Beating tissue factor at its own game: Design and properties of a soluble tissue factor–independent coagulation factor VIIa

Two decades of research have uncovered the mechanism by which the complex of tissue factor (TF) and the plasma serine protease factor VIIa (FVIIa) mediates the initiation of blood coagulation. Membrane-anchored TF directly interacts with substrates and induces allosteric effects in the protease domain of FVIIa. These properties are also recapitulated by the soluble ectodomain of TF (sTF). At least two interdependent allosteric activation pathways originate at the FVIIa:sTF interface are proposed to enhance FVIIa activity upon sTF binding. Here, we sought to engineer an sTF-independent FVIIa variant by stabilizing both proposed pathways, with one pathway terminating at segment 215–217 in the activation domain and the other pathway terminating at the N terminus insertion site. To stabilize segment 215–217, we replaced the flexible 170 loop of FVIIa with the more rigid 170 loop from trypsin and combined it with an L163V substitution (FVIIa-VYT). The FVIIa-VYT variant exhibited 60-fold higher amidolytic activity than FVIIa, and displayed similar FX activation and antithrombin inhibition kinetics to the FVIIa.sTF complex. The sTF-independent activity of FVIIa-VYT was partly mediated by an increase in the N terminus insertion and, as shown by X-ray crystallography, partly by Tyr-172 inserting into a cavity in the activation domain stabilizing the S1 substrate-binding pocket. The combination with L163V likely drove additional changes in a delicate hydrogen-bonding network that further stabilized S1–S3 sites. In summary, we report the first FVIIa variant that is catalytically independent of sTF and provide evidence supporting the existence of two TF-mediated allosteric activation pathways.

Serine proteases play a vital role in diverse physiological processes, such as development, digestion, coagulation, inflammation, and immunity (1). The coagulation factor VIIa (FVIIa), 5 a serine protease involved in blood coagulation, and its inactive precursor, zymogen factor VII (FVII), circulate in the bloodstream in a 1:100 ratio (2). Upon vascular injury, extravascular cells carrying the membrane protein tissue factor (TF) (3), are exposed to the blood stream, where they bind to FVIIa and FVII. The complex formation between TF and FVIIa enhances the catalytic activity of the latter ϳ100,000-fold on the membrane surface, thereby triggering the onset of the coagulation cascade (4 -6). The FVIIa:TF complex is a potent activator of zymogen coagulation factors IX (FIX) and X (FX), providing an initial burst of FIXa and FXa (7). FXa in turn facilitates initial thrombin generation, which leads to activation of cofactors and platelets. The activated platelets serve to generate a massive thrombin burst capable of cleaving sufficient fibrinogen to fibrin to yield a stable clot (8).
FVII is activated by cleavage of the peptide bond between residues Arg-15 and Ile-16 (chymotrypsin numbering). The newly formed N terminus (Ile-16), unlike prototypical trypsin, fails to insert into the catalytic domain of FVIIa-an essential conformational rearrangement necessary to stabilize the FVIIa active site (9). Therefore, the majority of FVIIa exists in a "zymogen-like" state due to an immature active site owing to (a) a disordered activation domain (residues 144 -152, 183-193, and 215-223) and (b) a poorly defined oxyanion hole (10). Binding to membrane-anchored cofactor TF results in allosteric enhancement of FVIIa catalytic activity by stabilizing the activation domain (5), facilitating an optimal localization of the protease domain above the cell membrane (11)(12)(13) and providing exosites for interaction with substrates FIX/FX (14). The allosteric stabilization of the FVIIa activation domain is fully achieved by the soluble version of TF (sTF). As illustrated in Fig.  1B, two principal allosteric pathways have been implicated in the sTF-mediated allosteric stabilization of FVIIa, both starting from Met-164 in FVIIa, located at the FVIIa:sTF interface (15,16) to the FVIIa active site (17). Recent findings have provided strong evidence for the first pathway and support a tethering of the 170 loop to the protease domain and an accompanying stabilization of segment 215-217 involved in substrate binding in the S2-S3 pockets (18 -21). The second pathway has been shown to facilitate N terminus insertion through stabilization of activation loop 2 (residues 183-193) and 3 (residues 215-223), enabling binding of substrate to the S1 pocket and formation of the oxyanion hole (10,22,23).
The clinical relevance of FVIIa has prompted several variant generation approaches aiming to enhance the activity of FVIIa in the absence of TF and enabling the development of more potent pharmaceutical compounds (24 -27). Two main groups of FVIIa variants have been described, either manipulating pathway I (28 -31) or increasing the N terminus insertion to stabilize pathway II (32)(33)(34)(35). A combination of these two approaches has generated highly active FVIIa variants, but none have reached the activity level of the FVIIa:sTF complex (32). We aimed to engineer an sTF-independent FVIIa variant by stabilizing both proposed pathways. We recently provided structural evidence for the molecular mechanism underlying the enhanced catalytic efficiency of a FVIIa variant with the 170 loop replaced by that of trypsin (21,30). The grafted loop provided significant stabilization of the activation domain, mainly by insertion of residue Tyr-172 into a cavity of the FVIIa protease domain promoting pathway I (Fig. 1, A-C). This orientation and interaction are highly similar to that observed in constitutively active trypsin, and we hypothesized that this variant could serve as a starting point for generating a fully matured FVIIa variant independent of TF-induced allosteric enhancement. We therefore combined this variant with previously described point mutations shown to increase activation domain stability and N terminus insertion (pathway II) (Fig. 1D). Based on a combined structural, functional, and biophysical evaluation, we show that a FVIIa variant with the 170 loop from trypsin and the L163V point mutation has obtained full catalytic activity and is functionally independent of sTF in several aspects.

