Stabilization of the E* Form Turns Thrombin into an Anticoagulant*

Previous studies have shown that deletion of nine residues in the autolysis loop of thrombin produces a mutant with an anticoagulant propensity of potential clinical relevance, but the molecular origin of the effect has remained unresolved. The x-ray crystal structure of this mutant solved in the free form at 1.55 Å resolution reveals an inactive conformation that is practically identical (root mean square deviation of 0.154 Å) to the recently identified E* form. The side chain of Trp215 collapses into the active site by shifting >10 Å from its position in the active E form, and the oxyanion hole is disrupted by a flip of the Glu192–Gly193 peptide bond. This finding confirms the existence of the inactive form E* in essentially the same incarnation as first identified in the structure of the thrombin mutant D102N. In addition, it demonstrates that the anticoagulant profile often caused by a mutation of the thrombin scaffold finds its likely molecular origin in the stabilization of the inactive E* form that is selectively shifted to the active E form upon thrombomodulin and protein C binding.

Previous studies have shown that deletion of nine residues in the autolysis loop of thrombin produces a mutant with an anticoagulant propensity of potential clinical relevance, but the molecular origin of the effect has remained unresolved. The x-ray crystal structure of this mutant solved in the free form at 1.55 Å resolution reveals an inactive conformation that is practically identical (root mean square deviation of 0.154 Å ) to the recently identified E* form. The side chain of Trp 215 collapses into the active site by shifting >10 Å from its position in the active E form, and the oxyanion hole is disrupted by a flip of the Glu 192 -Gly 193 peptide bond. This finding confirms the existence of the inactive form E* in essentially the same incarnation as first identified in the structure of the thrombin mutant D102N. In addition, it demonstrates that the anticoagulant profile often caused by a mutation of the thrombin scaffold finds its likely molecular origin in the stabilization of the inactive E* form that is selectively shifted to the active E form upon thrombomodulin and protein C binding.
Serine proteases of the trypsin family are responsible for digestion, blood coagulation, fibrinolysis, development, fertilization, apoptosis, and immunity (1). Activation of the protease requires the transition from a zymogen form (2) and formation of an ion pair between the newly formed amino terminus of the catalytic chain and the side chain of the highly conserved residue Asp 194 (chymotrypsinogen numbering) next to the catalytic Ser 195 . This ensures substrate access to the active site and proper formation of the oxyanion hole contributed by the backbone N atoms of Ser 195 and Gly 193 (3). The zymogen 3 protease conversion is classically associated with the onset of catalytic activity (3,4) and provides a useful paradigm for understanding key features of protease function and regulation.
Recent kinetic (5) and structural (6,7) studies of thrombin, the key protease in the blood coagulation cascade (8), have drawn attention to a significant plasticity of the trypsin fold that impacts the function of the enzyme in a decisive manner. The active form of the protease, E, coexists with an inactive form, E*, that is distinct from the zymogen conformation (9). The E* form features a collapse of the 215-217 ␤-strand into the active site and a flip of the peptide bond between residues Glu 192 and Gly 193 that disrupts the oxyanion hole. Importantly, the ion pair between Ile 16 and Asp 194 remains intact, suggesting that E* is not equivalent to the zymogen form of the protease and that the E*-E equilibrium is established after the conversion from the zymogen form has taken place. Indeed, existing structures of the zymogen forms of trypsin (10), chymotrypsin (11), and chymase (12) feature a broken Ile 16 -Asp 194 ion pair but no collapse of the 215-217 ␤-strand. Stopped-flow experiments show that the E*-E conversion takes place on a time scale of Ͻ10 ms (5), as opposed to the much longer (100 -1000 ms) time scale required for the zymogen-protease conversion (13,14).
The E* form is not a peculiarity of thrombin. The collapse of the 215-217 ␤-strand into the active site is observed in the inactive form of ␣I-tryptase (15), the high temperature requirement-like protease (16), complement factor D (17), granzyme K (18), hepatocyte growth factor activator (19), prostate kallikrein (20), and prostasin (21). A disrupted oxyanion hole is observed in complement factor B (22) and the arterivirus protease Nsp4 (23). The most likely explanation for the widespread occurrence of inactive conformations of trypsin-like proteases is that the E*-E equilibrium is a basic property of the trypsin fold that fine tunes activity and specificity once the zymogen 3 protease conversion has taken place (9).
