Structural basis for the specific inhibition of glycoprotein Ibα shedding by an inhibitory antibody

Ectodomain shedding of glycoprotein (GP) Ibα is thought to mediate the clearance of activated, aged or damaged platelets. A monoclonal antibody, 5G6, has been developed recently to specifically bind to the GPIbα shedding cleavage site and to inhibit its shedding. However, the molecular mechanism underlying antigen recognition and inhibitory specificity is not clear. To elucidate the structural basis for 5G6 binding to GPIbα, we determined the crystal structure of 5G6 Fab fragment in complex with its epitope peptide KL10 (GPIbα residues 461–470, KLRGVLQGHL), to 2.4-Å resolution. Key residues in both 5G6 and KL10 were mutated to validate their effects in antibody binding by using isothermal titration calorimetry. The 5G6 Fab-KL10 peptide complex structure confirmed the direct association of 5G6 with its target GPIbα residues and elucidated the molecular basis underlying its binding specificity and high affinity. The similar binding properties of 5G6 Fab fragment to GPIbα on human platelets as those to KL10 suggests that such an interaction may not be affected by the plasma membrane or nearby GPIbβ. This structural information may facilitate further antibody optimization and humanization.

Scientific RepoRts | 6:24789 | DOI: 10.1038/srep24789 GPIb-IX complex and inhibited GPIbα shedding with a similar potency as that of GM6001 without affecting the shedding of other receptors, platelet activation or aggregation 19 .
To elucidate the structural basis underlying antigen recognition and the high binding affinity of 5G6 for GPIbα , we solved the crystal structure of 5G6 Fab in complex with its epitope peptide KLRGVLQGHL (hGPIbα residues 461-470, short for KL10) at 2.4-Å resolution. Comparison of our structure with the structures of other Fab fragments revealed vital features of antigen recognition by 5G6 Fab, which were further confirmed by mutagenesis of 5G6Fab and KL10. In summary, we identified the features of the high binding affinity of 5G6 Fab for KL10 and shed light on the mechanisms of antibody-mediated shedding inhibition.
Crystallization and structural determination. Purified 5G6 Fab from hybridoma cells was mixed with the epitope peptide KL10 at a 1:2 molar ratio in crystallization solution (50 mM NaCl, 20 mM Tris·HCl, pH 7.4) and concentrated to approximately 10 mg/mL using a membrane with a 10-kDa molecular weight cut-off. After the initial screening, the complex was crystallized by the sitting drop method at 16 °C in 1.6M (NH 4 ) 2 SO 4 , 0.01M NiCl 2 , and 0.1M Tris·HCl, pH 8.5. Single crystals were transferred to the same solution containing 20% (v/v) glycerol and were flash-frozen in liquid nitrogen. X-ray diffraction data were collected at 100 K at the Shanghai Synchrotron Radiation Facility (beamline BL17U1). The collected data were integrated and scaled using the program HKL-2000 20 . The complex structure was determined by molecular replacement with the structure of mAb 1479 (PDB code: 3U9U) as a search model 21 . Most of the residues in 5G6 Fab were located correctly, and the remaining portion was built manually using the program COOT 22 . The model was refined to 2.38 Å using the program sphenix.refine 23 is the intensity of the ithmeasurement, and < I(hkl)> is the mean intensity for that reflection. c R factor = ∑ ∑ hkl ||Fo|-|Fc||/∑ ∑ hkl |Fo|, where |Fo| and |Fc| are the observed and calculated structure factor amplitudes, respectively. d R free was calculated with 5.0% of the reflections in the test set.
Scientific RepoRts | 6:24789 | DOI: 10.1038/srep24789 Cloning, expression and purification of recombinant Fab fragment. Total mRNA was extracted from frozen 5G6 hybridoma cells, and reverse transcription PCR was performed to amplify the variable regions of the antibody. The cDNA fragments encoding the heavy and light chains of wild-type 5G6 Fab were cloned into a pcDNA3.1 vector with a C-terminal 6 × His tag (Invitrogen) and were then sequenced. Both vectors were cotransfected at a 1:1 ratio into suspended CHO cells in serum-free medium (Wuxi App Tec). Then, the transiently transfected cells were cultured at 37 °C with 6% CO 2 and shaken at 125 rpm in disposable flasks. Seven days after transfection, the culture medium was collected and filtered, and the Fab proteins in the media were purified using a Ni-NTA Superflow column (Qiagen) and then further purified using a Hiload 16/60 Superdex 200 column (GE Healthcare) with Buffer A (200 mM NaCl, 20 mM Tris·HCl, pH 8.0). The eluted protein was collected and concentrated for isothermal titration calorimetry (ITC) and flow cytometry experiments.
