Figure 5D). We were also able to build a complete backbone conformation of the C1-C2 loop in the extended I-EGF2 domain, although some of the side chain conformations were not well defined because of the weak electron density (supplemental Figure 5E). Structural superimposition on the I-EGF2 domain revealed the extension of I-EGF1 afforded by the conformational rearrangement of the flexible C1-C2 loop and rotational movement at the C1 position (Figure 2B-D).

In the bent conformation, the Cα-Cα distance between the N-terminal C435 of I-EGF1 and the C-terminal E522 of I-EGF2 was 25.3 Å (Figure 2B). This distance changed to 33.6 Å in the intermediate conformation (Figure 2C) and to 45 Å in the extended conformation (Figure 2D). In coordination with these changes, the Cα-Cα distance between the C1 and C2 residues changed from 16.9 Å in the bent, to 14.9 Å in the intermediate, and to 13.8 Å in the extended states (Figure 2B-D). Along with the extension of the I-EGF1/2 junction, there was a rotational movement of the I-EGF1 domain using the C1-C5 disulfide bond as a hinge. There were large conformational rearrangements for both backbone and side chains of the I-EGF2 C1-C2 loop during the I-EGF1 extension (Figure 2B-D; supplemental Figure 4). The rotation of the C1-C5 disulfide bond of I-EGF2 accompanied with the swinging of the C1-C2 loop (Figure 2B-D). Interestingly, a small portion of the C1-C2 loop at the C2 end became α helical in the extended conformation (Figure 2D). No disulfide bond exchange was observed. All the disulfide bonds of I-EGF1 and I-EGF2 domains remained the same as in the bent conformation during the I-EGF1 extension (Figure 2B-D).

Two crystal structures of β₂ fragments were determined previously.³⁶  One contains the PSI, hybrid, and I-EGF1-2 domains and is bent at the I-EGF1/2 junction. The other contains an extra I-EGF3 domain and exhibits an extended I-EGF1/2 junction. When superimposed onto the I-EGF2 domain of the β₂ ectodomain structure in the bent conformation,³⁷  the bent β₂ fragment, in fact, showed an intermediate conformation, in which the I-EGF1 orientated close to the extended state (Figure 2E). Compared with our extended β₃ headpiece, the I-EGF1 in the extended β₂ fragment was in a less straight position relative to the I-EGF2 domain (Figure 2F). The Cα-Cα distance between I-EGF2 C1 (C461) and C2 (C475) was 15.2 Å, which was longer than that of the extended β₃ (Figure 2D). An α-helical conformation at the C2 end of the β₂ I-EGF2 C1-C2 loop presented at both the bent and intermediate states, but it became less α helical in the extended state (Figure 2E-F).

Previous crystallographic studies have determined the crystal structures of β₃ ectodomain in the bent conformation and β₃ headpiece (containing βI, hybrid, PSI, and I-EGF1 domains) either in the closed or open conformation.^4-6,38  Of note, EM studies have suggested that the integrin headpiece extends away from the leg domains on the cell surface, either remaining in the closed conformation, the same as in the bent state, or becoming an open configuration.^7,8,11  However, the conformations of I-EGF2-4 and β-tail domains likely remain unchanged after headpiece extension and leg separation, as shown by the EM structures. Our current cβ₃-E2 crystal structures revealed the extension of β₃ at the I-EGF1/I-EGF2 junction. Using our intermediate and extended I-EGF1/I-EGF2 structures as templates, and by superimposing the β₃ headpiece structure onto the I-EGF1 domain and superimposing the I-EGF2-4 and β-tail structures onto the I-EGF2 domain, we were able to build a complete high-resolution structure of β₃ ectodomain at both the intermediate and extended conformations (Figure 3A-D).

At the intermediate state, the β₃ headpiece extended halfway to the fully extended conformation (Figure 3A). In addition to the acute bent conformation (Figure 3Bi), the intermediate conformations of α_Vβ₃ (Figure 3Bii) and α_IIbβ₃ complexes (Figure 3Biii-iv) were observed by the EM studies.^5,7,11  At this intermediate state, the leg domains were still associated with each other, as in the bent conformation, and the headpiece remained closed. The structural model of the α_IIb subunit at the intermediate conformation was built by the structural superimpositions of the closed α_IIbβ₃ headpiece and the associated α_IIbβ₃ leg domains onto the modeled intermediated β₃ subunit with the closed headpiece (Figure 3B). Remarkably, this structural model of α_IIbβ₃ in the intermediate conformation agreed well with the intermediate EM structures (Figure 3B), demonstrating the reliability of our structure.

