Interactions Outside the Proteinase-binding Loop Contribute Significantly to the Inhibition of Activated Coagulation Factor XII by Its Canonical Inhibitor from Corn

Background: Canonical inhibitors interact with cognate proteases through a protease-binding loop. Results: Protease-binding loop of corn Hageman factor inhibitor (CHFI) does not inhibit its cognate enzyme-activated coagulation factor XII (FXIIa). Conclusion: Nonloop regions of CHFI are required for specific inhibition of FXIIa. Significance: CHFI is the first canonical inhibitor whose protease-binding loop is not sufficient for cognate enzyme inhibition.

activated factor XI with a K i of 5.4 ؎ 0.2 M. Our results suggest that the protease-binding loop is not sufficient for the interaction between FXIIa and CHFI; other regions of the inhibitor also contribute to specific inhibition.
Corn Hageman factor inhibitor (CHFI) 2 is a 14-kDa serine protease inhibitor that belongs to the I6 family, according to MEROPS (1). The mature protein consists of 127 amino acids, with a scissile bond between Arg 34 and Leu 35 (2). This bond is involved in interactions with both trypsin and activated coagulation factor XII (FXIIa) (2,3). CHFI selectively inhibits FXIIa (4). FXIIa is a serine endopeptidase in the S1 family, according to MEROPS (1), and is a key enzyme of the contact pathway of plasma coagulation. However, FXIIa fails to play an important role in hemostasis in vivo, because patients and animals deficient in FXIIa do not suffer from bleeding disorders (5)(6)(7)(8). Moreover, studies using experimental thrombosis (9) and ischemic stroke (10,11) models demonstrated that mice deficient in FXIIa were resistant to experimental stress. The restoration of FXIIa to plasma concentrations led to thrombosis formation in mice.
Thus, Factor XIIa represents a promising target for antithrombotic agents because it plays no significant role in hemostasis but is involved in pathological clot formation. This scenario provides the opportunity to develop novel antithrom-botic drugs and nonthrombogenic implantable medical devices (12) based on the selective inhibition of FXIIa with no risk of bleeding. Thus, interactions between FXIIa and CHFI are of great interest because of their high selectivity.
CHFI is believed to be a canonical inhibitor. Canonical inhibitors interact with their cognate proteases via a standard substrate-like mechanism (13). A convex extended protease-binding loop with a canonical conformation is responsible for inhibition (14,15). This loop is highly complementary to the active site of the cognate protease. The scissile bond P1-PЈ1 (15) is located in the outermost part of the loop (16). The P1 amino acid residue is the primary determinant of the specificity of the inhibitor (13). Hydrolysis of the scissile bond does not lead to a substantial loss of inhibitory activity. Both uncleaved (i.e. one-chain) and modified (i.e. two-chain) forms of the inhibitor are active (13). The protease-binding loop of canonical inhibitors is closed, with at least one disulfide bond (17). In rare exceptions (18), this bond is replaced by strong noncovalent interactions. Although the amino acid sequences of the proteasebinding loop vary greatly, inhibitory function is defined by the main chain conformation (13). Canonical inhibitors differ in folding and size, varying from 14 to ϳ200 amino acids (19).
In recent decades, studies of serine protease-canonical inhibitor interactions suggested that the protease-binding loop is a minimal and sufficient base for inhibitory activity. This concept was demonstrated using both synthetic (20,21) and recombinant (22) protease-binding loops from Bowman-Birk inhibitors. Native canonical serine protease inhibitors containing one disulfide bridge have also been described in other species, such as STFI-1 (23) from sunflower and peptides from Odorrana grahami (24,25). The amphibian peptide (ORB) was further shortened to a hendecapeptide trypsin inhibitory loop that not only retained but also drastically increased its initial inhibitory activity against trypsin (K i ϭ 306 M for ORB and K i ϭ 710 nM for the trypsin inhibitory loop) (26). Thus, an isolated proteasebinding loop from a canonical inhibitor appears promising as a base for the design of new serine protease inhibitors.
Although the structure of the CHFI-FXIIa complex is not available, evidence suggests that CHFI is a canonical inhibitor. Both the uncleaved one-chain and cleaved two-chain forms of CHFI are reported to inhibit trypsin (27,28) and FXIIa (3,4). However, the two-chain form exhibits only 20 -25% of the activity of the one-chain form (3,4). The crystal structure (29) revealed that CHFI has a typical protease-binding loop that is closed via a disulfide bond and supported by an additional cysteine bridge.
