Residues Tyr253 and Glu255 in Strand 3 of β-Sheet C of Antithrombin Are Key Determinants of an Exosite Made Accessible by Heparin Activation to Promote Rapid Inhibition of Factors Xa and IXa*

We previously showed that conformational activation of the anticoagulant serpin, antithrombin, by heparin generates new exosites in strand 3 of β-sheet C, which promote the reaction of the inhibitor with the target proteases, factor Xa and factor IXa. To determine which residues comprise the exosites, we mutated strand 3C residues that are conserved in all vertebrate antithrombins. Combined mutations of the three conserved surface-accessible residues, Tyr253,Glu255, and Lys257, or of just Tyr253 and Glu255, but not any of these residues alone, was sufficient to reproduce the exosite defects of a strand 3C antithrombin-α1-proteinase inhibitor chimera in reactions of the heparin-activated variants with both factor Xa and factor IXa. Importantly, the exosite-defective antithrombins bound heparin with nearly wild-type affinities, and the heparin-activated mutants showed near normal reactivities with thrombin, a protease that does not utilize the exosite. Mutation of the conserved but partially buried strand 3C residue, Gln254, the reactive loop P6′ residue, Arg399, which interacts with Glu255, or a residue proposed to constitute the exosite from modeling studies, Glu237, all produced minimal effects on antithrombin reactivity with thrombin, factor Xa, and factor IXa in the absence or presence of heparin. Together, these results indicate that Tyr253 and Glu255 are key exosite determinants responsible for promoting the reactions of conformationally activated antithrombin with both factor Xa and factor IXa.

Antithrombin, a member of the serpin superfamily of protein protease inhibitors, functions as a key anticoagulant regulator of blood clotting proteases in vertebrates (1). It inhibits its main target clotting proteases, thrombin, factor Xa, and factor IXa (2)(3)(4), by forming highly stable equimolar complexes with the enzymes through a novel conformational trapping mechanism (5,6). In this mechanism, the protease recognizes and binds an exposed reactive loop of the serpin and proceeds to cleave a reactive bond in the loop to generate a covalent serpinprotease acyl-intermediate as in a regular serine protease substrate reaction. However, major conformational changes of both the serpin and protease are induced at this stage that arrest further cleavage of the bond and stabilize the acyl-intermediate complex. In these changes, the N-terminal end of the cleaved serpin-reactive loop inserts into the major ␤-sheet of the serpin core, the A sheet, dragging the acyl-linked protease along with it to the opposite end of the serpin, where steric forces distort and inactivate the protease catalytic machinery (6 -9).
Antithrombin is unusual among serpins in that its reactions with its target clotting proteases are unusually slow and require the sulfated polysaccharide, heparin, to accelerate them to the physiologically significant diffusion limit. The acceleration of antithrombin-protease reactions by heparin results from the polysaccharide binding to the serpin through a unique pentasaccharide sequence and inducing an activating conformational change in the protein (2,10,11). The conformational change greatly increases the affinity of antithrombin for heparin and disrupts an intramolecular reactive loop-sheet A interaction to allow full exposure of the loop in a manner like that of other serpins (12)(13)(14).
One puzzling aspect of the function of antithrombin has been its ability to target three clotting proteases with very different substrate specificities. Whereas it was initially thought that the determinants of the antithrombin specificity for thrombin, factor Xa and factor IXa, were encoded in the reactive loop sequence and that heparin activation served to make this sequence accessible to protease, later studies proved that this was not the case. Mutagenesis of the reactive loop sequence, particularly the conserved P2-P1Ј residues, was thus shown to cause marked changes in the reactivity of antithrombin with proteases in the absence of heparin but to have little effect on the heparin enhancement of this reactivity (3,15,16). Such findings suggested that the reactive loop sequence only provided the determinants of the slow reactivity of antithrombin with its three target proteases in the absence of heparin and that the heparin enhancement of this reactivity resulted from the generation of new exosites that promoted the interaction of the three target proteases with the inhibitor (17). These exosites were made accessible on antithrombin upon conformational activation by the heparin pentasaccharide in the case of factors Xa and IXa, whereas additional exosites on heparin outside the pentasaccharide acted to further promote the antithrombin-protease interaction by bridging the two proteins in a ternary complex for all three target proteases (3, 18 -20).
