Amino Acid Residues in Thrombin-sensitive Region and First Epidermal Growth Factor Domain of Vitamin K-dependent Protein S Determining Specificity of the Activated Protein C Cofactor Function*

Human protein S (PS) potentiates the anticoagulant activity of human but not bovine activated protein C (APC), whereas bovine PS is a cofactor to APC from both species. The structural requirements for the specificity of the APC cofactor function of human PS are located in its thrombin-sensitive region (TSR) and the first epidermal growth factor (EGF1)-like module. To elucidate which residues in these two modules determine the specificity of the APC cofactor activity, 41 human PS mutants were expressed. All mutants were cofactors to human APC and some also to bovine APC. Residues in TSR (positions 49 and 52) and EGF1 (residues 97 and 106) together determined the specificity of the APC cofactor function, whereas substitution of individual residues did not change specificity. Bovine PS, and mutants expressing cofactor activity to bovine APC, stimulated phospholipid binding of bovine APC. In contrast, human PS and mutants lacking cofactor activity to bovine APC failed to support binding of bovine APC to phospholipids. These data indicate that residues in TSR and EGF1 cause the specificity of the APC cofactor activity and support the concept that key residues in these two modules interact with APC on the phospholipid surface.

Protein S (PS) 1 is a vitamin K-dependent plasma protein that functions as an anticoagulant cofactor to activated protein C (APC) in the degradation of the activated forms of coagulation factor V and factor VIII (1,2). The biological importance of PS as an anticoagulant is clearly demonstrated by the massive thrombotic complications that affect infants with homozygous PS deficiency and by an increased risk of venous thrombosis in individuals with heterozygous PS deficiency (3)(4)(5)(6).
The clear association between thrombosis and PS deficiency stands in sharp contrast to the weak APC cofactor function of PS in vitro. Thus, PS was reported to yield only a 2-fold increase in the rate of APC-mediated degradation of factor Va in a purified system (7). However, more recent results suggest that PS expresses rather specific APC cofactor function, because in the degradation of factor Va by APC, PS was found to stimulates only the APC-mediated cleavage at Arg 306 and not the cleavage at Arg 506 (8). That the protein C system indeed is more complicated than previously known became even more evident when factor V was found to function in synergy with PS as cofactor to APC (9,10). Degradation of factors Va/VIIIa thus depends on the assembly of multimolecular complexes on the phospholipid membrane via molecular mechanisms that are still incompletely understood.
PS has a high affinity for negatively charged phospholipid (11), a property that is important for its APC cofactor activity because PS increases the affinity of APC for negatively charged phospholipid vesicles (12), endothelial cells (13), platelets (14,15), and platelet microparticles (16). Recently, the interaction between APC and PS on the phospholipids was found to relocate the active site of APC about 10 Å closer to the membrane surface (17). The PS-induced conformational changes may be important for selectivity of the APC function. PS is also reported to have APC-independent anticoagulant function, which depends on direct interactions of PS with phospholipid membranes as well as with factor Va and factor Xa, and which leads to inhibition of factor X and prothrombin activation (18 -20). However, the biological significance of the APC-independent functions of PS is still uncertain.
In human plasma, approximately 60% of PS is noncovalently complexed to C4b-binding protein (C4BP), a regulator of the classical complement pathway (21). Upon complex formation, PS loses its anticoagulant function (22) but retains its ability to bind negatively charged phospholipids (23).
The primary structures of human, monkey, bovine, rabbit, porcine, mouse, and rat PS have been elucidated (24 -29). Mature PS is a mosaic protein composed of multiple modules. Starting from the NH 2 terminus, it contains a ␥-carboxyglutamic acid (Gla)-rich module, a thrombin-sensitive region (TSR), four EGF-like modules, and a sex hormone-binding globulin-like module. The Gla module binds calcium and negatively charged phospholipids. The integrity of the TSR has been shown to be essential for the APC cofactor activity (30). Probing the structure-function relationships with well characterized monoclonal antibodies has suggested the TSR and EGF1 mod-ules to be important for expression of full APC cofactor activity (31). FabЈ fragments of monoclonal antibodies directed against these two modules completely inhibited the APC cofactor activity, whereas antibodies against other modules had no or very little effect. Additional support for the involvement of EGF1 in the PS-APC interaction has recently been gained from two studies. Thus, a thrombotic patient with type II protein S deficiency (functional defect) was found to carry an intron mutation in the PS gene resulting in exon 5 skipping and as a consequence a truncated PS molecule lacking EGF1 (32). The truncated PS was present at normal concentration in the patient plasma suggesting that the EGF1 deletion does not influence the synthesis of the protein. In the second report, a recombinant PS fragment comprising EGF1-4 was found to inhibit the APC cofactor activity of protein S, whereas an EGF2-4 construct had no such inhibitory activity (33).