Based on the available data related to small peptide serine protease inhibitors, we propose that the isolated protease-binding loop of CHFI is a promising primary structure for the development of new FXIIa inhibitors. In this study, we tested the inhibitory activity of a synthetic peptide that resembles the CHFI protease-binding loop and five recombinant truncation mutants of CHFI. Surprisingly, the cyclic peptide CHFI-2, which represents the CHFI protease-binding loop bridged with one disulfide bond, is unable to inhibit FXIIa but retains its inhibitory activity against bovine pancreatic trypsin and activated coagulation factor XI (FXIa). Our results suggest that regions outside the protease-binding loop of CHFI are likely to contribute to its inhibitory potency toward FXIIa. We also report the first simple protocol for soluble expression of CHFI in Escherichia coli.

EXPERIMENTAL PROCEDURES
Strains and Vectors-The host strain E. coli Rosetta-Gami 2 DE3 (EMD Millipore Corporation, Billerica, MA) was used. The expression vector pET-28a was also obtained from EMD Millipore. Recombinant CHFI and its fragments were expressed under the control of a T7 promoter and induced using isopropyl ␤-D-thiogalactopyranoside.
Primer Design, PCR Amplification, and Site-directed Mutagenesis-The pLA-TA plasmid containing a synthetic version of the CHFI gene with codon usage optimized for E. coli was obtained from Eurogen (Moscow, Russia). The control CHFI protein from Zea mays was obtained from Enzyme Research Laboratories (South Bend, IN).
The pLA-TA plasmid containing the CHFI gene was used as a PCR template for the construction of the pET28a vector containing the CHFI gene. The forward and reverse primers used in this process are as follows, with mismatches in bold type: 5Ј-TGCGGATCCTCTGCTGGTACCAGCTG-3Ј and 5Ј-TGC-AAGCTTAGATCTGCTCGGCATGG-3Ј, respectively. Specific oligonucleotides were designed to perform PCR mutagenesis for each recombinant CHFI fragment from the pET28a/ CHFI template. PCR fusion was achieved as previously described (30), using forward and reverse primers and two mutagenesis primers for each mutant gene (Table 1). Vent DNA-polymerase was obtained from New England Biolabs (Ipswich, MA). The synthetic peptide CHFI-2 was obtained from "Syneuro" (Moscow, Russia).
Restriction Cloning-The BamHI and HindIII restriction enzymes (New England Biolabs) were used to digest the obtained PCR products and the pET28a vector. T4-DNA ligase (New England Biolabs) was used to ligate the pET28a vector to each CHFI fragment gene.
Transformation of Host Cells-E. coli Rosetta-Gami 2 DE3 cells were transformed with the resulting vectors via heat shock transformation, as previously described (31), and colonies were selected on LB agar plates containing 25 g/ml of kanamycin. Positive clones were selected by induction in 4 ml of LB containing 1 mM isopropyl ␤-D-thiogalactopyranoside for 2.5 h at 37°C and confirmed by DNA sequencing.
Expression and Detection of Mutant Proteins-The expression and detection of CHFI mutants were accomplished by inducing the expression of E. coli Rosetta-Gami 2 DE3 cells harboring the appropriate plasmid. A single colony was inoculated into 4 ml of LB containing 25 g/ml of kanamycin and grown overnight at 37°C. A total of 1.5 liters of LB medium containing the appropriate antibiotics was inoculated with a 1:50 dilution of an overnight culture. The cells were grown at 37°C until they reached an optical density at 600 nm of ϳ0.6 -0.7. Expression was induced via the addition of 0.02 mM isopropyl ␤-D-thiogalactopyranoside and subsequent incubation at 25°C for 20 h. The samples were analyzed via 12% SDS-PAGE and stained with R-250 Coomassie Brilliant Blue.
Purification of CHFI Mutants-After overexpression of the CHFI mutant proteins, E. coli cells were harvested by centrifu-gation for 20 min at 5,000 ϫ g and 4°C. The cells were subsequently lysed in a solution of 25 mM Tris-HCl, pH 8.0, 300 mM NaCl, 1 mg/ml of lysozyme, 10 mM imidazole, and 0.1 mM PMSF with sonication. The lysates were centrifuged for 40 min at 10,000 ϫ g and 4°C. The supernatants were filtered through a 0.2-m cellulose acetate membrane (EMD Millipore) and purified on a Ni 2ϩ -agarose column (Qiagen) using a gravity flow format. The column was washed with a solution of 25 mM Tris-HCl, pH 8.0, 300 mM NaCl, and 50 mM imidazole. The samples were subsequently eluted in a solution of 25 mM Tris-HCl, pH 8.0, 300 mM NaCl, and 200 mM imidazole and further analyzed via SDS-PAGE. The concentrations of the purified proteins were measured by UV absorbance at 280 nm. The extinction coefficient for each protein was calculated using Vector NTI Advance Software (Invitrogen).