Functional characterization reveals an sTF-independent variant
Because FVIIa variants incorporating the 170 loop from trypsin have previously shown compromised sTF affinity, we evaluated sTF binding to all FVIIa variants using both a functional activity-based assay and SPR steady-state kinetics to ensure saturation with sTF in relevant experiments ( Fig. 2A). Values obtained for FVIIa-WT and previously reported variants were in agreement with published results (21,28,33). The replacement of the FVIIa 170 loop with that from trypsin resulted in a significant decrease in affinity (Y T , K d ϭ 141 Ϯ 8 nM), with no observable alteration by the addition of the L163V point mutation (VY T , K d ϭ 141 Ϯ 7 nM).
To evaluate the functional activity of the generated FVIIa variants, we utilized three different assays: (see Figs. S1-S5 for raw data) (a) an amidolytic activity approach using a small, synthetic substrate (peptidomimetic S-2288) to probe the mat-uration of the S1-S3 pockets and the active site in general; (b) an assay based on inhibition of substrate conversion by a small molecule inhibitor (p-aminobenzamidine (pABA)) to estimate S1 maturity specifically; and (c) an amidolytic activity-based assay as a readout for the extent of N terminus insertion, by measuring the loss of activity as a carbamate ion (KNCO 3 ) reacts with the solvent-exposed Ile-16 primary amino group that in turn prevents N-terminal insertion and active site stabilization. FVIIa-WT, with a significantly exposed N terminus, is highly labile in this assay, whereas the FVIIa-WT:sTF complex shows almost no sensitivity, due to the sTF-mediated activity enhancement of FVIIa-WT, which drives stable N terminus insertion (Table 1 and Fig. 2 (B-D)). From these data, we observed an additive effect of the two mutational strategies in the QY T variant (S-2288 -based k cat /K m of 16.7 Ϯ 0.4 s Ϫ1 mM Ϫ1 ), resulting in a ϳ18-fold increase in catalytic efficiency when compared with that of FVIIa-WT (k cat /K m of 0.9 Ϯ 0.02 s Ϫ1 mM Ϫ1 ). This increase in amidolytic activity was accompanied by a correspondingly tighter inhibition of FVIIa-QY T by pABA (K i of 38.4 Ϯ 2.3 M) compared with FVIIa-WT (K i of 1485 Ϯ 88 M), suggesting a significantly matured S1 pocket. These effects are likely to be mediated by Tyr-172 inserting into the cavity between activation loop (AL)2 and AL3, stabilizing the S1-S3 sites. Saturating levels of sTF potentiated the variant to an activity level 1.5-fold higher than the FVIIa-WT:sTF complex and resulted in ϳ3.0-fold reduction in K i . Interestingly, sTF fails to promote full N terminus insertion in QY T (t1 ⁄ 2 of 124 Ϯ 5.6 min), even with the 2.5-fold increase observed when comparing the single Q variant (t1 ⁄ 2 of 1364 Ϯ 270 min) to FVIIa-WT (t1 ⁄ 2 of 536 Ϯ 40 min) in the presence of sTF.
The combination of the trypsin 170 loop and the L163V mutation (FVIIa-VY T ) resulted in a remarkable synergistic effect as catalytic efficiency (k cat /K m of 52.9 Ϯ 2.0 s Ϫ1 mM Ϫ1 ) was increased ϳ60-fold compared with that of the FVIIa-WT, and 1.8-fold over the FVIIa-WT:sTF complex. The increased activity was mediated by a 12-fold decreased K m and 5-fold higher k cat . The improved K m value was also reflected in a tremendous increase in pABA inhibition (100-fold) compared with that of FVIIa-WT, reaching a 3-fold higher inhibition than the FVIIa-WT:sTF complex. The stronger binding is likely to arise from the stabilizing effects of Tyr-172, working in unison with the introduced valine to facilitate a ϳ3-fold lower N terminus exposure level, as this effect was not observed in either of the variants containing the individual changes. No effects were observed on any of these functional parameters by adding saturating levels of sTF, confirming that we indeed developed FVIIa-VY T into a sTF-independent FVIIa variant with a fully matured active site and substrate-binding machinery.

