Elucidation of Structural Requirements on Plasminogen Activator Inhibitor 1 for Binding to Heparin*

Plasminogen activator inhibitor 1 (PAI-1), a member of the serpin superfamily of proteins, has been demonstrated previously to interact functionally with the glycosaminoglycan heparin Biochemistry 30,1021-1028). Heparin specifically enhances the rate of association between PAI-1 and thrombin about 2 orders of magnitude, whereas no effect is detected with other serine proteases (e.g. factor Xa). For the heparin-dependent serpins antithrombin I11 and heparin cofactor 11, basic amino acid residues in and around the helix D subdomain were proposed to be involved in the binding of glycosami-noglycans. Here we employed site-directed mutagenesis of full-length PAI-1 cDNA to identify the amino acid residues that mediate heparin binding. To that end, 15 single-point mutants of PAI-1, each having individual arginyl, lysyl, or histidyl residues replaced by a neutral (alanyl) residue (“ala-scan”), and one dou-ble mutant were constructed, expressed in Escherichia coli, and purified to apparent homogeneity. The purified

Plasminogen activator inhibitor 1 (PAI-1), a member of the serpin superfamily of proteins, has been demonstrated previously to interact functionally with the glycosaminoglycan heparin (Ehrlich, H. J., Keijer, J., Preissner, K. T., Klein Gebbink, R., and Pannekoek, H. (1991) Biochemistry 30,[1021][1022][1023][1024][1025][1026][1027][1028]. Heparin specifically enhances the rate of association between PAI-1 and thrombin about 2 orders of magnitude, whereas no effect is detected with other serine proteases (e.g. factor Xa). For the heparin-dependent serpins antithrombin I11 and heparin cofactor 11, basic amino acid residues in and around the helix D subdomain were proposed to be involved in the binding of glycosaminoglycans. Here we employed site-directed mutagenesis of full-length PAI-1 cDNA to identify the amino acid residues that mediate heparin binding. To that end, 15 single-point mutants of PAI-1, each having individual arginyl, lysyl, or histidyl residues replaced by a neutral (alanyl) residue ("ala-scan"), and one double mutant were constructed, expressed in Escherichia coli, and purified to apparent homogeneity. The purified biologically active proteins were subjected to the following analyses: (i) heparin-dependent inhibition of thrombin; (ii) heparin-dependent formation of sodium dodecyl sulfate-stable complexes with thrombin; and (iii) binding to and elution from heparin-Sepharose. Based on the data presented, we propose that the amino acid residues Lys", Lys", Arg", LysBo9 and Lyses constitute major determinants for heparin binding of PAI-1. These residues are located in and around the helix D domain and are conserved in the other heparindependent thrombin inhibitors, antithrombin I11 and heparin cofactor 11.
Heparin is a sulfated glycosaminoglycan that interacts with a variety of proteins such as growth factors (Maciag et al., 1984;Shing et al., 1984), coagulation factors (Fujikawa et al., 1973), fibronectin (Stathakis and Mosesson, 1977), vitronectin (Preissner and Muller-Berghaus, 1987), tissue-type plas-* Part of this work was been presented at the XIIIth International Congress on Thrombosis and Haemostasis (Amsterdam, June 30-July 6, 1991). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "aduertlsement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
5 To whom correspondence should be addressed Dept. of Molecular Biology, Central Laboratory of the Netherlands Red Cross Blood Transfusion Service, Plesmanlaan 125, 1066 CX Amsterdam, The Netherlands. Tel.: 020-5123125; Fax: 020-5123170. minogen activator (t-PA)' (Andrade-Gordon and Strickland, 1986), and the serine protease inhibitors (serpins) antithrombin I11 (Rosenberg and Damus, 1973), heparin cofactor I1 (Tollefsen and Blank, 1981), protease nexin I (Baker et al., 1980), and plasminogen activator inhibitor 1 (PAI-1) (Ehrlich et al., 1991a). One of the most dramatic consequences resulting from these interactions is the potent anticoagulant effect observed in uiuo after heparin administration (Brinkhous et al., 1939). Extensive studies on the underlying molecular mechanism revealed that highly negatively charged heparin molecules accelerate the inhibition of thrombin and other serine proteases of the coagulation system by antithrombin I11 (Rosenberg and Damus, 1973). The cofactor function of heparin has been explained by the observations that (i) heparin provides for a template in the assembly of both the protease and the serpin (Pomerantz and Owen, 1978;Griffith, 1982;Nesheim, 1983); and (ii) heparin induces a conformational change in the reactive site of the serpin, thereby enhancing the reactivity toward thrombin (Rosenberg and Damus, 1973;Olson et al., 1981).
