Previous crystal structures of thrombin indicate that the 60‐insertion loop is a rigid moiety that partially occludes the active site, suggesting that this structural feature plays a decisive role in restricting thrombin‘s specificity. This restricted specificity is typified by the experimental observation that thrombin is not inhibited by micromolar concentrations of basic pancreatic trypsin inhibitor (BPTI). Surprisingly, a single atom mutation in thrombin (E192Q) results in a 10−8 M affinity for BPTI. The crystal structure of human thrombin mutant E192Q has been solved in complex with BPTI at 2.3 Å resolution. Binding of the Kunitz inhibitor is accompanied by gross structural rearrangements in thrombin. In particular, thrombin's 60‐loop is found in a significantly different conformation. Concomitant reorganization of other surface loops that surround the active site, i.e. the 37‐loop, the 148‐loop and the 99‐loop, is observed. Thrombin can therefore undergo major structural reorganization upon strong ligand binding. Implications for the interaction of thrombin with antithrombin and thrombomodulin are discussed.
The actions of the serine proteinase α‐thrombin (EC 188.8.131.52) are central to haemostasis (Davie et al., 1991). Thrombin is a versatile enzyme, exhibiting both pro‐ and anti‐coagulatory effects: the former through its conversion of fibrinogen to fibrin, its activation of the platelet thrombin receptor and several coagulation cofactors, the latter via the thrombomodulin‐mediated activation of protein C (Esmon, 1989). Thrombin catalytic activity is controlled by antithrombin and other serpins (Olson and Björk, 1992), but is notably unaffected by the Kunitz‐type inhibitors basic pancreatic trypsin inhibitor (BPTI) and tissue factor pathway inhibitor (TFPI) (Ascenzi et al., 1988; Guinto et al., 1994).
Despite its many substrates, thrombin exhibits a narrow specificity in its cleavage sites compared with the related enzyme trypsin. Structural studies on the thrombin molecule reveal the characteristic fold of the trypsin‐like serine proteinases (Bode et al., 1989, 1992). A major part of thrombin‘s restricted specificity may come from two pronounced insertion loops that border the active site cleft: the ’60‐insertion loop‘, containing residue Trp60D (chymotrypsinogen numbering) and the ’148‐insertion loop', containing residue Trp148 (see Figure 3). Comparison of structures of thrombin in complex with a wide variety of substrates and inhibitors indicates that the 148‐loop can adopt a range of conformations (Priestle et al., 1993; Stubbs and Bode, 1993); the 60‐loop, on the other hand, appears to possess a rigid structure that differs by <1.5 Å in previously determined structures (Engh et al., 1996). Haematophages such as the medicinal leech Hirudo medicinalis, the assassin bug Rhodnius prolixus and the soft tick Ornithodoros moubata all possess potent thrombin inhibitors that display novel inhibition strategies, accommodating themselves to the restricted active site with concomitant binding to the other unique feature of thrombin, the basic fibrinogen recognition exosite (Rydel et al., 1991; van de Locht et al., 1995, 1996).
The actions of thrombin have hitherto been explained in terms of a static molecular model (Stubbs and Bode, 1993, 1995). Although there have been some reports on allosteric regulation of thrombin activity [linkage between the active site and the fibrinogen recognition exosite (Parry et al., 1993; De Cristofaro et al., 1995)], the sodium‐mediated fast–slow transition (Wells and Di Cera, 1992), and the thrombomodulin‐dependent activation of protein C (Ye et al., 1991), movements in the thrombin molecule have been assumed to be minimal. In particular, it has been an implicit assumption that the 60‐loop represents a rigid feature of thrombin, which would explain its poor inhibition by the paradigmatic serine proteinase inhibitor BPTI (Ascenzi et al., 1988; Bode et al., 1992). Deletion of the 60‐insertion loop (thrombin desPPW) results in nanomolar inhibition by both BPTI (Le Bonniec et al., 1993) and TFPI (Guinto et al., 1994), while deletion of the 148‐loop (thrombin desETW) results in a modest 25‐fold decrease in Ki for BPTI compared with wild‐type thrombin (Le Bonniec et al., 1992).