To prove the existence of the hypothesized exosites on antithrombin, which interact with factors Xa and IXa, we used a chimeric approach in which linear regions of antithrombin circumscribing the reactive loop were substituted with the homologous regions of ␣ 1 -proteinase inhibitor, a serpin not activated by heparin (21). Of six chimeras prepared, one was shown to be defective in the rate enhancement produced by heparin pentasaccharide of the antithrombin reaction with both factors Xa and IXa. These results thus suggested that an exosite, specific for both factors Xa and IXa, resides in strand 3 of ␤-sheet C just below the serpin reactive loop. The aim of the present study was to further localize the exosite to specific amino acid residues in strand 3C. Our results demonstrate that two residues in this strand, Tyr 253 and Glu 255 , which are completely conserved in the 13 vertebrate antithrombins that have been sequenced (22), are critical determi-nants of this exosite and responsible for mediating the rapid inhibition of factors Xa and IXa by heparin-activated antithrombin.

Construction, Expression, and Purification of Wild Type and Variant
Antithrombins-Recombinant antithrombin variants were constructed on an N135Q background to block glycosylation of Asn 135 and thereby prevent the heparin binding heterogeneity that results from incomplete glycosylation at this site (1). Variants were produced in baculovirusinfected insect cells using the expression system from Invitrogen as in previous studies (21). DNA mutagenesis was carried out by PCR employing the Quick Exchange kit from Invitrogen and utilizing custom made oligonucleotides ordered from Sigma Operon. All mutations were confirmed by DNA sequencing. Recombinant antithrombins were purified on a 5-ml Hi-Trap heparin-Sepharose column followed by ion exchange chromatography on monoQ as previously described (21,23). Recombinant antithrombin concentrations were measured from the absorbance at 280 nm using an extinction coefficient of 37,700 M Ϫ1 cm Ϫ1 (24). Corrections were made for small amounts of inactive, presumably latent, antithrombin in some variants based on the results of stoichiometric titrations of thrombin with the variants (see below).
Proteases and Heparin-Human ␣-thrombin was prepared by activating purified prothrombin (25) with a snake venom activator followed by purification of the active protease as described (26). Human factor Xa (a mixture of ␣ and ␤ forms) and factor IXa␤ were obtained from Enzyme Research Laboratories (South Bend, IN). A synthetic heparin pentasaccharide corresponding to the antithrombin binding sequence in heparin was provided by Dr. Maurice Petitou (Sanofi Recherche, Toulouse, France). A full-length heparin with reduced polydispersity, having a chain-length of about 50 saccharides and containing the pentasaccharide binding sequence, was prepared from commercial heparin by size exclusion and affinity chromatography on antithrombin-agarose as described (27). Concentrations of heparins were obtained by stoichiometric binding titrations of plasma antithrombin with the polysaccharides monitored from the 35-40% enhancement in tryptophan fluorescence, which reports binding as in previous studies (27). Stoichiometric binding conditions were achieved by using antithrombin concentrations at least 10 times the K D for the interaction. Human antithrombin was purified from blood plasma as in previous studies (27).
SDS-PAGE-The purity of recombinant antithrombins and their ability to form SDS-stable complexes with proteases was evaluated by SDS-PAGE using 10% acrylamide gels and the Laemmli discontinuous buffer system (28).
Experimental Conditions-Antithrombin-protease reaction stoichiometries and kinetics were done at 25°C in either 20 mM sodium phosphate, 0.1 M NaCl, 0.1 mM EDTA, 0.1% polyethylene glycol 8000, pH 7.4, buffer for reactions with thrombin and factor Xa or in 100 mM Hepes, 0.095 M NaCl, 5 mM CaCl 2 , 0.1% polyethylene glycol 8000, pH 7.4, buffer for reactions with factor IXa. The ionic strength of both buffers was 0.15.
Heparin Binding to Antithrombin Variants-Equilibrium binding of the heparin pentasaccharide to recombinant antithrombins was evaluated by titrating antithrombin, at a concentration close to the K D for the binding interaction, with the pentasaccharide and monitoring the tryptophan fluorescence enhancement, which reports binding as in prior studies (2). Because of the tight binding of recombinant antithrombin to heparin at physiologic ionic strength, binding studies were conducted in the phosphate buffer used for studying antithrombin-protease reactions but at a higher ionic strength (I ϭ 0.35, attained by adjusting the NaCl concentration to 0.3 M) to allow more accurate measurements of K D . Fluorescence titrations were computer-fit by the quadratic equilibrium binding equation assuming a 1:1 binding stoichiometry to determine values for K D and the maximal fluorescence change (2).
Stoichiometry of Antithrombin-Protease Reactions-The stoichiometries for the reactions of recombinant antithrombins with thrombin and factor Xa were determined as described previously (27). Briefly, to fixed concentrations of 100 nM protease was added increasing concentrations of inhibitor to give molar ratios of inhibitor to protease of up to 1.6 in a final volume of 100 l. Reactions were done both in the absence and presence of the full-length heparin, which was fixed at a concentration equimolar with the protease. After incubating for times sufficient to complete the reaction (Ͼ95%), 4 l of the reaction mixture was added to 1 ml of substrate (100 M S-2238 for thrombin or 100 M Spectrozyme FXa for factor Xa), and the residual enzymatic activity was measured from the initial linear rate of change of absorbance at 405 nm. The decrease in protease activity with increasing molar ratio of inhibitor/ protease was fit by linear regression to obtain the stoichiometry from the abscissa intercept.
