GpIbα interacts exclusively with exosite II of thrombin.

Activation of platelets by the serine protease thrombin is a critical event in haemostasis. This process involves the binding of thrombin to glycoprotein Ibα (GpIbα) and cleavage of protease-activated receptors (PARs). The N-terminal extracellular domain of GpIbα contains an acidic peptide stretch that has been identified as the main thrombin binding site, and both anion binding exosites of thrombin have been implicated in GpIbα binding, but it remains unclear how they are involved. This issue is of critical importance for the mechanism of platelet activation by thrombin. If both exosites bind to GpIbα, thrombin could potentially act as a platelet adhesion molecule or receptor dimerisation trigger. Alternatively, if only a single site is involved, GpIbα may serve as a cofactor for PAR-1 activation by thrombin. To determine the involvement of thrombin's two exosites in GpIbα binding, we employed the complementary methods of mutational analysis, binding studies, X-ray crystallography and NMR spectroscopy. Our results indicate that the peptide corresponding to the C-terminal portion of GpIbα and the entire extracellular domain bind exclusively to thrombin's exosite II. The interaction of thrombin with GpIbα thus serves to recruit thrombin activity to the platelet surface while leaving exosite I free for PAR-1 recognition.

Introduction

Thrombin is the final protease of the coagulation cascade [1], [2], and its activity is crucial for the formation of stable blood clots [3]. In addition to the cleavage of fibrinogen to fibrin, thrombin also feeds back to upregulate its own production by activating cofactors fV and fVIII, the protease fXI, and through cleavage of the protease-activated receptor (PAR)-1 [4], [5]. PAR-1 is a seven-transmembrane G-protein-coupled receptor expressed on the surface of platelets. Upon cleavage of the N-terminal extracellular peptide at Arg41, the newly formed N-terminus inserts into the body of the receptor triggering platelet activation. The rate of platelet PAR-1 cleavage by thrombin is increased by approximately 5-fold in the presence of glycoprotein Ibα (GpIbα) [6], a member of the platelet GpIb-IX-V receptor complex [7], that has been identified as the high-affinity receptor for thrombin on the surface of platelets [8]. The crystal structure of the extracellular domain of GpIbα, composed of the 300 N-terminal residues, revealed an elongated banana-like shape dominated by a leucine-rich repeat (LRR) region, followed by a flexible acidic stretch at the very C-terminus (Gly271–Glu282) containing three sulfated tyrosine residues (276, 278 and 279) [9]. The acidic residues in this region and the posttranslational sulfation of the tyrosines are crucial for binding of GpIbα to thrombin [10], [11], [12].

Thrombin has two anion binding exosites, both of which are essential for its activity and specificity (Fig. 1a and b) [13]. Exosite I is the fibrinogen recognition exosite, but it also interacts with a variety of other substrates, cofactors and inhibitors, including the N-terminal region of PAR-1, the cofactor thrombomodulin (TM; shown in Fig. 1a and b) and the C-terminal peptide of the inhibitor hirudin (known as hirugen). Exosite II binds to glycosaminoglycans such as heparin (shown in Fig. 1a and b), to the γ′-chain of fibrinogen and to the second kringle domain of prothrombin, known as fragment-2 (F2). Although both exosite I and exosite II interactions are mediated by electrostatics, there is a surprising degree of exosite selectivity, with ligands often binding exclusively to one site or the other [13]. At high concentrations, however, there is evidence that some ligands, such as heparin, can interact with both sites [14], [15], although this is not thought to be of physiological relevance. It has been demonstrated that the two exosites can bind independently and simultaneously to different ligands; for example, when hirugen is bound to exosite I, GpIbα or fibrinogen γ′ peptides are still able to bind to exosite II [16]. However, particularly in the absence of active-site occupation, there is a considerable degree of allosteric communication between exosites I and II, with binding at one exosite decreasing affinity of ligands for the other [17].

Fig. 1

Structure of thrombin and its exosite interactions with GpIbα. (a) A ribbon diagram of thrombin (heavy chain only) is shown in the standard orientation coloured from its N-terminus to its C-terminus, from blue to red. In magenta are classic exosite binding ligands, TM as ribbons (exosite I) and heparin as sticks (exosite II). The anion binding exosites are indicated. (b) Thrombin is oriented as in (a) but shown in surface representation coloured according to electrostatic potential, with blue positive and red negative. The active-site cleft is the deep red pocket between the exosites. (c) The crystal structure 1OOK is shown, with thrombin in cyan and the acidic C-terminal region of GpIbα as magenta sticks. The chosen asymmetric unit contained the copy of GpIbα shown binding to exosite I (right); however, a crystallographically related copy binds to exosite II (left). (d) The crystal structure 1P8V is shown with thrombin in magenta and GpIbα in green. The acidic C-terminal region (green sticks) binds only to exosite II, and exosite I is occupied by the LRR region (green ribbons) of a symmetry-related GpIbα molecule.