On the basis of the extended cβ₃-E2 structure, we built the extended structure of the β₃ ectodomain with either the closed (Figure 3C) or the open (Figure 3D) headpiece conformation. Similar to the intermediate conformation of β₃, the extended structures of β₃ fit near perfectly into the EM structures of extended β₃ with a closed or open headpiece (Figure 3E-F), again demonstrating the reliability of our structural models. We did not model the extended conformation of α_IIb ectodomain because of the lack of α_IIb and α_V crystal structures at the extended state.

The β₃ PSI domain has been shown to have some functional relevance.^39,40  Of interest, a Leu33Pro polymorphism at the β₃ PSI domain forms the HPA-1a/b alloantigen.^15,17  The β₃ PSI domain also contains the epitope of active conformation-specific mAb AP5,⁴¹  indicating its conformational changes upon β₃ activation. Leu33Pro locates at the C26-C38 loop of the PSI domain (Figure 4A-B). Previous studies have suggested that β₃ harboring Pro33 could be a risk factor for thrombosis because of the elevated function of α_IIbβ₃ in platelets, but the conclusions are controversial.^18-20  Our structures of the β₃ headpiece in both Leu33 and Pro33 forms as well as in the intermediate and extended states provide new structural information of the PSI domain.

Structural superimposition on the I-EGF1 domain showed that the overall conformations of the PSI and I-EGF1 domains were nearly identical between the intermediate and bent states of β₃ (Figure 4A). Interestingly, in the extended conformation, the C26-C38 loop of the PSI domain moved close to the I-EGF1 domain (Figure 4A). In addition, the C437-C448 loop of I-EGF1 domain also moved toward the PSI C26-C38 loop in the extended conformation (Figure 4A). These structural characters were observed in the cβ₃-E2 structures with either Pro33 or Leu33 (supplemental Figure 4C-D), demonstrating that Leu33Pro substitution does not affect the local conformation of PSI domain.

The sequences of the C26-C38 loop are relatively conserved among different species, but variations are seen at the positions equivalent to β₃-Leu33 (Figure 4B). We examined whether the Leu33Pro polymorphism affects the activation of β₃ integrin. Ligand mimetic mAb PAC-1 was used to measure the activation of α_IIbβ₃ stimulated either by Mn²⁺ or by the cytoplasmic activating mutation α_IIb-R995A that mimics inside-out activation.⁴²  No differences in PAC-1 binding were observed between β₃-Leu33 and β₃-Pro33 (Figure 4C). Similarly, no difference was observed in PAC-1 binding to α_IIbβ₃-Leu33 and α_IIbβ₃-Pro33 induced by the overexpression of EGFP-tagged talin head domain (Figure 4D). Moreover, no difference was found between β₃-Leu33 and β₃-Pro33 in Mn²⁺-stimulated fibronectin binding to α_Vβ₃ (Figure 4E). These data demonstrate that Pro33 does not render β₃ integrin more active than the Leu33 form.

The β₃ PSI domain connects with the hybrid domain through a short loop (Figure 5A). During the headpiece extension and hybrid domain swing-out, the PSI domain moves along with the hybrid domain (supplemental Figure 1C). We found no significant structural differences at the PSI/hybrid interface between the bent and extended β₃ headpieces (Figure 5A-B). The side chain of Arg93 at the hybrid domain A’B loop inserts into the hybrid/PSI interface and forms multiple hydrogen bonds with the backbone carbonyl oxygens of the PSI domain and the hybrid-PSI connecting loop (Figure 5A). These hydrogen bonds are maintained in the extended β₃ conformation (Figure 5B) and in the open headpiece of β₃.⁶  Such hydrogen bond interactions may be critical in maintaining the rigidity of the hybrid/PSI interface. Of interest, the equivalent Arg is completely conserved among all the integrin β subunits (Figure 5C). Mutagenesis studies were performed to investigate the function of this conserved Arg in integrin activation. As shown by the ligand-binding assay, β₃-R93A and β₃-R93Q mutations remarkably augmented Mn²⁺-stimulated PAC-1 binding to α_IIbβ₃ integrin, although the expression level was also greatly decreased (Figure 5D). Basal PAC-1 binding under the physiological Ca²⁺/Mg²⁺ concentration was also increased by β₃-R93 mutations (Figure 5D). Consistently, the β₃-R93A mutation greatly enhanced fibronectin binding to α_Vβ₃ (Figure 5E). Similar results were obtained for β₂ and β₁ integrins. The β₂-R87A mutation greatly enhanced ICAM-1 binding to α_Lβ₂ (Figure 5F). β₁-R104A greatly enhanced fibronectin binding to α₅β₁ (Figure 5G). Consistent with the fibronectin-binding assay, the β₁-R104A mutation enhanced both basal- and Mn²⁺-stimulated binding of 9EG7 to α₅β₁ (Figure 5H), demonstrating that disruption of the hybrid/PSI interface induces the extension of β₁ integrin.