Inhibitory Activity Measurements-The purified proteins were diluted to reduce the imidazole concentration and concentrated on 3-kDa Millipore filters. The final protein samples were obtained in a solution of 110 mM Tris-HCl, pH 8.0, 250 mM NaCl, and 4 mM imidazole.
Chromogenic assays were performed according to the substrate manufacturer's instructions, with minor changes. Recombinant enhanced green fluorescent protein (EGFP) that was obtained under the same conditions (i.e. vector, host, and purification procedure) was used as a negative control. Substrates S2302 (for XIIa), S2765 (for trypsin), and S2366 (for XIa) were obtained from Chromogenix (Orangeburg, NY). Factors XIIa and XIa were obtained from Enzyme Research Laboratories and Hematologic Technologies Inc. (Essex Junction, NY), respectively. Bovine pancreatic trypsin was obtained from Sigma-Aldrich.
Activity assays for factor XIIa, factor XIa, and trypsin were performed at 37°C in reaction buffer (110 mM Tris-HCl, pH 8.0, 250 mM NaCl, and 1% BSA) in 96-well flat-bottomed plates (Corning Inc., Corning, NY). The rate of substrate hydrolysis was recorded at 405 nm using a Sunrise microplate spectrophotometer (Tecan Group AG, Männedorf, Schweiz, Austria). We used 1 nM FXIIa and 500 M S2302 in the FXIIa assays, 200 pM FXIa and 500 M S2366 in the FXIa assays, and 1 nM trypsin and 500 M S2765 in the trypsin assays. To calculate the K i values for the interaction of the CHFI mutants with FXIIa, trypsin, and FXIa, the residual activity of each protease was determined in triplicate for at least six inhibitor concentrations.
The rate of substrate hydrolysis at 405 nm (v) was plotted against the concentration of inhibitor on reversed coordinates to give 1/v (s ϫ OD 405 Ϫ1 ), which is dependent on C (nM or M). Dependence was approximated linearly to obtain parameters a (y intercept) and b (slope).
The inhibitory constant K i for each experiment was calculated from the following transformed Michaelis-Menten equation, where K m is the Michaelis constant provided by the substrate manufacturer, and [S] is the substrate concentration.

JOURNAL OF BIOLOGICAL CHEMISTRY 14111
Statistical Analysis-Unpaired t tests and one-way analysis of variance were performed using GraphPad Prism software. Groups of three or more inhibitory constants representing the mean values of n ϭ 2 or more measurements were compared.
The p values are indicated on the figures.
Structure Preparation-The x-ray structure of CHFI-12345 at 1.95 Å resolution (Protein Data Bank (PDB) (32) code 1BEA (29)) was used. To model the CHFI-1245 mutant, Cys 29 was manually replaced with aspartic acid. To model the CHFI-2 loop, all residues before Cys 20 and after Cys 44 were removed. VMD software (34) was used to prepare the model systems and further analyze the molecular dynamics (MD) trajectories (57). Missing hydrogen atoms were added, and crystallographic water molecules surrounding CHFI-12345 were included in the model. All three models of CHFI and its mutants were surrounded with TIP3P water molecules, which formed a box with boundaries at least 10 Å from the solute. Sodium and chloride ions were added to a final ion concentration of 0.15 M. NAMD 2.9 software (35) was used with a CHARMM36 force field (36). Hydrogen atoms, ions, and water molecules added to CHFI-12345, and side chains of the mutated Asp 29 in CHFI-1245 and CHFI-2 were energy-minimized in 1,000 steps. The coordinates of atoms from the x-ray data were kept fixed. These structures of the inhibitor variants were also used in protein-protein docking and are further referred to as "model x-ray structures." Molecular Dynamics-The water box of the prepared and optimized system was equilibrated during a 1-ns MD run, with all coordinates of CHFI fixed (NPT ensemble, 298 K, 1 atm, periodical boundary conditions). The obtained model system parameters are provided in supplemental Table S1. The preequilibrated systems were fully energy-minimized in 5,000 steps. This step was followed by a 50-ns MD run at 298 K with parameters similar to those used in the equilibrating runs. All atoms were allowed to move. These 50-ns trajectories were used to analyze the conformational changes. To obtain stable conformations of the CHFI variants for further protein-protein docking, after the MD run at 298 K, the systems were slowly cooled to 4 K, with a temperature reduction of 1 K/1 ns. The resulting structures of the CHFI variants are further referred to as "MD structures." The MD simulations were performed using the Lomonosov Moscow State University supercomputer (37).