How thrombin binds to GpIbα has been studied extensively for the last 30 years; however, the issue remains unresolved. Initial experiments by cross-linking thrombin to washed platelets showed that the interaction was inhibited by addition of hirugen, suggesting exosite-I-mediated interactions [18]. Other exosite I ligands such as TM or fibrin(ogen) were also shown to block interactions between GpIbα and thrombin [19]. In addition, the exosite-I-deficient autolysis product γ-thrombin did not appear to interact with GpIbα [18]. However, the experiments with fibrinogen and hirugen could not be reproduced by other groups, and subsequent studies showed that GpIbα binding was not blocked by the PAR-1 peptide or the exosite-I-specific aptamer HD1 [20], [21], [22]. Originally, these contradicting results were explained by binding of the various peptides to different subsites of thrombin's exosite I [19]. Crystal structures, however, showed that fibrinogen, TM, hirugen and PAR-1 cover similar regions of exosite I [23], [24], [25], [26]. Arguing against the involvement of exosite I, mutations within the exosite (Arg67Ala, Arg73Glu or Arg77aGlu; chymotrypsin numbering scheme used throughout) were shown not to affect GpIbα binding, and addition of a GpIbα acidic region peptide did not inhibit the exosite-I-mediated activation of the PAR-1 peptide [19], [22], [27]. Consistent with these results, several experiments supported exclusive binding of GpIbα to exosite II. Binding was completely abolished by ligands such as fibrinogen γ′-peptide, heparin and the exosite-II-specific aptamer HD22 and by chemical modification of exosite II [20], [21], [22], [28], [29], [30]. Mutations in exosite II were also shown to weaken binding to GpIbα [22], [27]. The biggest effects were observed for mutations at Arg93, Arg233 and Lys236. Recent biophysical studies, including H/D (hydrogen/deuterium) exchange, showed that the binding of the acidic GpIbα peptide to thrombin resembled fibrinogen γ′-peptide binding and was different from the effects of hirugen, implying interactions with exosite II, but not exosite I [31]. Other researchers found an exosite II interaction in solution but suggested a different binding mode involving both exosites on surfaces, such as surface plasmon resonance (SPR) sensor chips [32].

In 2003, two crystal structures of the GpIbα–thrombin complex were published (1OOK and 1P8V) (Fig. 1c and d) [33], [34]. The structures were of similar GpIbα constructs with two of the three tyrosine residues sulfated, and crystals were obtained after purifying a 1:1 GpIbα–thrombin complex by size-exclusion chromatography. Both crystals contained a single complex in the asymmetric unit and had contacts between crystallographically related GpIbα molecules in both exosites of thrombin (Fig. 1c and d). In 1OOK, the most intimate complex was chosen as the asymmetric unit in spite of the fact that it had no interactions with exosite II. The total contact surface area of the complex was 2120 Å2, with less than half contributed by the acidic C-terminal region. A symmetry-related molecule of GpIbα buried a total of 1093 Å2 but had a larger total contact surface for the acidic region (1046.7 versus 1007.9 Å2). Both exosite contacts are thus plausible, and both involve the acidic C-terminal region (Fig. 1c). In 1P8V, the asymmetric unit again was chosen based on the most intimate complex (1867.6 Å2), but this time, it involved exosite II. A symmetry-related GpIbα molecule buried 1408.2 Å2 and contacted exosite I. However, this exosite I interaction did not involve the acidic region of GpIbα; rather, the LRR region bound towards the “top” of the exosite (Fig. 1d). The authors of both papers concluded that the bivalent contacts observed in their crystals would require thrombin to bind to two separate GpIbα molecules as seen in the crystal contacts. It was hypothesised that interactions with thrombin would effectively cross-link GpIbα molecules and their associated complexes together, either on the same or adjacent platelets.