sTF independence is observed with a physiologically relevant substrate and inhibitor
To investigate whether the generated FVIIa variant was indeed sTF-independent, we relied on the natural substrate FX, where the activation peptide docks into both prime and nonprime sites during activation (36) (Fig. 3 and Table 1). FXa generation in the absence of sTF revealed that the QY T variant had a catalytic efficiency (k cat /K m ϳ1373 Ϯ 12 M Ϫ1 s Ϫ1 ), equivalent to that of the FVIIa-WT:sTF complex (k cat /K m , 1307 Ϯ 10 To further examine the sTF-independent nature of the FVIIa-VY T variant, we evaluated the ability of the plasma serine protease inhibitor (serpin) antithrombin (AT) to irreversibly bind through a reactive center loop that engages extensively with the active site (37). During the assay, we became aware that the AT inhibition rate of the designed variants in the presence of sTF was too rapid for proper measurement, and the variants alone were therefore compared with FVIIa-WT in the absence

Exosite inhibitor binding supports a fully matured FVIIa variant
To further evaluate the sTF-independent variants, we employed two peptide-based exosite inhibitors, E-76 (38) and A-183 (39), shown to bind at exosites I and II, respectively, located distant from the sites of mutagenesis (Fig. 4, A and B). As both inhibitors mainly work through a noncompetitive allosteric mechanism the inhibitor binding (K D ) was assessed separately from actual functional inhibition, to separate out these two components. The functional inhibition of E-76 and A-183 is believed to be mainly driven by stabilization of the zymogen-like state of FVIIa, in agreement with the reduced effect observed with the FVIIa:sTF complex, thus being a surrogate measure of sTF independence (Fig. 4, C and D). Binding kinetics of the interactions between the inhibitors and FVIIa variants were measured by SPR. There was little effect on affinity, with K D values of ϳ6 -10 nM for E-76 and ϳ4 -11 nM for A-183 (Table 2), in agreement with the location of exosites I and II. We evaluated the functional inhibition at saturating levels of E-76 (600 nM) and found a small effect of the L163V mutation alone (44.1 Ϯ 4.4% residual activity) when compared with FVIIa-WT (32.7 Ϯ 0.85%), in agreement with this modification conferring some level of increased activity. A synergistic effect was again observed when combining the L163V mutation with the 170 loop of trypsin, resulting in an activity level of FVIIa-VY T (52.4 Ϯ 2.5%) similar to the FVIIa:sTF complex (50.3 Ϯ 3.9%). In further support of the sTF-independent nature of FVIIa-VY T , the inhibitory effect of saturation levels of A-183 (600 nM) was completely abolished (116.7 Ϯ 6.3%). Interestingly, the function of this inhibitor was significantly affected by Y T alone (residual activity 76.8 Ϯ 9.4%). The addition of the M156Q mutation (QY T ) also completely eliminated inhibitor function (101.2 Ϯ 7.2%), even surpassing FVIIa:sTF (76.8 Ϯ 9.4%).

Thermal stability studies and ITC binding experiments reveal underlying thermodynamics
From the functional studies and inhibitor-binding experiments, it seemed likely that the activation domain of the designed FVIIa variants was stabilized in a productive conformation. To evaluate whether this translated into increased thermal stability, we conducted thermal unfolding experiments with all variants (Fig. 5A). The transition point (T m ) was determined using first-derivative analysis, selecting the highest tem-

Figure 2. Functional characterization of FVIIa variants.
A, binding affinities for sTF measured using either a functional amidolytic activity-based assay or SPR. B, amidolytic activity (using S-2288) shown as k cat /K m . C, inhibition constants (K i ) for a small molecule S1 pocket inhibitor (pABA) measured in the presence of 1 mM S-2288. D, the effects of KNCO on FVIIa variant activity on 1 mM S-2288 reported as activity half-life (t1 ⁄2 ). All functional experiments were conducted at 25°C in the presence or absence of 3 M sTF, with data shown as mean with range bars (n ϭ 2 on the same day). The SPR data were also collected at 25°C and shown as the mean Ϯ S.D. (n ϭ 3, three separate runs over 2 days). All data were collected in conjunction with FVIIa-WT and FVIIa-Y T , for which the data were previously published (21).

