Mutagenesis of the (cid:1) -Carboxyglutamic Acid Domain of Human Factor VII to Generate Maximum Enhancement of the Membrane Contact Site*

Site-directed mutagenesis of the 40 N-terminal residues ( (cid:1) -carboxyglutamic acid domain) of blood clotting factor VII was carried out to identify sites that improve membrane affinity. Improvements and degree of change included P10Q (2-fold), K32E (13-fold), and insertion of Tyr at position 4 (2-fold). Two other beneficial changes, D33F (2-fold) and A34E (1.5-fold), may exert their impact via influence of K32E. The modification D33E (5.2-fold) also resulted in substantial improvement. The combined mutant with highest affinity, (Y4)P10Q/K32E/D33F/ A34E, showed 150–296-fold enhancement over wild-type factor VIIa, depending on the assay used. Undercarboxylation of Glu residues at positions 33 and 34 may result in an underestimate of the true contributions of (cid:1) -carboxyglutamic acid at these positions. Except for the Tyr 4 mutant, all other beneficial mutations were located on the same surface of the protein, suggesting a possible membrane contact region. An initial screening assay was developed that provided faithful evaluation of mutants in crude mixtures. Overall, the results suggest features of membrane binding by vitamin K-dependent proteins and provide reagents that may prove useful for research and therapy.

Interaction between the vitamin K-dependent plasma proteins and a membrane surface is essential for hemostasis (1,2). This interaction is mediated through contact of the ␥-carboxyglutamic acid (Gla) 1 domains of these proteins with membranes containing acidic phospholipids (3,4). The Gla domain consists of ϳ40 N-terminal residues of which 9 -12 glutamic acid residues are post-translationally modified to Gla (5)(6)(7). In the presence of calcium, the Gla domain adopts structure and binds to membranes by a mechanism that is not fully understood.
The Gla domains of the different vitamin K-dependent plasma proteins show striking sequence homology, yet have quite different affinities for phospholipid membranes (8). If the Gla domain constitutes the membrane contact region, individ-ual amino acid residues within this domain should contribute to these different affinities. This suggestion has been supported by mutations of the Gla domain that affect membrane affinity. For example, the P10H mutant of bovine protein C shows an ϳ10-fold enhancement in membrane affinity, whereas the H10P mutant of human protein C shows an ϳ3-fold decline in membrane affinity (9). Although a major difference among the vitamin K-dependent proteins is a Gla residue versus another amino acid at position 32, the Q32E mutation of protein C has no impact on membrane affinity (10,11). Similarly, replacement of Glu 32 in factor X has little impact on protein function (12), and Glu 32 of human prothrombin was described as only moderately important (13), having little impact on either K m or V max of the prothrombinase reaction. However, a double mutant of human protein C, S11G/Q32E, shows an ϳ10-fold enhancement in membrane affinity (11), and a double mutant of factor VII, P10Q/K32E, has a 25-fold enhancement in membrane affinity (14). Under appropriate conditions, these proteins show similar improvement in function (11, 14 -16).
