Inhibition of Six Serine Proteinases of the Human Coagulation System by Mutants of Bovine Pancreatic Trypsin Inhibitor*

A series of 12 bovine pancreatic trypsin inhibitor variants mutated in the P4 and P3 positions of the canonical binding loop containing additional K15R and M52L mutations were used to probe the role of single amino acid substitutions on binding to bovine trypsin and to the following human proteinases involved in blood clotting: plasmin, plasma kallikrein, factors Xa and XIIa, thrombin, and protein C. The mutants were expressed inEscherichia coli as fusion proteins with the LE1413 hydrophobic polypeptide and purified from inclusion bodies; these steps were followed by CNBr cleavage and oxidative refolding. The mutants inhibited the blood-clotting proteinases with association constants in the range of 103–1010 m − 1. Inhibition of plasma kallikrein, factors Xa and XIIa, thrombin, and protein C could be improved by up to 2 orders of magnitude by the K15R substitution. The highest increase in the association constant for P3 mutant was measured for factor XIIa; P13S substitution increased the K a value 58-fold. Several other substitutions at P3 resulted in about 10-fold increase for factor Xa, thrombin, and protein C. The cumulative P3 and P1 effects onK a values for the strongest mutant compared with the wild type bovine pancreatic trypsin inhibitor were in the range of 2.2- (plasmin) to 4,000-fold (factors XIIa and Xa). The substitutions at the P4 site always caused negative effects (a decrease in the range from over 1,000- to 1.3-fold) on binding to all studied enzymes, including trypsin. Thermal stability studies showed a very large decrease of the denaturation temperature (about 22 °C) for all P4 mutants, suggesting that substitution of the wild type Gly-12 residue leads to a change in the binding loop conformation manifesting itself in non-optimal binding to the proteinase active site.

A series of 12 bovine pancreatic trypsin inhibitor variants mutated in the P 4 and P 3 positions of the canonical binding loop containing additional K15R and M52L mutations were used to probe the role of single amino acid substitutions on binding to bovine trypsin and to the following human proteinases involved in blood clotting: plasmin, plasma kallikrein, factors X a and XII a , thrombin, and protein C. The mutants were expressed in Escherichia coli as fusion proteins with the LE1413 hydrophobic polypeptide and purified from inclusion bodies; these steps were followed by CNBr cleavage and oxidative refolding. The mutants inhibited the blood-clotting proteinases with association constants in the range of 10 3 -10 10 M ؊1 . Inhibition of plasma kallikrein, factors X a and XII a , thrombin, and protein C could be improved by up to 2 orders of magnitude by the K15R substitution. The highest increase in the association constant for P 3 mutant was measured for factor XII a ; P13S substitution increased the K a value 58-fold. Several other substitutions at P 3 resulted in about 10-fold increase for factor X a , thrombin, and protein C. The cumulative P 3 and P 1 effects on K a values for the strongest mutant compared with the wild type bovine pancreatic trypsin inhibitor were in the range of 2.2-(plasmin) to 4,000-fold (factors XII a and X a ). The substitutions at the P 4 site always caused negative effects (a decrease in the range from over 1,000-to 1.3-fold) on binding to all studied enzymes, including trypsin. Thermal stability studies showed a very large decrease of the denaturation temperature (about 22°C) for all P 4 mutants, suggesting that substitution of the wild type Gly-12 residue leads to a change in the binding loop conformation manifesting itself in non-optimal binding to the proteinase active site.
Blood coagulation is a series of proteolytic events resulting in clot formation. Equally important are the processes of anticoagulation and fibrinolysis that are also mediated by proteolytic enzymes. Research during the past decade has resulted in the determination of spatial structures for many of these enzymes, including thrombin, factors VII a , IX a , and X a , protein C, tissue plasminogen activator, and plasmin (1)(2)(3)(4)(5)(6). The specificity of a particular enzyme toward its cognate sequence results from well defined subsites on the enzyme surface recognizing not only the scissile peptide bond but often also more extended regions. These structural studies nicely explain earlier data on sequence-specific cleavage of natural substrates and provide a framework for drug design efforts.
