A Sequential Mechanism for Exosite-mediated Factor IX Activation by Factor XIa*

Background: Factor XIa proteolytically activates factor IX. Results: XIa cleaves IX after Arg145, forming IXα, and then after Arg180, forming IXaβ. Both reactions require substrate binding to the XIa A3 domain. Conclusion: XIa activates IX by an exosite-mediated release-rebind mechanism. Efficiency of the second cleavage is enhanced by changes resulting from the first cleavage. Significance: The data support a new model for IX activation by XIa. During blood coagulation, the protease factor XIa (fXIa) activates factor IX (fIX). We describe a new mechanism for this process. FIX is cleaved initially after Arg145 to form fIXα, and then after Arg180 to form the protease fIXaβ. FIXα is released from fXIa, and must rebind for cleavage after Arg180 to occur. Catalytic efficiency of cleavage after Arg180 is 7-fold greater than for cleavage after Arg145, limiting fIXα accumulation. FXIa contains four apple domains (A1–A4) and a catalytic domain. Exosite(s) on fXIa are required for fIX binding, however, there is lack of consensus on their location(s), with sites on the A2, A3, and catalytic domains described. Replacing the A3 domain with the prekallikrein A3 domain increases Km for fIX cleavage after Arg145 and Arg180 25- and ≥90-fold, respectively, and markedly decreases kcat for cleavage after Arg180. Similar results were obtained with the isolated fXIa catalytic domain, or fXIa in the absence of Ca2+. Forms of fXIa lacking the A3 domain exhibit 15-fold lower catalytic efficiency for cleavage after Arg180 than for cleavage after Arg145, resulting in fIXα accumulation. Replacing the A2 domain does not affect fIX activation. The results demonstrate that fXIa activates fIX by an exosite- and Ca2+-mediated release-rebind mechanism in which efficiency of the second cleavage is enhanced by conformational changes resulting from the first cleavage. Initial binding of fIX and fIXα requires an exosite on the fXIa A3 domain, but not the A2 or catalytic domain.

Factor IX (fIX) 2 is the zymogen of a protease, factor IXa␤ (fIXa␤), which is required for proper formation and maintenance of blood clots at sites of vascular injury (1,2). Congenital fIX deficiency causes the severe bleeding disorder hemophilia B (1). Human fIX is a 57-kDa protein composed of an N-terminal calcium-binding Gla domain, two epidermal growth factor domains, an activation peptide, and a trypsin-like catalytic domain (1,2). It shares this structure with coagulation factors VII and X (3). The Gla domains of these proteins facilitate binding to phospholipid membranes.
Conversion of fIX to fIXa␤ requires cleavage of the Arg 145 -Ala 146 and Arg 180 -Val 181 peptide bonds to release the activation peptide (2,4,5). The physiologic mediators of this process are the serine proteases factor VIIa (fVIIa) and factor XIa (fXIa). FIX activation by fVIIa requires Ca 2ϩ and phospholipid (6 -8).
In the presence of the cofactor tissue factor, fVIIa cleaves fIX first after Arg 145 , forming the inactive intermediate fIX␣. FIX␣ is released from fVIIa and must rebind to the protease to be cleaved after Arg 180 to form fIXa␤. As the second cleavage is rate-limiting, fIX␣ accumulates during fIX activation by fVIIa.
FXIa appears to activate fIX by a different process than fVIIa. FXIa, is a dimer of identical subunits, each containing a trypsinlike catalytic domain and a heavy chain comprised of four apple domains (A1-A4) (9 -12). In contrast to fVIIa, fXIa lacks a Gla domain. Perhaps as a result, fIX activation by fXIa, although Ca 2ϩ -dependent, does not require phospholipid. During fIX activation by fXIa an intermediate does not accumulate (13,14). This is unrelated to the dimeric structure of the protease, as isolated fXIa subunits convert fIX to fIXa␤ without intermediate accumulation (14,15). FXIa readily cleaves fIX with an R180A substitution after Arg 145 , whereas fIX with an R145A substitution undergoes insignificant cleavage after Arg 180 (14). This indicates that the structure of fIX facilitates ordered bond cleavage, with the Arg 145 -Ala 146 bond presented first to the fXIa active site. Changes in conformation resulting from cleavage after Arg 145 may then facilitate cleavage after Arg 180 . A similar process was reported previously for prothrombin activation (16).
There are conflicting hypotheses regarding the manner in which fXIa is able to activate fIX without fIX␣ accumulation. It has been proposed that fIX␣ may not be released from fXIa prior to conversion to fIXa␤ (13,17). However, at least a portion of the intermediate must be released, as active site-blocked fXIa competes with fXIa for fIX␣ binding during fIX activation (14). Active site-blocked fIXa␤ is a competitive inhibitor of fIX cleavage by fXIa (18), and fIX␣ may have similar inhibitory properties. If fIX␣ is released from fXIa, the lack of fIX␣ accumulation implies that cleavage after Arg 145 is rate-limiting, although this has yet to be demonstrated.