Although previous studies showed that proteins with enhanced membrane affinity can be created, the contribution of individual amino acid residues was not determined, and possible improvement by changes in other sites was not shown. This study used human factor VII (FVII) as a model Gla domain to investigate these questions and presents a simple functional assay that allows evaluation of minute quantities of impure protein as an initial screen of mutant protein activity. Singlesite mutants as well as those with the highest affinity were purified and characterized. The results indicate that nearly all functional improvements arose from membrane binding affinity. The largest enhancements in membrane affinity resulted from mutation in the region of residue 32, a position that is located quite far from the N-terminal end of the protein, where membrane contact is generally suggested (39). The mutant with highest function, (Y4)P10Q/K32E/D33F/A34E, showed 150 -296-fold improvement over wild-type FVII. Overall, the proteins described in this study contribute to a better understanding of the membrane contact site and the binding forces involved while providing novel reagents to probe coagulation reaction mechanisms that may be of use in therapy.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-Cloning and mutagenesis were performed by ATG Inc. (Eden Prairie, MN) following standard procedures (17). Human FVII cDNA was cloned from a human liver cDNA library and then subcloned into the vector pRc-CMV. Mutagenesis was verified by sequencing of the entire Gla domain of all variant FVII proteins, including untranslated pre-and propeptide segments. Proteins were expressed in fetal human kidney 293 cells that were stably transfected using the agent LipofectAMINE TM 2000 (Invitrogen) following the manufacturer's instructions. Following previously outlined procedures (9), Geneticin-resistant colonies were selected, and high pro-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The filtered medium was diluted 1:1 with distilled and deionized water containing 4.0 mM EDTA (pH 7.4) and 2.0 mM benzamidine hydrochloride and then applied to a High Q Macro-Prep anion-exchange column (Bio-Rad). The column was equilibrated prior to loading and washed extensively with Tris buffer (50 mM Tris, 100 mM NaCl, and 0.02% (w/w) NaN 3 (pH 7.4)) containing 2.0 mM EDTA and 2.0 mM benzamidine hydrochloride. The column was eluted isocratically with the same buffer containing 400 mM NaCl and no EDTA.
Eluted fractions containing FVII activity were pooled and diluted 1:1 with Tris buffer containing 30 mM CaCl 2 and 2 mM benzamidine hydrochloride. The pooled and diluted fractions were subjected to immunoaffinity chromatography using a calcium-dependent anti-human FVII monoclonal antibody (CaFVII22, provided by Dr. Walter Kisiel) coupled to Affi-Prep ® Hz support (Bio-Rad). The column was washed with Tris buffer containing calcium and then eluted with Tris buffer containing 15 mM EDTA and 2.0 mM benzamidine hydrochloride. Fractions containing FVII activity were pooled and subjected to a final ion-exchange chromatography step using a Mono Q HR5/5 anion-exchange column (Amersham Biosciences). The column was equilibrated and washed extensively with Tris buffer. Proteins were eluted with a linear gradient of 100 -500 mM NaCl over 30 min (flow rate of 1.0 ml/min). Eluted protein was concentrated by centrifugation filtration (Millipore Ultrafree, M r 10,000 cutoff) and then stored at Ϫ70°C. SDS-PAGE analysis indicated that the proteins were highly pure and contained only zymogen FVII and FVIIa in both nonreducing and reducing gels. The percentage of FVIIa ranged from 40 to 95%. Prior to assay for enzyme activity, the proteins were fully activated using a commercial source of tissue factor (TF; Innovin, Dade Behring Corp.).
FVII Protein Determination-A commercial preparation of FVIIa (No-voSeven, Nordisk) was used as the standard for all measurements. Protein concentrations were determined by the Bradford assay (40). The amidolytic activity for the chromogenic substrate S-2288 (Chromogenix) was also used as a standard of comparison. Plasma clotting assays, described previously (15), were used to determine FVIIa concentration. To ensure that FVIIa was the limiting component, the assay was conducted in the presence of excess TF (Innovin). TF concentrations in the Innovin preparation were determined from the concentration of NovoSeven needed to generate maximum activity. The TF concentration in the preparation used in this study was 2.7 nM. FVII was activated by mixing either crude or purified protein with 20 l of Innovin solution, followed by incubation at 37°C until activation was complete. (Typically, 1.0 l of solution resulted in a final concentration of ϳ0.3 nM FVII.) Activation of FVII was monitored by adding 2.65 l of the activation mixture to 112.5 l of prewarmed standard Tris buffer (0.05 M) containing 100 mM NaCl, 1.0 mg/ml bovine serum albumin, and 6.67 mM CaCl 2 . To initiate coagulation, 37.5 l of prewarmed FVII-deficient plasma (Sigma) was added, and clotting time was determined by the hand tilt-test method. Full activation occurred within 1 h, and concentrations of FVIIa were determined by comparison with the NovoSeven standard. The results for the different assay comparisons gave the same protein concentration within the estimated error of the assays (Ϯ20%).
