Crystal Structure of the Complex of Plasminogen Activator Inhibitor 2 with a Peptide Mimicking the Reactive Center Loop*

The structure of the serpin, plasminogen activator inhibitor type-2 (PAI-2), in a complex with a peptide mimicking its reactive center loop (RCL) has been determined at 1.6-Å resolution. The structure shows the relaxed state serpin structure with a prominent six-stranded β-sheet. Clear electron density is seen for all residues in the peptide. The P1 residue of the peptide binds to a well defined pocket at the base of PAI-2 that may be important in determining the specificity of protease inhibition. The stressed-to-relaxed state (S → R) transition in PAI-2 can be modeled as the relative motion between a quasirigid core domain and a smaller segment comprising helix hF and β-strands s1A, s2A, and s3A. A comparison of the Ramachandran plots of the stressed and relaxed state PAI-2 structures reveals the location of several hinge regions connecting these two domains. The hinge regions cluster in three locations on the structure, ensuring a cooperative S → R transition. We hypothesize that the hinge formed by the conserved Gly206 on β-strand s3A in the breach region of PAI-2 effects the S → R transition by altering its backbone torsion angles. This torsional change is due to the binding of the P14 threonine of the RCL to the open breach region of PAI-2.

The serpins are an unusual class of protein in that they fold into a metastable structure (the stressed state) that can usually undergo a major structural change to an extremely stable form (the relaxed state) on cleavage by a target protease (1). Most serpins are serine (or cysteine) protease inhibitors that act in a suicide fashion, using the stability of the relaxed state to trap the protease in a nonfunctional covalent serpin-protease complex. The target protease cuts the accessible reactive center loop (RCL) 1 on the serpin to produce a covalent acyl interme-diate (2,3). The RCL then inserts into ␤-sheet 2 A to form the relaxed state. Attached to the RCL, the target protease moves to the base of the serpin, where it is partially unfolded, thus inhibiting its activity (4,5). This mode of action, whereby one protein uses the increased stability of an altered conformer to partially unfold a second, covalently bound target protein, is unprecedented in biology.
Plasminogen activator inhibitor 2 (PAI-2) is a serine protease inhibitor that belongs to the ov-serpin branch of the serpin superfamily (6). PAI-2 is an effective inhibitor of urinary plasminogen activator (urokinase or uPA) and, to a lesser extent, an inhibitor of tissue-type plasminogen activator. PAI-2 has several features that distinguish it from the more widely studied serpins. PAI-2 lacks an N-terminal secretory signal but contains a relatively inefficient internal signal sequence (7). This produces a distribution where a majority (ϳ70%) of the protein is nonglycosylated and intracellular, with the rest in a glycosylated, extracellular form. PAI-2 has a unique insertion of ϳ30 residues between helices hC and hD (CD loop), which is the site of glycosylation and the site of cross-linking to the cell membrane and also fibrin via transglutaminase action (8). The CD loop also binds a number of largely unidentified cytoplasmic proteins (9). PAI-2 is expressed by specific cell types including epithelial cells, monocytes/macrophages, and keratinocytes (10). It is expressed at high levels in the skin, hair follicles, gingival cells, the cervix, and placenta. PAI-2 is crosslinked to create the cornified envelope in terminally differentiated epithelial cells, where it forms a proteinaceous coat, with PAI-2 in both the active and relaxed forms (11,12).
PAI-2 appears to have several distinct biological functions (10). To date, complexes between PAI-2 and uPA have not been observed in vivo. Several studies have indicated that the intracellular pool of PAI-2 plays a role in apoptosis (13,14). The up-regulation of PAI-2 expression by inflammatory mediators and its high level of expression in macrophages and keratinocytes suggests a role in the regulation of the inflammatory response and wound healing. The high levels of PAI-2 observed in human pregnancy plasma indicate a role in maintenance of the placenta or perhaps embryonic development (15). However, PAI-2 (Ϫ/Ϫ) null mice exhibit essentially normal development, survival, and fertility (16).
