Quaternary organization of GPIb-IX complex and insights into Bernard-Soulier syndrome revealed by the structures of GPIbβ and a GPIbβ/GPIX chimera

GPIb-IX-V complex is an abundant membrane receptor complex on the platelet surface that plays a critical role in mediating platelet adhesion to the damaged vessel wall under conditions of high shear stress.¹  Platelets adhere, and integrins are subsequently activated by interactions of GPIb-IX-V with VWF bound to the subendothelium. How GPIb-IX-V transmits the VWF-binding signal across the membrane is not clear, partly because the structure and organization of this complex receptor remain to be elucidated. Because GPV is only weakly associated with the receptor complex and is not essential for complex expression, assembly, VWF binding, or signal transduction,^2,3  we focus on the GPIb-IX complex here.

The GPIb-IX complex contains 3 subunits, GPIbα, GPIbβ, and GPIX, with a 1:2:1 stoichiometry.⁴  Each subunit is a type I transmembrane (TM) protein, containing an ectodomain with leucine-rich repeats (LRRs),⁵  a single TM helix, and a relatively short cytoplasmic tail. The GPIbα ectodomain contains binding sites for a growing list of hemostatically important ligands, including VWF and thrombin.^6-8  Covalent and noncovalent interactions are important to the quaternary stabilization of the receptor. GPIbα links to 2 GPIbβ subunits through membrane-proximal disulfide bonds to constitute the GPIb complex.⁴  GPIX tightly associates with GPIb through noncovalent interactions.⁹  Assembly of these subunits into a tightly integrated complex is also supported by genetic evidence. Bernard-Soulier syndrome (BSS) is a hereditary bleeding disorder that is characterized in most cases by giant platelets, low platelet counts, and little or no expression of GPIb-IX on the platelet surface.^10,11  More than 30 mutations have been identified from patients with BSS and mapped to GPIbα, GPIbβ, and GPIX,^12,13  indicating that all 3 subunits are required for proper surface expression of the complex.

Consistent with genetic evidence, efficient surface expression of GPIb-IX also requires all 3 subunits in transfected mammalian cells.¹⁴  For instance, GPIX alone cannot be expressed on the surface of transfected Chinese hamster ovary (CHO) cells. Only when coexpressed with GPIbβ can it be detected on the cell surface, indicating that GPIbβ interacts with and stabilizes GPIX.^15,16  Coexpression with GPIbβ and GPIbα produces even higher surface expression levels of GPIX.¹⁴  Thus, the surface expression level of individual subunits can be used as an indicator for the assembly of GPIb-IX and, implicitly, quaternary interactions among the subunits. Using this approach, we had previously shown that TM domains of GPIb-IX are essential for complex assembly.^17,18  Biophysical characterization of recombinant GPIb-IX–derived TM peptides in detergent micelles indicated that they form a parallel 4-helical bundle and that their association leads to formation of membrane-proximal disulfide bonds between GPIbα and GPIbβ, further stabilizing the complex.^4,19 

In addition to TM domains, GPIbβ and GPIX ectodomains, termed GPIbβ_E and GPIX_E in this study, respectively, are required for proper assembly and efficient surface expression of GPIb-IX, because numerous BSS-causing mutations have been mapped to these domains. However, GPIX_E is intrinsically unstable, impeding structural and biochemical investigation. Taking advantage of the high sequence homology between GPIbβ_E and GPIX_E (Figure 1A), we had earlier identified a GPIbβ_E/GPIX_E chimera (abbreviated to GPIbβ_Eabc) that interacts with GPIbβ in a manner that mimics the GPIX ectodomain.¹⁶  Here, we report X-ray crystal structures of human GPIbβ_E and GPIbβ_Eabc. These structures, in combination with assays to probe inter-subunit interactions in the context of full-length subunits, give insight into GPIb-IX quaternary assembly and provide greater understanding of the molecular mechanism of BSS.