Kinetics of Antithrombin-Protease Reactions-Association rate constants for reactions of recombinant antithrombins with proteases in the absence or presence of pentasaccharide or full-length heparins were measured under pseudo-first order conditions by using at least a 10-fold molar excess of inhibitor over protease as in previous studies (3,21). For reactions with all proteases in the absence of heparin and those reactions with proteases in the presence of pentasaccharide whose rate constants were 10 4 M Ϫ1 s Ϫ1 or less (or in some cases as high as 10 5 M Ϫ1 s Ϫ1 ), full reaction time courses of the loss of enzyme activity were obtained and fit by a single exponential function with a zero activity end point. Pentasaccharide was present in these full time course experiments at levels that saturated antithrombin. A nonzero end point was used for factor IXa reactions due to a small amount of degraded protease in the preparation (Ͻ5%), which was resistant to being inhibited (3). Assays of residual enzyme activity were done by removing aliquots of the reaction at different times, diluting into appropriate chromogenic or fluorogenic substrates, and measuring the initial rate of substrate hydrolysis as in previous studies (20). Association rate constants for free antithrombin or antithrombin-pentasaccharide complex reactions were obtained from fitted exponential rate constants by dividing by the concentration of antithrombin. For all other reactions in the presence of pentasaccharide or full-length heparins, reactions were done for a fixed time as a function of the concentration of heparin, and the loss in enzyme activity was fit by a single exponential function with heparin concentration instead of time as the independent variable (21). The association rate constant for the reaction of the antithrombin-heparin complex was obtained from the fitted exponential rate constant by dividing by the fixed reaction time and by a factor that corrected for the fraction of heparin that was bound by antithrombin.

Mutagenesis Strategy for Mapping the Putative Antithrombin Exosite-
We previously showed that an exosite that enhances the reactivity of antithrombin with factors Xa and IXa following activation of the inhibitor by a specific heparin pentasaccharide is localized to strand 3 of ␤-sheet C of the serpin just underneath the reactive center loop ( Fig. 1) (21). To further pinpoint which of the eight altered residues in the strand 3C antithrombin-␣ 1 -protease inhibitor chimera was responsible for the observed exosite defect of the chimera, we focused our attention on those residues that are exposed on the protein surface and are highly conserved in the 13 vertebrate antithrombins that have been sequenced (22). Three highly conserved surface residues in strand 3C, Tyr 253 , Glu 255 , and Lys 257 , and a conserved but less surface-accessible residue, Gln 254 , were selected for mutation ( Fig. 1). Additional mutations of another highly conserved residue in the reactive loop, Arg 399 , which appeared to interact with the conserved strand 3C residues in the x-ray structure of heparin-activated antithrombin and a residue implicated by modeling to be part of a factor Xa exosite, Glu 237 (29), were also made ( Fig. 1).
Antithrombin variants were all expressed at a level similar to the wild-type protein. Purification of the variants by heparin-agarose chromatography showed that they all bound and were eluted from the heparin matrix at a salt concentration reasonably close to that of the wildtype inhibitor, indicating wild type-like affinities for heparin. After an additional ionic exchange chromatography step on monoQ, the variants were Ͼ95% pure by SDS-PAGE analysis. All antithrombin mutants were found to inhibit thrombin enzymatic activity and selected exositedefective mutants were shown to inhibit factor Xa activity in the absence or presence of heparin with stoichiometries indistinguishable from the wild-type protein (Table 1 and data not shown). Only in a few cases were thrombin inhibition stoichiometries in the absence of hepa-rin slightly elevated above 1, indicative of a small amount of inactive latent inhibitor in the preparation (23). The elevated stoichiometries observed in the presence of heparin are due to enhancement of a competing substrate reaction of the inhibitor with the protease (30). SDS-PAGE analysis showed that all mutants formed SDS-stable equimolar complexes with thrombin, and selected exosite-defective mutants were observed to form complexes with factor Xa (Fig. 2) and with factor IXa (not shown) indistinguishable from the complexes formed with the wild-type inhibitor. Equilibrium binding of the heparin pentasaccharide activator to the mutant antithrombins, measured from the ϳ40% intrinsic fluorescence increase that reports binding and activation (2), showed that the fluorescence increase and binding affinities for pentasaccharide were not greatly perturbed from that of the wild-type inhibitor ( Table  1). To determine which of the conserved strand 3C antithrombin residues corresponded to the factor Xa and factor IXa-specific exosite, association rate constants for the reactions of wild-type and mutant antithrombins with thrombin, factor Xa, and factor IXa in the absence and presence of pentasaccharide and full-length heparins were determined ( Fig. 3 and Table 2).