Isothermal titration calorimetry (ITC). ITC was used to measure the binding affinities of various Fab fragments with KL10 peptide and wild-type 5G6 Fab fragment with various KL10 peptides. All Fab protein samples were purified in Buffer A at a final concentration of 0.02 mM, and all peptides were dissolved in Buffer A at 0.4 mM. The samples were centrifuged before the experiments to remove any protein precipitate. All measurements were performed at 25 °C using a MicroCal iTC200. Various KL10 peptides were titrated into the experimental cell that contained various Fab fragments in twenty 2 μL-injection increments. The titration data were analyzed using the ORIGIN data analysis software (MicroCal Software).
Flow cytometry. Human platelets were seeded into a 96-well plate (Corning 3365, 1.5 × 10 6 cells/well, 50 μL/well), incubated with various 5G6 Fab fragments at room temperature for 20 min, labeled with fluorescein isothiocyanate (FITC)-conjugated goat anti-mouse IgG, and washed with PBS buffer when necessary. The binding of Fab fragments was analyzed on a Becton-Dickinson FACS Canto II instrument (BD Biosciences, San Jose, CA, USA). The signal was quantitated by mean fluorescence intensity (MFI) for the entire cell population (10,000 cells).

Results
Crystal structure of 5G6 Fab in complex with the epitope peptide. To elucidate the structural basis for the binding of 5G6 to the GPIbα shedding cleavage site, we determined the crystal structure of 5G6 Fab fragment in complex with its epitope peptide to 2.4-Å resolution ( Table 1). The N-terminally acetylated epitope peptide contains 10 residues, KLRGVLQGHL, corresponding to residues 461-470 of human GPIbα (referred to as KL10). The crystal belongs to the space group C222 1 , with two Fab-peptide complexes in each asymmetric unit. The two complexes in the unit take on very similar structures and have a root-mean-square deviation (RMSD) value of 0.5 Å. In the complex (Fig. 1), 5G6 Fab adopts the canonical structure of IgG1 and belongs to the C H 1/C L κ subtype, with 4 of the 6 CDR (complementarity-determining region) loops conforming to canonical classes, as defined by Chothia and colleagues 25 . The epitope peptide is not fully extended in the complex structure and covers the interface with CDR loops from both the heavy and light chains of 5G6. Notably, 5G6 Fab binds only to one side of the epitope peptide without obstructing both termini and it interacts with the entire peptide at an approximate right angle. Thus, it is very likely that the peptide conformation revealed in the complex structure may mimic the native conformation adopted by the same sequence as an accessible part of the mechanosensitive domain (MSD) in GPIbα 26 and that binding of 5G6 to GPIb-IX may not be hindered by the juxtaposition of its binding epitope with the plasma membrane.
Interface. Close inspection of the complex structure yielded the following observations about the binding interface between 5G6 Fab and the epitope peptide ( Fig. 2A). First, the CDR loops of 5G6 form a groove at the binding interface. Over the groove residues, Gly 464 -Gln 467 of the epitope peptide adopts a helix-like structure. As a result, the shedding cleavage site, the Gly 464 -Val 465 peptide bond 10,27 , is not in direct contact with 5G6. Second, the Fab-peptide interface spans 647.7 Å 2 , of which 350.8 Å 2 is with the heavy chain and 296.9 Å 2 is with the light chain (as calculated using PISA and PDBsum, respectively) 28,29 . The extensive interface helps explain the high binding affinity of 5G6 for its epitope. Third, all polar residues in the epitope peptide, Lys 461 , Arg 463 , Gln 467 and His 469 , are located on the same side of the peptide that is in direct contact with 5G6 Fab, and they constitute the majority of the binding interface ( Fig. 2A). In particular, Arg 463 and Gln 467 supply atoms for 8 pairs of hydrogen bonds and hydrophobic contacts with 5G6 (Fig. 2B). It is notable that these two residues are specific for human GPIbα . Their murine counterparts are Pro and Leu, respectively, which may explain the inability of 5G6 to bind to murine GPIbα 19 . Fourth, it is also notable that the side chains of all hydrophobic residues in the epitope peptide, Leu 462 , Val 465 , Leu 466 and Leu 470 , point away from the binding interface ( Fig. 2A right panel), supporting the idea that several of these residues are part of the hydrophobic core of the juxtamembrane MSD of GPIbα 26 . Consistent with this notion, these hydrophobic residues are conserved in the murine sequence ( Supplementary Fig. S1). Overall, the Fab-peptide complex structure confirmed the direct association of 5G6 with its target GPIbα residues and elucidated the molecular basis underlying the binding specificity and high affinity.
