Serpin-protease complexes are trapped as stable acyl-enzyme intermediates.

The serine protease inhibitors of the serpin family are an unusual group of proteins thought to have metastable native structures. Functionally, they are unique among polypeptide protease inhibitors, although their precise mechanism of action remains controversial. Conflicting results from previous studies have suggested that the stable serpin-protease complex is trapped in either a tight Michaelis-like structure, a tetrahedral intermediate, or an acyl-enzyme. In this report we show that, upon association with a target protease, the serpin reactive-center loop (RCL) is cleaved resulting in formation of an acyl-enzyme intermediate. This cleavage is coupled to rapid movement of the RCL into the body of the protein bringing the inhibitor closer to its lowest free energy state. From these data we suggest a model for serpin action in which the drive toward the lowest free energy state results in trapping of the protease-inhibitor complex as an acyl-enzyme intermediate.

referred to as the strained or stressed loop, is capable of adopting markedly different conformations relative to the rest of the protein structure (2)(3)(4)(5). This conformational flexibility appears to be necessary for function but can also lead to inactivation when the loop inserts into the main body of the inhibitor, becoming the central strand of the major serpin structural motif, ␤-sheet A (6 -8). This inactive conformation was first observed in RCL-cleaved ␣ 1 -antitrypsin (␣ 1 AT) (9) and more recently in the related structure of latent plasminogen activator inhibitor 1 (PAI-1) (10). Loop insertion leads to a large increase in thermal stability, presumably due to reorganization of the five-stranded ␤-sheet A from a mixed parallel-antiparallel arrangement to a six-stranded, predominantly antiparallel, ␤-sheet (11)(12)(13)(14). This dramatic stabilization has led to the suggestion that native inhibitory serpins may be metastable structures, kinetically trapped in a state of higher free energy than their most stable thermodynamic state. Such an energetically unfavorable structure would almost certainly be subject to negative selection, and thus its retention in all inhibitory serpins implies that it has been conserved for functional reasons.
Currently, the role of loop mobility in serpin function and the structure of the serpin-protease complex are controversial (15)(16)(17)(18)(19)(20)(21)28). In the late 1970s, it was reported that serpins were unlike other tight binding protease inhibitors and formed covalent ester linkages with enzymes (15). However, these conclusions were based on SDS-PAGE analysis of denatured complexes leaving the nature of the native complex open to question. Later investigations suggested that the native serpin-protease complex may be reversible and therefore could not be covalent but instead might form a Michaelis-like complex similar to the tight binding inhibitors of the Kunitz and Kazal families (16). Finally, a recent NMR study suggests a stable tetrahedral configuration (19). To distinguish between these alternative structural models, we developed methods to monitor the position of the RCL in the inhibitor-enzyme complex and to determine the chemical nature of the association between serpins and their target proteases.

EXPERIMENTAL PROCEDURES
Materials-PAI-1 mutants containing Cys substitutions at either the P9 position (Ser-338) or the P1Ј (Met-347) 2 of the RCL were constructed by site-directed mutagenesis as described (14). Both mutants were purified and labeled with the environmentally sensitive probe NBD (Molecular Probes) as described (21). The labeled mutants retained full inhibitory activity toward both urokinase (uPA) and tissue-type plasminogen activator (tPA), with second order rate constants for inhibition Ͼ10 6 M Ϫ1 s Ϫ1 in all cases. Recombinant high molecular weight uPA was a generous gift of Dr. J. Henkin of Abbott Laboratories, and two-chain tPA was prepared from Activase (Genentech) as described previously (21). Porcine pancreatic elastase was from Elastin Products (Owensville, MO), and human ␣ 1 AT was from Athens Research and Technology (Athens, GA). 125 I-labeled Bolton-Hunter reagent (22) (monoiodinated) was from DuPont NEN, and N-hyroxysuccinimide acetic acid (NHS) was from Sigma.