There is compelling evidence that one or more exosites (substrate binding sites distinct from the protease active site) on fXIa are required for normal fIX activation (17)(18)(19)(20)(21)(22)(23), but there is a lack of agreement on their locations. It is clear that Ca 2ϩ -dependent fIX activation requires the fXIa heavy chain (19,20). One report indicated that the A2 domain contains a fIX binding site (21), whereas others point to A3 (22,23). More recent work suggests a distinct fIX binding site may be located on the fXIa catalytic domain (17). K m for fIX activation by the isolated fXIa catalytic domain (a species lacking the heavy chain) was reported to be similar to K m for fIX activation by full-length fXIa, implying the catalytic domain was largely responsible for recognition and specificity of fIX binding. These data are not compatible with those showing that A3 domain substitution results in a marked increase in K m for fIX activation (18,22,23).
We conducted a kinetic analysis of fIX and fIX␣ activation to fIXa␤ by fXIa and fXIa variants, to address the conflicting information regarding the mechanism of fIX activation and the locations of substrate binding exosites on fXIa. Binding studies were used to support the kinetic data, which show that fIX activation by fXIa is a largely sequential process in which fIX is converted to fIX␣ and then to fIXa␤, with a substantial increase in catalytic efficiency of the second cleavage explaining the absence of fIX␣ accumulation. The A3 domain serves as the major binding site for both fIX and fIX␣. The dramatic decrease in catalytic efficiency of forms of fXIa lacking the functional exosite illustrates the critical role of Ca 2ϩ -dependent fIX and fIX␣ binding to fXIa.
FXI chimeras in which apple domains are replaced with prekallikrein (PK) apple domains have been described (22,26). This study used fXI with the PK A2 or A3 domain (fXI/ PKA2 or fXI/PKA3) (22), and fXI in which the entire heavy chain is replaced with the PK heavy chain (FXI-CD/PK-HC) (26). In fXI/PKA2, Lys 140 was changed to Ser to prevent autoproteolysis.
FXI was converted to fXIa by incubation with fXIIa (100:1 substrate to enzyme molar ratio) at 37°C for 24 h in 50 mM Tris-HCl, pH 7.4, 100 mM NaCl (TBS). Conversion of the 80-kDa zymogen fXI subunits to the 45-kDa heavy chain and 35-kDa catalytic domain of fXIa was confirmed by SDS-PAGE. In studies of fIX activation by fXIa or fXIa variants, removal of fXIIa, or inhibition of fXIIa, had no effect on fIX or fIX␣ cleavage.
To prepare isolated fXIa heavy chain (fXIa-HC) and isolated fXIa catalytic domain (fXIa-CD) (18), fXI with Cys 362 and Cys 482 changed to Ser (fXI-Ser 362 /Ser 482 ) was activated and passed over a soybean trypsin inhibitor-agarose column. FXIa-CD binds to the column and is eluted with TBS containing 200 mM benzamidine, whereas fXIa-HC flows through the column.
Preparation of FIX␣-FIX (14.3 M) was incubated with 50 nM fXIa in TBS, pH 7.4, 20 mM EDTA for 12 min at 37°C. FIXa␤ was inhibited by incubating 90 min at 37°C with biotinylated EGR-CK (10-fold molar excess over fIX). After dialysis against PBS, strepavidin immobilized on agarose resin was added, and incubated for 1 h at room temperature with mixing. Resin was removed by centrifugation, and the supernatant containing fIX␣ was dialyzed against 50 mM HEPES, pH 7.4, 125 mM NaCl, 1 mg/ml of polyethylene glycol (PEG) 8000.
Hydrolysis of S-2366 by FXIa-FXIa (6 nM) was incubated with S-2366 (50 -2000 M) in TBS at room temperature. Free p-nitroaniline formation was followed by continuous monitoring of absorbance at 405 nm on a SpectraMax 340 plate reader (Molecular Devices, Sunnyvale, CA). Rates of p-nitroaniline generation (nM/s) were determined using an extinction coefficient of 9920 OD units (405 nm) per mol per cm of p-nitroaniline. K m and k cat for S-2366 cleavage were determined by nonlinear least squares fitting performed with MicroMath Scientific Software.
