Canonical Inhibitor-like Interactions Explain Reactivity of α1-Proteinase Inhibitor Pittsburgh and Antithrombin with Proteinases*

The serpin antithrombin is a slow thrombin inhibitor that requires heparin to enhance its reaction rate. In contrast, α1-proteinase inhibitor (α1PI) Pittsburgh (P1 Met → Arg natural variant) inhibits thrombin 17 times faster than pentasaccharide heparin-activated antithrombin. We present here x-ray structures of free and S195A trypsin-bound α1PI Pittsburgh, which show that the reactive center loop (RCL) possesses a canonical conformation in the free serpin that does not change upon binding to S195A trypsin and that contacts the proteinase only between P2 and P2′. By inference from the structure of heparin cofactor II bound to S195A thrombin, this RCL conformation is also appropriate for binding to thrombin. Reaction rates of trypsin and thrombin with α1PI Pittsburgh and antithrombin and their P2 variants show that the low antithrombin-thrombin reaction rate results from the antithrombin RCL sequence at P2 and implies that, in solution, the antithrombin RCL must be in a similar canonical conformation to that found here for α1PI Pittsburgh, even in the nonheparin-activated state. This suggests a general, limited, canonical-like interaction between serpins and proteinases in their Michaelis complexes.

Although antithrombin is the principal inhibitor of the blood coagulation proteinases thrombin and factor Xa, its rate of reaction with each of these proteinases is slow in the absence of heparin (1,2). In marked contrast, a variant of the serpin ␣ 1 -proteinase inhibitor that contains a P1 mutation of methionine to arginine, named ␣ 1 PI 1 Pittsburgh, was found to be a very effective inhibitor of thrombin, with a second order rate constant for thrombin inhibition 17-fold higher than that of antithrombin that had been conformationally activated by pen-tasaccharide heparin (3,4). In the individual who carried the Pittsburgh mutation, the potent thrombin inhibitory capability of this variant serpin resulted in hemorrhage and ultimately death.
In antithrombin, heparin is a multifactorial enhancer that can not only act as a bridging cofactor to bring serpin and proteinase together but can also induce a conformational change in antithrombin that results in expulsion of the two distal residues of the RCL from ␤-sheet A (5,6). It has been proposed that this loop expulsion permits alteration of the RCL conformation in the vicinity of the scissile bond to a more optimal one for interaction with proteinase and so contributes significantly to the heparin-induced acceleration of proteinase inhibition. In keeping with this proposal, the RCL conformation of antithrombin in published x-ray structures has the P1 arginine residue turned inwards toward the body of the protein, where it forms a salt bridge to Glu 255 (7,8). However, these x-ray structures are all of antithrombin dimers that contain one molecule of functionally active antithrombin and one in an inactive, latent conformation. Importantly, the RCL of the functional molecule forms the interface with the latent molecule, as an extra strand of ␤-sheet C of that molecule. This raises the question of whether the conformation of the active molecule has been artificially altered by heterodimer formation and thus of whether it represents the conformation of antithrombin in solution.
In an attempt to understand the structural basis for ␣ 1 PI Pittsburgh being able to inhibit thrombin at a high rate, and thus also to provide insight into the low basal reactivity of antithrombin, we have determined crystal structures of free and S195A trypsin-bound forms of the Pittsburgh variant of the serpin ␣ 1 -proteinase inhibitor and correlated these with rate constants for inhibition of thrombin and trypsin by ␣ 1 PI Pittsburgh, by antithrombin, and by P2 variants of each of these serpins.

MATERIALS AND METHODS
Protein Expression and Purification-␣ 1 PI Pittsburgh, containing seven stabilizing mutations that hinder polymerization (9) and a C232S mutation that prevents intermolecular disulfide bond formation, was expressed and refolded as previously described (10). The P2 Pro 3 Gly variant was made by site-directed mutagenesis and expressed and purified as for ␣ 1 PI Pittsburgh. S195A bovine trypsinogen was expressed in Escherichia coli as inclusion bodies and refolded, purified, and activated as previously described (11). The noncovalent complex between S195A trypsin and ␣ 1 PI Pittsburgh was formed at pH 7.5 in 25 mM Tris-HCl, 50 mM NaCl, and the complex was purified on Superdex 200 (Amersham Biosciences), dialyzed, and concentrated to 15 mg ml Ϫ1 in 10 mM MOPS, pH 6.0, containing 25 mM NaCl.
