An αIIbβ3 monoclonal antibody traps a semiextended conformation and allosterically inhibits large ligand binding

Monoclonal antibodies (mAbs) have provided valuable information regarding the structure and function of platelet integrin αIIbβ3. A number of mAbs inhibit ligand binding via steric hindrance by interacting with sites at, or near, the ligand binding site in the head region comprising the αIIb β-propeller domain and β3 I domain^1-4 or serving as ligand mimetics;^5-7 among the latter is 1 that only binds to platelets after platelet activation,⁵ whereas the others do not require platelet activation.^6,7 An mAb that binds to the plexin-semaphorin-integrin (PSI) domain of the β3 subunit partially inhibits ligand binding, perhaps by dimerization of the receptor, thus limiting access to large ligands,⁸ or by inhibition of the endogenous thiol isomerase-like activity of αIIbβ3.⁹ 

Other mAbs have been reported to bind selectively to conformation(s) that are induced by ligand binding to the receptor, and thus have been grouped together under the title ligand-induced binding site (LIBS) mAbs.^10-12 Most LIBS antibodies are directed against the β3 subunit, which is consistent with the more dramatic changes in its conformation upon ligand binding compared with αIIb.⁴ Similarly, most LIBS antibodies recognize conformational changes in the ectodomain of the receptor, but 1 has been reported to recognize a change in the cytoplasmic domain of αIIb,¹³ highlighting the ability of the integrin to transmit signals from outside to inside in addition to inside to outside. Because these antibodies also stabilize the activated conformation(s), which may be adopted spontaneously due to thermal motion, some can prime the receptors to bind ligands.^12,14,15 Another mAb can prime the receptor to bind ligand by binding to the αIIb β-propeller domain and preventing the receptor from adopting an inactive, bent conformation.¹⁶ 

Protein disulfide isomerase (PDI) is a thiol oxidoreductase that binds to αIIbβ3 and has been implicated in αIIbβ3 activation and platelet-mediated thrombosis.^17-21 Because the binding site for PDI on αIIbβ3 is unknown, we sought to identify a murine anti-αIIbβ3 mAb that can inhibit PDI binding to activated αIIbβ3. In the process, we identified a new mAb that prevents the binding of PDI, but also prevents the binding of ligands that bind to the arginine-glycine-aspartic acid (RGD)-binding pocket via an allosteric mechanism that traps a semiextended conformation of the receptor. Here, we describe the effects of the mAb on αIIbβ3 function and cryogenic electron microscopy (cryo-EM) structure of its Fab fragment bound to αIIbβ3.

Reagents, mAbs, and plasmids are detailed in the supplemental Materials.

Human platelet studies were performed according to a protocol approved by The Rockefeller University Institutional Review Board and in accordance with the Declaration of Helsinki. Washed platelets were prepared from blood anticoagulated with acid citrate dextrose as described previously.²² For studies on platelet-rich plasma (PRP), blood was anticoagulated with 3.2% sodium citrate and prepared as previously described.²² 

Balb/C mice were immunized with human platelets, and mAbs were prepared using the standard techniques described in detail in the supplemental Materials.

The interaction of the mAbs with αIIbβ3 was tested by enzyme-linked immunoassay (ELISA) with purified human αIIbβ3 protein (Innovative Research) and is described in the supplemental Materials.

Washed platelets (2 × 10⁸/mL) were preincubated with hybridoma culture supernatants or purified mAb for 15 minutes at 22°C and then were either left untreated or treated with 25 μM T6 for 10 minutes at 22°C. Alexa Fluor 488–labeled PDI (1 μM) was added and incubated for 15 minutes at 22°C. The platelets were diluted 40-fold with phosphate-buffered saline (10 mM Na₂HPO₄, 1.8 mM KH₂PO₄, 2.7 mM KCl, 150 mM NaCl, pH 7.4) and analyzed immediately by flow cytometry (BD Accuri C6 Plus). To determine the nonspecific binding of PDI, the binding of labeled PDI was assessed in the presence of 10 μM of unlabeled PDI. The measurement of Alexa Fluor 488–labeled fibrinogen (200 μg/mL) binding was performed as for the labeled PDI.