To produce recombinant GPIbβ_E and GPIbβ_Eabc, a gene fragment encoding human GPIbβ and GPIbβ_Eabc residues Cys1-Leu121, respectively, was cloned into a pBlueBac4.5-derived baculovirus expression vector (Invitrogen) that appended the signal sequence of baculovirus envelope gp67 (amplified from pAcGP67A vector; BD Biosciences) and a Ser-Ser-hexahistidine tag to the N-terminus and C-terminus of GPIbβ_E, respectively. The chimeric GPIbβ_Eabc contains 3 stretches of GPIX sequences: Ala29-Arg36, Ser49-Gln60, and Ser76-Arg87 (Figure 1A). The mature protein with the hexahistidine tag was secreted from the infected insect cells into the culture media. After 40%-85% ammonium sulfate fractionation of the collected culture media, the protein was dissolved in 50mM Tris·HCl, 20mM imidazole, 4mM EGTA, pH 7.6, at 4°C overnight; loaded onto the Ni-sepharose column (QIAGEN); and subsequently eluted with 50mM Tris·HCl, 300mM NaCl, 100mM imidazole, and 4mM EGTA, pH 7.6. The eluent was concentrated and further purified by gel filtration chromatography (Superdex 75; GE Healthcare) in 20mM Tris·HCl and 100mM NaCl, pH 7.0 (supplemental Figure 1, available on the Blood Web site; see the Supplemental Materials link at the top of the online article). N-terminal sequencing analysis of the purified protein confirmed the proper removal of the signal sequence.

Purified GPIbβ_E with the hexahistidine tag was dialyzed into 20mM Tris·HCl, and 100mM NaCl, pH 7.0, and concentrated to 18 mg/mL for crystallization at 19°C. Sparse matrix screens (QIAGEN) obtained initial conditions and refinement resulted in the crystallization condition of 1.6M (NH₄)₂SO₄, 0.4M LiCl, and 0.1M MES, pH 6.5. For GPIbβ_Eabc the sample was concentrated to 1.5 mg/mL, and sparse matrix screens (PACT; Molecular Dimensions) identified initial conditions at 10°C from D2 (form 1) and A9 (form 2). The D2 condition is 0.1M MMT buffer pH 5 (MMT buffer is a mixture of DL-malic acid, MES, and Tris base in the molar ratios 1:2:2, DL-malic acid/MES/Tris base), 25% (w/v) polyethylene glycol 1500 and A9 is 0.2M lithium chloride, 0.1M sodium acetate, pH 5, and 20% polyethylene glycol 6000. Single crystals were transferred to the same solution containing 25% and 10% glycerol, respectively, and flash cooled in liquid nitrogen. Diffraction data were collected with beam line ID23-2, ID29-1, and ID23-2, respectively, for GPIbβ_E, GPIbβ_Eabc (form 1), and GPIbβ_Eabc (form 2) at the European Synchrotron Radiation Facility. The GPIbβ_E structure was determined by molecular replacement with the use of the structure for C-terminal 133 residues of the Nogo-66 ectodomain (PDB code, 1P8T)²⁰  and programs MrBUMP²¹  and PHASER.²²  Initial electron density maps were improved with 2-fold noncrystallographic symmetry and solvent flattening with the use of the CCP4 program suite. Model rebuilding was performed with COOT²³  and crystallographic refinement was performed in REFMAC.²⁴  Crystallographic statistics are listed in Table 1. A Ramachandran plot shows 115 residues in preferred regions, 2 in allowed regions, and none in outlier regions. The GPIbβ_Eabc form 1 structure was determined by molecular replacement with the use of the GPIbβ_E structure. The model was built with COOT and refined with REFMAC. A Ramachandran plot shows 116 of residues in favored regions and 3 in allowed regions with none in outlier regions. The form 2 structure was determined by molecular replacement with the use of the form 1 structure and is identical with the exception of side chains involved in crystal packing.

Vectors expressing hemagglutinin (HA)–GPIbβ (full-length GPIbβ with N-terminal HA epitope tag, YPYDVPDYA), HA-GPIbβ_E (GPIbβ extracellular residues Cys1-Leu121 with N-terminal HA tag), HA-GPIbβ_Eabc-GPIX_TC (HA-tagged GPIbβ_Eabc fused to GPIX TM and cytoplasmic domains), HA-GPIbβ_Eabc (HA-tagged GPIbβ_Eabc), GPIX, and GPIbα had been described.^16,17,25  Site-directed mutagenesis was performed by the overlap extension PCR procedure with the use of the above-mentioned vectors as the template. Each PCR fragment was digested by appropriate restriction enzymes and subcloned into its target vector as described earlier.¹⁷  All constructs were confirmed by DNA sequencing.