Effects of Mutating Conserved Strand 3C Residues on Antithrombin Reactivity with Factor Xa and Thrombin-Mutation of the three conserved surface residues in strand 3C, Tyr 253 , Glu 255 , and Lys 257 , to their counterparts in ␣ 1 -proteinase inhibitor largely duplicated the factor Xa exosite defect previously observed when the entire strand 3C of antithrombin was replaced with the corresponding region of ␣ 1 -proteinase inhibitor (21). Whereas the association rate constant for inhibition of factor Xa by the triple mutant antithrombin was minimally perturbed from the wild-type value in the absence of heparin, the rate constant was markedly reduced to 9% of the wild-type value after activation of the mutant antithrombin by the heparin pentasaccharide, an impairment approaching that observed for the pentasaccharide-activated chimera reaction (2% of the wild-type rate constant). The triple mutant inhibitor showed a somewhat lesser impairment in reactivity upon activation with a full-length heparin, as did the chimera (i.e. association rate constants were reduced to 14 and 3% of the value for the wild-type inhibitor reaction, respectively). The triple mutant also resembled the chimera in showing a ϳ5-fold higher affinity for the heparin pentasaccharide than the wild-type inhibitor (Table 1) and a ϳ20-fold greater reactivity with thrombin in the absence of heparin but a thrombin reactivity closer to that of the wild-type inhibitor in the presence of either pentasaccharide or full-length heparins.
Mutating the three conserved strand 3C residues to alanine again reproduced the factor Xa exosite defect observed with both the chimera and the triple Y253K/E255L/K257M mutant. The reactivity of the triple alanine mutant with factor Xa was thus normal in the absence of heparin but was drastically reduced to 3% that of the wild-type inhibitor following activation by the heparin pentasaccharide. Again, the pentasaccharide activation defect was somewhat overcome when the mutant was activated by a full-length bridging heparin (association rate constant 10% of the value for the similarly activated wild-type inhibitor reaction). The triple alanine mutant also behaved like the chimera and Y253K/ E255L/K257M mutant with respect to its reaction with thrombin in the presence of pentasaccharide or full-length heparins in showing essentially wild-type values for association rate constants. However, it differed from the chimera or chimera-like triple mutant in showing no elevation of the rate constant for the reaction with thrombin in the absence of heparin.
Double mutations of the three conserved strand 3C residues to alanine showed that the effect of the triple mutation could be replicated by mutating just Tyr 253 and Glu 255 . The Y253A/E255A double mutant thus FIGURE 1. Antithrombin residues targeted for mutation within and outside strand 3C. A, ribbon diagram of activated antithrombin (taken from the ternary antithrombinheparin-anhydrothrombin ternary complex structure, Protein Data Bank code 1SR5) highlighting the strand 3C region (green) and the residues within (green) and nearby (red) that were mutated (in a stick representation). For reference, the reactive center loop is shown in yellow with the P1 residue in a stick representation, and the heparin binding helix D is shown in blue. B, space-filling view of the reactive center loop face of activated antithrombin depicting amino acid residues that were mutated in this study. Tyr 253 , Gln 254 , Glu 255 , and Lys 257 residues in strand 3 of ␤-sheet C are shown in green, whereas the P6Ј reactive loop residue, Arg 399 , and Glu 237 in strand 4 of ␤-sheet C are depicted in red (side chain atoms only). The reactive loop is yellow and shown in a stick representation except for the P1 Arg, which is depicted in a space-filling view.
showed a comparable major impairment in the heparin-activated inhibitor reactivity with factor Xa as the triple mutants (i.e. the association rate constant was reduced to 2 and 6% of the wild-type rate constant with inhibitor activation by pentasaccharide and full-length heparins, respectively), although the reactivity in the absence of heparin was also reduced ϳ2-fold with this mutant. The thrombin reactivity of the double mutant was elevated ϳ4-fold in the absence of heparin but approached a wild-type reactivity in the presence of either pentasaccharide or full-length heparin. By contrast, mutating Glu 255 and Lys 257 together to Ala resulted in a much smaller impairment in the association rate constant for the pentasaccharide-activated antithrombin reaction with factor Xa to 41% of the activated wild-type inhibitor rate constant. The reactions of this latter double mutant inhibitor with factor Xa or thrombin in the absence of heparin or the rate enhancements of these reactions due to heparin bridging were minimally affected.