In addition to its anticoagulant effects, heparin was shown to enhance fibrinolysis by binding to t-PA, thereby stimulating the plasminogen activating activity of this enzyme in the absence of fibrin (Andrade-Gordon and Strickland, 1986;Piiques et al., 1986;Stein et al., 1989). Moreover, as we have shown recently, there is also a profibrinolytic effect resulting from a functional interaction between heparin and the serpin PAI-1, the fast inhibitor of t-PA and urokinase-type plasminogen activator (urokinase-type PA) (Ehrlich et al., 1991a). The striking selectivity of PAI-1 for the inhibition of t-PA and urokinase-type PA is compromised in the presence of either heparin (Ehrlich et al., 1991a) or vitronectin , resulting in an enhancement of the association rate between PAI-1 and thrombin of 2-3 orders of magnitude. The effect of heparin on PAI-1-mediated inhibition is apparently restricted to thrombin inasmuch as no enhancement was observed with other serine protease tested, e.g. factors Xa and XIIa (Keijer et al., 1991). The interaction between PAI-1 and thrombin ultimately leads to neutralization of PAI-1 because of the formation of inactive, SDS-stable thrombin-PAI-1 complexes and the proteolytic cleavage of PAI-1 by thrombin (Ehrlich et al., 1991a(Ehrlich et al., , 1991b. The molecular mechanism of the effect of vitronectin, the PAI-1-binding protein in plasma (Declerck et al., 1988;Wiman et al., 1988) and in the endothelial cell matrix (Mimuro and Loskutoff, 1989;Preissner et al., 1990), on this reaction remains to be elucidated. For heparin, however, experimental evidence suggests that this glycosaminoglycan functions as a template for the assembly of both the protease and the protease inhibitor The abbreviations used are: t-PA, tissue-type plasminogen activator; PAI-1, plasminogen activator inhibitor 1; SDS, sodium dodecyl sulfate; RF, replicative form.

Heparin Binding Site
on PAI-1 11607 (Ehrlich et al., 1991a). Thus, the effect of heparin on the interaction between PAI-1 and thrombin resembles its effect on the acceleration of the inhibition of thrombin by the serpin antithrombin 111.
Recently, several studies have attempted to characterize the nature of the heparin binding site in antithrombin I11 and of heparin cofactor 11, a related serpin that also inhibits thrombin in a reaction accelerated by heparin, heparan sulfate, and dermatan sulfate. Chemical modification experiments suggest that several basic amino acid residues in antithrombin I11 (Lys'07, L y P , and Lys13'j) mediate binding to heparin (Chang, 1989). Using site-directed mutagenesis, Blinder and Tollefsen (1990) demonstrated the critical importance of Lys'= in heparin cofactor I1 (which aligns to L~S '~~ of antithrombin I11 (Huber and Carrell, 1989)) in glycosaminoglycan binding by heparin cofactor 11. These crucial amino acid residues are partially conserved in PAI-1 (Ehrlich et al., 1991a). In this study we have constructed an extensive set of PAI-1 point mutants that were analyzed to establish the location of the amino acid residues that constitute the heparin binding site of PAI-1.

EXPERIMENTAL PROCEDURES
Materid-Restriction endonucleases were purchased either from New England Biolabs or Bethesda Research Laboratories. T4 polynucleotide kinase was obtained from Pharmacia LKB Biotechnology Inc. and the Klenow fragment of DNA polymerase I from Boehringer Mannheim. The replicative forms (RF) of M13mp18am4 and M13mp19am4 DNA were from Anglian Biotechnology Limited (Colchester, United Kingdom). Sequenase DNA polymerase was purchased from United States Biochemical Corp. The synthetic substrates H-D-isoleucyl-prolyl-arginyl-p-nitroanilide (S2288) and H-Dphenylalanyl-pipecolyl-arginyl-p-nitroanilide (S2238) were obtained from KabiVitrum (Stockholm, Sweden). Heparin-Sepharose and Q-Sepharose Fast Flow were from Pharmacia. Unfractionated heparin (H-3125; average molecular weight, 15,000-18,000), isolated from porcine intestinal mucosa, was purchased from Sigma and Comp.
