Structural and Mutational Analyses of the Molecular Interactions between the Catalytic Domain of Factor XIa and the Kunitz Protease Inhibitor Domain of Protease Nexin 2 *

Factor XIa (FXIa) is a serine protease important for initiating the intrinsic pathway of blood coagulation. Protease nexin 2 (PN2) is a Kunitz-type protease inhibitor secreted by activated platelets and a physiologically important inhibitor of FXIa. Inhibition of FXIa by PN2 requires interactions between the two proteins that are con-fined to the catalytic domain of the enzyme and the Kunitz protease inhibitor(KPI)domainofPN2.RecombinantPN2KPIandamutant form of the FXI catalytic domain (FXIac) were expressed in yeast, purified to homogeneity, co-crystallized, and the structure of the complex was solved at 2.6 Å (Protein Data Bank code 1ZJD). In this complex, PN2KPI has a characteristic, disulfide-stabilized double loop structure that fits into the FXIac active site. To determine the contributions of residues within PN2KPI to its inhibitory activity, selected point mutations in PN2KPI loop 1 11 TGPCRAMISR 20 and loop 2 34 FYGGC 38 were tested for their ability to inhibit FXIa. The P1 site mutation R15A completely abolished its ability to inhibit FXIa. IC 50 values for the wild type protein and the remaining mutants were as follows: PN2KPI

A variety of important control mechanisms exist for regulating the activity of coagulation proteases in plasma. Several members of the serpin family have been proposed as physiological regulators of FXIa activity in plasma, including C1 inhibitor (11,12), ␣-1-protease inhibitor (13,14), antithrombin III (15), ␣-2-antiplasmin (16), and protease nexin 1 (17). However, a platelet secretory protein and member of the class of Kunitz-type inhibitors, protease nexin 2 (PN2), has recently been shown to be a much more potent and physiologically relevant FXIa inhibitor based on detailed kinetics studies (18 -23). PN2 is a ϳ120-kDa isoform of the Alzheimer ␤-amyloid protein precursor (APP) that contains a Kunitz-type serine protease inhibitor (PN2KPI) domain (22). Platelets are an important source of several isoforms of APP, including the APP 751 isoform of PN2 (23,24). Full-length APP is membraneassociated (25) and is processed by proteases in platelets (26,27). Upon platelet activation by physiological stimulators, PN2 is secreted from ␣-granules into plasma and inhibits FXIa (22)(23)(24), suggesting a role for this protein in blood coagulation. PN2 is a slow, tight binding inhibitor of FXIa with K i of ϳ300 -500 pM (18,20,22). The KPI domain of PN2 is 57 amino acids in length (Glu 289 -Ile 345 in the 751-amino acid isoform of PN2) and is known to contain the entire FXIa inhibitory function of PN2 (19 -21, 28). Similarly, all of the information required for PN2KPI inhibition of FXIa is contained within the catalytic domain, residues 370 -607 (19 -21, 28).
We have determined the crystal structures of the catalytic domain of FXIa (rhFXI-(370 -607)) with ecotin, an Escherichia coli serine protease inhibitor (29), as well as benzamidine, a small molecule inhibitor of trypsin-like serine proteases. 5 In the present study, we report the crystal structure of a complex of FXIac (FXI-(370 -607)-S434A,T475A,C482S) and PN2KPI. This structure provides insight into the interaction between the catalytic domain of FXIa and PN2KPI. We have utilized this structural information to carry out a detailed mutational analysis to identify the key residues of PN2KPI required for inhibition of FXIa.

EXPERIMENTAL PROCEDURES
Preparation of the Recombinant FXI Catalytic Domain-The recombinant human FXI catalytic domain, amino acid residues 370 -607 (in FXI numbering) with three mutations (S434A, T475A, and C482S in FXI sequence numbering; S75A, T115A, and C123S in chymotrypsin sequence numbering), was expressed as a secreted protein in the methylotropic yeast Pichia pastoris. For ease of discussion, recombinant human FXI-(370 -607)-S434A,T475A,C482S will be termed FXIac, whereas the unmutated form will be referred to as FXI-(370 -607), consistent with our previous publication (29). The protein was cloned, expressed, and purified as described previously. Throughout this paper, the chymotrypsin sequence numbering system is used for FXIac in order to be consistent with other published data on trypsin-like serine proteases (see supplemental material in Jin et al. (29) for the corresponding residue numbers for the FXI sequence and chymotrypsin numbering system) 5 .