Single mutations of Tyr 253 or Glu 255 produced smaller reductions in antithrombin reactivity with factor Xa than when these residues were mutated together. Mutations of Tyr 253 to Lys or Phe or of Glu 255 to Leu, Arg, or Gln resulted in no impairment or slight (up to ϳ2-fold) impairment of factor Xa reactivity in the absence or presence of pentasaccharide or bridging heparins. Mutation of Glu 255 to Ala produced a some-what larger 3-4-fold reduction in factor Xa reactivity in the presence of pentasaccharide but close to normal reactivity in the absence of heparin and a normal heparin bridging rate enhancement. The largest single mutation defect resulted from mutating Tyr 253 to Ala, which caused significant 10 -20-fold reductions in association rate constants for factor Xa reactions relative to wild type, but these defects were observed both in the absence and presence of pentasaccharide and full-length heparin activators. All Tyr 253 single mutants exhibited wild-type association rate constants for reactions with thrombin in the absence or presence of heparin. The reactivities of the Glu 255 mutants with thrombin were at most 2-fold altered from wild-type except for reactions in the absence of heparin which showed ϳ3-5-fold elevated reactivities similar to that exhibited by the Y253A/E255A double mutant. Mutation of Lys 257 alone to Met or Glu produced small effects on association rate constants for reactions with factor Xa and thrombin in the absence or presence of heparin except for the charge reversal mutation, which significantly decreased the rate constant for the reaction with thrombin in the absence of heparin.
Effect of the Strand 3C Mutations on Antithrombin Reactivity with Factor IXa-The series of triple, double, and single mutations of Tyr 253 , Glu 255 , and Lys 257 produced relative impairments in antithrombin reactivity with factor IXa that largely paralleled those observed with factor Xa with the notable exception that the reactivity impairments no longer depended on heparin activation of the serpin. This contrasted with the behavior of the chimera, which showed a large loss in factor IXa reactivity in the presence of heparin (0.2-0.9% of wild type) and only modest loss in its absence (ϳ40% of wild type). Mutation of all three conserved strand 3C residues to those in the chimera or to alanine or just Tyr 253 and Glu 255 to alanine thus produced marked decreases in antithrombin reactivity, which approached or approximated the reactivity losses of the heparin-activated chimera (1-9%, 0.6 -0.9%, and 0.4 -1% of wild type, respectively) for both the unactivated and heparin-activated inhibitor. Mutation of Tyr 253 appeared to be most responsible for these reactivity defects, since single mutations to Ala or Lys, although not to Phe, produced decreases in reactivity approaching that of the double and triple alanine mutants (1-4% of wild type). The Lys mutation in particular resulted in a much more pronounced reactivity impairment with

TABLE 1 Dissociation constants and maximal fluorescence changes for heparin pentasaccharide binding to antithrombin variants and stoichiometries of inhibition (SI) for the reaction of antithrombin variants with thrombin in the absence and presence of full-length heparin
Dissociation constants and maximal fluorescence changes for antithrombin-pentasaccharide interactions were measured by fluorescence titrations at pH 7.4, I ϭ 0. 35 factor IXa than with factor Xa. Glu 255 made a lesser contribution to the reactivity defects of the double mutant based on the modest effects of mutating this residue together with Lys 257 or alone. Interestingly, whereas reactivity impairments for pentasaccharide and full-length heparin-activated antithrombin variants were typically of comparable magnitude, the modest reactivity impairments of the single Glu 255 mutants in the presence of the pentasaccharide were magnified severalfold in the presence of the full-length heparin, suggesting that these mutations impaired heparin bridging. Mutation of Lys 257 produced no significant decreases in reactivity except for the K257E mutant in the presence of a bridging heparin, consistent with this residue not being critical for factor IXa reactivity.

Effects of Mutating Residues
Neighboring the Putative Exosite-Mutation of the remaining conserved but less surface-accessible strand 3C residue, Gln 254 , to the residue in the strand 3C chimera resulted in minimal differences from the wild-type inhibitor in the reactivity of the mutant with factor Xa, thrombin, or factor IXa with or without heparin activation. The P6Ј residue, Arg 399 , in the reactive loop points toward the putative exosite residues in strand 3C in the crystal structure of the antithrombin-heparin pentasaccharide complex and was therefore mutated. Factor Xa reactivity was only slightly affected by mutating Arg 399 to Met, and somewhat larger but still modest effects on reactivity were observed when the charge of this residue was reversed by mutating  Table 2. The dashed lines indicate wild-type rate constants. Rate constants were measured under pseudo-first-order conditions from the extent of protease inactivation as a function of time or as a function of heparin concentration over a fixed time as described under "Experimental Procedures." Bridging effects of the full-length heparin were measured in the absence of calcium for reactions of antithrombin with thrombin and factor Xa and in the presence of calcium for reactions with factor IXa. Although the former conditions reduce the bridging effect for the full-length heparin-catalyzed antithrombinfactor Xa reaction, a significant bridging effect is still observed under these conditions (4,19).
