A Binding Site for Heparin in the Apple 3 Domain of Factor XI*

Since heparin potentiates activated factor XI (FXIa) inhibition by protease nexin-2 by providing a template to which both proteins bind (Zhang, Y., Scandura, J. M., Van Nostrand, W. E., and Walsh, P. N. (1997) J. Biol. Chem.272, 26139–26144), we examined binding of factor XI (FXI) and FXIa to heparin. FXIa binds to heparin (K d ∼0.7 × 10−9 m) >150-fold more tightly than FXI (K d ∼1.1 × 10−7 m). To localize the heparin-binding site on FXI, rationally designed conformationally constrained synthetic peptides were used to compete with 125I-FXI binding to heparin. A peptide derived from the Apple 3 (A3) domain of FXI (Asn235–Arg266) inhibited FXI binding to heparin (K d ∼3.4 × 10−6 m), whereas peptides from the A1 domain (Phe56–Ser86), A2 domain (Ala134–Ala176), and A4 domain (Ala317–Gly350) had no such effect. The recombinant A3 domain (rA3, Ala181–Val271) inhibited FXI binding to heparin (K i ∼1.4 × 10−7 m) indicating that all the information necessary for FXI binding to heparin is contained entirely within the A3 domain. The A3 domain also contains a platelet-binding site (Asn235–Arg266), consisting of three surface-exposed loop structures, Pro229–Gln233, Thr741–Leu246, and Thr249–Phe260 (Baglia, F. A., Jameson, B. A., and Walsh, P. N. (1995) J. Biol. Chem.270, 6734–6740). Only peptide Thr249–Phe260 (which contains a heparin binding consensus sequence, RIKKSKA) inhibits FXI binding to heparin (K i = 2.1 × 10−7 m), whereas peptides Pro229–Gln233 and Thr241– Leu246 had no effect. Fine mapping of the heparin-binding site using prekallikrein analogue amino acid substitutions of the synthetic peptide Thr249–Phe260 and alanine scanning of the recombinant A3 indicated that the amino acids Lys252 and Lys253 are important for heparin binding. Thus, the sequence Thr249–Phe260 which contains most of the binding energy for FXI interaction with platelets also mediates the binding of FXI to heparin.

Factor XI (FXI) 1 is a zymogen that circulates in plasma in a non-covalent complex with high molecular weight kininogen (HMWK) (1,2) and participates in the contact phase of blood coagulation (3). The active enzyme, activated factor XI (FXIa), is a trypsin-like serine protease that is generated when FXI is cleaved by FXIIa (3), thrombin, or FXIa (4,5) at an internal Arg 369 -Ile 370 bond (2). Upon activation, FXIa is capable of converting FIX into its active form, FIXa (2,6). FIXa, in the presence of FVIII and platelets, can activate FX (7,8), resulting in the generation of thrombin and, subsequently, a fibrin clot.
The structure of FXI is unique among the plasma coagulation enzymes (9,10), since it exists as a homodimer consisting of two subunits each of which contains 607 amino acids. During proteolytic activation, each of these subunits is cleaved to generate a heavy chain of 369 amino acids and a light chain or catalytic domain of 238 amino acids. The heavy chain of FXI provides binding surfaces for several blood coagulation proteins and is organized into four tandem repeat Apple domains designated A1, A2, A3, and A4 (9,10). Each of these four domains contains 90 to 91 amino acids, the sequences of which are 23-34% identical. The A1 domain provides a binding site for HMWK (11), a protein that promotes the activation of FXI by FXIIa (12). Both the A2 domain (13) and the A3 domain (14) have been implicated in the binding of FXIa to its substrate, FIX. The A4 domain contains Cys 321 that is responsible for the dimerization of FXI (15,16). In addition, the A4 domain binds to FXIIa (10).
Examination of the cellular site of FXI activation by FXIIa has resulted in evidence that platelets promote the proteolytic activation of FXI by FXIIa (17) and that FXI, in the presence of zinc ions, calcium ions, and HMWK, binds to activated platelets in a specific, reversible, and saturable manner with a dissociation constant (K d ) of ϳ10 nM (18). Previous work from this laboratory showed that a synthetic peptide (Asn 235 -Arg 266 ) from the A3 domain inhibited 125 I-FXI binding to platelets (inhibition constant (K i ) ϭ 10 nM) in the presence of HMWK, ZnCl 2 , and CaCl 2 (19,20). Hence, these studies indicate that the A3 domain mediates FXI binding to platelets.
A second possible site of FXI activation is the endothelial cell surface. Berrettini et al. (21) demonstrated that FXI binds to endothelial cells in the presence of HMWK, CaCl 2 , and ZnCl 2 with a K d(app) ϳ4.5 nM, whereas FXIa binds with higher affinity (K d(app) ϳ1.5 nM). The binding of FXI to the endothelial cell surface may be of functional significance in localizing coagulation to the site of vascular injury since activation of FXI can occur on the endothelial cell surface (21), where proteoglycans such as heparan sulfate are known to be present. Both Naito and Fujikawa (4) and Gailani and Broze (5) showed that, in the presence of an appropriate surface such dextran sulfate or sulfatides, FXI can be activated by thrombin or autoactivated by FXIa. In the presence of heparin, FXI can be activated by thrombin or by FXIa (5).
Another important function of heparin is to promote the inhibition of FXIa by protease nexin II (PN-2) by increasing the association rate of FXIa binding to PN-2 (22). This increased association rate results in a decrease in the K i from 300 pM, in the absence of heparin, to 30 pM in the presence of heparin (22). The effect of heparin and glycosaminoglycans on the binding of FXI to cell surfaces and other proteins is, therefore, an impor-tant area of study. The primary structure of FXI was examined for amino acid sequences that could bind heparin. This sequence analysis was based on work by Cardin et al. (23) which showed that, for several proteins, there is a consensus sequence for glycosami-noglycan recognition, BXBBXBX, where B represents a basic residue and X is a hydrophilic residue. Three possible heparin-binding sites were identified in FXI: -R 250 IKKSKA 256within the A3 domain, -Y 509 RKLRDK 515 -, and -Q 528 KRYRGHKI 536in the catalytic domain. The present study was undertaken to examine the interaction of FXI with heparin and to begin a detailed mapping of the putative heparin-binding site that mediates this interaction.