sTF-independent FVIIa
perature transition to correspond to the FVIIa protease domain (40). The data demonstrated that the FVIIa-Y T protease domain had a higher stability (T m of 62.5 Ϯ 0.3°C) when compared with that of FVIIa-WT (59.6 Ϯ 0.1°C). By including a previously published variant (FVIIa-S T ) with Tyr-172 replaced by a serine, we were able to show that the increase in thermal stability was provided by the tyrosine side chain (T m of FVIIa-S T 60.0 Ϯ 0.1°C). The M156Q point mutation provided a small stabilization of the protease domain (60.6 Ϯ 0.3°C). The incorporation of either the L163V or M156Q mutations into FVIIa-Y T , however, did not provide further thermal stability of the protease domain, suggesting that the increased activity may come from more subtle changes in the activation domain.
To probe for more subtle changes in the activation domain, we employed ITC to determine the thermodynamics underlying sTF binding to FVIIa-WT, -Y T , and -VY T (Fig. 5B). Data from sTF binding to FVIIa-WT corresponded well to previously published values (41), showing a large enthalpic contribution (⌬H ϭ Ϫ32.8 Ϯ 0.8 kcal/mol), matched by a counteracting large entropic penalty (ϪT⌬S ϭ 22.2 Ϯ 1.1 kcal/mol) resulting in a K d of 14 Ϯ 8 nM. Values obtained for the Y T variant showed a markedly different thermodynamic landscape for sTF binding, with reductions in both enthalpy (⌬H ϭ Ϫ16.7 Ϯ 0.7 kcal/ mol) and entropy penalty (ϪT⌬S ϭ 8.4 Ϯ 0.7 kcal/mol). These effects may stem from the variant being prestabilized prior to sTF binding, as suggested from the thermal unfolding, resulting in less protease domain rearrangement during complex formation. This also seems to be true for FVIIa-VY T , where a further reduction in both enthalpy (⌬H ϭ Ϫ12.6 Ϯ 0.4 kcal/mol) and entropy penalty (ϪT⌬S ϭ 4.2 Ϯ 0.7 kcal/mol) is observed, in agreement with the decreased N-terminal exposure prior to sTF binding.

The crystal structure of FVIIa-VY T
To shed light on the structural features underlying the fully matured active site of FVIIa-VY T , we determined the crystal structure of this variant in the presence of an active site inhibitor and sTF (FVIIa-VY T -FFR:sTF) to a resolution of 1.25 Å ( Table 3). The structure was highly similar to the FVIIa-WT-FFR:sTF complex (42) with changes restricted to the 170 loop and AL2 and -3 (Fig. 6, A and B). As observed from a previous structure of FVIIa-Y T (21), Tyr-172 inserts into a cavity between AL2 and -3, forming hydrogen bonds with the backbone carbonyl of Gln-217 and Phe-225. The additional point mutation of L163V is clearly resolved in the high-resolution map, and the mutation leads to a 1.0-Å move of Phe-225 toward Val-163. This subtle change, combined with the insertion of Tyr-172, results in a rearrangement of the water network stabilizing AL2 and -3 (Fig. 6B). A more detailed look at these changes (Fig. 6, C-E) reveals a more extensive hydrogen bond network in VY T compared with WT, due to the preservation of all six water molecules and the additional bond/acceptor hydroxyl group of Tyr-172. This effect is not observed in Y T , where the presence of Tyr-172 displaces two of the water molecules (numbers 1 and 3) in the network. The extended network observed for VY T is likely to stabilize both interactions with the P1 arginine and may further stabilize the 215-217 segments vital for interaction with P2-P3 of the substrate. Minimal differences were seen in crystal contacts and B-factors for the two structures, as shown in Fig. S6.

Table 2 SPR data of exosite inhibitor binding to FVIIa variants
On-rate (k on ), off-rate (k off ) and equilibrium binding constant (K D ) were calculated from SPR data of E-76 and A-183 binding to FVIIa variants, fitted to a 1:1 Langmuir binding model. Results are shown as mean Ϯ S.D. of n ϭ 3.