Stopped-flow Fluorescence Analysis-Stopped-flow fluorimetry was performed on both PAI-1 NBD derivatives as described (21). Bolton-Hunter Labeling-Wild type PAI-1 or substrate PAI-1 (Thr-333 3 Arg (7)) at a concentration of 1 M was incubated Ϯ 1 M uPA or tPA in 50 mM sodium phosphate, pH 7.6, 150 mM NaCl at 23°C for 30 min, followed by the addition of 125 I-labeled Bolton-Hunter reagent to 100 nM. The samples were incubated an additional 30 min followed by the addition of 1/10 volume of stop solution (1 M glycine, 50 mM Tris, 150 mM NaCl, pH 7.75) and incubation continued for an additional 30 min. The labeled proteins were then separated from free 125 I-Bolton-Hunter by precipitation with 50% saturated ammonium sulfate for 30 min at 4°C followed by centrifugation at 14,000 ϫ g for 20 min. The supernatant, containing free 125 I-Bolton-Hunter, was discarded, and the pellets were washed twice with 50% saturated ammonium sulfate. The pellets were resuspended in 20 l of 1% SDS, 10 mM Tris, pH 7.4, 1 mM EDTA, and ϳ250,000 cpm of each sample was subjected to SDS-PAGE on a 20% homogeneous gel (PhastGel, Pharmacia Biotech Inc.) followed by staining with Coomassie Brilliant Blue and autoradiography.
Blocking of Free Amino-terminal Residues-Inhibitor (1 M) was incubated with enzyme (1 M) for 30 min at 23°C in 50 mM sodium phosphate, pH 7.6, 150 mM NaCl, after which NHS was added to either 100 M or 1 mM. The samples were incubated an additional 30 min followed by the addition of 1/10 volume of 1 M Tris, pH 7.5, and continued incubation for 30 min. The Tris was then removed by ultrafiltration and washing with distilled water in a Prospin column (Applied Biosystems) prior to automated amino-terminal sequence analysis using Edman chemistry (Applied Biosystems model 473A). Pretreatment of PAI-1 or tPA with NHS at both NHS concentrations tested did not significantly affect the activity of either protein, indicating that, under these conditions, NHS treatment alone does not result in protein denaturation.
Molecular Modeling-The model for active PAI-1 has been described previously (23). The model for cleaved PAI-1 was generated using Quanta (Burlington, MA) from the coordinates of latent PAI-1 (generously provided by Dr. E. Goldsmith) and from the coordinates of cleaved ␣ 1 AT obtained from the Brookhaven data base. The root mean square difference of the C␣ trace with latent PAI-1 is Ͻ0.04 Å and 2.2 Å with cleaved ␣ 1 AT.

RESULTS AND DISCUSSION
Stopped-flow Fluorescence Analysis-In the first series of experiments, site-directed mutants of the serpin PAI-1 were constructed with Cys residues at either the P9 or the P1Ј position of the RCL. Each mutant was then labeled with the fluorescent probe NBD. This probe shows a large enhancement in fluorescence when moved from an aqueous environment to a hydrophobic milieu. Both mutants were then reacted with tissue-type plasminogen activator (tPA) in a stopped-flow fluorimeter, and these results are shown in Fig. 1. Reaction of the P9-NBD PAI-1 with tPA resulted in a large and rapid enhancement of the relative fluorescence ( Fig. 1) together with a 13 nm blue spectral shift (data not shown), consistent with our previous data (21). The extent of this change is nearly identical with that observed during the transition to the latent conformation (21) and is consistent with insertion of the RCL into ␤-sheet A and the resultant burying of the P9 residue beneath ␣-helix F. The observed rates of loop insertion over a range of tPA concentrations are shown in the inset of Fig. 1. These data yield a limiting rate constant for this reaction of ϳ4 s Ϫ1 (t1 ⁄2 for insertion of ϳ250 ms). In contrast, reaction of the P1Ј-NBD PAI-1 with tPA resulted in a 30% decrease in relative fluorescence occurring at approximately the same rate (Fig. 1). This quench indicates that, unlike the P9 position of the RCL, the P1Ј side chain is exposed to a more hydrophilic environment upon reaction with tPA. Although such a shift in position could result from a minor conformational change in the RCL, this explanation seems unlikely given the close association of the serpin and protease via the directly adjacent P1 Arg residue of PAI-1 and the S1 subsite of tPA. Alternatively, cleavage of the PAI-1 RCL by tPA between the P1 and P1Ј residues could permit the NBD reporter group to move away from the enzyme into a more aqueous environment. In either case, the limiting rate of the reaction, calculated from the data in the inset to Fig. 1 is similar to that observed with the P9-NBD mutant (ϳ8 s Ϫ1 versus 4 s Ϫ1 ), indicating that changes at the P1Ј site must occur either immediately preceding or concurrent with loop insertion. Since the maximum changes in fluorescence are stable over time (data not shown), the reaction being monitored in the stopped flow is proceeding to completion and thus represents formation of the stable inhibitor-enzyme complex and not a transient or intermediate reaction.