Crystallographic studies have been central to understanding * 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: RCL, reactive center loop; PAI, plasminogen activator inhibitor; uPA, urinary plasminogen activator or urokinase; S 3 R, stressed-to-relaxed state; LEI, leukocyte elastase inhibitor; r.m.s., root mean square. 2 All serpin secondary structure elements are labeled as per Loebermann et al. (17). serpin function. It was clear from the initial structure of cleaved ␣ 1 -proteinase inhibitor (relaxed state) that the mechanism of protease inhibition was unusual (17). Subsequently determined structures of serpins in various states have led to models for the stressed to relaxed state (S 3 R) transition, which is central to protease inhibition (18,19). The determination of the stressed state structure of a PAI-2 CD loop deletion mutant has contributed to the understanding of the mechanics of this transition, especially with regard to the conserved shutter region (20).
We have determined the structure of the complex formed by the PAI-2 CD loop deletion mutant with a peptide mimicking the RCL at 1.6-Å resolution. The complex adopts the relaxed state, facilitating a high resolution analysis of the S 3 R transition in PAI-2. By comparing the Ramachandran plots of PAI-2 in the two states, we have identified key hinge regions that facilitate the transition. These are clustered into three regions coordinating the transition. In particular, we hypothesize that the binding of P14 threonine 3 in the breach region of PAI-2 triggers the transition by altering the backbone torsion angles of conserved Gly 206 (Gly 192 ). 4

EXPERIMENTAL PROCEDURES
Protein Expression, Purification, and PAI-2⅐Peptide Complex Formation-Recombinant human PAI-2 loop deletion mutant was cloned and expressed in Escherichia coli and purified as described by Jensen et al. (8). This mutant lacks the loop between helices hC and hD (residues 66 -98, inclusive, in the PAI-2 gene sequence). In order to generate a relaxed form of the protein, a synthetic peptide with an acetylated N terminus (N-acetyl-Thr-Glu-Ala-Ala-Ala-Gly-Thr-Gly-Gly-Val-Met-Thr-Gly-Arg-OH (Auspep, Australia)), which mimics the RCL loop, was incubated with PAI-2 (1 mg/ml) at a 100-fold molar excess of the RCL peptide in 50 mM NaCl, 50 mM Tris, pH 8.0, at 37°C for 48 h (21).
Data Collection and Structure Determination-Crystals were cryoprotected by serially transferring them through a set of solutions containing 17% polyethyleneglycol 8000, 0.1 M Mg(CH 3 COOH) 2 , 0.1 M Hepes, pH 7.5, and increasing concentrations of ethylene glycol (1% up to 16% in 1% steps) prior to freezing directly in liquid nitrogen. Synchrotron data were collected on station 9-2 at Stanford Synchrotron Radiation Laboratory using an ADSC Quantum 4 CCD detector with x-rays at 1.0-Å wavelength. Data were integrated using the program MOSFLM (22) and then scaled and reduced with the program SCALA, which is part of the CCP4 suite of programs (23). The structure of the stressed form of PAI-2 was used as a molecular replacement model (20) to provide initial phasing using the program AmoRe (24). Initial electron density maps showed that the PAI-2⅐peptide complex adopted the relaxed serpin structure with the peptide inserted as ␤-strand s4A in ␤-sheet A. A model was built using the interactive graphics program O (25) and then improved by several cycles of model building and refinement using both REFMAC (26) and CNS (27). The final model consists of PAI-2 residues 2-61, 102-215, 220 -270, 274 -367, and 378 -415, where Cys 161 is modified by ␤-mercaptoethanol; the 14 residues of the RCL peptide plus its N-terminal acetylation; and 440 water molecules. For 15 residues, alternative side chains have been modeled (seven of which are serines). Data reduction and refinement statistics are pre-sented in Table I. The coordinates and structure factors have been deposited with the Protein Data Bank (accession code 1JRR).

Structure Determination
The structure of a complex between a PAI-2 deletion mutant (where residues 66 -98, which form the CD loop, have been genetically removed) and a peptide mimicking the RCL loop has been determined at 1.6-Å resolution. The data reduction and refinement statistics are given in Table I. The electron density maps showed unambiguous electron density for all residues of the protein with the exception of the N terminus (residue 1); the remains of the CD loop (residues 62-65 and 99 -101); a loop before ␤-strand s4C (residues 216 -219); a loop joining ␤-strand s3B to helix hG (residues 271-273); and a portion of the endogenous RCL (residues 368 -377). Clear density was observed for all 14 residues in the RCL peptide.

Description of the Structure
The structure of the PAI-2⅐RCL peptide complex is typical of that seen in relaxed state serpins. The prominent ␤-sheet A contains six strands, with the RCL peptide forming strand s4A in the center of the sheet (Fig. 1). The structure is similar to that of the complex formed between antithrombin and a P14 -P3 RCL peptide (28).