EXPERIMENTAL PROCEDURES
Materials-Human FXI and XIa were purchased from Hematologic Technologies, Inc. (Essex Junction, VT). Unfractionated heparin, p-nitrophenyl phosphate, HEPES, fluorescein mono-p-guanidinobenzoate HCl, and sodium m-periodate were purchased from Sigma. The mono-clonal antibody, 5F4, was prepared in our laboratory (13). Alkaline phosphatase-conjugated goat anti-mouse antibody was purchased from Jackson ImmunoResearch Laboratories, Inc. (West Grove, PA.). 1,9-Dimethyl-Methylene Blue was purchased from Aldrich. The chromo-genic substrate for measurement of FXIa, pyro-EPR-p-nitroanilide (S2366), was obtained from Chromogenix (M148 Indal, Sweden). Re-striction enzymes were from New England Biolabs (Beverly, MA). DNA markers, Pfu polymerase, ligase, and nucleotide triphosphates were from Amersham Pharmacia Biotech. Amine-coated plates were pur-chased from Corning Costar (Cambridge, MA). Streptavidin-coated plates were obtained from Pierce. Biotin-XX-hydrazide was purchased from Calbiochem. Radiolabeling of FXI-Purified FXI was radiolabeled by a minor modification of the IODO-GEN method (24) to a specific activity of 5 ϫ 10 6 cpm/g. The radiolabeled protein retained Ͼ98% of its biological activity. Peptide Synthesis-Peptides were synthesized on an Applied Biosys-tems 430A Peptide Synthesizer (Foster City, CA) by a modification of the procedure described by Kent and Clark-Lewis (25). The sequences of the synthetic peptides utilized in this study are given in Table I. All the peptides utilized in this work were rationally designed, conformation-ally constrained synthetic peptides based upon a previously published molecular model for the A3 domain of FXI (26). Each peptide was separately modeled using energy minimization calculations (26) that confirmed that the modeled peptides assumed a conformation similar to that of the A3 domain. Refolding, Reduction, and Alkylation of Peptides-A previously pub-lished method (27) was used to refold peptides containing cysteine residues. Alternatively, peptides were reduced with dithiothreitol and alkylated with iodoacetamide as described previously (27,28). High Performance Liquid Chromatography (HPLC)-The HPLC sys-tem employed was from Waters (Waters 600 Gradient Module, model 740 Data Module, model 46K Universal Injector, and Lambda-Max model 481 Detector, Milford, MA). Reverse-phase chromatography was performed using a Waters C8 Bondapak column, whereas gel filtra-tion was carried out using a Waters Protein-Pak 60 column as described previously (27,29,30). Characterization of Synthetic Peptides-All peptides utilized in this study were examined by HPLC (both reverse phase and gel filtration) and all demonstrated a single homogeneous peak (data not shown). When the refolded peptides were examined by HPLC (both reverse phase and gel filtration), single peaks with identical retention times to the original mixtures were observed, demonstrating the presence of a single homogeneous mixture of refolded peptides. The results were the same after reduction and alkylation of these same peptides. All reduced and alkylated or refolded peptides were examined for free SH groups using the Ellman reagent (5,5Ј-dithiobis(2-nitrobenzoic acid)). It was determined (31) that there was less than 0.02 mol of free SH per mol of peptide, which further verifies that these refolded peptides were homo-geneous preparations consisting of intramolecular disulfide-bonded peptide. The upstream primer (U) contains a BamHI restriction site and a cleavage site for FXa (IEGR). Its sequence is as follows: 5ЈGCGGATC-C(ATCGAGGGTAGA)GCTTGTATTAGGGACATTTTCCCT. The downstream (D) primer contains a HindIII restriction site and its sequence is as follows: 5ЈGCAAGCTTTACACTGGGATGCTGTGCCTGCA. The mutagenic primers are as follows: 1) R250-A, 5ЈTTGCCCAGTAC-AGCCATTAAAAAGAGC; 2) K252-A, 5ЈAGTACACGCATTGCAAAGA-GCAAAGCT; 3) K253-A, 5ЈACACGCATTAAAGCGAGCAAAGCTCTT; 4) K255-A, 5ЈATTAAAAAGAGCGCAGCTCTTTCTGGT.
For polymerase chain reaction amplification of the wild type and mutant A3 domains, the procedures were based on a technique developed by Picard et al. (32). The insert was cut with BamHI and HindIII restriction enzymes using the procedures from New England Biolabs (Beverly, MA) (supplied with the enzymes). The insert was ligated to the QIAexpress pQE-9 Vector, and transformations were carried out using competent K12-derived Escherichia coli M15 (pREP4). Expression from the pQE-9 vector was induced by the addition of isopropyl-␤-D-thiogalactopyranoside.
Expression, Isolation, and Folding of the Recombinant A3 Domain (rA3)-Isolation of the rA3 was carried out using a nickle-nitrilotriacetic acid (Ni-NTA) resin that binds the 6-His tag on the expressed protein. The E. coli containing the A3 or mutant A3 inserts (as verified through DNA sequence analysis) were inoculated in 20 ml of LBKA media (10 mg/ml Bacto-tryptone, 5 mg/ml Bacto-yeast extract, 5 mg/ml NaCl, 100 g/ml ampicillin, and 25 g/ml kanamycin) overnight at 37°C with shaking (225 rpm). To express the rA3, 20 ml of the overnight culture was inoculated in 1 liter of LBKA and grown at 37°C with shaking, until the A 600 reached 0.6 -0.7. Expression was induced by adding isopropyl-␤-D-thiogalactopyranoside to a final concentration of 1 mM, and the culture was grown for an additional 3 h. The bacteria were harvested by centrifugation at 4000 ϫ g for 20 min.