Our high-resolution structural studies reveal how β₃ becomes extended at the I-EGF1/2 junction. Our data suggest that the multiple conformational states of β₃ observed in many EM studies^5,8-11,43-45  could be afforded by the conformational plasticity of the I-EGF2 C1-C2 loop. Previous studies have demonstrated that the length of the I-EGF2 C1-C2 loop modulates β₃ integrin activation.⁴⁶  The long and short loops favor the resting and active states, respectively.⁴⁶  The C1-C2 loop of I-EGF2 was the longest among the I-EGF domains. The I-EGF3 C1-C2 loop had 6 residues, whereas the I-EGF4 C1-C2 loop had 4 residues. The angle between I-EGF3 and I-EGF4 was closer to straight than the others (Figure 2G-H). The C1-C2 loop of β₂ I-EGF2 had 13 residues (Figure 2F). Consistently, the β₂ I-EGF1/2 junction was less extended than that of β₃, which had 12 residues spanning the C1-C2 loop (Figure 2F). These structural features of I-EGF domains are compatible with the long-range conformational changes of integrin. The rigidity of I-EGF2/3 and I-EGF3/4 interfaces provides a support for the extension of the integrin headpiece and for the resistance of mechanical force.⁵  The flexibility of the I-EGF1/2 interface allows the headpiece to pose at different conformations, which is required for affinity regulation. Given the critical role of the I-EGF2 C1-C2 loop in integrin extension, an intriguing strategy would be to target the C1-C2 loop, such as by antibodies, which may modulate integrin conformational change and cellular function.

Our structural studies have important implications for integrin conformational activation. Integrin headpiece extension is afforded by the extension of the β subunit at the I-EGF1/2 junction (β knee) and the extension of the α subunit at the thigh/calf-1 junction (α knee; Figure 3B-F; supplemental Figure 1). Because of the large interface formed between the β-propeller and βI domains, the α and β extensions need to occur simultaneously. Our high-resolution structures of β₃- knee at intermediate and extended conformations demonstrate the flexibility of this region. Extension of the I-EGF1 domain did not generate new interactions at the I-EGF1/2 junction (Figure 3D). Indeed, only a very small interface formed at the β₃ I-EGF1/2 junction, even in the bent conformation (Figure 3B). By contrast, the thigh/calf-1 junction in bent α_IIb formed a larger interface composed of hydrogen bonds and hydrophobic interactions (Figure 3G). This suggests that the bent α_IIb knee may resist the force of extension more than β₃ knee. EM and our modeled structure show that the α_IIb and α_V subunits can adopt an intermediate (half-extended) conformation (Figure 3B). After complete extension, α knee becomes straight and the thigh and calf-1 domains are expected to form new close contacts at their junction. Unlike extended β₃ knee, the new interface of extended α knee is expected to be rigid. These structural features are important for the conformational regulation of integrin by cytoskeletal force under conditions of cell adhesion and migration.^5,47,48  The flexibility of an extended β subunit afforded by the flexible I-EGF1/2 junction is required to sense and resist the changes of cellular forces, whereas the rigidity of an extended α leg may provide support for the sensing and resisting of tensile force by the β subunit.^47,49-51 

