Intrinsic Specificity of the Reactive Site Loop of α1-Antitrypsin, α1-Antichymotrypsin, Antithrombin III, and Protease Nexin I*

Members of the serpin (serine protease inhibitor) family share a similar backbone structure but expose a variable reactive-site loop, which binds to the catalytic groove of the target protease. Specificity originates in part from the sequence of this loop and also from secondary binding sites that contribute to the inhibitor function. To clarify the intrinsic contribution of the reactive-site loop, α1-antichymotrypsin has been utilized as a scaffold to construct chimeras carrying the loop of antithrombin III, protease nexin 1, or α1-antitrypsin. Reactive-site loops not only vary in sequence but also in length; therefore, the length of the reactive-site loop was also varied in the chimeras. The efficacy of the specificity transfer was evaluated by measuring the stoichiometry of the reaction, the ability to form an SDS-stable complex, and the association rate constant with a number of potential targets (chymotrypsin, neutrophil elastase, trypsin, thrombin, factor Xa, activated protein C, and urokinase). Overall, substitution of a reactive-site loop was not sufficient to transfer the specificity of a given serpin to α1-antichymotrypsin. Specificity of the chimera partly matched that of the loop donor and partly that of the acceptor, whereas the behavior as an inhibitor or a substrate depended upon the targeted protease. Results suggest that, aside from the contributions of the loop sequence and the framework-specific secondary binding sites, an intramolecular control may be essential for productive interaction.

Serpins 1 are protease inhibitors that are involved in the regulation of numerous protease cascades (e.g. blood coagula-tion and complement activation); they consist of 400 -450 amino acids organized into three ␤-sheets and eight or nine ␣-helices connected by surface loops (1). The connection between the A and C ␤-sheets constitutes the reactive-site loop. Although the precise inhibitory mechanism of serpins remains to be elucidated, all proposed models involve an initial interaction between the reactive-site loop of the serpin and the catalytic groove of the protease. Within the reactive-site loop, a P 1 residue 2 plays a crucial role in determining serpin specificity (2,3). However, many serpins have an arginine for P 1 residue, yet inhibit different targets; thus other residues modulate the inhibitor function (4 -13).
To document further the role of the reactive site loop in controlling the specificity, we have prepared and characterized various chimeras, using ACT as a framework to carry the loop of other serpins with overlapping specificities. Results suggest that, in addition to the sequence of the reactive site loop, the specificity of serpins originates from secondary binding sites and conformational constraints.
ACT Variants-Mutants of ACT were produced and purified essentially according to Rubin et al. (5,28). Reactive-site loops were exchanged by cassette mutagenesis using the expression vector pACT * This work was supported by Medical Research Council of the UK and by British Heart Foundation. 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.  1 The abbreviations used are: serpin, serine protease inhibitor; ACT, recombinant ␣ 1 -antichymotrypsin; ACT (P1ϭArg) , ACT where the P 1 residue (Leu 358 ) has been replaced by an arginine; ACT (P1ϭArg; P2ϭGly) , ACT (P1ϭArg; P2ϭPro) , and ACT (P1ϭArg; P2ϭAla) , ACT mutant having an arginine for P 1 residue and a glycine for P 2 residue (proline and alanine for P 2 residue, respectively); ACT (des-TIVR) , ACT where residues Thr 366 -Arg 369 (positions P 8 Ј to P 11 Ј of the reactive-site loop) have been deleted; ACT/ATIII, chimera in which the ACT framework carries the reactivesite loop of antithrombin III; ACT/␣AT, chimera carrying the reactivesite loop of ␣ 1 -antitrypsin; ACT/PN1 