Crystallization, Data Collection, Structure Determination, and Refinement-␣ 1 PI Pittsburgh crystals were obtained at 20°C using the hanging drop method, by mixing equal volumes of protein (25 mg/ml) and well solutions (0.2 M potassium citrate, 20% polyethylene glycol * This work was supported by National Institutes of Health (NIH) Grants HL49234 and HL64013 (to P. G. W. G.). Use of the Advanced Photon Source is supported by the United States Department of Energy. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The  1 The abbreviations used are: ␣ 1 PI, ␣ 1 -proteinase inhibitor; BPTI, basic pancreatic trypsin inhibitor; RCL, reactive center loop; 5-mer, pentasaccharide heparin that binds to antithrombin with high affinity; MOPS, 4-morpholinepropanesulfonic acid. 3350, pH 6.30). An irregular rod-shaped crystal (0.9 ϫ 0.25 ϫ 0.1 mm) grew within 1 month. Diffraction data were collected at room temperature from a single crystal on a Rigaku RAXIS II detector, with a Rigaku RU-H2R rotating anode x-ray generator. Data were indexed with DENZO and scaled and reduced with SCALEPACK. Initial phases were determined by molecular replacement using the structure of wild type ␣ 1 PI (1QLP). Molecular replacement calculations were performed with CNS. Electron density maps were interpreted using Quanta. The model was progressively refined using simulated annealing at constant temperature, followed by energy minimization and manual inspection and rebuilding. 22 N-terminal amino acids and the histidine tag were not visible in the final electron density maps. Crystallographic data and statistics are shown in Table I. S195A trypsin-␣ 1 PI Pittsburgh complex crystals (0.15 ϫ 0.20 ϫ 0.20 mm) were obtained within 10 days at 18°C from hanging drops in 0.2 M potassium citrate buffer, pH 8.3-8.5, containing 17.5% (w/v) polyethylene glycol 3350 and 2% glycerol. Diffraction data were collected from two single, flash frozen, crystals, cryoprotected in oil, using synchrotron radiation at the SERCAT beamline (Advanced Photon Source, Argonne National Laboratory). Data were indexed and processed as described above. The structure of the complex was solved by molecular replacement using CNS programs with the structures of human ␣ 1 PI (1QLP) and bovine trypsin (1EJM) as search models. The final refined model of the complex was obtained as described for ␣ 1 PI Pittsburgh. Statistics are given in Table I.
Kinetic Measurements-Inhibition of trypsin by ␣ 1 PI Pittsburgh was followed at 25°C by progress curve analysis using a continuous assay. 1 nM trypsin was mixed with 10 -50 nM ␣ 1 PI Pittsburgh in 10 mM sodium phosphate, pH 7.40, 2 mM EDTA, 0.1 M NaCl, 0.1% polyethylene glycol 8000, containing 250 M S-2222 chromogenic substrate. Product release was followed by change in absorbance at 405 nm. Curves obtained at different ␣ 1 PI concentrations were fitted by a single exponential function with Sigma-Plot software, and values for k obs were plotted against [I] 0 /(1 ϩ [S] 0 /K m ), where [I] 0 and [S] 0 are the initial concentrations of ␣ 1 PI and S-2222, respectively. The second-order rate constant was obtained from the slope.
The trypsin-P2 Pro 3 Gly ␣ 1 PI Pittsburgh complex had very low stability. The rate of complex formation was therefore determined from SDS-PAGE analysis of the reaction products obtained under conditions where dissociation of the covalent complex was negligible (12). The ␣ 1 PI Pittsburgh P2 Gly variant (0.2 g) was mixed with trypsin (0.05 g) at room temperature (22°C) in 10 l of assay buffer. Samples were removed at intervals from 3 to 80 s, the reaction was quenched by mixing immediately with 10 l of preheated reduced SDS-loading buffer and heating for 2 min, and reaction products were immediately analyzed by 10 -20% gradient Tris/glycine SDS-PAGE. Band intensities were determined by densitometry and converted to relative concentrations. The second-order rate constant was determined from the slope of a plot of time against log([⌬I]/[⌬E]), where I represents serpin and E is proteinase. The rate of thrombin inhibition by the P2G variant was determined by discontinuous assay. 25 nM thrombin was mixed with 250 nM P2G ␣ 1 PI Pittsburgh, and the residual proteinase activity was determined at intervals by diluting the reaction mixture into assay buffer containing 75 M S-2238 chromogenic substrate. Residual proteinase activity was plotted against time and analyzed by nonlinear least-squares fitting in Sigma-Plot software using the following equation: The second order rate constant, k, was determined by dividing k obs by the initial ␣ 1 PI concentration.