The activation of protein C is an important regulator of thrombotic activity (Esmon, 1989). In the presence of thrombomodulin, protein C becomes the major substrate for thrombin; in its absence, however, protein C is a poor substrate. In an attempt to explain this behaviour, it was suggested that the acidic P3 and P3′ side chains in protein C might be repulsed by residues Glu192 or Glu39 of thrombin (Le Bonniec and Esmon, 1991; Le Bonniec et al., 1991). Accordingly, thrombin mutants E192Q (Le Bonniec and Esmon, 1991) and E39K (Le Bonniec et al., 1991) activate protein C 22 and 2.2 times more rapidly than wild‐type thrombin, respectively. These data provided the basis for a simple mechanical model for the switching of thrombin activity upon thrombomodulin binding; the binding induced a movement of Glu192 and/or Glu39, which avoided unfavourable interactions with the substrate (Stubbs et al., 1992). Quite unexpectedly, however, the single atom mutation E192Q, originally proposed by Ascenzi et al. (1988), allowed the binding of BPTI and TFPI, with Ki values in the 10−8 M range (Guinto et al., 1994). Docking of BPTI into the active site of thrombin revealed severe steric clashes on complex formation (Bode et al., 1992). Clearly, either E192Q–BPTI did not exhibit the ‘canonical interaction’, or Trp60D had to move.
The major physiological regulator of thrombin is the serpin antithrombin (Olson and Björk, 1992). Serpins exhibit a complex inhibition mechanism, in which an initial loose association of the proteinase and serpin evolves into a tight irreversible complex. The interaction of thrombin with antithrombin is enhanced considerably through the action of heparin, which binds the two components together via a so‐called ‘template mechanism’. The mutants desETW and desPPW both interact poorly with antithrombin (Le Bonniec et al., 1995), although this can be compensated for by the addition of heparin. Recent experiments concerning the interaction of antithrombin with the thrombin mutant Trp60D→Ala (Rezaie, 1996) suggest that Trp60D destabilizes the formation of the initial reversible complex, yet favours the transition to the final stable complex. This suggests that a rearrangement of the 60‐loop could be fundamental to the thrombin–serpin interaction.
Here, we describe the 2.3 Å crystal structure of human thrombin variant E192Q in complex with BPTI. The structure reveals a major reorganization of the surface loops surrounding the active site cleft, suggesting that binding of BPTI to thrombin results in destabilization of the molecule.
The complex of BPTI with thrombin E192Q exhibits the same overall features as the homologous trypsin complex (Huber et al., 1974) (Figure 1). BPTI binds to the active site of thrombin with its binding loop found in a canonical interaction with the serine protease. The BPTI molecule exhibits no significant differences from its trypsin bound form, with an r.m.s. difference of 0.48 Å for the 58 Cα positions. The orientation of the inhibitor with respect to the protease is also very similar with almost no difference in the binding loop region. The slight displacement of 1.5 Å for BPTI residues farthest away from the enzyme represents a negligible tilting of the inhibitor from its position in the trypsin–BPTI complex.
The binding loop of BPTI is in direct contact with thrombin, and binds to its active site like a substrate, with the characteristic main‐chain conformation and intermolecular hydrogen bond interactions of canonical binding proteinase inhibitors (Bode and Huber, 1992). The four residues preceding the scissile peptide bond exhibit an antiparallel arrangement with the enzyme segment Ser214–Gly216, forming the typical hydrogen bonds Ser214 O to N Lys15I (P1) and Gly216 N to O Pro13I (P3) (BPTI residues are distinguished by the suffix I). As P3 is a proline in BPTI, the second typical hydrogen bond displayed by Gly216 to the nitrogen atom of the P3 residue is not possible. The carbonyl group of Lys15I (P1) is situated in the oxyanion hole and displays the typical bifurcated hydrogen bonds to Gly193 N and Ser195 N. The following three residues run antiparallel to thrombin's segment Leu40–Leu41 and form the usual hydrogen bond Leu41 O to N Arg17I(P2′).
The side chain of Lys15I extends into the specificity pocket with atom Nζ more than 4 Å away from the carboxylate group of Asp189. Instead, it forms two hydrogen bonds to the carbonyl groups of Ala190 and Gly219. No density corresponding to a buried water molecule is observed, reflecting the increased hydrophobic nature of thrombin's S1 pocket. The scissile peptide bond is intact, and there is no electron density between Ser195 Oγ and the carbonyl C of Lys15I that might indicate an attack of the reactive oxygen onto this group.
Secondary contacts outside of the binding loop are found for all four surface loops of thrombin framing the active site, i.e. the 60‐, 39‐, 148‐ and 99‐loops. Altogether, 12 hydrogen bonds and three salt bridges are formed between thrombin E192Q and BPTI (Table I), with the burial of an accessible surface of 1200 Å2 on complex formation. This is significantly more than the seven hydrogen bonds and the single salt‐bridge, burying a total of 800 Å2, observed in the trypsin–BPTI complex.