to Glu, although only for heparin-dependent reactions. Factor IXa reactivity was affected insignificantly by the Arg 399 mutations except for the charge reversal mutant, which showed a 7-fold reduced reactivity when activated by the full-length bridging heparin. Mutating Arg 399 together with Tyr 253 to Ala produced effects on factor Xa and factor IXa reactivity that were indistinguishable from those produced by mutating Tyr 253 alone to Ala. Thrombin reactivity was affected significantly (ϳ5-fold) by mutating Arg 399 to neutral Met or Ala residues (the latter in the R399A/ Y253A double mutant) but only modestly by the charge reversal to Glu for both unactivated and heparin-activated inhibitor reactions. Glu 237 in strand 4 of sheet C has been implicated as being a part of the factor Xa exosite by modeling studies (29). Because this residue is the same in ␣ 1 -proteinase inhibitor and antithrombin and therefore was not mutated in the strand 4C chimera we previously made, it was mutated to alanine. Minimal effects of the mutation on factor Xa, thrombin, or factor IXa reactivity either in the absence or presence of heparin were observed.

DISCUSSION
We have used a site-directed mutagenesis approach to localize an exosite on antithrombin, previously shown to reside in strand 3C, which is responsible for accelerating the reaction of antithrombin with factors Xa and IXa following its conformational activation by a specific heparin pentasaccharide (17,21). To identify which of the eight residues we had replaced in the exosite-defective strand 3C chimera actually comprise the exosite, we focused our attention on the four that are highly conserved in the 13 antithrombins, representing all five vertebrate families, that have been sequenced (22). Our findings demonstrate that two of the highly conserved residues, Tyr 253 and Glu 255 , are key contributors to the reactivity of antithrombin with both factor Xa and factor IXa. Mutation of Tyr 253 and Glu 255 together was thus found to replicate the exosite defect of the strand 3C chimera in showing a nearly complete abrogation of the 100-fold rate enhancement produced by the heparin pentasaccharide on the reaction of antithrombin with factor Xa. These mutations had modest effects on the factor Xa reaction in the absence of heparin and produced equal or greater than wild-type rate enhancements due to bridging by a fulllength heparin, indicating that the mutations specifically affected the determinants on antithrombin responsible for the increased reactivity of the inhibitor toward factor Xa following its conformational activation by the pentasaccharide. The combined mutation of Tyr 253 and Glu 255 also produced major decreases in the reactivity of antithrombin with factor IXa like those of the strand 3C chimera, but, unlike the reaction with factor Xa, the losses in reactivity were observed both for the native and the pentasaccharide-activated inhibitor. Importantly, the mutations did not appreciably affect the reactivity increase due to full-length heparin bridging. Tyr 253 and Glu 255 therefore also appear to comprise an exosite that is critical for antithrombin to react with factor IXa, but in this case, the contribution of the exosite is not dependent on pentasaccharide activation of the inhibitor.
These apparently paradoxical findings can be rationalized when one considers that the slow reactivity of antithrombin with proteases in the absence of heparin derives at least in part from a minor equilibrium fraction of activated antithrombin (13,34). The extent to which this preexisting equilibrium fraction of activated antithrombin contributes to the basal reactivity with different proteases was only recently revealed by studies of a variant antithrombin locked in its native state by engineering a disulfide bond between the N-terminal hinge of the reactive loop and the serpin body (31). The disulfide bond served to fix the reactive loop hinge-sheet A interaction, a defining feature of the native state (13,32), in a manner that blocked the activating extension of the loop away from sheet A. The reactivity of the native state-locked antithrombin variant in the absence of heparin was normal with thrombin, somewhat reduced with factor Xa, and barely detectable with factor IXa, indicating that factor Xa and thrombin are both reactive with the native inhibitor conformation, whereas factor IXa is effectively unreactive. Moreover, heparin pentasaccharide activation of the native state-locked antithrombin failed to enhance its reactivity with either factor Xa or factor IXa, in keeping with the variant being unable to make the exosite for these proteases accessible because of the need to fully extend the reactive loop for factor Xa and factor IXa to engage the exosite. These findings support the view that the observed reactivity of factor IXa with native antithrombin in the absence of heparin reflects a reaction of the protease with a minor fraction of activated inhibitor (Ͻ1%) in equilibrium with a largely unreactive native inhibitor (association rate constant Ͻ Ͻ1 M Ϫ1 s Ϫ1 ), and the pentasaccharide enhancement in reactivity is due to all of the inhibitor becoming activated; i.e. the exosite is fully responsible for antithrombin