Discussion
The allosteric signal that arises from sTF (or TF) binding toFVIIa leads to a 30 -50-fold enhanced FVIIa catalytic activity on small synthetic substratesand has been studied for several decades (15,17). These studies have implicated two interdependent pathways traveling more than 25 Å from the FVIIa:sTF interface to the active site of FVIIa. Mutational approaches to manipulate these pathways have yielded FVIIa variants with significantly increased activity, albeit never reaching the ultimate goal of an activity comparable with that of the FVIIa:sTF complex (29,32,43). Studies using hydrogen-deuterium exchange MS (18) as well as all-atom molecular dynamics simulation (20) have revealed the detailed structural underpinnings of this remarkable allosteric communication between FVIIa and its cofactor sTF. We benefitted from this increased understanding and designed two new FVIIa variants using the grafted 170 loop from trypsin (Y T ) as a starting template. Both variants showed increased N terminus insertion and a concurrent significant increase in amidolytic activity in the absence of sTF. One variant (FVIIa-QY T ) showed a simple additive effect of combining the Y T loop graft with an M156Q mutation, reaching 60% of the FVIIa-WT:sTF complex amidolytic activity. The second variant, FVIIa-VY T , showed a synergistic effect of combining the trypsin 170 loop with a single L163V mutation, resulting in 180% activity compared with FVIIa-WT:sTF and an almost 60-fold increase in activity when compared with the FVIIa-WT molecule. The increased activity seemed to be mediated without the participation of N terminus insertion when compared with levels measured for FVIIa-WT:sTF as judged by the carbamylation assay (9). It is likely that FVIIa-VY T does not support AL1 stabilization, and hence the N terminus may be less protected from carbamylation than in FVIIa-WT:sTF, while still allowing full functional activity. The obtained enzymatic activity levels and S1 pocket maturity suggested that we had succeeded in generating a fully sTF-independent variant.
We have examined the susceptibilities to inhibition by the serpin AT and the abilities to activate the physiological substrate FX. From the collected data, it was evident that the active site of FVIIa-VY T was fully matured, as the variant alone was inhibited 5 times faster (639 Ϯ 43 ϫ 10 4 s Ϫ1 ) than FVIIa bound to sTF (130 Ϯ 20 ϫ 10 4 s Ϫ1 ). To our surprise, FVIIa-VY T achieved an FX activation level without sTF (k cat /K m , 1895 Ϯ 29 M Ϫ1 s Ϫ1 ) that was higher than that of the FVIIa-WT:sTF complex (1307 Ϯ 10 M Ϫ1 s Ϫ1 ). The second combination variant, FVIIa-QY T , achieved a similar FX activation level (1376 Ϯ 12 M Ϫ1 s Ϫ1 ), suggesting that the exosites provided by soluble tissue factor are negligible in the absence of a membrane surface and are likely to be more pronounced with full-length, lipid-embedded TF (14,44). Published studies (32,45,46) indicate that the FVIIa activation domain loops are important for the direct interaction with the FX protease domain, enabling a productive engagement of the FX activation peptide. The relatively efficient pABA and AT inhibition observed for both FVIIa-VY T and -QY T variants suggest that AL2 and -3 are fully stabilized, thus maturing the S1 pocket and providing the exosites needed for productive interaction between FVIIa and FX in the absence of sTF. This would in turn suggest that the FX light-chain interactions with sTF contribute little to the catalytic activity and are largely dominated by interactions in the protease domain, contrary to previous findings (47,48).
We further explored the FVIIa variants by examining their susceptibility to peptide-based exosite inhibitors (38,39) known to allosterically suppress the functional activity of FVIIa. SPR analysis revealed that the affinity for neither E-76 nor A-183 was affected for the generated variants, confirming that the inhibitor-binding sites were intact. Interestingly, the functional inhibition was markedly reduced for several of the FVIIa variants. E-76 binds to FVIIa exosite I, and we found that the FVIIa-VY T variant was protected from this inhibitor to the same extent as the FVIIa-WT:sTF complex. Even more remarkable was the abolishment of inhibitor influence of A-183 on FVIIa-VY T . This inhibitor binds to exosite-II (49) and significantly reduces FVIIa-WT:sTF activity. This shows that the mutational strategy has resulted in almost complete stabilization of the FVIIa-VY T activation domain, rendering the inhibitors incapable of inducing a nonproductive conformation in the FVIIa active-site region.
In support of the inhibitor-binding data, we found a small increase in thermal stability of FVIIa-VY T . We dissected this observation further by collecting data for FVIIa-Y T and a variant with residue Tyr-172 replaced with serine (21). It was evident that the increased transition temperature was conferred solely by Tyr-172 in both FVIIa-VY T and -Y T . This suggested that the large increase in activity for FVIIa-VY T compared with -Y T is likely to be a more subtle change in activation domain mobility. We investigated this in more detail using ITC to monitor binding of sTF to FVIIa variants. ITC data indicated a 2-fold