Recombinant PN2KPI Wild Type (WT) Domain Gene Segment and Its Mutant Gene Constructs-The full-length human PN2 gene in pcDNA3.1 vector (kind gift from William E. Van Nostrand, SUNY, Stony Brook, NY) served as template for the synthesis by PCR of the WT PN2KPI domain. WT PN2KPI 5Ј primers had an XhoI site followed by a Kex2 signal cleavage site, and 3Ј primers had stop codons followed by a NotI site. This WT PN2KPI PCR product was restriction-digested with XhoI and NotI (New England BioLabs, Inc.), ligated into the similarly treated yeast expression vector, pPICZ␣A (Invitrogen), and propagated in XL1-Blue bacteria. PCR-based site-directed mutagenesis (QuikChange; Stratagene), using mutagenic primers, was used for incorporating desired mutations by inserting codons preferentially used by yeast for appropriate amino acids in the WT PN2KPI sequence. The pPICZ␣A plasmids containing WT PN2KPI and mutant inserts were sequenced from 5Ј to the ␣-mating factor secretion signal using an AOX5 primer to confirm mutations at desired sites as well as the reading frame integrity.
Recombinant PN2KPI Expression in Yeast-pPICZ␣A plasmid constructs containing WT PN2KPI and mutant gene inserts were treated with the SacI (New England BioLabs, Inc.) restriction enzyme. The linearized plasmids were incorporated into the methylotrophic yeast P. pastoris X33 (Invitrogen) competent cell genome (Invitrogen) by recombinational cloning. Selected yeast clones were lysed either by repeated heat-thaw cycles or by lyticase (Sigma), and the lysates were analyzed by PCR using PN2KPI-specific forward and reverse primers for correct insert size. Successful clones were grown in yeast growth medium containing glycerol, BMGY (buffered medium containing glycerol and yeast nitrogen base) at 30°C until obtaining sufficient cell density. These cells were transferred to methanol-containing expression medium BMMY (buffered medium containing methanol and yeast nitrogen base) and incubated at 30°C for 96 h under vigorous agitation, supplemented with 0.5% methanol every 24 h.
Purification of Recombinant WT and Mutant PN2KPI Domains-Yeast cultures containing WT PN2KPI domain and mutant proteins were centrifuged to remove yeast cells. The supernatant was precipitated at room temperature using saturating concentrations of ammonium sulfate and centrifuged at 11,000 ϫ g for 30 min. Pellets were resuspended in 50 mM Tris buffer, pH 7.8. These samples were loaded onto large desalting columns (G25 16/60; Amersham Biosciences) followed by anion exchange chromatography on 20-ml columns (HiTrap Q HP; Amersham Biosciences) in 50 mM Tris, pH 7.8. The eluted samples were concentrated in dialysis tubing (Float-a-lyser, Spectrum Laboratories, Inc.) using cross-linked sodium polyacrylate gel (Spectragel; Spectrum Laboratories, Inc.) as an external dehydrant at 4°C. Concentrated samples were loaded onto gel filtration columns (HiLoad 30, 16/60; Amersham Biosciences). Recovered samples were pooled, and the protein concentration was estimated by the bicinchoninic acid assay (Pierce). All PN2KPI proteins were analyzed by SDS-PAGE to determine the size followed by Western blotting and probing blots using rabbit anti-APP polyclonal antibody (Chemicon International). To confirm mutations at desired sites, purified PN2KPI proteins were sequenced from the N terminus by Edman degradation using a PerkinElmer Life Sciences protein sequencer and amino acid analyzer (protein facility, Iowa State University).