The PAI-2 molecule in the complex is intact; thus, it has a complete endogenous RCL, which is distinct from the peptide. Clear density is seen for approximately half of the endogenous RCL of the molecule (residues 364 -367 and 378 -386), which adopts an extended conformation (Fig. 1). The observed endogenous RCL residues are stabilized by crystal contacts with two neighboring molecules in the crystal. Such stabilization appears to be essential, since crystal forms lacking appropriate crystal contacts show no density for the stressed state, endogenous RCL (20,29).

Peptide Binding
Breach Region-In the breach region (19), P14 Thr of the peptide forms a hydrogen bond with the side chain of Tyr 258 3 Residues in the RCL (specifically the RCL peptide) are labeled P1, P2, etc. following the protease convention (42). 4 Residue numbers prefixed with a three-letter amino acid code are based on the translated PAI-2 gene sequence. To aid comparison with other serpins, these are sometimes accompanied by canonical serpin numbers (37) in parentheses. Residues in the RCL peptide are numbered using the convention for protease targets (42) followed by the three-letter amino acid abbreviation. (Tyr 244 ) ( Fig. 2A), displacing the three water molecules seen in the stressed state structure of PAI-2 (20). The P14 Thr backbone forms two hydrogen bonds with Gly 206 (Gly 192 ), which may be important in initiating the S 3 R transition. The acetyl modification of the peptide N terminus mimics the peptide backbone by continuing the ␤-sheet hydrogen bonding to Asn 363 (Asn 341 ). This chemical modification is essential for peptide binding (21). P1 Binding Pocket-The structure shows that PAI-2 has a well defined binding pocket for the P1 arginine residue of the inserted RCL peptide (Fig. 2B). The site is formed by the long loop that joins helix hI to ␤-strand s5A at the base of the molecule. The guanidinium group of the arginine is stabilized by a network of hydrogen bonds to the backbone carbonyl groups of Lys 335 (Lys 314 ), Ala 338 (Ala 316 ), and Asn 347 (Asn 325 ) as well as a hydrogen bond to the side chain oxygen of Asn 347 (Asn 325 ). An examination of the stressed state structure shows that the P1 binding pocket is essentially intact in both states, requiring only small backbone torsional changes between residues Ser 344 and Leu 349 and rotamer changes in Asn 347 and Asp 348 to accommodate the incoming arginine.

Comparison with Other Serpin Structures
An overlay of the PAI-2⅐RCL peptide structure with that of other relaxed state serpin structures shows considerable similarity (data not shown). Least-squares comparisons show that r.m.s. deviations between relaxed state serpin structures and the PAI-2⅐RCL peptide complex range between 1.0 and 1.4 Å, while the r.m.s. deviation between stressed state structures and the complex are ϳ2.0 Å (Table II). As a comparison, the stressed state PAI-2 structure has an r.m.s. deviation compared with other stressed state structures of 1.5-1.7 Å, and relaxed state structures range from 1.5 to 2.2 Å. The alignments of the PAI-2⅐RCL peptide complex with other relaxed state structures overlay most of the secondary structural ele-ments with nearly all of the structural differences confined to the loops. There are, however, several regions of structural difference.
There is a significant structural rearrangement of the PAI-2⅐RCL peptide complex in the neighborhood of Trp 294 (Trp 275 ), which is located at the C terminus of helix hH (Fig. 3A). Trp 294 is conserved in many serpins, and an overlay of the structures of leukocyte elastase inhibitor (LEI) (30), ovalbumin (31), antithrombin (32), and the stressed state of PAI-2 shows that its side chain is located in a pocket formed by helix hH and ␤-strands s1B and s2B (Fig. 3A, red). Trp 294 in the PAI-2⅐RCL peptide complex differs from this consensus, with the side chain adopting a different rotamer (Fig. 3A, green). The electron density and the B-factors for this region indicate that it may be strained compared with the neighboring regions (mean B-factor for residues Asp 289 -Glu 302 is 31 Å 2 compared with 18 Å 2 for Asn 242 -Ile 246 (s1B) and 15 Å 2 for Pro 393 -Ser 413 (s4B and s5B)).