Fibrinogen binding to primed platelets was performed as previously described^23,24 and is described in the supplemental Materials.

Washed platelets or PRP were incubated with mAbs for 15 minutes at 22°C and aggregation was induced by adding T6 or adenosine 5'-diphosphate (ADP; supplemental Materials).

Immunoblotting was performed as described previously²⁵ and the detailed method is provided in the supplemental Materials.

Platelet αIIbβ3 was purified according to a modification of the method described previously (supplemental Materials).²⁶ 

αIIbβ3 purified from platelets was mixed with R21D10 Fab at a 1:4 molar ratio in buffer containing 20 mM HEPES (N-2-hydroxyethylpiperazine-N′-2-ethanesulfonic acid), 150 mM NaCl, 2 mM CaCl₂, and 1 mM MgCl₂ (pH 7.4), incubated for 1 hour at 4°C, and gel-filtered on a Superdex Increase 200 column. Fractions containing the αIIbβ3-R21D10 Fab complex were identified by sodium dodecyl sulfate–polyacrylamide gel electrophoresis and imaged by negative-stain EM using a Philips CM10 transmission electron microscope.

Cryo-EM sample preparation, data collection and processing, and model building and refinement were performed as previously described^3,16 with minor differences, as detailed in the supplemental Materials.

The 874 hybridoma culture supernatants were screened using flow cytometry for their ability to inhibit PDI binding to T6-activated platelets. A total of 97 supernatants inhibited PDI binding, of which 40 were positive for binding to αIIbβ3 using the ELISA. We selected antibody R21D10 for further analysis because of the consistency of its strong inhibition of PDI binding to T6-activated platelets, reducing the binding of labeled PDI to the level found when labeled PDI was tested in the presence of 10-fold excess unlabeled PDI (Figure 1A). R21D10 partially inhibited the binding of fibrinogen and the αIIbβ3 ligand–mimetic mAb PAC-1 to T6-activated platelets, but was much less potent than mAb 7E3 (Figure 1B-C), which binds to β3 near the RGD-binding pocket.⁹ Neither R21D10 nor 7E3 affected P-selectin expression, indicating that they did not prevent the release reaction induced by T6 (Figure 1D). R21D10 inhibited T6-induced aggregation of washed platelets in a concentration-dependent manner, with the highest concentration (40 μg/mL) inhibiting maximal platelet aggregation by 53 ± 19% (mean ± standard deviation; Figure 1E). R21D10 also inhibited the ADP-induced aggregation of PRP in a dose-dependent manner (Figure 1F).

The Fab fragment of R21D10 potently inhibited T6-induced fibrinogen binding (supplemental Figure 1A), and at 40 μg/mL, partially inhibited both T6-induced aggregation of washed platelets (supplemental Figure 1B) and ADP-induced aggregation of PRP (supplemental Figure 1C) to the same extent as intact R21D10 at the same mass concentration.

Antibodies 10E5 and 7E3, which inhibit fibrinogen binding to αIIbβ3 and bind near the RGD-binding pocket, did not affect the binding of R21D10 to unactivated platelets (Figure 2A), and reciprocally, R21D10 did not inhibit the binding of either 10E5 or 7E3 to unactivated platelets (Figure 2B-C). Intact R21D10 immunoglobulin G (IgG) partially inhibited Alexa Fluor 488–labeled 7E3 binding to T6-activated platelets, whereas R21D10 Fab did not (Figure 2D). mAb 7H2, which binds to the PSI domain of β3, did not affect the binding of R21D10 to unactivated platelets (Figure 2E) and vice versa (Figure 2F).

R21D10 reacted by immunoblotting of a platelet lysate with a band that was also recognized by the anti-β3 PSI domain antibody 7H2, but not the band detected by the anti-αIIb mAb PMI-1 (supplemental Figure 2A). However, R21D10 did not bind to β3 after disulfide bond reduction (supplemental Figure 2A). R21D10 also reacted with chymotrypsin fragments of β3, including those with molecular weight <66 kDa (supplemental Figure 2B). There was no obvious reduction in the total intensity of R21D10 staining, even after overnight digestion of αIIbβ3 with chymotrypsin.