To lyse the E. coli, the pellet was suspended in 6 M Gdn-HCl, 0.1 M sodium phosphate, 0.01 M Tris, pH 8.0 (2 g wet weight per 10 ml of buffer), and stirred at room temperature for 1 h. The lysate was centrifuged at 10,000 ϫ g for 15 min at 4°C to pellet the cellular debris. The supernatant was mixed with 50% Ni-NTA slurry (1 ml of Ni-NTA/4 ml of lysate) and stirred at room temperature for 45 min. The lysate/ resin mixture was loaded onto a 5-ml polypropylene column (Qiagen Inc., Chatsworth, CA) and washed with 50 ml of buffer A (8 M urea, 0.1 sodium phosphate, 0.01 M Tris), pH 8.0, and again with buffer A containing 10 mM ␤-mercaptoethanol. The column was further washed with 50 ml of buffer A, pH 7.5, and 50 ml of buffer A, pH 6.5.
The rA3 was eluted with 25 ml of buffer A at pH 5.5, and protein concentration was measured spectrophotometrically (⑀ ϭ 6400 M Ϫ1 at 280 nm absorption). Typically, for a 1-liter preparation, 20 -25 mg of rA3 was isolated. The pH was adjusted to 8.8, and the protein was concentrated with Centriprep 10 (Amicon, Inc., Beverly, MA) at 2600 ϫ g to a final concentration of 2 mg/ml. To increase the solubility of rA3 and to aid in the folding process, the rA3 was sulfonated (33, 34) by adding 200ϫ molar excess of Na 2 SO 3 , 50ϫ molar excess of cysteine and EDTA to a final concentration of 10 mM. The mixture was stirred at room temperature overnight.
To decrease the urea concentration and remove Na 2 SO 3 , cysteine, and EDTA, the rA3 was dialyzed against 100ϫ its volume in 4 M urea, 10 mM Tris buffer, pH 8.8, using Slide-A Lyzer 10K dialysis cassette (Pierce) for 1 h. The dialysis (1 h) was repeated five times with the same buffer but with different concentrations of urea (2, 1, 0.5, 0.25, and 0 M urea). To stabilize the rA3, 10 M ␣-cyclodextrin (␣-CD) was added into the solution containing the rA3 and stirred at 4°C overnight (data from our laboratory have shown that ␣-CD can prevent rA3 aggregation, data not shown). To remove the 6-His tag, 30 g of bovine FXa was added to each aliquot (2 mg/ml) of rA3 and incubated at 37°C for 3 h or overnight.
FXa was separated from the rA3 using Centriprep-30 (Amicon, Inc., Beverly, MA) at 2600 rpm with the rA3 filtered through the membrane (filtrate) and FXa retained on the membrane. To separate the cleaved from the uncleaved rA3, the filtrate was mixed with 50% Ni-NTA slurry (1 ml of Ni-NTA/4 ml of filtrate) and stirred at room temperature for 45 min. The filtrate/resin mixture was loaded onto a 5-ml polypropylene column (Qiagen Inc., Chatsworth, CA), and the flow-through (which contained the cleaved A3) was collected. The column was washed again with 25 ml of 10 mM Tris, pH 8.8, and the flow-through was also collected. The rA3 protein concentration was measured spectrophotometrically (⑀ ϭ 6400 M Ϫ1 at 280 nm absorption).
The concentration of the cleaved rA3 was adjusted to 0.05 mg/ml with 10 mM Tris, pH 8.8, containing final concentrations of 10 mM DTT and 10 M ␣-CD. The cleaved rA3 was stirred at 4°C for 48 h to promote intra-disulfide bonding. Before use, the rA3 was applied to a Sephadex G-25 (Sigma) spin column to remove the ␣-CD and concentrated with Centricon 3 (Amicon, Inc., Beverly, MA) to the desired experimental concentrations.
The rA3 was examined by HPLC and matrix-assisted laser desorption ionization time of flight-mass spectrometry (analysis conducted by the Protein Chemistry Laboratory at the University of Pennsylvania, Philadelphia). Data indicated a single peak demonstrating the presence of a single homogeneous species of protein with the molecular weight of 10,519 which is consistent with the calculated molecular weight of the rA3 which is 10,423. Ellman's reagent (5,5Ј-dithiobis(2-nitrobenzoic acid)) was used to determine free SH groups (31), which indicated less than 0.02 mol of free SH per mol of rA3.
Coating Heparin to the Amine-coated Plates-Assays were performed in 96 removable well amine-coated plates (Costar, Cambridge, MA). Wells were incubated with 200 l of heparin (2 mg/ml) (Sigma) in 10 mM Tris, pH 7.4, or with 10 mM Tris in water, pH 7.4 (blank plate), overnight at 37°C. Wells were washed three times with 0.2 ml of 10 mM Tris, pH 7.4, and incubated for 1 h at room temperature with 0.2 ml of 1% (w/v) non-fat dry milk in 10 mM Tris, pH 7.4, to block all nonspecific binding sites. The amount of heparin bound was determined according to the method of Klompmakers and Hendriks (35). The amount of heparin bound in the amine-coated plate was determined to be 3-5 g per well, whereas no heparin was detected in the blank plate. In each experiment, the amount of heparin bound per well and the amount of 125 I-labeled FXI bound per well were determined. The amount of 125 I-FXI bound per heparin-containing well was a linear function of the amount of heparin bound per well (r ϭ 0.99, p Ͻ 0.01). Any variations from experiment to experiment in the competition binding studies were obviated by expressing the results as percentage of total 125 I-FXI bound.