It is clear from biochemical and structural studies that exosite II is involved, and most agree that it is the principal high-affinity interaction. However, the possibility of an additional exosite I interaction has been supported by multiple biochemical and structural studies and has important implications for the function of thrombin on platelets. The goal of this study is to determine if exosite I of thrombin is involved in GpIbα binding by employing complementary methods of mutagenesis, binding studies, X-ray crystallography and NMR spectroscopy. These methods allowed us to obtain a detailed picture of the interactions between thrombin and GpIbα and show that, even at high concentrations or on a surface, GpIbα binds exclusively to thrombin's exosite II. On platelets, this would leave exosite I free to recognise PAR-1 and other substrates, such as fibrinogen.

Results

Binding of thrombin variants to the GpIbα peptide

In order to determine the binding site of the GpIbα acidic region on thrombin, we performed studies using a fortéBIO Octet Red instrument. This technology uses biolayer interferometry to monitor ligand binding to proteins or peptides immobilised onto glass fibre tips. Measurements are taken in 96-well plates, with results generally comparable to SPR [35]. We fixed biotinylated peptides onto streptavidin sensor tips and monitored binding of S195A thrombin at various concentrations. Dissociation constants were determined by plotting steady-state response versus thrombin concentration (Fig. 2 and Table 1). The importance of the two exosites was tested by creating charge reversal mutations within exosites I and II, Arg73Glu and Arg93Glu, respectively. These residues have been shown by crystallographic and mutagenesis studies to be important for ligand binding at the two exosites [13]. The exosite-I-specific PAR-1 peptide (Fig. 2a) and exosite-II-specific ligand F2 (Fig. 2b) were used as controls. As expected, thrombin bound to the PAR-1 peptide with moderate affinity (Kd of 5.44 ± 1.76 μM). The exosite II mutation had a small effect on binding (Kd of 11.86 ± 1.52 μM), possibly due to the altered electrostatics of the protein surface or allosteric effects of the mutation. The exosite I mutation resulted in a linear response with no sign of saturation even at high thrombin concentrations, and we were able only to get a poor estimate of Kd for PAR-1 binding to this mutant. The effect of mutations on the exosite-II-specific ligand F2 also yielded the expected results. The affinity of thrombin for F2 (Kd of 0.77 ± 0.17 μM) was only slightly affected by the exosite I mutation (Kd of 1.88 ± 0.16 μM), consistent with what was observed with the PAR-1 peptide, whereas the exosite II mutation had a profound effect on binding. After these successful control experiments, we tested the binding of a GpIbα peptide corresponding to Gly268–Thr284, with all three tyrosine residues phosphorylated, mimicking the naturally occurring sulfotyrosines. The exosite I mutation weakened the affinity for the GpIbα peptide (Kd from 1.03 ± 0.05 μM to 5.63 ± 0.69 μM) in line with what was observed for the control ligands, whereas the exosite II mutation resulted in a lack of saturation of signal, even when higher thrombin concentrations were used (up to 40 μM). Taken together, these results suggest that the acidic C-terminal GpIbα peptide binds to thrombin's exosite II with a similar degree of “specificity” as F2, the quintessential exosite II ligand.

Fig. 2

Steady-state binding curves of thrombin to exosite ligands. The binding of PAR-1 peptide (a), F2 (b) and the acidic GpIbα peptide (c) to S195A thrombin (●), and S915A thrombin with additional mutations R73E (■) and R93E (▲) were assessed by plotting steady-state responses against thrombin concentration. Dissociation constants were calculated as indicated in Materials and Methods. Linear or low magnitude responses are indicative of weak binding.

Table 1

Dissociation constants (μM) for thrombin and exosite variants

	Control	R73E	R93E
PAR-1	5.44 ± 1.76	> 70	11.86 ± 1.52
Fragment-2	0.77 ± 0.17	1.88 ± 0.16	> 100
GpIbα	1.03 ± 0.05	5.63 ± 0.69	> 1000