chimera carrying the reactive-site loop of protease nexin 1; elastase, human leukocyte elastase; SI, stoichiometry of inhibition; pNA, p-nitroanilide. generously given by Dr. H. Rubin (University of Pennsylvania, Philadelphia, PA), in which two restriction sites (KpnI and MluI) had been engineered within the ACT coding sequence. The engineered KpnI site, between the codon for the P 9 and P 10 residues, results in the replacement of the alanines of wild-type ACT with glycine and threonine; the MluI site, between the codons for the P 10 Ј and P 11 Ј residues, exchanges the P 10 Ј valine to threonine (28 -29). To express chimeras ACT/ATIII, ACT/␣AT, ACT/␣AT (P1ϭArg) and ACT/PN1 (Table I) (31). Chimera ACT (des-TIVR) , in which the reactive-site loop of ACT was shortened by four residues (Thr 366 , Ile 367 , Val 368 , and Arg 369 ), was prepared by removing the corresponding codons from pACT and reconstruction of an MluI site between the codons for the P 6 Ј and P 7 Ј residues. Briefly, pACT was used as a template, with oligonucleotides 5Ј-GAC-CCC-CAA-GAT-ACT-CAT-CAG-3Ј and 3Ј-A-AAC-CAA-CTT-TGC-GCA-TGG-TAG-CAT-GC-5Ј as forward/reverse primers in a first polymerase chain reaction, and oligonucleotides 5Ј-T-TTG-GTT-GAA-ACG-CGT-ACC-ATC-GTA-CG-3Ј and 3Ј-CGG-AAG-TTG-GGT-CAG-TCG-AGG-AAG-5Ј as forward/reverse primers in a second polymerase chain reaction. The resulting fragments (517 and 624 base pairs) were digested with MluI, ligated with T4 DNA ligase, and utilized as template for a third amplification, using as primers the forward of the first and the reverse of the second polymerase chain reactions. The resulting 1124-base pair fragment was cloned into pCRII vector (Invitrogen, Abingdon, UK), according to the manufacturer's instruction. The pACT (des-TIVR) expression vector with a "short" reactive-site loop was prepared by exchanging the KpnI/SphI fragment of pACT for the corresponding 618-base pair KpnI/SphI fragment of pCRII. The resulting expression vector pACT (des-TIVR) was used for preparation of ACT (des-TIVR) and to construct, by cassette mutagenesis, the expression vectors for ACT/␣AT (des-TIVR) , ACT/␣AT (P1ϭArg; des-TIVR) , and ACT/ PN1 (des-TIVR) . Recombinants ACT variants were purified in mg quantities by a combination of anion-exchange and DNA-cellulose chromatography, as described previously (30). Final products appeared homogeneous by Coomassie Blue staining, following SDS-polyacrylamide gel electrophoresis.
Amidolytic Assays-Kinetics were performed in 50 mM Tris-HCl, pH 7.8, containing 0.1 M NaCl, 0.2% poly(ethylene glycol) M r 6000, and 1 mg/ml bovine serum albumin (protease-free, Sigma). Assays with activated protein C contained in addition 5 mM CaCl 2 . The chromogenic substrates S-2222 (benzyl-CO-Ile-Glu-(␥-OR)-Gly-Arg-pNA, where R is H or CH 3 , and pNA indicates p-nitroanilide) was used with factor Xa (K m ϭ 547 M). S-2238 (H-D-Phe-Pip-Arg-pNA) (Pip indicates L-pipecolyl) was used with thrombin and activated protein C (K m ϭ 3.6 and 315 M, respectively), as well as S-2266 (H-D-Val-Leu-Arg-pNA; K m ϭ 486 and 221 M, respectively). S-2288 (H-D-Ile-Pro-Arg-pNA) and S-2302 (H-D-Pro-Phe-Arg-pNA) were used with trypsin (K m ϭ 19 and 134 M, respectively). S-2444 (ϽGlu-Gly-Arg-pNA) was used with urokinase (K m ϭ 90 M). The above peptidyl-pNA substrates were purchased from Chromogenix (Mölndal, Sweden). Succinyl-Ala-Ala-Pro-Phe-pNA was used with chymotrypsin (K m ϭ 53 M), and N-methoxysuccinyl-Ala-Ala-Pro-Val-pNA was used with elastase (K m ϭ 121 M); both were purchased from Sigma. Substrate concentrations were determined from their absorbance at 342 nm using a molar extinction coefficient of 8270 cm Ϫ1 (32). Choice of the pNA substrate depended whether competition between hydrolysis and inhibition was desired; substrates with low K m values were used to increase the apparent half-life of complex formation, and those with high K m values were used to minimize competition.