EDTA treatment of platelets at 22°C and pH 7.4 alters the conformation of αIIbβ3 by chelating divalent cations but does not dissociate αIIb from β3, whereas EDTA treatment at 37°C and higher pH leads to dissociation of the subunits.¹⁴ Treatment of platelets with 10 mM EDTA at 22°C had no effect on R21D10 binding, nor did it affect the binding of mAb 10E5 (Figure 3A-B). Dissociating αIIb from β3 with 10 mM EDTA at 37°C and pH 8.0 did not affect R21D10 binding, whereas, as we previously reported,²⁷ it dramatically decreased 10E5 binding (Figure 3C-D).

To assess whether the epitope for R21D10 is readily accessible to unactivated platelet αIIbβ3, we assessed the binding kinetics of R21D10 and found that the binding rate of R21D10 was similar to that of mAb 10E5, which is readily accessible on the αIIb cap domain but more rapid than the binding of mAb 7E3, which binds to a partially activation-dependent epitope primarily on β3 (supplemental Figure 3).¹⁵ 

Negative-stain EM images of full-length αIIbβ3 showed a monodispersed particle population, with particles varying in size and shape (supplemental Figure 4). After picking and classifying ∼10 000 particles into 100 classes, the averages showed that the integrin in conformations varied from the compact, V-shaped bent-closed conformation to the extended-closed conformation (supplemental Figure 4B-C). We then imaged the αIIbβ3-R21D10 Fab complex and observed an increase in the size of the particles compared with that of αIIbβ3 alone (supplemental Figure 4D). We again classified ∼10 000 particles into 100 classes (supplemental Figure 4E-F). Several of the averages showed an integrin in the canonical bent-closed V-shaped conformation but with an additional density extending from the tip of the V, which is the junction between the upper and lower legs, that we presumed represented the bound R21D10 Fab. Other averages showed the integrin in a more extended conformation with an additional density, presumably representing the bound R21D10 Fab near the junction between the upper and lower legs. The densities representing the Fab and the αIIbβ3 headpiece were clearly defined in most of these averages but the density of the integrin lower legs was often blurred. Collectively, these results showed that the R21D10 Fab binds to αIIbβ3 near the lower leg region of the integrin and appears to be able to bind to the integrin in its bent-closed conformation as well as in a somewhat extended conformation.

We vitrified the αIIbβ3-R21D10 Fab complex and imaged the sample using a K3 direct electron detector on a Titan Krios electron microscope (supplemental Figure 5). After image processing, we determined the structures of the αIIbβ3-R21D10 Fab complex with the integrin adopting either a semiextended or a bent conformation at resolutions of 3.3 Å and 4.4 Å, respectively (Figure 4; supplemental Figures 5-7; supplemental Table 1). The cryo-EM structures of the αIIbβ3-21D10 Fab complex confirmed that R21D10 indeed binds to αIIbβ3 in 2 different conformations that differ in the arrangement of the leg domains, with the legs partially extended in 1 and bent in the other.

The headpiece region structures, including the αIIb propeller and thigh domains and the β3 PSI, hybrid, and β-I domains, are very similar in the 2 conformations with a root-mean-square deviation between the Cα atoms of 1.1 Å, and both are very similar to the same region in the previously reported crystal structure (Protein Data Bank [PDB]: 3FCS) and cryo-EM structure of αIIbβ3 (PDB: 8T2V) with root-mean-square deviations compared with our bent conformation of 1.4 Å and 1.0 Å, respectively. In sharp contrast, the αIIb and β3 knee regions of the head and legs adopted very different conformations (Figure 4A-C). The map of the complex with αIIbβ3 in the semiextended conformation had sufficient resolution to establish that R21D10 Fab primarily interacts with the integrin-epidermal growth factor 1 (I-EGF1) domain through its heavy chain (Figure 4A-C). Specifically: (1) Asp31 in the heavy chain CDR1 binds to the β3 Arg447 in the I-EGF1 domain through an electrostatic interaction, (2) Thr55 in the heavy chain CDR2 interacts with Gln440 in I-EGF1 via a hydrogen bond, and (3) Arg104 in the heavy chain CDR3 interacts with Glu472 in the I-EGF1 domain via charge interaction (Figure 4C, left panels). The heavy chain is also very close to the PSI domain and may make good contact, but the resolution was insufficient to assign specific atomic interactions. Although the resolution of the map was not sufficient to be certain, we presumed that most, if not all, these interactions were the same when R21D10 Fab was bound to αIIbβ3 in the bent conformation.