Activation of FIX and FIX␣ by FXIa-FIX or fIX␣ (25 nM to 5 M) in assay buffer (50 mM HEPES, pH 7.4, 125 mM NaCl, 5 mM CaCl 2 (or 25 mM EDTA), 1 mg/ml of PEG 8000) was incubated at room temperature with fXIa (1 to 240 nM active sites, depending on the fXIa species and substrate) in tubes coated with PEG 20,000. At various times (0 to 640 min), aliquots were removed into nonreducing SDS-sample buffer, size fractionated on 17% polyacrylamide-SDS gels, and stained with GelCode Blue (Pierce). Gels were imaged on an Odyssey Infrared Imaging System (LI-COR Biosciences, Lincoln, NE) using an excitation of 685 nm, and emission of 720 nm. Conversion of fIX to fIX␣ and fIXa␤, and fIX␣ to fIXa␤ was assessed by densitometry. To determine the amount of protein per band, standards were run on a separate gel, with one standard also run on the gel with the time course samples. Full progress curves were constructed from data for disappearance of fIX, and appearance of fIX␣ and fIXa␤. In some reactions, antibodies to fXIa or fIX were included.
Kinetic Analyses-Steady-state kinetic parameters for fIX or fIX␣ cleavage were obtained by numerical integration fitting of full progress curves of substrate depletion, and intermediate and product formation at substrate concentrations from 25 to 5000 nM; and by analysis of initial rate dependence of substrate depletion as a function of substrate concentration. Rates of cleavage of the Arg 145 -Ala 146 and Arg 180 -Val 181 bonds were analyzed simultaneously with KinTek Explorer Version 2.5 software (27) using the reaction mechanism shown in Equations 1 and 2. IX␣* indicates the intermediate bound to fXIa in the correct orientation for cleavage after Arg 180 .
We reported that the product fIXa␤ binds to fXIa and is a competitive inhibitor of fIX activation by fXIa (18). A similar assumption was made for product inhibition by fIX␣. K d and K i values were initially constrained to those obtained from surface plasmon resonance studies (below) and full progress curve analysis of fIXa␤ formation, and no rapid equilibrium assumptions were imposed on association and dissociation rates. By fitting full progress curves for fIX disappearance, and fIX␣ and fIXa␤ appearance, K m , k cat , K d , K i , and catalytic efficiency (k cat / K m ) were determined for both cleavages in fIX. The same method was used with fIX␣ as substrate. As fIX␣ preparations have ϳ20% fIX contamination, analyses of conversion of fIX␣ to fIXa␤ included the cleavage of the fIX. Results of numerical analysis were compared with those obtained by conventional Michaelis-Menten analysis of both cleavages. With fIX as substrate, initial velocities (v 0 ) of cleavage after Arg 145 were determined from the initial slopes of progress curves for disappearance of fIX, normalized to 1 nM fXIa active sites. Values for v 0 were analyzed with the Michaelis-Menten equation, and values for K m and k cat were obtained from direct nonlinear least squares analysis using Scientist Software. With fIX␣ as substrate, K m and k cat for cleavage after Arg 180 were determined in a similar manner. Competitive binding of fXIa to fIX in the fIX␣ preparation was calculated by a cubic equation, and was taken into account for determining the fXIa concentration available for the reaction with fIX␣.
Surface Plasmon Resonance-Binding studies were performed on a Biacore T100 flow biosensor (Biacore, Uppsala, Sweden) at 25°C. FIX, fIX␣, or fIXa␤ were immobilized on carboxymethyl-dextran flow cells (CM5 sensor chips, GE Healthcare) using amine-coupling chemistry. To prevent cleavage of bound fIX or fIX␣, fXIa active sites were blocked with FPR-CMK. FIXa␤ active sites were blocked with EGR-CK. Flow cell surfaces were activated with a mixture of 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide and N-hydroxysulfosuccinimide for 5 min (flow rate 10 l/min), after which the protein (30 g/ml in sodium acetate, pH 4.0) was injected onto the surface. Unreacted sites were blocked for 5 min with 1 M ethanolamine. Analytes (fXIa species, 1 to 5000 nM) were perfused through flow cells in HBS-P buffer (10 mM HEPES, pH 7.4, 150 mM NaCl, 2 mM CaCl 2 , 1 mM benzamidine, 0.005% (v/v) P20) at 10 l/min for 6 min. After changing to HBS-P buffer without fXIa, analyte dissociation was monitored for 10 min. Flow cells were regenerated with HBS-P containing 30 mM EDTA (no CaCl 2 ). FXIa binding to fIX species was also tested in the absence of Ca 2ϩ (10 mM EDTA). Data were corrected for nonspecific binding by subtracting signals obtained with analytes infused through a flow cell without coupled protein.