Because the coagulation/fibrinolysis processes are of vital importance, they are precisely controlled. Protein inhibitors are one of the most essential regulating factors. Interestingly, the scaffold of Kunitz-type inhibitors is used both to control the natural coagulation process in human blood and to prevent blood clotting in a blood-sucking organism, the soft tick. In the former case, tissue factor pathway inhibitor (TFPI) 1 composed of three tandemly arranged Kunitz-type domains inhibits different clotting proteinases (e.g. factor X a at the second domain) through the canonical proteinase binding loop. On the contrary, the amino terminus of tick anticoagulant peptide, which also possesses a Kunitz-domain fold, is involved in non-canonical binding to the active site of factor X a (7). Both TFPI and tick anticoagulant peptide were recognized as potential antithrombotic drugs (8,9).
Bovine pancreatic trypsin inhibitor (BPTI) is the best known example of Kunitz-type inhibitors. It consists of 58 amino acid residues cross-linked by three disulfide bridges that contribute to a compact tertiary structure and remarkable thermodynamic stability (10,11). Wild type BPTI is a powerful competitive inhibitor of several serine proteinases including trypsin, chymotrypsin, plasmin, and tissue kallikrein (12). BPTI is not a strong immunogen because specific antibodies have not been observed in humans (12). The wild type protein has been used as an antifibrinolytic agent in cardiac surgery (13) and in acute pancreatitis (12). Because of its low immunogenicity and well known mechanism of action, BPTI appears to be an ideal candidate for protein engineering studies with the aim of directing its specificity toward blocking unbalanced proteolysis. So far, several studies have shown that through semisynthetic (14,15) or recombinant approaches (16 -18) it is possible to change or enhance the BPTI inhibition spectrum. Additionally, phage display technology has been applied to generate potent inhibitors directed toward different serine proteinases based on either the BPTI sequence (19) or sequences of homologous inhibitors: amyloid ␤-protein precursor (20 -22) and the first domain of TFPI (23,24). Stassen et al. converted BPTI into a potent inhibitor of factors VII a -TF, X a , XI a and plasma kallikrein through grafting into BPTI binding loop sequences of individual TFPI domains (17).
In this paper 12 mutants of BPTI located in the P 3 and P 4 positions (nomenclature according to Schechter and Berger (25)) of the binding loop segment were constructed with the aim of improving binding to several proteinases of the human blood clotting system: ␣-plasmin, plasma kallikrein, factor X a , ␣-factor XII a , protein C, and ␣-thrombin. In all the mutants the wild type Lys-15 was replaced with Arg because of its more favorable binding to the studied enzymes. Although the inhibitor contacts different serine proteinases using about seven amino acid residues of the binding loop (26), it was decided to mutate just the P 1 , P 3 , and P 4 positions. Cys at P 2 forms a disulfide bond with Cys-38. It is generally recognized that the unprimed side of the binding loop is energetically more essential than the prime side (27). In addition, much less substrate kinetic data is available for the prime side residues. Furthermore, our data on seven P 1 Ј mutants showed that in BPTI-serine proteinase complexes the P 1 Ј side chain sterically conflicts with the side chain of Ile (P 3 Ј) leading to a large decrease in the association energy (28).

EXPERIMENTAL PROCEDURES
Materials-Guanidinium chloride, urea, dimethyl sulfoxide (Me 2 SO), N,N-dimethylformamide, methanol, and acetonitrile were purchased from Merck. Trifluoroacetic acid and cyanogen bromide were from Fluka. Tris, sodium acetate, dithiothreitol, chloramphenicol, ampicilin, and reduced and oxidized glutathione were from Sigma. Merck supplied the basic components of culture media. DNA-modifying enzymes (T4 polymerase, T4 DNA ligase, T4 polynucleotide kinase) were purchased from Roche Molecular Biochemicals. A DNA sequencing kit from Amersham Pharmacia Biotech and a DNA purification kit from Qiagen were used. Oligonucleotides were chemically synthesized by Ransom Hill. Isopropyl-␤-D-thiogalactopyranoside and oligopeptide substrate Suc-Ala-Ala-Pro-Arg-pNA were from Bachem. The other substrates: Bz-Arg-pNA, Bz-Pro-Phe-Arg-pNA, and Tos-Gly-Pro-Arg-pNA were from Sigma; H-D-Pro-Phe-Arg-pNA (S-2302) was supplied by Chromogenix. 4-nitrophenyl 4Ј-guanidinobenzoate (NPGB) was from Merck. The human proteinases ␣-plasmin, plasma kallikrein, ␣-factor XII a , and ␣-thrombin were from Kordia and factor X a was supplied by Roche Molecular Biochemicals. Recombinant human protein C was purified from the milk of transgenic swine and was a generous gift from Dr. H. Lubon (American Red Cross) (29). Bovine ␤-trypsin was supplied by Worthington Biochemical.