Characterization of the Chimera Specificity-Stoichiometry of inhibition (SI, i.e. the number of moles of serpin required to inhibit 1 mol of protease) was determined as described previously (30), using chymotrypsin and elastase for the chimeras having a P 1 leucine or methionine and using trypsin and thrombin for the chimeras having a P 1 arginine. The ability of the chimeras to form SDS-stable complexes was examined by incubating the variant (2 M) with the protease (1 M) in 50 mM Tris-HCl buffer, pH 7.8, containing 100 mM NaCl, and 0.2% (w/v) poly(ethylene glycol) M r 6000 for 30 min at room temperature. The sample was denatured at 65°C for 10 min in 0.37 M Tris-HCl, pH 8.8, containing 1% SDS (w/v), 10% glycerol (v/v), and 5% ␤-mercaptoethanol (v/v) and analyzed by SDS-polyacrylamide gel electrophoresis (gradient 10 -20% acrylamide). The overall association rate constant for the formation of a protease-serpin complex (k on ) was estimated by analysis of data from progress curve kinetics completed with a large excess of serpin (6 concentrations, between 0.32 and 10 M) over the enzyme (a single concentration between 0.01 and 1 nM, depending upon the protease). Inhibition reactions were followed for up to 3 h using a Hewlett-Packard diode array spectrophotometer, but only data corresponding to less than 10% substrate hydrolysis were analyzed. Reactions were initiated by addition of the enzyme at a concentration such that the velocity in the absence of inhibitor was about 0.2 M min Ϫ1 with the appropriate peptidyl-substrate (100 M). Data were analyzed according to the equation for slow binding inhibition to yield estimates for k on as described previously (30).

RESULTS
We have constructed ACT chimeras in which the length of the reactive-site loop corresponds to that of either the donor or acceptor (ACT/␣AT and ACT/PN1 carrying long reactive-site loop, as well as ACT/␣AT (des-TIVR) and ACT/PN1 (des-TIVR) carrying shorter loop); because the reactive-site loop of antithrombin III is only one amino acid shorter than that of ACT, only the ACT/ATIII chimera was constructed ( Table I). The selectivity TABLE I Nomenclature of the ACT mutants, sequence of their reactive-site loop, and stoichiometry of inhibition (SI) Comparison with ACT of the P10 -P13Ј residues of the various serpins and chimeras used in this study. Amino acids that differ from ACT are underlined; dot indicates that the corresponding residue is absent. SI value represents the number of moles of serpin required to inhibit 1 mol of enzyme; its estimation was not reliable (ND) for elastase inhibition by ACT (des-TIVR) . SI values of the ACT double mutants (P 1 arginine and either P 2 proline, alanine, or glycine) were taken from Djie et al. (30).

ACT variant
Reactive-site loop sequence SI and effectiveness of each chimera toward a number of possible targets were evaluated using three criteria: SI value, ability to form an SDS-stable complex, and k on value. Chimeras with proline in P 2 and arginine in P 1 exhibited SI values higher than 150 with trypsin and lower than 10 with thrombin (Table I). This is consistent with our observation (30) that replacement of the P 2 leucine of ACT (P1ϭArg) with proline causes a dramatic increase of the SI value with trypsin but not with thrombin. When P 1 was either methionine or leucine, SI values were higher than 60 with elastase and lower than 3 with chymotrypsin (regardless of the P 2 residue). When the loop of the ACT/␣AT chimeras was shorter, SI values were higher for chymotrypsin, trypsin, and thrombin inhibition; the opposite was true for elastase. In contrast, the length of loop did not change the SI value with the ACT/PN1 chimeras. Thus SI value depended upon the sequence and the length of the reactive-site loop, as well as upon the target considered.
Attempts to detect SDS-stable complex formation followed incubation of a slight excess of serpin with M quantities of the potential target (i.e. in conditions where cleavage reaction occurs substantially). Three patterns were observed (Fig. 1) as follows: 1) the formation of SDS-stable complex, but the presence of intermediate size fragments attributable to their degradation by remaining (active) protease (38); 2) absence of detectable complex, but accumulation of material migrating as ACT having the reactive-site loop cleaved; and 3) absence of detectable reaction (neither complex formation nor cleavage). Chimeras with leucine or methionine in P 1 position all formed an SDS-stable complex with chymotrypsin (pattern 1), whereas consistent with the high SI value, chimeras were mainly cleaved following incubation with elastase (pattern 2). Chimeras having arginine in P 1 position exhibited pattern 1 with thrombin, whereas the pattern was either 1 or 2 following incubation with trypsin (again, consistent with the high SI value, pattern was 2 with proline in P 2 , and pattern 1 was otherwise). Regardless of the chimera, there was no clear evidence of SDS-stable complex formation nor of cleavage reaction with factor Xa, activated protein C, and urokinase; the predominant bands were those of the intact proteins (pattern 3). Finally, ACT/ATIII exhibited a mixed pattern. Thus, with the exception of trypsin, the pattern was specific for the target rather than for the chimera.