Bolton-Hunter Labeling of Free Amines-To distinguish between cleavage of the P1-P1Ј peptide bond versus solely a conformational change in the intact RCL, the PAI-1⅐PA complex was reacted with the amino-specific 125 I-Bolton-Hunter reagent (22). Two different plasminogen activators were used, tPA and uPA. These experiments were conducted with trace amounts of 125 I-Bolton-Hunter reagent under nondenaturing conditions, followed by treatment of the unreacted label with glycine and removal prior to SDS-PAGE analysis. This procedure should report the presence of a cleaved RCL in the complex by the appearance of a novel labeled peptide fragment of the correct size. Furthermore, since the unreacted Bolton-Hunter reagent was blocked with glycine and removed before the samples were denatured by exposure to SDS, any labeled peptide must have been formed while the complexes were in their native state. Although all accessible amines could potentially be labeled, including ⑀-NH 3 ϩ groups of internal Lys residues, the only position that can incorporate label in the PAI-1 RCL carboxyl-terminal peptide would be its amino terminus, since PAI-1 contains no Lys residues in the 33-residue peptide produced by cleavage of the P1-P1Ј bond. SDS-PAGE analysis of the labeled complexes shown in Fig. 2 demonstrated a unique ϳ3.0-kDa band with both PAI-1⅐PA complexes which was not present with PAI-1 or either PA alone (B). The observed mobility of this novel peptide is consistent with the predicted molecular mass (3.8 kDa) of the PAI-1 carboxylterminal peptide. As a positive control for cleavage of the RCL and the labeling efficiency of the C-terminal peptide, we also tested a mutant PAI-1 that we have previously shown is a pure substrate for plasminogen activators and is completely cleaved at its RCL P1-P1Ј peptide bond (7). Consistent with previous observations, the mutant PAI-1 fails to form stable complexes with either PA and is instead completely cleaved (Fig. 2A,  compare lanes 2 and 3 with lanes 7 and 8). Furthermore, a labeled peptide identical with that observed with wild type PAI-1⅐PA complexes is also seen (Fig. 2B, compare lanes 2 and  3 with lanes 7 and 8). This similar efficiency of peptide labeling for the mutant PAI-1 cleaved by either uPA or tPA compared to wild type PAI-1 in association with each enzyme indicates that the RCL within the stable serpin-protease complex is cleaved and suggests that this cleavage is complete.