Crystal structures of the thrombin–GpIbα peptide complex

To determine how the acidic GpIbα peptide binds to thrombin when in excess and unconstrained by the LRR region, we established crystal trials with a 10-fold molar excess of peptide over thrombin inhibited with d-phenylalanyl-l-prolyl-l-arginine chloromethyl ketone (PPACK). PPACK reduces thrombin's flexibility and thus PPACK-thrombin has better crystallisation properties than apo-thrombin. Diffraction quality crystals grew in two space groups, C2 and P21, with two and four copies of thrombin in the asymmetric units, respectively. The first crystal grew in 50 mM sodium formate (“low salt”) and diffracted to 1.60 Å. The structure is of high quality and was refined to final Rwork/Rfree of 15.45/18.73% (full crystallographic statistics are reported in Table 2). The asymmetric unit contains two nearly identical thrombin molecules with a Cα RMSD of 0.486 Å (residues 1c–14j, 16–147 and 150–246) and only small differences in the N-terminal region of the light chain and the γ-loop. For both molecules, clear electron density was observed for 7–8 residues (Thr273/Asp274–Pro280) of the GpIbα peptide at thrombin's exosite II (Fig. 3a and b). The second structure, in slightly higher salt concentration (100 mM sodium formate, “high salt”) and space group P21, diffracted to 1.75 Å and was refined to Rwork/Rfree of 20.95/24.23% (Table 2). It contained four thrombin molecules in the asymmetric unit, and as with the other structure, all molecules are nearly identical. However, only 3–4 GpIbα residues (Asp274/Leu275–Asp277) could be built into electron density at thrombin's exosite II for each of the molecules, probably reflecting the decreased affinity for GpIbα at higher salt concentrations. In all six copies of thrombin, the peptide was observed to make contact exclusively with exosite II in spite of the 10-fold molar excess of the peptide, the high concentrations used and the fact that exosite I is not making crystal contacts and is therefore available for peptide binding.

Table 2

Crystals, data processing, refinement and models

	Low salt	High salt
Crystal
Space group	C2	P21
Cell dimensions
a, b, c (Å)	149.57, 50.56, 76.28	74.22, 51.33, 150.08
β (°)	96.13	96.13
Solvent content (%)	37.3	36.9
Data processing statistics
Wavelength (Å)	0.80	0.80
Resolution (Å)	1.60 (1.69–1.60)	1.75 (1.84–1.75)
Total reflections	182,431	275,332
Unique reflections	72,363	105,859
〈I/σ(I)〉	9.6 (1.8)	9.2 (1.7)
Completeness (%)	96.8 (98.4)	93.2 (95.8)
Multiplicity	2.5 (2.5)	2.6 (2.4)
Rmerge	0.075 (0.504)	0.056 (0.448)
Model
Number of protein/other atoms	4680/727	8857/700
Mean B-factor (Å2)	10.3	17.3
Refinement statistics
Reflections in working/free set	68,655/3646	100,532/5305
R-factor/Rfree	15.45/18.73	20.95/24.23
RMSD of bonds (Å)/angles (°) from ideality	0.020/2.07	0.017/1.80
Ramachandran plota (%)
Favoured	97.5	97.7
Outlier	0.0	0.09

Calculated using MolProbity [36].

Fig. 3

Crystal structure of the acidic GpIbα peptide bound to thrombin. All panels are from the low-salt structure. (a) Thrombin is shown in cartoon representation with the GpIbα peptides in magenta and PPACK as yellow sticks. The 2Fo − Fc electron density for the peptide (contoured at 1 × the RMSD of the map) is shown as blue mesh. (b) A close-up of the electron density is shown in stereo. (c) Detail of the binding site with GpIbα residues labelled in red and thrombin side chains interacting with the ligand are shown as sticks and labelled black. (d) Thrombin's exosite II is shown in surface representation and coloured according to the electrostatic potential with the GpIbα peptide as magenta sticks.

The first complex of the low-salt structure is described here in detail because more of the GpIbα peptide is visible in electron density (Fig. 3a and b). The interface between thrombin and GpIbα mainly involves charged interactions between the basic exosite II residues (Arg93, Arg101, Arg126, Lys235, Lys236 and Lys240) and acidic GpIbα side chains (p-Tyr276, Asp277, p-Tyr278 and p-Tyr279). Several hydrogen bonds are also observed between the peptide backbone (carboxyl group of Asp274, Leu275 and p-Tyr276) and thrombin (side chains of Arg126 and Arg233). Additionally, Phe232 of thrombin makes pi-stacking interactions with p-Tyr276 of the peptide (Fig. 3c). The complementarity of the interface is evident when the electrostatic surface of thrombin is shown (Fig. 3d).