Expression and Purification of Mutant Proteins-All the mutants of BPTI were overexpressed as fusion proteins in Escherichia coli strain BL21 (DE3) pLysS, using the T7 promotor system (30) as described (31). The plasmids, derived from pAED4, bearing a portion of the E. coli trp operon, which serves as a leader sequence, followed by a Met residue and the mutant BPTI encoding sequences, were prepared by site-directed mutagenesis (32). All recombinant variants contained the additional mutation M52L to allow CNBr cleavage of the fusion protein.
Details of the purification and oxidative refolding protocols have been published elsewhere (33,34). Electrospray mass spectrometry of all variants used in this study was performed with a Finnigan MAT TSQ-700. Samples were dissolved in a methanol:water:acetic acid mixture (50:49:1).
Thermal Transitions-Thermal denaturation was monitored by following the ellipticity at 223 nm with continuous monitoring of the thermal transition at a constant rate of 1°C/min. The temperature of the sample was monitored directly using a probe immersed in a cuvette and was controlled with PFD. A band slit of 2 nm and a response time of 4 s was applied. Automatic Peltier accessory PFD 350S allowed 350S/350L Peltier type FDCD attachment. Protein was dissolved in 10 mM glycine-HCl, pH 2.0 and passed through a 0.22-m Millipore filter before measurements. The data were analyzed assuming a 2-state reversible equilibrium transition, where T is the absolute temperature in K, R is the gas constant (1.98 cal/mol K), (T) is the ellipticity signal at 223 nm, ⌬H vH is van't Hoff enthalpy, F is the value of the folded signal extrapolated to 0 K, m F is the slope of the temperature dependence of the CD signal for the folded protein, U is the value of the unfolded signal extrapolated to 0 K, m U is the slope of the temperature dependence of the CD signal for the unfolded protein, and T den is the denaturation temperature. Standardization of Enzymes, Inhibitors, and Substrates-Human plasmin was dissolved in 1 mM HCl containing 20 mM CaCl 2 . Plasmin concentration was determined by titration with wild type BPTI, which in turn was standardized with NPGB-titrated trypsin (35). Plasma kallikrein and thrombin were dissolved in MilliQ water. The concentrations of both enzymes were determined using respective molar absorbance coefficients ⑀ 280 ϭ 93,280 M Ϫ1 cm Ϫ1 and 24,240 M Ϫ1 cm Ϫ1 (36). Protein C was dissolved in 50 mM Tris-HCl, 50 mM NaCl, 7.5 mM CaCl 2 , pH 7.4, and its concentration was determined from the 56,640 M Ϫ1 cm Ϫ1 molar absorbance coefficient at 280 nm (37). The concentration of factor X a was calculated from the Briggs-Haldane equation applying published values of catalytic parameters k cat and K m for the Tos-Gly-Pro-Arg-pNA substrate (38). The concentration of factor XII a was determined from kinetic parameters k cat ϭ 15 s Ϫ1 and K m ϭ 1.9 ϫ 10 Ϫ4 M for H-D-Pro-Phe-Arg-pNA (39). The following substrates were used to measure the residual enzyme activities: Tos-Gly-Pro-Arg-pNA (bovine trypsin), H-D-Pro-Phe-Arg-pNA (␣-plasmin), Tos-Gly-Pro-Arg-pNA (factor X a ), H-D-Pro-Phe-Arg-pNA (␣-factor XII a ), Bz-Pro-Phe-Arg-pNA (plasma kallikrein), Suc-Ala-Ala-Pro-Arg-pNA (␣-thrombin), and Tos-Gly-Pro-Arg-pNA (protein C). All inhibitor concentrations were determined by titration with NPGB-standardized trypsin.