Quantitation of Free Amino-terminal Residues-To confirm complete cleavage of the PAI-1⅐PA complex and exclude substrate behavior by a subset of the inhibitor molecules, the extent of RCL cleavage was directly quantitated by microsequencing of the PAI-1⅐uPA complex. Since complex denaturation during the sequencing reaction could potentially induce cleavage, the extent of cleavage in native complexes was determined by a subtractive method. PAI-1⅐uPA complexes were first reacted with the amino-specific reagent NHS under nondenaturing, physiological conditions. This compound is similar to Bolton-Hunter reagent in its reactivity and covalently binds to free amines. Treatment of PAI-1⅐uPA complexes with NHS should therefore block available amino termini in a dose-dependent manner. The excess NHS was then reacted with Tris and removed by ultrafiltration prior to direct amino-terminal sequence analysis of remaining unreacted amino termini. This analysis is quantitative, and the relative reactivity of natural amino termini from both the inhibitor and protease serve as internal controls for NHS reactivity and sequencing efficiency. The results of this analysis are shown in Table I. The yield of RCL peptide amino terminus at each dose of NHS is very similar to those of the natural PAI-1 and uPA amino termini. Thus, the RCL peptide amino terminus generated upon complex formation is as reactive as the natural amino termini and therefore is very likely to be fully cleaved and exposed, consistent with the quenched fluorescence of the P1Ј-NBD PAI-1 and the 125 I labeling results. Interestingly, the least NHS-reactive amino terminus tested is that of the uPA heavy chain (uPAhc ,  Table I), a relatively hydrophobic sequence Ile-Ile-Gly-Gly which is likely to be oriented toward the interior of the molecule (24). The latter observation suggests that this approach is quite sensitive to amino termini solvent accessibility, further supporting the conclusion that the PAI-1 RCL must be completely cleaved when in complex with uPA. Similar results were also obtained with PAI-1⅐tPA complexes (data not shown).
To test the general relevance of these observations for other serpin-protease complexes, the ␣ 1 AT⅐elastase complex was also treated with NHS and subjected to microsequencing. The results shown in Table II are similar to the data obtained with the PAI-1⅐uPA complexes (Table I), demonstrating that ␣ 1 AT is also cleaved in its RCL when in complex with elastase. Taken together with the known stability of these complexes to SDS-PAGE, these observations strongly suggest that the serpinprotease complex is trapped in the form of a covalent acylenzyme intermediate. Serpin inhibition appears to be a twostep process with an initial reversible encounter complex followed by formation of an apparently irreversible stable com-  plex (21,25). 3 It is likely that the previous studies suggesting complex reversibility were observing this first step (16). Peptide bond cleavage by serine proteases is known to proceed through two tetrahedral intermediates (26). Although our data rule out a complex frozen at the point of the first tetrahedral intermediate (19), prior to RCL cleavage, it is possible that the serpin-protease complex is frozen at the point of the second tetrahedral, following RCL cleavage, or that it is trapped in a distorted conformation intermediate between these two forms.
A Model of Serpin Function-Based on these data, we propose that upon encountering a target protease, serpins bind to the enzyme forming a Michaelis-like encounter complex. This relationship is similar to a protease's interaction with its substrate, progressing to the point of cleavage of the P1-P1Ј peptide bond and formation of the first covalent acyl-intermediate. Cleavage is coupled to a rapid insertion of the RCL into ␤-sheet A, with exposure of the P1Ј residue to solvent and burying of the P9 residue. Fig. 3 shows the relative position of the RCL and the P1Ј and P9 residues in molecular models of active and cleaved PAI-1. Since the inserting RCL is covalently linked to the enzyme via the active site Ser, this transition should also affect the protease and significantly change its position relative to the inhibitor. Such a rapid shift in the relative positions of the two molecules, centered at the enzyme's active site, might sufficiently distort the active site geometry to prevent efficient deacylation and thus trap the complex. Alternatively, the new position of the acylated protease's active site may prevent the water necessary for deacylation from entering the active site. This model is consistent with our previous results demonstrating that RCL insertion is not required for protease binding but is necessary for stable inhibition (7) as well as the observation that only an active enzyme can induce RCL insertion. 3 We suggest that native serpin structures are kinetically trapped in a conformation which is not their most stable structure, and that the stored drive toward a lower energy structure results in trapping of the protease-inhibitor complex in the acyl-enzyme intermediate form.