Preparation of Small Unilamellar Vesicles-Highly pure phospholipids in organic solvent (bovine brain phosphatidylserine (PS) and egg phosphatidylcholine (PC), Sigma) were mixed at proper ratios and thoroughly dried, first under a stream of argon gas and then under vacuum for 14 h. Dried phospholipids were solubilized in Tris buffer and sonicated on ice for 3-s bursts every 10 s for 3 min (Heat Systems-Ultrasonics W-385). This was repeated four times. The sonicated vesicles were applied to a Sepharose 4B size-exclusion column (Amersham Biosciences). Eluted fractions containing vesicles with average diameters of 32-38 nm (LSA2 photon correlation spectrometer, Langley Ford Co.) were pooled for use in membrane binding studies. The concentrations of phospholipid were determined by phosphorus analysis (18) using a phospholipid/phosphorus weight ratio of 25:1.
Light Scattering Measurements-Protein-membrane binding was determined by relative light scattering at 90°using methods previously described (19). Briefly, for a constant particle concentration and for particles that are small compared with the wavelength of light, the ratio of the light scattering intensity of a protein⅐vesicle complex (I 2 ) to that of the vesicles alone (I 1 ) is related to the ratio of the molecular weight of the protein⅐vesicle complex (M 2 ) to that of the vesicles alone (M 1 ) by the relationship in Equation 1, where Ѩn/Ѩc is the change in refractive index as a function of concentration of the light scattering species and was estimated as described (19). Light scattering contributions of the buffer and protein were subtracted to obtain I 1 and I 2 . Values of M 2 /M 1 were estimated at various amounts of added protein and were plotted versus the protein (P)/phospholipid (PL) ratio (w/w). Dissociation constants (K D ) for protein-membrane binding were estimated from the relationship in Equation 2.
The amount of free protein ([P] free ) was estimated from the known weight concentrations of phospholipid and protein in the solution and the difference between M 2 /M 1 and the theoretical value of M 2 /M 1 if all of the added protein were bound. Weight concentrations of bound and free FVII were converted to molar concentration with a M r of 50,000 for all FVII variants. The concentration of P⅐PL was estimated from the known concentration of phospholipid and the ratio M 2 /M 1 . P⅐PL max was calculated by assuming saturation binding at a ratio of 1:1 protein/ phospholipid in the complex (M 2 /M 1(max) ϭ 2.0) (19).

Matrix-assisted Laser Desorption Ionization Time-of-Flight (MALDI-TOF)
Analysis-MALDI-TOF mass spectrometry was used to confirm the identity of the variant FVII proteins tested and to assess the extent of carboxylation. Purified proteins were incubated at 37°C for 30 min in the presence of either chymotrypsin or trypsin at a ratio of 500:1 (w/w) FVII protein/protease. Reaction solutions were dried by vacuum centrifugation and solubilized in a 5:95 acetonitrile/water solution containing 0.1% trifluoroacetic acid. The solutions were desalted using reversephase chromatography (ZipTip, Millipore Corp.), mixed 1:1 with a saturated matrix solution (5-methoxysalicylic acid in 50:50 ethanol/ water solution), spotted on the spectrometer target, allowed to dry, and then subjected to MALDI-TOF mass spectrometry (Biflex TM III, Bruker). Minimum laser power was used to obtain spectra. Moderate increases in power above this minimum did not result in changes in the distribution of the various carboxylated species observed. The percentage of each carboxylation species was determined by measurement of peak areas using integration software provide by the spectrometer manufacturer.