Determination of Association Constants-The association constant (K a ) values were determined by a modified method of Green and Work, as described by Empie and Laskowski (40) and Otlewski and Zbyryt (41). All measurements were performed in 100 mM Tris-Cl, 150 mM NaCl, 20 mM CaCl 2 , 0.05% Triton X-100, pH 8.3. Although pH 8.3 is above the optimum pH for most of the proteinases used, at this pH the K a values are not perturbed by the pK of His-57, and it is a reference pH routinely used in our laboratory. To determine the association constant, increasing amounts of the inhibitor were added to a constant concentration of the proteinase. K a determinations were done singly; for each K a value 7-10 data points were measured. The enzyme concentration was chosen to fulfil the condition: 1 Ͻ [E 0 ] ϫ K a Ͻ 50. For the determination of low K a values (10 3 -10 7 M Ϫ1 ), the inhibitors were used at much higher concentrations than the enzyme, to force complex formation. After a suitable incubation time the residual enzyme activity was measured for about 500 s by monitoring the linear release of p-nitroanilide with an HP 8452A diode array spectrophotometer. The recorded signal was the average value of absorbance in the 380 -410-nm range corrected for the background average absorbance in the 480 -510-nm range. The value of K a was calculated by a 3-parameter algorithm, (E) ϭ f(E 0, K a , F), using the non-linear regression analysis program GraFit, according to the equation, where [E 0 ] and [I 0 ] are the total enzyme and inhibitor concentrations, respectively, [E] is the residual enzyme concentration, and F is the enzyme-inhibitor equimolarity factor. In the case of weak associations (K a Ͻ 10 7 M Ϫ1 ) only the 2-parameter algorithm (E) ϭ f(E 0, K a ) and the default value of F ϭ 1 were applied.

RESULTS
The BPTI molecule contains a solvent-exposed proteinase binding loop spanning the residues P 3 through P 3 Ј (Fig. 1). Thirteen mutations were installed in this segment using sitedirected mutagenesis. All the mutants were expressed in E. coli with an average yield of 5 mg of pure protein per 1 liter of culture. Mutant proteins were purified on a C 18 Vydac HPLC column in both their reduced and oxidized states. Experimental molecular masses determined by electrospray spectrometry of all variants were within 1.0 atomic mass unit of the expected values.
Stability of the Mutants-All the mutant proteins showed similar CD spectra both in the near-and far-UV range, so no major conformational changes occurred upon installing the amino acid substitutions (data not shown). Furthermore, the thermodynamic stability of the mutants was studied by thermal denaturation monitored by a 223-nm CD signal. All transitions were well reversible, as judged by the similarity of their CD spectra recorded at 20°C before and after the experiment. Differential scanning calorimetry experiments showed a ⌬H cal to ⌬H vH ratio close to unity, which is indicative of a 2-state transition (data not shown). Fig. 2 shows an example of denaturation curves for the pseudo wild type BPTI (K15R, M52L) and its two P 3 and P 4 mutants: P13A and G12V. Table I summarizes T den and ⌬H vH values determined for all the mutants at pH 2.0. It can easily be noticed that both the K15R and M52L mutations, which were introduced into all the mutants, did not lead to stability changes. However, all P 3 and P 4 mutants were systematically destabilized by about 12.5 and 22°C, respectively.
Association with Trypsin-All the mutants were found to be functionally active, because they inhibited bovine trypsin. For all P 3 mutants the association constants were very high (over 8 ϫ 10 11 M Ϫ1 ) and could not be determined exactly (Table II). However, the P 4 mutants showed large drops in their association constants with trypsin, in the range of 1300-to 4500-fold.
K15R Substitution- Table II shows the effect of the K15R substitution at the P 1 position of BPTI measured for six human proteinases. For five of the proteinases Arg is significantly better than Lys. The largest increase of the association constant was observed for factor X a (484-fold); the lowest increase was for thrombin (17-fold). Factor XII a inhibition by wild type BPTI and its Arg-15 mutant is shown in Fig. 3. Plasmin was the only enzyme that did not differentiate between the two basic side chains, and the measured association constants were the same, within experimental error.