Structural basis underlying the high binding affinity. To fully elucidate the molecular basis underlying the binding specificity and high affinity of 5G6 for its epitope, sequence alignment of the heavy and light chains of 5G6 with several known antibodies was performed (Fig. 3A,B, respectively). Among the residues that  participate in the interaction with KL10, CDRH3 and Asp 93 of CDRL3 are not found in other antibodies but are specific to 5G6, whereas residues in CDRH2 and CDRL1 are either conserved in other antibodies or provide a hydrophobic environment. Structural comparison indicates that CDRH3 adopts a canonical β -sheet conformation with a relatively rigid loop region (Fig. 3C). In particular, the four threonine residues (Thr 102-105 ) in this loop region provide atoms for 4 pairs of hydrogen bonds and hydrophobic contacts with KL10. Therefore, CDRH3 and Asp 93 of CDRL3 may play decisive roles in the binding specificity and high affinity of 5G6 for GPIbα .
To further validate that Thr 102-105 and Asp 93 were the key residues mediating 5G6 binding specificity and affinity, two mutant Fab fragments were generated, and their binding affinities for KL10 were determined using ITC. In the mutant 5G6 Fab 5G6-M1, Asp 93 of CDRL3 was replaced with Ala, whereas in the mutant 5G6-M2, both the Thr 102-105 of CDRH3 and the Asp 93 of CDRL3 were replaced with Ala (Fig. 4A). To ensure the same origin of the Fab proteins, both the wild-type 5G6 Fab fragment and its mutants were expressed in CHO cells and purified by affinity chromatography. Consistent with results from our previous report 19 , the binding affinity of the wild-type 5G6 Fab fragment for KL10, measured by ITC, was 5.7 nM (Fig. 4B), approximately 4-fold higher than that of 5G6-M1 (21.2 nM, Fig. 4C). In contrast, 5G6-M2, with five point mutations in the interface, completely lost its capacity to interact with KL10 (Fig. 4D). These results demonstrated that Asp 93 of CDRL3and Thr 102-105 of CDRH3 play decisive roles in the binding specificity and high affinity of 5G6 for its epitope.
To verify the role of CDRH3, antibody 15C6, which contains a similar CDRH3 sequence, was studied. We previously reported that the affinity of 15C6 for GPIb-IX (~28 nM) was slightly weaker than that of 5G6 (~3 nM), but much higher than those of other antibodies 19 . Consistently, in the present study, the dissociation constant (K d ) of 15C6 with KL10 measured by ITC assay was 35 nM. 5G6 and 15C6 differ by only seven amino acids, of which only three are in the CDR loops ( Supplementary Fig. S2A). Sequence comparison revealed that Thr 102 of 5G6 CDRH3 and Tyr 30 of 5G6 CDRL1 are replaced by Ser and His, respectively, in 15C6. It was interesting to note that Tyr 30 is specific to 5G6, but that His 30 is conserved in 15C6 and the other four antibodies (Fig. 2B   and Supplementary Fig. S2B). To test whether these two sites are responsible for the difference in 5G6 and 15C6 binding affinity, a T102S and Y30H double mutant (5G6-M5) was generated, and the mutant protein was expressed and purified ( Supplementary Fig. S2A). 5G6-M5 exhibits a similar binding capacity as that of 5G6, suggesting that these two residues are not primarily responsible for the weaker affinity of 15C6 ( Supplementary  Fig. S2B). It is possible that the other three amino acids slightly alter the conformation of 15C6 and result in the fine distinction. Fig. 2B, Arg 463 of KL10 supplies atoms for 6 pairs of hydrogen bonds and hydrophobic contacts with 5G6. In particular, the Nη atom of Arg 463 in the interface is extremely important.