Local conformational changes were observed at the PSI and I-EGF1 domains upon β₃ extension. It is not readily known whether such conformational changes are functionally significant. Nonetheless, structural diversities of the PSI and I-EGF1 domains, especially at the C26-C38 loop, have been observed among the reported β₃ crystal structures (A.M.M.T., Z.W., D.Z., Y.Z., Brian R. Curtis, and J.Z., manuscript submitted April 2018). Moreover, the biological relevance of such conformational diversities is implicated by the binding of heterogeneous anti–HPA-1a alloantibodies. All anti–HPA-1a alloantibodies require Leu33 at the β₃ PSI C26-C38 loop.^16,17  However, some anti–HPA-1a alloantibodies require noncontiguous sequences and the integrity of the PSI domain to form the HPA-1a epitope.^52-56  In particular, certain anti–HPA-1a alloantibodies require both the PSI and I-EGF1 domains.^52,56,57  We have unpublished data demonstrating that the specific residues of the I-EGF1 domain are required for the binding of certain anti–HPA-1 alloantibodies. Of significance, recent studies have demonstrated that a subtype of maternal anti–HPA-1a alloantibodies could distinguish the α_IIbβ₃ vs α_Vβ₃ complex.⁵⁸  α_Vβ₃-specific anti–HPA-1a alloantibodies were indicated to be the cause of intracranial hemorrhages in fetus and neonates through the blockage of the α_Vβ₃ function in endothelial cells.⁵⁸  We have data showing that α_IIbβ₃-specific anti–HPA-1a alloantibodies can specifically recognize the bent but not extended conformation of β₃ (A.M.M.T., Z.W., D.Z., Y.Z., Brian R. Curtis, and J.Z., manuscript submitted April 2018). In addition, a recent study showed that EDTA treatment of α_IIbβ₃, known to disturb α_IIbβ₃ conformation,^41,59-61  reduced the binding of anti–HPA-1a.⁶²  All these data suggest the conformation-dependent feature of anti–HPA-1a alloantibodies. The conformational changes of the PSI and I-EGF1 domains observed in our extended β₃ structure may account for the differences in anti–HPA-1a binding.

It is remarkable that the anti–HPA-1a alloantibodies could distinguish the subtle conformational differences of the PSI and I-EGF1 domains. More interestingly, some but not all anti–HPA-1a alloantibodies were found to block the function of α_IIbβ₃ and/or α_Vβ₃ integrins.^58,63  Our current studies provide a structure-based interpretation for this observation. The conformational differences of the PSI and I-EGF1 domains may determine the binding poses of the alloantibodies, which in turn exert either a blocking or nonblocking effect on β₃ function. In particular, the bent conformation-dependent anti–HPA-1a alloantibodies may block the integrin function by stabilizing β₃ in the bent conformation. This raises an intriguing possibility that the inhibitory antibodies targeting the PSI and I-EGF1 domains may be developed to specifically target α_IIbβ₃ or α_Vβ₃ integrin.

Polymorphism substitution β₃-R93Q has been shown to disrupt the binding of certain anti–HPA-1a (β₃-Leu33) alloantibodies.⁶⁴  β₃-R93 is more than 28-Å distal from β₃-Leu33 and is buried into the hybrid/PSI interface, and thus cannot directly participate in the HPA-1a epitope. Here, we found that β₃-R93 mutations induced the active conformation of β₃ integrin, which may have led to the structural changes of the PSI C26-C38 loop and the I-EGF1 domain as observed in our extended β₃ structure (Figure 4A). As discussed, such conformational changes may alter the binding of anti–HPA-1a alloantibodies.

We found no differences in either conformation or ligand binding between the Leu33 and Pro33 forms of β₃. Studies using human platelets or cell lines have shown that β₃-Pro33 enhanced cell adhesion and outside-in signaling.^18,65-68  A recent study using knock-in mice showed the Pro33 form of mouse β₃ increased platelet adhesion, clot formation, and aggregation, probably as a result of elevated Src activation.⁶⁹  However, agonist-induced soluble ligand binding was not affected, consistent with our data. Interestingly, the talin association with mouse β₃ was also increased by the presence of Pro33.⁶⁹  Because Pro33 does not directly alter the conformation of the PSI domain (Figure 4A), it remains unknown by which mechanism Pro33 affects β₃ function. Molecular dynamics simulation studies have indicated that Pro33 substitution might affect β₃ function by modifying the structural equilibrium of the PSI and the I-EGF domains and probably integrin extension,^70,71  but such an effect may not directly affect ligand binding according to our study.

The authors thank Peter Newman for helpful discussions and for providing antibodies.

This work was supported by grants HL122985 and HL131836 (J.Z.) from the National Heart, Lung, and Blood Institute of the National Institutes of Health.

Contribution: D.Z. collected X-ray diffraction data and refined the structures; A.M.M.T. performed functional assays and collected X-ray diffraction data; Y.Z. contributed to protein purification; Z.W. contributed to the generation of DNA constructs; J.Z. designed the study, did protein crystallography, and wrote the manuscript; and all authors contributed to manuscript preparation.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Jieqing Zhu, Blood Research Institute, BloodCenter of Wisconsin, 8727 Watertown Plank Rd, Milwaukee, WI 53226; e-mail: jieqing.zhu@bcw.edu.