The structure of BPTI and its interaction with the enzyme are therefore unaffected by the different serine protease. BPTI is found in the same position as that observed in its trypsin complex on which the modelling studies for a hypothetical thrombin–BPTI complex were based. The only way out of this dilemma of steric collision is a substantial movement of thrombin's 60‐loop.
The thrombin E192Q structure
Although it is the mutation of Glu192 to Gln that provides the driving force behind complex formation with BPTI, the conformation of Gln192 is not particularly unusual. As observed in the BPTI–trypsin complex, the carboxamide nitrogen atom of Gln192 makes hydrogen bonds with the carbonyl groups of Cys14I (P2) and Gly12I (Table I), while its oxygen hydrogen bonds to a buried water molecule coordinated by the backbone carbonyl and amide groups of Thr147 and the side chain Nδ2 of Asn143. The Gln192 carboxamide group is effectively buried in the interface between thrombin E192Q and BPTI. The polar groups that line this pocket are all main‐chain atoms: the carbonyl groups of Thr147, Gly12I, Pro13I, Cys14I and Ala16I and the amide nitrogen of Lys15I. The electronegative character of this cavity is therefore ill‐suited to the native glutamate side chain.
The most striking difference of the thrombin structure observed here compared with all previous thrombin structures is the large displacement of the 60‐loop. The Cα positions of the 12 residues Leu60 to Asn62 deviate by >1 Å from the PPACK structure; at the apex of the loop, the Cα atom of Trp60D is displaced by 8 Å (see Figure 3). Displacement is not a simple rigid body movement; the main‐chain conformation is altered significantly, indicating that major disruptions of the loop must have occurred for binding to BPTI.
The side chain of Trp60D is sandwiched between Pro60B and Arg42I, the latter being in a parallel stacking arrangement. Trp60D displays the second‐highest number of van der Waals contacts to BPTI, so that its interaction probably plays a major role in the stabilization of the loop. Asp60E faces Asp50I, with a mediating Lys46I between them.
The major remodelling of the 60‐loop throws the hydrophobic side chain of Phe60H into the solvent area; this unfavourable environment presumably accounts for the lack of density observed for Phe60H (Figure 2). The ensuing cavity left in thrombin is filled partially by the reorientation of Tyr60A; the rest is occupied by the side chains of Arg35 and Leu40 (Figure 3B). This latter interaction is accomplished via a substantial rearrangement of the 39‐loop, and is stabilized further by the close approach of the acidic side chains of Glu39 and Asp63 to the guanidino group of Arg35. The Cα positions of residues Arg35 to Leu41 are dislocated by >1 Å with respect to the PPACK (d‐Phe‐Pro‐Arg‐chloromethylketone) structure (Bode et al., 1989). Glu39 makes a further ionic interaction with Arg20I, and van der Waals contacts to Ile18I and Ile19I.
The 148‐loop, which has been noted for its structural variability (Priestle et al., 1993), packs against the body of the BPTI molecule in a position not far removed from that in PPACK‐thrombin (Figure 2) (Bode et al., 1989). The side chain of Trp148 makes a large number of van der Waals contacts with BPTI, thus stabilizing both the conformation of the loop, and also probably the thrombin–BPTI interaction.
Finally, binding of BPTI also causes a dislocation of the 99‐loop. Residues Trp96 to Leu99 deviate significantly from the PPACK structure, with the largest displacement observed for Asn98. This movement is probably the result of steric hindrance between Leu99 and the disulfide bridge 14I–38I, itself a consequence of the one residue insertion Glu97A found in thrombin.
Previous crystal structures of thrombin indicate that the 60‐insertion loop is a rigid moiety that partially occludes the active site. It has therefore been assumed tacitly that this structural feature plays a decisive role in restricting thrombin‘s specificity. The results presented here show that a seemingly minor change in thrombin—the exchange of a single oxygen atom for nitrogen—is enough to allow complex rearrangements of thrombin's substrate recognition apparatus upon ligand binding. The observed reorganization of the 60‐loop, the 148‐loop, the 37‐loop and the 99‐loop is in agreement with thermodynamic data on the thrombin–hirudin interaction (Ayala et al., 1995), which suggest that these loops may exhibit a degree of inherent flexibility in free thrombin.