Binding was analyzed with BIAevaluation software (Biacore) using a bivalent binding model for dimers (fXIa-WT, fXIa/ PKA2, fXIa/PKA3) and a 1:1 binding model for monomers (fXIa-CD, fXIa-CD/PK-HC). K d values were calculated from the quotient of the derived dissociation (k d ) and association (k a ) rate constants. In addition, a steady state affinity model was used in which K d was derived from nonlinear regression fitting of the response at equilibrium to fXIa concentration.
FIX Activation in the Presence of FXIa-HC-FIX (100 nM) was incubated with fXIa-WT (1 nM) with or without fXIa-HC (1 M) in assay buffer at room temperature. At various times, aliquots were removed into nonreducing SDS sample buffer, size fractionated on 12% polyacrylamide-SDS gels, and transferred to a nitrocellulose membrane. Western blots were performed using an HRP-conjugated goat polyclonal anti-human fIX IgG and chemiluminescence for detection.

RESULTS
Recombinant FXIa-All fXI species migrate at ϳ160 kDa on nonreducing SDS-PAGE, except for fXI-CD/PK-HC, which is an 80-kDa monomer (Fig. 1A). FXI-Ser 362 /Ser 482 lacks the disulfide bond that links the A4 and catalytic domains after conversion to fXIa. Activation of this variant results in an ϳ90 kDa heavy chain dimer and 35-kDa catalytic domains that dissociate (Fig. 1B). Under reducing conditions the heavy chain dimer dissociates into 45-kDa monomers (Fig. 1B). Chroma- tography on soybean trypsin inhibitor-agarose separates the heavy chains and catalytic domains. All species of fXIa cleaved the tripeptide S-2366 (18 -20) with similar K m (190 -290 M) and k cat (140 -150 s Ϫ1 ), indicating domain manipulations did not significantly alter active site conformation.
FIX␣-FIX␣ was prepared from fIX by limited digestion with fXIa in the absence of Ca 2ϩ (Fig. 1C). A disulfide bond connects the polypeptides comprising fIX␣, which has essentially the same molecular mass as fIX. FIX␣ migrates slightly faster than fIX on nonreducing PAGE, likely due to conformational changes that accompany cleavage (Fig. 1C). FIX␣ preparations contain ϳ10 -20% fIX (Fig. 1C), and this is taken into account as an initial substrate in kinetic analyses.
Cleavage of FIX and FIX␣ by FXIa-WT-FIX activation can be followed by measuring cleavage of a chromogenic substrate by fIXa␤ (18); however, this method lacks sensitivity, making it difficult to measure initial activation rates. Sensitivity can be improved by using a coupled assay in which fIXa␤ converts fX to fXa, which is then detected with a chromogenic substrate (17,22). Chromogenic substrate-based techniques share a limitation. They are uninformative about fIX␣ generation. To address this, we monitored fIX activation by densitometry of Coomassie Blue-stained gels imaged at infrared wavelengths. FIX, fIX␣, and fIXa␤ migrate differently under nonreducing conditions on SDS-PAGE, and the low signal to noise ratio facilitated studying reactions with substrate concentrations down to 25 nM.
Initially, we assessed fIX conversion to fIXa␤ by fXIa-WT in the presence of Ca 2ϩ ions (Fig. 2, A-C). As expected, little fIX␣ accumulation was observed. Although fIX␣ is usually not detected when fIX is activated by plasma-derived fXIa, traces may be seen with recombinant fXIa. Results of the kinetic analysis by numerical integration and by the conventional Michaelis-Menten approach are summarized in Tables 1 and 2. The catalytic efficiency (k cat /K m ) is ϳ7-fold higher for cleavage after Arg 180 compared with Arg 145 (400 Ϯ 100 and 60 Ϯ 6 M Ϫ1 min Ϫ1 , respectively), with an estimated ϳ2.5-fold lower K m and ϳ3-fold higher k cat . The catalytic efficiency of cleavage after Arg 180 in purified fIX␣ (130 Ϯ 20 M Ϫ1 min Ϫ1 , Fig. 2, A, D, and E, and Tables 1 and 2) was ϳ3-fold lower than for fIX␣ formed in situ during fIX activation, mainly due to a lower k cat , suggesting conformational differences between the nascent intermediate and purified fIX␣ bound to fXIa. Dynamic simulation indicated that k cat for cleavage after Arg 145 is very sensitive to changes in v 0 for disappearance of fIX, and good agreement was found with the kinetic parameters determined by Michaelis-Menten analysis of fIX activation (Fig. 2C). Taken as a whole, the results are consistent with a mechanism in which Ca 2ϩ -dependent cleavage of Arg 145 -Ala 146 increases the efficiency of cleavage of Arg 180 -Val 181, without fIX␣ accumulation.