Crystallization and Structure Determination of the FXIac-PN2KPI Complex-The FXIac-PN2KPI complex was formed by incubating stoichiometric quantities of FXIac and PN2KPI at 4°C and isolated by Superose 12 column chromatography (HR10/300; Amersham Biosciences) at a flow rate of 1 ml/min in a buffer containing 20 mM Tris-HCl, pH 7.8, and 0.1 M NaCl on a BioCAD 700E. The FXIac-PN2KPI complex was concentrated using a Centricon YM-10 concentrator (Amicon) to an absorbance of 36 at 280 nm with a 1-cm path length. The complex was crystallized at 10°C by hanging drop vapor diffusion from 4 M sodium formate (Crystal Screen 33; Hampton Research). Rodlike crystals were obtained in 2-3 days with dimensions of 0.2 ϫ 0.2 ϫ 0.5 mm. Diffraction data, to 2.6 Å resolution, were measured on an R-AXIS IV ϩϩ imaging plate detector with an RU-H3R generator and an X-stream TM 2000 low temperature system (Rigaku/MSC) and processed by HKL2000 (30). The crystal belongs to the space group P3 2 21 with unit cell parameters of a ϭ b ϭ 92.8 Å and c ϭ 107.0 Å. There was one FXIac-PN2KPI complex in the asymmetric unit.
The structure of FXIac-PN2KPI was solved by molecular replacement using AMoRe (31). The initial search model was the catalytic domain of FXI from the FXIac-benzamidine structure (Protein Data Bank code 1ZHM) 5 with all of the side chain atoms beyond C ␤ and all water molecules truncated. A clear solution was obtained using data of 10 to 5 Å resolution with a correlation coefficient of 0.591 and R factor of 0.495. Structure refinement was carried out using CNX (Accelrys), and model building was performed in Quanta (Accelrys). The electron density map clearly showed additional densities for PN2KPI. When the catalytic triad of FXIac (His 413 , Asp 462 , and Ser 557 in FXI numbering; His 57 , Asp 102 , and Ser 195 in chymotrypsin numbering) was superimposed with that of thrombin, derived from the structure of thrombin in complex with bovine pancreatic trypsin inhibitor (BPTI) (Protein Data Bank code 1BTH) (32), the BPTI backbone fitted into the PN2KPI density. Clear densities for the side chains of PN2KPI were observed after changing the BPTI sequence to alanine, combining the polyalanine coordinates with the coordinates of FXIac and further refining with CNX. The structure of the FXIac-PN2KPI complex was obtained after several rounds of refinement and model building. Detailed data and refinement statistics are listed in TABLE ONE.
Determination of IC 50 Values of FXIa Inhibition by PN2KPI-Increasing concentrations of the WT PN2KPI domain or its mutants (0 -100 nM) in 50 mM Tris, 150 mM NaCl, 0.5% bovine serum albumin, pH 7.5, were incubated with FXIa (0.1 nM) in a 285-l volume for 30 min at 37°C in a microtiter plate for establishing equilibrium between the inhibitor and the enzyme. To this preincubation mixture, 15 l of substrate L-pyro-Glu-Pro-Arg-p-nitroanalide-HCl (S-2366; Chromogenix) was added to a final concentration of 0.25 mM. Initial reaction velocity readings for 20 min at 37°C in a microplate reader (Molecular Devices Thermo Max) were converted to the fraction of amidolytic activity remaining. Values of IC 50 were determined using KaleidaGraph version 3.5 software.
Activated Partial Thromboplastin Time (APTT) Assay-In plastic microcuvettes (Sigma), 50 l of normal pooled plasma and 25 l of APTT reagent (Alexin; Sigma Diagnostics) were incubated for 15 min at 37°C with 50 l of Tris-buffered saline containing increasing concentrations of WT PN2KPI or mutants. Twenty-five l of CaCl 2 was added (6.7 mM, final concentration) to initiate clot formation, and clotting time was determined in an Amelung KC4 microcoagulometer (Amelung GmbH, Germany).