Coating Heparin to Streptavidin-coated Plates-Heparin was labeled with biotin by sodium m-periodate oxidation (10 mM final concentration), desalting on a G-25 column, and incubation with biotin-XXhydrazide (17 mM final concentration). Excess biotin was removed by dialysis against 0.05 M HEPES, 0.15 M NaCl (HBS) in a Centricon-3 apparatus (Amicon, Inc., Beverly, MA). Biotin incorporation was determined to be ϳ1.4 mol/mol heparin using 4Јhydroxyazobenzene-2-carboxylic acid and avidin. Biotin-heparin (10 g/100 l) was incubated in the wells of a streptavidin-coated plate for 2 h. After washing twice with 10 mM Tris, 0.15 M NaCl (TBS), pH 7.4, either FXIa or 125 I-labeled FXI was applied to the wells and incubated as described below. The binding capacity of the streptavidin plate is stated by the manufacturer (Pierce) to be 25 pmol of biotin per well.
Heparin Binding Assays-Heparin-containing wells were washed twice with 0.2 ml of 10 mM Tris, pH 7.4 (for amine coated plate), or TBS (for streptavidin coated plate) and incubated with 125 I-FXI for 2 h. Each well was washed three times with 0.2 ml of 10 mM Tris, pH 7.4, and wells were removed and counted in a Wallac 1470 Wizard (Wallac Inc., Gaithersburg, MD) gamma counter to determine the total binding. Binding of 125 I-FXI was compared with nonspecific binding of 125 I-FXI determined in the presence of a 50-fold molar excess of unlabeled FXI. Specific binding was obtained by subtracting the nonspecific binding from total binding. The calculated values for specific binding were used to determine the K d for FXI binding to heparin.
Surface Plasmon Resonance-The binding affinity of FXI for heparin was determined by kinetic analysis using a Biacore 2000 flow biosensor (Biacore, Inc., Uppsala, Sweden). Biotin-heparin, 50 g/ml, was bound to a streptavidin-coated sensor chip to a response unit value of 280. Biotin-XX-hydrazide in TBS was applied to a second flow cell to determine nonspecific binding. Serial dilutions of FXI were flowed across the cells at a rate of 30 l/min with an association time of 1 min. Dissociation was determined over a 3-min period, with an initial HBS flow, followed by a 1 M NaCl flow for 2 min. Data were analyzed using Biaevaluation version 3.0 (Biacore, Inc., Uppsala, Sweden) according to a model which simultaneously fits association and dissociation rates to a value of 2 Ͻ10 to determine the binding constants based on a two-state conformational change equation (36). The 2 value is the square of the differences between the theoretical ideal curve and the actual curve and was calculated according to Equation 1.
where r f indicates fitted value at a given point; r x indicates experimental value at that point; n ϭ the number of data points; and p indicates the number of fitted parameters. Competition Experiments-The 125 I-FXI (22 nM) was mixed with samples of various competitor ligands at the indicated concentrations and incubated in a heparin-containing amine-coated plate or an aminecoated plate without heparin (blank plate). After 2 h, samples were removed, and each well was washed three times with 0.2 ml of 10 mM Tris, pH 7.4. Wells were removed and counted in a Wallac 1470 Wizard gamma counter. One hundred percent binding of 125 I-FXI represented an average of 250,000 cpm bound, whereas 0% binding represented background binding (250 cpm) in which labeled FXI was incubated in the blank plate wells. The concentration of competitor which displaced 50% of bound 125 I-FXI (IC 50 ) was determined by plotting the amount of 125 I-FXI bound to heparin versus the amount of competitor ligand added. The K i was calculated using the following equation: IC 50 ϭ (1 ϩ [S]/K d ) K i ; where S was the concentration of 125 I-FXI used in these experiments (held constant at 22 nM) and K d was the value used from "Heparin Binding Assays." Identical experiments were carried out using heparin-containing streptavidin-coated plate or streptavidin-coated plate without heparin with the exception that the wash buffer used was TBS. Separate experiments revealed that the amount of FXI bound to heparin was the same using these two elution buffers.
Binding of FXIa to Heparin-A concentration range of FXIa was added in 100-l aliquots to heparin-containing microtiter wells in triplicate. For determination of nonspecific binding, the same concentration range was added to a set of blank plate wells. The wells were then incubated at 37°C for 2 h. The wells were washed twice with PBS with 0.05% Tween 20 (PBST) and once with PBS alone. The wells were then incubated with 100 l/well of 1 g/ml of 5F4, a FXI light chain-specific monoclonal antibody, for 1 h at 37°C. Since the 5F4 monoclonal antibody binds with high affinity to FXIa (K d ϳ0.78 ϫ 10 Ϫ10 M) (37) and a saturating concentration of antibody (6.3 ϫ 10 Ϫ9 M) was used, this assay provides an excellent determination of bound FXIa. After washing, 100 l per well of a 1:1000 dilution of alkaline phosphatase-conjugated goat anti-mouse antibody was added for a 1-h incubation at 37°C. After washing four times with PBST and once with PBS, 150 l of a 1 mg/ml solution of p-nitrophenyl phosphate was added, and color development was allowed to proceed for 20 -30 min. The reaction was stopped with 2 N NaOH and absorbance read at 405 nm. The absorbance of FXIa bound nonspecifically to blank plates was subtracted from the absorbance for wells containing heparin.

