Identification of the phospholipid binding site in the vitamin K-dependent blood coagulation protein factor IX.

The blood coagulation and regulatory proteins that contain γ-carboxyglutamic acid are a part of a unique class of membrane binding proteins that require calcium for their interaction with cell membranes. Following protein biosynthesis, glutamic acids on these proteins are converted to γ-carboxyglutamic acid (Gla) in a reaction that requires vitamin K as a cofactor. The vitamin K-dependent proteins undergo a conformational transition upon metal ion binding, but only calcium ions mediate protein-phospholipid interaction. To identify the site on Factor IX that is required for phospholipid binding, we have determined the three-dimensional structure of the Factor IX Gla domain bound to magnesium ions by NMR spectroscopy. By comparison of this structure to that of the Gla domain bound to calcium ions, we localize the membrane binding site to a highly ordered structure including residues 1–11 of the Gla domain. In the presence of Ca2+, Factor IX Gla domain peptides that contain the photoactivatable amino acid p-benzoyl-L-phenylalanine at positions 6 or 9 cross-link to phospholipid following irradiation, while peptides lacking this amino acid analog or with this analog at position 46 did not cross-link. These results indicate that the NH2 terminus of the Gla domain, specifically including leucine 6 and phenylalanine 9 in the hydrophobic patch, is the contact surface on Factor IX that interacts with the phospholipid bilayer.

The vitamin K-dependent blood coagulation and regulatory proteins interact with phospholipid membranes in the presence of Ca 2ϩ . This interaction plays an important biological role in the formation of protein complexes with enhanced catalytic properties (Furie and Furie, 1988). For instance, the catalytic activity (k cat /K m ) of Factor IXa for the activation of Factor X is increased almost 10 9 -fold by the presence of phospholipids, calcium ions and Factor VIIIa (van Dieijen et al., 1981). The ␥-carboxyglutamic acid (Gla)-rich domain of the vitamin K-de-pendent proteins mediates phospholipid recognition upon the liganding of calcium ions and the subsequent formation of a phospholipid binding site (for review, see Furie and Furie (1988)). ␥-Carboxyglutamic acid, which defines the calcium ion binding sites of this domain, is synthesized posttranslationally from glutamic acid in an enzymatic reaction directed by a propeptide on the precursor form (Jorgensen et al., 1987) and catalyzed by the vitamin K-dependent carboxylase (Suttie, 1985). In contrast to Ca 2ϩ , other divalent metal ions induce a stable conformer of the Gla domain that does not bind phospholipids (Nelsestuen et al., 1976). A two-step sequential conformational transition model for the metal ion-induced structural changes has been proposed based on observations using conformation-specific antibodies (Borowski et al., 1986;Liebman et al., 1987). In the first step, the binding of a variety of divalent metal ions to the Gla domain results in a conformational change with the expression of new antigens common among the metal ion-protein binary complexes. This conformer does not bind phospholipid vesicles. In the second step, saturation with calcium ions induces a specific conformer that binds to phospholipids and expresses an additional neoantigenic determinant. Antibodies and their Fab fragments directed against the Ca 2ϩ -stabilized epitope block phospholipid binding (Borowski et al., 1986;Liebman et al., 1987). The differences between the metal-free structures and the structures of Ca 2ϩbound Gla domains from Factor IX and a homologous protein, prothrombin, have been determined in the absence of phospholipid (Soriano-Garcia et al., 1992;Freedman et al., 1995aFreedman et al., , 1995b. However, the location of the phospholipid binding site within the Gla domain remains unknown. A synthetic peptide, Factor IX (1-47), which contains the Gla domain and the aromatic amino acid stack domain of Factor IX, has been used to study the structural requirements of phospholipid binding by Factor IX (Jacobs et al., 1994). We show that the Factor IX (1-47) peptide, like Factor IX, does not interact with phospholipid membranes in the presence of magnesium ions but does interact with phospholipids in the presence of calcium ions. Here, we have localized the phospholipid binding site by determining the solution structure of the Factor IX Gla domain peptide when fully saturated with Mg 2ϩ . Since the Ca 2ϩ -and Mg 2ϩ -bound Gla domains undergo a common metal-induced fluorescence transition and share a structural epitope (Borowski et al., 1986;Liebman et al., 1987), the differences between the Ca 2ϩ -and Mg 2ϩ -bound structures identify the NH 2 -terminal 11 amino acids as critical for phospholipid binding. The covalent cross-linking of a Factor IX Gla domain analog containing a photoactivatable amino acid, pbenzoyl-L-phenylalanine, at position 6 or position 9 to phospholipid proves that residues 6 and 9 are within or in close proximity of the membrane.
NMR spectroscopy was performed on three different samples of Factor IX (1-47) in H 2 O: 1) 0.45 mM peptide, 8 eq of MgCl 2 , pH 5.8; 2) 1.5 mM peptide, 8 eq of MgCl 2 , pH 5.8; 3) 1.5 mM peptide, 3 M urea, 2.5 M guanidine HCl, 120 eq of MgCl 2 , 0.5 M deuterated (d 4 ) acetate, pH 5.4. The latter sample was prepared to compare the spectra to that of the Ca 2ϩ -bound peptide analyzed in the same solvent (Freedman et al., 1995b). The samples also contained 10% D 2 O for the deuterium lock signal. A sample in D 2 O was prepared by lyophilizing 1 mM peptide with 14 eq of MgCl 2 , 0.1 M deuterated (d 4 ) acetate, pH 5.3, in 99.8% D 2 O, and then redissolving in 99.96% D 2 O. Spectra were collected on a Bruker AMX-500 spectrometer with a proton frequency of 500.14 MHz. The carrier frequency was set on the water resonance, which was suppressed using presaturation.