CHO K1 cells were grown in DMEM supplemented with 10% FCS at 37°C and 5% CO₂. Transient transfection of vectors encoding GPIbα, GPIbβ, and GPIX-derived constructs, in desired combinations, into CHO K1 cells was performed with lipofectamine 2000 (Invitrogen) as described earlier.¹⁷  Key parameters of the transfection, such as cell density and DNA amount, were kept the same as previously described to allow proper comparison among various constructs.¹⁷  After transfection, the cells were grown for an additional 48 hours before being analyzed.

The secretion and folding of mutant HA-GPIbβ_E and HA-GPIbβ_Eabc proteins were characterized as previously described.¹⁶  Briefly, the N-terminally HA-tagged protein secreted into the culture medium was collected by coimmunoprecipitation with the use of the anti-HA Ab, resolved in 12% Bis-Tris SDS gel in the presence or absence of reducing agents, and immunoblotted by HRP-conjugated anti-HA Ab. The cellular expression and assembly of the GPIb-IX complex was assessed by Western blot of each subunit as previously described.^17,18  The surface expression levels of GPIbα, HA-tagged GPIbβ, and GPIX in transiently transfected cells were measured with WM23, anti-HA, and FMC25 Abs, respectively, on a Beckman-Coulter Gallios flow cytometer as described.^17,25  To assess the mutational effect, the measured mean fluorescence value of the entire cell population (10 000 cells) is normalized with the value of CHO cells expressing wild-type GPIb-IX complex (GPIbα/HA-GPIbβ/GPIX) being 100% and that of empty vector-transfected cells 0%.¹⁷  Groups were compared with the 2-tailed Student t test.

To explore the architecture of the GPIb-IX receptor complex we crystallized recombinant GPIbβ_E and solved the crystal structure to 1.25-Å resolution (Table 1). The topology of GPIbβ_E spanning residues 1-118 is shown in Figure 1B, showing the first structure with only a single LRR repeat. A typical example of the refined electron density is shown in stereo in Figure 1C and supplemental Video 1. As seen in other LRR proteins, GPIbβ_E assumes a right-handed coiled structure with a parallel β-sheet on one side (the concave face) and connecting loops containing β-turns and short 3₁₀ helices on the opposite (convex) face. The central LRR is covered on both ends by N- and C-terminal capping regions that contain several short α- and 3₁₀-helices. Four disulfide bonds are observed in the GPIbβ_E structure: 2 (Cys1-Cys7 and Cys5-Cys14) are located in the N-terminal cap and the other 2 (Cys68-Cys93 and Cys70-Cys116) in the C-terminal cap, which are topologically equivalent to those in the Nogo-66 and SLIT receptor.^20,26  With only a single LRR it assumes a compact rectangular shape with a relatively flat, rather than a curved concave, face commonly observed in multi-LRR structures. Moreover, the single LRR accommodates a unique feature in GPIbβ_E that has not been observed in previously reported LRR structures; interactions of side chains bridging N- and C-terminal capping regions. Extending over the convex face, the aromatic ring in Trp21 is locked between the amino group of Pro46 and the guanidinium group of Arg71 by cation-π interactions (Figure 1B). These interactions exemplify numerous interloop interactions on the convex face, which probably add stability to the structure in a manner similar to the buried GPIbβ Asn residues, Asn40 and Asn64, on the concave face.²⁷ 

Eight missense mutations in GPIbβ_E (C5Y,²⁸  R17C,²⁹  P29L,³⁰  N64T,³¹  P74R,³²  Y88C,^33,34  P96S,³⁵  and A108P³⁴ ) have been identified in patients with BSS. We have examined the context of these 8 mutations with the use of the GPIbβ_E structure. Residues affected are shown in a ribbon diagram of the structure in Figure 2A and also in supplemental Video 2. Information on surface localization and conservation of each affected residue is summarized in supplemental Table 1. Cys5 and Cys14 form a disulfide bond. Substitution of Cys5 with a tyrosine would result in loss of the disulfide bond and leave an unpaired Cys14. Asn64 is fully buried in the structure and bridges the C-terminal cap and the LRR with 2 hydrogen bonds, both of which would be lost if substituted by a smaller Thr residue. The other GPbβ_E residues (Arg17, Pro29, Pro74, Tyr88, Pro96, and Ala108) are present on the protein surface, although Arg17, Tyr88, and Pro96 are partially buried by surrounding side chains (Figure 2B). Both R17C and Y88C would introduce an additional Cys residue to the domain, which will probably interfere with formation of the 4 native disulfide bonds. The other 4 mutations involve either removal or addition of a Pro residue, which can affect local conformation or global stability.