The anticoagulant activity of bovine APC is reported to be species-restricted, which has been found to be due to the interaction between APC and PS (22,34). Thus, bovine APC is inefficient as anticoagulant in human plasma unless bovine PS is also present. The interpretation of this is that bovine APC and human PS are unable to form an anticoagulant complex on the phospholipid surface. Human APC on the other hand is found to function both in human and bovine plasma, suggesting human APC to interact with both human and bovine PS (22). The structural determinants in human PS responsible for the species-restriction have been demonstrated to be located in both the TSR and EGF1 modules (35). Thus, human/bovine recombinant PS chimeras containing both these modules of bovine origin were found to function like bovine PS, whereas those chimeras having only one of the two modules of bovine origin did not. With respect to specificity of the APC cofactor function, porcine PS functions like bovine PS, whereas monkey PS is similar to its human counterpart (25). Rabbit PS is not an efficient cofactor to bovine APC but is stimulating the anticoagulant properties of human APC (27). By sequence comparison, a number of amino acid residues have been suggested to play a role for the species restriction. The candidate residues are Arg 49 , Gln 52 , Thr 53 , Gln 61 , Pro 77 , Ser 81 , Ser 92 , Lys 97 , Ser 99 , Thr 103 , and Pro 106 (25,35).
Elucidation of the molecular basis of the specificity of the APC-PS interaction may provide valuable information as to which residues are functionally important in the interaction between APC and PS. The aim of this work was to identify amino acid residues in the TSR and EGF1 regions that are involved in determining the specificity of the APC cofactor function. This was achieved by expressing human recombinant PS mutants in which individual amino acids in the TSR and EGF1 were replaced with amino acid residues present at corresponding positions in bovine PS. The mutations were mapped on a preliminary model structure for the Gla-TSR-EGF1 region of human PS in order to further analyze our results.

EXPERIMENTAL PROCEDURES
Materials-Kits for DNA sequencing and the T7-Gen in vitro mutagenesis were from U. S. Biochemical Corp.; Q-Sepharose Fast Flow was obtained from Amersham Pharmacia Biotech, and hygromycin B was from Calbiochem. Saline of phosphate buffer without calcium and carbonate, Lipofectin, Opti-MEM medium, and Dulbecco's modified Eagle's medium were from Life Technologies, Inc. Polystyrene latex beads were from Sigma. The chromogenic substrate pCa was from American Diagnostica. 1-Palmitoyl-sn-glycero-3-phosphatidylserine and 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphatidylcholine were from Avanti Polar Lipids, Inc. 1-Palmitoyl-2-[1-14 C-oleoyl]PC was from NEN Life Science Products. The human PS deficiency plasma was prepared by incubating normal human plasma with immobilized polyclonal anti-human PS at 4°C overnight as described (22). Bovine PS-deficient plasma was prepared in a similar way but using immobilized polyclonal anti-bovine PS.
Proteins-PS and activated protein C were prepared from human and bovine plasma using methods previously described (31). The monoclonal antibodies that reacted with human PS have been previously characterized (31). cDNA Clones-Full-length human and bovine cDNA clones were previously isolated and characterized (26,36). The original bovine PS cDNA clone was missing one nucleotide, an A in the second EGF-like module. To express wild-type bovine PS, site-directed mutagenesis was used to introduce the missing codon. BamHI sites were introduced at both 5Ј-and 3Ј-end of the coding regions of full-length human and bovine cDNA clones as described previously (35). The numbering of the nucleotides of human and bovine PS cDNA used in this paper is based on sequences which are available in the GenBank/EBI Data Bank having accession numbers M15036 and M13044, respectively.
In Vitro Mutagenesis-Three human/bovine PS chimeras were prepared in which the bovine TSR, EGF1, or both modules were introduced into human PS, i.e. the bovine modules replaced corresponding human modules. Mutant 1 containing the bovine EGF1 module was prepared from the previously described chimeras V (35). This chimeras contained the Gla, TSR, and EGF1 modules of bovine origin with the rest being of human origin. A cleavage site for HincII (position 454) located between TSR and EGF1 in both human and bovine PS cDNA was used to create the mutant 1. BamHI-XbaI (XbaI cleaves at 1481) fragments of both human PS cDNA and chimeras V were cleaved with HincII. The isolated larger HincII-XbaI fragment from chimera V was ligated with the smaller BamHI-HincII fragment from human PS cDNA and the 3Ј XbaI-BamHI fragment to create full-length cDNA in BamHI-cleaved pUC18. This full-length construct, which contained EGF1 of bovine origin in a human PS background, was transferred to the expression vector pGT-h as described below. Mutant 6, which contained both TSR and EGF1 of bovine origin, was constructed in a similar way utilizing an NciI cleavage site which was present in both human (position 372) and bovine (position 261) PS cDNA between the Gla and TSR modules. In short, chimeras V was cut with NciI and BamHI, and the large fragment was isolated and ligated to the small BamHI-NciI fragment obtained from the human cDNA clone. Mutant 5, which contained only the TSR of bovine origin, was made from mutant 6 and human PS cDNA. BamHI-XbaI fragments of both these cDNAs were cut with HincII and the BamHI-HincII fragment from mutant 6 was ligated to the HincII-XbaI fragment from human PS together with the 3Ј-part XbaI-BamHI fragment to create full-length molecules. As a result, the mutant contained the TSR of bovine origin, and the rest was human PS.