FXI Catalytic Domain Structure in the FXIac-PN2KPI Complex-In
order to identify the key interactions between PN2KPI and the FXIa catalytic domain, we determined the structure of the FXIac-PN2KPI complex to 2.6 Å resolution. The recombinant catalytic domain of FXI used for the crystallization experiment contains the following mutations: S75A (S434A, FXI sequence number in parenthesis) and T115A (T475A) to remove the glycosylation sites and C123S (C482S) to remove an unpaired Cys residue. 5 When the C␣ atoms of residues Interactions between FXIac and PN2KPI-PN2KPI has extensive interactions with FXIac via two loops (Fig. 4A), those containing resi-dues 11 TGPCRAMISR 20 (P5-P5Ј) and residues 34 FYGGC 38 , linked by a disulfide bond between Cys 14 and Cys 38 (Fig. 4B). The equivalent two loops of BPTI are residues 11 TGPCKARIIR 20 and 34 VYGGC 38 . The general orientation of the two loops in PN2KPI is similar to the orientation of the corresponding loops in BPTI. Although the primary binding site of ecotin also interacts with FXIac via two loops (29), the orientation of the two loops in PN2KPI and ecotin is different relative to the active site of FXIa, with parts of the loops (P1Ј-P3) overlapping (Fig. 2).  (TABLE TWO).

Purification and Characterization of PN2KPI WT and Mutant
Domains-All mutants and WT PN2KPI domains were expressed in a Pichia yeast expression system and purified as described under "Experimental Procedures." The expression levels of all mutants and WT PN2KPI were between 7 and 10 mg/liter, with the exception of the mutations at Tyr 35 (Y35A and Y35I), which yielded negligible quantities of proteins and hence were insufficient for detailed kinetic studies. Purified proteins gave a sharp, single band at 6.3 kDa by silver-stained SDS-PAGE in the presence of reducing agent, ␤-mercaptoethanol. Edman degradation results confirmed the expected amino acid sequences of these recombinant proteins.
The WT PN2KPI was characterized by a prolonged transient phase for inhibition of FXIac, typical of slow, tight binding Kunitz inhibitors (20). Increasing concentrations of either WT PN2KPI or mutant PN2KPI domains, incubated with 0.1 nM FXIac, resulted in plots of fractional amidolytic activity versus the inhibitor concentration (Figs. 5-8), which correlated well with the results of an assay to measure the in vitro prolongation of clotting time (APTT) by WT and mutant PN2KPI domains (Fig. 9).
P1 Site (Arg 15 ) PN2KPI Mutations (Fig. 5)-The IC 50 value for WT PN2KPI domain inhibition of FXIac was 1.28 nM. When the P1 site amino acid (Arg 15 ) was mutated to Ala (R15A), the mutant protein completely lost its inhibitory activity (Fig. 5A). However, a substitution with Lys in the P1 site (R15K) resulted in only partial loss of inhibitory activity (IC 50 of 42 nM). This substitution resulted in a 33-fold reduction in inhibitory activity as compared with WT PN2KPI. Similarly, in the APTT assay, the R15A mutant was inert (i.e. it failed to prolong the clotting time), whereas R15K mutant showed a partial loss of inhibitory activity (Fig. 5B).  ( Fig. 6A), in agreement with the results of the APTT assay (Fig. 6B). On the contrary, Glu (IC 50 of 13.5 nM) and Gln (IC 50 of 11.79 nM) mutations disrupted the function 11-and 9-fold, respectively (Fig. 6A), results also confirmed by those of the APTT assay (Fig. 6B).   similar results in the APTT assay (Fig. 7B). In contrast, mutation at the P5Ј site with an Ala replacement for Arg (R20A) resulted in a 4-fold loss of activity (IC 50 of 5.67 nM) compared with the WT PN2KPI (Fig. 7A) and a similar observation of loss of activity in the APTT assay (Fig. 7B). F34A PN2KPI Mutation in Loop 2 (Fig. 8)-In addition to PN2KPI loop 1 amino acids (Figs. 5-7), the loop 2 region amino acid, Phe 34 , was also replaced with Ala, resulting in an 8-fold loss of activity (IC 50 of 9.85 nM) as compared with WT PN2KPI (IC 50 of 1.28 nM), which agrees with the loss of activity in the APTT assay (Fig. 8, A and B).
Correlation between Results of Kinetic Assays and APTT Results (Fig.  9)-The results of our in vitro kinetic assays to measure the inhibitory properties of PN2KPI mutants were compared with the prolongation of clotting times measured by APTT assays (Fig. 9). The APTT assay monitors the activity of intrinsic pathway of blood coagulation, in which the primary target for PN2KPI, FXIa, is located. These two methods of monitoring the inhibitory properties of PN2KPI mutants were found to correlate well (r ϭ Ϫ0.8).