Factor X Activation by FVIIa-The relative activities of FVIIa variants were determined by a method outlined previously (15). Full activation of FVII was first ensured by incubation with TF (18 pM; Innovin) in 50 l of Tris buffer containing 5.0 mM CaCl 2 and 1.0 mg/ml bovine serum albumin. The mutant FVII proteins with higher membrane affinity (K32E, P10Q/K32E, P10Q/D33E, and (Y4)P10Q/K32E/D33F/ A34E) were added to a final protein concentration of 3.0 nM. WT-FVIIa, P10Q, and K32E were added to a final concentration of 10 nM. Full activation of the FVIIa preparation was achieved after incubation for 1 h. A range of WT-FVIIai concentrations was added, and the mixture was allowed to equilibrate for another 2 h at 37°C. Factor X (Enzyme Research Laboratories) was added to a concentration of 200 nM to initiate the reaction. After 10 min, the reaction was stopped by addition of excess EDTA (15 mM). The concentration of factor Xa was measured as activity for chromogenic substrate (0.32 mM S-2222, Chromogenix) by monitoring the absorbance change at 405 nm in a Beckman DU-70 UV-visible spectrophotometer. The amount of FVIIa that was bound to tissue factor (FVIIa⅐TF) was estimated from the activity observed relative to that of a standard with no WT-FVIIai. The concentration of WT-FVIIai bound to TF (WT-FVIIai⅐TF) was estimated from the fraction activity that was lost. Results are presented in a Hill-type plot represented by Equation 3 (15),

Enhancement of the Gla Domain of Human Factor VII
where K D(VIIai) is the dissociation constant for the WT-FVIIai⅐TF complex and K D(VIIa) is the dissociation constant for the FVIIa⅐TF complex.

Comparison of a plot of log([WT-FVIIai⅐TF]/[FVIIa⅐TF]) versus log[WT-
FVIIai] for two FVIIa variants at an identical and constant FVIIa concentration will allow estimation of their relative affinities for TF. Equation 3 represents free protein concentrations. Therefore, conditions were selected to ensure that the total protein concentration was in large excess over TF, so total protein was approximately equal to free protein.
Clotting Activity of FVII Proteins from Purified and Crude Sources-Pure FVII preparations or conditioned medium containing FVII were added to Innovin (0.1 ml) in an amount to generate ϳ0.3 nM FVII. This concentration of pure FVII provided a final coagulation time of 25 s for all samples. Because all FVII variants have the same clotting time in the absence of inhibitor (see pure protein analysis below and Ref. 15), use of a constant clotting time allowed the determination of the FVII concentration in an unpurified sample. It was important that TF was maintained in excess over FVII. Activation of FVII was allowed to proceed to completion (60 min at 37°C). Activated FVII solution (containing 2.5 l of Innovin) plus varying amounts of WT-FVIIai were mixed with Tris buffer containing 6.67 mM CaCl 2 and 1.0 mg/ml bovine serum albumin to create 112.5-l aliquots, which were incubated for 1 h at 37°C to achieve equilibrium binding between TF, WT-FVIIai, and FVIIa. Coagulation was initiated by addition of 37.5 l of prewarmed FVII-deficient plasma. Clotting times (CT) were determined, and results were evaluated by a plot of log where CT o is the clotting time without inhibitor. Relative function of the different FVIIa variants was estimated by offset of the plots for the two proteins.

Analysis of Protein Carboxylation States by MALDI-TOF
Mass Spectrometry-The identities of purified recombinant FVII proteins were verified by mass spectrometry of Gla domains released from the intact proteins by limited protease digestion. The Gla domains consisted of either 1-40 (containing all Gla residues) or 1-32 (less one carboxylation site at residue 35) N-terminal residues. Use of the methoxysalicylic acid matrix and the lowest practical laser power resulted in a low level of undercarboxylated peptide species (Fig. 1). In most cases, the fully carboxylated peptide (theoretical m/z of 5235 for the ϩ1 charge state of K32E) was the most abundant peak (Fig.  1A). However, in-source decarboxylation of Gla residues occurred during MALDI-TOF analysis and reached quantitative levels when the sinapinic acid matrix was used (20).