Homology Model-Because no x-ray structure is currently available for the FXIIa catalytic domain, a homology model was built using the sequence of this domain (UniProt number P00748). The BLAST tool (38) on the UniProt web server (39) was used to find a suitable structure. The highest similarity among the proteins with available crystallographic data were observed for hepatocyte growth factor activator (UniProt number Q04756) (47% similarity). Several structures of the catalytic domain of this protein are available in the PDB, with the best resolution of 2.20 Å for structure PDB code 2R0L (40). This structure was used to build a homology model using Modeler 9v12 (41). To check the validity of the generated FXIIa homology model, we docked the tripeptide D-Pro-Phe-Arg, which is analogous to the FXIIa-specific chromogenic substrate S2302 (D-Pro-Phe-Arg-p-nitroaniline), in the current study. We used the protein-protein docking ClusPro server (42,43) to estimate the initial positioning and Rosetta FlexPepDock for refinement (44,45).

Cloning, Expression, and Purification of Recombinant CHFI and Its
Mutants-To investigate the role of CHFI regions outside the protease-binding loop in the interaction with FXIIa, a number of mutants were generated (Fig. 1) and tested. Based on the published (29) CHFI structure (PDB code 1BEA), the mutants were designed to lack different N-and/or C-terminal CHFI regions and/or disulfide bridges. CHFI-12345 is a fullsize CHFI without any truncations. CHFI-2345 lacks 11 amino acids and the first disulfide bond at the N terminus, and Cys 55 is replaced by Asp. This mutant is identical to a previously described shortened form of CHFI (49) except for the replacement of Cys 55 . Mutant CHFI-1234 lacks 24 amino acids at the C terminus and the fifth disulfide bond, with unpaired Cys 57 replaced by Asp. Mutant CHFI-234 is a combination of the previous two truncations. This mutant lacks 11 and 24 amino acids at the N and C termini, respectively, and the first and the fifth disulfide bonds. Unpaired Cys 55 and Cys 57 were replaced by Asp. CHFI-123 lacks 34 amino acids at the C terminus and the fourth and the fifth disulfide bonds (Fig. 1).
To exclude interactions outside the inhibitor protease-binding loop, a cyclic 25-amino acid peptide was synthesized, with a disulfide bridge at its termini. This CHFI-2 peptide contains amino acid residues 20 -45 of CHFI. Cys 29 was replaced by Asp to avoid unwanted disulfide bond formation between unpaired cysteines. This peptide resembles the CHFI protease-binding loop. To study the effect of Cys 29 replacement in the loop, the CHFI-1245 mutant was designed. This mutant is a full-size CHFI with Cys 29 substituted by Asp.
DNA sequences for CHFI-12345, CHFI-1234, CHFI-2345, CHFI-234, CHFI-1245, and CHFI-123 were amplified from the synthetic CHFI gene using primers with mismatches to introduce BamHI and HindIII restriction sites and a single amino acid substitution. An additional stop codon was introduced before the HindIII restriction site in mutants CHFI-12345 and CHFI-1234; therefore, these mutants carry only an N-terminal His 6 tag (Fig. 1C). After cloning into the pET28a vector, the constructs were transformed into the E. coli Rosetta-Gami 2 DE3 expression host.