RESULTS
Binding to Heparin of FXI and FXIa-In the presence of a negatively charged surface such as heparin, dextran sulfate, or sulfatides, FXI can be activated to FXIa by thrombin or autoactivated by FXIa (4,5). Therefore, before carrying out experiments designed to examine the binding of FXI to heparin and to determine the affinity of this interaction, it was important to determine whether FXI exposed to a heparinized amine-coated plate undergoes conversion to FXIa. Consequently, we conducted an experiment to determine the amount of FXIa generated when the zymogen FXI was added to a heparinized plate versus a non-heparinized plate (Fig. 1A). The data indicated that after 2 h incubation at 37°C, about 75% of the 22.5 nM FXI added was converted to FXIa in the heparin-containing plate but not the blank plate. The generation of FXIa is most likely to be the result of autoactivation of FXI by a trace amount of FXIa present in the FXI added to heparin-containing wells, although no FXIa amidolytic activity was detected in the FXI preparation. To confirm this conclusion, FXI was incubated in heparin-containing wells and examined by SDS-polyacrylamide gel electrophoresis (Fig. 1B). The results indicated that after a 2-h incubation at 37°C on a heparinized plate, the majority of FXI added was converted to FXIa (lane 2) as indicated by the appearance of the heavy chain (50 kDa) and the light chain (30 kDa) as compared with intact FXI (a single band at 80 kDa, lane 1) on the blank plate under reducing conditions. Additionally, when a 50-fold molar excess of the serine protease inhibitor 4-(2-aminoethyl)-benzenesulfonyl fluoride (AEBSF) was added to FXI and incubated in heparin-containing wells, no conversion to FXIa was observed (a single band at 80 kDa, lane 3) indicating that AEBSF is able to prevent autoactivation of FXI on the heparinized plate.
To study the binding of FXI to heparin, it is therefore necessary to prevent autoactivation of FXI to FXIa on the heparin surface. Initially, the binding of 125 I-FXI to heparin-coated microtiter plates was studied in the presence or absence of a 50-fold molar excess of AEBSF. In the presence of AEBSF, saturable binding of 125 I-FXI to heparin was observed with an apparent K d ϳ1.1 ϫ 10 Ϫ7 M, whereas in the absence of AEBSF, the apparent K d was 1.5 ϫ 10 Ϫ8 M (Fig. 2A). This suggests that the binding of FXIa to heparin is tighter than that of FXI and that the binding observed in the absence of AEBSF is due to the presence of a mixture of FXI and FXIa.
To examine more specifically the affinity of FXIa binding to heparin, direct binding studies were performed. The K d for FXIa was determined to be ϳ0.7 ϫ 10 Ϫ9 M using amine-coated plates, where heparin was non-covalently immobilized on the amine surface (Fig. 2B), compared with 1.1 ϫ 10 Ϫ7 M for FXI ( Fig. 2A). The K d for FXIa using streptavidin plates, where heparin was covalently immobilized on the streptavidin surface, was 1.45 ϫ 10 Ϫ9 M (Table II), in close agreement with the amine plate results. Thus, the affinity of FXIa for heparin is ϳ150-fold greater than that of FXI. This also suggests that there are additional heparin-binding sites exposed in FXIa that are not accessible in FXI. The identification and characterization of these putative heparin-binding sites in FXIa is the subject of a separate study in our laboratory. The remainder of the present studies focused on the identification and characterization of heparin-binding sites within the zymogen, FXI.
Binding of FXI to Heparin Using Various Binding Assays-The binding of heparin to the amine-coated plates occurs through a charge interaction, in which the negatively charged groups of heparin are bound to the positively charged amine groups on the surface of the plates. When the plate undergoes extensive washing to carry out the binding assays, there is a possibility that some of the heparin may be displaced from the amine-coated surface. To address this question, we compared the K d and K i values of FXI binding to heparin covalently linked to streptavidin-coated plates and to non-covalently linked heparin on the amine-coated plates. All subsequent studies were carried out in the presence of a 50-fold molar excess of AEBSF to prevent conversion of FXI to FXIa. When heparin was covalently immobilized to a surface, the apparent K d ϳ90 ϫ 10 Ϫ9 M for FXI binding to heparin was similar to that obtained when heparin was non-covalently immobilized on the amine-coated surface (K d ϳ110 ϫ 10 Ϫ9 M) (Table II). We also carried out competition studies in which unlabeled FXI was used to compete with 125 I-FXI binding to heparin attached covalently (K i ϳ105 ϫ 10 Ϫ9 M) or non-covalently (K i ϳ110 ϫ 10 Ϫ9 M) to a solid surface (Table II). These results confirm the conclusion that the saturable binding observed with 125 I-FXI is readily reversible and that the labeled and unlabeled FXI bind with equal affinity. Furthermore, we also examined the direct binding of FXI to heparin using surface plasmon resonance. Biotinylated heparin was covalently immobilized on a streptavidin biosensor chip, and the K d values of FXI binding to   2. Factor XI and factor XIa binding to heparin. A, increasing concentrations of 125 I-factor XI in the presence (E) and absence (q) of AEBSF were incubated for 2 h at 37°C in heparin-coated wells. Each point is an average of triplicate determinations, and the maximum variation of cpm bound for each observation was Ͻ2% of total cpm bound. Results represent the data of three separate studies with nonspecific binding subtracted. B, increasing concentrations of FXIa were added to amine plates on which heparin had been coated. The microtiter plate was incubated for 2 h at 37°C. Using ELISA methodology (see "Experimental Procedures"), the amount of factor XIa bound was detected through use of 5F4, a monoclonal antibody that binds to the catalytic domain of factor XIa. Results represent the combined data of 10 separate studies with nonspecific binding subtracted. heparin were measured at three different concentrations (Table III). Using a two-state conformation change (36) equation, the 2 of Ͻ10 indicated that the measured K d values closely fit the model (Table III). This analysis provided an average K d ϳ118 ϫ 10 Ϫ9 M for FXI binding to heparin. Therefore, based on combined results from both equilibrium and kinetic binding studies, the true K d for FXI binding to heparin is ϳ110 ϫ 10 Ϫ9 M. For the subsequent binding studies, we chose to employ the amine-coated plates since competition binding studies were complicated by nonspecific binding of the rA3 to the streptavidin-coated microtiter plates and to the streptavidin-coated biosensor chips.