Undercarboxylation was detected by peaks separated by 44 mass units and a small peak corresponding to the fully decarboxylated peptide (m/z 4751) (Fig. 1A). The quantitative distribution among the different species was very consistent for replicate samples as well as for many plasma-derived versus recombinant proteins (20). This consistency suggested that comparison of quantitative MALDI-TOF data could be used to detect relative differences in the carboxylation states of various proteins. For example, repeated measurement of different preparations of plasma-derived bovine factor X gave 77 Ϯ 4% signal intensity in the fully carboxylated peptide. This level of fully carboxylated peptide corresponded to 97-98% carboxylation of all Glu residues. Similar results were obtained for bovine prothrombin and human protein C. Consequently, the somewhat lower yield of the fully carboxylated peptide of recombinant WT-FVII (46%) suggested undercarboxylation of the parent protein (Fig. 1B). In fact, undercarboxylation at position 35 of recombinant WT-FVIIa has been previously observed (21,22). That position 35 of recombinant WT-FVII was the major site of undercarboxylation was also suggested by MALDI-TOF analysis of peptide 1-32, which gave a high yield of the fully carboxylated state (70%) (Fig. 1B). Undercarboxylation at position 35 of FVII (and a corresponding residue in factor IX (23)) had no detected impact on the activity of the mature proteins (21,22). The total carboxylation levels of WT-FVII and P10Q/K32E, determined by this procedure, were 9.3 (of 10 theoretical) and 10.4 (of 11 theoretical) residues per Gla domain, respectively. These estimates were nearly identical to the values of 9.6 Ϯ 0.9 and 10.7 Ϯ 0.8 obtained by amino acid analysis after base hydrolysis (14).
Comparative analysis by MALDI-TOF was used to estimate the carboxylation states for the FVII mutants produced in this study. Five of the seven proteins showed Ͼ93% of theoretical Gla levels (Fig. 1B, last column), suggesting a carboxylation state of the parent protein similar to that of commercial FVIIa. Two mutants showed substantially lower levels of carboxylation, 8.9 of 11 theoretical residues for P10Q/D33E and 10.9 of 12 theoretical residues for (Y4)P10Q/K32E/D33F/A34E. The latter mutants contained additional Glu residues beyond position 32. Given that undercarboxylation occurs at position 35 of recombinant WT-FVIIa, it was possible that the additional Glu Interaction of Purified FVII Variants with Phospholipid Vesicles-When small unilamellar vesicles of 25:75 PS/PC were used to measure protein binding, the variant proteins displayed increasing membrane affinity in the order WT-FVII Ͻ P10Q Ͻ K32E ( Fig. 2A). Mutants with higher affinity (K32E) bound at the theoretical limit ( Fig. 2A), so equilibrium binding constants could not be estimated. Because binding affinity is dependent on PS content of the membrane, use of a lower PS content (10:90 PS/PC) (Fig. 2B) provided an equilibrium of bound and free protein for most mutants. Affinity increased in the order WT-FVII Ͻ P10Q Ͻ K32E Ͻ P10Q/D33E Ͻ P10Q/ K32E. Dissociation constants estimated from these results are reported in Table I (14). Even lower PS content was needed to estimate the binding constant for the highest affinity mutant (5:95 PS/PC) (Fig. 2C). Estimated K D values indicated a 3-fold enhancement in membrane affinity for the (Y4)P10Q/K32E/D33F/A34E mutant over the P10Q/K32E variant.
Activation of Factor X-Functional evaluation of the FVIIa mutants was carried out in a purified system that detected factor X activation. Experiments were performed under equilibrium competition conditions in which the FVIIa variants must compete with inhibitor, WT-FVIIai, for TF (described in Ref. 15). To allow comparison of results for different proteins, the FVIIa species and WT-FVIIai concentrations were maintained in great excess over TF, so the total FVIIa and FVIIai concentration approximated the respective free concentration. The ability of the lowest affinity FVIIa variants (10 nM) to displace WT-FVIIai increased in the order WT-FVIIa Ͻ P10Q Ͻ K32E (Fig. 3A). Mutants with higher function were evaluated at 3 nM and gave increasing affinity in the order K32E Ͻ P10Q/D33E Ͻ P10Q/K32E Ͻ (Y4)P10Q/K32E/D33F/ A34E (Fig. 3B). The WT-FVIIai concentrations required to reach 50% inhibition (FVIIai⅐TF/FVIIa⅐TF ϭ 1.0) are reported in Table I. In agreement with the higher affinity of WT-FVIIai for TF (15,24,25), its concentration at 50% inhibition was lower than that of WT-FVIIa.