The new paradigm established by the E*-E equilibrium has obvious physiological relevance. In the case of complement factors, kallikreins, tryptase, and some coagulation factors must be kept to a minimum until binding of a trigger factor ensues. Stabilization of E* may afford a resting state of the protease waiting for action, as seen for other systems (24 -28). For example, factor B is mostly inactive until binding of complement factor C3 unleashes catalytic activity at the site where amplification of C3 activation is most needed prior to formation of the membrane attack complex (29). Indeed, the crystal structure of factor B reveals a conformation with the oxyanion hole disrupted by a flip of the 192-193 peptide bond (22), as observed in the E* form of thrombin (6,7).
The allosteric equilibrium as shown in Scheme 1, involves the rates for the E* 3 E transition, k 1 , and backward, k Ϫ1 , that define the equilibrium constant r ϭ k Ϫ1 /k 1 ϭ [E*]/[E] (5). The value of k cat /K m for an enzyme undergoing the E*-E equilibrium is as shown in Equation 1 (30), where s E is the value of s for the E form, and obviously s E* ϭ 0. Likewise, the binding of an inhibitor to the enzyme undergoing the E*-E equilibrium is shown in Equation 2, where K E is the value of the equilibrium association constant K for the E form, and K E* ϭ 0. As the value of r increases upon stabilization of E*, the values of s and K in Equations 1 and 2 decrease without limits. Hence, stabilization of E* has the potential to completely abrogate substrate hydrolysis (s 3 0) or inhibitor binding (K 3 0). However, binding of a suitable cofactor could restore activity by triggering the E* 3 E transition. This suggests a simple explanation for the anticoagulant profile observed in a number of thrombin mutants that have poor activity toward all physiological substrates but retain activity toward the anticoagulant protein C in the presence of the cofactor thrombomodulin (31)(32)(33)(34). Here we report evidence that stabilization of E* provides a molecular mechanism to turn thrombin into an anticoagulant.

MATERIALS AND METHODS
The human thrombin mutant ⌬146 -149e was constructed, expressed, and purified to homogeneity as reported elsewhere (32,35,36), using the QuikChange site-directed mutagenesis kit from Stratagene (La Jolla, CA) in an HPC4-modified pNUT expression vector containing the human prethrombin-1 gene. Values of s ϭ k cat /K m for the hydrolysis of the chromogenic substrates H-D-Phe-Gly-Arg-p-nitroanilide, H-D-Phe-Pro-Phe-p-nitroanilide, H-D-Phe-Pro-Lys-p-nitroanilide, and H-D-Phe-Pro-Arg-p-nitroanilide (FPR), 2 the release of fibrinopeptide A from fibrinogen, cleavage of the protease-activated receptors PAR1, PAR3, and PAR4, and activation of protein C in the absence or presence of 100 nM thrombomodulin and 5 mM CaCl 2 were determined as reported elsewhere (32,37,38) under physiological experimental conditions of 5 mM Tris, 0.1% PEG8000, 145 mM NaCl, pH 7.4, at 37°C.
Binding of the active site inhibitor argatroban (39) was studied directly by isothermal titration calorimetry under experimental conditions of 5 mM Tris, 0.1% PEG8000, 145 mM NaCl, pH 7.4, at 37°C, using an iTC200 calorimeter (MicroCal Inc. Northampton, MA) with the sample cell containing thrombin and the syringe injecting argatroban. The sample volume for iTC200 is 204.6 l and the total volume of injected ligand is 39.7 l. The thermal equilibration step at 37°C, was followed by an initial 60-s delay step and subsequently an initial 0.2-l injec-tion. Typically, 19 serial injections of 2 l and 1 last injection of 1.5 l of ligand were performed at an interval of 180 s. The stirring speed was maintained at 1000 rpm, and the reference power was kept constant at 5 cal/s. The heat associated with each injection of ligand was integrated and plotted against the molar ratio of ligand to macromolecule. Thermodynamic parameters were extracted from a curve fit to the data using the software (Origin 7.0) provided by MicroCal according to a onesite binding model. Experiments were performed in triplicate with excellent reproducibility (Ͻ10% variation in thermodynamic parameters).