Although it could be argued that the E192Q mutation alters the conformation of thrombin and its 60‐loop, this seems unlikely for several reasons. In the native structure, Glu192 extends into the solvent and is not involved in contacts to other residues. More importantly, the observed perturbation of the structure around the 60‐loop would disrupt the S2 binding site, altering dramatically the cleavage kinetics of typical thrombin substrates, including synthetic substrates with P2 proline residues and fibrinogen. On the contrary, thrombin E192Q maintains a strong preference for synthetic substrates with proline at the P2 position and retains enhanced activity towards fibrinogen (Le Bonniec and Esmon, 1991). Thus, the mutation in itself is unlikely to account for the major conformational changes in the 60‐loop observed with BPTI in complex with thrombin E192Q.
Conceptually, it is possible to break down the interaction of thrombin with BPTI into three consecutive (possibly simultaneous) parts: (i) approach of BPTI to thrombin and alignment of reactive and active sites; (ii) opening up of the active site through displacement of characteristic surface loops, in particular the 60‐loop; (iii) tight binding of BPTI in the active site cleft together with reorganization of loops to stabilize the interaction. Comparison of free and complexed thrombin and BPTI allows visualization of states (i) and (iii), whilst state (ii) is hypothetical. Clearly, such an opening of thrombin‘s active site cleft must exact an energetic cost. This energy may be estimated from the published Ki values for thrombin, thrombin variants and trypsin with BPTI (Green et al., 1957; Ascenzi et al., 1988; Le Bonniec et al., 1992, 1993; Guinto et al., 1994), which have been plotted in the form of free energy changes ΔG° (Figure 4). Comparing thrombin–BPTI with desPPW–BPTI (or E192Q–BPTI with desPPW/E192Q–BPTI) suggests that the intact 60‐loop impedes the binding of BPTI at a cost of ∼20 kJ/mol. Similarly, a comparison of desPPW/E192Q–BPTI with trypsin–BPTI suggests that opening of desPPW itself probably costs a further 20 kJ/mol. This latter value is probably an overestimate, as BPTI is specially suited to trypsin (in particular, residue Lys15I is not favourable for thrombin); the apparent isomerization of the Cys14I–Cys38I disulfide bond (Otting et al., 1993) could also ease complexation. Nevertheless, a total energy of 40 kJ/mol to expose fully thrombin's active site would appear a reasonable estimate.
The binding of BPTI to thrombin is in stark contrast to that of hirudin, where structural changes in thrombin are minimal (Rydel et al., 1991). Although hirudin undergoes a reorganization of its tail to bind, the energetic cost must be relatively small, as the binding of hirudin (73–75 kJ/mol) is the sum of binding N‐terminal (38–45 kJ/mol) and C‐terminal (33–36 kJ/mol) components (Dennis et al., 1990; Schmitz et al., 1991). The latter value is of particular interest, in that it suggests that the energy gained in binding at the fibrinogen recognition exosite is roughly comparable with that necessary to open thrombin's active site cleft. Moreover, the data presented here demonstrate a structural link between the expulsion of the 60‐loop and reorganization of the 39‐loop, and thus an effect on the fibrinogen recognition exosite. It is therefore conceivable that the reverse process could occur: that strong binding at the fibrinogen recognition exosite could facilitate partial expulsion of the 60‐loop. Thus, the estimated binding energy of 52 kJ/mol for the thrombomodulin–thrombin interaction (Hofsteenge et al., 1986; Liu et al., 1994) would be more than enough to compensate for the expulsion of the 60‐loop.
Evidence in favour of local conformational changes induced by thrombomodulin, chondroitin sulfate or heparin include changes in the fluorescence properties of probes located near the active site of thrombin and alterations in the activity towards chromogenic substrates (Ye et al., 1991; Liu et al., 1994). These changes could be mediated by interactions with either the fibrinogen recognition exosite or the heparin‐binding site. Such changes in the active site might partially explain why the thrombin–thrombomodulin complex prefers to activate the Ca2+‐stabilized conformation of protein C, while free thrombin has an extremely strong preference for the Ca2+‐free protein C conformation (Esmon et al., 1983). Effective protein C activation requires the presence of the thrombomodulin domains corresponding to EGF domains 4–6. The complex of thrombin with a thrombomodulin peptide from EGF 5 (Mathews et al., 1994) suggests that these thrombomodulin EGF domains extend out over the active site of thrombin.
However, it should be emphasized that binding to the fibrinogen recognition exosite does not lead to expulsion of the 60‐loop per se. The position of this loop has not been found to vary greatly between structures with the fibrinogen recognition exosite complexed or uncomplexed (Stubbs and Bode, 1993; Engh et al., 1996). The role of fibrinogen recognition exosite binding can be seen as a facilitator; it would not directly induce the expulsion, but may promote it. In this regard, it is interesting to note that exosite binding stabilizes a cleavage‐resistant conformation of the 148‐loop (Parry et al., 1993).