To elucidate the molecular pathogenesis of these BSS mutations, we have characterized systematically the mutational effects on GPIb-IX expression and assembly in transiently transfected CHO cells. Key parameters of transient transfection were kept constant to ensure proper comparison of protein expression levels among different experiments.¹⁷  CHO cells largely recapitulated the reported clinical observations that all 8 mutations led to a significant decrease in surface expression of the GPIb-IX complex, albeit to various degrees (supplemental Figure 2). To test whether these BSS mutations are detrimental to the structural integrity of GPIbβ_E, each mutation was introduced to the HA-tagged GPIbβ_E (with the native signal sequence), and the resulting gene was transfected transiently into CHO cells. Western blot analysis of the cell lysate indicated that translation of the HA-GPIbβ_E gene was not affected by any of the mutations (Figure 2C). However, C5Y, P29L, and P96S mutant proteins failed to secrete from the cells because no HA-tagged protein was detected in the culture media. The secretion of R17C was significantly decreased. In N64T-expressing cells, only the ectodomain with a higher molecular mass was detected in the culture media. The other mutants, P74R, Y88C, and A108P, were secreted similar to the wild type. Of the mutants that were secreted, R17C, N64T, and Y88C exhibited significant formation of intermolecular disulfide bonds as detected in SDS-PAGE under nonreducing conditions. Because GPIbβ_E contains 4 intramolecular disulfide bonds and no intermolecular ones (Figure 1B), the existence of the latter indicated misfolding of these mutant proteins. In contrast, P74R and A108P GPIbβ_E mutant proteins, like the wild-type, contained only intramolecular disulfide bonds and should therefore be well folded.

The effects of mutations on the interaction between GPIbβ and GPIX were analyzed next. GPIX does not express on the cell surface in isolation but becomes detectable when coexpressed with GPIbβ.¹⁴  GPIbβ is thought to interact with and stabilize GPIX, which involves the ectodomain and TM domain of both proteins.^15,16  As shown in Figure 2D, when HA-tagged GPIbβ (HA-GPIbβ) and GPIX were cotransfected into CHO cells, both HA-GPIbβ and GPIX were detected on the cell surface. Mutations that disrupt secretion or folding of GPIbβ_E, such as P29L, resulted in little cell surface expression of HA-GPIbβ and, as a consequence, GPIX as detected by flow cytometry. By contrast, a distinctive feature of P74R and A108P was that HA-GPIbβ was present on the cell surface but GPIX was not. Overall, these results showed that BSS missense mutations disrupt assembly and surface expression of the GPIb-IX complex via different mechanisms.

Despite high sequence similarity between GPIbβ_E and GPIX_E, attempts to express recombinant GPIX_E have not been successful.¹⁶  However, sequence analysis did allow the engineering of GPIbβ_Eabc, a chimera of GPIbβ_E and GPIX_E, as a stable protein that is readily secreted.¹⁶  Built on the GPIbβ_E scaffold, GPIbβ_Eabc incorporates 3 discontinuous stretches of GPIX_E that correspond to the 3 convex loops, including the α1 helix (termed loops a, b, c; spanning GPIX residues Ala29-Arg36, Ser49-Gln60, and Ser76-Arg87 as shown in Figure 1A). GPIbβ_Eabc was purified from insect cells and crystallized, and the structure was solved to 2.35-Å resolution, using the GPIbβ_E crystal structure for a molecular replacement calculation identifying 4 molecules in the asymmetric unit (Figure 3A; Table 1). To avoid confusion, residues of GPIbβ_Eabc are labeled in this study by their residue numbers in respective source domains (eg, GPIX-Asp56 in loop b) rather than by GPIbβ_Eabc's own.