sTF-independent FVIIa
reduction in enthalpy and 2.8-fold decrease in entropy when comparing sTF binding with FVIIa-WT and -Y T variant. If the functional and thermal stability data are considered, FVIIa-Y T seems to exist in a more stable and catalytically competent conformation prior to sTF binding with the allosteric pathways partially matured. This would result in a significant reduction in both enthalpic gain and entropic penalty upon sTF binding to the variant. If the ITC data for FVIIa-VY T is considered, yet another reduction is seen in both the enthalpy and entropy terms when compared with FVIIa-Y T , in agreement with further stabilization of the allosteric pathways in this variant as observed from the functional and inhibitor binding data. However, effects from manipulating the 170 loop cannot be ruled out, as it may result in an unfavorable TF-helix conformation adding to the observed reduction in binding enthalpy.
To elaborate on the findings from the ITC experiments, we solved the crystal structure of FVIIa-VY T variant in the presence of sTF and an irreversible active-site inhibitor. The crystal structure showed a topology similar to the recently reported structure of FVIIa-Y T , with Tyr-172 inserting into the pocket between AL2 and -3, stabilizing the 215-217 segment. The L163V mutation site was well-resolved in the electron density and shifted Phe-225 by ϳ1 Å. This minute change, combined with the insertion of Tyr-172, did, however, lead to a significant restructuring of the water network stabilizing AL2 and -3. The new configuration of the network is likely to be more rigid, due to the introduced hydroxyl group of Tyr-172, facilitating a more prominent stabilization of both the P1 arginine and the backbone of P2/P3 even in the absence of sTF. These subtle changes to the water net-

sTF-independent FVIIa
work are in good agreement with both the thermal stability findings and the ITC-binding data.
In conclusion, we have significantly substantiated the existence of both allosteric pathways suggested to be mediated by TF. We report the first FVIIa variant that is independent of sTF and exhibits a fully mature activation domain. This variant may be a useful tool to dissect the evolutionary role of the 35-fold increase in FVIIa activity mediated by TF. Recent studies have shown that the allosteric enhancement is not necessary for TFmediated coagulation, as exosites and membrane surface localization contribute much of the increase in catalytic turnover (44). Protease-activated receptor 2 signaling has, however, been found to be dependent on active site interactions (50), and the need for allosteric activation could therefore play an important role in the intracellular signaling pathways of FVIIa:TF that mediate the link between coagulation and inflammation.

Materials
H-D-Phe-Phe-Arg-chloromethylketone TFA salt (FFR) was from Bachem (Basel, Switzerland), and S-2288 (D-Ile-Pro-Argp-nitroanilide) was from Chromogenix (Milan, Italy). All other chemicals were from Sigma-Aldrich (Munich, Germany) and of analytical grade or highest quality commercially available. Recombinant WT human FVIIa was prepared as described previously (29). Recombinant human sTF 1-219 was prepared as described (51), with the modification of using the reductasedeficient Escherichia coli strain BL21 Origami (Novagen). FXa used in the purification process was purchased from Molecular Innovations (Novi, MI), whereas FX, FXa, and AT for the functional assays were acquired from Hematologic Technologies (Essex Junction, VT).

FVII mutagenesis and protein expression
Human WT FVII cDNA was cloned into a pLN174 vector or a QMCF vector (Icosagen AS, Tartu, Estonia), and the variants were generated with a QuikChange Lightning XL kit (Agilent Technologies, Santa Clara, CA) according to the manufacturer's instructions. Introduction of the desired mutations was verified by DNA sequencing of the entire FVII cDNA region (MWG Biotech). Expression of FVII-V and FVII-Q was conducted as described previously using the pLN174 vector and a baby hamster kidney cell line (16). Expression of FVII-Y T , FVII-VY T , and FVII-QY T was performed using the QMCF Technology, a semistable episomal mammalian expression system, obtained from Icosagen AS in a QMCF CHO cell line (CHOE-BNALT85) cultivated according to the manufacturer's instructions. Cells were expanded for 3-4 weeks to achieve sufficient volume, and the medium were harvested by centrifugation and 0.22-m filtration.

Protein purification and verification
For all FVII variants, expression medium pH was adjusted to 6.0, CaCl 2 was added to 5 mM, and benzamidine-HCl was added to a final concentration of 10 mM. All purification steps were essentially performed as described (21). Activation was performed by passing the protein solution through a custom-packed Tricon column (GE Healthcare) with FXa coupled to Sepharose 4B FF CNBr (GE Healthcare). Protein identity was verified using intact electrospray ionization-TOF MS, and purity was shown to be Ͼ95% by SDS-PAGE on a Novex 4 -12% NuPAGE gel (Invitrogen). The amount of active protein was determined by active site titration using FFR and measuring residual S-2288 hydrolytic activity (52).