NMR chemical shift perturbation experiments

To map the footprint of the GpIbα peptide on thrombin, we conducted NMR chemical shift perturbation experiments. This method is sensitive to local environmental changes and longer-range conformational changes caused by the binding of ligands, even when the affinity is extremely low (Kd in the millimolar range) [37]. The experiments with the GpIbα peptide were based on the NMR chemical shift backbone assignments for PPACK-thrombin that we published in 2010 [38]. PPACK-thrombin was used to reduce the conformational dynamics of thrombin and thus improve the fraction of residues with resonances assigned. PPACK binding in the active site also reduces the allosteric linkage between the two exosites so that direct thrombin–peptide interactions are measured without complicating the data analysis by long-range allosteric effects. Two-dimensional 1H,15N transverse relaxation optimised spectroscopy (TROSY) NMR spectra were recorded with increasing amounts of the phosphorylated GpIbα peptide (0.5-, 1.1- and 5-fold molar ratio compared to thrombin) (Fig. 4a). Only a small subset of resonances is perturbed by the addition of the peptide, confirming the specificity of the interaction between thrombin and the peptide. Affected resonances either are shifted or disappear after addition of the peptide, indicating altered environment or dynamical behaviour (Fig. 4a). The residues with perturbed resonances were mapped onto the structure of thrombin and show that the changes at low (1.1-fold) excess of peptide cluster in exosite II (Fig. 5a). The largest perturbations are observed for typical exosite II residues such as Arg93, the regions around Arg101 and Arg126 and in the C-terminal helix (Phe232–Asp243). At a higher peptide concentration (5-fold excess, ~ 0.5 mM peptide), a few additional perturbations are detectable for residues near exosite I (Fig. 5b). To determine the significance of these additional perturbations, we performed control experiments with the exosite II ligand fibrinogen γ′-peptide (Fig. 5c and d). Comparison of the results with the GpIbα peptide and the fibrinogen γ′-peptide shows that both induce very similar chemical shift perturbations in exosite II. At high concentrations, both peptides induce additional perturbations near exosite I (Fig. 5b versus d). These perturbations, however, differ significantly from those induced by the known exosite I ligand hirugen [38] (Fig. 5e) and are likely explained by the allosteric linkage between the two exosites. These results suggest that the small perturbations in the vicinity of exosite I when using a large excess (500 μM) of the GpIbα and γ′-peptides do not reflect bona fide exosite I interactions. This, however, still leaves open the possibility that the LRR region of GpIbα mediates an exosite I interaction, as found as a crystal contact in 1P8V.

Fig. 4

Two-dimensional (2D) NMR spectra of thrombin with GpIbα. (a) Overlay of 2D TROSY 1H–15N NMR spectra of free thrombin (blue) and thrombin with 0.5-fold (green), 1.1-fold (orange) and 5-fold (red) molar excess of the GpIbα peptide. The residues with the largest chemical shift perturbation with a 1.1-fold addition of GpIbα are labelled in black. Examples of residues with a small perturbation for 1.1-fold excess GpIbα, but a larger perturbation at 5-fold excess of GpIbα are labelled grey. Circles indicate residues with disappearing resonances either due to exchange broadening (indicative of changed dynamics) or due to movement of the cross-peak into a crowded region. (b) Overlay of 2D TROSY 1H–15N NMR spectra of free thrombin (blue) and thrombin with 1.1-fold molar excess (red) of the extracellular domain of GpIbα. The residues with the highest chemical shift perturbation after addition of GpIbα are labelled in black.

Fig. 5

Footprint of chemical shift perturbations plotted on the structure of thrombin. Thrombin crystal structures (heavy chain only) are coloured according to the chemical shift perturbations after addition of the indicated ligand. Residues are coloured from blue to orange with increasing magnitude of chemical shift perturbation. Residues whose resonances appear or disappear after addition of the ligand are coloured red, and those that are unassigned are white. In all cases, thrombin was inhibited by PPACK (yellow sticks) and spectra were recorded in the presence of Na+ (magenta ball). Thrombin with a 1.1-fold excess of the GpIbα peptide (a), 5-fold excess of the GpIbα peptide (b), 1.1-fold excess of the fibrinogen γ′-peptide (c), 5-fold excess of the fibrinogen γ′-peptide (d), 1.1-fold excess of hirugen (e) and 1.1-fold excess of the full extracellular domain of GpIbα (f). Ligands are shown as semitransparent sticks to indicate binding locations observed in crystal structures.