Cleavage of FIX and FIX␣ by FXIa Species Lacking the A2 or A3 Domains-Although the fXIa heavy chain clearly contains a substrate-binding exosite (18 -20), different studies indicate the site is on the A2 (21) or A3 (22,23) domain. A second fIX-binding exosite may reside on the catalytic domain (17). We used recombinant fXIa variants to identify the region of the molecule involved in high-affinity binding to fIX and fIX␣.
FXIa in which the A3 domain is replaced with the PK A3 domain (fXIa/PKA3), and the isolated fXIa catalytic domain (fXIa-CD) were tested for their ability to cleave fIX and fIX␣ ( Fig. 3 and Tables 1 and 2). The cleavage rates of the Arg 145 -Ala 146 and Arg 180 -Val 181 bonds were markedly reduced compared with fXIa-WT, with pronounced fIX␣ accumulation. K m values for fIX cleavage after Arg 145 by fXIa/PKA3 and fXIa-CD were increased ϳ25-fold compared with fXIa-WT, with 60and 90-fold decreases in catalytic efficiency, respectively. The catalytic efficiency of cleavage after Arg 180 was impaired to an even greater degree, but only k cat /K m could be estimated accurately because saturation was not achieved for the reactions (Fig. 3E). The values for K m in Tables 1 and 2 should, therefore, be considered lower limits for the actual K m . These data indicate that the A3 domain is important for binding of fIX and fIX␣ to fXIa, and that loss of the A3 binding site has a deleterious effect on both cleavages. FIX␣ accumulation is the result of an 11-14-fold higher catalytic efficiency for Arg 145 -Ala 146 cleavage compared with Arg 180 -Val 181 cleavage. Progress curve simulations for cleavage of isolated fIX␣ by fXIa/PKA3 and fXIa-CD suggested that Arg 180 -Val 181 is cleaved with similar, low catalytic efficiency when fIX or fIX␣ is the starting  NOVEMBER 2, 2012 • VOLUME 287 • NUMBER 45 substrate, indicating that the exosite is important for efficient binding of both fIX and fIX␣ as substrates. Cleavage of fIX or fIX␣ by fXIa in which the A2 domain was replaced with PK A2 (fXIa/PKA2) was similar to cleavage by fXIa-WT (Table 1) indicating that the A2 domain does not contain a high affinity fIX or fIX␣ binding site.

Factor IX Activation Mechanism
The fXIa catalytic domain is attached to the heavy chain by the Cys 362 -Cys 482 disulfide bond (11,12). The absence of the heavy chain or the substitution of Cys 482 with serine may have caused changes to fXIa-CD that reduced its ability to activate fIX, independent of the loss of the A3 exosite. FXIa-CD/PK-HC consists of the fXIa catalytic domain attached to the heavy chain of PK. Like the fXI heavy chain, the PK heavy chain contains four apple domains, and it is connected to the catalytic domain through a single disulfide bond. Activation of fIX and fIX␣ by fXIa-CD/PK-HC (Tables 1 and 2) was similar to activation by fXIa/PKA3 and fXIa-CD. Taken as a whole, the data indicate that a major exosite for binding of fIX and fIX␣ is located on the fXIa A3 domain. Loss of this exosite, either in chimeras (fXIa/PKA3 or fXIa-CD/PK-HC) or through absence of the heavy chain (fXIa-CD) results in loss of the high affinity site and a marked increase in K m for fIX activation.
The anti-human fXI monoclonal IgG O1A6 binds to the A3 domain (24). In the presence of O1A6, fIX activation by fXIa-WT appears similar to activation by fXIa/PKA3, with decreased rates of cleavage for both bonds and fIX␣ accumulation (compare Figs. 4, A and B, to 3B). An IgG that binds to the A2 domain (14E11; Ref. 24) had no effect on fIX activation by fXIa (not shown).
Cleavage of FIX and FIX␣ in the Absence of Calcium Ions-Although most reactions catalyzed by fXIa do not require Ca 2ϩ , fIX activation by fXIa is Ca 2ϩ -dependent. The fIX Gla domain is involved in the interaction with fXIa (25), and Ca 2ϩ is likely required for proper Gla domain conformation. In the absence of Ca 2ϩ , fIX bond cleavage is slow, with accumulation of fIX␣ (compare Fig. 4, D and E). K m values for fIX activation in the absence of Ca 2ϩ are similar to those with fXIa/PKA3 in the presence of Ca 2ϩ (Tables 1 and 2). Absence of Ca 2ϩ had little effect on the overall rate of fIX cleavage by fXIa/PKA3 or fXIa-CD (not shown). The data support the hypothesis that the a Due to the low affinity between the enzyme and substrate in certain reactions, we were not able to reach saturation for reactions with fXIa-CD, fXIa/PKA3, fXIa-CD/PK-HC, or fXIa-WT in the absence of Ca 2ϩ . Fits for k cat and K m are linked, with the listed K m values representing lower estimates for the reaction. k cat /K m is the appropriate parameter for comparing these reactions.