Conformation of KL10. As shown in
To confirm the role of Arg 463 in epitope recognition, two mutant KL10 peptides were synthesized. In the mutant KL10 peptide KL10-M3 and KL10-M4, Arg 463 was replaced by Lys and Ala, respectively (Fig. 4A). The binding affinities of these mutant peptides for the wild-type 5G6 Fab fragment were determined using the ITC technique. When Arg 463 was replaced by Ala, the residue lacked the side chains involved in the interaction with 5G6, and thus, as expected, KL10-M4 had no interaction with 5G6 (Fig. 4F). In contrast, when Arg 463 was replaced by Lys, the guanidine group in the side chain of Arg 463 changed into an amino group that may supply similar pairs of hydrogen bonds as Arg 463 . However, to our surprise, the dissociation constant of KL10-M3 with 5G6 was approximately 1000-fold lower (16 μM, Fig. 4E) than the wild-type KL10 peptide (5.7 nM). These ITC results indicate that Arg 463 is indispensable for the binding: it provides key elements for the interaction with 5G6 and may also be important for peptide conformation.

5G6 mutants have lower affinity to human platelets. Our previous study has revealed that both 5G6
and 5G6 Fab fragments exhibit strong binding to human platelets, as detected by flow cytometry 19 . To address whether the two mutant 5G6 Fab fragments 5G6-M1 and 5G6-M2 also have weaker affinities for human platelets as observed in ITC measurements, similar flow cytometry assays were performed described previously 19 . Compared to wild-type 5G6 Fab fragment, 5G6-M1 showed relatively lower affinity for human platelets, whereas 5G6-M2 had no binding (Fig. 5A). These results are consistent with ITC observations, because 5G6-M2 lost its capacity to interact with KL10 peptide (Fig. 4D).

Discussion
GPIbα , a major receptor in platelets, is continuously cleaved from the platelet surface and releases its extracellular domain, glycocalicin, into the plasma, which has a glycocalicin concentration of 1-3 μg/mL 30 . Such cleavage, also known as GPIbα ectodomain shedding, has been reported since 1981 and is mediated by ADAM17 10,11 . In addition to constitutive shedding, GPIbα ectodomain shedding can be up-regulated under certain circumstances, such as platelet activation 14 . The biological significance of GPIbα shedding remains to be defined, but it has been suggested to serve as a biomarker for atherosclerosis, since an increased glycocalicin level has been reported in patients with atherosclerosis 31 . More importantly, GPIbα shedding has been suggested to be related to platelet production, storage and clearance [32][33][34] . Various studies have consistently reported that glycocalicin accumulates during platelet storage. Thus, GPIbα shedding is thought to be a specific characteristic of the platelet storage lesion, one of the major reasons for the short shelf-life of platelets 12,27,32,33,35 . Treatment with the broad-spectrum metalloprotease inhibitor GM6001 or inhibition of ADAM17 activity with p38 mitogen-activated protein kinase (MAPK) inhibitors improves murine platelet recovery during storage 12,16 . However, because ADAM17 has a broad range of substrates, these studies do not specifically target GPIbα shedding. Therefore, to explore the functions of GPIbα shedding, a monoclonal antibody, 5G6, that specifically inhibits GPIbα shedding was screened in our previous study 19 .
The extracellular domain of GPIbα consists of three regions: an N-terminal ligand-binding domain (LBD) for the interaction with vWF and other ligands, a sialomucin domain that extends for tens of nanometers, and a quasi-stable MSD (Fig. 5B) 26,36,37 . The MSD, which harbors the GPIbα shedding cleavage site, is structured but relatively unstable. It is postulated to undergo unfolding upon ligand binding and mechanical pulling, leading to transmembrane signaling 26 . Because the proper expression of GPIbα on the platelet membrane requires its correct association with GPIbβ and GPIX, it has been hypothesized that the folded MSD in GPIbα is surrounded and constrained by its neighboring GPIbβ and GPIX subunits, leaving a potential entry site for ADAM17 19,26 . This entry site for ADAM17, which is demonstrated can be blocked by 5G6 in the previous study, is supposed to be available when MSD is unstructured 19,26 . Because both the folded MSD in the full length GPIb-IX complex and KL10 can be recognized by 5G6 and bound at similar affinity, the structure of KL10 in 5G6 Fab-KL10 complex may partly provide the structural information of MSD in its native condition. In the complex structure, KL10 adopts a partially extended conformation with four polar residues (Lys 461 , Arg 463 , Gln 467 and His 469 ) constituting the majority of the binding interface and four hydrophobic residues (Leu 462 , Val 465 , Leu 466 and Leu 470 ) pointing away from the interface. Therefore, Leu 462 , Val 465 , Leu 466 and Leu 470 in KL10 are suggested to be part of the native MSD structure.