The altered structure of Trp 294 in the PAI-2⅐RCL peptide complex appears to be due to a packing interaction that transmits a structural change over ϳ10 Å. The side chain of Tyr 245 (Tyr 231 ) on ␤-strand s1B is at the surface of the molecule. An overlay of the stressed state and relaxed state structures of PAI-2 shows that the side chain of Tyr 245 has moved significantly (1.4 Å), along with the backbone of ␤-strands s1B and s2B (Fig. 3A). The cause of this rearrangement appears to be a steric clash between Tyr 245 and the salt bridge formed between Glu 140 (Glu 129 ) and Arg 143 (Arg 132 ) in a neighboring molecule. Although this structural rearrangement appears to be a result of the transmission of a distal packing interaction, the observed change indicates that there is some plasticity in this region of PAI-2.

Comparison of the Relaxed and Stressed States of PAI-2
A comparison of the relaxed state PAI-2⅐RCL peptide structure with the stressed state PAI-2 structure shows all of the general features of the S 3 R transition that have been obtained from comparisons of different serpins (18) or the same serpin at lower resolution (19). Given this, we will focus mainly on differences that are specific to PAI-2 and comparisons that are warranted by the higher resolution of our structures.
An important difference between stressed and relaxed state serpins is the existence of a cavity beneath ␤-sheet A in the stressed state, which originates between s2A, hD, and hE (20,29,33,34). The closure of this cavity in the relaxed state is likely to contribute to its enhanced stability and thus to its mechanism of protease inhibition. In the stressed state of PAI-2, this cavity contains three ordered water molecules (20). It is also partially filled by Trp 33 (Trp 55 ), which adopts two conformers, with the major one facilitated by the opening of the cavity ( 1 ϭ 167°, 2 ϭ Ϫ99°). In the relaxed state, Trp 33 is well ordered, adopting only the minor conformer seen in the stressed state ( 1 ϭ 160°, 2 ϭ 66°), where the indole ring is flipped by 180°, allowing the closure of the cavity between ␤-sheet A and the body of the serpin.
Several changes occur in the position of helix hF in the PAI-2 S 3 R transition, which differ from those observed in other serpins. In most serpins, helix hF moves along with ␤-strands s3A and s2A. However, in PAI-2, helix hF appears to move upwards and closer to ␤-sheet A during the S 3 R transition (Fig. 3, B and C). A comparison of the PAI-2⅐RCL peptide complex with other relaxed state structures shows that the position of helix hF in relaxed state PAI-2 is typical; however, a comparison of stressed state PAI-2 with other stressed state structures shows that helix hF is different in PAI-2.
To preclude packing artifacts, we have examined crystal  with a second molecule via van der Waals interactions between Glu 187 (Glu 175 ) and Ser 189 (Ser 175B ) and between Pro 221 (Pro 207 ) and Arg 229 (Arg 215 ), respectively. None of these packing interactions appears to account for the difference in the orientation of helix hF with respect to ␤-sheet A in the stressed state of PAI-2. Thus, we believe that this is a genuine reflection of the structure of stressed state PAI-2 in solution.
The closure of the cavity below ␤-sheet A and the movement of helix hF appear to be correlated with a gross change in ␤-strands s2A and s3A. In the stressed state structure, both ␤-strands are relatively straight, while in the relaxed state structure, there appears to be a kink centered at residues Asn 127 (Asn 116 ) and Asn 200 (Asn 186 ) on ␤-strands s2A and s3A, respectively. The angle of the kink is 158°and 164°for ␤-strands s2A and s3A, respectively (cf. 178°for each strand in the stressed state).
Apart from these specific changes that occur on transition between the stressed and relaxed states, there is a general ordering of the structure in the relaxed, PAI-2⅐RCL peptide structure compared with the stressed state. This is reflected in the average temperature factors for all atoms being 20 and 38 Å 2 for the relaxed and stressed state structures, respectively. The stressed state also contains more regions that are disordered such as the loop joining helix hD to ␤-strand s2A.

