Crystal structures of the FXIa catalytic domain in complex with ecotin mutants reveal substrate-like interactions.

Thrombosis can lead to life-threatening conditions such as acute myocardial infarction, pulmonary embolism, and stroke. Although commonly used anti-coagulant drugs, such as low molecular weight heparin and warfarin, are effective, they carry a significant risk of inducing severe bleeding complications, and there is a need for safer drugs. Activated Factor XI (FXIa) is a key enzyme in the amplification phase of the coagulation cascade. Anti-human FXI antibody significantly reduces thrombus growth in a baboon thrombosis model without bleeding problems (Gruber, A., and Hanson, S. R. (2003) Blood 102, 953-955). Therefore, FXIa is a potential target for anti-thrombosis therapy. To determine the structure of FXIa, we derived a recombinant catalytic domain of FXI, consisting of residues 370-607 (rhFXI370-607). Here we report the first crystal structure of rhFXI370-607 in complex with a substitution mutant of ecotin, a panserine protease protein inhibitor secreted by Escherichia coli, to 2.2 A resolution. The presence of ecotin not only assisted in the crystallization of the enzyme but also revealed unique structural features in the active site of FXIa. Subsequently, the sequence from P5 to P2' in ecotin was mutated to the FXIa substrate sequence, and the structures of the rhFXI370-607-ecotin mutant complexes were determined. These structures provide us with an understanding of substrate binding interactions of FXIa, the structural information essential for the structure-based design of FXIa-selective inhibitors.

Thrombosis can be divided into an initiation phase and an amplification phase (1). The initiation phase is triggered by vascular injury, which leads to exposure of tissue factor. Tissue factor and activated Factor VII (FVIIa) 1 together activate Factor X to FXa, which in turn activates prothrombin to thrombin. This process is normally rapidly shut down by tissue factor pathway inhibitor and anti-thrombin III. Trace amounts of thrombin initiate the amplification phase of thrombosis and can convert platelet-bound FXI to FXIa, which, in turn, promotes Factor IX (FIX) activation on the platelet surface, resulting in local, explosive generation of thrombin and the formation of hemostatic thrombi at sites of vascular injury (see Ref. 2 for review). Thus, FXI plays an essential role as an amplifier in the blood coagulation process. Although FXI can be converted to FXIa by thrombin (3) or FXIa itself (4,5) in vitro, activation by thrombin predominates over other activators such as activated Factor XII (FXIIa) or FXIa on the activated platelet surface (6). In addition, individuals with FXI deficiency have abnormal hemostasis, whereas deficiencies in FXII, prekallikrein, and high molecular weight kininogen are not associated with hemostatic abnormalities (7,8). Therefore, thrombin, rather than FXIIa, is the likely activator of FXI under normal physiological conditions.
Malfunctions of the blood coagulation cascade can lead to life-threatening conditions such as acute myocardial infarction, pulmonary embolism, and stroke resulting from arterial or venous thrombosis. Commonly used anti-coagulant drugs include unfractionated heparins, low molecular weight heparins, and warfarin. Although these drugs are efficacious, they have serious limitations. They target multiple coagulation factors, carry a significant risk of inducing severe bleeding complications, and have narrow therapeutic windows for safe administration, requiring routine monitoring. To decrease bleeding liabilities, new anti-coagulants have been designed to target specific steps in the coagulation pathway, such as thrombin and FXa; however, their safety profile has yet to be established in clinical trials (9). Because FXIa is a key enzyme involved in the amplification phase of the pathway, modulating FXIa activity should not affect the initiation phase but should reduce thrombus formation without compromising hemostasis and thus have few or no bleeding liabilities. Supporting evidence comes from the report that an anti-human FXI antibody did not prevent thrombus initiation, but significantly reduced intraluminal thrombus growth in a baboon thrombosis model (10). The anti-FXI antibody prolonged the partial thromboplastin time (a measure of the coagulation function involving FXIa) but did not affect prothrombin time (a measure of the coagulation function involving FVIIa and tissue factor). Most significantly, the anti-FXI antibody did not prolong bleeding time. In addition, individuals with FXI deficiency do not show severe bleeding problems (8). Furthermore, FXIa has been implicated in maintaining clot integrity (1,11). Thus, FXIa appears to be an attractive target for developing novel anti-coagulant agents that might be advantageous over existing inhibitors for prophylaxis and treatment of thrombotic disorders.