Under the conditions of the slow binding assays (low enzyme concentration and very large excess of serpin), depletion of the inhibitor by the cleavage reaction becomes negligible, as the total amount of cleaved inhibitor depends on the absolute amount of enzyme (30, 34 -35). The k on values obtained (Table  II) did not reveal a clear relationship between the nature of the reactive-site loop and the specificity or function of the chimera, but in two instances only, the k on values were higher when length of the reactive-site loop was shorter than that of ACT.
Reactive-Site Loop of Antithrombin III-Substitution of the reactive-site loop of ACT with that of antithrombin III transposed quite effectively the specificity of antithrombin III to ACT (Table II). ACT/ATIII exhibited k on values within 10-fold those of antithrombin III for trypsin, thrombin, and factor Xa inhibition, whereas k on values were lower than 10 M Ϫ1 s Ϫ1 with activated protein C and urokinase. Thus, ACT/ATIII exhibited a selectivity resembling that of antithrombin III. However, this apparent success must be tempered by the observation that simple substitution of the P 1 leucine with arginine largely accounts for the inhibitory properties of ACT/ATIII with trypsin or thrombin, whereas further substitution of the P 2 leucine with glycine is sufficient to mimic antithrombin III with every protease (30). Thus, substitution of the whole reactive-site loop did not alter the selectivity of the ACT mutant having the P 1 and P 2 residues of antithrombin III.
Reactive-Site Loop of Protease Nexin 1-Except for factor Xa inhibition, introducing the reactive-site loop of protease nexin 1 into ACT did not result in an appropriate transfer of specificity (Table II). Failure was most evident with urokinase; k on values were at least 4 orders of magnitude lower than with protease nexin 1. In fact, regardless of the reactive-site loop, none of the ACT variants neutralized urokinase. The lack of urokinase inhibition was due to an absence of reaction rather than to the ACT variants acting as substrates; no cleavage reaction was detected by polyacrylamide gel electrophoresis. The k on values for thrombin inhibition were also dramatically lower than with the loop donor (875-and 5000-fold with ACT/PN1 and ACT/ PN1 (des-TIVR) , respectively), and activated protein C inhibition was hardly detectable. However, the reactive-site loop of protease nexin 1 was functional in the context of the ACT framework. Less than 5-fold separated the k on values of factor Xa inhibition by protease nexin 1 and ACT/PN1 and less than 27-fold separated those for trypsin inhibition. Factor Xa was also a remarkable exception, because shortening the loop increased the inhibitory activity of the chimera. Nevertheless, as was observed with the antithrombin III reactive-site loop, the effectiveness of transfer could be attributed to a predominant role of the P 1 and P 2 residues. Replacement of the P 2 leucine of ACT (P1ϭArg) with alanine essentially mimics the behavior of ACT/PN1 (30).
Reactive-site Loop of ␣ 1 -Antitrypsin-␣ 1 -Antitrypsin with a methionine in P 1 position inhibits elastase much more efficiently than ACT (k on value 10 3 -fold higher) and chymotrypsin 14-fold more efficiently (36,37). Yet the k on values obtained with the ACT/␣AT chimera were 2-to 3-fold lower than those with ACT, and truncation of the loop further decreased the k on values 2-to 4-fold (Table III). Thus, grafting the ␣ 1 -antitrypsin loop into ACT not only failed to reproduce the specificity of ␣ 1 -antitrypsin, but it was rather detrimental for the inhibitory activity of ACT toward chymotrypsin and elastase.