Transition between the Stressed and Relaxed States
A least squares superposition of the stressed state and relaxed state structures of PAI-2 gives a remarkably good fit when most structural elements are included. Using LSQMAN (35) in the program O (25), the optimized r.m.s. deviation between the structures was 1.38 Å over 298 C-␣ atoms including all secondary structural elements with the exception of ␤-strands s1A, s2A, and s3A and helix hF. Visual examination of the superposition shows that all remaining secondary structure elements fit well with the exception of the following: helix hE, which is translated by ϳ2 Å and tilted by 8.3°; helix hD, which is tilted by 10.5°; and helix hH, which moves as a result of the distortion caused by Trp 294 (Table III). Based on the above, PAI-2 can be partitioned into three segments: a quasirigid unit comprising helix hF and ␤-strands s1A, s2A, and s3A; helix E; and a quasirigid core formed by all remaining secondary structure elements. An optimized least-squares superposition of the quasirigid core of PAI-2 in the stressed and relaxed states results in an r.m.s. deviation of 0.85Å over 250 C-␣ atoms (cf. 1.0 -1.6 Å between different relaxed state serpins; Table II). Thus, the S 3 R transition can be viewed as the relative motion of two quasirigid domains, with an adjustment of helices hD and hE, which is similar to previous analyses (18,19). The figure compares the structure of the relaxed state PAI-2⅐RCL peptide complex (green) with the stressed state structure of PAI-2. A, an examination of the region around Trp 294 . The structure of the stressed state in this region most closely resembles that of other serpins. The structure of the relaxed state appears to be perturbed by crystal packing, which forces Tyr 245 to move down, displacing ␤-strands s2B and s3B. This structural change is accommodated by Trp 294 , adopting a different side chain rotamer as well as a distortion in the loop connecting helix hH with ␤-strand s2C. B, a comparison of the position of helix hF in the stressed state (red) with the relaxed state (green). The structures are aligned using the quasirigid core domain of PAI-2 (see "Results"), which results in a close superposition of the A ␤-sheets. The RCL peptide is shown in blue. C, a side view of the same overlay showing that while the A ␤-sheets are well aligned, helix hF in the stressed state (red) has moved down and away from the body of the serpin.
Protein structural transitions involving quasirigid units are often mediated by the existence of hinges. If the S 3 R transition of PAI-2 is a quasirigid two-domain motion as described above, then one would expect to find hinges in the following locations: the loop between helix hD and s2A; the loop between s2A and hE; the loop between hE and s1A; and the C terminus of s3A. These four hinges would be sufficient to allow the two domains to move in a quasirigid fashion.
In order to locate possible hinges in an unbiased fashion, we compared the Ramachandran plots of PAI-2 in the stressed and relaxed states and computed a Ramachandran distance, D, for each residue as follows.
For each region, we examined the electron density maps for both stressed state and relaxed state structures and compared the overlay of the structures. For all regions with the exception of Glu 187 -Gly 188 (Glu 175 -Gly 175A ), the electron density was unambiguous, and the structural transition was judged to be real. The electron density for Glu 187 -Gly 188 in the stressed state is ambiguous. This segment is in the loop joining helix hF with ␤-strand s3A (Fig. 4C), and there is no structural reason for a transition. Almost all of the large transitions involve a peptide flip between the key residues.
Ala 259 -Gly 260 -Asp 261 (Ala 245 -Gly 246 -Asp 247 ) forms a turn between ␤-strands s1B and s2B. The observed transition involves peptide flips on either side of Gly 260 . The transition in this region is likely to be coupled to the disorder-order transition that occurs in the loop joining helix hD to ␤-strand s2A (Fig.  4C).
Asp 298 -Lys 299 -Met 300 (Asp 279 -Lys 279A -Met 279B ) lie in the region containing Trp 294 (Trp 275 ). The structural changes observed in this region are all linked to the side chain rotamer change of Trp 294 . This in turn is likely to be a crystal-packing artifact as discussed above.
The transition at Arg 346 -Asn 347 (Arg 324 -Asn 325 ) is in the loop joining helix hI and ␤-strand s5A. These residues form part of the P1 binding pocket (Fig. 2B) and are intimately linked with RCL insertion. The structural change is also coupled to that observed between helix hE and ␤-strand s1A (Fig. 4C). Fig. 5 shows this region in detail in both the stressed and relaxed states. Changes in Asn 347 (Asn 325 ) are directly linked to the binding of P1 Arg to the P1 pocket, where the Asn 347 carbonyl group forms a hydrogen bond to the guanidinium moiety of P1 Arg. P1 binding also displaces the side chain of Asp 348 (Asp 326 ), which forms a hydrogen bond to the side chain of Ser 344 (Ser 322 ) in the relaxed state. The switch in Ser 344 rotamer breaks the link with Ser 136 (Ser 125 ) (observed in the stressed state), altering the relationship between ␤-strand s2A and helix hE. Thus, there appears to be a coordination between the binding of P1 Arg to the loop between hI and s5A and the transitions between s2A and hE.