DISCUSSION
The Kunitz-type inhibitor, PN2, is postulated to have an important role in the regulation of FXIa at the site of blood vessel injury where platelets form a hemostatic thrombus and play an essential part in the activation of FXI by thrombin (19, 20, 22, 23). Thus, it has been observed that normal human plasma contains very little (Ͻ60 pM) PN2 (i.e. concentrations well below the reported K i value) (300 -500 pM) for FXIa inhibition by PN2 (19,20,(22)(23)(24). However, the protein is secreted from platelet ␣-granules in sufficient quantities (2-30 nM) to result in rapid and complete inhibition of FXIa in solution (20,23). In contrast, the FXIa bound to the GPIb-IX-V complex on the activated platelet surface (5, 6) is completely protected from inactivation by PN2 (21,35). Therefore, it is likely that PN2 has an important role in regulating the initiation of the consolidation phase of blood coagulation and localizing it to the hemostatic thrombus. These facts emphasize the importance of understanding the mechanism and structural determinants of FXIa inhibition by PN2.
Previously in this laboratory, a detailed quantitative study (21) of FXIa and FXIac with PN2 and PN2KPI suggested that the totality of interactions that result in inhibition reside within the catalytic domain of FXIa and the KPI domain of PN2. Thus, in the current study, it is assumed that co-crystals of PN2KPI and FXIac domains will yield as much information as studies of the respective complete molecules.
FXI Catalytic Domain Structure and Interactions between FXIac and PN2KPI-Binding of PN2KPI causes a slight enlargement of the active site of FXIac compared with the active site containing benzamidine (Fig.  1B). This is similar to the enlargement observed when the structures of rhFXI-(370 -607)-ecotinM84R (Protein Data Bank code 1XX9) and FXIac-benzamidine (Protein Data Bank code 1ZHR) 5 were compared. The catalytic triad and Asp 189 at the bottom of the S1 pocket superimpose well in the two structures (Fig. 1B), whereas Lys 192 is in different  conformations. Lys 192 in the FXIac-benzamidine structure interacts with a sulfate molecule from the crystallization medium. It forms hydrogen bonds with PN2KPI in the FXIac-PN2KPI structure (see below) (Fig. 4B). The hydrogen bond between Glu 98 and His 174 in FXIac-benzamidine structure is absent in the FXIac-PN2KPI structure. The aliphatic portion of the Glu 98 side chain is in van der Waals contact with Trp 215 as it is in the FXIac-benzamidine and the rhFXI-(370 -607)-ecotinM84R structures, whereas the carboxylic acid group is packed against Cys 38 and Pro 13 of PN2KPI. Glu 217 and Arg 224 point away from the S4 pocket, similar to their orientation in the rhFXI-(370 -607)-ecotinM84R structure; however, they bend inward in the FXIac-benzamidine structure. All other residues surrounding the active site do not move significantly, including Tyr 59A , although the latter interacts with PN2KPI (see below).
The P1 residue (Arg 15 ) of PN2KPI extends into the S1 pocket of FXIac. One nitrogen atom of the guanidinium group forms a hydrogen bond with Asp 189 of FXIac, whereas the other nitrogen atom forms a hydrogen bond with the main-chain carbonyl oxygen atom of Gly 218 in FXIac (Fig. 4B) (Fig. 4B). Most interestingly, Pro 13 (P3) of PN2KPI bends the main chain of PN2KPI. That prevents Gly 12 (P4) from interacting with the S4 pocket of FXIac   . Comparison of kinetic assay with APTT assay data for mutants. Equilibrium inhibition kinetics data for all PN2KPI mutants determined by microtiter plate assays were compared with APTT assay data for the same mutants. These two methods correlate well (r ϭ Ϫ0.8).