Crystals of human thrombin ⌬146 -149e were obtained using the hanging drop vapor-diffusion method. A solution of ⌬146 -149e (8 mg/ml in 1 l) in 50 mM NaCl, 20 mM Tris, pH 7.5, was mixed with an equal volume reservoir solution containing 20% PEG20000 and 100 mM Tris, pH 8.5, at 25°C. Crystals were tetragonal, space group P4 3 , with unit cell parameters a ϭ b ϭ 58.2 Å, c ϭ 119.6 Å, and contained one molecule in the asymmetric unit. Crystals were flash-frozen in liquid nitrogen after soaking in artificial mother liquor containing 15% (v/v) glycerol. X-ray diffraction data were recorded on an ADSC Quantum 315 CCD at beamline 14-BM-C at BIOCARS (Argonne, IL). One pass of 150°with steps of 0.5°was collected and processed to 1.55 Å resolution. Integration and scaling of diffraction data were carried out with the HKL-2000 package (40). The structure was solved by molecular replacement using the CCP4 suite (41) and Protein Data Bank accession code 2GP9 (7) as a search model. Rounds of positional and isotropic temperature factor refinement in REFMAC (42) were alternated with model building in COOT (43). After most of the model was found, TLS tensors modeling rigid-body anisotropic temperature factors were calculated and applied to the model using REFMAC. This was alternated with more model building in COOT until the final model was produced. Ramachandran plots were calculated using PROCHECK (44). Statistics for data collection and refinement are summarized in Table 1. Coordinates of the structure of the human thrombin mutant ⌬146 -149e have been deposited to the Protein Data Bank (accession code 3GIC).

RESULTS
Once generated in the blood from its inactive precursor prothrombin, thrombin acts as a procoagulant when it converts fibrinogen into an insoluble fibrin clot (45) and acts as a prothrombotic when it cleaves protease-activated receptors (PARs) (46,47). However, upon interaction with the endothelial cell receptor thrombomodulin, thrombin loses both procoagulant and prothrombotic functions and increases its activity Ͼ1,000-fold toward the anticoagulant protein C (48). A thrombin mutant stabilized in the E* form would have little or no activity toward physiological substrates. If this mutant could be converted to the E form upon binding of thrombomodulin, then a selective anticoagulant response would be elicited upon activation of protein C in the vascular endothelium where thrombomodulin is present.
The mutant ⌬146 -149e carries a deletion of the nine residues 146 ETWTANVGK 149e in the autolysis loop and was originally constructed to assess the role of this highly flexi-ble domain in thrombin function (49). The sequence 149a ANVGK 149e is not present in trypsin and chymotrypsin and can be deleted without functional consequences (49). Also, swapping the entire 146 ETWTANVGK 149e sequence of thrombin with the 146 SSGT 149 sequence of trypsin (50) or deleting the 146 ETW 148 sequence (51) produces perturbations of function that are recapitulated by the single mutations E146A and R221aA (32,35). Residue Glu 146 makes an important ion pair interaction with Arg 221a in the adjacent 220-loop in the E form but not in the E* form (6,7,35). The entire autolysis loop likely participates in the long range communication seen in the E*-E equilibrium (6) between the 220-loop, the active site, and exosite I where thrombomodulin binds (8). Deletion of the entire sequence 146 ETWTANVGK 149e in the ⌬146 -149e mutant causes a significant loss of activity toward chromogenic and physiological substrates, but binding of thrombomodulin almost completely restores activity toward the anticoagulant protein C (Fig. 1). Importantly, thrombomodulin has only a modest effect on the hydrolysis of a chromogenic substrate (Fig. 1), as already documented for other anticoagulant thrombin mutants (52) and wild type (53). These properties suggest that the ⌬146 -149e mutation shifts the E*-E equilibrium of thrombin in favor of E*, and thrombomodulin in complex with protein C can switch the mutant back into the active conformation E.
Evidence that the thrombin mutant ⌬146 -149e is stabilized in the E* form in solution comes from inspection of the values of k cat /K m for chromogenic and natural substrates. The data in Fig. 1 reveal a remarkable similarity in the loss of activity toward fibrinogen, PAR1, PAR3, and protein C for the ⌬146 -149e mutant compared with wild type. Likewise, a comparable loss of activity is observed toward several chro-mogenic substrates bearing replacements at the P1 or P2 positions (54). On the average, the loss is about 200-fold. For a mutation that selectively shifts the E*-E equilibrium in favor of E*, without introducing additional effects on substrate or inhibitor recognition, the values of s and K in Equations 1-2 must decrease by the same amount. Specifically, the ratio shown in Equation 3 between the wild-type (WT) and mutant values of s and K must be the same for all substrates and inhibitors.