Twelve EGF1 mutants and three TSR mutants were prepared. The sequences of the oligonucleotides used in the mutagenesis are given in Table I. Mutants containing multiple mutations were made using combinations of two oligonucleotides. All EGF1 mutations were made using the BamHI-XbaI fragment, whereas a BamHI-HindIII fragment (Hin-dIII cleaves at position 559) was used for the TSR mutations. These fragments of human PS cDNA were subcloned into M13mp18, and single-strand templates were prepared using standard methods (37). Site-directed mutagenesis was accomplished by the "gapped duplex" method (T7-Gen In Vitro Mutagenesis kit from U. S. Biochemical Corp.) (38,39). The M13mp18 clones carrying the PS cDNAs inserts were annealed with each specific oligonucleotide, subjected to the second strand synthesis with T7 DNA polymerase, and ligation with T4 DNA ligase. The DNA was used to transfect competent Escherichia coli SDM (mcrAϪ mcrBϪ) cells. Single-stranded DNA from individual resultant plaques was isolated and sequenced by the dideoxy chain termination method using M13 or PS-specific primers (40). The mutated PS cDNA inserts were isolated from double-stranded phage DNA (HincII-XbaI for EGF1 mutants and BamHI-HincII for TSR mutants), and the appropriate fragments of human PS cDNA were used to construct full-length human PS cDNA in pUC18. With the help of the HincII site between TSR and EGF1, combinations between the TSR and EGF1 mutants were prepared using standard restriction enzyme cleavage and fragment ligation procedures, as outlined above. Mutant cDNAs were isolated after BamHI digestion and subcloned into the BclI site of the expression plasmid pGT-h, which was a kind gift from Dr. B. W. Grinnell (Lilly) (41). The resultant PS cDNA expression plasmids were prepared by CsCl gradient ultracentrifugation (37) and used to transfect 293 cells.
Cell Culture and Expression-The adenovirus-transfected human kidney cell line 293 was grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, 2 mM L-glutamine, 100 units/ml penicillin and streptomycin, 10 g/ml vitamin K1. Transfection was performed using the Lipofectin method (42). DNA (2-4 g) was diluted to 0.1 ml with sterile water. Lipofectin was added (1 g/l), and samples were left at room temperature for 10 -15 min. Cell monolayers (40 -50% confluent in a 5-cm Petri dish) were washed twice in serum-free Opti-MEM medium (Life Technologies, Inc.). The DNA/lipid mixture was diluted to 1 ml in Opti-MEM medium, added to the cells, and incubated overnight (16 -20 h). The cells were fed with 2 ml of complete medium containing 10% calf serum and left to recover for another 48 -72 h. They were then trypsinized and seeded into 10-cm dishes with selection medium (Dulbecco's modified Eagle's medium containing 10% calf serum and 200 g/ml hygromycin B) at 1:5 (43). Hygromycinresistant colonies were obtained after 3-5 weeks of selection, pooled (30 colonies on average), grown to confluency, and the media screened for PS expression with an enzyme-linked immunosorbent assay (ELISA). Conditioned media were collected in the presence of 10 g/ml vitamin K1.
Purification of Recombinant Wild-type and Mutant PS-Tissue culture medium (approximately 2 liters) was filtered through glass wool and diluted with an equal volume of 10 mM Tris-HCl, pH 7.5, containing 10 mM benzamidine HCl and incubated with 50 ml of Q-Sepharose Fast Flow (Amersham Pharmacia Biotech) during gentle stirring at 4°C overnight. The gel was allowed to settle for 1 h and the supernatant discarded. The gel was packed in a column (3 ϫ 50 cm) and washed with 10 mM Tris-HCl, pH 7.5, containing 10 mM benzamidine HCl, and then with the same buffer also containing 50 mM NaCl. Elution was performed with 1.0 M NaCl in the same buffer. The column was run at 4°C at a flow rate of 80 ml/h; 15-ml fractions were collected. Fractions containing PS were identified by electroimmunoassay or ELISA, pooled, and dialyzed against 50 mM Tris-HCl, 0.15 M NaCl, pH 7.5, containing 5 mM benzamidine HCl. After the addition of Ca 2ϩ to a final concentration of 2 mM and incubation for 2 h, the material was applied to an affinity column containing the Ca 2ϩ -dependent antibody HPS-21 (31,44). The column was equilibrated with 50 mM Tris-HCl, 0.15 NaCl, pH 7.5, containing 5 mM benzamidine HCl and 2 mM Ca 2ϩ . After application, the column was washed with the same buffer containing 1.0 M NaCl, and PS was then eluted with 50 mM Tris-HCl, 1.0 M NaCl, 10 mM EDTA, pH 7.5, containing 5 mM benzamidine HCl. Fractions containing PS were pooled and dialyzed against 50 mM Tris-HCl, 0.15 M NaCl, pH 7.5, overnight. After concentration by ultrafiltration using YM 10 filters (Amicon), the proteins were stored at Ϫ70°C.
Enzyme-linked Immunosorbent Assay (ELISA)-Quantification of recombinant wild-type and mutant PS was performed by ELISA using rabbit anti-PS IgG (human or bovine) as catching antibodies and biotinylated monoclonal antibody HPS-21 as detecting antibody. Polyclonal antibodies were used at appropriate dilution in 50 mM Na 2 CO 3 , pH 8.5, to coat microtiter plates (Costar, Cambridge, MA) overnight at 4°C. Plates were washed three times with 50 mM Tris-HCl, 150 mM NaCl, pH 7.5, and incubated with 200 l of 1% BSA in the same buffer for 30 min. This was followed by three washes using 50 mM Tris-HCl, 150 mM NaCl, pH 7.5, containing 0.1% Tween 20. Samples containing PS (diluted in 150 mM NaCl, 20 mM Tris-HCl, 0.1% BSA, pH 7.5, containing 2 mM CaCl 2 ) were incubated in the wells for 2 h at room temperature. After three washes, 100 l of biotinylated HPS-21 (final concentration 1 g/ml) was added and incubated for 1 h at room temperature. After three washes, a mixture of streptavidin and biotinylated horseradish peroxidase (50 l) was added and incubated for 30 min. After three washes, 50 l of 0.04% 1,2-phenylenediamine dihydrochloride and 0.015% hydrogen peroxide in 0.1 M sodium phosphate was added. Hydrolysis of p-nitrophenylphosphate was stopped by the addition of 50 l of 2 N sulfuric acid, and the absorbance was measured at 492 nm using a V max plate reader (Victor, Molecular Devices Corporation, Menlo Park, CA).