sTF binding by surface plasmon resonance (SPR)
sTF binding to FVIIa variants measured by SPR was performed by immobilization of sTF-E219C-biotin (60 -80 resonance units) on a SA sensor chip using a Biacore T200 instrument (GE Healthcare Bio-Sciences AB, Uppsala, Sweden) according to the manufacturer's instructions. Prior to immobilization, sTF-E219C was biotinylated at its free terminal cysteine by reacting the protein with 2 mM biotin-PEO-iodoacetamide in 50 mM Hepes, pH 7.0, for 20 min. Excess reagent was subsequently removed by gel filtration on a NAP-5 column (GE Healthcare) and equilibrated in running buffer (HBS-P buffer containing 10 mM CaCl 2 ). Binding analyses were performed on the same chip repeatedly (regenerated for 2 min with HBS-P buffer containing 20 mM EDTA between runs) at eight protein concentrations (FVIIa-WT, -V, -Q: 128 to 0 nM; FVIIa-Y T , -VY T , -QY T : 320 to 0 nM) in HBS-P with 10 mM CaCl 2 at a flow rate of 10 l/min. The association and dissociation phases lasted for 8 and 11 min, respectively, and the temperature was 25°C. Steady-state data were fitted by nonlinear regression using a 1:1 steady-state affinity equation in Biacore T200 evaluation software 2.0 to obtain K d values.

Functional evaluation of FVIIa variants
All functional assays were carried out in 50 mM Hepes, pH 7.4, 0.1 M NaCl, 5 mM CaCl 2 , and 0.01% Tween 20 (assay buffer) and monitored at 405 nm in a microplate reader (SpectraMax 340, Molecular Devices, Sunnyvale, CA) using a Nunc F96 well plate (Non-Treated-clear) and 200-l assay volume at 25°C. sTF-binding studies using S-2288 were performed essentially as described (30), using 0 -3 M sTF. Kinetic parameters of S-2288 hydrolysis were determined for the FVIIa variants with 0 -12.5 mM S-2288, and the K i for inhibition by pABA was determined in a competitive activity assay using 1 mM S-2288 as described (29). Carbamylation of the N-terminal Ile-16 was investigated by incubating with 0.2 M KNCO and measuring residual activity at 1 mM S-2288 (29). All functional studies were performed in the absence and presence of 3 M sTF. Data analysis and curve fitting were performed using GraphPad Prism version 6.0.

AT inhibition and FX activation
AT inhibition and FXa generation assays were performed at 25°C in 50 mM Hepes, pH 7.4, 0.1 M NaCl, 10 mM CaCl 2 , 0.1% (w/v) PEG 8000, and 1 mg/ml BSA. To assess AT inhibition, 200 nM FVIIa variant was incubated with 12 M low-molecularweight heparin (Enoxaparin) with or without 5 M sTF, and the reaction was initiated by the addition of 2.5 M AT in a total volume of 250 l. At different time points, 20 l of the reaction mixture was transferred to a plate containing 250 nM sTF, 0.6 mg/ml Polybrene, and 1 mM S-2288 in a total assay volume of 200 l, and the hydrolysis of S-2288 was immediately moni-sTF-independent FVIIa tored at 405 nm for 5 min. The inhibition rate was projected by normalizing the data to a sample lacking AT and fitting the inhibition curve to a nonlinear regression one-phase decay model in GraphPad Prism version 7.0. FXa generation, in the absence of phospholipid membranes, was evaluated by adding 0 -312 nM FX to either FVIIa alone (100 nM FVIIa-WT, 50 nM FVIIa-Q or FVIIa-V, and 5 nM FVIIa-Y T , FVIIa-VY T , or FVIIa-QY T ) or 5 nM FVIIa variant with 2.5 M sTF. After a 20-min incubation, 1 mM S-2765 was added to the mixture, and the absorbance was monitored at 405 nm. The resulting S-2765 hydrolysis was converted to FX activation velocity, nM FXa/s, by using the equation obtained from an FXa (0 -6 nM) standard curve and dividing by the duration of the reaction (20 min). k cat /K m parameters were extracted from the slope of the initial activation rate.