Notably, our 2 cryo-EM structures show that the adjacent β3 I-EGF2 domain adopts 2 drastically different conformations. This domain occupies a crucial position because of its proximity to the αIIb thigh domain (Figure 4C). A review of the crystal structures of both αIIbβ3 and αVβ3 in the bent conformation identified an interaction between I-EGF2 Glu500 and thigh domain Lys514 in αIIbβ3 (PDB: 3FCS) and the equivalent Lys and Glu residues in αVβ3 (PDB: 3IJE); the reported cryo-EM structure also shows these regions close to each other, but the resolution was insufficient to assign a specific interaction (PDB: 8T2V) (supplemental Figure 8). This interaction has not been highlighted in the literature but has the potential to provide intersubunit stability. The resolution of our cryo-EM structure of the bent conformation in complex with R21D10 Fab (4.4 Å) is inadequate to unequivocally identify the same interaction, but because our structure is very similar to both the crystal and cryo-EM structures in this region, it is likely that it is also present in our cryo-EM structure of the bent conformation. In sharp contrast, in the semiextended conformation, this interaction was lost due to a major change in the conformation of the I-EGF2 472-486 loop, which led to the creation of an interaction between I-EGF2 Glu475 and Arg516 in the thigh domain (Figure 4C, right panels). There were minimal changes in the thigh domain, but there was a reorganization of the Lys514-Arg516 segment (Figure 4). Supplemental Figure 7 provides a view of each conformation with the bound R21D10 Fab, as they are likely to be oriented in the platelet membrane.

To further explore the possibility that R21D10 traps αIIbβ3 in a semiextended conformation, we assessed the effect of altering the conformation of αIIbβ3 on R21D10 binding and the ability of R21D10 binding to affect the binding of a LIBS mAb (AP5), which demonstrates low-level binding to unactivated platelets and much higher binding to activated αIIbβ3.²⁸ The αIIbβ3 antagonist eptifibatide induces an extended-open conformation in αIIbβ3 and thus increases AP5 binding.²⁹ 

AP5 did not affect R21D10 binding to platelets, either in the absence or presence of eptifibatide (Figure 5A-B), indicating that neither AP5 nor eptifibatide inhibited the binding of R21D10. In sharp contrast, R21D10 essentially abolished the small amount of AP5 binding to unactivated platelets (Figure 5C) and partially inhibited AP5 binding to platelets pretreated with eptifibatide to induce the extended-open conformation (Figure 5D). These data suggest that AP5 cannot react with the conformation trapped by R21D10, that R21D10 can shift the subpopulation of αIIbβ3 that is reactive with AP5 in untreated platelets to the trapped conformation, and that R21D10 can bind to and induce eptifibatide-treated αIIbβ3, which is in the extended-open conformation, to partially adopt the trapped conformation. We also assessed the effect of adding R21D10 before adding eptifibatide (Figure 5E) and found that, although there was an increase in AP5 binding, indicating that R21D10 did not prevent eptifibatide binding, the increase in AP5 binding was less than that in the absence of R21D10, indicating that R21D10 limited the effect of eptifibatide.

We assessed whether R21D10 could partially reverse the ability of eptifibatide to prime αIIbβ3 to bind to fibrinogen. We found that when R21D10 was added before fixation, it inhibited fibrinogen binding to eptifibatide-primed platelets by 73 ± 3% (Figure 6A,C). In contrast, when R21D10 was added after fixation, it produced much less (23 ± 11%) inhibition of fibrinogen binding (Figure 6B-C).