Stoichiometries of Inhibition-The stoichiometry of inhibition of ␣ 1 PI Pittsburgh P2G variant for thrombin was determined by plotting residual proteinase activity against the serpin/proteinase ratio, after incubating the serpin (20 -400 nM) with 200 nM thrombin in assay buffer at 25°C and allowing the reactions to go to completion. Residual thrombin activity was determined by diluting the reaction mixture 100-fold with assay buffer containing 100 M S-2238 and measuring the rate of substrate hydrolysis at 405 nm. The stoichiometry of inhibition of ␣ 1 PI Pittsburgh for trypsin was determined analogously, except that residual activity of trypsin was measured by diluting the reaction mixture 500-fold into assay buffer containing 250 M S-2222.

RESULTS
Structure of the ␣ 1 -PI Pittsburgh-S195A Trypsin Noncovalent Complex-Although the serpin-proteinase reaction cannot readily be halted at the stage of the initial noncovalent Michaelis complex, this complex is thought to be closely approximated by the noncovalent complex formed between the dehydroalanine or S195A form of the proteinase and the serpin. We have determined the structure of such a complex formed between S195A trypsin and the serpin ␣ 1 PI Pittsburgh (Fig. 1a).
The ␣ 1 PI Pittsburgh RCL in the complex had a canonical conformation (Fig. 1b) similar to that found previously for noncomplexed wild-type serpin (13,14). As expected, the active site of the proteinase interacted with the serpin RCL, centered on the P1-P1Ј scissile bond. No contacts were present between the serpin body and the proteinase, unlike the noncovalent complex of another serpin, heparin cofactor II, with S195A thrombin. Surprisingly, considering the high stability of the ␣ 1 PI Pittsburgh-S195A trypsin complex (K d ϳ5 nM) (15), contacts between the serpin RCL and the proteinase are quite limited (Fig. 1c). Only the P1 arginine side chain has extensive interaction with the proteinase, being buried in the S1 specificity pocket. The remaining contacts involve hydrogen bonds between the backbone of serpin residues P2 to P2Ј with backbone atoms of the proteinase, in a manner almost identical to that of classical canonical serine proteinase inhibitors (16). This differs from an earlier structure of S195A rat trypsin with Manduca sexta serpin 1B A353K variant, which showed some additional contacts between more distal parts of the RCL and surface loops of trypsin that were suggested to be important for binding (17). We do not, however, find such a difference for our complex, in which the interaction of the serpin with S195A trypsin is equivalent to that of BPTI, and consequently the serpin has an RCL con-formation superimposable on that of the interaction loop of BPTI.
Structure of Free ␣ 1 -PI Pittsburgh-We also determined the x-ray crystal structure of native ␣ 1 PI Pittsburgh and found minimal differences between the structure of the free serpin and the serpin in noncovalent complex with S195A trypsin. The serpin body in the two species overlaid with root mean square deviation of only 0.6 Å for 240 non-RCL C-␣ atoms. This lack of conformational change for both the serpin body and the trypsin moiety upon forming complex is what had been found previously from NMR measurements for the same complex (10,11). Even within the RCL, the only significant differences were in the region where the loop joins the serpin body, distally from the scissile bond. Within P5 to P5Ј (Glu-Ala-Ile-Pro-Arg-Ser-Ile-Pro-Pro-Glu), however, which encompasses the whole of the region that interacts with S195A trypsin, there were almost identical backbone conformations in free and complexed states (Fig. 2). This RCL conformation was also very similar to that of FIG. 1. Structure of the ␣ 1 PI Pittsburgh-S195A complex. a, stereo ribbon representation of the noncovalent complex of S195A trypsin with ␣ 1 PI Pittsburgh. Trypsin is in gold, and ␣ 1 PI Pittsburgh is in brown, with the serpin RCL shown in cyan and ␤-sheet A in red. b, final 2F o Ϫ F c electron density map of RCL of ␣ 1 PI Pittsburgh from P4 to P4Ј. c, stereo representation of the contact interface between the ␣ 1 PI Pittsburgh RCL (cyan) and S195A trypsin, rotated 90°c lockwise around the vertical axis relative to the orientation shown in a. Hydrogen bonds are shown as dashed black lines. Residues in trypsin are numbered according to the protein sequence. Residues in the serpin RCL are given P and PЈ designations.
wild-type ␣ 1 PI, which has previously been pointed out to have a canonical conformation (13,14).