The GPIbβ_Eabc structure shares the same topology as GPIbβ_E, but it has 3 changes in the structure (Figure 3B). The first, as expected, is in the convex loops. Because of the substantial sequence difference, the 3₁₀ helix in loop a and the cation-π interloop interaction between Trp21 and Arg71 in GPIbβ_E are not present in the GPIbβ_Eabc structure (Figure 3A). Second, in helix α1 of loop c GPIX-Tyr79 replaces GPIbβ-Pro74, which surprisingly does not affect the kinked main chain conformation but instead forms a substantial new interaction with loop b (Figure 3A). Here, the GPIX-Tyr79 side chain is buried under the main chain of GPIX-Gly53 and the Tyr OH group forms a hydrogen bond to the main chain nitrogen of adjacent GPIX-Phe55 (Figure 3A). The extensive interactions observed in these convex loops substantiate our earlier observation that only when all 3 convex loops are grafted together onto the GPIbβ_E scaffold did the GPIbβ_Eabc chimeric protein become stable and well folded.¹⁶  A third difference occurs in the C-terminal cap, despite GPIbβ_E and GPIbβ_Eabc sharing the same amino acid sequence in this region. Residues 86-89 that form the 3₁₀ helix in loop c of GPIbβ_E unravel, and residues 80-86 coil up to form a novel turn of helix in GPIbβ_Eabc with GPIbβ-Glu84 forming a new salt bridge to GPIbβ-Arg57. In addition, the C-terminal helix α2 makes a rigid body movement of 3 Å away from helix α1. Compared with GPIbβ_E, the angle between helices α1 and α2 in GPIbβ_Eabc is widened (Figure 3C).

The re-coiling conformational change in the loop containing residue Tyr88 is in a topologically equivalent region of the LRR fold to the β-switch loop in GPIbα known to alter conformation when binding to VWF-A1 domain or peptide inhibitor.^6,36  In GPIbα the conformational change involves unraveling of a short α-helix to form an extended β-hairpin, whereas in GPIbβ the α-helix unravels and a second stretch is formed, displaced toward the C-terminus in a twisting helical motion (Figure 3D).

Previous studies with the GPIbβ_Eabc construct have shown that the 3 loops (a, b, c) from GPIX form ectodomain interactions with GPIbβ and that this is required for surface expression of the GPIb-IX complex.¹⁶  The side chains from GPIbβ involved in this interaction with GPIX were unknown. In the crystal structure we observe 4 GPIbβ_Eabc molecules in the asymmetric unit form a tetrameric ring structure. Here, the C-terminal cap region of one subunit packs against the convex loops (b, c) of another placing the α-helices toward the center of the ring where they are partially buried (Figure 4A). At this interface the side chain of GPIbβ-Tyr106 from one subunit lies at the center of a shallow pocket created between loop b and helix α1 (loop c) from a second subunit that is rotated by 90 degrees. Each interface buries a surface area of ∼ 1100 Å,²  which is significantly greater than values typical for crystal contacts³⁷  or the contacts found in the GPIbβ_E structure. A second monoclinic crystal form (space group C2₁) of GPIbβ_Eabc grown under different conditions has an identical structure and tetrameric arrangement (labeled form 2 in Table 1).

Figure 4B shows hydrophobic contacts contribute to the GPIX pocket, including GPIX-Tyr79, GPIX-Leu82, and GPIX-Trp83, and more peripheral contacts to the interface come from GPIbβ-Ala108 and GPIX-His57. The OH group of GPIbβ-Tyr106 forms a hydrogen bond to the guanidinium of GPIX-Arg87. GPIbβ-Tyr106 is further surrounded by 3 salt bridges between GPIbβ residues Arg102, Asp90, and Arg89 and GPIX residues Asp56, Arg87, and Asp86, respectively. GPIbβ-Arg92 and GPIX-Arg87 side chains form hydrogen bonds to the main chain carbonyls GPIX-Asp56 and GPIbβ-Ala86, respectively.

The interface between GPIbβ_Eabc molecules is composed of the GPIbβ_E-derived C-terminal region and GPIX_E-derived convex loops. To test whether this interface is a valid representation of full-length GPIbβ and GPIX in the whole complex, 3 residues were selected for mutagenesis. GPIbβ-Tyr106 lies at the center of the interface, and GPIX-Leu82 and GPIX-Asp86 are examples of residues that contribute hydrophobic and electrostatic contacts to the interface, respectively (shown as underlined in Figure 4B). In the GPIbβ_E crystal structure, the side chain of Tyr106 is exposed to the solvent and flanked by Arg85 and Arg89. Mutating Tyr106 to Phe, Asn, Asp, Ala, or Val largely preserved proper secretion and folding of HA-GPIbβ_E that were expressed transiently from CHO cells (Figure 5A). However, when expressed as a full-length subunit on the surface of cells, all Tyr106 mutations resulted in the loss of the ability of HA-GPIbβ to enhance surface expression of GPIX in transfected CHO cells (Figure 5B). Thus, as predicted by the GPIbβ_Eabc crystal structure, these results pinpoint GPIbβ-Tyr106 as a critical part of a GPIX_E binding site.