TABLE 2 Association rate constants for the reactions of wild-type and variant antithrombins with thrombin, factor Xa, and factor IXa in the absence and presence of pentasaccharide and full-length heparins
Association rate constants were measured under pseudo-first-order conditions at 25°C in pH 7.4, I ϭ 0.15 buffer without calcium for reactions with thrombin and factor Xa and containing 5 mM calcium for reactions with factor IXa as described under "Experimental Procedures." Average values Ϯ S.E. from at least three determinations are reported except for some reactions with factor IXa in the absence of heparin, in which case only a single value was obtained. reactivity with factor IXa both in the absence and presence of heparin (31). By contrast, the reactivity of factor Xa with antithrombin in the absence of heparin reflects contributions of an intrinsic reactivity with the native inhibitor conformation (ϳ10 3 M Ϫ1 s Ϫ1 ) as well as the reactivity of the minor activated inhibitor fraction; i.e. the generation of the exosite makes an important contribution to the pentasaccharide enhancement of antithrombin reactivity. Mutations of exosite residues are thus expected to produce modest reductions in antithrombin reactivity with factor Xa in the absence of heparin but large decreases in the pentasaccharide-activated inhibitor reactivity, whereas large reductions in reactivity for reactions with factor IXa are expected both in the absence and presence of the pentasaccharide, in accordance with our findings. Our results thus support the conclusion that Tyr 253 and Glu 255 together contribute a major part of the exosite that is responsible for enhancing antithrombin reactivity toward both factor Xa and factor IXa and that the accessibility of this exosite is dependent on conformational activation of the inhibitor for both protease reactions. The specificity of the antithrombin exosite for factors Xa and IXa is supported by the observation that the mutations of Tyr 253 and Glu 255 minimally impaired the reactivity of antithrombin with thrombin in the absence or presence of pentasaccharide or bridging heparins. Mutation of Glu 255 was found to enhance the reactivity of antithrombin with thrombin most significantly in the absence of heparin and more modestly for the pentasaccharide-activated inhibitor in a manner that depended on the context of this mutation. These findings confirm those of previous studies in suggesting that the reactivity of native antithrombin with thrombin is influenced by the surface charge of the inhibitor surrounding the reactive loop due to the close contact of the loop and bound protease with the serpin body in the native state (21). Upon pentasaccharide activation, the reactive loop moves away from the inhibitor surface, and the effects of surface charge are less pronounced as evidenced by the normalization of reactivities of the mutant antithrombins to those of wild type in the presence of pentasaccharide or bridging heparins. The elevated reactivity produced by all single mutations of Glu 255 suggests that this residue makes an unfavorable interaction with thrombin and functions to down-regulate antithrombin reactivity with thrombin in the absence of heparin. This unfavorable interaction may involve the close approach of Glu 255 to Glu 192 of thrombin in the native antithrombin-thrombin Michaelis complex based on the crystal structure of the complex bound to a bridging heparin (33,34). No unfavorable interaction is expected in the Michaelis complexes with factor Xa and IXa, since residue 192 is Gln in these proteases. The effect of mutating Glu 255 was found to depend on the adjacent Lys 257 and Tyr 253 residues, since when paired with the K257A mutation, the elevated reactivity was lost, and when combined with K257M and Y253K mutations, the elevated reactivity was magnified. The context dependence of these effects is clear from the fact that single Y253K and K257M mutations minimally affected the reactivity. Mutation of Lys 257 alone caused significant reductions in antithrombin reactivity with thrombin when the charge of this residue was reversed, this being most pronounced in the absence of heparin (ϳ50-fold), possibly due to the favorable interaction of Lys 257 with Asp 60 in thrombin. Mutation of Arg 399 produced modest effects on reactivity of less than 10-fold, which were context-dependent and least severe when the charge was reversed. Since factor Xa, like thrombin, is reactive with the native conformation of antithrombin, which is most sensitive to the surface charge distribution (31), similar context-dependent effects of mutations may account for some of the observed effects on antithrombin reactivity with factor Xa in the absence of heparin, such as the larger than expected effect of the single Y253A mutation.