Comparison of RCL Conformations-Whether this same conformation of the RCL of ␣ 1 PI Pittsburgh could also fit into the active site of thrombin without need for structural change was evaluated from a comparison of the present structure with the recent x-ray structure of the equivalent noncovalent complex of S195A thrombin with the thrombin-specific serpin heparin cofactor II (18). The corresponding P5-P5Ј sequences of heparin cofactor II and (for later comparison) of antithrombin are Gly-Phe-Met-Pro-Leu-Ser-Thr-Gln-Val-Arg and Val-Ile-Ala-Gly-Arg-Ser-Leu-Asn-Pro-Asn, respectively. Although the P1 residue of heparin cofactor II is a leucine residue rather than arginine, it is nevertheless a good serpin for comparison with ␣ 1 PI Pittsburgh, since, in addition to both serpins being thrombin inhibitors, both have P2 proline residues and P1Ј serine residues, and the P1 leucine of heparin cofactor II occupies the S1 pocket of S195A thrombin in the same way that arginine in ␣ 1 PI Pittsburgh does in our noncovalent complex with S195A trypsin. Superpositioning of the RCLs of ␣ 1 PI Pittsburgh and heparin cofactor II, both from their noncovalent complexes with S195A proteinase, shows that the backbone conformations are almost identical in the region flanking the scissile bond (Fig. 3). Even the P1 side chain, which is very different in the two serpins, projects in the same direction, so as to fit in the S1 pocket in both cases.
Modeling of ␣ 1 PI Pittsburgh RCL into Thrombin Active Site-Although trypsin and thrombin have the same fold, they differ significantly at the entrance to the active site. Whereas trypsin has relatively unhindered access, thrombin has two insertion loops, one on either side of the active site, that limit access and are thought to contribute to the much greater protein substrate selectivity of thrombin. Prior to the x-ray structure determination of the noncovalent complex of E192Q thrombin with BPTI, it was thought that thrombin underwent no conformational change in binding to inhibitors. Indeed, all of the structures of thrombin with low molecular weight inhibitors had shown the same conformation for the insertion loops as well as for the core catalytic domain. The structure of the E192Q thrombin-BPTI complex, however, showed a major displacement of the "60 loop" and "150 loop," which permitted access of BPTI to the active site (19). This distortion accounts for the lower binding energy of the BPTI-thrombin complex compared with the BPTI-trypsin complex. In the heparin cofactor II-S195A thrombin noncovalent complex, these insertion loops are also displaced relative to uncomplexed thrombin, although to a slightly different degree than in the BPTI complex (Fig. 4a). It is therefore expected that a similar displacement of these insertion loops would allow ␣ 1 PI Pittsburgh to interact with thrombin in a tightly binding canonical-like manner (Fig. 4b), without need for changing the ␣ 1 PI Pittsburgh RCL conformation.

Modeling of Known Antithrombin RCL into Thrombin-The
RCL conformation of antithrombin found in all x-ray structures is dramatically different from that of ␣ 1 PI Pittsburgh and from heparin cofactor II (Fig. 5a), implying that there would be very poor complementarity between thrombin and an antithrombin RCL that had such a conformation in solution (Fig. 5b). With such a solution conformation, the RCL of antithrombin would require a major conformational change to become like that of ␣ 1 PI Pittsburgh and hence to be able to bind satisfactorily with thrombin.