FXI is a 160-kDa protein composed of two identical 80-kDa subunits that are linked by a disulfide bond and is activated by cleavage after residue 369 of each of the two subunits, producing a protein that contains two heavy and two light chains. Each heavy chain contains four tandem repeat sequences, designated "apple" domains, containing binding sites for platelets (12), heparin (13), and many proteins such as thrombin (14), high molecular weight kininogen (15), factor XIIa (16), and glycoprotein Ib␣ (17). Each light chain, consisting of residues 370 -607, contains a catalytic domain. The catalytic domain of FXIa is a trypsin-like serine protease, highly homologous to the catalytic domains of other factors in the blood coagulation cascade, such as FIXa, FXa, FVIIa, and thrombin. To develop FXIa-selective inhibitors using structure-based ligand design, it is essential to obtain the three-dimensional structure of FXIa and to identify unique features of the FXIa catalytic domain.
To obtain the first structure of the catalytic domain of FXIa (rhFXI 370 -607 ), we formed a complex between rhFXI 370 -607 and ecotin. Ecotin is a 142-amino acid pan-serine protease inhibitor found in the periplasm of Escherichia coli (18). The protein exists as a dimer and inhibits a broad range of serine proteases. The ecotin dimer interacts with two serine protease molecules to form a tetrameric complex. Several crystal structures of serine proteases in complex with ecotin have been reported (19 -23). The unique arrangement of the interactions in the serine protease-ecotin tetramer and the adaptability of the ecotin dimer suggested that ecotin could assist in the crystallization of many serine proteases, including rhFXI 370 -607 . Here we report three rhFXI 370 -607 -ecotin structures. The first is the rhFXI 370 -607 -ecotinM84R structure, in which the P 1 residue of ecotin has been mutated from methionine to arginine. The second is rhFXI 370 -607 -ecotinD, in which the ecotin residues P 5 through P 2Ј have been replaced with residues from the FXIa substrate FIX ( 80 NDFTRVV 86 ). The last is rhFXI 370 -607 -ecotinP, in which ecotin residues P 4 to P 2Ј were mutated to the substrate sequence of FXIa ( 81 DFTRVV 86 ), which helped to explain structural differences between rhFXI 370 -607 -ecotinM84R and rhFXI 370 -607 -ecotinD. All together, these structures not only reveal three-dimensional structural information for FXIa, but also demonstrate the possible interaction of the enzyme with its substrate.

EXPERIMENTAL PROCEDURES
Preparation of Recombinant FXI Catalytic Domain-The recombinant FXI catalytic domain, amino acid residues 370 -607, was expressed as a secreted protein in the methanolotropic yeast Pichia pastoris. A detailed report on cloning and expression will be presented elsewhere. 2 The enzyme was initially separated from conditioned medium using ammonium sulfate precipitation (50 -80% saturation). The ammonium sulfate precipitate was dialyzed against two changes (100 volumes each) of 20 mM Tris-HCl, pH 7.4. The dialyzed enzyme was then purified by a combination of immobilized Zn ϩ2 -chelate chromatography (Zn-IMAC, Chelating Sepharose Fast Flow, Amersham Biosciences) and cation exchange chromatography (SP-Sepharose Fast Flow, Amersham Bioscience). To remove high mannose-type oligosaccharide side chains, the enzyme was treated with endoglycosidase H (New England Biolabs) at a concentration of 500 units/mg rhFXI 370 -607 for 90 min prior to the ion exchange step. The purity of the protein was estimated by SDS-PAGE analysis on 8 -16% polyacrylamide gels (Invitrogen) and by RP-HPLC analysis (data not shown). Protein concentrations were also estimated by RP-HPLC using a calibration curve constructed with commercially purchased, native FXIa (Hematologic Technologies, Inc., Essex Junction, VT). The protein was reduced by treatment with 10 mM dithiothreitol to allow separation of the heavy and light chains during chromatography, and only the peak derived from the light chain was used for calibration purposes. Enzymatic activity was measured using a peptide substrate pyroGlu-Pro-Arg-7methylamidocoumarin. This is a modification of the standard peptide substrate S2366 in which the chromogenic moiety p-nitroaniline was replaced with the fluorogenic reporter, 7-methylcoumarin.