TABLE 2 Kinetic parameters for cleavage of fIX and fIX␣ by fXIa determined from initial velocities
The initial velocities (v 0 ) of cleavage after Arg 145 in fIX, and cleavage after Arg 180 in fIX␣ were plotted against initial substrate concentration, and analyzed using the Michaelis-Menten equation. K m and k cat were obtained from direct nonlinear least squares analysis using Scientist Software. Due to the low affinity between the enzyme and substrate, we were not able to reach saturation reactions with fXIa-CD, fXIa/PKA3, and fXIa-CD/PK-HC. Thus, values for K m are approximate and values for k cat were not determined.

Factor IX Activation Mechanism
Ca 2ϩ -dependent interaction between the fXIa A3 domain and fIX requires the fIX Gla domain. Consistent with this, fIX activation by fXIa in the presence of IgG SB 249417 (25), which binds to the fIX Gla domain, is characterized by significant accumulation of fIX␣ and slow generation of fIXa␤ (Fig. 4C).
Binding of FXIa to FIX and FIX␣-We studied binding of active site-blocked fXIa to immobilized fIX, fIX␣, and fIXa␤ using surface plasmon resonance. Zymogen fXI-WT did not bind to any species in the presence of Ca 2ϩ , nor did fXIa-WT in the presence of 10 mM EDTA (data not shown). In the presence of Ca 2ϩ , fXIa-WT bound to fIX, fIX␣, and fIXa␤. Binding at equilibrium was plotted as a function of the fXIa-WT concentration (Figs. 5, A-C). K d for binding to fIX (130 Ϯ 20 nM), fIX␣ (140 Ϯ 20 nM), and fIXa␤ (60 Ϯ 10 nM) were comparable (Table  3). Isolated fXIa heavy chain also bound to these species in a Ca 2ϩ -dependent manner (Fig. 5, D and E) with K d values of 90 Ϯ 20, 70 Ϯ 10, and 70 Ϯ 10 nM, respectively (Table 3). We were unable to demonstrate binding of fXIa-CD, fXIa/PKA3, or fXIa-CD/PK-HC ( Fig. 5F and Table 3) to fIX, fIX␣, or fIXa␤ at analyte concentrations up to 5 M. These results indicate that most of the binding energy for the fIX-fXIa interaction requires the fXIa A3 domain. The Western blots in Fig. 5G show that FXIa-HC inhibits fIX conversion to fIXa␤ by fXIa, indicating the A3 exosite engages fIX in the absence of the catalytic domain, and accumulation of FIX␣ in the presence of fXIa-HC supports the premise that fIX␣ is released from fXIa.

DISCUSSION
The current study was undertaken to establish the mechanism by which fXIa converts fIX to fIXa␤. The data need to be interpreted in light of published work in the field, some of which is in disagreement with the results presented here. It is clear that fIX activation by fVIIa or fXIa involves sequential proteolysis of the Arg 145 -Ala 146 and Arg 180 -Val 181 peptide bonds (2,5). The observation that fIX containing an R145A substitution is cleaved poorly by fXIa (14) supports the premise that the Arg 145 -Ala 146 cleavage site is presented to the fXIa active site after binding of fIX, and that conformational changes resulting from cleavage after Arg 145 , and perhaps repositioning of fIX␣ on the protease, enhances cleavage after Arg 180 .
Although fIX␣ forms during fIX activation by either fVIIa or fXIa, it does not accumulate during the latter reaction (13,14). The fact that fXIa is a dimer initially suggested that the two subunits may cleave the two fIX bonds prior to releasing fIXa␤; however, subsequent work demonstrated that the ability to cleave fIX without fIX␣ accumulation is intrinsic to each fXIa subunit (14,15). Wolberg et al. (13) observed that cleavage rates of Arg 145 -Ala 146 and Arg 180 -Val 181 in the intermediates fIXa␣ (fIX cleaved after Arg 180 ) and fIX␣ (cleaved after Arg 145 ) are similar and comparable with the overall rate for conversion of fIX to fIXa␤. They suggested that any intermediate formed may not be released prior to conversion to fIXa␤. However, significant fIX␣ accumulation was subsequently observed in competitive reactions in which fIX is activated by fXIa in the presence of active site-inhibited fXIa (14). Some portion of fIX␣ must be released from fXIa, arguing against a mechanism based strictly on the intermediate remaining bound to fXIa. There are parallels between fIX activation and prothrombin activation, in this regard. During ordered bond cleavage of prothrombin by fXa, the 150-fold higher catalytic efficiency of the second cleavage relative to the first results in formation of an intermediate (meizothrombin) that does not accumulate, but that can be captured with a tripeptide chloromethylketone (28).