Rates of Reaction of ␣ 1 PI Pittsburgh and Antithrombin with Trypsin and Thrombin-A recent extensive study of the rates of cleavage of peptide substrates by trypsin, thrombin, and factor Xa has shown that the rates of cleavage of unconstrained peptides by trypsin show very little dependence on residues flanking P1, whereas for cleavage of thrombin, there is very much higher specificity, particularly at the P2 position (20). Thus, change of the P2 residue from Pro to Gly results in almost no change in rate for cleavage by trypsin but a 20-fold reduction in rate for cleavage by thrombin. To determine whether the same pattern of specificity is shown in the more constrained RCL of ␣ 1 PI Pittsburgh, we created a P2 Pro 3 Gly variant of ␣ 1 PI Pittsburgh and compared the rates of reaction of the P2 Pro and P2 Gly forms of the serpin with both trypsin and thrombin.
For reaction with trypsin, both ␣ 1 PI variants gave similar rates of reaction, as was seen with peptide substrates (Table  II). In addition, the absolute values of the rate constants were less than 1 order of magnitude lower than for the peptides. For reaction with thrombin, a similar relationship was observed between rates of peptide substrate cleavage and serpin reaction for these two ␣ 1 PI variants. Thus, ␣ 1 PI Pittsburgh (P2 Pro) reacted about 30-fold faster than the P2 Gly variant, which was similar to the equivalent peptide substrates (Table II). In addition, the absolute values of the rate constants showed only a slightly larger decrease between free peptide and serpin RCL reactions than did the reactions with trypsin.
Compared with the ␣ 1 PI Pittsburgh analog, pentasaccharide-activated wild-type antithrombin (P2 Gly) reacts about as fast or faster with either trypsin or thrombin (Table II). The antithrombin RCL conformation in this heparin-bound form is therefore likely to require no more change in conformation to be able to bind to either of these proteinases than does ␣ 1 PI Pittsburgh. Similarly, for the P2 Pro variant of antithrombin bound to heparin pentasaccharide, the rate of reaction with thrombin is almost the same as for the reaction of ␣ 1 PI Pittsburgh, again implying equivalent minimal change in conformation upon binding. Whereas it is true that the rate of reaction of antithrombin with thrombin is lower in the absence of heparin pentasaccharide than in its presence, the reduction is only 1.6-fold for the wild-type and a little more (6-fold) for the P2 Pro variant (Table II).

FIG. 2. Comparisons of RCL from
␣ 1 PI Pittsburgh in free and complexed states. Superposition, in stereo, of the backbone conformations of the RCL P5-P5Ј residues of ␣ 1 PI Pittsburgh backbone in the free state (brown) and in noncovalent complex with S195A trypsin (cyan) is shown.

DISCUSSION
Inhibition of serine proteinases by serpins involves initial noncovalent binding of the proteinase to the RCL of the serpin, followed by attack of the active site serine on the P1-P1Ј peptide bond of the serpin. The rate-limiting step for inhibition is typically formation of the initial Michaelis-like complex that involves docking of the proteinase active site with the serpin RCL (21). Based on the x-ray structures of ␣ 1 PI Pittsburgh, alone and in noncovalent complex with S195A trypsin presented here, and on previous structures of wild-type ␣ 1 PI, we can conclude that the RCLs of these serpins are already in a preferred, canonical conformation to interact with trypsin or elastase, respectively, and form the initial, rate-determining Michaelis-like complex. This may account both for the high affinity of the noncovalent complexes and the very high rates of reaction with active trypsin and elastase, for the Pittsburgh and wild-type proteins, respectively.
Although we did not determine the structure of the noncovalent complex of ␣ 1 PI Pittsburgh with S195A thrombin, the availability of such a structure for another serpin, heparin cofactor II, allowed us to compare the RCL conformations of heparin cofactor II in complex with S195A thrombin with that of ␣ 1 PI Pittsburgh in complex with S195A trypsin. Based on a near identity of conformation of the two RCLs, it is expected that ␣ 1 PI Pittsburgh could also form an optimal, canonical noncovalent complex with thrombin without need for change in conformation of the RCL compared with that in the free state. In contrast, the conformation of the antithrombin RCL from published x-ray structures is noncanonical and consequently would not fit into the active site of thrombin without major distortion. This would presumably require significant energy and consequently result in a major rate reduction for formation of this complex compared with its formation by ␣ 1 PI Pittsburgh. That this is not the case, either for P2 Gly or P2 Pro forms of antithrombin compared with the equivalent ␣ 1 PI Pittsburgh species, strongly suggests that the solution conformation of the antithrombin RCL around the scissile bond is not like that in the crystal structure but is already like that of ␣ 1 PI Pittsburgh and hence requires little further conformational change to bind to thrombin.
