Structure-function analysis of the reactive site in the first Kunitz-type domain of human tissue factor pathway inhibitor-2.

Human tissue factor pathway inhibitor-2 (TFPI-2) is a Kunitz-type proteinase inhibitor that regulates a variety of serine proteinases involved in coagulation and fibrinolysis through their non-productive interaction with a P(1) residue (Arg-24) in its first Kunitz-type domain (KD1). Previous kinetic studies revealed that TFPI-2 was a more effective inhibitor of plasmin than several other serine proteinases, but the molecular basis for this specificity was unclear. In this study, we employed molecular modeling and mutagenesis strategies to produce several variants of human TFPI-2 KD1 in an effort to identify interactive site residues other than the P(1) Arg that contribute significantly to its inhibitory activity and specificity. Molecular modeling of KD1 based on the crystal structure of bovine pancreatic trypsin inhibitor revealed that KD1 formed a more energetically favorable complex with plasmin versus trypsin and/or the factor VIIa-tissue factor complex primarily due to strong ionic interactions between Asp-19 (P(6)) and Arg residues in plasmin (Arg-644, Arg-719, and Arg-767), Arg-24 (P(1)) with Asp-735 in plasmin, and Arg-29 (P(5)') with Glu-606 in plasmin. In addition, Leu-26 through Leu-28 (P(2)'-P(4)') in KD1 formed strong van der Waals contact with a hydrophobic cluster in plasmin (Phe-583, Met-585, and Phe-587). Mutagenesis of Asp-19, Tyr-20, Arg-24, Arg-29, and Leu-26 in KD1 resulted in substantial reductions in plasmin inhibitory activity relative to wild-type KD1, but the Asp-19 and Tyr-20 mutations revealed the importance of these residues in the specific inhibition of plasmin. In addition to the reactive site residues in the P(6)-P(5)' region of KD1, mutation of a highly conserved Phe at the P(18)' position revealed the importance of this residue in the inhibition of serine proteinases by KD1. Thus, together with the P(1) residue, the nature of other residues flanking the P(1) residue, particularly at P(6) and P(5)', strongly influences the inhibitory activity and specificity of human TFPI-2.

Proteinase inhibitors play a critical role in the regulation of several physiological processes such as blood coagulation, com-plement fixation, fibrinolysis, and fertilization (1). Most of these inhibitors are proteins having characteristic polypeptide scaffolds, and are grouped into a number of families, including the Kunitz (2), Kazal (2), Serpin (3), and mucus (4) families. The Kunitz-type family, serine proteinase inhibitors that include one or more Kunitz-type inhibitory domains, includes tissue factor pathway inhibitor (TFPI) 1 and type-2 tissue factor pathway inhibitor (TFPI-2). These two inhibitors have been investigated extensively in the past decade and have been shown to play an important role in inhibiting serine proteinases involved in coagulation and fibrinolysis (5)(6)(7)(8). Human TFPI-2, originally isolated from placenta and designated as placental protein 5, is a matrix-associated inhibitor consisting of three tandemly arranged Kunitz-type proteinase inhibitor domains flanked by a short acidic amino terminus and a highly basic C-terminal tail (7,9). A wide variety of cells, including keratinocytes (10), dermal fibroblasts (10), smooth muscle cells (11), syncytiotrophoblasts (12), synoviocytes (13), and endothelial cells (14), synthesize and secrete TFPI-2 primarily into their extracellular matrix. Three variants/isoforms of molecular mass 32, 30, and 27 kDa are synthesized by these cells and are thought to represent differentially glycosylated forms (15). TFPI-2 exhibits inhibitory activity toward a broad spectrum of proteinases, including trypsin, plasmin, chymotrypsin, cathepsin G, plasma kallikrein, and the factor VIIa-tissue factor complex. However, TFPI-2 exhibits little, if any, inhibitory activity toward urokinase-type plasminogen activator, tissue-type plasminogen activator, and ␣-thrombin (16). TFPI-2 presumably inhibits proteinases through a P 1 arginine residue (Arg-24) in its first Kunitz-type domain, as an R24Q TFPI-2 mutant exhibited only 5-10% inhibitory activity toward trypsin, plasmin, and the factor VIIa-tissue factor complex (17). Recently, TFPI-2 expression by select tumors has been shown to play a significant role in inhibiting tumor growth and metastasis by a mechanism that involves its inhibitory activity (18,19).