Mutations at the P 3 Site-In the following analysis all association constants are compared with the pseudo wild type BPTI containing two mutations: K15R and M52L. This allows the calculation of single amino acid substitution effects at the P 3 and P 4 sites. The P 3 position (Pro-13 in the wild type) was probed with nine side chains: Gly, Ala, Val, Ile, Phe, Ser, Asp, His, and Arg. A set of P 3 variants was chosen to include different properties of the side chain, such as polarity, charge, hydrophobicity, size, and branching. The observed effects on proteinases are reported in Table III. The largest effects with respect to the substitutions at the P 3 site were observed for factor XII a . With the exception of Ile (5-fold decrease), all P 3 mutants showed stronger inhibition of this proteinase than did the pseudo wild type BPTI. The largest increase (58-fold) was observed for Ser; the weakest effect, which still increased the K a value, was observed for the P 3 Val residue (2-fold). In the case of plasma kallikrein, the best residue was Ile (2.2-fold increase), and the worst was Asp, which decreased K a by a factor of 32.
Three hydrophobic side chains (Val, Ile, and Phe) installed at P 3 improved binding to thrombin by 4.6-, 5.7-, and 1.5-fold, respectively. Other mutations were destabilizing in the range of 1.3-to 23-fold. In the case of plasmin no clear effects of the substitution at P 3 were observed. Two residues, His and Arg, improved binding to this protease 1.5-and 1.8-fold, respectively. The most destabilizing side chain was that of Val (5.6fold). Three mutations (Val (8-fold), Phe (8-fold), and His (4fold)) substantially improved binding of BPTI to factor X a . Gly and Ser at the P 3 site were clearly destabilizing (6.4-and 3.5-fold, respectively). For protein C the most stabilizing were ␤-branched side chains of Val (6-fold) and Ile (4.4-fold) but also  3

and P 4 mutants of BPTI
⌬T den values of P 3 and P 4 mutants were related to the pseudo wild type BPTI. All the P 3 and P 4 mutants and the pseudo wild type also contained an M52L substitution. The protein was dissolved in 10 mM glycine-HCl, pH 2.0 to the final concentration of 20 g/ml. The unfolding experiments were monitored on a Jasco model J-715 spectrometer at 223 nm in the temperature range of 35-100°C; the temperature increase rate was 1°C min.

BPTI mutant
T den ⌬T den ⌬H den P 4 P 3  Substitutions at P 4 -In this study the wild type Gly-12 was substituted with Ser, Val, or Phe residues. These three BPTI mutants could be oxidatively refolded, although with an average yield of 20%, compared with 80% for the P 3 mutants. In addition, as described above, all P 4 mutants showed much greater destabilization compared with P 3 mutants (Table I).
Generally, mutations at P 4 affected the association energy with the studied enzymes to a larger extent than did mutations at the P 3 position. In all studied cases the effect of Gly-12 substitution led to a rather drastic lowering of K a values. The largest drops were found for trypsin (over 3 orders of magnitude), the largest effect reported in this paper. In the case of plasma kallikrein, installation of the ␤-branched Val residue also decreased K a by 3 orders of magnitude. Phe at P 4 was almost as poor as Val. Substitution of Gly with Ser did not significantly lower the association constant. The relatively small effect of G12S substitution was also found for other enzymes studied here. The interactions of protein C and plasmin with three P 4 mutants were similar to those described for plasma kallikrein; the largest (several hundred-fold) decrease of the K a value occurred in the case of Val, and a smaller but still substantial decrease (30 -40-fold) occurred upon mutation to Phe.
For the remaining three proteinases (factors X a and XII a and thrombin) the drops in the association constants were smaller. The effects on factor X a did not exceed 6-fold, and they were somewhat larger for factor XII a and thrombin. In the case of thrombin, Phe at P 4 was the worst residue (23-fold decrease of K a ). DISCUSSION In this paper the inhibition of six human proteinases involved in hemostasis and fibrinolysis (plasma kallikrein, factors X a and XII a , thrombin, plasmin, and protein C) by 13 mutants of BPTI was studied. In addition, bovine ␤-trypsin was investigated as a reference enzyme. All the P 3 and P 4 variants contained a fixed Arg residue at P 1 , because most of the studied enzymes were known to be Arg-specific.
The effect of the K15R substitution was calculated by comparing the association constant values for wild type BPTI and its Arg variant (Table II). The other mutations were installed at the P 3 (nine mutations: Gly, Ala, Val, Ile, Phe, Ser, Asp, His, and Arg) and P 4 (three mutations: Ser, Val, and Phe) sites.