Several of the smaller changes (Ramachandran distance Ͼ 30°) lie in interesting regions of PAI-2. Gly 336 (Gly 314A ) forms part of the P1 binding pocket and the observed change is directly linked to P1 Arg binding (Fig. 5). Asn 127 (Asn 116 ) and Asn 200 (Asn 186 ) are adjacent to each other on ␤-sheet A (Fig.  4C). They are at the apex of the kink that is observed when ␤-strands s2A and s3A move to the relaxed state. Asn 200 is also part of the shutter region of PAI-2 (20). Gly 206 (Gly 192 ) and Lys 207 (Lys 193 ) line the breach region on ␤-strand s3A ( Fig. 2A). The transition in these residues appears to be the key to opening ␤-sheet A, and hence, initiating the S 3 R transition.

DISCUSSION
By determining the structure of the PAI-2⅐RCL peptide complex at high resolution, we are able to examine the structural transition that occurs in PAI-2 as the protein moves from the stressed to the relaxed state. A comparison of the two structures supports the simple model that the S 3 R transition involves the quasirigid motion of a small domain consisting of helix hF and ␤-strands s1A, s2A, and s3A, relative to a quasirigid core formed by the remaining secondary structure elements, as proposed by previous analyses (18,19).
By analyzing large scale changes in the Ramachandran plot, we have been able to identify most of the hinge regions that facilitate the relative motion of the two quasirigid segments of PAI-2. These include segments linking both ends of helix E and the top of ␤-strand s3A. The hinge connecting helix hD to ␤-strand s2A undergoes a disorder-to-order transition, and it is linked to the neighboring hairpin connecting ␤-strands s1B to s2B.
The regions that are identified by analyzing the Ramachandran plot tend to cluster within PAI-2, indicating that these transitions are correlated. The hinge region between helix hE and ␤-strand s2A is adjacent to the two hinge regions that straddle the loop between helix hI and ␤-strand s5A, which forms the P1 binding site for the inserted RCL peptide. This indicates that signaling occurs across the bottom of PAI-2. Similarly, the hinge region linking helix hE to ␤-stand s1A is adjacent to Asn 127 (Asn 116 ) and Asn 200 (Asn 186 ), which are at the apex of the kink seen in ␤-sheet A in the relaxed state. Finally, the loop between helix hD and ␤-strand s2A, the hairpin between ␤-strands s1B and s2B, and the hinge at the top of ␤-strand s3A in the breach region are all adjacent to each other.

TABLE III
Movement of ␣-helices in the S 3 R transition Changes in position and orientation of the ␣-helices of PAI-2 were determined by first overlaying the quasirigid core using the program LSQMAN (35). Each helix was then mapped from the stressed to the relaxed state structure using LSQMAN. The helices of PAI-2 are as defined in Harrop et al. (20). Thus, a picture emerges of three regions that are critical for the S 3 R transition in PAI-2. An examination of the structural details in each of these hinge regions shows that, for the most part, they appear to be passive, facilitating the transition but not initiating it. There is clear communication across the bottom of the molecule sensing the binding of P1 Arg to the P1 pocket; however, this event occurs after much of the transition has occurred. The question remains as to which event triggers the S 3 R transition.
Our hypothesis, based on these structures and the results of the accompanying paper (36), is that the binding of P14 Thr to the breach region is the critical event that triggers the S 3 R transition. Several pieces of evidence support this hypothesis. First, the breach site is open in the stressed state, and P14 Thr is the only residue in the RCL that can enter its relaxed state site as part of ␤-strand s4A. On binding, P14 Thr forms two backbone hydrogen bonds with Gly 206 (Gly 192 ), and these bonds effect a torsional change within the residue resulting in the transition detected by the Ramachandran analysis. This glycine residue is highly conserved in all serpins (20,37,38). The backbone transition of Gly 206 is then propagated in the Nterminal direction along ␤-strand s3A, resulting in the opening of ␤-sheet A and the insertion of the RCL.