formed by Trp 215 , His 174 , and Glu 98 . Fig. 2B clearly shows that the C␣ atoms of the P3 residue from PN2KPI and ecotin are located close to each other; PN2KPI Pro 13 bends the loop away from the direction that the ecotin loop is heading. Other members of the KPI family, such as BPTI (32) and soybean protease inhibitor (36), also have proline at the P3 position and adopt the same backbone conformation as PN2KPI. PN2KPI forms a turn around Pro 13 and allows Lys 192 of FXIac to insert into PN2KPI and form two hydrogen bonds with carbonyl oxygen atoms of Cys 14 (P2) and Gly 12 (P4) (Fig. 4B). Both FXIa and FVIIa contain a lysine at position 192, whereas that position is either Glu or Gln in the other serine proteases in the coagulation cascade. After surveying all of the serine protease structures in complex with APPI and BPTI in the Protein Data Bank, we observed that Glu/Gln 192 of the serine protease forms either no hydrogen bond or one hydrogen bond with the carbonyl oxygen atom of Cys 14 (P2) of APPI or BPTI. In the structure of FVIIa with a BPTI mutant (Protein Data Bank code 1FAK (37)), which mimics the first Kunitz domain of tissue factor pathway inhibitor (the natural inhibitor of the tissue factor FVIIa complex), Lys 192 is hydrogen-bonded to Asp 11 (a mutated residue, T11D) and the carbonyl oxygen atom of Tyr 34 (V34Y) of the BPTI mutant. We suspect that Lys 192 plays an important role in FXIa activity and is essential for the interaction between FXIa and PN2KPI.
In PN2KPI, the prime side residues (located C-terminal to the scissile bond, leaving group for the proteolysis reaction) 16 -20 have intricate interactions with residues 36 -32 by forming an anti-parallel ␤-sheet. This hydrogen bond network as well as the side chain van der Waals packing in PN2KPI stabilizes the leaving group and thus prevents the hydrolysis reaction from occurring. This is consistent with the clogged gutter mechanism for protease inhibitors (38). Although PN2KPI binds to FXIa very tightly (IC 50 of 1.28 nM) in a substrate-like manner, the combination of the hydrogen bond network, an acyl-enzyme (detected in subtilisin-chymotrypsin inhibitor 2 complex (38), not yet tested in FXIac-PN2KPI complex), and the correct orientation of the religating amide of PN2KPI arrest the hydrolysis reaction and make PN2KPI an effective inhibitor rather than substrate.
There are two unique loops in FXIa that are longer than those of other serine proteases (except thrombin, which has a signature S2 insertion loop (TABLE THREE)). These loops contain residues 59 -64 and residues 36 -37D, and interact with PN2KPI (Fig. 4A). As shown in Fig. 4B, Tyr 59A of FXIac is in van der Waals contact with Tyr 35 , Ile 18 (P3Ј), and Arg 20 (P5Ј) of PN2KPI, whereas Arg 37D is hydrogen-bonded with the main-chain nitrogen atom of Met 17 (P2Ј) and within hydrogen bond distance but orienting poorly to the O␥ of Ser 19 (P4Ј) of PN2KPI. Among the residues in the prime side, PN2KPI Ala 16  Comparison of PN2KPI Structure with Other APPI Structures-Superimposition of C␣ atoms of PN2KPI with all of the APPI crystal structures reveals that Arg 15 (P1) inserts into the S1 pocket of serine proteases. The different conformations of Arg 15 are dictated by the different environment of the S1 pocket in different serine proteases. Thr 11 (P5) is a different rotamer in PN2KPI than the other APPIs. Thr 11 does not form polar interactions with other residues. It is not clear why Thr 11 adopts a different conformation in PN2KPI. It is close to Lys 192 (Fig. 4B), a residue that is unique in the active site of FXIa. PN2KPI Met 17 (P2Ј) is in different conformation in all four structures compared herein. The most interesting changes are in PN2KPI Arg 20 and Asp 46 , which move in a concerted fashion away from FXIac relative to those in the APPI structures (Fig. 3B). These changes are clearly caused by the presence of Tyr 59A of FXIac; therefore, those are unique features of the PN2KPI interaction with FXIa.
Mutation Studies on PN2KPI Domain-Selective Ala mutations at P3, P1, P2Ј, P4Ј, and P5Ј sites of loop 1 and Phe 34 of loop 2 of PN2KPI domain were studied, and extended mutations at P3 and P1 sites were also done for understanding the molecular mechanism of FXIa inhibition by PN2KPI. All mutants (except the P1 site mutant, R15A) were found to exhibit varying degrees of inhibitory activity against FXIa (TABLE TWO).