To investigate this, we performed chemical shift perturbation experiments with the extracellular domain of GpIbα, GpIbα(1-289). Because of limited availability and solubility of GpIbα(1-289), we were only able to achieve a 1.1-fold excess of the ligand. The NMR TROSY heteronuclear single quantum coherence spectrum was comparable in quality to that recorded with the peptide ligand (Fig. 4b), and a further TROSY HNCA experiment was run to help re-assign the perturbed resonances. As before, only a specific subset of thrombin resonances is affected by addition of the ligand (Fig. 4b). Mapping the chemical shift perturbations onto the structure of thrombin reveals a very similar pattern to what was observed for the isolated acidic peptide (Fig. 5f). The main changes centre on exosite II, including residues around Arg93 (Ile88, Tyr89, Tyr94), Arg126 (Leu123, Tyr129–Ser129b) and the C-terminal helix (Phe232–Asp243), with a few changes near exosite I. No new perturbations are visible, suggesting that the LRR region does not contribute to the binding of thrombin. In addition, one-dimensional relaxation measurements indicated a 1:1 complex of PPACK-thrombin and GpIbα(1-289), with an apparent molecular mass of 67 kDa, ruling out the possibility of the presence of higher-order complexes. The weighted chemical shift differences for all experiments are given in Supplementary Table 1.

Discussion

It is not possible to review every experiment ever conducted regarding the binding site of GpIbα on thrombin within this manuscript. The attention this issue has received reflects the importance of thrombin and platelets in physiological and pathological blood clotting. Suffice it to say that there is a plethora of biochemical and structural data concerning the GpIbα–thrombin interaction, some experiments properly designed, conducted and interpreted and others that require closer examination. The goal of this work is to see if, by using simple and sensitive biophysical and structural methods, we can clear up the mechanistic controversy that seems to be deeply entrenched regarding the role of exosite I in GpIbα binding.

It is useful to step back and review what we really know about the interaction between thrombin and GpIbα. Firstly, GpIbα is the high-affinity thrombin receptor on platelets, and the interaction is mediated by the C-terminal acidic region of the extracellular domain. The interaction results in a 1:1 complex that can be purified as such by size-exclusion chromatography, even at high concentrations and with an excess of thrombin. It is also accepted that, at least in solution, the principal interaction site for the acidic region of GpIbα is exosite II of thrombin. Several reports employing competition with various “exosite-specific” ligands have also implicated exosite I, and these reports were used to interpret the two crystal structures of the thrombin–GpIbα complex that were published in 2003. At first glance, it is remarkable that the two structures were so very different, since it is reasonable to expect that the important interactions allowing both groups to purify a 1:1 complex would be preserved. Indeed, the exosite II interactions are remarkably similar (Fig. 6a), while the exosite I interactions are completely different, one binding to the opposite side of the acidic peptide and another binding to the LRR domain (see Fig. 1c and d). The six exosite II contacts with the acidic peptide observed in our crystal structures are also very similar to the contacts seen in the first two structures (Fig. 6b). Indeed, there is a remarkable conservation of structure and interactions in a core three-residue motif, 275LY⁎D277 (Y⁎ indicating phosphorylated or sulfated tyrosine), in all eight structures (Fig. 6c). The variability seen in the flanking regions likely reflects the non-directional nature of ionic interactions and the different environments (crystal contacts) in the various crystals. With the simple and reasonable assumption that important contacts will be preserved in all structures, we can conclude that the binding of the 275LY⁎D277 motif of GpIbα to exosite II of thrombin is the core interaction, in good agreement with previously published NMR data that identified residues D274–Y279 as critical for thrombin binding [31]. We also know that the acidic region of GpIbα is flexible relative to the LRR domain. This is evident not only from the two crystal structures of GpIbα with thrombin but also from the other nine structures of the same construct of GpIbα (two copies each for 1GWB [9], 1M0Z [39] and 1QYY [40]; one copy each for 1P9A, 1M10 [39] and 1SQ0 [41]). The acidic C-terminal tail can only be placed in electron density of one of these nine structures (1GWB, chain B), and in that case, it was making extensive contacts with the other copy of GpIbα in the asymmetric unit. Indeed, this lone copy of GpIbα is the structure used to hypothesise a “conformational change” in the C-terminus to allow binding via exosite I [34]. Finally, we know that crystals cannot form without crystal contacts. This means that there will always be interactions that help to stabilise the crystal lattice but do not reflect physiologically relevant protein–protein interactions. Accommodating thrombin in a crystal lattice requires compensating for its two large electrostatically positive exosites, and in 1OOK, this is likely what happened, with the acidic peptide interposing between exosites I and II. There is no evidence that such a bivalent interaction happens in solution. In fact, if it did, thrombin would aggregate in the presence of soluble GpIbα, rendering it impossible to purify the 1:1 complex by size-exclusion chromatography, and precluding NMR studies.