Several approaches have been employed to elucidate the structure-function relationship and broad specificity of Kunitztype inhibitors using the well characterized bovine pancreatic trypsin inhibitor (BPTI) as a model. Detailed biophysical and biochemical studies have provided a greater insight into the structural basis for the association of BPTI, or its homologues, to proteinases. Moreover, using semisynthetic (20,21) or recombinant approaches (22,23), it has been possible to change or enhance the inhibitory activity and spectrum of BPTI, as well as its homologues. Kunitz-type inhibitors possess a compact pear-shaped structure stabilized by three disulfide bonds containing a reactive site region featuring the principal determinant P 1 residue in a rigid conformation. These inhibitors competitively prevent access of the serine proteinase for its physiologically relevant macromolecular substrate through insertion of the P 1 residue into the active site cleft (24). In addition to the P 1 residue, other residues within the reactive site region of BPTI (P 4 -P 4 Ј) have been shown to interact with different serine proteinases, and it is generally recognized that the N-terminal side of the reactive site (P) is energetically more important than the PЈ C-terminal side (25). In all, about 10 -12 amino acid residues in the inhibitor and 20 -25 residues in the proteinase are in direct contact in the formation of a stable proteinase-inhibitor complex, and provide a buried area of 600 -900 Å (26). Although many proteins structurally similar to BPTI, such as TFPI KD2 (27), APPI (28), and bikunin (29), have been isolated and their three-dimensional structures determined, there are few studies that have assigned the relative contribution of residues flanking the reactive site residue in the formation of the proteinase-inhibitor complex and their affect on inhibitory activity and specificity (30 -32).
In the case of TFPI-2, it is generally believed that its first Kunitz-type domain, in a BPTI-like manner, harbors most of its inhibitory activity, although no studies have definitively shown that this domain is sufficient to mediate this activity. In the present study, the complete first Kunitz domain of human TFPI-2 was expressed and purified, and its inhibitory activity toward selected proteinases was compared with that of fulllength TFPI-2 and BPTI. In addition, molecular modeling was employed to obtain three-dimensional structural information on complexes of TFPI-2 KD1-plasmin, TFPI-2 KD1-trypsin, and TFPI-2 KD1-factor VIIa to identify residues in KD1 involved in its molecular recognition of each proteinase. From this analysis, residues primarily responsible for the interaction and proteinase specificity were then selected for mutagenesis. Select amino acid residues on both the N-and C-terminal side of the reactive site residue (P 1 ) were substituted individually, and the effects of these point-mutations on the proteinase specificity and inhibitory activity were investigated.
Molecular Modeling-Three-dimensional structural information on complexes formed between KD1 and plasmin, KD1 and trypsin, and KD1 and factor VIIa was obtained using molecular modeling strategies. The crystallographically determined structures of factor VIIa-TF inhibited with a BPTI mutant ((33); pdb code 1fak), factor VIIa-TF ((34); pdb code 1dan), trypsin inhibited with TFPI KD2 ((27); pdb code 1tfx), the NMR-determined structure of TFPI KD2 ((27); pdb code 1adz), trypsin inhibited with BPTI ((35); pdb code 2ptc), and plasmin ((36); pdb code 1bml) served as templates in building these models. Bulk solvent was excluded from the proteinase-inhibitor complex, and, accordingly, it was anticipated that hydrogen bonds and ionic interactions that may play an important role in specificity could be accurately evaluated. The protocols for modeling these complexes have been described in detail earlier (8). Briefly, the relative positions of the inhibitor and proteinase domains were maintained and adjustments were only made to the side chains. Hydrophobic/van der Waals, hydrogen bonds, and ionic interactions were observed between each proteinase-inhibitor complex. All of these interactions were taken into consideration in evaluating each proteinase-inhibitor complex, and it was assumed that all potential hydrogen bond donors and acceptors would participate in these interactions.