Localization of the Heparin-binding Site on FXI-We have determined that the A3 domain of FXI contains a heparin binding consensus sequence, -R 250 IKKSKA 256 -. In order to verify that the A3 domain of FXI is the heparin-binding site for FXI, the rA3 was expressed in E. coli, purified, and used in competition binding experiments to determine its capacity to inhibit 125 I-FXI binding to heparin (Fig. 3A). It was found that the rA3 inhibited 125 I-FXI binding to heparin with a K i value of ϳ1.4 ϫ 10 Ϫ7 M which is similar to the inhibitory effect of unlabeled FXI (K i ϳ1.1 ϫ 10 Ϫ7 M, Fig. 3A) on 125 I-FXI binding to heparin and also similar to the K d (ϳ1.1 ϫ 10 Ϫ7 M, Fig. 2A) of 125 I-FXI binding to heparin. When FXIa was used to compete with 125 I-FXI binding to heparin in the presence of AEBSF, it inhibited with a K i ϳ1.5 ϫ 10 Ϫ8 M which confirms the conclusion that FXIa binds to heparin more tightly than FXI.
Our previous studies of the interactions of FXI with HMWK, FIX, FXIIa, thrombin and, the activated platelet surface (13,26,29,30,38,39) have delineated binding sites for these ligands within the carboxyl-terminal half of each of the four Apple domains of FXI. Molecular modeling with all four Apple domains suggests that a similar pattern of folding is present in each domain (11,13,21,26,38,39). Because of these predicted structural similarities and the 23-34% sequence identity among these four tandem repeat domains, we previously prepared a panel of heavy chain-derived peptides (Table I). Initially, we screened these peptides to determine whether they affected 125 I-FXI binding to heparin. Peptide Asn 235 -Arg 266 from the A3 domain inhibited 125 I-FXI binding to heparin with a K i ϳ3.4 ϫ 10 Ϫ7 M (Fig. 3B). In contrast, the concentration of the peptide derived from the A2 domain (Ala 134 -Ala 176 ) required to inhibit FXI binding to heparin 50% was ϳ1,000-fold higher (K i ϳ1 ϫ 10 Ϫ4 M). Peptides from the A1 domain (Phe 56 -Ser 86 ) and the A4 domain (Ala 317 -Gly 350 ) had no effect on FXI binding to heparin at concentrations as high as 10 Ϫ4 M (Fig. 3B).
Previously, we have determined that the A3 domain peptide, Asn 235 -Arg 266 , interacts with the platelet surface (26). Although detailed structural information is not available for the A3 domain, we have utilized a molecular model of this domain to make testable predictions about the structure and function of the platelet-binding site in the A3 domain (26). Our experiments revealed that the sequence of amino acids (Pro 229 -Arg 266 ) within the A3 domain of FXI contains three anti-parallel ␤-strands connected by ␤-turns that comprise a continuous surface utilized for binding to the platelet surface (26). Since peptide Asn 235 -Arg 266 inhibits the binding of 125 I-FXI to heparin (Fig. 3B) and this sequence of amino acids contains a consensus sequence for a heparin binding domain (Arg 250 -Ala 256 , -RIKKSKA-), we prepared conformationally constrained cyclic peptides comprising these peptide loop structures to determine whether they might assume a conformation . 125 I-Factor XI (22 nM) was incubated for 2 h at 37°C in heparin-coated wells with the designated protein peptide at the concentration specified or with buffer solution. Details of the assay are described under "Experimental Procedures." Each point is an average of triplicate determinations, and the maximum variation of cpm bound for each observation was Ͻ2% of total cpm bound. One hundred percent binding of factor XI represents an average of 250,000 cpm bound, whereas 0% binding of factor XI represents 0% bound upon subtracting an average of 250 cpm representing a control in which labeled factor XI was incubated in wells coated with non-fat dry milk.
that may also comprise a heparin binding domain. These peptides are identical to those tested to delineate the plateletbinding site (26). Cysteine residues were introduced at the amino and carboxyl terminus of each peptide so that the resulting disulfide bond might stabilize the loop-like structure (26). The peptides designated Thr 241 -Leu 246 and Pro 229 -Gln 233 had no effect on FXI binding to heparin at concentrations up to 1 ϫ 10 Ϫ4 M (Fig. 3C). By comparison, these peptides inhibited FXI binding to platelets (26). The peptide, Thr 249 -Phe 260 , inhibits FXI binding to heparin with a K i ϳ2.1 ϫ 10 Ϫ7 M, which is similar to that of the rA3 (K i ϳ1.4 ϫ 10 Ϫ7 M) (Fig. 3A). These experiments suggest that there are amino acid sequences in the A3 domain that bind platelets but are not important for heparin interaction as well as common sequences involved in binding both ligands.
Fine Mapping of the Heparin-binding Site in the A3 Domain of FXI-A comparison was made between the heparin-binding site on the A3 domain of FXI and the A3 domain of prekallikrein (PK) (PK-A3). Although FXI and PK are 58% identical, the heparin-binding site in the FXI-A3 domain was not found in the PK-A3 domain (Table I). Therefore, to fine map the heparin site on the A3 domain, a series of synthetic peptides were made in which basic amino acids possibly involved in heparin binding within the putative heparin binding consensus sequence of the FXI-A3 were replaced with the corresponding amino acids from PK-A3 and used in competition heparin-binding experiments with 125 I-FXI (Fig. 4A). The peptide Thr 249 -Phe 260 , K253Q failed to compete with 125 I-FXI, whereas the other peptides (Thr 249 -Phe 60 , R250S; Thr 249 -Phe 260 , K252P, and Thr 249 -Phe 260 , K255A) competed normally with 125 I-FXI binding to heparin. These data show that residue Lys 253 plays a major role in heparin binding, and its absence may abolish the heparin binding capacity of the rA3.