The proteins were partially expressed in soluble form (data not shown), and after one-step nickel affinity purification, the samples were run on SDS-PAGE (Fig. 2). The theoretical molecular masses of the recombinant proteins with purification tags (i.e. N-and/or C-terminal tag with molecular masses of 3.8 and 1.5 kDa, respectively) were calculated as: 17.5 kDa for CHFI-12345, 19 kDa for CHFI-1245, 15.2 kDa for CHFI-1234, 18 kDa for CHFI-2345, 16 kDa for CHFI-234, and 15.5 kDa for CHFI-123. The theoretical molecular mass of wild type CHFI was calculated as 13.6 kDa. These values are consistent with the SDS-PAGE results (Fig. 2). The molecular mass discrepancy between control CHFI-ER from Z. mays seeds and recombinant CHFI (CHFI-12345) is caused by the remaining N-terminal purification tag. The approximate yields of the individual mutants ranged from 2.5 to 75 mg of purified protein/liter of cell culture. Recombinant EGFP was also expressed for further use as a negative control in chromogenic assays to exclude possible unspecific chromogenic substrate cleavage by traces of proteases from E. coli.
Previous attempts to obtain recombinant CHFI resulted in the preparation of nonfunctional protein because of inclusion body formation during expression in E. coli (49,50). Because CHFI has five disulfide bonds (29), correct folding of the molecule requires an oxidative environment (51). However, the E. coli cytoplasm is slightly reducing (52). Ordinary E. coli strains, such as BL21 DE3 (49,50), that were previously used for CHFI expression do not provide optimal conditions for disulfide bond formation because of the functions of the antioxidant glutathione and thioredoxin systems (52). Based on these observations, the E. coli Rosetta-Gami 2 DE3 strain with mutations in the glutathione and thioredoxin systems (53) was chosen as a host for the production of CHFI and its mutants. Thus, the observed soluble expression of CHFI is consistent with the data obtained in proof of method studies (54).
Inhibitory Activity Assay-To investigate inhibitory activity against FXIIa, FXIa, and trypsin, the purified mutant pro-FIGURE 1. CHFI and its mutants. A, schematic diagrams of CHFI mutants. Cysteine residues and terminal amino acids are marked with numbers below each scheme. Numbered square brackets indicate disulfide bridges, triangles indicate the reactive site (i.e. the scissile bond between Arg 34 and Leu 35 ), crosses indicate Cys to Asp substitutions, and the ellipse indicates the unpaired cysteine C86. B, schematic diagrams of CHFI mutant conformations. Visualizations were performed using Discovery Studio software (Accelrys, Inc., San Diego, CA). ␣-Helices are indicated in red, and disulfide bridges are indicated as yellow balls and sticks. Cysteines and Cys to Asp mutations are labeled, and the inhibitory loop is located at the top. C, amino acid sequence alignment of CHFI mutants obtained using ClustalW software (38). The cysteine positions are indicated with blue boxes, and the positions of Cys to Asp mutations are presented in red. Purification tags and amino acids from the expression vector are marked with gray shading. The numbers above the sequence indicate the first and last amino acids from CHFI that are present in the mutants and the Cys residues, according to 1BEA in the PDB. The last number in each sequence line corresponds to the total length of the recombinant protein.
FXIIa-The function of recombinant CHFI is indistinguishable from that of wild-type CHFI (CHFI-ER) from corn seeds. This finding suggests that the remaining purification tags do not influence the inhibitory activity of CHFI against FXIIa. Shortened mutants with the truncation of only one disulfide bond at the N (i.e. CHFI-2345) or C terminus (i.e. CHFI-1234) also inhibit FXIIa with the same activity as wild-type CHFI. The truncation of one disulfide bond from both the N terminus and the C terminus (i.e. CHFI-234) slightly decreases the efficiency of the inhibitor (K i ϭ 3.2 Ϯ 0.4 nM versus K i ϭ 1.0 Ϯ 0.1 nM for wild type; p value Ͻ 0.05) (Fig. 4A). The truncation of the fourth and fifth disulfide bonds (i.e. CHFI-123) dramatically reduces the ability of the mutant to inhibit FXIIa (K i ϭ 116 Ϯ 16 nM) compared with the wild-type protein (p value Ͻ 0.001). A fullsize CHFI mutant lacking the third disulfide bond also exhibits reduced inhibitory activity against FXIIa (K i ϭ 50 Ϯ 8 nM, p value Ͻ 0.05) (Fig. 4B). Inhibition of FXIIa by the synthetic peptide CHFI-2, which represents the protease-binding loop, was not observed (Fig. 3B and Table 2) for peptide concentrations up to 1 mM (data not shown). Thus, the mutants retain wild-type inhibitory activity until the third and fourth disulfide bonds are deleted from the molecule, with a drastic decrease observed for CHFI-123.