The data in Fig. 3 are presented in the manner of a Hill plot. The plots have slopes of ϳ1.0, as anticipated. Validity of this analysis was supported by titration at different FVIIa levels. For example, a 3.4-fold difference in inhibitor concentration (Table I) was observed for titration of K32E at 3 nM (Fig. 3B) versus 10 nM (Fig. 3A). The concentration of WT-FVIIai at the midpoint of each titration curve was used to estimate the relative binding affinity of FVIIa variants. The K32E, P10Q/ K32E, and (Y4)P10Q/K32E/D33F/A34E mutants displayed 13.8-, 45-, and 149-fold increases in activity over WT-FVIIa, respectively. If the sites make independent contributions to membrane affinity, the 13.8-fold enhancement for K32E and the 3.1-fold enhancement for P10Q would suggest a 42.8-fold enhancement for P10Q/K32E. This was very nearly the value that was observed. Thus, this functional assay suggested that the P10Q and K32E modifications functioned in a manner that was independent of each other.
Overall, the results of the functional assay in the purified system under conditions in which the protein ligands (FVIIa and WT-FVIIai) were in large excess over TF mirrored the differences in membrane affinity observed in the phospholipid binding studies (Fig. 2). It appeared that all improvements in function arose from changes in the membrane contact site.
Functional Evaluation by Coagulation-Coagulation assays were conducted under conditions in which TF was more abundant than FVIIa. To compare the relative function of different FVIIa variants versus inhibition by WT-FVIIai, the concentration of free FVIIai needed to approximate the concentration of total FVIIai. Low affinity variants such as WT-FVIIa were displaced at low inhibitor concentration, at which most of WT-FVIIai was bound to TF. Consequently, comparison of protein function by this assay was limited to FVIIa variants with high affinity. This included K32E and better. These variants were displaced at WT-FVIIai concentrations that greatly exceed the TF concentration, so total WT-FVIIai approximated free WT-FVIIai in the solution.
This coagulation assay appeared useful for screening mutants in conditioned media (Fig. 4B). In fact, results obtained using this crude protein source were very reproducible for different batches of media and were indistinguishable from the results obtained for purified proteins (Fig. 4A). Screening of FVIIa variants in conditioned media was used as a facile first method to estimate protein function to identify the best proteins for purification. This assay appeared amenable to high throughput analysis and might be used in future studies to identify beneficial mutants from large libraries of cells expressing proteins with different modifications. A logical modification would be to quantify FVIIa activity by measuring the product, factor Xa, by amidolytic assay.
The Tyr 4 , D33F, and A34E mutants were introduced sequentially into P10Q/K32E to estimate their individual impact.  These proteins were evaluated only in the preliminary screening test. The result showed that the Tyr 4 insertion conferred a 2-fold functional enhancement (data not shown) and that this was independent of other changes in the Gla domain. Introduction of D33F also increased the function of P10Q/K32E by 2-fold. Incorporation of D33F into WT-FVIIa was not attempted, so it was not possible to determine whether this change was independent of other changes. Introduction of A34E into proteins that did not contain K32E had no detected influence on protein function, but showed a 1.5-fold increase in function when introduced into P10Q/K32E/D33F. The order of observed enhancements of function generally followed changes in membrane affinity, but appeared to slightly exceed changes in K D values (Table I). DISCUSSION Human FVII displays one of the lowest affinities for phospholipid membranes among the vitamin K-dependent plasma proteins (8). Modifications within the Gla domain of FVII (P10Q/K32E) have been shown to enhance membrane affinity (14) and, under appropriate conditions, to increase function of the protein (15). This study identified several new sites that also enhance membrane affinity and protein activity. Mutations were designed to test the independence and relative contribution of each site. A simple assay, amenable to screening large numbers of variant proteins, that used unpurified FVIIa was shown to faithfully report changes in activity due to mutagenesis of the Gla domain.