Proteins-Bowes melanoma-derived two-chain t-PA (910,000 IU/ mg) was obtained from Biopool (Umea, Sweden). Purified human thrombin was kindly provided by Dr. K. Mertens (Department of Blood Coagulation, this institute). Active-site titration with p-nitrophenyl p'-guanidinobenzoate (Chase and Shaw, 1970) yielded a concentration of 5.0 mg/ml, a value that is consistent with a determination of the protein concentration (4.5 mg/ml) following the Bradford procedure (Bradford, 1976). Thrombin was radiolabeled with lz5I, using the Iodogen method, resulting in a specific radioactivity of 1.6 pCi/pg protein. Vitronectin was purified from human plasma to apparent homogeneity as described previously (Preissner et al., 1985). The protein concentration of the purified preparation was determined assuming an extinction coefficient of E;& = 13.0 and a molecular weight of 75,000 (Dahlback and Podack, 1985).
General Methods-Standard molecular biological techniques, including plasmid and bacteriophage M13 DNA isolations, restriction fragment isolation, enzyme reactions with nucleic acids, and bacterial transformations, were performed as described (Maniatis et al., 1982). Nucleotide sequence determinations were carried out according to the dideoxy chain termination method (Sanger et al., 1977) with Sequenase DNA polymerase. Synthetic oligonucleotides were synthesized by solid-phase phosphoramidite chemistry on an automated synthesizer (Applied Biosystems, model 381A).
Construction of Expression Plasmids-The construction of the plasmid suitable for expression of biologically active recombinant ("wild-type") PAI-1 in transformed Escherichia coli has been described (Ehrlich et al., , 1991a. This plasmid, designated pMBLll/PAI-l, carries the E. coli tryptophan promotor-operator regulatory elements, the gene for @-lactamase, and full-length PAI-1 cDNA (Pannekoek et al., 1986). Oligonucleotide-directed Mutagenesis-Oligonucleotide-directed, site-specific mutagenesis was carried out as described (Kramer et al., 1984), utilizing three different M13 vectors as single-stranded template. First, a 384-base pair EcoRI-SstI restriction fragment of pMBLll/PAI-l, covering part of pMBLll and the 5' portion of the cDNA coding for mature human PAI-1 (positions 52-436), was inserted into M13mp19am4 RF DNA. Second, a 609-base pair SstI-Sal1 restriction fragment, corresponding to positions 436 and 1045, was inserted into M13mp19am4 RF DNA. For the construction of the third vector, both strands of a linker (sequence of "upper" strand, 5'-

GAATTCGAATGCCGGCCCACCTGGCCATGCATCCATGGGTC-GAC-3'), containing the restriction sites EcoRI-BsmI-SfiI-NsiI-NcoI-
SalI, were synthesized and employed to substitute for the EcoRI-SalI portion of the multiple cloning site of phage M13mp18am4 RF DNA, yielding M13mp18am4' RF DNA. A 330-base pair SalI-NsiI restriction fragment from full-length, human PAI-1 cDNA (corresponding to positions 1045 and 1375, respectively) was then inserted into M13mp18am4' RF DNA. The oligonucleotides and constructs used for the construction of the individual mutants are summarized in Table I. The fidelity of the entire PAI-1 coding sequences in the mutated EcoRI-SstI, SstI-SalI, or SalI-NsiI restriction fragments was verified by DNA sequencing. Finally, the mutated EcoRI-SstI, SstI-SalI, and NsiI-Sal1 fragments were isolated to substitute the nonmutated corresponding fragments of the expression plasmid Expression of PAI-I and PAI-1 Mutants in E. coli and Purification of the Recombinant Proteins-Expression of PAI-1 wild-type and mutants in E. coli strain 1046 and purification of the recombinant proteins were carried out as described previously , except that heparin-Sepharose was utilized instead of an immunoaffinity matrix for the final purification step. The supernatants of the sonicated bacterial pellets from a 250-ml culture were treated with pancreatic DNase I and subsequently fractionated by ammonium sulfate precipitation (25-45%). The resulting pellets were resuspended in 20 mM Tris-HC1 (pH 8.0), 0.1% (v/v) Tween 80 and, after