5G6 Fab shares high sequence homology with four other antibodies and adopts the canonical structure of IgG1. Sequence alignments show that CDRH1 and CDRH2 of 5G6 are conserved among different antibodies, indicating their limiting roles in binding specificity. In contrast, CDRH3 is specific to 5G6 and accounts for the majority of the hydrogen bonds between 5G6 and KL10. In the complex structure, Thr 102-105 in CDRH3, along with Asp 93 in CDRL3 were the key residues for 5G6 in mediating binding specificity and affinity, while Arg 463 was the corresponding key residue for peptide KL10 (Fig. 2B). To further validate this hypothesis, two mutants for 5G6 Fab and two mutants for KL10 were generated and applied to ITC experiments (Fig. 4). In 5G6-M1, where Asp 93 in CDRL3 was mutated to alanine, only the corresponding hydrogen bond between CDRL-Asp 93 and KL10-Gln 467 was lost, with the main hydrogen bonds and hydrophobic interactions remained. Therefore, only a subtle change in ∆ H and ∆ S was caused, and as a result a 3-4 fold of decrease in its binding affinity to KL10. In contrast, when the four threonine residues 102-105 were also mutated to alanine in 5G6-M2, the hydrogen bonds between CDRH-Thr 102 and KL10-Arg 463 , KL10-Leu 466 , the hydrogen bond between CDRH-Thr 103 and KL10-Arg 463 , as well as the hydrogen bond between CDRH-Thr 104 and KL10-Arg 463 were all absent, abolishing the binding of the Fab protein to KL10. On the other hand, guanidine group of Arg 463 in KL10, responsible for two hydrogen bonds with CDRL-Phe 91 and one hydrogen bond each with CDRH-Thr 102 , Thr 103 and Ser 106 , plays a pivotal role in epitope recognition. When Arg 463 was replaced by a lysine in KL10-M3, the corresponding amine group in lysine failed to sustain the two hydrogen bonds with CDRL-Phe 91 and the other hydrogen bonds with CDRH-Thr 102 , Thr 103 and Ser 106 were also supposed to be compromised, leading to a significant increase of ∆ H and a nearly 3000-fold decrease in its binding affinity with 5G6. Additionally, the orientation of KL10-M3 with 5G6 might be deviated compared to KL10-WT, and the hydrophobic interactions between Arg 463 and CDRL-Trp 92 is likely to be attenuated as a consequence, resulting in significant increase of ∆ S. When Arg 463 was replaced by an alanine in KL10-M4, all the five hydrogen bonds mentioned above were supposed to be absent, which led to a complete loss of its capacity to interact with 5G6 as expected.
As indicated above, ITC data for the mutant Fab proteins demonstrated that CDRH3 is the key fragment in the binding affinity and specificity of 5G6 for its epitope (Fig. 4C,D). Moreover, the importance of Thr 102-105 and Asp 93 in CDRH3 in binding to the full length GPIbα expressed on the cell surface was also confirmed in flow cytometry studies (Fig. 5A). Therefore, our structural and biochemical data indicate the above crucial residues in the antigen recognition and high binding affinity of 5G6 for GPIbα and may provide the molecular basis for further antibody optimization and antibody humanization. Furthermore, because the specific inhibition of GPIbα shedding through treatment with 5G6 Fab markedly improves post-transfusion recovery in vivo (data unpublished), the data presented herein may facilitate the development of a new generation of GPIbα shedding inhibitors that may be utilized in platelet storage.
Overall, in the present study, we determined the 5G6 Fab-KL10 complex structure, which confirmed the direct association of 5G6 with GPIbα residues flanking the shedding cleavage site, and elucidated the molecular basis underlying the binding specificity and high affinity.