To test further this hypothesis, a mutational analysis of the putative heparin-binding site was carried out on the background of the recombinant FXI-A3 domain. The mutants rA3 R250A, rA3 K252A, rA3 K253A, and rA3 K255A were constructed and tested for their capacity to compete with 125 I-FXI in heparin binding (Fig. 4B). The mutants rA3 K253A and rA3 K252A were deficient in their inhibitory capacities, with K i values of ϳ1.4 ϫ 10 Ϫ4 M and ϳ1.7 ϫ 10 Ϫ5 M, respectively, whereas the mutants rA3 R250A (K i ϳ1.8 ϫ 10 Ϫ7 M) and rA3 K255A (K i ϳ4.6 ϫ 10 Ϫ7 M) were entirely normal in their capacity to compete with 125 I-FXI in heparin binding. The binding and the inhibitory constants from these studies are summarized in Table IV.  Details of the assay are described under "Experimental Procedures." Each point is an average of triplicate determinations, and the maximum variation of cpm bound for each observation was Ͻ2% of total cpm bound. One hundred percent binding of factor XI represents an average of 250,000 cpm bound, whereas 0% binding of factor XI represents 0% bound after subtracting an average of 250 cpm representing a control in which labeled factor XI was incubated in wells coated with non-fat dry milk.

DISCUSSION
There is considerable interest in determining the specific mechanism by which heparin and heparin-like glycosaminoglycans interact with coagulation proteins to regulate the hemostatic process. In one of the most completely studied systems, the binding of heparin to a specific domain on human antithrombin III involves specific basic residues within the molecule (40). Additionally, heparin has been found to bind to a number of other proteins in human plasma (41). Consensus sequences for heparin recognition were determined as -BXB-BXBX-and -XBBBXXBX-, where B is a basic residue and X is a hydrophilic residue (23). Since FXI was shown to bind to immobilized heparin, there should be binding sites present within its structure for heparin-like glycosaminoglycans. The aims of the present study were to examine the binding of FXI to heparin, to investigate the role of the Apple domains in the binding of FXI to heparin, and to delineate a sequence of amino acids within FXI which mediates heparin binding.
Studies conducted by Naito and Fujikawa (4) and by Gailani and Broze (5) demonstrated that when FXI is exposed to a negatively charged surface such as dextran sulfate or sulfatides, it can be readily converted to FXIa by a trace amount of FXIa. Data from our laboratory (22) demonstrated that heparin functions by a template mechanism to increase the association rate of FXIa and PN-2 with a resultant decrease in the inhibition constant. It can be concluded from these observations that there are heparin-binding sites within FXI and FXIa; however, it is not clear whether the heparin-binding site utilized by the zymogen is the same or different from that utilized by the enzyme. Initially, we conducted experiments to determine whether or not FXI can be converted to FXIa on a heparin surface and to determine the relative affinities of FXI and FXIa binding to heparin. Since it is clear from these experiments that FXI exposed to a heparin-containing plate is autocatalytically activated (Fig. 1, A and B) and that the affinity of FXIa binding to heparin (K d ϳ0.7 ϫ 10 Ϫ9 M, Fig. 2B) is considerably higher than that of FXI (K d ϳ1.1 ϫ 10 Ϫ7 M, Fig. 2A), we have focused our present studies on a determination of the structural requirements for interaction of the zymogen (FXI) with heparin. Rational design and interpretation of these experiments require the inclusion of an effective serine protease inhibitor (AEBSF) to prevent the conversion of FXI to FXIa.
Our subsequent studies were focused upon the identification and fine mapping of the site within FXI (the zymogen) which mediates its binding to heparin. By having developed two equilibrium binding techniques (utilizing heparin non-covalently bound to amine-coated microtiter plates and biotinylated heparin bound to streptavidin-coated plates) and a kinetic method (utilizing surface plasmon resonance with streptavidin-coated biosensor chips with bound biotinylated heparin) for examining FXI binding to heparin, we next chose the most appropriate method for our subsequent studies. These three methods give very similar estimates of the affinity of FXI binding to heparin (K d ϭ 0.9 -1.18 ϫ 10 Ϫ7 M, Tables II and III). Moreover, the values of K d (0.9 -1.1 ϫ 10 Ϫ7 M) in equilibrium binding experiments were almost identical to the values for K i (1.05-1.10 ϫ 10 Ϫ7 M) in competition studies with unlabeled FXI. Therefore, we conclude that FXI binds in a saturable and reversible manner to immobilized heparin and that the fidelity of FXI as a probe for quantitative assessment of this interaction is unaffected by the radiolabeling procedure, thereby justifying the subsequent competition studies designed to map the heparin binding domain(s) in FXI. However, we found (data not shown) that the isolated rA3 domain binds very tightly and nonspecifically to streptavidin-coated surface either in microtiter plates or in surface plasmon resonance studies. We were therefore forced to choose the amine-coated plates to immobilize heparin for our subsequent studies. The theoretical disadvantage of heparin elution during the wash steps of the FXI binding and competition assays does not appear to pose a significant problem since estimates of affinity are almost identical using all three methods.
Since the recombinant A3 domain of FXI competes with 125 I-FXI for binding to heparin with the same K i (ϳ1.4 ϫ 10 Ϫ7 M) as does FXI (K i ϳ1.1 ϫ 10 Ϫ7 M, K d ϳ1.1 ϫ 10 Ϫ7 M, see Table  III), we conclude that all of the binding energy mediating the interaction of FXI with heparin resides within the A3 domain. This conclusion is confirmed by the fact that two A3 domain peptides (Asn 235 -Arg 266 and Thr 249 -Phe 260 ) have nearly equivalent potency (K i ϳ3.4 ϫ 10 Ϫ7 M and ϳ2.1 ϫ 10 Ϫ7 M, respectively) in competing with FXI for heparin binding, whereas peptides derived from the A1, A2, and A4 domains have no binding activity (Fig. 3B). We further conclude that the heparin-binding site within the A3 domain resides exclusively within a conformationally constrained loop structure (Thr 249 -Phe 260 ) containing a heparin binding consensus sequence -R 250 IKKSKA 256 -.