Amino Acid Analysis and Protein Sequencing-After SDS-PAGE, proteins were transferred to Immobilon membranes, stained with Coomassie Brilliant Blue, and sequenced in a gas phase sequencer as described by Matsudaira (46). The amino acid composition was determined after acid hydrolysis (6 M HCl, 24 h, 110°C in vacuum) using a Beckman 6300 amino acid analyzer. Gla was measured following alkaline hydrolysis as described (47). All measurements were performed at least in duplicate.
Measurement of APC Cofactor Activity of Different PS Mutants-The APC cofactor activities of recombinant wild-type and mutant PS were characterized in a plasma-based activated partial thromboplastin time (APTT) system. Human PS-deficient plasma (100 l) was incubated with 100 l of APTT reagent (Chromogenix) for 5 min at 37°C, after which 100 l of 25 mM calcium chloride containing human or bovine APC (final concentration in assay of 0.3 g/ml) and wild-type or mutant PS samples (final concentration in assay of 0 -10 g/ml) were added, and the clotting time was measured in an Amelung-Coagulometer KC4A. All dilutions were made in Michaelis buffer (0.036 M sodium acetate, 0.036 M sodium barbitone, 0.14 M NaCl, pH 7.4). Each data point was the mean of three determinations; mutants 7-18 were measured on two occasions each including three individual measurements. Control experiments with bovine PS-deficient plasma were also performed.
Interaction of PS Mutants with Human C4BP-Binding of PS to C4BP was measured in a microtiter plate assay using immobilized human C4BP and detection of bound PS with biotinylated HPS-21. Human C4BP (10 g/ml in 50 mM Na 2 CO 3 , pH 9.6) was incubated overnight at 4°C in polyvinyl microtiter plates. After three washes with 50 mM Tris-HCl, 150 mM NaCl, 0.2% Tween 20, pH 7.5, containing 2 mM CaCl 2 , wells were blocked with 200 l of 1% BSA in the same buffer for 30 min at room temperature. After washing three times using the same buffer, 50 l of different amounts of wild-type or mutant PS samples (0 -10 g/ml) in 50 mM Tris-HCl, 150 mM NaCl, 0.2% Tween 20, pH 7.5, containing 2 mM CaCl 2 were added and incubated overnight at 4°C. Bound PS was detected with HPS-21 as described for the ELISA.
Adsorption of Phospholipid Vesicles to Polystyrene Latex Beads-Phospholipid vesicles composed of 20% phosphatidylserine and 80% phosphatidylcholine (plus a tracer of [ 14 C]phosphatidylcholine) were prepared using the method of Barenholz et al. (48) as modified by Smirnov and Esmon (49). In brief, the lipids were suspended (1 mg/ml) by vortexing in 0.15 M NaCl, 0.02 M Tris-HCl, pH 7.4 (TBS), containing 0.02% sodium azide and sonicated for 20 min in an ice-water bath under argon flow. After sonication, the suspension was centrifuged at 8000 ϫ g for 15 min and passed through a 100-nm nucleopore membrane in an extruder (Avestin) to obtain homogenous vesicles. The vesicles were used immediately or stored at Ϫ80°C. The vesicles were adsorbed to latex beads as described by Smirnov et al. (50). In brief, polystyrene latex beads (100 l) were pelleted in an Eppendorf tube by centrifugation (8000 ϫ g for 1 min), washed three times with TBS, and resuspended in 100 l of TBS containing 5 mM CaCl 2 . Liposomes (100 l) were added and incubated for 2 h at 37°C with shaking. After washing twice with TBS, the beads were suspended in TBS containing 1 mg/ml gelatin, 1 mg/ml ovalbumin, and 10 mg/ml BSA, pH 7.4. The suspension was continuously mixed at room temperature with shaking for 2 h. After washing twice with TBS, the beads were suspended in 1 ml of TBS. The concentration of phospholipid was determined by counting the [ 14 C]phosphatidylcholine tracer which was included in the phospholipid mixture (Beckman model LS 6000 SE scintillation counter). The beads were used immediately or stored at ϩ4°C for up to 3 weeks  Facscan Analysis of Fluorescein-labeled APC Binding to Liposomes-A procedure recently described by Smirnov et al. (50) was followed. In this method, latex beads carrying 0.5 g of phospholipid/ml were suspended in TBS, 2.5 mM CaCl 2 containing 1 mg/ml gelatin, 1 mg/ml ovalbumin, and 10 mg/ml BSA, pH 7.4. Protein S (100 nM final concentration) and fluorescein-labeled bovine APC (between 0 and 2000 nM) were added. The protein/liposome mixtures were incubated at 25°C for 20 min in the dark with gentle mixing. Binding of labeled APC to the phospholipid was measured using a FACScan flow cytometer (Becton Dickinson). Calcium-independent binding of the fluorescent-labeled APC to phospholipid was determined after addition of EDTA (10 mM final concentration). Each experiment was performed twice, and mean values were used for data analysis.
