Molecular basis of thrombomodulin activation of slow thrombin

Summary

Background

Coagulation is a highly regulated process where the ability to prevent blood loss after injury is balanced against the maintenance of blood fluidity. Thrombin is at the center of this balancing act. It is the critical enzyme for producing and stabilizing a clot, but when complexed with thrombomodulin (TM) it is converted to a powerful anticoagulant. Another cofactor that may play a role in determining thrombin function is the monovalent cation Na⁺. Its apparent affinity suggests that half of the thrombin generated is in a Na⁺-free ‘slow’ state and half is in a Na⁺-coordinated ‘fast’ state. While slow thrombin is a poor procoagulant enzyme, when complexed to TM it is an effective anticoagulant.

Methods

To better understand this molecular transformation we solved a 2.4 Å structure of thrombin complexed with EGF domains 4–6 of TM in the absence of Na⁺ and other cofactors or inhibitors.

Results

We find that TM binds as previously observed, and that the thrombin component resembles structures of the fast form. The Na⁺ binding loop is observed in a conformation identical to the Na⁺-bound form, with conserved water molecules compensating for the missing ion. Using the fluorescent probe p-aminobenzamidine we show that activation of slow thrombin by TM principally involves the opening of the primary specificity pocket.

Conclusions

These data show that TM binding alters the conformation of thrombin in a similar manner as Na⁺ coordination, resulting in an ordering of the Na⁺ binding loop and an opening of the adjacent S1 pocket. We conclude that other, more subtle subsite changes are unlikely to influence thrombin specificity toward macromolecular substrates.

Introduction

Thrombin is the central enzyme of the hemostatic cascade, generating the initial platelet-rich plug via cleavage of protease-activated receptors, PARs 1 and 4, stabilizing the clot through cleavage of fibrinogen, and increasing production of itself by activating cofactors V and VIII [1,2]. With a classic chymotrypsin family serine protease fold, thrombin contains a deep negatively charged active-site cleft and two basic exosites that help direct its multiple interactions with substrates and cofactors [3–5]. Following formation of a stable blood clot, procoagulant activity is attenuated by cleavage inactivation of cofactors [6], inhibition of serine proteases by serpins [7], and initiation of the fibrinolytic system [8]. A critical step in this attenuation is the activation of protein C by the thrombin–thrombomodulin (TM) complex. Activated protein C (APC) inactivates cofactors Va and VIIIa to shut down further thrombin generation [9]. The complex also stabilizes the formed clot by activating the thrombin-activated fibrinolysis inhibitor (TAFI), which delays fibrin degradation and links coagulation with fibrinolysis [10].

TM is an integral membrane glycoprotein expressed on the surface of the vascular endothelium [11]. The extracellular portion of the protein consists of an N-terminal lectin domain followed by six epidermal growth factor-like repeats (EGF1–6) and a proteoglycan rich region. A chondroitin sulfate moiety within the proteoglycan region has been shown to increase the affinity of TM for thrombin by ~10-fold [12,13]. The TM binding site for thrombin has been identified using limited proteolysis and alanine-scanning mutagenesis as residues within EGF5 and EGF6 [14]. Biochemical and mutational analysis identified exosite I on thrombin as critical for TM binding [15–17]. A recent study showed that the basic residues that make up exosite I are not required for the interaction [18]; instead several hydrophobic residues within EGF5 and EGF6 of TM and exosite I of thrombin mediate the stability of the complex.

When complexed with TM, thrombin becomes incapable of cleaving most of its macromolecular substrates; however, the rate of activation of protein C is increased by three orders of magnitude [6,19]. The precise mechanism by which this switch in substrate specificity occurs is unclear. The tight binding of TM to exosite I is sufficient to out-compete most procoagulant substrates that depend on the same site, but there is also evidence that conformational changes in and around the active site of thrombin may play a role in favoring protein C cleavage [20–23]. In 2000, the crystal structure of the complex between human thrombin and the minimal cofactor fragment of TM, EGF4–6 (TM456), revealed the features of the thrombin-TM interaction and corroborated earlier biochemical information [24], although the presence of a covalently bound active-site thrombin inhibitor clouded any potential allosteric effects of TM binding.