Wild type BPTI binds to the studied human enzymes with association constants in the range of 2.9 ϫ 10 3 M Ϫ1 (factor X a ) to 5.3 ϫ 10 9 M Ϫ1 (plasmin) (Table II). Compared with the wild type, the strength of inhibition was improved for all human enzymes in the range of 2-to 4000-fold (Tables II and III). In many cases the enzymes showed opposite effects in response to the installed mutations, revealing differences in their subsite specificities.
The largest positive effects on the association energy were due to the K15R mutation at the P 1 site. Except for plasmin and trypsin, all other proteinases studied here contain Ala-190 at the bottom of their S 1 specificity pocket (Table II). Conse-  quently, as it was originally noticed for tissue plasminogen activator (5), the S 1 pockets of these enzymes are slightly larger, with one less hydrogen bond partner for P 1 Lys, thus favoring Arg at this position. The result of the K15R substitution on factor X a , a 485-fold increase (Table II), is in agreement with the k cat /K m values for oligopeptide substrates, which show a 50 -260-fold preference for Arg (38). Natural plasma inhibitors of factor X a (antithrombin III and the second domain of TFPI), as well as the natural factor X a protein substrate prothrombin, contain an Arg at their P 1 sites. Additionally, it is worth noticing that potent inhibitors of factor X a isolated from exogenous sources (antistasin and ghilanten) also contain an Arg residue at the P 1 site (42). Site-directed mutagenesis of antistasin showed a 20-fold preference for Arg over Lys (43), and the K15R mutation in human pancreatic secretory trypsin inhibitor led to a Ͼ100-fold improvement of inhibition (44).
Similarly, there is an agreement between our results (Table  II) and oligopeptide kinetic data for human plasma kallikrein, which show a 15-50-fold preference for Arg over Lys (45). Phage display of amyloid ␤-protein precursor libraries and selection on human plasma kallikrein showed full selection of P 1 Arg over 19 other coded residues (24). An enhancement of the antiproteinase activity of BPTI after the K15R mutation was also observed for protein C (20-fold) and thrombin (17fold). Both proteinases contain Ala-190 in the S 1 specificity pocket and are known to recognize the Arg side chain at the P 1 position of their substrates and inhibitors (46).
Bovine trypsin and human plasmin contain Ser at position 190 and do not differentiate between the two basic side chains at the P 1 site of protein inhibitors and substrates (Table II). Phage display selection of the first Kunitz domain of TFPI against human plasmin showed a negligible preference for Arg over Lys at P 1 when this position was randomized with only these two basic residues (23). Very similar effects to those observed here for BPTI were also found for Lys3 Arg substitution in squash inhibitors interacting with plasmin, plasma kallikrein, and factors X a and XII a (Table II). Observed preferences for P 1 Lys in squash inhibitors can result from formation of direct hydrogen bonds between this side chain and Asp-189 (47), compared with water-mediated bonds observed in the BPTI-trypsin complex (26).
Two hydrophobic substitutions at P 4 (G12V and G12F) were chosen to probe the binding properties of the S 4 aryl binding site, which is structurally defined for factor X a and thrombin (2,3). The S 4 pocket of protein C is more polar and flanked by Thr-99 and Asn-174 (4), and this pocket was tested with Ser. For other proteinases the S 4 site is not well defined. Gly-12 is strongly conserved among Kunitz-type inhibitors and, together with three disulfide bridges and the residues Phe-33, Gly-37, and Phe-45, shows the critical locations in the BPTI fold (48). Inspection of the crystal structure of BPTI reveals that the main chain ( ϭ 88.6 o and ϭ178.5 o ) angles of Gly-12 in free BPTI (49) cannot be adopted by non-Gly residues without conformational adjustment. Both the poor yield of oxidative refolding and the Ͼ20°C decrease of T den values observed for P 4 mutants suggest a significant conformational change in free mutant structure upon replacement of Gly-12.