Preparation of Ecotin Mutants-Expression plasmids of wild-type ecotin and the ecotinM84R ( 80 PVSTRMA 86 , native sequence 80 PVST-MMA 86 ) were the kind gift of Dr. Robert J. Fletterick and have been described before (24). The ecotin substitutions containing the FIX cleavage site, ecotinD ( 80 NDFTRVV 86 ) and ecotinP ( 80 PDFTRVV 86 ), were generated by site-directed mutagenesis. The ecotinD construct was generated in two steps. In the first step, ecotinM84R was mutated to ecotinNDF ( 80 NDFTRMA 86 ), and in the second step ecotinNDF was used as a template to generate ecotinD. EcotinP was generated using ecotinD as a template. The methods for the expression and purification of ecotin mutants were similar to those described previously (24). Minor modifications included transformation of the ecotin plasmids into the E. coli strain BL21 and further purification of the ecotin mutants using a POROS HQ (Applied Biosystems) column (4.6 ϫ 50 mm) mounted on a BioCAD 700E in buffer containing 5 mM MgCl 2 and 10 mM Tris-HCl, pH 7.8. Purified ecotin was collected in the unbound fraction.
Crystallization and Structure Determination of rhFXI 370 -607 -Ecotin Mutant Complexes-The rhFXI 370 -607 -ecotinM84R complex was crystallized at room temperature by the hanging drop vapor diffusion method, from 0.2 M NaCl, 20% (w/v) polyethylene glycol-1000, 0.1 M sodium potassium phosphate, pH 6.2. Diffraction data to 2.2 Å resolution were measured at beamline X12C of the National Synchrotron Light Source at the Brookhaven National Laboratory and processed by HKL2000 (25). The crystals belong to the space group P2 1 2 1 2 1 with unit cell parameters of a ϭ 44.6 Å, b ϭ 92.7 Å, and c ϭ 186.9 Å. There are two rhFXI 370 -607 and two ecotinM84R molecules forming a tetramer in the asymmetric unit.
The structure of rhFXI 370 -607 -ecotinM84R was determined by molecular replacement using the program Amore (26). The search model was created from the coordinates of the tetrameric rat granzyme B [N66Q]ecotin 81 IEPD 84 complex with Protein Data Bank (PDB) code 1FI8 (21). The conserved core structure of serine proteases was identified first by structural alignment of several serine proteases. All side-chain atoms beyond C␤, all surface loops in granzyme B, and all water molecules were deleted from the model. A clear solution was obtained using data of 20 -5 Å resolution with a correlation coefficient of 52.4 and an R factor of 48.9. This solution also persisted in calculations carried out using data in different resolution ranges. Structure refinement was carried out using CNX (Accelrys), and model building was performed in Quanta (Accelrys). Detailed data and refinement statistics are listed in Table I.
The rhFXI 370 -607 -ecotinD complex was crystallized at room temperature by the hanging drop vapor diffusion method from 0.1 M ammonium sulfate, 18% (w/v) polyethylene glycol-monomethyl ether 2000 and 0.1 M sodium cacodylate, pH 6.3. Diffraction data to 2.9 Å resolution were measured on a RAXIS IV ϩϩ with an RU-H3R generator and an X-stream 2000 system (Rigaku/MSC) and processed by HKL2000. The crystals belong to the space group P2 1 2 1 2 1 with unit cell parameters of a ϭ 45.3 Å, b ϭ 91.0 Å, and c ϭ 188.5 Å.
The rhFXI 370 -607 -ecotinP complex was crystallized at room temperature by the hanging drop vapor diffusion method from 0.1 M ammonium sulfate, 22% (w/v) polyethylene glycol-monomethyl ether 2000, 0.1 M sodium cacodylate, pH 6.2, and 3% (v/v) methanol. Diffraction data to 2.6 Å resolution were measured on a RAXIS IV ϩϩ with an RU-H3R generator and an X-stream 2000 system and processed by HKL2000. The crystals belong to the space group P2 1 2 1 2 1 with unit cell parameters of a ϭ 44.5 Å, b ϭ 90.1 Å, and c ϭ 189.1 Å.
Structures of both rhFXI 370 -607 -ecotinD and rhFXI 370 -607 -ecotinP were determined using the coordinates of rhFXI 370 -607 -ecotinM84R. Structure refinement was carried out using CNX, and model building was performed in Quanta. Detailed data and refinement statistics are listed in Table I.