Another regulator of thrombin activity is Na⁺ [25]. Na⁺-free (‘slow’) thrombin has reduced efficiency for the cleavage of many substrates relative to the Na⁺-bound (‘fast’) form. Structures of wild-type and mutant thrombins in the absence of Na⁺ revealed disruption of the Na⁺ binding site, reordering of the active site loop, flipping of Trp215, and blockage of the S1 pocket [26]. These structural changes would predictably inactivate thrombin. However, when the slow form binds to TM its ability to activate protein C is restored, and variants (including W215A/E217A [27] and E217K [28]) that are deficient in Na⁺ binding show only a moderate reduction in protein C activation in the presence of TM. To determine how TM converts slow thrombin from an inactive enzyme into an active anticoagulant we crystallized the thrombin-TM complex in the absence of Na⁺ and solved the structure to 2.4 Å resolution. The structure shows that binding of TM allosterically reorders the Na⁺ binding loop so that it resembles the Na⁺-bound state and opens the active site cleft, rendering it competent for substrate binding and hydrolysis. The potential functional role of TM-induced active site changes in thrombin is discussed.

Experimental

Reagents

Crystallization reagents were purchased from Hampton Research (Aliso Viejo, CA, USA). Buffers and salts were purchased from either Sigma-Aldrich (Gillingham, UK) or BDH (Poole, UK).

Thrombin expression and purification

The catalytically inert thrombin mutant (S195A) was produced as described previously [29]. Briefly, wild-type human pre-thrombin-2 in the pET23 (+) vector (Novagen, Beeston, UK) was converted to the S195A (chymotrypsin numbering) variant using site-directed mutagenesis and transformed into the BL21 STAR (DE3) pLysS E. coli strain (Novagen). The protein was expressed into inclusion bodies then solublized in 10 mL 20 mM Tris–HCl, pH 8.0, 6 M guanidine–HCl, 5 mM reduced glutathione, 2 mM oxidized glutathione, 0.5 mM EDTA. Following refolding by dropwise addition into 100 volumes of 50 mM Tris–HCl, pH8.5, 0.6 ML-arginine, 20 mM CaCl₂, 10% glycerol, 0.2% Brij-58 the protein was loaded onto a 5 mL HiTrap heparin Sepharose column (GE Healthcare, Little Chalfont, UK) and eluted with a 0.25–1 M NaCl gradient over five column volumes. Fractions containing prethrombin-2 were pooled and activated by overnight incubation with 1:50 (w/w) E. carinatus venom at room temperature. The thrombin was re-purified on heparin Sepharose using the conditions mentioned above, concentrated, and extensively dialyzed against 20 mM Tris, pH 7.4, 100 mM LiCl.

TM expression and purification

TM456 was expressed into the periplasm of the BL21 (DE3) E. coli strain using the pET39b vector (Novagen) as a fusion protein with DsbA and purified as described previously [30]. To minimize oxidation [31], the Met at position 388 was mutated to Leu [32], and to prevent proteolysis Arg456 and His457 were mutated to Gly and Glu, respectively (as previously [24]). Briefly, following protein expression overnight at 25 °C, the periplasmic extract was prepared. The extract was then loaded onto a 5 mL HiTrap Chelating column (GE Healthcare) charged with Ni²⁺ ions and eluted with 20 mM Tris, pH 7.4, 1 M NaCl, 250 mM imidazole, 2.5 mM CaCl₂. The fusion protein was then buffer exchanged into 20 mM Tris, pH 7.4, 100 mM NaCl, 5 mM CaCl₂ and digested with 0.5 U recombinant factor Xa per mg fusion protein at 37 °C for 24 h. The TM456 was then loaded onto a 1 mL HiTrap IMAC column to bind the proteolysed DsbA fusion tag. The flow-through was then diluted with water and loaded onto HiTrap Q-Sepharose, followed by elution with a 0.1–0.5 M NaCl gradient over 10 column volumes. Fractions containing TM456 were pooled, concentrated and extensively dialyzed against 20 mM Tris, pH 7.4, 100 mM LiCl, 5 mM CaCl₂.