extensive dialysis against the same buffer, adsorbed batchwise onto 2 ml of Q-Sepharose Fast Flow matrix that was then packed into a column, washed with 10 ml of 20 mM Tris-HC1 (pH 8.0), 0.1% (v/v) Tween 80, and eluted with 5 ml of 20 mM Tris-HC1 (pH 8.0), 0.1% (v/v) Tween 80, 200 mM NaC1. The eluate was dialyzed against 20 mM Tris-HC1 (pH 7.4), 0.1% (v/v) Tween 80,50 NaCl, and adsorbed for 16 h at 4 "C by end-over-end rotation to 1.5-ml packed beads of heparin-Sepharose. Finally, the matrix was packed into a column, washed with 5 column volumes of 20 mM Tris-HC1 (pH 7.4), 0.1% (v/v) Tween 80, 100 mM NaCl, and eluted with 20 mM Tris-HC1 (pH 7.41, 0.1% (v/v) Tween 80, 400 mM NaCl. The eluates were then dialyzed extensively against 20 mM Tris-HC1 (pH 8.0), 0.1% Tween 80, 100 mM NaCl (TST buffer), and PAI-1 antigen in the purified fractions was determined using an immunoradiometric assay that utilized two different monoclonal antibodies raised against human PAI-I, as described previously (Lambers et al., 1988;Ehrlich et al., 1990). PAI-1 mutant K154A could not be detected using this assay: its concentration was estimated from the relative intensity of the PAI-1-related band on a silver-stained gel. SDS-polyacrylamide gel electrophoresis followed by silver staining essentially showed a single band with a molecular weight of about 42,000 for the purified PAI-1 preparations.
Activation of PAI-1 and PAI-1 Mutants and Titration of their Respective Activities-PAI-1 or PAI-1 mutants (100-500 pg/ml) were activated by incubation for 2 h at room temperature in 4 M guanidinium C1 (Hekman and Loskutoff, 1985). The denaturant was removed by overnight dialysis at 4 "C against TST buffer. Increasing amounts of activated PAI-1 or PAI-1 mutants were incubated at 37 "C for 1 h in a total volume of 50 pl with 1.5 nM two-chain t-PA in TST buffer. Then, 200 pl of the chromogenic substrate S2288 (0.6 mM in TST buffer) was added, and residual t-PA activity was determined from a linear plot of the increase of absorbance at 405 nm over time.
Heparin-mediated Inhibition of Thrombin by PAI-1 and PAI-1 Mutants-Thrombin (0.3 nM) was incubated with PAI-1 or PAI-1 mutants (5 nM active PAI-1, titrated on t-PA as described above) in a total volume of 30 p1 of TST buffer in the absence or in the presence of 45 nM vitronectin or 1 unit/ml heparin. After 1 h at 37 "C, 200 pl of the synthetic substrate S2238 was added, and thrombin activity was determined from the linear increase of absorbance at 405 nm. The increase of absorbance measured for the sample containing only thrombin was taken as 100%.

K288A
SstI-SalI/mplS The one-letter code for amino acids is used to designate the mutants. The alignment with the structure of cleaved a,-proteinase inhibitor (Loebermann et al., 1984) follows the proposal of Huber and Carrel1 (1989). The abbreviations hx, s, and coil are utilized to designate the helical subdomains, the 0-sheets, and the random coil areas, respectively. hxI/10.17

5'GTCGACCTCAGGGGCCCCTAGAGAAC3'
SalI-NsiI/mpl8* Over a window of 6 residues according to Emini. polyacrylamide gel (Laemmli, 1970). The relative position of the lZ5Iradiolabeled material, as compared with molecular weight markers, was visualized by autoradiography.
Elution of PAI-I and PAI-I Mutants from Heparin-Sepharose-One ml of PAI-1 wild-type or mutants (100 pg/ml in 20 mM Tris-HCl (pH 7.4), 0.1% (v/v) Tween 80) was incubated for 16 h at 4 "C by end-over-end rotation with 100 p1 of a slurry of heparin-Sepharose. Quantitative binding was observed for all PAI-1 proteins. Elution was then carried out by washing the beads with 0.5 ml of 20 mM Tris-HCl (pH 7.4), 0.1% (v/v) Tween 80 and then by stepwise increasing the NaCl concentration (steps of 50 mM). The amounts of wild-type PAI-1 or mutants in each fraction were assessed using the immunoradiometric assay described above. The optimal NaCl concentration for the elution of PAI-1 and the various PAI-1 derivatives was obtained by drawing symmetrical profiles, using the values of the immunoradiometric assays.