RESULTS AND DISCUSSION
Ecotin Facilitates the Crystallization of rhFXI 370 -607 -Several crystal structures of serine protease-ecotin complexes, such as trypsin-ecotin, thrombin-ecotin, FXa-ecotin, collagenase-ecotin, and granzyme B-ecotin (19 -23), have been reported. There is a unique arrangement between the ecotin dimer and serine proteases that allows ecotin to facilitate the crystallization of the different serine proteases forming a complex with it. Because the catalytic domain of FXIa is a serine protease, we proposed that ecotin would be an inhibitor for FXIa and that ecotin would facilitate the crystallization of rhFXI 370 -607. Furthermore, ecotinM84R was designed to interact better with the aspartate residue that all trypsin-like serine proteases possess in the substrate binding site. The ability of each of the mutants to inhibit the peptidolytic activity of FXIa was assessed. The IC 50 values for ecotinM84R, ecotinP, and ecotinD were 3, Ͼ200, and 600 nM, respectively. However, subsequent results suggest that ecotinP and ecotinD were acting as competitive substrates, see below. The rhFXI 370 -607 protein did not produce diffraction-quality crystals with small molecule inhibitors such as benzamidine but yielded diffraction-quality crystals in complex with the wild-type and all three mutated ecotin proteins. Thus, the use of ecotin allowed us to obtain a structure of our target enzyme.
Tetrameric Complex of rhFXI 370 -607 -EcotinM84R-There is a tetrameric complex in the asymmetric unit of rhFXI 370 -607 -ecotinM84R crystal. The quaternary structure of the rhFXI 370 -607 -ecotinM84R tetramer is similar to other serine protease-ecotin structures in that it is composed of one ecotinM84R dimer (chain C and D in the coordinate file) and two rhFXI 370 -607 molecules (chain A and chain B in the coordinate file), which form three distinct types of binding interfaces: the dimerization interface between the two ecotinM84R molecules, a primary substrate-like interaction and a smaller secondary interaction between the corresponding ecotinM84R and rhFXI 370 -607 molecule (Fig. 1A). The ecotin monomers have their primary and secondary FXIa binding sites located at the two ends of the molecules. The ecotin dimer collectively provides binding surfaces for two FXIa molecules.
For each rhFXI 370 -607 molecule, there is a primary and a secondary ecotin binding site. The primary binding site involves ecotin residues 81-88 (the ecotin 80s loop) and 52-54 (the ecotin 50s loop) of one ecotin molecule, which extend into the active site of rhFXI 370 -607 . The secondary binding site is about 25 Å away from the primary site; it includes two surface loops of the other ecotin molecule, ecotin residues 63-70 (the ecotin 60s loop) and 108 -113 (the ecotin 110s loop), which bind to part of the C-terminal helix and residues 91-95 of rhFXI 370 -607 . We adopted the trypsin numbering system for rhFXI 370 -607 for easy comparison with other members in this serine protease family. The corresponding residue numbers for the FXI sequence and trypsin numbering system are listed in the Supplemental Material (Table S1).
The rhFXI 370 -607 Structure in the rhFXI 370 -607 -EcotinM84R Complex-The overall structure of rhFXI 370 -607 is similar to other serine proteases of the chymotrypsin fold (Fig. 1B). The globular structure is formed by two ␤ barrels with a few helical segments and many loops. The catalytic triad, composed of Ser 195 , His 57 , and Asp 102 is located at the junction between the two ␤-barrel domains. A unique feature of rhFXI 370 -607 revealed in the rhFXI 370 -607 -ecotinM84R structure is that the side chain of Glu 98 (Fig. 1B) points into the active site, where it does not appear to have a complementary polar environment. One of the oxygen atoms of the Glu 98 side chain forms a hydrogen bond with its main-chain nitrogen atom, and the other oxygen atom is within hydrogen bonding distance to the carbonyl oxygen of ecotin Leu 52 . Because Glu 98 interacts with ecotin Leu 52 , this orientation of Glu 98 might be ecotin-induced. Another unique feature is the orientation of Tyr 59A , located near the catalytic His 57 (Fig. 1B), which points away from the protein and is in van der Waals contact with the ecotin 50s loop. Although it was not clear from this structure whether the position of Tyr 59A in rhFXI 370 -607 is ecotin-induced, subsequent crystal structures have shown that this is a feature of rhFXI 370 -607 and is not induced by either ecotin or crystal packing. 3 However, the side-chain position of Glu 98 varies from structure to structure.