In our attempt to accomplish fine mapping of this A3 heparin-binding site, we have pursued a dual approach. First, we have prepared conformationally constrained synthetic peptides with each basic residue within the putative heparin binding consensus sequence replaced by the corresponding amino acid present in PK, which has 58% sequence identity with FXI and yet does not contain a heparin binding consensus sequence and does not bind to heparin ("homolog scanning"). Second, we have prepared recombinant A3 constructs in which each of the basic residues comprising the putative heparin binding consensus sequence was replaced by an alanine ("alanine scanning"). The two approaches yielded complementary information ( Fig. 3 and Table II). Both implicated a single lysine residue (Lys 253 ) as essential for heparin binding since replacement either by a glutamine in the synthetic peptide or by an alanine in rA3 greatly diminished or abolished binding activity, whereas replacements of Arg 250 or Lys 255 with the corresponding PK residues in synthetic peptides or alanines in rA3 had no effect on heparin binding. When Lys 252 was replaced by alanine in rA3, heparin binding activity was decreased 100-fold (K i ϳ1.7 ϫ 10 Ϫ5 M, Fig. 3B, Table II), whereas when it was replaced by a proline in the synthetic peptide, only a minimal loss of heparin binding activity was observed. Since proline has the capacity to induce turn formation in the ␣-carbon backbone of a protein or peptide, it is possible that the conformation of the proline containing peptide (Thr 249 -Phe 260 , K252P) may actually favor heparin binding, whereas replacement with alanine in rA3 is likely to be a more conservative replacement of a basic amino acid side chain with an uncharged methyl group. We conclude that of the four basic residues within the A3 heparin binding consensus sequence (-R 250 IKKSKA 256 -), Lys 253 is essential and Lys 252 is probably important in heparin binding.
There are additional possible heparin-binding sites within the FXI molecule, including -Y 509 RKLRDK 515 -and -Q 528 KRYR-GHKI 536 -, both of which reside in the catalytic domain. Since the present studies support the view that the A3 domain site -R 250 IKKSKA 256 -contains all the binding energy for FXI (zymogen)-heparin interaction, we believe it is unlikely that the catalytic domain sites play a part in zymogen binding to heparin. However, since the enzyme (FXIa) binds to heparin with ϳ150-fold higher affinity (K d ϳ0.7 ϫ 10 Ϫ9 M) than the zymogen, it is possible that these catalytic domain sites mediate this high affinity FXIa-heparin interaction. This possibility is currently being investigated in our laboratory.
The potential physiological relevance of our current studies relates not only to the interactions of FXI with heparin, which is present in circulating blood only when administered therapeutically as an anticoagulant but also to the interactions of FXI with intravascular surfaces which have heparin-like membrane receptors for FXI that are important for FXI activation (17)(18)(19)26). Thus the endothelial cell surface membrane contains heparan sulfate glycosaminoglycans which bind both FXI and FXIa and can promote the activation of FXI and FIX (21). The affinity (K d ϳ1.5 ϫ 10 Ϫ9 M) of FXIa binding to endothelial cells (26) was found to be nearly identical to the affinity of FXIa binding to heparin (K d ϳ0.7 to 1.45 ϫ 10 Ϫ9 M) in the present study. However, the affinity (K d ϳ4.5 ϫ 10 Ϫ9 M) of FXI binding to endothelial cell (21) was found to be 45-fold higher than the affinity of FXI binding to heparin (K d 1.1 ϫ 10 Ϫ7 M) in the present study. When we did not include the serine protease inhibitor AEBSF with FXI in our heparin binding studies, the FXI was converted to FXIa (Fig. 1A), and the observed binding was that of a mixture of FXI and FXIa ( Fig. 2A). Furthermore, preliminary studies in our laboratory indicate that the K d for binding of FXI to human umbilical vein endothelial cells in the presence of AEBSF is much closer to that for binding to heparin (i.e. K d Ͼ10 Ϫ7 M), whereas when AEBSF is excluded, FXI is converted to FXIa and the binding affinity increases (data not shown). Therefore, we suspect that FXI binds to heparin-like molecules on endothelium and that the mechanisms of these interactions may be similar.
Previous studies from our laboratory have defined a plateletbinding site, consisting of three antiparallel ␤-strands connected by ␤-turns (Pro 229 -Gln 233 , Thr 241 -Leu 246 , and Thr 249 -Phe 260 ) within the A3 domain of FXI (30). Our present study provides evidence that, while contiguous and partially overlapping, these sites have separate and distinct structural determinants. Thus, whereas the platelet-binding site consists of the three above-mentioned loop structures which act synergistically to generate a contact surface for interaction with a platelet receptor (30), the entirety of the heparin-binding site resides within the sequence Thr 249 -Phe 260 , and more specifically within the heparin binding consensus sequence -R 250 IKKS-KA 256 -. A detailed mutational analysis of the structure of the platelet-and heparin-binding sites is in progress in our laboratory which provides definitive evidence (data not shown) that the amino acid side chains that mediate binding of FXI to platelets are different from those that mediate binding to heparin. It will be important to carry out a similar analysis of the residues involved in FXI binding to endothelium.
Ultimately, it will be important to obtain detailed threedimensional structural information about the conformation of the A3 domain of FXI in complex with heparin by employing techniques such as circular dichroism, nuclear magnetic resonance, and/or x-ray crystallography. Some information exists concerning the conformation of heparin-binding domains of apolipoprotein E (apoE) and apolipoprotein B-100 (apoB-100), each of which increases its helical and ␤-strand content upon binding heparin (23) suggesting that heparin induces amphipathic helical structures in apoE and apoB-100 heparinbinding peptide regions. It has been hypothesized that peptide regions exposed to heparin segregate their basic residues primarily on one side of the helical face, forming a region of high positive charge density (23). The -XBBXBX motif ensures the segregation of the basic residues in the consensus sequence to one side of the helical face. Thus, the binding of heparin to the consensus sequence of the A3 domain of FXI may increase its ␤-strand and ␣-helical character depending on the precise organization of basic and non-basic residues in the A3 domain sequence. It also remains to be seen whether the binding of the A3 domain to a putative polyanionic platelet receptor will induce significant changes in secondary structure. The data presented in the present paper and previously (26) indicate that the heparin-binding site, although juxtaposed or superimposed upon the platelet-binding site, has separate and distinct structural determinants.