The results reported here support an exosite-and Ca 2ϩ -mediated release-rebind mechanism for fIX activation by fXIa (summarized in Fig. 6), in which the efficiency of the second cleavage is enhanced by conformational changes resulting from the first cleavage. We propose that binding of fIX to the fXIa A3 domain is followed by docking with the active site and cleavage after Arg 145 to form fIX␣. This is followed by a second docking step with the active site after rebinding of fIX␣ to A3, and cleavage after Arg 180 . The ratio of the catalytic efficiencies of the two cleavages determine whether fIX␣ accumulates. If the A3 domain exosite is available, catalytic efficiency for cleavage after Arg 180 is 7-fold greater than for cleavage after Arg 145 , and fIX␣ does not accumulate. In the absence of a functional exo- site, the rates of both bond cleavages are markedly decreased, but the predominance of the catalytic efficiency of the second cleavage is lost, leading to fIX␣ accumulation.
The observations that fIX␣ is captured by active site-inhibited fXIa (14) or fXIa-HC in competition assays, and that a small amount of fIX␣ is transiently observed on gels during fIX cleav-age by fXIa-WT, support a mechanism involving fIX␣ release. However, the data do not exclude the possibility that release is partial, with a fraction of fIX␣ repositioned for cleavage after Arg 180 , while still bound to fXIa (a transition from fIX␣⅐XIa in Equation 1 directly to fIX␣*⅐fXIa in Equation 2). Such a mechanism has been termed "processive" in prior studies (13). Using  our data, we simulated a model that includes conformational repositioning of fIX␣⅐fXIa to the productive complex fIX␣*⅐fXIa without release (represented by dashed arrows in Fig. 6), in parallel with release rebinding. Because the true ratio of fIX␣⅐fXIa to fIX␣*⅐fXIa is unknown, we examined a range of values for the ratio (0.05-1), and found that the model fits data sets for cleavage after Arg 180 for both nascent and purified fIX␣ by fXIa-WT reasonably well at a ratio of 0.5, a K d of ϳ100 nM for binding of fIX␣ to fXIa in a productive complex, and a k cat of ϳ20 min Ϫ1 . Independently determined parameters for cleavage of the first bond, obtained by Michaelis-Menten analysis, and K i values for surface plasmon resonance binding of fIX, fIX␣, and IXa␤ to active site-blocked fXIa-WT were kept constant. The results suggest a possible upper limit of ϳ40% (the difference between k cat of 35 min Ϫ1 and 20 min Ϫ1 ) of fIX␣ proceeding to fIXa␤ through repositioning without release-rebinding. If the mechanism only involved repositioning of fIX␣ without release, no transient intermediate should be observed in time courses with fXIa-WT, no fIX␣ would competitively bind to active site-blocked fXIa, and k cat for the second cleavage would have to be exceedingly large (up to 60 to 150 min Ϫ1 ). Our data indicate this latter scenario is not likely.
The 3-fold greater efficiency of cleavage of fIX␣ generated in situ compared with purified fIX␣ recruited from the aqueous phase in the release-rebind model deserves comment. Although it must be recognized that this result may simply reflect structural perturbations in fIX␣ acquired during purification, the difference in catalytic efficiency disappears in reactions with exosite-impaired proteases, perhaps weakening this argument somewhat. The difference in catalytic efficiency may actually reflect some degree of fIX␣ repositioning without release, or perhaps induced fit related to conformational changes in fXIa, fIX␣, or both triggered by fIX cleavage after Arg 145 .