Expression, Purification, and Titration of PAI-1 and Mu-
tants Thereof-In agreement with our previous observations, PAI-1 and mutants thereof were efficiently produced in E. coli strain 1046 that had been transformed with the different expression plasmids derived from pMBLll/PAI-1 . These plasmids carried the entire coding sequence either for mature, wild-type PAI-1 or for the mutants as depicted in Table I. Our choice for mutating basic (arginyl, lysyl, or histidyl) residues relied on: (i) a comparison between the amino acid sequences of antithrombin 111 and PAI-1, in particular of the region of antithrombin I11 implicated either in heparin binding (helix D) or in a heparin-induced conformational change (Arg'87/Arg'ss and Lys"'); and (ii) a computer program that predicts surface probability. Each 250-ml culture of transformed E. coli cells yielded at least 500 pg of purified protein, irrespective of the nature of the PAI-1 variant. Upon activation with guanidinium C1, the purified PAI-1 proteins displayed similar specific activities in a titration assay, measuring inhibition of the amidolytic activity of t-PA (data not shown). These observations indicate that none of the mutations grossly affected the structure of the respective PAI-1 proteins.
Cofactor-dependent Thrombin Inhibition by PAI-1 and PAI-1 Mutants-The effect of heparin on the inhibition of thrombin by PAI-1 and PAI-1 mutants was analyzed as described (Ehrlich et al., 1991a). Experiments, carried out in the pres-ence of vitronectin or in the absence of either cofactor, served as controls. In the absence of heparin or vitronectin, no inhibition of thrombin by PAI-1 wild-type or by any of the PAI-1 mutants was observed (data not shown). The results obtained in the presence of either one of the cofactors are depicted in Fig. 1. Upon adding vitronectin to the reaction mixtures, PAI-1 wild-type and each of the PAI-1 mutants were endowed with thrombin inhibitory properties. Under the conditions employed, approximately 50-75% inhibition of the amidolytic activity of thrombin was observed, indicating that both wild-type and any of the PAI-1 mutants can inhibit thrombin in the presence of vitronectin. In contrast, the effect of heparin as a cofactor for thrombin inhibition is markedly different for the various PAI-1 derivatives. A similar thrombin inhibition in the presence of heparin is detected for wild-type However, hardly any inhibition of thrombin in the presence of heparin is observed using the mutant proteins PAI-1 R76A, K80A, and K88A, whereas these proteins fully inhibit thrombin in the presence of vitronectin. Partial inhibition of thrombin in the presence of heparin is detected with the mutant proteins PAI-1 K65A and PAI-1 K69A again full inhibition of thrombin by these latter mutant proteins is apparent in the presence of vitronectin. It should be noted that three mutants, i.e. PAI-1 HlOA, PAI-1 K28A (His" and Lys" comprise the two positively charged amino acid residues in the helix A subdomain of PAI-1, the counterpart of which in antithrombin I11 contains the Arg47 residue, proposed to be involved in heparin binding (Owen et al., 1987)), and PAI-1 H77A (the latter histidyl residue is located between the crucial Arg76 and Lys@" residues in PAI-1) displayed similar thrombin inhibitory properties in the presence of either vitronectin or heparin as wild-type PAI-1 (data not shown).

Cofactor-dependent Formation of SDS-stable Complexes between PAI-l or PAI-l Mutants and Thrombin-A typical
feature of the inhibition of a serine protease by its cognate serpin is the ultimate generation of equimolar, inactive complexes that do not dissociate in the presence of SDS. Consequently, we examined the formation of complexes between '2'I-labeled thrombin and PAI-1 (derivatives) in the presence of either heparin or vitronectin, using SDS-polyacrylamide gel electrophoresis. The results are depicted in Fig. 2. Both heparin and vitronectin efficiently promoted the generation of SDS-stable complexes between thrombin and either PAI-1 wild-type or the mutants PAI-1 K28A, K154A, R186A/ R187A, K207A, and K243A. In addition, the mutants PAI-1 HlOA, H77A, K263A, K277A, and K288A also displayed no significant difference in complex formation from wild-type PAI-1 (data not shown). In contrast, hardly any complex formation is observed with the mutant proteins PAI-1 R76A, K80A, and K88A, whereas only part of the '251-labeled thrombin formed complexes with the mutants PAI-1 K65A and K69A in the presence of heparin. Again, it should be noted that complex formation of thrombin with any of the PAI-1 derivatives was comparable in the presence of vitronectin. These observations are consistent with the data presented on the inhibition of thrombin by the different PAI-1 species in the presence of either heparin or vitronectin (Fig. 1). An exception is represented by the mutant protein PAI-1 K191A: this protein, although fully inhibitory in a suitable amidolytic assay toward t-PA, urokinase-type PA, and thrombin in the presence of either heparin or vitronectin, reproducibly failed to form SDS-stable complexes with thrombin in the presence of either cofactor.