Fluorescence studies

Fluorescence spectra were collected at room temperature on a PerkinElmer Life Sciences 50B fluorometer, exciting at 345 nm and recording emission from 355–420 nm with slits of 3 and 5 nm, respectively. Background spectra of 50 μM p-aminobenzamidine (pAB) in phosphate buffered saline (20 mM NaPi or LiPi, pH 7.4, with 150 mM NaCl or LiCl, plus 0.1% PEG8000) were subtracted from those collected in the presence of 1.26 μM thrombin to obtain the pAB signal reporting the state of the S1 pocket. The effect of TM456 was determined by stepwise additions to ensure saturation. Dissociation constants (Kd) for pAB binding to thrombin and the thrombin-TM456 complex in the presence of 150 mM LiCl or NaCl were obtained essentially as previously [33]. Briefly, pAB was titrated into solutions containing 1 μM thrombin in 20 mM MPi, pH 7.4, with 150 mM MCl, 0.1% PEG8000, where M = Li⁺ or Na⁺. Readings were averaged over 30 s while exciting at 345 nm and recording emission at 370 nm. Fluorescence change was plotted against pAB concentration and fitted to a one-site specific binding equation using PRISM (Fig. S1).

Crystallization and data collection

The S195A thrombin-TM456 complex was prepared by mixing the proteins at 1:1 ratio followed by purification on S75 Superdex 10/30 (GE Healthcare) in 10 mM Tris, pH 7.4, 100 mM LiCl, 2 mM CaCl₂. The complex was concentrated to 4.5 mg/mL and mixed 1:1 with precipitant containing 0.2 M lithium sulfate and 22% PEG3350. The drops were placed over a reservoir containing 50% PEG3350 and diffraction quality crystals were observed after 7–14 days. Crystals were flash cooled in liquid N₂ and data were collected at SRS beamline 14.2 (Daresbury, UK) and indexed with Mosflm [34]. The data were processed with the CCP4 program suite using Scala and Truncate [35] followed by molecular replacement using Phaser [36]. Initial refinement was carried out iteratively with successive rounds of model building using XtalView [37] or Coot [38] and refinement with CNS [39], using NCS restraints throughout. Data processing and refinement statistics are given in Table 1. Figures were made using Pymol [40]. Thrombin numbering is based on chymotrypsin, with insertion loops indicated by sequential letters. Coordinates and structure factors are deposited in the Protein Data Bank under PDBID code 3GIS.

Table 1.

Data processing and refinement statistics

Crystal
Space group	P212121
Cell dimensions (Å)	a = 66.25, b = 100.34, c = 229.28, α = β = γ = 90
Solvent content (%)	53.6
Data processing
Wavelength (Å)	0.98 (Daresbury SRS 14.2)
Resolution (Å)	92.5–2.40	2.53–2.40
Total reflections	269 396	23 632
Unique reflections	57 921	7315
Multiplicity	4.7	3.2
< I/σ(I) >	8.9	3.5
Completeness (%)	96.0	95.0
Rmerge	15.0	33.9
Model details
# of atoms:
Protein	10 192
Water	575
Ca+2 ions	3
${\text{SO}}_4^{-2}$ ions	15
Average B-factor (Å2)	28.2
Refinement statistics
Resolution (Å)	40.0–2.40	2.49–2.40
Reflections (working/free)	57 399/2875	5004/240
Rfactor /Rfree (%)	21.1/25.9	25.8/29.6
r.m.s. deviation from ideality bonds (Å)/Angles(°)	0.006/1.34
Ramachandran plot (%)
Most favoured	84.0
Additionally allowed	14.7
Generously allowed	1.3
Disallowed	0.0