The EcotinM84R Structure-The modified jellyroll fold of the ecotin monomer is preserved in the rhFXI 370 -607 -ecotinM84R structure. The extended C-terminal regions of the two ecotins form anti-parallel ␤-sheets and wrap around each other at the  1A and 2). When the C␣ atoms of an ecotinM84R monomer (chain C, residues 9 -60, 71-75, and 105-134) are compared with other ecotin monomers derived from the structure of the ecotin dimer (PDB code: 1ECZ), the trypsin-ecotin complex (1EZS), the collagenase-ecotin complex (1AZZ), the granzyme B-ecotin complex (1FI8), the thrombin-ecotin complex (1ID5), and the FXa-ecotin complex (1P0S), the root mean square deviation is between 0.53 and 1.19 Å. The ecotin monomer in the rhFXI 370 -607 -ecotin M84R complex is most similar to that in the thrombin-ecotin structure and most divergent from the granzyme B-ecotin structure. As illustrated in Fig. 2, most of the differences are located in ecotin residues 80 -100 to accommodate the uniqueness of the active sites in different proteases. There is a slightly altered spatial orientation of the primary and secondary binding sites with respect to each other. When the C␣ atoms of one ecotin monomer in the different ecotin dimers structures described above were superimposed, the relative positions of the second monomer diverged considerably (Fig. 2). The difference between ecotin dimers arises from small local shifts among different ␤ sheets, accumulating to large distal positional changes. Similar observations were discussed in more detail by Perona et al. (20) for the collagenase-ecotin structure. This flexibility of the ecotin dimer is responsible for the ability of ecotin to inhibit a variety of serine proteases. The structural differences in the ecotin dimer are also reflected in the differences in the protease-ecotin tetrameric structures. In fact, molecular replacement searches did not produce a reasonable solution using either the tetrameric structures of trypsin-ecotin or collagenase-ecotin as search models against the diffraction data of rhFXI 370 -607 -ecotinM84R. The tetrameric structure of granzyme B-ecotin, however, yielded an unambiguous solution from the rotational and translational search of molecular replacement.
Interactions between EcotinM84R and rhFXI 370 -607 at the Primary Binding Site-The primary binding site of ecotinM84R interacts extensively with the active site of rhFXI 370 -607 in a substrate-like manner (Fig. 3). Eight residues of the ecotin 80s loop, residues 81-88, on both sides of the scissile bond, are in direct contact with rhFXI 370 -607 . The binding pockets of serine proteases are defined according to the residues of its substrate, thus the P 1 residue of the substrate (located N-terminal to the scissile bond) occupies the S 1 site of the protease, P 1Ј (located C-terminal to the scissile bond) re- sides in the S 1Ј site, and so on. Cys 50 of ecotin forms a disulfide bond with ecotin Cys 87 (P 3Ј ) to anchor the 80s loop, whereas many hydrogen bonds between the ecotin 50s and 80s loops provide additional support.
The S 1 binding pocket of rhFXI 370 -607 is the only deep pocket in rhFXI 370 -607 used for substrate binding (Fig. 3). Ecotin Arg 84 fits into the S 1 pocket of rhFXI 370 -607 . In one pair of the rhFXI 370 -607 -ecotinM84R complexes, the guanidinium group of ecotin Arg 84 (chain C) forms salt bridges with both oxygen atoms of Asp 189 of rhFXI 370 -607 (chain A). The conformation of ecotin Arg 84 and its interactions with rhFXI 370 -607 , involving two water molecules, are very similar to the corresponding residues in the structures of trypsin-ecotin and thrombin-ecotin complexes. In the other pair of the rhFXI 370 -607 -ecotinM84R complexes, the guanidinium group of ecotin Arg 84 (chain D) forms only a single salt bridge with one of the oxygen atoms of Asp 189 (chain B). Furthermore, there is only one water molecule mediating the interaction between ecotin Arg 84 and the S 1 pocket. In both pairs, the main chain oxygen of ecotin Arg 84 is stabilized by the oxyanion hole of rhFXI 370 -607 formed by the main-chain nitrogen atoms of Gly 193 and Ser 195 .
The S 2 , S 3 , and S 4 binding sites are less well defined (Fig. 3), although the S 2 loop of rhFXI 370 -607 (residues 96 -100, including Glu 98 ) is in a unique conformation compared with other serine proteases (see discussion below, Fig. 4). Ecotin Thr 83 (P 2 ) is in van der Waals contact with His 57 of the S 2 area but does not form any direct polar contacts with rhFXI 370 -607 . The  gen bond with the carbonyl oxygen of Leu 39 of rhFXI 370 -607 in chain B. The guanidinium groups of Arg 37D in both chains A and B of rhFXI 370 -607 pack against ecotin Pro 88 (P 4Ј ). FXIa is very selective for its substrate. More research is required to identify which site or sites in FXIa are the key selectivity elements for its substrate binding.