Data Analysis-Binding data were analyzed by fitting the calciumdependent binding parameter to the equation for single site binding model. The K d was calculated using the ENZFITTER program (Elsevier Biosoft, Cambridge, UK). The free concentration of fluorescent labeled APC was calculated as described by Nelsestuen et al. (11).

Comparison of TSR and EGF1 Sequences in Different Species
and Site-directed Mutagenesis Strategy-Bovine PS supports the activity of both human and bovine APC, whereas human PS is restricted in its activity and only expresses activity to human APC. Sequence differences in the TSR and EGF1 regions between human and bovine PS explain this difference in specificity (22,34,35). Species restriction of the APC-PS interaction is not only expressed by the human-bovine pair but also by other species combinations. Thus, monkey PS (mPS) is functionally similar to hPS (neither of them stimulate bovine APC), whereas porcine PS (pPS) is similar to bPS (stimulates both human and bovine APC) (25). Comparison of PS amino acid sequences from these four species (human, monkey, bovine, and porcine) reveals a limited number of amino acid differences in the TSR and EGF1 which possibly are involved in the species restriction of the PS function (Fig. 1). In the TSR, there are three amino acid residues that are identical in hPS and mPS and different from the bovine/porcine PS sequences. Position 49 in hPS and mPS is occupied by Arg, whereas bPS and pPS have Gly; at position 52, hPS and mPS have Gln, whereas bPS and pPS have Arg; at position 61, hPS and mPS have Gln, whereas Leu is found in both bPS and pPS. In the EGF1 module, the following six positions have a consensus sequence for the human/monkey pair that is different from that of the bovine/ porcine one: Ser 81 3 Asn, Tyr 90 3 Phe, Ser 92 3 Thr, Lys 97 3 Gln, Thr 103 3 Ile, and Pro 106 3 Ser. In addition, at positions 77 and 99 there are some differences between the four sequences which, however, do not obey the human/monkey and bovine/ porcine pairing rule. Thus, at position 77, Pro is found in hPS, mPS, and pPS, whereas bPS has Ser, and at position 99 hPS and mPS have Ser, and bPS has Thr, and pPS has Met.
Expression and Characterization of Recombinant Wild-type and Mutant PS-In addition to the discussed four primary structures, amino acid sequences of rabbit, rat, and mouse PS are known, and the sequences of the TSR and EGF1 of these species are also given in Fig. 1. It has been demonstrated that rabbit PS potentiates the anticoagulant activity of human APC, whereas rabbit PS is a poor cofactor to bovine APC. In this context it is interesting to note that the rabbit sequence is identical to the bovine/porcine pair at positions 49, 52, 53, 81, 97, and 103. However, it is identical to the human/monkey pair in having a Pro at position 106. Moreover, it is identical to hPS, mPS, and pPS in having a Pro at position 77, a position at which there is a Ser in bPS. In addition, rabbit PS is different from both human and bovine PS at five positions in the TSR (positions 54, 59, 61, 63, and 65) and four in the EGF1 (positions 86, 88, 91, and 92). The rat and mouse sequences are not discussed in detail as they have not been fully characterized with respect to their cofactor activity to human and bovine APC.
Unique restriction enzyme sites located between the Gla-TSR and TSR-EGF1 modules were used to construct human/ FIG. 1. Comparison of the TSR and EGF1 amino acid sequences of protein S. A, schematic diagram of amino acid sequences of the human TSR and EGF1 modules. The shaded residues indicate amino acid residues that are different in human and bovine PS, and the arrows point to the amino acid residues present in bovine PS. B, alignment of the amino acid sequences (TSR and EGF1) from human, monkey, bovine, porcine, rabbit, rat, and mouse origins. Asterisks denote positions that are different between human/monkey and bovine/porcine. bovine chimeras for TSR and EGF1 individually, as well as for the two in combination. In addition, site-directed mutagenesis was used to create a number of mutants (Table II)  The mutant recombinant proteins were expressed as stable cell lines in human 293 cells as described previously for the human/bovine chimeras (35). The level of PS expression was estimated with an ELISA and found to vary between 0.6 and 1.2 g/10 6 cells/24 h. The mutants were purified and characterized. In unreduced SDS-PAGE, the proteins migrated as single chain bands, like the wild-type human and bovine PS and the respective plasma-derived proteins (not shown). As previously noted, the human PS and the mutants had slightly higher molecular weight than the bovine PS, which was due to differences in number of carbohydrate side chains in human (three carbohydrate sites) and bovine (two carbohydrate sites) PS (24,26,52). After reduction, the human and bovine wildtype proteins and some of the recombinants migrated as closely spaced doublets, just like the human and bovine plasma-derived PS preparations. The purified wild-type and mutant recombinant PS were fully post-translationally modified as judged by their Gla and Hya/Hyn residue contents, which were found to be similar to those of the plasma-derived proteins (results not shown). Human and bovine plasma-derived PS contained 2.4 and 3.9 mol of Hya/Hyn per mol of protein, respectively. Recombinant PS mutants yielded values close to that of plasma-derived human PS suggesting that the mutations in the TSR and EGF1 modules in human PS did not affect the hydroxylation reaction. Amino-terminal protein sequence analysis yielded the expected sequences of recombinant wildtype and mutant PS (data not shown). All recombinant PS variants were found to bind to immobilized human C4BP with high affinity (data not shown), half-maximum binding occurred at 5-20 nM PS, suggesting the different recombinant proteins to be correctly folded.