Expression and Purification of Wild-type and Mutant Proteins-The first Kunitz-type proteinase inhibitor domain of human TFPI-2 (KD1), and its mutants were overexpressed as N-terminal His-tagged fusion proteins in E. coli strain BL21(DE3)pLys using the T7 promoter system (37). The recombinant plasmid derived from pET19b bearing a decahistidine tag leader sequence followed by an enterokinase cleavage site and cDNA encoding the first Kunitz-domain of TFPI-2 was prepared according to standard procedures (38). Using this recombinant vector as a template, several other constructs containing the desired point mutations were generated using a QuikChange® site-directed mutagenesis kit according to the manufacturer's instructions. Each recombinant construct was examined for in-frame orientation, integrity, and desired mutation by nucleic acid sequencing. Wild-type and mutant His tag KD1 preparations were expressed in E. coli grown in rich media containing 100 mg/liter ampicillin, and induced at 37°C with 1 mM isopropyl thiogalactopyranoside at mid log-phase (A 600 ϭ 0.6 -0.8). The overexpressed proteins were recovered from the cell lysates in the form of inclusion bodies following sonication in 50 mM Tris-HCl (pH 8.0) containing 0.5 M NaCl, 5 mM 2-mercaptoethanol, and 10 mM imidazole (buffer A). Inclusion bodies were recovered by high speed centrifugation (20,000 ϫ g for 60 min) and thoroughly washed overnight at room temperature before solubilizing in buffer A containing 6 M guanidine hydrochloride. The solubilized inclusion bodies were recovered by high speed centrifugation and were filtered through 0.22-m Nalgene® filters before application to His-Trap® column individually dedicated to each expressed protein. His-Trap affinity columns were used in a Amersham Biosciences FPLC system and purification was carried out following the manufacturer's protocol. Peak fractions were identified by SDS-PAGE, pooled, and oxidatively refolded by initial dialysis against 50 mM Tris-HCl (pH 8.0) containing 3 mM 2-mercaptoethanol, followed by extensive dialysis against 50 mM Tris-HCl (pH 8.0).
The refolded proteins were purified to homogeneity by Mono Q FPLC at room temperature. KD1 proteins were eluted from this column in a linear NaCl gradient consisting of 50 mM Tris-HCl (pH 8) and 50 mM Tris-HCl (pH 8) containing 1 M NaCl. Peak fractions were subjected to SDS-PAGE analysis, and pure fractions were pooled and concentrated on YM-3 ultrafiltration membranes.
General Methods-The concentration of each purified KD1 protein was determined by measuring its absorbance at 280 nm using calculated values for E 1% derived from its Tyr, Trp, and Cys content (39). The concentration of plasmin was provided by the supplier, whereas the concentrations of all other proteins used in this study were determined according to Bradford (40) using BSA as the reference protein. SDS-PAGE was performed according to Laemmli (41) using 4 -20% polyacrylamide gradient gels.
Trypsin and Plasmin Inhibition Assays-Trypsin and plasmin inhibition assays were performed as described elsewhere (16). Briefly, trypsin and plasmin were incubated with various concentrations of inhibitor preparations for 15 min at 37°C in a 96-well microtitration plate. The chromogenic substrate S-2251 was then added, and residual amidolytic activity was measured in a Molecular Devices UV max kinetic microplate reader.
Inhibition Assay for Factor VIIa-Tissue Factor Amidolytic Activity-Recombinant soluble human tissue factor (100 nM) and factor VIIa (50 nM) were incubated in a TBS (50 mM Tris-HCl (pH 7.5) containing 100 mM NaCl)-BSA buffer/5 mM CaCl 2 for 15 min at 37°C. Following this incubation, aliquots (100 l) were dispensed into a 96-well microtitration plate and treated with serial dilutions of inhibitors dissolved in TBS buffer. After 15 min of incubation, 30 l of S-2288 (final concentration, 1 mM) was added to each well and the absorbance at 405 nm was determined as described earlier.
Inhibition Kinetics-The apparent inhibition constant, K i Ј was determined using the non-linear regression data analysis program Ultrafitƒv3.0 (Biosoft). Trypsin and plasmin inhibitory data were analyzed according to the following equation for a tight-binding inhibitor, where v i and v 0 are the inhibited and uninhibited rates, respectively, and [I] 0 and [E] 0 are the total concentrations of inhibitor and enzyme, respectively. Factor VIIa-tissue factor inhibition data, where K i Ͼ Ͼ [E] 0 , were analyzed according to Equation 2.