Structure Comparison of rhFXI 370 -607 with Other Serine Proteases-To identify features specific to FXIa, we compared the structure of rhFXI 370 -607 to human trypsin (1TRN) and other related proteases in the coagulation cascade: thrombin (1H8I), FVIIa (1QFK), FIXa (1RFN), and FXa (1HCG) (Fig. 4). All the structures were aligned with rhFXI 370 -607 by superimposing the C␣ atoms of the catalytic triads. The core structure is similar and superimposes well in all of these serine proteases; however, the lengths and conformations of surface loops are quite different. The following discussion is based on the orientation shown in Fig. 4.
The conformation of the S 1 pocket of FXIa resembles that of the other trypsin-like proteases. Asp 189 , at the base of the S 1 pocket, interacts with an arginine residue of the substrate. The residues forming the S 1 pocket are conserved except for three residues at 190, 192, and 213 (Table II). It is useful to distinguish three classes of trypsin-like serine proteases: those with alanine, serine, or threonine at position 190. The subtle difference between these residues changes the hydrogen-bonding network in the S 1 pocket and influences inhibitor binding and selectivity as determined for urokinase-type plasminogen activator, a trypsin-like serine protease that has Ser 190 (28 -30). FXIa has an alanine, which lacks the polar group of serine or threonine at that position, but provides a slightly larger S1 pocket and is preferred for hydrophobic interactions with ligands. The residue at 192 is involved in binding to ligands and protein inhibitors. The Gln or Glu at position 192 contributes to inhibitor binding in trypsin, FXa, and thrombin (31)(32)(33). FXI, as well as FVIIa, has a lysine at position 192. Lys 192 could be important for substrate specificity and as a potential selectivity element. Residue 213, in close proximity to residue 190, forms the back wall of the S 1 pocket. FXIa has a threonine at this position rather than the valine or isoleucine residue found in the other enzymes. However, the O␥ points away from the S 1 pocket, and the side chain adopts the same rotamer as valine in other enzymes. Therefore, Thr 213 is unlikely to be useful as a selectivity element for FXIa.
Directly facing the S 1 pocket in rhFXI 370 -607 and perpendicular to the catalytic triad, there is a shallow channel lined with hydrophobic residues that we have named the prime-side channel (Fig. 4). It is formed by residues 34 -39 (the 37s loop) and residues 59 -65 (the 60s loop). The lengths of the 37s and 60s loops vary among different serine proteases (Table III), and the amino acids in the loops are different. These variations contribute to the distinct sizes, shapes, and properties of the region. Although it is not clear what role the prime-side channel plays, it may be related to substrate recognition and protein-protein interactions for each protease. Of particular interest is the positioning of Tyr 59A of FXIa at the mouth of this channel forming an opened door allowing access to the catalytic triad; however, additional structural analysis is required to confirm that this is the natural position of this residue. At least for this complex, this residue helps to define the shape of S 2 in FXIa.
The prime-side channel of FXIa may provide selectivity elements for designing specific inhibitors.
The S 2 loop of FXIa folds toward the active site cleft from the bottom of the catalytic triad (Fig. 4). The numbers of residues forming this loop do not vary significantly (Table III), but the conformation of the loop differs among the serine proteases. The S 2 loop of FXIa is most similar to that of tryptase (34), with Ala 97 and Glu 98 /Gln 98 (rhFXI 370 -607 /tryptase). In both enzymes, Glu 98 /Gln 98 orients toward the center of the active site cleft. Although the position of the acidic head group of Glu 98 varies in different FXIa structures, the aliphatic portion of the Glu 98 side chain is always in van der Waals contact with Trp 215 and obstructs solvent access to its indole side chain. The unique position of Glu 98 reduces the size of both the S 2 and S 4 regions in FXIa and could be a key selectivity element for FXIa.
FXIa has a small S 4 pocket with charged and polar amino acids. Trp 215 is conserved in many trypsin-like serine proteases and forms the back wall of the S 4 subsite; however, the numbers and properties of the residues forming the front of the S 4 site (residues 170 -174, part of the S 4 loop consisting of residues 170 -176 in FXIa) differ in serine proteases. These differences, in turn, determine the different shapes and properties of the region. FXIa cleaves FIX at two locations that have either a lysine or an aspartate at P 4 . In the rhFXI 370 -607 -ecotinM84R structure, ecotin has a valine at P 4 . From this structure, it is not clear how the charged P 4 residue in the natural substrate interacts with the S 4 pocket.