The inhibition mechanism of serine proteinases by serpins is still a matter of some debate (Engh et al., 1995). It is generally agreed, however, that one stage of the interaction involves canonical complex formation. Until now, it has been assumed that the thrombin–serpin interaction provides the best test for modelling inhibitory serpins due to thrombin‘s restrictive active site cleft. Although the Pittsburgh variant of α1‐antitrypsin reacts rapidly with thrombin (Owen et al., 1983), antithrombin–thrombin complex formation proceeds very slowly in the absence of heparin (Olson and Björk, 1992). The results presented here, together with those for thrombin variants Trp60D→Ala (Rezaie, 1996), desPPW (Le Bonniec et al., 1993) and desETW (Le Bonniec et al., 1992) suggest the following scenario: (i) approach of antithrombin to thrombin, (ii) expulsion of thrombin occluding loops, (iii) canonical complex formation, stabilization of thrombin loops and (iv) transition to stable, non‐reversible complex. Thus, steps (i) to (iii) resemble closely those proposed for BPTI. Furthermore, the present complex indicates stabilizing interactions for both Trp60D and Trp148 of the 60‐ and 148‐loops, respectively. The E192Q–BPTI complex may therefore serve as a guide to understanding the thrombin–antithrombin interaction. The addition of heparin as a template, linking the enzyme and inhibitor components, could provide sufficient binding energy to counterbalance the energetically unfavourable 60‐loop expulsion. The requirement for a reorganization of the 60‐loop, which is also responsible for thrombin's extracellular matrix binding and chemotactic activities, might explain why extracellular matrix bound thrombin is incapable of complex formation with antithrombin (Bar‐Shavit et al., 1989).
It is, therefore, necessary to revise our picture of thrombin as a rigid and restrictive molecule. The closed form represents the ground state structure; the large scale rearrangements observed here probably represent more the exception than the rule. Given sufficiently favourable interactions, thrombin‘s active site may be made accessible to otherwise unsuitable substrates and inhibitors. More experimental investigations are necessary to establish the precise nature of thrombin's ‘allostery’.
Materials and methods
Human thrombin E192Q was prepared as described previously (Le Bonniec and Esmon, 1991). BPTI was a generous gift from Dr Hans Dietrich Hörlein (Bayer AG, Wuppertal, Germany). Thrombin E192Q was co‐crystallized with a slight excess of a 1:1 molar ratio of BPTI to thrombin. Monoclinic crystals (space group P21, containing two complexes per asymmetric unit) were grown at 20°C from 20% PEG 5000 monomethyl ether, 0.1 M HEPES, pH 7.0–7.5, in hanging drops using the vapour diffusion technique. Diffraction data up to 2.3 Å were collected on a MAR imaging plate system and evaluated using the Mosflm package (Leslie, 1994) and programs from the CCP4 Suite (CCP4, 1994).
The structure was solved using Patterson search techniques. Rotational and translational searches for the orientation and position of the thrombin molecules were performed with the program AMoRe (Navaza, 1994) using data up to 3.5 Å and the bovine thrombin model as obtained previously in complex with rhodniin (van de Locht et al., 1995). The rotational search showed two solutions with correlation values of 9.3 and 8.8 σ. Translational search and rigid body fitting for these solutions resulted in a correlation value of 0.54 and an R‐value of 40.9%. An envelope was created using thrombin from its rhodniin complex and BPTI from its trypsin complex after optimal superposition of the trypsin and thrombin components, and the density was averaged using routines from the RAVE program package (Kleywegt and Jones, 1994). The BPTI model as taken from its trypsin complex could be fitted to the averaged electron density, instantaneously. Thrombin surface loops framing the active site had to be rebuilt. Several refinement cycles, consisting of model building using O (Jones et al., 1991) and simulated annealing minimization with X‐PLOR (Brünger, 1992) using the parameters of Engh and Huber (1991), reduced the Rvalue to 19.6%. One hundred and thirteen water molecules were added to the model. Data statistics are given in Table II. The coordinates of the BPTI–thrombin E192Q complex have been deposited with the Brookhaven Protein Data Bank (accession No. 1BTH).
BPTI was a generous gift from Dr Hans Dietrich Hörlein (Bayer AG, Wuppertal, Germany). This work was supported by the Sonderforschungsbereich 207 and EU contract BMH4‐CT96‐0937.
↵† This paper is dedicated to the memory of our friend and colleague Professor Stuart R.Stone, whose untimely death (December 16, 1996) is a tragic loss to us and to the scientific community.
- Copyright © 1997 European Molecular Biology Organization