The interactions responsible for binding of the Gla domain of vitamin K-dependent plasma proteins to phospholipid membranes are not fully understood. The location of amino acids that enhance affinity as well as the magnitude of the changes can offer information about the membrane contact site and binding mechanism. All but one of the beneficial changes reported here were clustered on the same surface of the Gla domain (Fig. 5), suggesting a possible membrane contact surface.
The exception was the Tyr 4 insertion, which is located on nearly the opposite surface in the protein, adjacent to the N-terminal structure referred to as the -loop (26). Insertion of an additional hydrophobic residue at position 4 was motivated by sequence comparison with bovine protein Z, a vitamin K-dependent protein with very high membrane affinity. The -loop of the Gla domain provides three solvent-exposed hydrophobic side chains (Phe 4 , Leu 5 , and Leu 8 for human FVII) (Fig. 5) (27). Although this region in bovine prothrombin fragment 1 is a contact point in the crystal (26) and may be subject to some distortion, these hydrophobic groups have been suggested to constitute a central feature of the membrane-binding site (28 -30). Insertion of tyrosine (or phenylalanine; data not shown) at this position resulted in a 2-fold increase in protein function. One possible explanation for enhanced membrane binding is contact of the Tyr 4 side chain with the hydrophobic region of the phospholipid membrane. However, hydrophobic insertion by the -loop has been questioned on many grounds (3,(31)(32)(33)(34)(35). The impact of the Tyr 4 insertion was small (2-fold; ⌬⌬G ϭ 0.4 kcal/mol) compared with the free energy for transfer of a large hydrophobic group such as the phenyl side chain from aqueous solution (to octanol; ⌬⌬G ϭ Ϫ3.6 kcal/mol for ⌬K D ϭ 400-fold) (36). The small impact of the Tyr 4 insertion suggested an indirect role in membrane association.
Replacement of Pro 10 , a residue present only in vitamin K-dependent proteins displaying low membrane affinity, also increased protein affinity by 2-3-fold. Assay by the preliminary screen showed that any of a number of amino acid substitutions (Gln, Tyr, Glu, Leu, or Asn) at this position gave similar enhancement. The small effect of Pro 10 replacement and the large range of permissible residues suggested a relatively peripheral effect. Thus, this level of affinity enhancement may result from increased mobility within this region and improved packing of hydrophobic groups near H-bonding networks. Complete removal of a H-bond from an aqueous environment results in a change in free energy of approximately Ϫ2.4 to Ϫ3.6 kcal/mol (37).
The largest single benefit was provided by introduction of glutamic acid at position 32, producing a 13-fold enhancement as a single independent site (⌬⌬G ϭ Ϫ1.6 kcal/mol). Basic residues occur at this position in human and bovine FVII. This change in membrane affinity, which resulted in a large improvement in protein function, equates with a free energy change of ϳ15% of the total binding energy for the protein (K D ϳ 10 Ϫ7 M; ⌬G ϳ Ϫ10 kcal/mol). Although not measured as independent changes, it was likely that the D33F (2-fold enhancement) and A34E (1.5-fold enhancement) mutations exerted their impact by indirect means such as by increasing the pK a of Gla 32 , thereby increasing affinity for ion pairs such as calcium binding. Gla 32 has been proposed to associate with an extra calcium ion during membrane binding (8). It was reasoned that another Gla residue at position 34 may enhance this interaction, whereas addition of a large hydrophobic side chain in this vicinity may serve to isolate this charge complex from water molecules. The total impact of changes at positions 32-34 (39-fold; ⌬⌬G ϭ Ϫ2.2 kcal/mol) may be sufficient to suggest a direct role for this region in membrane association. This conclusion is necessarily tentative and must be confirmed by additional study.