The K i values were obtained by correcting for the effect of substrate according to Bieth et al. (42),

Molecular Modeling and Selection of Mutations-Previous
studies demonstrated that human TFPI-2 is a strong inhibitor of plasmin and trypsin, and a relatively weak inhibitor of the factor VIIa-tissue factor complex (16). The molecular basis of the specificity of TFPI-2 for plasmin and trypsin relative to the serine proteinase factor VIIa is unclear, but presumably involves residues other than the P 1 Arg in the first Kunitz-type domain of TFPI-2, as well as residues in the active site region of the proteinase. To address whether other residues in the reactive site region of TFPI-2 may play a role in its inhibitory potency and specificity, we employed a molecular modeling approach to guide subsequent mutagenesis studies designed to provide information on the functional importance of these residues. Because our preliminary data indicated that a recombinant preparation of the first Kunitz-type domain of TFPI-2 (KD1) exhibited better inhibitory activity in comparison to the intact parent molecule (see below), we decided to model complexes of KD1 with plasmin, trypsin, and factor VIIa based on the crystal structure of BPTI and each proteinase, respectively. A preliminary inspection of the amino acid sequences surrounding the P 1 residue in a number of Kunitz-type inhibitors revealed highly conserved residues at the P 6 , P 1 , P 5 Ј, and P 18 Ј positions (Fig. 1), and our molecular modeling studies thus initially focused on the contributions of these residues in the formation of an energetically stable complex between KD1 and the above proteinases. In the model structures of these complexes, no unfavorable contacts between atoms and no unnatural chiral centers were observed. In the Ramachandran plot of the main-chainangles, all of the non-glycine residues are in the most favored or permissive regions. Moreover, there are no gross steric clashes that preclude the interaction of proteinases with KD1. For simplicity and consistency, the residue numbering system employed for KD1 and each proteinase is that of its linear sequence position. In addition, to relate each proteinase residue number to its corresponding position in the prototypical proteinase chymotrypsin, each proteinase residue number is followed by its position in chymotrypsin in braces and is preceded by the letter "c." Finally, the relationship between residues mutated in KD1 and their position in BPTI is indicated in Table I. 2 In the KD1-proteinase complex, there is an interactive hydrophobic patch and an internal hydrophobic patch in KD1 (Fig. 2, A and B). The plasmin-interactive hydrophobic interface is formed by a number of residues in KD1, including Leu-26, Leu-27, Leu-28, Leu-43, and Tyr-55. The residues Leu-18, Tyr-20, Tyr-31, Phe-42, and Tyr-44 in KD1 are buried within and contribute to the formation of an internal hydrophobic pocket. Within the interactive patch, Leu-27 in KD1 interacts with Phe-583{c37}, Met-585{c39}, Phe-587{c41}, and C B of Lys-607{c61} in plasmin (Fig. 2, A and B). In addition, Leu-28 interacts with Met-585{c39} in plasmin. Furthermore, the side-chain C D and C E of Lys-607{c61} could make hydrophobic interactions with C D1 and C E1 of Tyr-55 in KD1 ( plimentarity when compared with analogous interactions in KD1-trypsin (Fig. 3A) and KD1-factor VIIa (Fig. 3B) complexes. In this regard, the hydrophobic patch observed between KD1 and plasmin does not exist with factor VIIa (Fig. 3B), whereas trypsin appears to have this hydrophobic patch interaction (Fig. 3A). Tyr-159{c151} in trypsin is probably involved in hydrophobic interactions with Leu-26 and Leu-43 of KD1 as is evident from the KD1-trypsin complex. In addition, Phe-49{c41} in trypsin is positioned to interact with Leu-27 of KD1. We further propose that Lys-68{c60} of trypsin may interact with Tyr-55 of KD1 and that the side-chain of Lys-68{c60} in trypsin may interact with Leu-27 via hydrophobic interactions. Finally, there are also main-chain interactions between the carbonyl O of Pro-22 in KD1 and the amide N of Gly{c216} in serine proteinases, as well as the amide N of Arg-24 in KD1 and the carbonyl O of Ser{c214} in serine proteinases. These interactions are common to a serine proteinase interacting with Kunitz-type inhibitors (26).