Trypsin-Because no FXIIa inhibition by CHFI-2 was observed, this peptide and the other mutants were assayed against trypsin (Fig. 3, D and E, and Table 2). The observed inhibitory activity of the purified mutant proteins is similar with respect to the level of magnitude and varies nonsignificantly (Figs. 3, D and E, and 4C) from 1 Ϯ 0.6 nM for CHFI-1234 to 2.9 Ϯ 0.5 nM for CHFI-2345, as long as disulfide bridges 2, 3, and 4 are present. CHFI-1245 exhibited a dramatically increased K i (250 Ϯ 70 nM), as did the synthetic peptide CHFI-2 (12 Ϯ 2 M). No inhibitory activity was observed for CHFI-123 at concentrations up to 75 M (data not shown). Because CHFI-2 retains partial inhibitory activity, the isolated proteasebinding loop might be in the active conformation.
FXIa-To determine whether the inhibitors with decreased activity exhibited weak interactions with the cognate enzyme, CHFI-2 and CHFI-1245, which serves as a control for the active loop conformation, were assayed against FXIa. CHFI-2 retains partial inhibitory activity against FXIa, with K i ϭ 94 Ϯ 11 M, whereas wild-type CHFI inhibited FXIa with K i ϭ 5.4 Ϯ 0.2 M. Unexpectedly, CHFI1245 lost the ability to inhibit FXIa (Fig. 3C and Table 3).
Modeling of the Interactions between the Isolated Proteasebinding Loop CHFI-2 and Cognate Enzymes-To analyze the possible causes of CHFI-2 inactivity against FXIIa, protein-protein docking was performed. Because the x-ray structure of the complete FXIIa molecule is not currently available, the FXIIa structure was modeled prior to the docking procedure.
Molecular Dynamics-To investigate whether the mutations cause conformational changes in the isolated protease-binding loop (CHFI-2) and CHFI-1245 (i.e. CHFI-12345 with a C29D mutation), MD simulations were performed. A 50-ns MD simulation was followed by cooling of the system to obtain conformations of all three inhibitor variants stable in solution. These conformations might differ from the conformation of CHFI-12345 in the crystalline phase that was provided by the x-ray data. The conformation of the Arg 34 side chain, which is a residue located at the scissile bond, is of particular interest.
During the MD simulation of CHFI-12345, the C29D mutant CHFI-1245, and the synthetic cyclic peptide CHFI-2, no dramatic conformational changes were observed. After cooling the CHFI-12345 system, the conformation of the Arg 34 side chain, which is one of the amino acids from the scissile bond Arg 34 -Leu 35 , differed from the initial x-ray conformation. In CHFI-1245 and CHFI-2, Arg 34 is similarly bent down to the backbone. However, in CHFI-1245 (i.e. the C29D mutant), Arg 34 forms a stable salt bridge with mutated Asp 29 (Fig. 5A), causing a change in the shape of the loop carrying Arg 34 (Fig. 5B). In CHFI-2, the angle between two ␣-helixes is different from that observed in the initial x-ray structure (Fig. 5C). Because of this conformational change, the structure of the cyclic peptide is more rigid; as a result, Arg 34 and Asp 29 are unable to form a salt bridge because they are separated by a distance of 7 Å, which is sufficient to dampen their interaction. These differences might alter the interactions of the CHFI variants with the active sites of their cognate enzymes. Because of the observed difference in the conformation of the Arg 34 side chain between the x-ray structure and the structure obtained after MD cooling, two forms of each inhibitor were used for protein-protein docking: the x-ray structures with minimization of modified parts and the stabilized geometries after 50-ns MD at 298 K and slow cooling of the system to 4 K. A detailed account of the molecular modeling results is presented in the supplemental materials.
Protein-Protein Docking-Homology modeling of the FXIIa structure provided a model with a serine-histidine-aspartic acid triad located in a cleft on the protein surface in a suitable position for performing catalytic substrate cleavage (Fig. 5D). Docking of the tripeptide analog of the FXIIa-specific substrate S2302 provided a position in the model active site that was suitable for peptide bond cleavage (see SI). Thus, this FXIIa  model was used for protein-protein docking without further improvements. A detailed description of the protein-protein docking results for the three inhibitor variants and their cognate enzymes is provided in the SI.