The data in Fig. 1 are consistent with the prediction from Equation 3. Perturbation of PAR4 recognition is significantly more pronounced compared with all other substrates, but this is consistent with the direct interactions that this substrate makes with residues of the autolysis loop (55). Further support for stabilization of E* in the ⌬146 -149e mutant comes from calorimetric measurements of the binding of the inhibitor argatroban (39) to the active site. The value of K (see Equation 2) drops 135-fold in the mutant compared with wild type (Fig. 2), as expected from Equation 3. Addition of thrombomodulin restores activity of the mutant toward protein C (Fig. 1), indicating that although the mutation stabilizes E*, the active form E is still present in solution and can be populated for protein C activation in the presence of cofac-  tor. Evidence that binding to exosite I, the major thrombin epitope for thrombomodulin recognition (56,57), can convert E* to E has been provided recently by the structure of the thrombin mutant D102N bound to a fragment of the platelet receptor PAR1 (6). Therefore, the thrombin mutant ⌬146 -149e likely functions as an allosteric switch stabilized into the inactive form E* until the combined binding of thrombomodulin and protein C shifts E* to E and restores activity.
To gain direct information on the conformational properties of the ⌬146 -149e mutant, its crystal structure was solved in the free form at 1.55 Å resolution. Consistent with the functional data, the mutant assumes a collapsed conformation that is practically identical (r.m.s.d. 0.154 Å) to the E* form identified recently from the structure of the thrombin mutant D102N (6, 7). The ⌬146 -149e mutant folds in a selfinhibited conformation (Fig. 3) due to a collapse of the 215-217 ␤-strand into the active site that moves the indole ring of Trp 215 Ͼ10 Å to engage the catalytic His 57 on the opposite side of the active site cleft (Fig. 4). The drastic rearrangement of the 215-217 ␤-strand propagates the perturbation up to the peptide bond between Glu 192 and Gly 193 that is flipped relative to the conformation of the E form (7) and destroys the architecture of the oxyanion hole (Fig. 4). These significant structural changes occur away from the site of mutation in the autolysis loop, where the shortened sequence 144 LKGQ 151 shows Lys 145 connected directly to Gly 150 (Fig. 4). Hence, the thrombin mutants D102N and ⌬146 -149e crystallize in the same E* form, notwithstanding differences in sequence and crystallization conditions. The inactive form E* is therefore a genuine conformation of thrombin accessible to the enzyme in addition to its active form E (9,35). The E* form is stabilized by mutations that compromise activity of the enzyme and is selectively converted to the active E form under suitable conditions.

DISCUSSION
The new paradigm emerged from analysis of recent crystal structures of trypsin-like proteases (6,7,(15)(16)(17)(18)(19)(20)(21)(22)(23) supports the existence of the E*-E equilibrium as a critical feature of the trypsin fold (9). This allosteric equilibrium explains several important aspects of protease biology. For proteases that are poorly active until interaction with a cofactor, as observed for some clotting and complement factors (29), the onset of catalytic activity can be attributed to the E* 3 E conversion. The E* form acts in this case as a resting state for the enzyme and a spring-loaded mechanism that can be turned on when required  by the biological context. The E*-E equilibrium also provides context to interpret the effect of mutations associated with loss of biological activity in highly active proteases. In some cases, as documented by thrombin (8,58), the molecular origin of the effect is unclear because the mutation does not affect residues in direct contact with substrate. Stabilization of E* through molecular conduits not necessarily involved in substrate recognition may offer a plausible explanation.
The allosteric E*-E equilibrium has far reaching implications for protein engineering. Stabilization of E* by selected mutations, coupled with a transition to E triggered by suitable cofactors, may result in expression of protease activity on demand in a biological context. Thrombin exists predominantly in the E form, which functions as an ensemble partitioned between Na ϩ -free and Na ϩ -bound conformations (5,8) because of the extremely fast rates of binding and dissociation of Na ϩ with the enzyme (59). However, a thrombin mutant stabilized in the E* form that converts to E upon interaction with thrombomodulin and protein C would be an anticoagulant of potential clinical relevance. Such mutant would show little or no activity toward fibrinogen and PAR1 but would retain activity toward protein C. A number of thrombin mutants have been reported with an altered specificity that favors protein C activation over fibrinogen cleavage (31-34, 49, 60, 61). Among these mutants, E217K and W215A/E217A are effective as anticoagulants and antithrombotics in non-human primates (33,34,62,63) and have been crystallized at 2.5-2.8 Å resolution (52,64). The structures show a partial collapse of the 215-217 ␤-strand and disruption of the oxyanion hole that resembles the conformation of E* (7). It is possible that these mutants are stabilized in an E*-like form, but it is equally possible that the perturbed structures are the result of the mutations introduced at residues on the critical 215-217 ␤-strand. Substantial crystallographic packing interactions especially evident in the E217K mutant also bias the conformation, unlike the structures of the mutants D102N (7) and ⌬146 -149e reported here. Additional struc-tural work is necessary to validate the collapsed conformations of E217K and W215A/E217A.