In addition to hydrophobic interactions, electrostatic attraction/repulsion also plays an important role in forming and stabilizing the KD1-proteinase complex (Fig. 2B). One of the reactive site residues at the P 6 position, Asp-19, together with Glu-48 of the secondary loop, forms an acidic patch in KD1. This acidic patch interacts with a basic patch in plasmin that consists of Arg-644{c98}, Arg-719{c173}, and Arg-767{c221}. Inasmuch as this basic patch is not present in either trypsin or factor VIIa (Fig. 3, A and B), we believe that Asp-19 in KD1 enhances the specificity of KD1 for plasmin through this electrostatic interaction. Tyr-20, the P 5 residue, lines the hydrophobic cavity of the internal hydrophobic patch of KD1 and contributes to its structural stability. The following P 4 residue, Gly-21 is in close proximity spatially to Asp-19 and, as observed in the structure N E2 of Gln-738{c192} in plasmin, makes a hydrogen bond with the backbone C-O of Gly-21 in KD1. At the P 3 position in KD1, Pro-22 is involved in a turn and also fits into a hydrophobic patch in plasmin, trypsin and factor VIIa. Pro-22 also sits in the S 3 /S 4 site of the proteinase. Arg-24, the P 1 residue, ion pairs with Asp-735{c189} at the bottom of the substrate binding pocket in plasmin and is further stabilized through hydrogen bonding to Ser-736{c190} O␥ and Gly-765{c219}O. Arg-29, the highly conserved Arg/Lys at the P 5 Ј of Kunitz-type inhibitors, makes hydrogen bonds with Glu-606{c60} in plasmin, and interacts with Tyr-67{c59} in trypsin and Asp-196{c60} in factor VIIa. Finally, the conserved P 18 Ј residue, Phe-42 is located in a hydrophobic pocket with Tyr-20, Leu-18, Tyr-31, and Tyr-44 in KD1 and probably contributes to the stabilization of the KD1 inhibitory structure. Based on the above molecular modeling studies, KD1 residues Asp-19, Tyr-20, Gly-21, Arg-24, Leu-26, Arg-29, and Phe-42 were selected for mutagenesis. A schematic model representation of the human TFPI-2 KD1 is illustrated in Fig. 4 and highlights residues mutated in this study at the P 6 , P 5 , P 4 , P 1 , P 2 Ј, P 5 Ј, and P 18 Ј.
Preparation and Purification of Human TFPI-2 KD1 and Various KD1 Mutants-Wild-type and site-specific mutant preparations of human TFPI-2 KD1 were overexpressed as His-tagged fusion proteins in E. coli. The in-frame orientation, integrity, and desired mutations in the recombinant constructs were confirmed by nucleic acid sequencing. Each of the recombinant KD1 preparations was expressed and purified from 4 liters of LB broth following induction at 37°C with 1 mM isopropyl thiogalactopyranoside. The KD1 preparations obtained from inclusion bodies were solubilized in 6 M guanidine hydrochloride and initially purified on a nickel-charged metal chelating column (His-Trap®) and refolded by sequential dialysis in the presence and absence of reducing agent. The partially purified and refolded KD1 preparations consisted mainly of monomeric KD1 (ϳ70%) with the remainder consisting of KD1 oligomers. Monomeric KD1 was subsequently separated from oligomeric KD1 by Mono Q FPLC. Each of the KD1 preparations migrated as a single band with an average apparent molecular mass of 16 kDa in a denaturing SDS-PAGE gel (Fig.  5). An average yield of 3 mg of purified KD1 was obtained per liter of broth. For each KD1 preparation, the precise molecular weight values were obtained from its amino acid composition, and their mass concentrations were determined spectrophotometrically at 280 nm using a calculated E 1% value based on its Trp, Tyr, and Cys content (39).
Inhibitory Properties of Human TFPI-2 KD1 and Various KD1 Mutants-Recombinant KD1 exhibited stronger inhibitory activity toward each of the three serine proteinases in comparison to the eukaryotically expressed recombinant fulllength TFPI-2 molecule (Table I), providing strong evidence that the other two Kunitz-type domains of TFPI-2 do not provide any significant effect on the inhibitory properties of TFPI-2 and that post-translational modifications are not essential for full expression of its inhibitory activity. Wild-type KD1 inhibited plasmin amidolytic activity with roughly a 3-fold higher K i value than BPTI (Table I), but was 3-fold more potent than full-length TFPI-2. On the other hand, KD1 exhibited ϳ2-fold higher trypsin inhibitory activity than full-length TFPI-2, but was 2-fold less potent than BPTI (Table I). Wildtype KD1 inhibited factor VIIa-tissue factor amidolytic activity with a comparable K i value to that of full-length TFPI-2, whereas BPTI failed to show any inhibitory activity toward this complex (Table I). Extra amino acids N-terminal to the mature protein, such as the His tag region, had little, if any, negative effect on its inhibitory activity. The higher inhibitory activity of the first Kunitz-type domain of TFPI-2 compared with fulllength TFPI-2 in all likelihood is attributed to its smaller size and/or flexibility.