Similarly, HA-GPIbβ_Eabc protein containing a GPIX-L82A or GPIX-D86E mutation was able to secrete to the culture media and fold well (Figure 6A). However, both mutations failed to retain the GPIbβ-binding ability of GPIbβ_Eabc in the context of the full-length subunit (Figure 6B). Because GPIX_E alone cannot express as a well-folded form,¹⁶  it would be difficult to assess the effect of either mutation on its folding and secretion. Nonetheless, full-length GPIX bearing either mutation could not be expressed on the cell surface even in the presence of GPIbβ (Figure 6C), which is consistent with the GPIbβ_Eabc crystal structure that both GPIX-Leu82 and GPIX-Asp86 participate directly in the interaction between GPIbβ_E and GPIX_E.

One of the fundamental but unanswered questions about the GPIb-IX complex is how the subunits organize and assemble. Earlier studies have shown that TM helices of the GPIb-IX complex interact with one another to form a parallel tetrameric α-helical bundle.^4,18,19  The GPIX TM helix bridges the 2 GPIbβ TM helices, and GPIbα performs a similar function providing noncovalent and covalent quaternary interactions and stability by formation of the 2 interchain disulfide bonds bridging 2 GPIbβs. Previously, it has also been shown that the LRR-containing ectodomains form an interface involving the convex loops of GPIX interacting with GPIbβ.¹⁶  Overall protein-protein interactions mediated by LRR domains use a diverse range of scaffolds with different numbers of LRRs (ranging from 1 to > 15) mediating homotypic and heterotypic interactions.⁵  These interactions are principally mediated by the LRR concave face that typically has a curved or horseshoe shape as observed in GPIbα.⁶  GPIbβ and GPIX are unusual in that they have only a single LRR. The GPIbβ_E structure is the first description of this fold and shows a compact shape with little LRR curvature.

A crystal structure for the SLIT receptor LRR ectodomain domain 4 (pdb code 2WFH) shows a homodimer with a concave face-to-face interaction of the LRRs. We initially speculated that the 2 copies of GPIbβ_E may also interact in a similar way to form a dimer; however, in the crystal structure we only observe a monomer. By contrast the structure of a chimeric GPIbβ_Eabc, where 3 loops from GPIX are added, shows a heterodimeric GPIbβ-GPIX interface. Because of the positioning of the 2 subunits in the dimer at right angles, this structure is then able to cyclize with a second dimer to form a ringlike tetramer where the same interface is observed 4 times. Rather than using the LRR concave face, this structure shows that interactions occur through the C-terminal cap of GPIbβ_E and convex face loops of GPIX_E. Here, a central side chain Tyr106 from the GPIbβ_E cap inserts into a pocket created by 2 loops from GPIX on the convex face. This is more familiar to the interfaces observed in Toll-like receptors in which heterodimeric interactions form between the C-terminal cap and one side of the LRRs.³⁸ 

An important spatial constraint for the GPIb-IX complex lies in the uniform topology of its subunits. Because GPIbα, GPIbβ, and GPIX are all type I TM proteins, the C-termini of all 4 ectodomains in the GPIb-IX complex should remain in close proximity to one another to enable formation of the adjacent parallel TM helix bundle. We noticed that in the GPIbβ_Eabc crystal structure, the C-termini of all 4 GPIbβ_Eabc molecules are located on the same side of the tetramer (Figure 4A; supplemental Video 3). If the GPIbβ₁-GPIX-GPIbβ₂ trimer has a similar cyclic arrangement positioned above the TM helices, then a schematic model of this trimer is shown in Figure 7A with the GPIbα TM helix added (shown in green) extending toward the N-terminal ligand binding domain. A second observation from the GPIbβ_Eabc structure is that GPIX can only bind one GPIbβ_E through its b,c loop pocket; thus, further studies will be required to define what the context and conformation of a second GPIbβ_E (GPIbβ_E2) is. Because we observe no interaction between 2 GPIbβ_E molecules in the crystal structure, the model in Figure 7A shows GPIbβ_E2 on the opposite side of GPIX to GPIbβ_E1.