In most cases, more favorable scoring energies of proteininhibitor complexes were obtained for conformations of CHFI and its analogs that were provided by comparing the MD simulations to the x-ray structure and its modifications (i.e. the C29D mutant and the truncated cyclic fragment) (supplemental Table S2). This finding indicates that conformations obtained after cooling the system in the MD simulation (i.e. stable in solution) for the inhibitor variants are more suitable for inhibition of the target enzymes than the crystalline phase conformation provided by the x-ray data (see "Molecular Dynamics" in the supplemental materials for details). The protein-protein docking results for these MD structures are consistent with the experimental observations of inhibitory activity described above. All three CHFI variants exhibit good binding with the trypsin active site in a position favorable for cleavage of the bond between Arg 34 and Leu 35 , but the binding of the cyclic peptide is weaker than that of the full-length inhibitors. This finding is consistent with the data obtained from trypsin inhibition assay (Table 2 and Fig. 3, D and E). The docked position of CHFI-12345 with FXIa places Arg 34 in the active site, but the position is relatively far from the catalytic serine. In addition, the scoring energy of the binding is rather poor compared with the energies obtained for the complexes with trypsin. Taken together, these data could explain the experimentally measured 3-fold increase in inhibition constants. In the case of CHFI-1245, the observed position is rather far from the active site and is stabilized by several salt bridges (Fig. 5E). The existence of such an alternative binding site could explain the poor experimentally observed inhibitory activity of CHFI-1245 against FXIa (Table 3 and Fig. 3C). The interaction of the cyclic synthetic peptide CHFI-2 with FXIa is similar to the binding of CHFI-2 to trypsin. The complexes between the FXIIa model and CHFI-12345 and its C29D mutant were similar to those of trypsin. However, in the case of CHFI-2, the most favorable position is located far from the active site and generated primarily by hydrophobic contacts (Fig. 5F).

DISCUSSION
Here we report that interactions outside the protease-binding loop of CHFI and the active site of FXIIa are likely to contribute to the strength and specificity of binding. This finding suggests that CHFI is likely the first canonical inhibitor whose protease-binding loop is not sufficient for cognate enzyme inhibition. According to our results, the synthetic cyclic peptide CHFI-2 lacks inhibitory activity against FXIIa. Moreover, FXIIa-CHFI-2 docking results suggest that the peptide is likely to bind far from the active site of FXIIa. This finding is not expected for a canonical inhibitor because an isolated proteasebinding loop is reported to be an independent structural element (20,21) and to determine interactions with the cognate enzyme. For instance, a cyclic peptide resembling the loop from SFTI-1 retains the inhibitory activity and structural features of the full-size inhibitor and has a K i (19 nM) ϳ40 times higher than that of the parental protein (0.5 nM) (20).
CHFI-2 (i.e. the loop) is rich in proline (Figs. 1C and 6) and should therefore have a stable conformation. We initially     CHFI-2 inhibitory activity was assayed for proteases that are specifically inhibited by CHFI (i.e. trypsin and FXIIa) and for FXIa, for which the inhibition is much weaker. MD simulation and proteinprotein docking were also performed to predict possible interactions between the peptide and its cognate enzymes.
The MD results suggest that the spatial structure of the cyclic peptide is quite rigid, and no dramatic conformational changes compared with the loop in the wild-type inhibitor are proposed (see Section 1.3, "CHFI-2: The Cyclic Synthetic Peptide," in the supplemental materials for details). Only slight alterations of the angle between two ␣-helixes in CHFI-2 were observed compared with the initial x-ray structure of the loop (Fig. 5C). These results do not support the initial assumption that the isolated loop would exhibit extra flexibility.
Our experimental results from the inhibition assay for the CHFI-2 peptide are consistent with previous findings (20,21). These results suggest that the isolated protease-binding loop retains partial inhibitory activity and acts as an independent structural element. CHFI-2 retains inhibitory activity against FXIa, but the K i for this interaction (94 Ϯ 11 M) is ϳ20 times higher than that of wild-type CHFI (5.4 Ϯ 0.2 M). CHFI-2 also retains its inhibitory activity against trypsin; however, the difference in K i between CHFI-2 and wild-type CHFI is greater (12 Ϯ 2 M versus 1.3 Ϯ 0.2 nM, respectively). Taken together with the MD results, these data suggest that the isolated protease-binding loop is likely to be in an active conformation because it can inhibit trypsin and FXIa. The docking results support these experimental data, because the scissile bond of the peptide (i.e. Arg 34 -Leu 35 ) is located in the active site of trypsin and FXIa.