The effect of mutations at the P 6 , P 5 , P 4 , P 1 , P 2 Ј, P 5 Ј, and P 18 Ј are also listed in Table I. An alanine substitution at the P 6 position, Asp-19, showed a dramatic (ϳ40-fold) loss of inhibitory activity toward plasmin but failed to show any significant loss of activity toward trypsin and factor VIIa. From the molecular graphics model (Fig. 1A), the most rational explanation for this effect is that Asp-19 interacts with a basic patch in plasmin consisting of Arg-644{c98}, Arg-719{c173}, and Arg-767{c221}. Because there are no corresponding basic patches in trypsin or factor VIIa at these positions, the interaction of Asp-19 with the basic patch in plasmin appears to confer KD1 with enhanced reactivity, and inhibitory potency, toward plasmin. Mutagenesis of the neighboring P 5 residue, Tyr-20, to Ala also had a significant negative effect on KD1 inhibition of plasmin, suggesting that this residue either contributes toward the formation of the KD1-plasmin complex, or is critical in maintaining the conformation of the KD1 reactive site toward plasmin. Mutagenesis of Tyr-20 to Ala, however, slightly enhanced its ability to inhibit trypsin and factor VIIa ( Table I).
The residue at the P 4 position, Gly-21, also seems to play a supportive role in the interaction of KD1 with proteinases as shown earlier using BPTI mutants (32). In this study, we mutated Gly-21 to aspartic acid to increase the acidic patch on KD1 and enhance its interaction with the basic patch in plas-min, because Gly-21 is in close spatial proximity to Asp-19. However, mutation of Gly with Asp did not have the desired effect, and this mutant lost inhibitory activity toward all proteinases tested most probably due to perturbation in the mainchain conformation of the KD1 reactive site. In this regard, the phi angle () of Gly is ϩ111°and the psi angle () is Ϫ174°, which places Gly-21 in a region of the Ramachandran plot accessible only to Gly residues (43). Accordingly, any other residue would conceivably alter the backbone structure by changing the phi and psi angles resulting in an altered mainchain conformation in the vicinity of Gly-21. The proteinase domain number of plasmin is based on chymotrypsin numbering. A, specific interactions between plasmin and TFPI-2 KD1. Red represents oxygen, blue represents nitrogen, and green represents carbon atoms. Plasmin is shown with cyan ribbons, and TFPI-2 KD1 is shown with yellow ribbons. TFPI-2 KD1 has an acidic patch (Asp-19 and Glu-48) that interacts with a basic patch on plasmin (Arg-644{c98}, Arg-719{c173}, and Arg-767{c221}). The NH 2 groups of Arg-719{c173} in plasmin could make H-bonds with both of the side-chain carboxylate groups of Glu-48 in TFPI-2 KD1. Gln-738{c192} N E2 in plasmin appears to make a H-bond with the carbonyl oxygen of Gly-21 in TFPI-2 KD1. In addition to these interactions, TFPI-2 KD1 contains a hydrophobic core consisting of Leu-18, Tyr-20, Tyr-31, and Phe-42. This hydrophobic core is connected to an interactive hydrophobic patch consisting of Leu-26, Leu-27, Leu-28, and Leu-43. This hydrophobic patch in KD1 makes hydrophobic interactions with the C B of Lys-607{c61}, Phe-583{c37}, Met-585{c39}, and Phe-587{c41} in plasmin. B, electrostatic potential between TFPI-2 KD1 and plasmin proteinase domain. The electrostatic potential between TFPI-2 KD1 and plasmin was determined using the program GRASP (54), and the orientation of the molecules is the same as in A. Blue represents positive, red represents negative, and white represents neutral residues. Region 1 refers to the interactions of the acidic patch on TFPI-2 KD1 (Asp-19 and Glu-48) and a basic patch on plasmin (Arg-644{c98}, Arg-719{c173}, and Arg-767{c221}) described above, and region 2 refers to the hydrophobic interactions between TFPI-2 KD1 and plasmin described above.

FIG. 3. Models of the interaction of TFPI-2 KD1 with trypsin and factor VIIa.
The orientation of the molecules is the same as in Fig. 1. Blue represents positive, red represents negative, and white represents neutral residues. A, electrostatic potential between TFPI-2 KD1 and trypsin using the program GRASP (54). Region 1 shows the absence of the interaction between the acidic region (Asp-19 and Glu-48) of TFPI-2 KD1 and a basic region present in plasmin but absent in trypsin. Instead, this region is acidic in trypsin. Region 2 corresponds to the hydrophobic interactions between TFPI-2 KD1 and trypsin, which is similar to those in the interaction of TFPI-2 KD1 with plasmin. B, electrostatic potential between TFPI-2 KD1 and factor VIIa determined using the program GRASP (54). Region 1 shows the absence of the interaction between the acidic region (Asp-19 and Glu-48) of TFPI-2 KD1 and a basic region, which is present in plasmin but not factor VIIa. Instead, this region is hydrophobic. Region 2 corresponds to the absence of hydrophobic patch interactions between TFPI-2 KD1 and factor VIIa, which are present in both plasmin and trypsin.