Exosite inhibitor binding to FVIIa 170-loop variants
Inhibitor binding kinetics were evaluated using SPR. 50 g/ml anti-FVIIa ␥-carboxyglutamic acid domain antibody (4F6A4 (53)) diluted in sodium acetate, pH 5, was used for immobilization by a standard amine-coupling procedure on a CM5 chip according to the manufacturer's instructions (GE Healthcare Bio-Sciences AB). FVIIa-WT or FVIIa variant was diluted to 50 nM in running buffer (HBS-P with 10 mM CaCl 2 and 1 mg/ml BSA at pH 7.4) and injected into the activated cell for 60 s at a flow rate of 10 l/min, resulting in a capture level of 350 -450 RU. Following the capture of FVIIa, a 0 -250 nM concentration of either E-76 or A-183 was applied at 30 l/min in both the reference and the activated cell with a 60-s association and dissociation time. Regeneration was performed using HBS-P buffer with 50 mM EDTA. The resulting binding curves were fitted to a 1:1 Langmuir binding model in the Biacore evaluation software, and k on , k off , and K D values were derived. Functional inhibition was measured at saturating conditions for FVIIa-WT, -V, -Q (50 nM) and FVIIa-Y T , -VY T , -QY T (10 nM) using either 600 nM E-76 (38) or 600 nM A-183 (39) peptide to evaluate the effects from the combined mutagenesis. Residual activity was measured after 5-min incubation using the same approach as described for the functional assays with 3 mM S-2288. As a comparative control for full allosteric activation of FVIIa, we used 5 nM FVIIa-WT and 50 nM sTF. Significantly different inhibition levels compared with FVIIa-WT were evaluated with one-way analysis of variance with Dunnett's multiple-comparison test in GraphPad Prism version 6.0.

Thermal unfolding of FVIIa variants
Thermal unfolding was measured at a concentration of 1 M (50 g/ml) using a Prometheus NT.48 instrument (NanoTemper Technologies, Munich, Germany). All experiments were performed in 100 mM Hepes, pH 7.4, 100 mM NaCl, 10 mM CaCl 2 , 0.01% (v/v) Tween 20 at 25°C using quartz capillaries in four replicates performed on two separate days. Thermal unfolding was measured from 20 to 90°C using a 1.5°C/min scan rate with 80% excitation power at 280 nm and monitoring the ratio of 330/ 350-nm fluorescence. Data analysis and curve smoothing was performed using the PR.Control version 1.11 (NanoTemper Technologies) software package followed by first-derivative analysis to determine the transition midpoint (T m ).

Isothermal titration calorimetry
FVIIa-WT, FVIIa-Y T , FVIIa-VY T , and sTF were dialyzed simultaneously into 20 mM Hepes, 150 mM NaCl, 5 mM CaCl 2 , pH 7.4, using Slide-A-Lyzer TM dialysis cassettes overnight. The FVIIa concentration was determined under reducing conditions with an X-Bridge C4 reverse-phase column (Waters, Milford, MA) on an Alliance HPLC instrument with a 474 fluorescence detector, separating FVIIa into heavy and light chain followed by integration of the fluorescence signal contained in the light-chain peak. A recombinant FVIIa (NovoSeven, Novo Nordisk A/S, Bagsvaerd, Denmark) standard curve was used to calculate the amount of protein from the integrated peak. The sTF concentration was determined using A280 on a NanoDrop ND-2000 (Thermo Fisher Scientific) and an extinction coefficient at 280 nm of 37,080 cm Ϫ1 M Ϫ1 . All ITC experiments were conducted at 20°C on an ITC200 instrument (Malvern Instruments, Malvern, UK), washing with Decon90 (Decon Laboratories, Hove, UK), water, and buffer between each experiment. FVIIa-WT (5 M) experiments were conducted with the enzyme in the cell and titrating with 50 M sTF in the syringe. Experiments with FVIIa-Y T and FVIIa-VY T (10 and 8 M, respectively) were conducted in the same manner using 100 and 80 M sTF, respectively. Experiments were performed as triplicates, and data treatment was done using the PEAQ-ITC analysis software (Malvern Instruments) using a 1:1 binding model with a fitted offset compensating for the heat of dilution.

FVIIa variant preparation, crystallization, and data collection
Samples for X-ray crystallography were prepared by inhibiting FVIIa-VY T with FFR-cmk overnight and removing excess inhibitor by a single ion-exchange purification step. An equimolar amount of sTF was then added, and the protein mixture was concentrated to 4 mg/ml. Protein integrity was verified using SDS-PAGE. Diffraction quality crystals were obtained using vapor diffusion and hanging drop at 22°C with 0.1 M sodium citrate, pH 5.6, 15.6% (w/v) PEG 3350, 12% (v/v) 1-propanol (Hampton Research, Aliso Viejo, CA). Crystals were cryoprotected by transfer to 3 l of trimethylamine N-oxide (Hampton Research) before flash-freezing in liquid N 2 . All diffraction data were collected at MaxLab IV beamline I911-3 (54). Data were integrated and scaled using the XDS package (55). Molecular replacement was performed with the Phenix.Phaser software (56) and a FVIIa-WT:sTF 1-219 -FFR complex as search model. The subsequent refinement and model building were performed using iterative cycles of Phenix.Refine (57) in the Phenix program package (58), utilizing MolProbity (59) and individual anisotropic B-factors, followed by computer graphic model corrections by the Coot software (60).