APC Cofactor Activity of Recombinant Wild-type and Mutant PS-The APC cofactor activity of the recombinant proteins was tested in an APTT-based assay using human or bovine PSdeficient plasma and human or bovine APC. The use of bovine and human PS-deficient plasma yielded similar results demonstrating that the species of factor V(a) did not influence the results. Only the results obtained with human PS-deficient plasma will be shown. In the absence of added PS, both human and bovine APC yielded approximately 10 s prolongation of the clotting time. In the absence of APC, the addition of protein S did not prolong the clotting time. In the presence of APC, wild-type human and bovine PS (and the plasma-derived proteins) yielded results in accordance with those on record, i.e. both human and bovine PS potentiated the anticoagulant activity of human APC, whereas only bovine PS functioned as cofactor to bovine APC. In this system, bovine PS was found to be more potent than its human counterpart even in the presence of human APC. The plasma-derived human and bovine PS gave results similar to those of their recombinant counterparts (results not shown).
The APC cofactor functions of the 41 mutants as well as the wild-type human and bovine recombinant PS and their plasmaderived counterparts were tested in the APTT-based assay at seven different PS concentrations (0 -10 g/ml final concentration) using both human and bovine APC. Fig. 2 illustrates the activities of some of the mutants. In order to extract the relevant information from the resulting 630 data points and simplify the interpretation, the following approach was adopted. The results obtained with the wild-type human PS were used as reference in regression analysis of the clotting results obtained with the different mutants. The slopes of the regression equations indicated the relative potency of the different mutants (i.e. their activities were in this way related to that of the human wild-type PS). Thus, PS mutants with potent APC cofactor activity resulted in steep slopes with high slope coefficients. The regressions slope coefficients obtained using both human and bovine APC are plotted in histogram form in Fig. 3.
Mutant 1, in which the whole EGF1 was of bovine origin, was slightly more active than wt-hPS both in the presence of human and bovine APC, but just like hPS it was a poor cofactor to bovine APC. The same was true for mutant 5 which had its TSR of bovine origin. When TSR and EGF1 were both of bovine origin (mutant 6), the recombinant product behaved like wt-bPS and functioned as a potent cofactor to both human and bovine APC. Like wt-bPS, this mutant expressed higher cofactor activity to human APC than did wt-hPS, suggesting sequence differences between the human and bovine TSR and EGF1 modules to be the cause for this difference.
To elucidate which amino acid residues were important in the TSR for the species restriction of hPS, different TSR mutants were introduced in a recombinant human PS having its EGF1 of bovine origin (mutants 2-4 and 6). Individual replacements of Gln 52 with Arg (mutant 3) or Gln 61 with Leu (mutant 4) did not lead to enhanced activity to bovine APC. However, mutant carrying the two replacements R49G and Q52R was functionally similar to bovine APC and expressed as high cofactor activity to bovine and human APC as did wt-bPS.
To determine which amino acid residues in the EGF1 module were important for the species restriction of hPS, three different groups of EGF1 mutants carrying different mutations in the TSR were expressed and characterized. In the first TSR group (GRA؊), the TSR had the R49G, Q52R, and T53A replacements; the second group (-L) had only the Q61L replacement; and the third group had the full bovine TSR (GRAL) with all four replacements (R49G, Q52R, T53A, and Q61L). All the recombinant mutants of the -L group behaved like hPS and expressed low activity against bovine APC, even though there were some minor differences in activity between the different EGF1 mutants. Some were distinctly more active than wild-type human PS. The results obtained with the GRAand GRAL groups were qualitatively similar although the GRAL group mutants were consistently more potent than corresponding mutants in the GRA-group. The presentation will focus on the GRAL group, but the conclusions are also valid for the GRA-group. In the GRA-and GRAL groups, three individual replacements in the EGF1 were expressed, the S81N (mutants 16 and 39), the K97Q (mutants 7 and 30), and the P106S (mutants 8 and 31). The S81N replacement did not affect the species restriction, whereas both the K97Q and the P106S replacements resulted in molecules functioning like bovine PS. In particular the P106S replacement in the GRAgroup was noteworthy, because the introduction of these few amino acid replacements in the human PS molecule led to a dramatic change in PS functional activity. Not only was the APC cofactor specificity changed from human to bovine type, but, in addition, the resulting molecule was distinctly more active to human APC than the wt-human PS, approximately 200% activity as compared with the wt-human PS (Fig. 3). Addition of the P77S (mutants 18 and 41) replacement or the S81N (mutants 17 and 40) replacement to the P106S containing mutants did not increase the APC cofactor activity further. Likewise, addition of the S92T replacement to the K97Q mutant (mutants 9 and 32) did not increase the cofactor activity to bovine APC further, whereas when the S99T replacement was introduced into the K97Q mutant (mutants 10 and 33) this led to an increased activity of the recombinant proteins up to the level of wt-bPS. The combination of S92T, K97Q, T103I, and P106S replacements (mutants 11 and 34) was not better than  Table II and the mutations in TSR and EGF1 are indicated by the one-letter amino acid residue code. the K97Q replacement alone and even somewhat less active than the P106S mutants. Among the most potent recombinants were mutants 13-14 and 36 -37 which expressed potent APC cofactor activity to both bovine and human APC. In common, these PS mutants had the S92T, K97Q, T103I, and P106S replacements and in addition either the S81N (mutants 13 and 36) or the S99T (mutants 14 and 37) replacements. The most potent mutants (15 and 38) "super PS" had all six replacements (S81N, S92T, K97Q, S99T, T103I, and P106S).