There are several loops, most of which are remote from the active site, that are different in FXIa compared with other serine proteases. The 80s loop (residues 71-81 of FXIa) is the calcium-binding loop in trypsin, FVIIa and FIXa. FXIa does not bind calcium and does not have calcium-chelating residues; therefore, the 80s loop of FXIa is in a different conformation. Residues 146 and 147 of FXIa are located on the 146s loop (residues 144 -149) and protrude over the S 1 pocket, creating a small area that has been demonstrated to participate in inhibitor binding in urokinase (35), as well as in thrombin (36). This loop shows a high degree of flexibility in the FXIa structures reported in this report, as well as in FXIa structures in complex with small molecule ligands. 3 Similarly, this loop is disordered in some thrombin (36) and FXa structures (37). The different shape and properties of this region in serine proteases provide opportunities to design inhibitors specific for FXIa versus other serine proteases.
The Heparin Binding Site on rhFXI 370 -607 Is in a Large Positively Charged Area-Heparin is reported to facilitate the inhibition of rhFXI 370 -607 by the protease nexin II, a multifunctional protein containing a Bowman-Burke type protease inhibitor domain that is secreted by activated platelets and has been suggested to be a physiologically relevant inhibitor of FXIa (38). In the presence of heparin, the K i of protease nexin II for FXIa decreased from 436 Ϯ 62 to 88 Ϯ 10 pM (39). Through mutation and peptide inhibition studies, the heparinbinding site has been located on the catalytic domain of FXIa and involves residues Lys 170 , Arg 171 , and Arg 173 (39). In our FXIa structure, these residues are part of the S 4 loop on the back side of the S 4 pocket. It is likely that the binding of heparin changes the conformation of the S 4 loop and the active site, thus facilitating the binding of protease nexin II. It is interesting that these three positively charged residues in rhFXI 370 -607 are part of a large, positively charged area formed by residues Lys 170 , Arg 171 , Arg 173 , Arg 185 , Arg 222 , and Arg 224 . The large size of this positively charged area is unique in FXIa compared with other serine proteases. Negatively charged glycosaminoglycans have been shown to affect the activation of FXI and inactivation of FXIa by serine protease inhibitors. FXI can be activated on immobilized dextran sulfate and heparin sulfate (4,40). In addition, glycosaminoglycans have been shown to enhance the inhibition of FXIa by C1 inhibitor, anti-thrombin III, and protease nexin II (39,41,42). Polyanions can inhibit the hydrolysis reaction of the peptide substrate S2366 (pyroGlu-Pro-Arg-p-nitroanilide) and FIX activation by FXIa in a concentration-dependent manner (43). Based on the fluorescence characteristics of 5-(dimethylamino)-1-(naphthalenesulfonyl)-glutamylglycylarginyl-FXIa in the presence or the absence of a negatively charged surface, it has been suggested that binding of FXIa to the polyanions heparin and dextran sulfate results in an allosteric modification of its functional activity (43). It is very likely that the function of this large positively charged patch on FXI is to interact with negatively charged surfaces and, in turn, to reg-ulate FXIa activity. Another patch of positive residues (Arg 144 , Lys 145 , Arg 147 , and Lys 149 ) has been shown not to be involved in heparin binding (39).