Behavior of the ACT chimera carrying the loop of ␣ 1antitrypsin (P1ϭArg) was somewhat erratic (Table II). Single replacement for proline of the P 2 leucine of ACT (P1ϭArg) does not change the k on value for trypsin inhibition (30); replacement of the entire loop increased this value 6-fold. In contrast, the P 2 mutation improves the k on values over 18-fold for thrombin and factor Xa inhibition (30), whereas substitution of the entire loop improved marginally (less than 5-fold) the k on values. Behavior with activated protein C was again different; replacement of the P 2 residue is neutral (30), but substitution of the entire loop was detrimental (4-fold decreased of the k on value). Thus, in contrast to the antithrombin III and protease nexin 1 loop transfers, replacement of the P 2 residue did not mimic the effect of swapping the entire loop with ␣ 1 -antitrypsin. In particular, three of the chimeras were less effective than the ACT mutant having the same P 1 and P 2 residues, suggesting that one or more detrimental elements outweighed the benefit of the P 1 and P 2 substitutions. The length of the loop was particularly critical for trypsin inhibition: ACT/␣AT (P1ϭArg; des-TIVR) inhibited trypsin with a k on value 80-fold lower than that of ACT/ ␣AT (P1ϭArg) , whereas the k on values decreased 6-fold at the most with thrombin and factor Xa.

DISCUSSION
The reactive-site loop of serpins undoubtedly defines in part their ability to inhibit a particular protease (6, 30, 39 -44). Therefore, we anticipated that exchanging loops between serpins might reassign the targets. Although swapping of reactive-site loops dramatically modified the specificity, in general it was not possible to transfer the inhibitory properties of a loop donor to ACT. Overall, the specificity of the chimera rarely matched that of the loop donor, indicating that other players must participate (directly or indirectly) in the specificity of serpins. In addition to residues P 3 to P 5 Ј, at least three factors could conceivably influence the serpin behavior: (i) length of the reactive-site loop, (ii) secondary binding sites remote from the reactive-site loop, and (iii) intramolecular control of the reactive-site loop conformation. Failure to transfer a given specificity to ACT would result from an impaired mechanism of action due to one or more of these factors.
Length of the Reactive-site Loop-There was no clear rela-tionship between length of the reactive-site loop and specificity, but as a general rule, ACT chimeras exhibited higher k on value with the longer loop. Thus our data favor the hypothesis that the length of the loop is important for the mechanism of inhibition within a given framework, rather than in governing the specificity per se. This is consistent with the observation that specificity does not correlate with length of the reactive-site loop in natural serpins. ACT and ␣1-antitrypsin both inhibit chymotrypsin, even though the length of their loops differs by 4 amino acids. Similarly, the loops of ␣1-antitrypsin and antithrombin III differ by 3 amino acids, but both inhibit trypsin. Also consistent with the length of the loop being important for the serpin mechanism of action rather than its specificity, Avron et al. (45) have shown that insertion of an alanine in P 5 Ј position decreases the k on value of ␣1-antitrypsin for elastase 10-fold and for trypsin nearly 100-fold. However, two exceptions prohibit generalization and suggest that deletion of 4 amino acids from the reactive-site loop of ACT did not hamper its folding nor its mechanism of action. Elastase was inactivated more rapidly by ACT (des-TIVR) than by ACT, and compared with ACT/PN1, factor Xa was inactivated more rapidly by ACT/PN1 (des-TIVR) . Thus, the length of the reactive-site loop appears important in the context of the serpin framework but alone does not account for the overall failure in transposing specificity. Secondary Binding Sites-Several serpins possess a secondary binding site, often remote from the reactive-site loop, which promotes interaction with the target. The binding site interacts with a complementary motif on the protease, often distant from the catalytic groove. The specificity would be controlled by the simultaneous presence of these motifs on the inhibitor and its target. Swapping frameworks would shuffle the secondary motifs, rendering the chimeras' behavior relatively unpredictable. For example, plasminogen activator inhibitor 1 uses its acidic residues P 4 Ј, P 5 Ј, and P 9 Ј to interact with a positively charged surface-loop (amino acids 35 to 41) 3 of tissue plasminogen activator (48 -50), and heparin cofactor II uses its negatively charged N-terminal extension to interact with anion-binding exosite 1 of thrombin (46,47). Likewise, antithrombin III uses an unknown motif to interact with the 60-loop insertion of thrombin (44), and protease nexin 1 uses an unknown motif to interact with urokinase (51). Precisely, regardless of the reactive-site loop sequence, none of our ACT variants inhibited urokinase. Thus, our results would be consistent with the hy-TABLE II k on values for various proteases of the chimeras having a P 1 arginine Each slow-binding inhibition experiment was performed at least twice; the k on value given represents the weighted mean of these determinations. The standard errors of the weighted means were 5% or less of the mean value. Abbreviations used are: FXa, factor Xa; APC, activated protein C; ND. not determined. Values of the ACT double mutants (P 1 arginine and either P 2 proline, alanine or glycine) were taken from Djie et al. (30).   