The transmission of the structural change in the N-terminal direction along ␤-strand s3A is due to the fact that Gly 206 is anchored in the C-terminal direction by the side chain of Trp 208 (Trp 194 ), which is also highly conserved (20,37,38). This pattern of anchoring a glycine-containing hinge by the binding of a neighboring aromatic side chain to the bulk of a protein has been observed in other protein systems where structural tran- FIG. 4. Hinge regions in the S 3 R transition in PAI-2. The structural changes that occur when PAI-2 goes from the stressed to the relaxed state are measured by examining changes in the Ramachandran plot. These are quantified by measuring a "distance" traveled in Ramachandran space by a residue as it goes from the stressed to the relaxed state structure. A, a plot of Ramachandran distance for all residues in the stressed and relaxed state PAI-2 structures. Six large changes are observed in the vicinity of residues: Ala 135 , Ser 151 , Glu 187 , Gly 260 , Lys 299 , and Asn 347 . B, a Ramachandran diagram plotting thetransitions for residues undergoing a large torsional change (Ramachandran distance Ͼ 50°). The direction of the arrow indicates the transition from the stressed to the relaxed state conformation. C, a stereogram overlaying the backbone structure of the stressed state (red) and the relaxed state (green) of PAI-2. All residues that undergo a change in Ramachandran space greater than 30°are colored blue on the relaxed state structure. sitions occur, including the zymogen-to-active enzyme transition in serine proteases (39) and the open-to-closed state transition in the enzyme, rubisco (40).
The closure of the cavity beneath ␤-sheet A is likely to contribute to the enhanced stability of the relaxed state compared with the stressed state. In the stressed state, ␤-strands s1A, s2A, and s3A and helix hF appear to form an independent unit with its own hydrophobic core, which is separate from the main molecule. The closure of the cavity results in the expulsion of ordered, buried water molecules and the loss of side chain flexibility and disorder beneath ␤-sheet A. This movement of ␤-sheet A is in part due to the formation of a kink in ␤-strands s2A and s3A on transition to the relaxed state. The Ramachandran analysis identified two adjacent residues in ␤-strands s2A and s3A that are at the apex of the observed kink: Asn 127 (Asn 116 ) and Asn 200 (Asn 186 ).
Asn 200 (Asn 186 ) is critical to the shutter region of the protein. This buried polar region undergoes a structural rearrangement during the S 3 R transition (20). The altered hydrogen bonding pattern beneath ␤-sheet A, involving Asn 200 , is likely to be the cause of the closer interaction between the sheet and the un-derlying protein. Thus, the structure of the shutter region may be the cause of the closure of the cavity during the S 3 R transition and, hence, the enhanced stability of the relaxed state.
One of the unexpected findings upon determining the structure of the PAI-2⅐RCL peptide complex was the highly ordered nature of the P1 Arg binding site. The site is identically situated to that observed in other relaxed state serpin structures; however, in most of these structures, the P1 residue has a nonpolar side chain, and hence it does not appear to be rigidly held in position at the base of the serpin.
Given the structural order of the RCL peptide, we have docked the uPA structure (41) to the base of PAI-2 using only the coordinates of the RCL peptide residues P3-P1 and the uPA peptidic inhibitor. The resulting model of the complex differed from the recently reported structure of the ␣ 1 -antitrypsin-trypsin complex only in the torsional orientation of the protease attached to the inserted RCL (4). It did show severe steric overlap between PAI-2 and residues 94 -101, 161-194, and 213-231 in uPA. These regions include key portions of the activation domain of uPA that are disordered in the trypsin- The interactions between the hinge segments at the base of PAI-2 in the stressed state and relaxed state structures are compared. A, the stressed state structure shows that residues in the hinge region between strand s2A and helix hE (Ser 135 ) interact directly with residues on the adjacent hinge region (Ser 341 , Met 343 , and Ser 344 ), while the P1 site (between Asp 348 and Lys 335 ) is empty. B, an identical view of the relaxed state structure showing that the interactions between Ser 136 and Ser 334 have been broken. Strand s1A is now separated from the adjacent hinge region, and the intervening gap is filled by water molecules. Structural changes in the region are coupled to P1 Arg filling the P1 binding site, which displaces Asp 348 , which now interacts with Ser 344 . ␣ 1 -antitrypsin complex (4). Limited proteolysis studies of the PAI-1⅐uPA complex demonstrate that many of these regions do become disordered upon complex formation (5). We conclude that the specific, well ordered P1 binding site on the base of PAI-2 may be important for the location and orientation of the target protease during inhibition. This site may contribute to the specificity of the serpin for its target.