We tested whether the new interaction between the thigh and I-EGF2 domains associated with R21D10 binding could stabilize the interaction between the 2 subunits by assessing whether R21D10 binding could prevent the dissociation of αIIbβ3 by EDTA at pH ∼8.0 and 37°C. The data in Table 1 show that R21D10, unlike mAb 7H2, can partially protect αIIbβ3 from EDTA-induced dissociation, using the loss of binding of 10E5 as an indicator of dissociation.²⁷ 

We report a new mAb to αIIbβ3 (R21D10) that is a partial inhibitor of large ligand binding to the RGD pocket on the receptor induced by multiple agonists, binds to the β3 I-EGF1 domain, and traps a semiextended conformation of the receptor. To our knowledge, this is the first mAb reported to inhibit the activation-induced binding of PDI to platelets.

PDI binding requires platelet activation, which may occur via a change in the conformation of the αIIbβ3 receptors, exposure to additional αIIbβ3 receptors, or a combination of these effects. The increase in platelet fluorescence intensity with the binding of fibrinogen and PAC-1 was greater than that with the binding of PDI, which is consistent with reported data on PDI binding to human and murine platelets stimulated with thrombin.^19-21 As with large ligand binding, it is possible that the PDI binding site is inaccessible in the semiextended conformation trapped by R21D10. Alternatively, R21D10 may inhibit PDI binding by steric hindrance given that the I-EGF domains are rich in disulfide bonds and may be a target of PDI. Thus, additional research on PDI binding is required.

R21D10 binds at the junction of the α and β knee domains, which in the bent integrin is opposite to the ligand-binding head region; thus, it acts as an allosteric inhibitor of ligand binding rather than steric inhibition. The I-EGF1 and I-EGF2 domains make a sharp angle at the knee, at which point the thigh domain “knob” approaches the I-EGF1 domain. Zhu et al suggested that the knob is important because it is conserved in αVβ3.³⁰ In fact, although not highlighted in the literature, we identified an interaction between the thigh domain Lys514 and the I-EGF2 Glu500 in the αIIbβ3 crystal structure and a similar interaction between the analogous Lys in the αV thigh domain and Glu500 in β3 (supplemental Figure 8). These interactions may provide intersubunit stability and prevent the swing-out of the hybrid domain. This interacting pair is unique to the β3 family of integrins (supplemental Figure 9) and is highly conserved among αIIb and β3 in different species. Thus, it is possible that this interaction plays an important role in determining which α-integrin subunits can pair with the β3 subunit. The αIIb thigh domain loop containing Lys514 also contains Arg516, which makes contact with β3 Glu475 after the binding of R21D10. Similar loops were present in only 10 of the18 integrin α subunits and they showed variable lengths and sequences.

The cryo-EM structure of the αIIbβ3-R21D10 Fab complex identified a novel, semiextended conformation of the integrin, which made up ∼75% of the particles and allowed the determination of a high-resolution structure (3.3 Å). This conformation has not been described before and is associated with decreased large ligand binding. The antibody could limit ligand binding by preventing the complete extension of the legs, thus limiting access to the RGD-binding site. Alternatively, or additionally, if the reorganization of the I-EGF2 loop and new interaction between the thigh and I-EGF2 domains hold αIIb and β3 together (as suggested by our data indicating that R21D10 diminishes the ability of EDTA to dissociate the subunits), then it might also prevent the swing-out of the β3 hybrid domain required for high-affinity ligand binding.

We previously reported the impact of preventing swing-out by creating disulfide bonds between the αIIb β-propeller (Lys321) and the β3 hybrid domain (either Glu358 or Glu360), and the consequences were similar to those observed with R21D10.²³ In particular, reduced binding of both PAC-1 and fibrinogen when the HEK293 cells expressing the recombinant mutant proteins were stimulated with the activating mAb PT25-2, or even when the mutations were combined with mutations that result in constitutively active receptors.²³ Kamata et al engineered a similar disulfide bond between αIIb Asp319 and β3 Val359 to prevent the swing-out motion and their mutant receptor also reduced fibrinogen binding when activated with PT25-2 that could not be overcome by mutations to induce receptor extension.³¹ Finally, because R21D10 inhibits PDI binding, either through an allosteric mechanism or by direct steric hindrance, it is possible that the reduction in ligand binding is a consequence of the reduction in PDI binding and the loss of the positive effect of PDI on αIIbβ3 activation.^18,19,21 