Our kinetic results also suggest that, for ␣ 1 PI Pittsburgh, the 30-fold reduction in rate upon replacement of proline at P2 by glycine is not a result of conformational differences, since the free peptides show a similar rate reduction. Since the rate of reaction of ␣ 1 PI Pittsburgh with thrombin is only 6-fold lower than with trypsin, it is likely that the RCL conformations are the same as those found for the complexes of HCII with S195A thrombin and ␣ 1 PI Pittsburgh with S195A trypsin and that the FIG. 3. Comparisons of RCLs of different serpins in noncovalent complexes. Superposition, in stereo, of the conformations of P5-P5Ј residues of ␣ 1 PI Pittsburgh (cyan) and heparin cofactor II (magenta), both from the structures of noncovalent complexes with S195A proteinases is shown.
FIG. 4. Thrombin and its interactions with inhibitors. a, superposition, in stereo, of thrombin or S195A thrombin structures of complexes with PPACK (gold) (Protein Data Bank code 1HAI), with BPTI (green) (Protein Data Bank code 1BTH), and with heparin cofactor II (red) (Protein Data Bank code 1JMO), showing the identical conformations of all regions except the surface loops flanking the active site, particularly the 60 loop and 150 loop. b, stereo representation of the P5-P5Ј residues of ␣ 1 PI Pittsburgh (cyan) from the noncovalent complex with S195A trypsin docked into the thrombin structure found in the noncovalent complex of S195A thrombin with heparin cofactor II (HCII in magenta). modest rate reduction results from the need to bring about the same kind of displacement of the 60 and 150 loops of thrombin as was seen in the HCII-S195A thrombin noncovalent complex. Taken together, these rate correlations imply that the RCL of ␣ 1 PI Pittsburgh, with either P2 Pro or P2 Gly, has the same canonical conformation in complexes with S195A trypsin and S195A thrombin and that differences in rates of reaction result from changes in P2 side chain contacts in thrombin and from the need to displace the 60 and 150 loops of thrombin to accommodate the serpin RCL.
A similar conclusion can be drawn with regard to pentasaccharide-activated antithrombin, thus providing an answer to the question of why ␣ 1 PI Pittsburgh is a faster inhibitor of thrombin than pentasaccharide-activated antithrombin. Since the 15-fold increase in the rate of inhibition of thrombin by pentasaccharide-activated antithrombin upon changing the normal P2 Gly to Pro parallels the behavior in ␣ 1 PI Pittsburgh and free peptides, it most likely reflects the changes in contacts between the proline and glycine side chains and thrombin, rather than differences in conformation of the RCLs of the two serpins. ␣ 1 PI Pittsburgh is therefore a faster inhibitor of thrombin than pentasaccharide-activated antithrombin primarily because it has a proline at P2 rather than the glycine that is present in antithrombin.
For antithrombin in the absence of activating heparin, the small reductions in the rate of reaction with thrombin are unlikely to result from major conformational differences between pentasaccharide-activated and native antithrombin in the vicinity of the scissile bond but may result from the greater restraint placed on the RCL in the native state by the anchoring of two RCL residues, P14 and P15, in the ␤-sheet A. Indeed, a fluorescence lifetime study of a N-((2-acetoxy)ethyl-N-methyl)amino-7-nitrobenz-2-oxa-1,3-diazole fluorophore covalently bound to cysteine introduced at P1 as a replacement for arginine showed that the fluorophore experiences a similarly solvent-exposed environment in heparin-free and heparin-bound forms of antithrombin (22). The fluorescence enhancement that the N-((2-acetoxy)ethyl-N-methyl)amino-7-nitrobenz-2-oxa-1,3-diazole fluorophore undergoes upon heparin binding results from a dequenching of about 50% of the N-((2-acetoxy)- FIG. 5. Antithrombin RCL conformation and modeled interaction with thrombin. a, superposition, in stereo, of the conformations of P5-P5Ј residues of ␣ 1 PI Pittsburgh from its noncovalent complex with S195A trypsin and of pentasacharide-activated antithrombin from a structure of the heterodimer (orange) (Protein Data Bank code 1AZX). These RCLs are aligned such that the P1 residues have the same horizontal position. The view is rotated 90°anticlockwise relative to Fig. 1b. b, stereo representation of the P5-P5Ј residues of antithrombin from the x-ray structure of antithrombin heterodimer docked into the thrombin structure found in the noncovalent complex of S195A thrombin with heparin cofactor II, showing inappropriate positioning relative to the catalytic triad and unfavorable interactions with surface loops. ethyl-N-methyl)amino-7-nitrobenz-2-oxa-1,3-diazole, probably as a result of movement of the RCL outward from the serpin body, but without need for a change in backbone conformation at and near to the P1-P1Ј bond.