The side chain of the P 1 residue primarily dictates the specificity of a proteinase inhibitor for its cognate proteinase. Systematic substitution at this position in a number of inhibitors revealed a large dynamic range of effects on its association with different proteinases (31, 32, 44 -49). The glutamine substitution at the P 1 site resulted in a decreased inhibitory activity in KD1 (Table I), as was observed with the full-length R24Q TFPI-2 mutant (17). As observed in the model structure of KD1 with plasmin, trypsin, and factor VIIa, the Arg-24 in KD1 forms a salt bridge, in addition to two hydrogen bonds, with the carbonyl backbone of Asp{c189} and Gly{c219} to stabilize the complex. Mutation of Arg to Gln eliminates interaction with the S 1 site residue Asp{c189} due to the shorter side chain and its lack of charge. Lysine substitution at this position restores the inhibitory activity of KD1, and the lower K i values obtained with R24K KD1 against plasmin and trypsin could be the result of an ionic interaction of the protonated amino group with the carboxylate group of Asp{c189}, as well as watermediated hydrogen bonding between the carbonyl group of Gly{c219} and the hydroxyl group of Ser{c190} with the P 1 Lys amino group. In view of this potential bonding pattern, it is curious as to why the inhibitory activity of R24K KD1 for factor VIIa was reduced ϳ5-fold, inasmuch as factor VIIa also has a Ser-326{c190}. The reason for its reduced inhibitory activity against factor VIIa is not known at this point but may be due other residues in the substrate binding pocket of factor VIIa as opposed to that of plasmin and trypsin.
As mentioned above, KD1 contains a cluster of hydrophobic Leu residues at the P 2 Ј-P 4 Ј region that interacts with a hydrophobic patch in plasmin, trypsin, and factor VIIa. To disrupt this cluster, Leu-26 was substituted with the highly hydrophilic residue, glutamine. This L26Q mutation resulted in at least a 10-fold reduction in inhibitory potency of KD1 toward each of the proteinases tested (Table I) and underscores the importance of this hydrophobic interaction in the inhibitory mechanism of KD1. Leu-26 is part of a hydrophobic patch and interacts with Leu-43 and Leu-28 of KD1. Leu-26 also has the potential to have hydrophobic interactions with Gln-738{c192} in plasmin, with C D and C G of Gln-200{c192} in trypsin, and with C D and C G of Lys-328{c192} in factor VIIa. Thus, changing this residue to a non-hydrophobic residue such as Gln will disrupt these interactions and be disruptive for each proteinase.
Virtually all Kunitz-type domains studied have a highly Mutant KD1 preparations are designated according to the notation by Shapiro and Vallee (56), in which the single letter code for the original amino acid is followed by its position in the sequence and the single letter code for the new amino acid.
conserved Lys/Arg at the P 5 Ј position (Fig. 1), and three point mutants were made at this position. In plasmin, Glu-606{c60} makes hydrogen bonds with Arg-29 in KD1, whereas Tyr-67{c59} and Asp-196{c60} in trypsin and factor VIIa, respectively, interact with this residue. Substitution of Arg-29 with alanine resulted in a marginal loss of inhibitory activity toward all three proteinases (Table I), whereas substitution with aspartic acid presumably caused charge repulsion, as well as disruption of hydrogen bonds, with a major effect on K i ( Table  I). Mutation of Arg-29 with lysine could possibly preserve the hydrogen bonding observed with Arg and resulted in minor changes in K i (Table I). Although the P 5 Ј Arg/Lys residue is important in the inhibitory mechanism of KD1, it does not appear to be a major determinant in KD1 specificity.
Finally, as expected, mutagenesis of the highly conserved Phe-42 at the P 18 Ј position with alanine resulted in similar losses of inhibitory activity toward all three proteinases (Table  I), presumably by disruption of the internal hydrophobic core in KD1 formed by Phe-42, Tyr-20, Leu-18, Tyr-31, Tyr-44, and the side chain of Arg-29.