Results

Overall structure

Using S195A thrombin (1JOU [41], molecule AB) and TM456 (1DX5 [24], molecule I) as search models, molecular replacement identified three copies of thrombin but no solutions were found for TM456. Deletion of EGF4 led to a successful molecular replacement, and the three copies of the 4th EGF domain were independently built into electron density. The structure was refined to 2.4 Å with a final R_factor of 21.1% and an R_free of 25.9% (Table 1). The three complexes constituting the asymmetric unit are shown in Fig. 1(A, B). Superposition of the complexes revealed identical thrombin structures with Cα RMSDs of 0.21 and 0.47 Å and all atom RMSDs of 0.59 and 0.78 Å, when the second and third thrombins were compared with AB. The contacts between thrombin and EGF domains 5 and 6 were also essentially identical for all three complexes (discussed later). The orientation of the EGF4 domains with respect to the rest of TM, however, was significantly different for each of the complexes (Fig. 1C). In fact, a rotation of 20° resulted in a 13 Å shift of EGF4 relative to the other two EGF domains, demonstrating significant flexibility in the linker region between the 4th and 5th EGF domains. This is in contrast to the previously solved thrombin-TM456 structure (1DX5), which showed identical TM456 conformations for the four complexes in the asymmetric unit, with a fixed 90° angle between EGF4 and the other domains [24]. All three EGF4 positions in our structure are all significantly different from what was previously observed. Our structure does not suffer from major crystal contacts involving EGF4 and thus may reflect more accurately its natural range of motion. Consistent with this, the B-factors of the 4th EGF domain are significantly higher than those of domains 5 and 6 (Fig. 1B).

Fig. 1.

Structure of the thrombin-TM456 complex. (A) Ribbon diagram of the three complexes in the asymmetric unit, with thrombin coloured grey and TM456 coloured from N-to-C terminus from blue (EGF4) to red (EGF6), with magenta balls indicating Ca²⁺ions. (B) The same image as in (A), but coloured according to B-factor to illustrate the inherent flexibility of the 4th EGF domains. (C) A stereo view of the three complexes superimposed reveals the flexibility of EGF4 relative to the other EGF domains and thrombin (grey surface representation).

Thrombin-TM456 interaction

In spite of the flexibility of EGF4, the interface involving anion-binding exosite I of thrombin and the 5th and 6th EGF domains of TM is essentially identical for the three complexes and when compared with the previous structure. Positively charged residues within exosite I, including Lys 36, 81 and 110 and Arg 67, all interact with TM456, but only Lys 110 forms a salt-bridge (to Asp 461). These and other basic residues primarily serve to steer the electronegative face of TM to the proper binding site, as previously shown for other exosite I binding proteins [42]. The interface between TM and thrombin is dominated by hydrogen bonding (10 in total, involving residues 407–409, 416–417, 425 and 428–429 of TM with 36, 36A, 38, 74, 76, 81 and 82 of thrombin), and a hydrophobic core composed of residues 414, 415, 424 and 432 from TM and 34, 38, 65, 67, 76, 82 and 84 of thrombin. Highlighting the importance of these hydrophobic contacts, mutation of Ile 414 and 424 and Leu 415 of TM to alanine results in a two-order of magnitude reduction in affinity for thrombin [43].

Na⁺-binding site

Na⁺ was rigorously excluded from protein and precipitant solutions before and during crystallization, and, as expected, when the structure was solved there was no electron density corresponding to coordinated Na⁺ (Fig. 2A). However, in all three thrombins the Na⁺ binding loop was found in a conformation indistinguishable from that of the Na⁺-bound state (Fig. 2A). This is similar to what was observed previously in the absence of Na⁺, either with or without an exosite I binding ligand or inhibitor [44,45]. Neither the Na⁺ binding loop nor the 186-loop in the structure is restricted by crystal contacts, suggesting that the conformation of this loop is the result of TM binding. Although Na⁺ is absent, waters that normally surround Na⁺ are observed in all three complexes. These waters are also conserved in other Na⁺-free thrombin structures with formed Na⁺ binding loops (1SGI [44], 1SHH [44], and 1HXF [45]) (Fig. 2B). The four waters that coordinate Na⁺ are present in these structures, with one observed migrating toward the Na⁺ site to some extent, effectively substituting for the Na⁺ ion. The waters forming the second shell around the normal Na⁺ position are conserved in all structures in the absence of Na⁺, in spite of variable active site occupancy. Thus Na⁺ is not required for the correct folding of this region, or for the formation of the water channel previously hypothesized to be the result of Na⁺ coordination [44].