From the presented results it can be concluded that the species restriction of APC cofactor activity depends on a combination of amino acid replacements in both the TSR and the EGF1 and that a limited number of amino acid replacements leads to a change in specificity. As the experimental approach was very time consuming and laborious, we were unable to test a large number of possible mutations or other combination of replacements, e.g. the individual replacements S92T, Y90F, S99T, and T103I were not tested.
Increased Binding of Bovine APC to Phospholipid in the Presence of Bovine Protein S and Some Human Protein S Mutants-As described above, bovine protein S and several of the human protein S mutants expressed cofactor activity to bovine APC. A likely explanation for this phenomenon was that these protein S variants stimulate binding of bovine APC to the phospholipid. To investigate this possibility, binding of bovine APC (fluorescein-labeled in its active site) to phospholipid vesicles was measured using FACScan flow cytometry. In the absence of protein S, binding of bovine APC to the vesicles was characterized by a K d of 626 nM (Table III). Bovine protein S yielded a 3.5-fold stimulation of the binding of bovine APC (K d was 177 nM). In contrast, human protein S did not affect the binding of bovine APC. Nine of the mutants were tested for their ability to stimulate binding of bovine APC to the phospholipid. Mutant 1 (human TSR ϩ bovine EGF1) did not stimulate binding of bovine APC. In contrast, mutant 2 (bovine EGF1 and the double mutant R49Q,Q52R in the TSR) yielded a 3-fold stimulation of bovine APC binding. The individual Q52R and Q61L mutations in TSR combined with bovine EGF1 (mutants 3 and 4) were not sufficient to stimulate binding of bovine APC. In mutants 9, 10, and 11, the TSR carried the triple mutation R49Q,Q52R,T53A (GRA-) in combination with point mutations at one or more of residues at positions 92, 97, 99, 103, and 106. These mutants all stimulated the binding of bovine APC. The mutant 35 yielded a 1.9-fold increase in bovine APC binding, which was somewhat lower than expected.
The results on record, i.e. those reported here taken together with results obtained with well characterized monoclonal an-tibodies (31), with human/bovine PS chimeras (35), with the thrombosis patient having the protein S lacking EGF1 (32), and with the recombinant protein S fragments containing EGF1 (33) suggest the TSR and EGF1 to be directly involved in the interaction between APC and PS and that mutations in these two domains of human PS lead to an increased interaction with bovine APC. It should also be stated that none of the so far reported data exclude the possibility that other modules in the PS molecule are crucially important for the expression of APC cofactor activity. In this context it is noteworthy that we recently found the sex hormone-binding globulin module of protein S to be required for the synergistic APC cofactor activity between protein S and factor V in the degradation of factor VIIIa (9,10,53).
Structural Interpretation of Mutations Using a Model for the Gla-TSR-EGF1 Region-Structural studies of PS using x-ray crystallography or NMR have so far been unsuccessful. However, it is possible to use some of the tools provided by computational biochemistry research to investigate a protein of unknown structure. We have used comparative model building and structural prediction analysis to develop a model for the Gla-TSR-EGF1 region of PS (54). Recently, the model has been refined based on experimental data, including monoclonal antibody epitope mapping (55), new information regarding sensitivity to proteolytic enzymes (56), and the description of a thrombosis patient carrying a truncated protein S molecule lacking EGF1 (32). These data support the conclusion that Arg 49 , Gln 52 , and Arg 60 are solvent-exposed because Arg 49 and Gln 52 are included in the epitope of the calcium-dependent monoclonal antibody HPS67, whereas factor Xa cleaves the TSR at Arg 60 (55,56). Lys 97 is part of an epitope for HPS54 suggesting also this residue to be solvent-exposed (55). Moreover, as deletion of EGF1 from the protein S molecule does not affect the synthesis or stability of the protein, we suggest that EGF1 has no specific or very limited contact with the Gla, TSR, and (possibly) EGF2 domains.
The structural analysis will focus mainly on residues 49, 52, 61, 81, 90, 92, 97, 103, and 106. Arg 49 is part of a short helical segment in the model structure and could be hydrogen-bonded to Gln 52 (Fig. 4). Arg 49 is fully accessible to the solvent. No acidic amino acids are noted nearby this arginine, and the residue does not appear to be involved in a salt bridge. From the modeling study and the experimental data it seems reasonable to suggest that residues 49 and 52 offer their side chains for macromolecular interaction with APC. A Gly at position 49 (as in bovine PS) may induce local flexibility. The Q52R replacement (e.g. mutant 3) may bring together two positively charged residues, but because they are solvent-exposed such mutation should not alter the overall structure of the TSR. The fact that mutant 2 (R49G,Q52R) modifies the cofactor activity of PS indicates that the location of the positively charged residue is critical for the interaction with APC (the positive residue could form a salt bridge with APC).