P1 Site (Arg 15 ) of PN2KPI-Among trypsin-like serine proteases that can be inhibited by KPIs, the S1 subsite in the binding pocket of FXIa can accommodate preferentially either Arg or Lys from the P1 site of the inhibitor. The primary interaction between Arg 15 of the PN2KPI and Asp 189 of FXIa is an optimal fit. This observation is substantiated by our experimental evidence that an Ala replacement for Arg at position 15 (R15A) in PN2KPI resulted in complete loss of inhibitory activity and that the R15K mutation resulted in a ϳ33-fold reduction in inhibitory activity. In the R15A mutant, the side chain is shortened as well as lacking the positive charge, thus disrupting the primary interaction. However, in the R15K mutant, a shortened side chain (compared with Arg) may not directly interact with the carboxyl group of Asp 189 but could still have an interaction with Asp 189 using a water molecule as a bridge, similar to the P1 Lys interaction in BPTI-trypsin complex (34). Both FXIa and plasma kallikrein have Ala at residue 190 in the S1 pocket instead of Ser as in trypsin, lacking an additional hydrogen-bonding interaction with the P1 residue. A Lys mutation at Arg 15 in the inhibitor combined with the absence of a Ser at position 190 in FXIa could pose severe constraints in their interactions, similar to plasma kallikrein, which less favorably accommodates R15K compared with trypsin (39). This may partly explain why R15K still retains its FXIa-inhibitory function, albeit with a ϳ33-fold reduction compared with the WT PN2KPI molecule.
Earlier studies have shown that the P1 site mutations of PN2KPI result in enhanced potency toward other proteases. Two separate single amino acid substitutions at the Arg 15 position of PN2 (R15K and R15V) resulted in increased inhibitory potency for human plasmin (40) and human neutrophil elastase (41), with decreased inhibitory activity against FXIa. BPTI has a Lys at position 15 (P1 site). When this site was mutated to Arg (K15R; called pseudo-WT BPTI), a severalfold increase in K a was observed with several key coagulation enzymes, such as FXa, FXIIa, plasma kallikrein, protein C, and ␣-thrombin (42). This indicates the importance and preference of Arg at the P1 site of PN2KPI for coagulation enzyme inhibition. Therefore, P1 of PN2KPI is important for its selectivity and potency in the inhibition of FXIa. P3 Site (Pro 13 ) of PN2KPI-There are two Pro residues in PN2KPI, one at position 13 (P3 site) and the other at position 32, each located at the beginning of characteristic loop regions that are involved in interaction with FXIac as observed in the PN2KPI-FXIac co-crystal structure. Pro 13 is a highly conserved residue among the Kunitz family of protease inhibitors. The ␤ carbon in Pro 13 is oriented toward a solventexposed region; therefore, this site could accept a variety of amino acids (34). A study utilizing phage display mutagenesis, where a large number of random mutations are generated within the PN2KPI domain, showed that amino acids found at position 13 were highly dependent upon the amino acid at position 39 (43). Having Gly at 39 allowed several substitutions at position 13, whereas having a non-Gly residue at 39 rendered Pro the requisite residue at position 13. In the current study, having Gly at position 39, varieties of mutations at position 13 should be tolerated. Here, the Pro 13 site was substituted individually with Ala, Glu, Gln, Lys, or Arg. The substrates of FXIa have either Asp or Lys at the P4 position. The charged residues were chosen to mimic FXIa substrate.
The P13A mutant showed a 4.6-fold decrease in IC 50 against FXIa, whereas P13R and P13K had a marginal decrease of 1.3-and 1.8-fold, respectively. Without structures of the mutants, we speculate that a larger drop of IC 50 with P13A might be caused by the lack of bulk of Ala compared with Pro to occupy the space and/or slight change of backbone conformation. Because the S4 pocket of FXIa has Glu 98 and Glu 217 , P13R and P13K mutants might interact with the negatively charged residues and therefore rescue the inhibition function of PN2KPI. However, P13E and P13Q are less favored at the P3 site, resulting in a 9 -11fold (IC 50 of 13.5 and 11.8 nM, respectively) reduction in activity against FXIa compared with the WT PN2KPI. In the crystal structure of FXIac in complex with ecotin that has P2Ј to P5 mutated to FXIa substrate sequence (29), the Asp of P4 is interacting with His 174 in the S4 pocket. Because the C␤ of Pro 13 in PN2-KPI points toward S4 in the FXIa-PN2KPI structure but its C␣ locates further away from His 174 compared with P4 of ecotin, we chose to mutate Pro 13 to the longer Glu in order to mimic FXIa substrate Asp at P4. Apparently, Glu failed to orient as Asp in the FXI-ecotin complex structure; moreover, it might exhibit charge repulsion with the negatively charged residues in S4, resulting in the loss of inhibitory activity.