DISCUSSION
In the present study, we have expressed and purified the human TFPI-2 Kunitz-type domain 1 (KD1) and compared its inhibitory activity toward plasmin, trypsin, and the factor VIIa-tissue factor (VIIa-TF) complex to that of full-length TFPI-2, BPTI, and nine human TFPI-2 KD1 constructs with mutations in the reactive site region (P 6 -P 5 Ј). The isolated TFPI-2 KD1 exhibited stronger inhibitory activity toward these proteinases in comparison to intact TFPI-2. Alanine substitution at the P 6 (D19A) and the P 5 (Y20A) positions had a marginal effect on its inhibitory activity toward trypsin and VIIa-TF but exhibited a marked decrease in activity toward plasmin. Substitution of aspartic acid for alanine was particularly deleterious to plasmin inhibition by KD1, and molecular modeling studies revealed that this was in all likelihood due to the modulation of an ionic interaction between an acidic patch in KD1, formed by Asp-19 and Glu-39, and a basic patch unique to plasmin composed of Arg-644{c98}, Arg-719{c173}, and Arg-767{c221}. Thus, Asp-19 and Tyr-20 in KD1 appear to play a major role in the specificity of TFPI-2 for plasmin. In contrast, point mutations at the P 4 (G21D), P 1 (R24Q), P 2 Ј(L26Q), and P 5 Ј(R29A) positions all exhibited substantial decreased inhibitory activity toward all of these proteinases. The importance for a highly conserved basic residue (Arg/Lys) at the P 5 Ј position was evident from a substantial loss of inhibitory activity in the R29D KD1, presumably through the loss of either a stabilizing ionic interaction between Arg-29 and Glu-606/Asp-196{c60} in plasmin/VIIa, or by hydrogen bonding of Arg-29 to Tyr-67{c59} in trypsin. Finally, mutation of a highly conserved phenylalanine at the P 18 Ј position (F42A) revealed the importance of this residue in the stabilization of the reactive site structure through internal hydrophobic interactions.
A lysine substitution at the P 1 position (R24K) in KD1 significantly increased its inhibitory activity toward both plasmin and trypsin, making it essentially as effective as BPTI toward these proteinases. In sharp contrast, R24K KD1 paradoxically exhibited approximately a 5-fold reduction in inhibitory activity toward VIIa-TF, a somewhat surprising result in consideration of the fact that VIIa contains a Ser-326{c190} that forms an additional water-mediated hydrogen bond with the protonated ␦-amino group in lysine (1) and that VIIa forms a stable interaction with the first Kunitz-type domain of TFPI through its interaction with a P 1 lysine residue. On the other hand, BPTI also contains a lysine in its P 1 position and failed to inhibit VIIa-TF (Table I), suggesting that VIIa prefers Arg P 1 residues and that other residues in the reactive site region of TFPI-2 KD1 somehow synergistically enhance VIIa inhibition, as has been shown for a BPTI mutant (33).
Of potential clinical relevance, R24K KD1 exhibited essentially the same inhibitory activity as BPTI, which is widely used as a plasmin inhibitor during surgery but, being of bovine origin, precipitates episodes of severe anaphylaxis on some occasions (0.5-1%). In this context, these studies may provide a template for the design of improved Kunitz-type serine proteinase inhibitors with considerable therapeutic potential. In this regard, our laboratory and other laboratories have demonstrated the importance of serine proteinase inhibition in the growth, migration, angiogenesis, and metastasis of a variety of human tumors (18,19,57,58). These tumor properties are presumably mediated in large part by proteinases such as plasmin and/or trypsin IV (59), and the secretion of inhibitory TFPI-2 by these tumors markedly inhibits their growth and metastasis in animal models (18,19). Moreover, in preliminary studies, we have shown that human KD1 exhibits dose-dependent inhibition of angiogenesis in a commercially available in vitro human endothelial cell angiogenesis assay. 3 In addition, intravenous administration of human KD1 to ovalbumin-sensitized asthmatic mice resulted in a significant decrease in the number of airway macrophages and lymphocytes relative to vehicle-treated asthmatic mice, suggesting that KD1 inhibits the proteinase-mediated transepithelial migration of mononuclear cells from the bloodstream to the airways. 4 Accordingly, administration of KD1, or a more potent KD1 mutant, may conceivably regulate these and other pathological processes dependent upon the activity of serine proteinases. In addition, the availability of human KD1 generated in these studies will facilitate x-ray crystallographic studies of either this inhibitor alone or in complex with serine proteinases, and these studies are currently ongoing in our laboratories.
In summary, these studies provide the initial, definitive evidence that the first Kunitz-type domain of human TFPI-2 contains all the structural elements for the inhibition of a variety of serine proteinases and underscores the importance of critical residues in its P 6 -P 5 Ј position in its inhibitory activity toward these proteinases. In addition, these studies reveal the importance of the Asp and Tyr residues at the P 6 and P 5 positions in the reactive site region of KD1 that appears to confer specificity for plasmin inhibition by TFPI-2.