Significance of Canonical Binding of Serpins-Our finding that ␣ 1 PI Pittsburgh interacts with S195A trypsin analogously to true canonical inhibitors, as had been predicted by Elliott and colleagues (13,14), and the further demonstration that this mode of interaction is also used in two other noncovalent serpin-proteinase complexes, those of heparin cofactor II with S195A thrombin and M. sexta serpin 1B with S195A rat trypsin, raises the question of what makes serpins different from canonical inhibitors. The present findings show that the difference is not due to a fundamental difference in the conformation of the RCLs presented to proteinase. In addition, since proteolytic cleavage of the RCL of canonical inhibitors can take place, cleavage versus noncleavage is not per se a fundamental difference between these types of inhibitor. Instead, the difference appears to be in the position of the equilibrium between the cleaved and intact forms of the inhibitor. With classical canonical inhibitors, the equilibrium constant between intact and RCL-cleaved forms is close to 1 at physiological pH, demonstrating that there is little if any energy difference between these two states as a result of the constraints placed on the RCL. In serpins, however, cleavage of the RCL at the acyl enzyme intermediate stage removes the physical constraint that maintains the RCL as a fully exposed loop at the "top" of the molecule. Since the serpin conformation in which the cleaved RCL has inserted into ␤-sheet A is very much more stable than that in which it remains exposed, there is an enormous thermodynamic push toward the loop-inserted state, once the acyl intermediate has been formed. That some of this energy is used to "crush" the proteinase and hence trap it kinetically is a consequence of the evolution of the complete serpin mechanism. However, it should not detract from the realization that serpins, in their initial interaction with the proteinase, even up to acyl enzyme formation, are behaving as canonical inhibitors.
This conclusion has relevance to the question of how the serpin inhibitory mechanism may have evolved, since the mechanism imposes severe requirements on the serpin for the mechanism to work efficiently. First, the serpin must fold into a metastable state, such that the loop insertion that follows reactive center loop cleavage is highly favorable energetically. Such metastable folding of proteins is extremely rare. Second, the rate at which loop insertion occurs must be fast enough that translocation of the proteinase and subsequent "crushing" of the proteinase must occur fast enough to compete with the alternative of simple hydrolysis of the peptide bond. Inappropriate residues within the hinge of the RCL that might slow down insertion or too slow an opening of ␤-sheet A to accommodate the inserting RCL would be sufficient to inactivate the mechanism or greatly reduce its effectiveness. It is hard to imagine that all of the structural requirements needed by a serpin to make it work properly came about simultaneously. Instead, serpins more likely evolved by a progressive incorporation of features of this mechanism. It is possible that a primordial serpin, which may have originally had an alternative function, had an exposed loop that could interact with a proteinase but was not constrained enough to inhibit it efficiently. Mutations both within the hinge of the loop and within and under ␤-sheet A may have promoted partial insertion of the loop, which would have reduced the conformational freedom of the loop and perhaps made it into a better canonical inhibitor (indeed, there is precedent for such partial insertion in antithrombin). Subsequent changes within the serpin body may have allowed further insertion upon cleavage of the scissile bond and therefore the start of movement of the proteinase, perhaps resulting in a partial compromise of activity. Gradually, full insertion might have evolved to more effectively compromise the proteinase. If all serpins thus started out more like canonical inhibitors with little or no ability to loop-insert into ␤-sheet A, it may be that some serpins whose RCLs are still not able to insert may be like these primordial serpins and actually be inhibitors of the canonical type. This may be especially true if residues within the RCL restrict conformational freedom. A possible example is pigment epithelium-derived factor, which appears incapable of facile loop insertion upon RCL cleavage but has a well exposed RCL that contains rigidifying prolines at positions P10, P8, P4, and P3Ј (23).