The genetic basis for BSS was established 3 decades ago.¹¹  More than 30 mutations have since been identified from patients with BSS, and the number will probably grow.³⁹  As we showed in this study, several novel mutations could produce BSS-like phenotypes in transfected cells (Figures 5–6). Some BSS mutations are located in the promoter region and are presumed to decrease gene transcription.⁴⁰  Some are frameshifting or nonsense mutations, resulting in a typically truncated and nonfunctional protein product.^41-43  In this study, we have for the first time systematically characterized all the reported missense BSS mutations in GPIbβ and observed distinct mechanisms that underlie the pathogenesis of the disease (Figure 2). Six of the 8 missense mutations are detrimental to proper tertiary folding or secretion of GPIbβ_E. However, the other 2 mutations A108P and P74R maintain the native disulphide bonds and probably fold normally.

The clinical data on the A108P mutation describe a patient who is compound heterozygous with mutation Y88C.³⁴  In these platelets GPIb-IX receptor complex is present on the surface at a reduced level and does bind VWF. Antibody SZ1 (binds GPIbβ/GPIX subunits in complex but not in isolation) recognizes the GPIb-IX present. As a compound heterozygote of GPIbβ, the mutations could form different combinations of the chains, that is, A108P/A108P, A108P/Y88C, and Y88C/Y88C. A separate family that is homozygous for Y88C has classic BSS giant platelets with no receptor present at the platelet surface.³³  As shown in Figure 4B, Ala108 is adjacent to Tyr106 which is central to the GPIbβ_Eabc interface. We showed mutating Tyr106 abolishes GPIX expression in the same way as A108P. Mutating Ala108 to Pro at the periphery of the interface may produce a less-stable subunit association that, in turn, gives rise to the reduction in surface expression of the whole complex, which is what we observe in CHO cells. The P74R mutation shows classic BBS platelets in homozygous patients with no receptor complex detected at the surface.³²  This mutation is more severe than A108P in our cell-based assays and does not support any GPIX expression at the cell surface even in the presence of GPIbα (supplemental Figure 2). Pro74 does not contribute directly to the Tyr106 interface but is located close by in helix α1, which does contribute through side chains from residues Ala86, Arg89, and Asp90 (Figure 4B). Proline residues are commonly found at the N-termini of helices where the special main chain properties influence the secondary structure formation. A change here could disrupt the orientation of the helix and thus its contributions to the interface.

Finally, not all mutations characterized in GPIbβ result in loss or reduction of receptor complex at the platelet surface. The polymorphism G15E is associated with the human platelet-specific alloantigen Iy^a and does not adversely affect the expression level of GPIb-IX.⁴⁴  Gly15 is exposed on the protein surface and located in the N-terminal cap. Figure 7B shows the locality of these mutations in the same model as Figure 7A and shows how 2 copies of GPIbβ in this arrangement present mutations in different contexts (supplemental Video 4). Overall these studies provide valuable insights on the assembly of GPIb-IX complex and will prove a scaffold for further investigation of this important platelet receptor.

The authors thank the European Synchrotron Radiation Facility for provision of synchrotron radiation source and thank David Flot and Gordon Leonard for assistance in using the beamline ID23-2.

This work was supported in part by the National Institutes of Health (grant HL082808, R.L.) and Cancer Research UK (grant C21418/A8573), British Heart Foundation (grant RG/07/002/23 132, J.E.).

Coordinates and structure factors deposited with the Protein Data Bank are 3RFE for GPIbβ_E and 3REZ for GPIbβ_Eabc.

National Institutes of Health

Contribution: P.A.M. and K.H.C. crystallized and determined the structures of GPIbβ_E and GPIbβ_Eabc, respectively; W.Y. produced recombinant proteins for crystallization, characterized and analyzed mutational effects on GPIb-IX expression and assembly, and wrote the paper; X.M. and X.Z. cloned many constructs and established the baculovirus protein expression system; R.L. initiated and designed research, analyzed results, and wrote the paper; and J.E. analyzed results and wrote the paper.

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

The current affiliation for R.L. is Division of Hematology, Oncology, and Bone Marrow Transplant, Department of Pediatrics, Emory University School of Medicine, Atlanta, GA.

Correspondence: Jonas Emsley, Centre for Biomolecular Sciences, School of Pharmacy, University of Nottingham, Nottingham, NG7 2RD, United Kingdom; e-mail: jonas.emsley@nottingham.ac.uk; and Renhao Li, Division of Hem/Onc/BMT, Department of Pediatrics, Emory University School of Medicine, 2015 Uppergate Dr, Rm 464, Atlanta, GA 30322; e-mail: renhao.li@emory.edu.