The effects at the P 4 site were deleterious for all the studied enzymes, but to different extents. The P 4 effect was very strong for trypsin; for all the mutants the K a value dropped more than 3 orders of magnitude. Interestingly, for enzymes with the aryl binding site (factor X a and thrombin), the decreases due to the G12V mutation were only about 6-fold, compared with e.g. Ͼ1,000-fold for plasma kallikrein, 500-fold for plasmin, or 230fold for protein C. This might indicate that the deleterious effect on the loop conformation is partially compensated for by a favorable interaction of the phenyl ring with the S 4 pocket. Significant destabilization of the inhibitor molecule and a lack of positive effects on proteinase inhibition were discouraging to further mutational studies.
Compared with the effects on P 4 , nine substitutions at the P 3 position provided differential data on proteinase binding. For example, with the exception of Ile, all the mutations at position P 3 resulted in a substantial improvement of association with factor XII a (Table III). The most favorable single substitution, P13S, yielded a 58-fold increase of the association constant, which is in agreement with k cat /K m data on p-nitroanilide substrates (50). Such comparisons should be taken with caution, because substrate kinetic (and probably also thermodynamic effects on inhibitor binding) data are dependent on the structural context of the mutated site. For example, when Phe is present at P 2 in tripeptide substrates, Pro3 Ser substitution at P 3 increases k cat /K m 97-fold, whereas when Gly is present at P 2 the effect is only 3.6-fold (50).
Three hydrophobic mutations introduced at the P 3 site (Val, Ile, Phe) improved binding of BPTI to thrombin severalfold (Table III), in agreement with the x-ray structure of human ␣-thrombin inhibited by D-Phe-Pro-Arg chloromethylketone, which shows that the S 3 site overlaps with the aryl binding site (51). Regarding these three side chains and Ala, quantitatively similar effects were also found for protein C. However, the two proteinases differ in P 3 Gly recognition; whereas Gly is neutral for protein C, its binding to thrombin is disadvantageous.
All other substitutions also lowered the K a value with thrombin. When Asp was introduced at the P 3 site of BPTI, a 15-fold decrease in the association energy was observed, whereas for factor X a the K a value for the P13D mutant of BPTI was 2-fold higher than for the pseudo wild type. The negatively charged residues at any position between P 3 and P 3 Ј could hamper the catalysis of thrombin in the absence of the anticoagulatory cofactor thrombomodulin. Contrary to thrombin, factor X a contains three basic amino acids at positions 222-224 (Arg-222, Lys-223, and Lys-224) that can interact with the acidic residue at P 3 (3). Both Glu and Asp are found in the activating sequences of prothrombin, together with the well conserved Gly residue at the P 2 site (52). Surprisingly, the effect of the P13D mutation in BPTI was not as positive as expected. This could be explained by modeling of the binding of the second domain of TFPI to factor X a (53). According to the model, the presence of the half-cystine at P 2 in the domain forces the side chain of Tyr-99, which normally restricts the S 2 pocket, to swing out toward Phe-174. This causes an increase of the hydrophobic area around the S 3 pocket, which does not allow the binding of acidic residues as easily as it would when the rearrangement of Tyr-99 did not occur (7). Cho et al. (50) also showed that the preference toward either acidic or hydrophobic side chains at P 3 depends on the nature of the P 2 residue. For a series of Z-Xaa-Gly-Arg-pNA substrates, a Glu residue at the P 3 position resulted in the highest k cat /K m value (3.9 ϫ 10 4 M Ϫ1 sec Ϫ1 ), whereas for the Z-Xaa-Phe-Arg-pNA series, Phe was best as the P 3 side chain.
Similar to thrombin, factor X a preferred hydrophobic Val and Phe at P 3 by 8-fold; the Ile side chain, however, was neutral (Table III). Contrary to thrombin, the hydrophobic S 3 pocket of factor X a is lined by the carbonyl oxygens of Lys-96 and Glu-97 that can favorably bind positive charges. Indeed, the P13H mutation provided a 4-fold increase in K a .
Only two basic side chains of His and Arg improved (slightly) the binding of P 3 mutants to plasmin. In agreement with this result, selection of phage display libraries of the first domain of TFPI on plasmin showed a strong preference for Pro at the P 3 site (23). The S 3 site of plasma kallikrein seems to favorably accept hydrophobic side chains of increasing size: GlyϽAlaϽValϽIle (Table III). The phenyl ring of Phe was already too large for this pocket. The acidic side chain of Asp at P 3 was particularly bad for this enzyme.