Fig. 2.

Na⁺ binding region and water clusters. (A) A stereo view of the Na⁺ binding loop (residues 221–225) from the current structure (yellow rods) shows an identical conformation as a representative Na⁺ bound structure (1JOU, semitransparent pink). A simulated annealing map generated in the absence of waters in the vicinity of the Na⁺ binding site (2Fo−Fc contoured at 1σ in blue, and Fo−Fc contoured at 3σ in green) proves the absence of Na⁺ (purple ball) and the presence of several water molecules (red balls). (B) Stereo view of the Na⁺ binding site and surrounding region of a Na⁺ coordinated state (semitransparent yellow rods for residues 184–191 and 215–225, with Na⁺ in purple, waters in red, and coordinating bonds as red lines). Waters coordinating Na⁺ and surrounding the site are conserved in Na⁺-bound structures (red) in several clusters. These are surprisingly conserved even in the absence of coordinated Na⁺, with those from the current structure in blue, orange and yellow, from 1HXF in cyan, from 1SGI in magenta and green, and from 1SHH in pink and grey. The only adjustment appears to be the migration of certain waters closer to the Na⁺ site to compensate for the missing ion.

Active site cleft

We set out to obtain a structure of active site free slow thrombin bound to TM. However, analysis of the structure revealed the presence of a crystal contact involving the C-termini of the light chains binding in a substrate-like fashion into the active sites of crystallographically related thrombins. We are thus unable to determine the precise active site rearrangements that occur solely in response to TM binding from this structure. Previous work suggested that the conformational effect of TM binding is similar to that induced by Na⁺ coordination for slow thrombin [46]. This was tested here using a probe specific for the S1 pocket of thrombin [33], p-aminobenzamidine (pAB, Figs 3 and S1). As expected from our earlier structure of Na⁺-free thrombin [47], the S1 pocket of slow thrombin is significantly less able to bind the probe, as reflected by a ~2-fold increase in Kd for pAB and a lower fluorescence at a fixed pAB concentration relative to the fast form. A similar improvement in pAB affinity upon Na⁺ coordination (from 60.5 to 37.3 μM) was seen upon binding of TM456 (from 60.5 to 21.5 μM). The addition of TM to Na⁺-bound thrombin did not result in any change in Kd for pAB (from 37.3 to 34.7 μM) or an additional fluorescence increase, indicating that the accessibility/properties of the S1 pocket is identical for the two ligands. We thus conclude that the opening of the S1 pocket is a major consequence of TM binding to slow thrombin.

Fig. 3.

Fluorescence studies probing the S1 pocket of thrombin. (A) The p-aminobenzamidine/thrombin fluorescence signal increases by about 2-fold upon changing conditions from 150 mM LiCl (solid red) to 150 mM NaCl (solid black). The same increase is seen upon addition of saturating amounts of TM456 (dashed red). However, addition of TM to thrombin coordinated by Na⁺ (dashed black) does not result in any further fluorescence enhancement. (B) A similar spectrum taken with prethrombin-2 in the presence of Na⁺ and TM (dashed green) indicates an obscured S1 pocket, consistent with what has been seen in the crystal structure [54]. The E217K variant in the presence of Na⁺ (solid blue) gives a similar baseline fluorescence as prethrombin-2, again consistent with the crystal structure that showed a blocked S1 pocket [54]. However, when TM was added (dashed blue) pAB binding was restored to about half the level of slow thrombin (solid red).