Fig. 6

Stereo views of acidic C-terminal region of GpIbα from crystal structures. The GpIbα peptides from various structures are shown after superposition of the thrombin molecules. (a) The exosite II contacts for the acidic regions (274–280) from 1P8V (green) and 1OOK (cyan) are remarkably similar. (b) The same region as in (a), now including the six thrombin–peptide interactions from the structures reported here (1P8V in green and 1OOK in cyan; molecules 1 and 2 of the low-salt structure in yellow and magenta, respectively. The four other colours indicate the molecules from the high-salt structure). (c) The conserved core is composed of the Leu275-pTyr276-Asp277 tripeptide, colours as in (b).

It has been argued that the bivalent interaction involving thrombin exosites I and II requires a surface, such as that of a platelet membrane or an SPR chip [32]. Here we show that, on a surface (Octet Red tip), thrombin binds to the acidic GpIbα peptide exclusively with exosite II. This was performed using single charge reversal mutations instead of competition; thus, the data are not affected by any lack of specificity of the competing ligand. However, we did not conduct the binding studies using the full-length GpIbα construct; thus, the possibility still remained that the LRR region could mediate further interactions. This was addressed using the sensitive technique of NMR spectroscopy. We found that an excess of peptide and full-length GpIbα at ~ 100 μM thrombin did not perturb resonances in exosite I, and when plotted on the structure of thrombin, the effects were essentially identical with those obtained with the exosite-II-specific ligand, fibrinogen γ′. The peptide-binding interface identified in the NMR experiments is larger than that observed in the crystal structures, likely reflecting the fact that the GpIbα peptide is longer than the maximum of 8 residues observed in the crystal structures and that NMR spectroscopy is able to identify weak and transient interactions and long-range structural perturbations. The chemical shift perturbations close to exosite I observed for both the fibrinogen γ′-peptide and the GpIbα peptide were also identified in a recent H/D exchange study [31] and were explained by allosteric linkage between the two exosites, as has been suggested by some researchers [42] or by non-specific binding to exosite I at high concentrations. The NMR experiments with the extracellular domain of GpIbα confirm the exosite II interaction but do not indicate any additional interactions of the LRR with thrombin's exosite I. This again is in agreement with a recent H/D exchange study [31]. It should be noted that NMR is capable of detecting interactions with Kd values in the millimolar range. Our data therefore argue strongly against any exosite I interaction with GpIbα in solution.

Taken together, our experiments add to a large body of data to show that GpIbα binds exclusively to thrombin's exosite II. This binding mode does not allow any additional non-catalytic function of thrombin on platelets, such as receptor cross-linking or platelet adhesion. Instead, GpIbα binding localises thrombin activity to the platelet surface, and by utilising exosite II for GpIbα binding, thrombin is free to engage PAR-1 and other substrates dependent on exosite I.

Materials and Methods

Proteins and peptides

Wild-type thrombin was expressed in BL21Star(DE3)pLysS Escherichia coli as prethrombin-2 in inclusion bodies, re-solubilised, refolded and purified using heparin Sepharose (GE healthcare), as described previously [43]. It was activated with PMSF-treated Echis carinatus venom for 1 h at 37 °C and inhibited with a 5-fold molar excess of PPACK for 30 min at room temperature. PPACK-thrombin was repurified with heparin-Sepharose. Thrombin mutants were generated using the QuikChange site-directed mutagenesis kit (Stratagene) with pET23-prethrombin-2 (S195A) as template. The GpIbα peptide (Ac-Gly268-Thr284-NH2), N-terminally biotin-labelled GpIbα peptide (biotin-Gly268-Thr284-NH2), the PAR-1 peptide (Ac-Ser16-Asn36-NH2) and the fibrinogen γ′-peptide (Ac-Val408-Leu427) were synthesised by Peptide Protein Research Ltd. (Fareham, UK). All peptides contained phosphorylated tyrosine residues instead of the naturally occurring sulfotyrosine. This is not expected to affect binding to thrombin [44], [45]. Hirugen (hirudin Gly54-Gln65) was purchased from Sigma or Cambridge Bioscience (Cambridge, UK). Fragment-2 was generated from plasma-derived prothrombin using human fXa (Haematologic Technologies). The GpIbα(1-289)-calmodulin fusion protein was a kind gift from Willem Ouwehand (Department of Haematology, University of Cambridge) and was expressed in Sf9 cells as previously described [46]. GpIbα was cleaved from calmodulin using porcine pancreatic elastase (1:60, w/w) in phosphate-buffered saline overnight, and GpIbα was separated from calmodulin using anion-exchange chromatography equilibrated in 20 mM Tris (pH 8) and 50 mM NaCl. Proteins were eluted with a NaCl gradient from 50 to 1000 mM over 10 column volumes. The PAR-1 peptide and fragment-2 were biotin-labelled using the EZ-Link NHS polyethylene glycol (PEG)-4 biotin kit (Thermo Scientific) with the protocol for preferential labelling of the N-terminus [47].