Antithrombin variant Association rate constants for antithrombin-thrombin reactions
Single mutations of Tyr 253 or Glu 255 suggested that Tyr 253 is more critical than Glu 255 for exosite function, in keeping with the findings of a previous Glu 255 mutagenesis study (35). Notably, replacement of Tyr 253 with Ala greatly perturbed both factor Xa and factor IXa exosite interactions, replacement with Lys only perturbed the factor IXa and not the factor Xa exosite interaction, and replacement with Phe had minimal effects on either exosite interaction. The two exosite interactions thus may involve hydrophobic and electrostatic contacts with a complementary exosite in the protease with significant context-dependent differences in the types of substitutions tolerated in the two interactions. However, such conclusions should be tempered by the fact that our principal measure of the contribution of antithrombin residues to the exosite interaction, changes in the association rate constant for antithrombin reactions with factor Xa and IXa produced by mutation of antithrombin residues, may not reflect the true contribution of a residue to the exosite interaction. Of particular relevance is our observation that engagement of antithrombin or heparin bridging exosite interactions in the Michaelis complex causes the overall association rate constant to become limited by the diffusional encounter of antithrombin and protease to form the Michaelis complex (i.e. the on-rate constant) (36). Under such conditions, the association rate constant would no longer be responsive to disruptions in the exosite that affect the off-rate constant, and thus, a threshold disabling of the antithrombin exosite may be required to produce detectable effects on the association rate constant, especially in the case of the factor Xa reaction, which approaches a diffusion-limited reaction already with pentasaccharide activation of antithrombin. Such threshold requirements could also account for why heparin bridging seemed to overcome some of the exosite defects observed for pentasaccharide-activated inhibitor reactions with either factor Xa or factor IXa. Equilibrium binding studies of the antithrombin-protease Michaelis complex interaction will therefore be required to directly assess the contributions of Tyr 253 and Glu 255 to the exosite interaction with factors Xa and factor IXa.
Two other conserved residues in strand 3C, Gln 254 and Lys 257 , did not appear to be important contributors to the exosite for factors Xa and IXa. Single Lys 257 mutations had minimal effects on antithrombin reactions with factor Xa or factor IXa, and combined mutations with Tyr 253 and/or Glu 255 caused effects similar to those observed when the latter residues were mutated without mutating Lys 257 . Mutation of Gln 254 also showed no significant effects on antithrombin reactions with any of the proteases whether heparin was present or not. Two residues outside of strand 3C were also mutated to assess their potential effects on the antithrombin exosite. Although Arg 399 , the P6Ј residue in the reactive loop, points toward the strand 3C exosite we identified in this study and this region contributes to an exosite in another serpin (37), Arg 399 mutations had modest effects on the exosite interaction even when the charge of this residue was reversed. Glu 237 in strand 4 of sheet C has been proposed as the factor Xa interaction exosite based on modeling of the antithrombin-factor Xa Michaelis complex using the crystal structure of the heparin cofactor II-thrombin Michaelis complex as a template (29). Mutation of Glu 237 showed that it does not contribute to the antithrombin exosite for factors Xa or IXa and argues against the heparin cofactor II-thrombin complex as an appropriate template for the factor Xa Michaelis complex. Determination of whether other neighboring residues outside of strand 3C, such as His 319 , which was not changed in the previous chimeras we prepared, contribute to the exosite will require further mutagenesis of these residues.
Attempts to localize a complementary exosite in factors Xa and IXa that interacts with the antithrombin exosite have identified Arg 150 in the autolysis loop of these proteases as an important determinant of the complementary exosite (38,39). Our modeling of the antithrombin-factor Xa Michaelis complex using the recently determined structures of the antithrombin-thrombin-heparin ternary complex as a template (33,34) place the autolysis loop of factor Xa in close proximity to the strand 3C exosite residues we have identified (Fig. 4). The reported involvement of the 60-loop residue, Gln 61 , in the complementary protease exosite, on the other hand (40), does not fit with our proposed localization of the antithrombin exosite nor with recent mutagenesis studies that show minimal effects of factor Xa 60-loop mutations on the antithrombin exosite interaction (41). Comparison with the structure of unactivated antithrombin suggests that the exosite residues are available in the native partially buried reactive loop conformation but cannot interact with the autolysis loop of the proteases until the inhibitor reactive loop is expelled and extends away from sheet A upon heparin activation. Our identification of Tyr 253 and Glu 255 as critical exosite residues in antithrombin has a precedent in tissue factor wherein a similar Tyr-Glu pair of residues was shown to contribute to an exosite that mediates interaction with and activation of factor VIIa (42). In this case, the Tyr 94 -Glu 91 pair were proposed to stabilize a disordered helix (residues 165-170, chymotrypsin numbering) in free factor VIIa that is part of the protease activation domain. The stabilization involved the capping of the N-terminal positive end of the helix dipole by these residues. The negative charge of Glu 255 and the ability of Tyr 253 to engage in -cation interactions (43) may be critical for promoting a productive interaction with Arg 150 of the autolysis loop of factors Xa and IXa. The final proof of the importance of this exosite will require crystallization of Michaelis complexes of heparin-activated antithrombin with either protease.