Coagulation proteases are trypsin-like enzymes, but with more restricted substrate specificity than trypsin. Binding interactions at exosites on the proteases distinct from the active site are key to the mechanisms of action of these enzymes, and have been shown to be primary determinants of substrate affin- Using surface plasmon resonance, fXIa species (1 to 5000 nM) in Ca 2ϩ -containing HBS-P buffer were perfused across sensor chips coated with fIX, fIX␣, or fIXa␤. HBS-P buffer without fXIa was then perfused for 10 min to follow dissociation. Data were corrected for nonspecific binding by subtracting signals obtained with analytes infused through a flow cell without coupled protein. Binding was analyzed using a bivalent binding model for all species except fXIa-CD/PK-HC and fXIa-CD, which were evaluated with a 1:1 binding model. K d values were calculated from the quotient of the derived dissociation (k d ) and association (k a ) rate constants.  ity and specificity during activation of prothrombin (29 -33) and factor X (34). Binding of the substrate at the exosite precedes substrate docking with the active site and catalysis (31,33). In the current study, high affinity binding between fIX or fIX␣ and fXIa was largely, if not completely, due to a Ca 2ϩ -dependent interaction with the A3 domain. FXIa/PKA3 and the isolated fXIa protease domain (fXIa-CD) did not engage fIX or fIX␣ with high affinity, resulting in Ͼ25-fold increases in K m for cleavage after both Arg 145 and Arg 180 . These results are not consistent with a model based on the fXIa catalytic domain containing a high affinity fIX binding exosite. Sinha et al. (17) reported that the slow rate of fIX conversion to fIXa␤ by an isolated fXIa protease domain was due to reduced k cat , with K m for the overall reaction similar to that for fXIa-WT. It was concluded that fIX, but not fIX␣, bound to an exosite on the catalytic domain in a Ca 2ϩ -independent manner. A mechanism was proposed that required fIX to bind to both the catalytic domain and heavy chain exosites, although K m data imply that the putative catalytic domain site would dominate the binding interaction. Our results do not support this model. Although we noted that fXIa-CD activates fIX similarly in the presence or absence of Ca 2ϩ , the kinetic and binding data fail to support the presence of a fIX binding site on fXIa-CD. Furthermore, our results for fIX activation by fXIa-WT and direct binding studies do not indicate that fIX binds with high affinity to fXIa-WT in the absence of Ca 2ϩ . The reason for the differences between our results and those of Sinha et al. (17) are not clear; however, their study used a chromogenic assay to examine fIX activation, which would not facilitate examination of the initial cleavage converting fIX to fIX␣ with the detail we were able to achieve with our densitometry-based approach.

Analyte
The same group posited that catalytic domain amino acids Glu 458 and Lys 550 (Glu 98 and Lys 192 in chymotrypsin numbering) may be part of a fIX binding exosite (35). However, substitutions for these residues resulted in Ͼ100-fold reductions in k cat for fIX conversion to fIXa␤, more consistent with a catalytic defect. Preliminary work from our group indicated that fXIa with a Lys 550 substitution activates fIX with normal K m , with the reduction in k cat likely due to disruption of the interaction between the Lys 550 side chain and the substrate P3Ј residue (36). Recently, Marcinkiewicz et al. (37) reported that cross-talk between the fXIa catalytic domain and heavy chain is required for expression of the fIX binding site on the heavy chain, based on observing that isolated fXIa heavy chain (fXIa-HC) does not bind to fIX. In the current study, in contrast, fXIa-HC and fXIa-WT bound to fIX and fIX␣ with similar affinity in the presence of Ca 2ϩ . Furthermore, fXIa-HC inhibited fIX activation by fXIa. These findings are most consistent with a mechanism in which fIX binds initially to the fXIa heavy chain, with binding being independent of a contribution from the catalytic domain.
Although our data do not support the presence of a fIXbinding exosite on the fXIa catalytic domain that is required for initial substrate recognition, they do not rule out the possibility that such a binding site is expressed as the result of initial binding of fIX to the A3 exosite (although such a process would not explain the differences between our results and published work). Previously, we noted that binding of fIX to fXIa results in a mixed-type inhibition of cleavage of a tripeptide substrate by fXIa, consistent with fIX binding to A3 causing changes in the protease domain, in addition to competing with the tripeptide at the active site (18).
The Gla domains of the coagulation proteases fVIIa, fIXa, and factor Xa contain ␥-carboxylated glutamic acid residues that bind Ca 2ϩ and facilitate binding to phosphatidylserinerich membranes. K m for reactions involving these proteases are strongly influenced by the Gla domain-phospholipid interaction. FXIa lacks a Gla domain (10 -12), and fXIa activation of fIX is not influenced appreciably by phospholipid. However, the fIX Gla domain is required for binding to fXIa (25), and may be the structure that interacts with the A3 domain. Supporting this is the observation that the kinetic parameters of fIX activation by fXIa-WT in the absence of Ca 2ϩ are similar to those for activation by fXIa/PKA3 ( Table 1), and that an antibody to the fIX Gla domain results in accumulation of fIX␣ (Fig. 4C). It is possible, therefore, that during fIX activation by fXIa, the A3 domain performs the Gla domain binding role that phospholipid performs during fIX activation by fVIIa.
FXI is a relatively new component of the blood coagulation mechanism, making its appearance during mammalian evolution as the result of a duplication of the PK gene (38). Human PK and fXI share a high degree of structural homology, and are 58% identical in amino acid sequence (10). However, the sequence around the putative fIX binding site in the fXI A3 domain, which is highly conserved in mammals, is distinctly different from the corresponding sequence in PK (38), supporting the conclusion that changes to the A3 domain were critical to the ability of fXIa to engage fIX as a substrate.