Completely independent contribution as a result of change at positions 10 and 32 was surprising. In the case of human protein C, simultaneous changes at residues 11 and 32 are required to realize enhancement (11). It is possible that Ser 11 of human protein C hinders development of a structure, allowing contact of position 32 with calcium or the membrane.
Another surprising benefit was the D33E mutation. This change introduced a 5.2-fold increase in function over P10Q. Taking into consideration the observed undercarboxylation of the P10Q/D33E mutant, possibly at position 33 (see below), it is possible that Gla 33 offers a significantly larger enhancement and may rival the impact of K32E. This was surprising for several reasons. First of all, the charge difference and electrostatic effect of the D33E mutation (⌬charge ϭ Ϫ1) were much smaller than those of the K32E mutation (⌬charge ϭ Ϫ3). Furthermore, the D33E mutation would leave a cationic group at position 32 and would move the Gla residue relatively far from any of the suggested membrane contact regions. Clearly, additional effort is needed to identify the mechanism and binding site of vitamin K-dependent proteins.

Enhancement of the Gla Domain of Human Factor VII
The highest affinity mutant that has been identified to date, (Y4)P10Q/K32E/D33F/A34E, provided 146 -296-fold enhancement over WT-FVIIa (⌬⌬G ϭ Ϫ3.0 to Ϫ3.4 kcal/mol). The range of functional enhancement arose from use of different assays. The benefit of P10Q/D33E also differed with these assay methods. It is possible that these method-dependent differences arose from undercarboxylation of these two mutants, producing preparations with different functional states. Assays that used excess FVIIa would detect an average function for the entire preparation, resulting in an overestimate of the function of the low affinity species and an underestimate of the function of the high affinity species.
The coagulation assays also suggested heterogeneity of these preparations. Titration of K32E and P10Q/K32E gave plots with parallel slopes, indicating similar differences between these proteins at all levels of inhibition. Parallel plots would arise if both protein preparations were homogeneous but had different functional levels. Nearly all of the much larger number of mutants that were assayed by the initial screening test (Fig. 4B) gave similar curve shapes and slopes. The exceptions were mutants that contained Glu residues beyond position 32. These proteins, including P10Q/D33E and (Y4)P10Q/K32E/ D33F/A34E, showed substantial inhibition at low WT-FVIIai concentrations, but great resistance to complete inhibition at high WT-FVIIai concentrations. This could arise from initial displacement of a low affinity FVIIa species (possibly an undercarboxylated protein), followed by competition with a high affinity FVIIa species (the fully carboxylated protein). MALDI-TOF analysis suggested undercarboxylation of these proteins, as well. Although further studies are needed, it is possible that the 5.2-fold increase in function arising from the D33E mutation (Table I) and the 149-fold increase in function of (Y4)P10Q/ K32E/D33F/A34E actually underestimate the value for the fully carboxylated protein. The best analysis of the fully carboxylated protein (296-fold enhancement) (Table I) may be obtained by the coagulation test at high levels of FVIIai. Several future approaches may used to address this matter. For example, although complete separation of proteins by charge has proven difficult, cell lines may be found that provide more complete carboxylation of FVII polypeptides to generate proteins with the highest function.
Overall, this study has outlined a number of sites that improve function of FVIIa and, by inference, suggested approaches that may enhance the membrane affinity of other vitamin K-dependent proteins. Mutations within the Gla domain have allowed for investigation of surfaces within this domain that may contribute to the protein-membrane interactions. In contrast to the view of membrane interaction centering around hydrophobic residues near the N terminus (39), we have found that amino acid changes within the region of the Gla domain encompassing residues 32-34 resulted in the greatest enhancements in membrane affinity (Fig. 5). Access to proteins with a range of function may prove useful in investigation of blood clotting reactions both in vitro and in vivo. The large enhancements obtained may also provide benefit for therapies that are based on the vitamin K-dependent proteins (38).