P2Ј (Met 17 ), P4Ј (Ser 19 ), and P5Ј (Arg 20 ) Sites and Loop 2 Residue 34 (Phe 34 ) of PN2KPI-There are defined complementing hydrophobic regions, contributed by Met 17 (P2Ј), Ile 18 , and Phe 34 from PN2KPI and Leu 39 , Tyr 143 , and Ile 151 from FXIa (S2Ј pocket) in the co-crystal complex. Of these six hydrophobic residues from either molecule, it appears that not all amino acids from PN2KPI are equally important in contributing to this hydrophobic region in maintaining the inhibitory function. M17A mutation at the P2Ј site results in no loss of inhibitory function (1.3-fold; IC 50 of 1.62 nM), whereas mutation at loop 2 amino acid Phe 34 to Ala (F34A) showed a significant loss, 7.7-fold (IC 50 of 9.85 nM), of inhibitory activity. From the perspective of FXIa, Tyr 143 is the only residue that interacts with Phe 34 of PN2KPI, via a van der Waals interaction. The side chain of Phe 34 also packs closely to the side chain of Met 17 of PN2KPI. The F34A mutant puts Tyr 143 of FXIac beyond its reach, resulting in complete loss of contact, and this mutation might also destabilize Met 17 in PN2KPI. This was evidenced from an elevation in IC 50 for the F34A mutation. Mutation at Met 17 alone is insufficient to create a large scale disturbance in structural integrity as evidenced from near total retention of FXIa-inhibitory activity. It can be inferred from these results that a double mutation at Met 17 and Phe 34 could possibly disturb the hydrophobic region created by Met 17 , Ile 18 , and Phe 34 , leading to a likely large scale functional loss of PN2KPI activity.
The P4Ј site amino acid Ser 19 is close to Arg 37D of FXIa, but the orientation of O␥ prevents it from forming a hydrogen bond with the guanidinium group of Arg 37D . The backbone nitrogen atom of Ser 19 of PN2KPI interacts with the OH group of Tyr 39 of trypsin in the PN2KPItrypsin complex (34). Such an interaction with FXIa is not possible, since Arg 37D occupies an equivalent position in FXIa. The S19A mutant resulted in a marginal decrease of 1.5-fold (IC 50 1.86 nM) inhibitory activity against FXIa compared with WT PN2KPI. This meager reduction in inhibitory function suggests that the P4Ј site Ser 19 residue has a minor role in FXIa inhibition.
The P5Ј site residue Arg 20 interacts only with Tyr 59A (a unique residue in FXIac), since no other residue from FXIa is within the vicinity of ϳ4 Å. Replacement with Ala in this location (R20A) causes the inhibitor to lose 4.4-fold of its activity compared with the WT PN2KPI, indicating an important contribution for this residue in FXIa inhibition.
The crystal structure of FXIac-PN2KPI provides detailed structural information for the interaction between FXIa and PN2KPI. Our current studies delineate details of some of the molecular interactions between PN2KPI and FXIa. Not all of the PN2KPI amino acids identified from the co-crystal structure appear to be required for the inhibitory function. For example, mutating Met 17 and Ser 19 to Ala had no major effect on their inhibition of FXIa. Among the mutants tested, the P1 site Arg 15 is the most important in FXIa inhibition, followed by Phe 34 , Pro 13 , and Arg 20 . Additional mutagenesis studies are currently in progress in our laboratory to identify residues within FXIa that interact with PN2KPI. The mutational studies in conjunction with the structural analysis reveal the structure-function relationship of PN2KPI inhibition of FXIa. This information, in turn, will be useful to the structure-based ligand design for the discovery of potent and selective small molecule FXIa inhibitors.