Discussion

This study had two goals: first, to determine how the binding of TM to exosite I activates slow thrombin; and second, to see if TM binding alters the fine features of the substrate binding pocket of thrombin. We crystallized the complex under conditions that rigorously excluded Na⁺ or other monovalent cations known to coordinate thrombin. Accordingly, the structure revealed an absence of Na⁺; however, the active site was open and the conformation of the Na⁺ binding loop was identical to structures of the fast form. This supports a conformational link between exosite I and the Na⁺ loop of thrombin, so that occupancy of either site effects a similar conformational change in slow thrombin [46]. The details of this conformational transition will vary depending on what crystallographic structure is taken to represent the slow form, but there are two features shared by all likely candidates (including variants deficient in Na⁺ binding) [26] – an obscured S1 pocket and a non-catalytic oxyanion hole. These features are also shared by the zymogen forms of serine proteases [48]. Accordingly, pAB does not bind to the zymogen form of thrombin, prethrombin-2, even in the presence of Na⁺ and TM(Fig. 3B). Thrombin variant E217K is considered constitutively slow due to its deficiency in Na⁺ binding and its exceedingly low catalytic activity (0.4% activity vs. fibrinogen [28]). However, when bound to TM, E217K thrombin recovers to about half the activity of slow thrombin. This is supported by our pAB binding study, which shows an increase from baseline to about half the fluorescence of slow thrombin in the presence of saturating amounts of TM (Fig. 3B). These data, together with our crystal structure, support the conclusion that the poor catalytic activity of thrombin in the absence of Na⁺ reflects an equilibrium between inert and active states [26]. Mutations and ligands, such as TM, can affect this equilibrium, either favouring the inert or the active state.

Early attempts to explain the 1000-fold improvement in protein C cleavage by thrombin in the presence of TM focused on possible allosteric changes in the active site of thrombin. The unusual P3 and P3′ Asp residues suggested an altered substrate binding mode, and fluorescent active site probes reported significant changes in quantum yield upon TM binding [20]. Subsequent reports revealed modest changes in hydrolysis rates for certain chromogenic substrates in the presence of TM or hirugen (the C-terminal, exosite I binding peptide from hirudin), solely through an improvement in k_cat [21]. However, subsequent studies showed that the unfavourability of the P3 Asp was not mitigated by TM binding [49], and that TM did not generally enhance the rate of small substrate hydrolysis by thrombin under physiological Na⁺ concentrations [50]. It is thus reasonable to conclude, based on the biochemical evidence preceding the first structure of the thrombin-TM456 complex, that TM subtly alters the conformation of thrombin, but that this ‘allosteric’ effect does not in itself transform substrate preference. This is supported by recent structures of thrombin with natural substrates, that show the P3 and P3′ residues pointing out into solution making no contacts with thrombin [30,51,52]. The negative effect of Asp at these positions is likely to be due solely to the electronegative character of thrombin’s active site cleft; no conformational change could alter that.

There are, however, persistent suggestions of allosteric changes in the S2 pocket of thrombin due to exosite I binding ligands, such as TM, with respect to the conformation of the 60-loop and Trp60D in particular. The original crystal structure of EGRCK-inhibited thrombin bound to TM revealed a slightly larger than normal S2 pocket [24], and a recent structure of murine thrombin bound to an exosite I binding peptide based on PAR3 showed a wide-open S2 with a flipped Trp60Dindole ring [53]. However, while there are no crystal contacts involving the 60-loop in the thrombin-TM structure (1DX5), both apo and PAR3-bound murine thrombin have extensive contacts in the 60-loop. Indeed, Trp60Ditself makes contacts with a symmetry-related PAR3 ligand (Fig. S2), rendering any conclusions concerning allostery problematic at best. In addition, structures of thrombin bound to macromolecular substrates (serpins) show that the 60-loop tends to mediate direct contacts with regions outside the reactive centre loop. Thus, the conformation of the 60-loop and Trp60D in particular will depend on active site and exosite interactions, in a manner that responds to the particular macromolecular substrate in question. A dramatic example of this is provided by our recent structure of thrombin bound to protein C inhibitor [30]. We conclude that the active site of thrombin is generally receptive to peptides and can alter its shape in a plastic fashion to accommodate a large variety of substrate sequences. Other factors, such as steric interference, co-localization and exosite interactions are more likely to govern thrombin recognition of physiological substrates, and to account for the ability of TM to convert thrombin from a pro- to an anticoagulant.

Supplementary Material

1

Figure S1. Representative pAB titrations of thrombin in the presence and absence of TM.

Figure S2. Stereo view of the crystal contacts involving the 60-loop in 2PUX.