rhFXI 370 -607 -EcotinD Structure-Most of the 18 known families of macromolecular serine protease inhibitors have a cysteine involved in a disulfide linkage located somewhere between P 5 and P 2Ј (44). Ecotin, along with the Kunitz soybean trypsin inhibitor and potato-I inhibitor, does not have this disulfide. These classes of inhibitors provide the opportunity to introduce the sequence of a natural substrate into a substratelike macromolecular inhibitor. In a previous study, the natural P 4 to P 1 substrate sequence for the serine protease granzyme B was engineered into ecotin (21). We designed EcotinD, with its P 5 to P 2Ј residues mutated to match the corresponding sequence of the FXIa substrate FIX, P5 80 NDFTRVV 86 P2Ј . This is the most extended natural sequence introduced into a proteinbased inhibitor to date. Since protease enzymes such as FXIa are highly selective for the sequences they cleave, EcotinD provides us with a model system to examine the details of the interactions between P 5 and P 2Ј of the natural substrate with the S 5 to S 2Ј subsites of FXIa. When reducing SDS-PAGE was performed on fractions from the gel filtration column used in  purifying the rhFXI 370 -607 -ecotinD complex, it was noted that the protein band for the full-length ecotinD was replaced by two lower molecular weight bands corresponding to cleavage fragments of ecotin (data not shown). The cleavage of an ecotin mutant by a serine protease was also reported for granzyme B-ecotin ( 81 IEPD 84 ) complex (21). In the crystal structure of granzyme B-ecotin ( 81 IEPD 84 ), the prime-side residues 85-96 of ecotin were not visible in the electron density maps. To our surprise, the rhFXI 370 -607 -ecotinD structure showed an intact peptide bond between residues 84 and 85 of ecotin. Ecotin and various ecotin mutants were reported to be cleaved by rat and/or bovine trypsin (45)(46)(47)(48). To reduce the proteolytic activity of trypsin, trypsin D102N was used to form complexes with ecotin mutants for crystallization studies (47). In our case, an intact ecotin mutant was present in the crystal structure, but showed evidence of being cleaved based on SDS-PAGE analysis. While this ecotin mutant is cleaved by rhFXI 370 -607 , the prime side of the ecotin molecule is still bound to rhFXI 370 -607 by secondary binding sites under non-denaturing conditions. We speculate that the leaving group (prime side residues) of ecotin is held in place by rhFXI 370 -607 , and thus hydrolysis does not take place or acyl-enzyme formation can be reversed. This is consistent with the clogged-gutter mechanism proposed for protease inhibitors (49) in which an acyl-enzyme intermediate of subtilisin and chymotrypsin inhibitor 2 was detected.
Protein-based inhibitors of serine proteases were proposed to bind more tightly than substrate, yet hydrolyze much more slowly due to the tight and oriented binding of the cleaved peptide, preventing acyl-enzyme hydrolysis and favoring the reverse, i.e. peptide bond forming, reaction. The overall structure of rhFXI 370 -607 -ecotinD is very similar to rhFXI 370 -607 -ecotinM84R. The peptide bonds between residues 84 and 85 in both copies of ecotinD are intact. In one ecotinD molecule (chain D), the prime side residues 88 -91 were too flexible to be seen, but the same residues were visible in the other ecotinD molecule (chain C) due to a symmetry related molecule of rhFXI 370 -607 packing against the area and stabilizing the loop. When the catalytic triads of rhFXI 370 -607 in rhFXI 370 -607 -ecotinM84R and rhFXI 370 -607 -ecotinD were superimposed (Fig. 5A), the main-chain conformations of residues 79 and 80 in ecotinD were different from those in ecotinM84R. There is a corresponding main-chain flip between Leu 52 and His 53 in the 50s loop of ecotinD. We suspected these changes might be caused by residue 80, Asn in ecotinD, and Pro in ecotinM84R. This was confirmed in the rhFXI 370 -607 -ecotinP structure (see below). P 1 and P 2 of ecotinD are the same as those of ecotinM84R and orient similarly. P 3 of ecotinD (Phe 82 ) is packed against the side of P 5 (Asn 80 ). P 4 (Asp 81 ) is a negatively charged residue, and its O␦ forms a hydrogen bond with N⑀ of rhFXI 370 -607 His 174 in the S 4 site.
rhFXI 370 -607 -EcotinP Structure-To confirm that the structural changes in the rhFXI 370 -607 -ecotinD structure were due to Asn 80 of ecotinD, we obtained the crystal structure of rhFXI 370 -607 -ecotinP with residue 80 of ecotinD changed back to the native Pro of ecotin. EcotinP was also cleaved by rhFXI 370 -607 as shown by SDS-PAGE analysis (data not shown). This suggested that P 5 is not important in the cleavage of these ecotin mutants by rhFXI 370 -607 . The backbone conformations of ecotinP residues 79 and 80, as well as residues 52 and 53, are the same as ecotinM84R (Fig. 5B). This confirmed that the two backbone changes in the rhFXI 370 -607 -ecotinD structure were indeed induced by the Pro-to-Asn mutation at residue 80. When the catalytic triads of rhFXI 370 -607 were superimposed, rhFXI 370 -607 -ecotinD showed more deviation from rhFXI 370 -607 -ecotinM84R than did rhFXI 370 -607 -ecotinP. This result also demonstrated the adaptability of the ecotin molecule. Although the side-chain orientation of Asp 81 (P 4 ) in ecotinP is different from that in ecotinD, Asp 81 still interacts with His 174 in S 4 .
All three structures of rhFXI 370 -607 with ecotin mutants provided detailed information on rhFXI 370 -607 and its substrate-binding interactions. This structural information is very useful for the development FXIa-selective inhibitors.