Binding assays

All experiments were performed in 20 mM Hepes (pH 7.4), 150 mM NaCl, 5 mM CaCl2, 0.1 mg/ml bovine serum albumin and 0.002% Tween20 (binding buffer) on an Octet RED instrument (fortéBIO) with streptavidin biosensors. Biotin-labelled peptides were loaded onto the biosensors at 1 μg/ml, and the biotin-fragment-2 was loaded at 5 μg/ml. S195A thrombin control and exosite variants were diluted in binding buffer with concentrations between 0.15 and 40 μM. Data traces were analysed using fortéBIO Analysis software and Prism (GraphPad). Steady-state responses were plotted versus concentration and fit to a 1:1 specific binding model to obtain dissociation constants (Kd).

Crystallisation

PPACK-inhibited thrombin (4.6 mg/ml) in 20 mM Tris (pH 7.4) and 100 mM NaCl was added to a 10-fold molar excess of lyophilised GpIbα peptide. Initial screens were set up in MRC 96-well plates using the sitting-drop vapour diffusion method with 200 nl protein solution and 200 nl reservoir solution. Crystals appeared after 3–4 days in 100 mM sodium formate and 12% (w/w) PEG-3350 and were optimised using a hanging-drop vapour diffusion grid screen with 1 μl + 1 μl drops. After 2 days, crystals from two conditions were harvested. Condition 1 contained 50 mM sodium formate and 16% PEG-3350, and condition 2 contained 100 mM sodium formate and 18% PEG-3350. Crystals from both conditions were cryoprotected in 100 mM sodium formate and 18% PEG-3350 and increasing amounts of glycerol up to 20% and were cryo-cooled in liquid nitrogen. Data were collected at 100 K on beamline I03 at Diamond Light Source (Didcot, UK). Diffraction markedly improved after annealing by blocking the cryo-stream three times for 3 s. Data were processed using Mosflm, Scala and Truncate [48] from the CCP4 [49] package, and structures were solved by molecular replacement using Phaser [50] with chains A and B from structure 1JOU [51] as search model. Rebuilding and refinement were conducted using Coot [52] and Refmac [53]. All figures were produced with PyMOL [54]. Crystallographic data for the two structures are given in Table 2, and the coordinates and structure factors have been deposited in the Protein Data Bank under accession codes 4ch2 and 4ch8.

NMR spectroscopy

For NMR spectroscopy, 2H, 13C, 15N-labelled prethrombin-2 was expressed, purified and activated as described previously [38]. PPACK-inhibited wild-type thrombin at a concentration of 80–100 μM in 50 mM sodium phosphate (pH 7.4), 5% D2O and 0.02% NaN3 was used for all NMR experiments. Increasing amounts of the GpIbα peptide or the fibrinogen γ′-peptide were added to the NMR sample from highly concentrated stock solutions to minimise sample dilution. For experiments with GpIbα(1-289), the protein was added to a thrombin solution at 1.1-fold excess and then concentrated and dialysed into NMR buffer. A TROSY-based HNCA experiment was measured to re-assign the shifted resonances. All NMR spectra were recorded on a Bruker Avance II + 700-MHz spectrometer with a TCI triple resonance cryoprobe at 37 °C, processed in TopSpin 2.1 (Bruker) and analysed in Sparky (University of California, San Francisco). Weighted chemical shift perturbations [55] were calculated with Δδ(1H) + Δδ(15N)/5. Assignments for the reference spectrum of unliganded PPACK-thrombin and PPACK-thrombin with hirugen were taken from our previous work [38]. The molecular weight of the thrombin–GpIbα(1-289) NMR sample was estimated using a 1H T2 relaxation experiment based on the Hahn spin-echo pulse sequence [56] with relaxation delays of 2 and 0.4 ms. The T2 relaxation time is determined by the rate of signal attenuation during the two relaxation delays. From this, the rotational correlation time τc can be determined [τc ≈ 1/(5T2) ns, where T2 is expressed in seconds] [57]; τc in turn is directly proportional to the molecular weight of the sample.

Accession numbers

The atomic coordinates and structure factors have been deposited in the PDB under accession codes 4ch2 and 4ch8.

The following are the supplementary data related to this article.

Supplemental Table 1

Weighted chemical shift differences.