pothesis that for urokinase inhibition a critical area of protease nexin 1 had not been implanted in the ACT framework. Alternatively, a motif in ACT could interact favorably with elastase, chymotrypsin, and perhaps factor Xa, while being detrimental for activated protein C and urokinase binding. In this regard, it is interesting to note that ACT binds double-stranded DNA, through a stretch of 3 lysines (residues 212-214), a property not shared by ␣1-antitrypsin, antithrombin III, or protease nexin 1 (15,52). From x-ray structure data, this DNA binding site must be close to the 35-41 surface loop of the target protease, at least during formation of the initial complex. The 35-41 surface loop of factor Xa is negatively charged, those of urokinase and activated protein C are positively charged, and those of chymotrypsin, elastase, trypsin, and thrombin are rather neutral. Thus, a putative interaction between the positively charged DNA-binding site and the 35-41 loop would favor interaction with factor Xa, while impeding binding of urokinase and activated protein C. This hypothesis would be consistent with our observation that all ACT chimeras reacted more rapidly with factor Xa than with activated protein C and urokinase. Intramolecular Control-In several instances, substitution of the entire loop was more detrimental than simple substitution of the corresponding P1 and P2 residues. This observation suggests that the specificity is also restrained by intramolecular interactions. Residues within the core of the inhibitor would control the specificity by ensuring proper folding and/or function of the reactive-site loop. Important intramolecular interactions between the reactive-site loop and the core of the molecule have been characterized in several canonical inhibitors (3,53). In Kunitz-type 1 inhibitors, for instance, the conformation of the reactive-site loop is firmly maintained by a disulfide bridge between Cys 14 (P 2 position) and Cys 38 in the core of the inhibitor (54). This covalent link allows the reactive-site loop of Kunitz inhibitors to withstand extensive substitution in most positions (other than P 2 ), with the mutant specificity following the preferences of the protease target (55,56). In eglin c, a hydrogen bonding network stabilizes the conformation of the P 1 -P 1 Ј bond (57); disruption of this network allows greater mobility of the reactive-site loop, resulting in eglin c becoming a substrate of elastase (58). Consistent with the hypothesis that serpin specificity relies on intramolecular control, the reactive-site loops of ACT and ␣1-antitrypsin are both suitable for chymotrypsin inactivation but are more efficient in the context of their original framework. The crystal structure of the ACT/␣AT chimera reveals that only a few contacts occur between the reactive-site loop and the underlying protein (16), whereas the same P 3 -P 3 Ј sequence in the context of ␣1-antitrypsin has developed many more contacts and packs close to the main body of the serpin (59 -60). The loose reactive-site loop of ACT/␣AT inactivates chymotrypsin 55-fold less efficiently than ACT (28); the same loop tightly packed in ␣1antitrypsin is 328 times more effective.
Thus, loss of an intramolecular control would explain that several specificity swaps failed, even though donor and acceptor have a common target. Yet, at least in one instance loop swapping seems feasible without constraint, because this is the natural process to generate multiple serpins from a single gene in Mandula sexta (61)(62). Using a unique N-terminal framework, the insect generates 12 serpins by mutually exclusive exon shuffling. Among the serpins produced, the reactive-site loop varies in sequence as well as in length (by as many as 6 residues). Perhaps a key difference between the chimeras described in this paper and those of M. sexta resides in the C-terminal portion of the molecule. Both the reactive-site loop and the C-terminal portions are systematically exchanged in the insect serpins, whereas in our ACT chimeras, the C-terminal portion remained that of ACT. It is possible that the Cterminal region (strands 4 and 5 of the B ␤-sheet) plays an unidentified role in serpin function.