Interaction of PAI-1 and PAI-1 Mutants with Heparin-Sephurose-It is thought that the function of heparin in the acceleration of the inhibition of thrombin by its cognate heparin-dependent serpin antithrombin I11 is to provide for a template to assemble both the enzyme and the inhibitor and/ or to induce a conformational change that increases the reactivity of the inhibitor. In either case, efficient binding of heparin to the serpin is a prerequisite for the observed enhanced inhibition. To determine relative affinities of wildtype PAI-1 and the various PAI-1 mutants for heparin, these species were separately bound to heparin-Sepharose and eluted from the gel by increasing the ionic strength. The results are summarized in Table 11. PAI-1 wild-type and the mutants PAI-1 K28A, K65A, K69A, R186A/R187A, K191A, K207A, and K243A eluted at a concentration between 293 and 318 mM NaC1. No significant difference in the affinity for heparin-Sepharose was detected between the mutants PAI-1 K65A and PAI-1 K69A, being partially defective in heparin-induced inhibition of thrombin, and e.g. wild-type PAI-1. In contrast, the mutants PAI-1 R76A, K80A, and K88A, being fully defective in heparin-induced stimulation of thrombin inhibition as well as in complex formation with the protease, eluted at a considerably lower ionic strength (between 175 and 238 mM NaCl) than wild-type PAI-l.

DISCUSSION
Evidence for the involvement of specific amino acid residues in the interaction between the serpins antithrombin III/ heparin cofactor I1 and heparin originates from three different experimental approaches: (i) chemical modification of specific residues in proteins (Rosenberg and Damus, 1973;Chang, 1989); (ii) analysis of the antithrombin I11 variants of patients suffering from thrombotic episodes caused by defective heparin binding (for review, see Huber and Carrell, 1989) (iii) construction, expression, and analysis of (recombinant) mutant inhibitors (Blinder and Tollefsen, 1990;Whinna et al., 1991). Although a growing volume of data suggests that particular basic amino acid residues are of critical importance for the interaction between heparin and antithrombin I11 or heparin cofactor 11, the individual contribution of specific amino acid residues to the binding of heparin remains to be clarified.
Recently, we showed that PAI-1 interacts with heparin, and we established a system for the construction, expression, and characterization of recombinant PAI-1 and mutants thereof (Ehrlich et al., 1991a). Furthermore, heparin endows PAI-1 with inhibitory properties specifically directed toward the serine protease thrombin (Keijer et al., 1991). Here, we have replaced a number of selected lysine, arginine, and histidine residues by the neutral residue alanine (denoted ala scanning) and subsequently characterized the purified variant proteins to determine the structural requirements of PAI-1 for its interaction with heparin. The following considerations were the basis for our choice to alter particular amino acid residues. First, based on the elucidation of the three-dimensional structure of cleaved al-proteinase inhibitor (Loebermann et al., 1984) and the striking homology of the amino acid sequences of serpins, it is generally accepted that the three-dimensional structure of serpins is similar and that an extrapolation of the location of subdomains is permitted (Huber and Carrell, 1989). Second, the antithrombin 111 of some patients suffering from thrombotic episodes has been shown to contain an alteration of Arg47, located in the helix A subdomain of this serpin. Although not directly proven, it has been suggested that this residue may be involved in heparin binding (Koide et al., 1984;Owen et al., 1987;Borg et al., 1988). In the helix A of PAI-1, there are two positively charged amino acid residues (i.e. His" and Lys"), and these were consequently selected as a target for mutagenesis. Third, as mentioned above, evidence has been presented which demonstrates the involvement in heparin binding of basic amino acids in and around the helix D domain of both antithrombin 111 and heparin cofactor I1 (Chang, 1989;Sun and Chang, 1989;Blinder and Tollefsen, 1990;Whinna et al., 1991). Hence, each basic residue of PAI-1 which corresponds to those of antithrombin I11 and heparin cofactor 11, i.e. L Y S~~, Lys6', Arg76, His77, LysBo, and L y P , has been separately altered into an alanine residue. Fourth, it has been shown that residue LysZ3'j of antithrombin I11 is exposed upon binding of heparin to the inhibitor (Chang, 1989). This observation has been interpreted as evidence for a conformational change of antithrombin I11 after binding to heparin. The corresponding residues of PAI-1, i.e. either Arg'87-Arg188 or Lysl'l, were thus also part of the analysis presented. Fifth, six lysyl residues (LYs'~~, LysZo7, LYS'~~, LyP3, L y P , and Lysm) that were predicted to display either high or low surface probability were separately replaced by an alanine residue and served as controls.