The T53A replacement is difficult to evaluate from a structural standpoint. Both Thr and Ala have short side chains with Thr displaying a hydroxyl group which could be involved in hydrogen bonding with APC. The Thr 53 side chain points toward the solvent and Arg 49 -Gln 52 area (Fig. 4) and thus could have contact with APC.
At the expected tip of the TSR, human PS displays Gln 61 , whereas bovine PS has Leu (Fig. 4). This region is fully solventaccessible, and because it contains several polar and short residues, we suggest that this part of the TSR forms a loose turn. Residue 61 is likely on the other side of the TSR when compared with the Arg 49 -Gln 52 area (Fig. 4). The Gln to Leu replacement does not seem essential for the specificity of the APC interaction as seen from the analysis of mutant 4. This suggests that the Gln 61 TSR "face" does not directly contact protein C. This is supported by the second set of mutants (-L) that behaved like human PS. Moreover, human and bovine PS have the same residues in this region from Ser 62 to Ile 76 . It would be expected that to display species-specific interaction; evolutionary pressure would have introduced several point mutations on this side of the loop. The fact that the GRAL group was more potent than the GRA-suggests that residue 61 could have an indirect effect in the interaction with APC. In EGF1, Ser 81 is solvent-accessible and can be easily replaced by Asn without inducing structural strains. This residue does not seem to play an important role in the interaction with APC (mutants 17 and 39). It is located (Fig. 4) on the EGF1 side opposite from the cluster of residues that produce important changes in the specificity (Lys 97 , Ser 99 , and Pro 106 , see below).
Pro 77 is several Angstroms below Lys 97 (Fig. 4). The data from mutants 18 and 41 suggest that Pro 77 is not important for the APC-PS interaction. This is confirmed when looking at the model and the results from mutants 17 and 40 which contain the S81N mutation. Ser 81 is on the same side as Pro 77 and on the other face as our expected binding site area (Fig. 4).
Ser 92 is relatively close to residue Pro 106 , and therefore the Ser 92 region could contact protein C (Fig. 4). Less conservative amino acid substitution than a Ser to Thr would be needed to test this hypothesis. The K97Q replacement (e.g. mutants 7 and 30) seems important for the specificity of the APC-PS interaction. This residue is solvent-accessible in the model structure and is located in the loop connecting the EGF's major ␤-sheet (Fig. 4). Residue 97 could thus have direct contact with APC.
The mutation S99T (mutants 10 and 33) further supports that the region of residue 97 is important for the interaction with APC since residues 97 and 99 are very close in space (Fig. 4).
The replacement P106S is important for the specificity of the APC-PS interaction. This residue is located in a loop, at the opposite end of the EGF1 module when compared with residue 97 but on the same face (Fig. 4). In this region, several surrounding side chains may be involved in the interaction with APC. These involve Lys 105 , Tyr 90 , Met 91 , Thr 103 , and Glu 109 . Such solvent-exposed charged and hydrophobic/aromatic residues can in fact offer a suitable surface area for protein-protein interaction. Moreover, the T103N substitution, leading to a type II deficiency (57), would suggest that this region is of importance for macromolecular interaction since this replacement can be easily accommodated in the structure. Moreover, FIG. 4. Three-dimensional model for the Gla-TSR-EGF1 region of protein S. The Gla module (residues 1-46), with the ⍀-loop (green, residues 1-11, important for membrane binding), the disulfide loop (pink, Cys 17 -Cys 22 ), and the helix forming part of the aromatic stack (red) are shown using a ribbon representation. The remaining part of the Gla module is in yellow. The TSR (white) is followed by the EGF1 (magenta). Some hydrophobic and ionic interactions between the Gla module and the TSR are expected at least in a calcium-loaded conformation (not shown to simplify the figure). The EGF-1 ␤-sheets are colored orange (major ␤-sheet) and green (minor ␤-sheet). The seven calcium ions are magenta and disulfide bridges blue (ball and stick representation). In EGF1, the disulfide bridges involve Cys 80 -Cys 93 , Cys 85 -Cys 102 , and Cys 104 -Cys 113 . The side chains of some important residues discussed in this paper are shown as stick model, with a color code not related to the one used for the ribbon trace. The residues found essential for the species-specific interaction by our site-directed mutagenesis study are Arg 49 , Gln 52 , Ser 99 , Lys 97 , and Pro 106 . Inset, the model has been rotated when compared with the main figure in order to see the molecule from the opposite face. The color code for the ribbon trace is unchanged. Only a few key residues are labeled to simplify the figure. The cell membrane is drawn in white, and the short arrow indicates that the orientation of the membrane plane is approximate. Clearly, the areas Arg 49 , Lys 97 , and Pro 106 could form one APCinteracting site, whereas Gln 61 and Ser 81 are definitely on opposite faces. The overall structure of the model is in good agreement with the experimental data. bovine PS has an isoleucine there, thus an extra surface-exposed hydrophobic residue, consistent with a possible binding site.
The data obtained from the above mutations suggest that APC could wrap around PS, making contact with the area 49 -52 and residues 97 and 106. Site-directed mutagenesis experiments on the protein C molecule are needed to enhance our understanding of this very complex biological system.