Our kinetic studies indicated that the 21D10 epitope is freely available on αIIbβ3, which is in accordance with our cryo-EM studies, indicating that the amino acids responsible for R21D10 binding (β3 Gln440, Arg447, and Glu472) are solvent exposed, and our identification of a complex between R21D10 and the αIIbβ3 bent conformation. Our data are consistent with a model in which the binding of R21D10 destabilizes the bent conformation, as suggested by the lower resolution of the cryo-EM structure, and traps the semiextended conformation, as suggested by the dominance of particles in this conformation and higher resolution of the cryo-EM map. It is possible that the semiextended conformation is an intermediate on the path to full αIIbβ3 activation, but it may be unique to changes associated with the binding of the antibody.

Ramsamooj et al reported a murine mAb (CS-1) with properties similar to those of R21D10^32,33 Thus, it bound to nonreduced, but not reduced β3, as well as similar chymotryptic fragments of β3; partially inhibited fibrinogen binding to activated platelets; and partially inhibited platelet aggregation. They localized the binding site to β3 349-422, which is near but does not overlap the β3 Gln440, Arg147, and Glu472 to which R21D10 binds. The authors did not provide any structural data regarding CS-1 binding.

In conclusion, we report a new murine mAb with a unique allosteric mechanism of inhibition of ligand binding to the RGD pocket, which may or may not be connected to its ability to inhibit the binding of PDI to platelets. It traps an intermediate conformation of the integrin in which the legs are partially extended. This conformation also involves a new interaction between the αIIb thigh domain and the I-EGF2 domain, which is not present in the crystal structure of the bent αIIbβ3 receptor. To our knowledge, this has not been previously reported. Thus, R21D10 may inhibit the large ligand binding to αIIbβ3 by limiting the extension and/or swing-out of the hybrid domain. Inhibition of PDI binding may or may not contribute to reduced binding of large ligands. Studies by Li and Springer and Li et al on other integrin receptors have established the value of mAbs that can stabilize the receptors in different conformations.^34,35 They demonstrated their value as crucial tools in establishing the thermodynamics of the activation mechanism, including the conclusion that ligands to other integrins have faster on-rates in the bent and extended-closed conformations than in the extended-open conformation, but their overall affinity is much higher to the extended-open conformation because the off-rate to that conformation is profoundly slower. Whether this is also true for the binding of large ligands to αIIbβ3 needs to be assessed. R21D10 thus provides a valuable new tool for probing the dynamics of αIIbβ3 activation and ligand binding.

The authors thank Frances-Weis Garcia and Louis Mattera for their expertise in producing the monoclonal antibodies described in this manuscript, which were developed at The Rockefeller University and Memorial Sloan Kettering Cancer Center Bi-institutional Antibody and Bioresource Core Facility (Research Resource Identification: SCR_01769). The authors also thank Mark Ebrahim, Johanna Sotiris, and Honkit Ng for cryo-EM data collection and the Cryo-Electron Microscopy Resource Center of The Rockefeller University for cryo-EM work. The authors also extend thanks to Lorena Buitrago and Julio Padovan for their support throughout this study and Suzanne Rivera for outstanding administrative assistance.

This study was supported in part by National Heart, Lung, and Blood Institute grant HL19278, National Center for Advancing Translational Sciences of the National Institutes of Health grant UL1 TR001866, and a pilot project from the Shapiro-Silverberg Fund for the Advancement of Translational Research at The Rockefeller University.

Contribution: L.W. designed, performed, and analyzed platelet experiments, and wrote the manuscript; J.W. designed, performed, and analyzed EM experiments and wrote the manuscript; J.L. purified antibodies and designed and performed the amino acid sequence analysis of the antibody; T.W. designed, oversaw, and analyzed EM studies, and wrote the manuscript; and B.S.C. designed and oversaw the study, analyzed data, and had primary responsibility for writing the manuscript.

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

Correspondence: Barry S. Coller, Laboratory of Blood and Vascular Biology, The Rockefeller University, 1230 York Ave, New York, NY 10065; email: collerb@rockefeller.edu.