The mutant ⌬146 -149e was previously identified for its significant anticoagulant profile (49) and carries a deletion of nine residues in the highly disordered autolysis loop, whose role in the function of the enzyme remains elusive. Importantly, residues of the autolysis loop are separate from the 215-217 ␤-strand or the oxyanion hole that undergoes substantial rearrangement in the E*-E transition (6,7). Yet the mutant ⌬146 -149e crystallizes in a collapsed conformation that is practically identical (r.m.s.d. 0.154 Å) to the E* form identified recently (6,7). This result is notable for two reasons. First, it confirms the existence of the inactive form E* in essentially the same incarnation as first identified in the structure of the thrombin mutant D102N under different crystallization conditions. Second, it provides proof of principle that the anticoagulant profile often caused by a mutation of the thrombin scaffold finds its likely molecular origin into stabilization of the inactive E* form. The mutant ⌬146 -149e was previously reported to feature reduced Na ϩ affinity (49), which supports the conclusion that its active E form is essentially Na ϩ -free. This confirms the tenet that abrogation of the procoagulant effect of Na ϩ (32, 65) is necessary to switch thrombin into an anticoagulant.
There is no contradiction between the modest effect of thrombomodulin on chromogenic substrate hydrolysis (Fig. 1) and structural evidence of the E* to E transition upon exosite I binding (6). Crystallographic evidence that the D102N mutant assumes the E* form when free and the E form when bound to exosite I does not imply an all-or-none distribution of E* and E in solution. In fact, a significant fraction of D102N does exist in the E form in solution (7), but this conformation is not favored under the crystallization conditions so far explored (7). Thermodynamic principles (66,67) establish that an allosteric effector can only shift the E*-E equilibrium in favor of E by an amount equal to the ratio of affinities of the two forms. Because exosite I, unlike the active site, is similarly accessible in the E* and E form (6 -8, 35), binding of thrombomodulin to the two forms cannot result in extreme perturbations of the E*-E equilibrium. Hence, when the equilibrium is shifted drastically in favor of E*, as seen for the ⌬146 -149e mutant but not the D102N mutant, binding of thrombomodulin is insufficient to populate significantly the active form E. On the other hand, the combined action of protein C and thrombomodulin may have a more profound effect on the E*-E equilibrium by accessing additional regions of the thrombin surface beyond exosite I (57), thereby ensuring almost complete restoration of activity.
Elucidation of the molecular mechanism underscoring the anticoagulant profile of thrombin mutants like ⌬146 -149e provides new impetus to the effort of rationally engineering thrombin mutants with exclusive activity toward protein C for clinical applications. The molecular underpinnings of the E*-E equilibrium have been revealed by structural biology (6), along with precise targets for mutagenesis aimed at stabilizing the E* form. An exclusive protein C activator would have several advantages compared with the direct administration of activated protein C currently marketed for the treatment of sepsis (68). Activated protein C acts as an anticoagulant when it inactivates clotting factor Va with the assistance of the cofactor  (6,7,9), causing disruption of the oxyanion hole contributed by the N atoms of Gly 193 and Ser 195 . The 2F o Ϫ F c electron density map (green mesh) is contoured at 2.0. Right, deletion of residues 146 ETWTANVGK 149e in the autolysis loop of the ⌬146 -149e mutant results in a new peptide bond connection between Lys 145 and Gly 150 (CPK, yellow). The autolysis loop is rarely seen in its entirety in thrombin structures, and considerable disorder remains in the mutant ⌬146 -149e where the sequence 144 LKGQ 151 must be contoured at 0.5 in the 2F o Ϫ F c electron density map (green mesh).
protein S and as a cytoprotective agent when it cleaves PAR1 on the surface of endothelial cells with the assistance of endothelial protein C receptor (69). Importantly, activated protein C generated in situ with the anticoagulant thrombin mutant W215A/E217A offers cytoprotective advantages over activated protein C administered to the circulation (70). Furthermore, the thrombin mutant W215A/E217A acts as a potent antithrombotic by blocking the interaction of von Willebrand factor with the platelet receptor GpIb (71), a property of which activated protein C is naturally devoid and that challenges the efficacy of low molecular weight heparins (63). These intriguing properties of the mutant W215A/ E217A, its well established potency as an anticoagulant in non-human primates (62,63) and the current elucidation of the role of E* in switching thrombin into an anticoagulant, will facilitate the rational engineering of a thrombin mutant with exclusive activity toward protein C.