Surprisingly, no inhibition was observed when CHFI-2 was assayed against FXIIa (Fig. 3B). Because the peptide is supposed to be in the active conformation, this lack of inhibition could be explained by an incorrect position of the scissile bond relative to the FXIIa catalytic site. We hypothesize that the functional position of the CHFI protease-binding loop in the cave of the FXIIa catalytic site is primarily determined by interactions between nonloop regions of CHFI and FXIIa. This hypothesis is supported by the CHFI-2-FXIIa docking results. The most favorable calculated position of the peptide is outside of the predicted active site on the FXIIa model structure (Fig. 5F). However, the disproportional loss of CHFI-2 inhibitory activity against FXIIa compared with trypsin and FXIa could also be partially caused by the elimination of the third disulfide bond and the other interactions between the loop and nonloop regions of CHFI.
To better understand these possible reasons for the lack of activity of the CHFI-2 peptide against FXIIa, the role of CHFI protease-binding loop stabilization by the third disulfide bond was studied in experiments with CHFI-1245, which is a full-size CHFI mutant with the third disulfide bridge eliminated. This mutant was also used as a control for the C29D replacement in the CHFI-2 peptide. CHFI-1245 exhibits reduced inhibitory activity against FXIIa and trypsin (K i ϭ 50 Ϯ 8 and 250 Ϯ 20 nM, respectively) compared with wild-type CHFI (K i ϭ 1.0 Ϯ 0.1 and 1.3 Ϯ 0.2 nM, respectively). Unexpectedly, CHFI-1245 lost the ability to inhibit FXIa (K i ϭ 490 Ϯ 110 M). This finding could be explained by the CHFI-1245-FXIa docking results, because an alternative CHFI-1245 binding site was present on the FXIa surface outside the active site (Fig. 5E). Nevertheless, the residual inhibitory activity of CHFI-1245 against FXIIa and trypsin suggests that although the third disulfide bridge contributes to the interaction of the CHFI loop with the FXIIa active site, the C29D substitution is not likely to be the only cause of the lack of activity of the CHFI-2 peptide against FXIIa.
Our further studies of the contact region in the interaction between CHFI and FXIIa suggest that either a region between CHFI residues Gly 93 and Thr 108 interacts with FXIIa or the fourth disulfide bond is essential for proper folding of CHFI. The elimination of the first and the fifth disulfide bonds does not affect the folding of the contact region of the truncated protein with FXIIa. Considering the conformation-based concept of the structure-activity investigation (55) and the observation (49) that a mutant that lacks 11 N-terminal amino acids and has a Cys 55 substitution is indistinguishable from full-size recombinant CHFI, we investigated the inhibitory activity of four truncated CHFI mutants in comparison with the wild-type and full-size recombinant CHFI (CHFI-12345). The shortened mutants CHFI-1234 (K i ϭ 1.0 Ϯ 0.3 nM for FXIIa) and CHFI-2345 (K i ϭ 1.0 Ϯ 0.2 nM) retain the ability to inhibit FXIIa to the same extent as wild-type CHFI (K i ϭ 1.0 Ϯ 0.1 nM); this finding is consistent with previously reported results (49). CHFI-234 is a combination of the truncations from CHFI-1234 and CHFI-2345. This mutant has a slightly increased K i against FXIIa (3.2 Ϯ 0.4 nM), which is similar to of CHFI-1234, CHFI-12345, and CHFI-2345 at a 0.01 level of significance (p value ϭ 0.0013, which is Ͻ0.01; Figs. 3A and 4A). Similar results were obtained in the trypsin inhibition assay (Table 2 and Figs. 3, D and E, and 4C). Together, these results suggest that the 11 N-terminal and 24 C-terminal amino acids of CHFI and the first and the fifth disulfide bonds are not involved in interactions with FXIIa and trypsin. Therefore, CHFI-234 is a minimally active CHFI fragment that is necessary for FXIIa inhibition, and CHFI-2 is an independent structural element that retains its inhibitory activity against trypsin and FXIa.
Interactions outside the FXIIa active site and the CHFI protease-binding loop contribute to the strength and the specificity of binding, because the isolated binding loop is unable to inhibit FXIIa but partially retains inhibitory activity against other proteases, such as trypsin and FXIa. These findings suggest that the mechanism of FXIIa inhibition by CHFI could deviate from the standard mechanism (13), with nonloop regions involved in the process in an unknown manner. In contrast, our results indicate that CHFI behaves according to the standard mechanism of inhibition, in which the proteasebinding loop plays a crucial role, during the inhibition of trypsin and FXIa.