Our data provide direct evidence that the amino acid residues Arg76, Lyssa, and L y P are of critical importance for the interaction between PAI-1 and heparin. Clearly, the mutant proteins PAI-1 R76A, K80A, and K88A have a significantly reduced affinity for heparin inasmuch as they require a lower ionic strength to dissociate from heparin-Sepharose than wild-type PAI-1. Reduced affinity for heparin presumably results in a deficiency of thrombin inhibition and a lack of heparin-induced complex formation between thrombin and the respective PAI-1 mutants. The control experiments, carried out with vitronectin as a cofactor, clearly demonstrate that these variants interact with thrombin in the presence of vitronectin. It is concluded that at least the amino acid residues Arg76, LysBo, and Lys=, located in the helix D subdomain of PAI-1, are part of its heparin binding site. Interestingly, the mutant proteins PAI-1 K65A and PAI-1 K69A are partially defective in thrombin inhibition in the presence of heparin and in SDS-stable complex formation. Nevertheless, under the conditions employed, their affinity for heparin is similar to wild-type PAI-1. It should be noted, however, that each PAI-1 species was allowed to interact with heparin-Sepharose for an extensive period of time to reach equilibrium of binding. Currently, more detailed kinetic experiments are being performed with PAI-1 K65A and PAI-1 K69A to determine whether the association and dissociation constants with heparin differ from those of wild-type PAI-1. Based on our observations on the partial inhibition of thrombin and SDSstable complex formation in the presence of heparin, we tentatively propose that, next to Arg7', Lyssa, and L y P , residues Lys65 and Lys69 are also part of the heparin binding site of PAI-1. The involvement of (at least) 5 basic residues of PAI-1 with a similar number of negatively charged groups on the heparin molecule has been proposed previously to constitute the high affinity heparin binding site on antithrombin I11 (Olson et al., 1981). We conclude that PAI-1 is a member of the heparin-dependent thrombin inhibitors: the location on helix D and the nature of the amino acid residues that constitute the PAI-1 heparin binding site are similar to those of antithrombin I11 and heparin cofactor 11. No evidence can be advanced for the participation of His" and Arg2' in heparin binding, indicating that in contrast to antithrombin I11 and heparin cofactor 11, PAI-1 may not contain a site in the helix A subdomain which would provide the molecule with additional heparin binding properties. Finally, it should be emphasized that none of the mutations grossly affected the general structure of the inhibitor as evidenced by similar specific inhibitory activity toward either t-PA or urokinase-type PA. Moreover, as shown in Figs. 1  and 2, the ability of the mutants to inhibit thrombin in the presence of vitronectin and to form SDS-stable complexes is unaltered. However, we consistently found that the mutant PAI-1 K191A, although fully able to inhibit thrombin and to bind to heparin efficiently, displays significantly reduced SDS-stable complex formation with thrombin in the presence of both heparin and vitronectin. The position of the mutation (Lyslgl replaced by Ala) coincides with the residue of antithrombin I11 which has been shown to become exposed upon binding of antithrombin I11 to heparin and which may play a direct role in the interaction between the heparinbound inhibitor and thrombin (Chang, 1989). The properties Heparin Binding Site on PAI-1 11611 of the mutant PAI-1 K191A would be consistent with normal formation of a Michaelis-Menten ( E I ) complex, whereas the subsequent formation of an SDS-stable complex (so-called EI' complex) could be hampered. Detailed kinetic experiments will provide more insight in the function of Lyslgl of PAI-1 in cofactor-dependent inhibition of thrombin.