We observed concentration-dependent aggregation of the Factor IX (1-47) peptide in the presence of magnesium ions that, under certain conditions, resulted in minor amounts of peptide precipitation. We have previously reported Mg 2ϩ -induced linewidth broadening in one-dimensional NMR spectra of the Factor IX (1-47) peptide (Freedman et al., 1995a). Despite these minor impediments to the acquisition of high quality spectra, two-dimensional 1 H NMR studies were performed on the Mg 2ϩ -saturated Factor IX (1-47) peptide. Spectra acquisition and processing were similar to that performed previously for the apo and Ca 2ϩ -bound peptides (Freedman et al., 1995a(Freedman et al., , 1995b. Briefly, two-dimensional NOESY spectra were recorded with mixing times of 150, 250, and 300 ms at 25°C, and 250 ms at 35°C. A total of 2048 real data points was acquired in t 2 , 384 -512 points in t 1 , a spectral width of approximately 7000 Hz, 128 -160 summed scans, and a relaxation delay of 1.3 s between scans. As an alternative to presaturation for H 2 O suppression, the jump-and-return technique was also used to maximize amide intensity (Plateau and Guéron, 1982). Total correlation spectroscopy (TOCSY) data were recorded using a mixing time of 55 ms, 80 summed scans, and an MLEV-17 mixing sequence (Bax and Davis, 1985). Double-quantum filtered correlation spectroscopy (DQF-COSY) data were recorded with 2048 real t 2 points, 64 summed scans, and 1024 increments (Piantini et al., 1982). As described previously, the NMR data sets were multiplied by sine bell window functions and zero-filled to 2000 by 1000 (real) matrices (Freedman et al., 1995a(Freedman et al., , 1995b. Sequence-specific resonance assignments were made using the methodology described for the metalfree Factor IX (1-47) and Ca 2ϩ -bound peptides (Freedman et al., 1995a(Freedman et al., , 1995b. Briefly, the NOE contacts were classified into five categories: 1) "intraresidue" for NOEs within a residue; 2) "sequential" for contacts between the backbone and side-chain protons of residue i with the backbone amide proton of residue iϩ1; 3) "short range" for all other contacts between residue i and iϩ1; 4) "medium range" for NOEs between protons on residues separated by 3 amino acid positions or less in the sequence; and 5) "long range" for contacts between residues that are separated by 4 amino acid positions or more in sequence.
Characteristic spin system patterns were observed in the Factor IX (1-47)-Mg 2ϩ peptide spectra that coincided with the spin system patterns for the 20 naturally occurring amino acids as described by Wü thrich (1986). ␥-Carboxyglutamic acid residues had the spin system pattern as described previously (Freedman et al., 1995a(Freedman et al., , 1995b. The main starting point for spectral assignment was the identification of the spin system characteristic for the single tryptophan in the peptide. Correlations from aromatic side chains from the ring protons to the ␤ protons (and backbone) were achieved primarily from the D 2 O NOESY spectrum. For some residues, the cross-peak patterns for the Ca 2ϩbound Factor IX (1-47) peptide and metal-free Factor IX (1-47) peptide were used as a guide in assigning the proton resonances. In TOCSY spectra, correlation from the amide protons generally extended to the ␤ protons. Side-chain identification beyond the ␤ protons were mainly completed from the aliphatic region of TOCSY (and DQF-COSY) spectra. Here correlations were generally observed from the ␣ and ␤ protons out to the end of the chain. The intraresidue information from these spectra in conjunction with the NOESY spectra, which displayed all sequential connectivities (d NN , d ␣N , d ␤N ; Wü thrich (1986)), made nearly complete proton resonance assignments possible. The proton resonance assignments are presented in Table I. NOESY cross-peak intensities were converted into three distance classes (strong, 1.7-3.0 Å; medium, 1.7-4.0 Å; weak, 3.0 -5.0 Å) and calibrated using published methods (Detlefsen et al., 1991;Hyberts et al., 1992). Comparison of NOESY spectra collected at both short and long mixing times was used to control for spin diffusion effects. Nonstereospecifically assigned atoms were treated as pseudoatoms and given correction distances according to the guidelines presented by Wü thrich (1986).
Structure determination used a set of 710 distance restraints (intraresidue and sequential, 380; short, medium, and long range, 330). A combination of distance geometry and simulated annealing methods (Havel, 1991) were used to generate 15 final structures using the DGII program of InsightII (Biosym Technologies, San Diego, CA). An average structure was calculated using the Analysis program of InsightII. The distance and energy violations were compared for each structure in the set. Backbone root-mean-square deviation values following superimposition with the geometric average reflected the quality of the structures determined. The Factor IX (1-47)-Mg 2ϩ structures were also compared to the metal-free and Factor IX (1-47)-Ca 2ϩ structures by superimposition of backbone atoms.
The three-dimensional structures of Factor IX (1-47)-Ca 2ϩ and Factor IX (1-47)-Mg 2ϩ were formally compared using QUANTA (Molecular Simulations). These structures were overlaid to minimize the backbone root-mean-square deviation values between residues 15 and 47. Using the Molecular Similarity utility, the molecular volumes of both structures were defined, and the volume common to both structures identified. All residues in Factor IX (1-47)-Ca 2ϩ within and outside of this common volume of matched atoms were color-coded.
Covalent cross-linking of Factor IX (1-47) to phospholipid was performed via photoactivation of p-benzoyl-L-phenylalanine incorporated into the polypeptide backbone. 125 I-Labeled Factor IX (1-47)/Bpa 6, 125 I-labeled Factor IX (1-47)/Bpa 9, and 125 I-labeled Factor IX (1-47)/ Bpa 46 were affinity-purified after iodination using anti-Factor IX-Ca(II)-specific antibodies bound to Sepharose and were eluted with EDTA (Liebman et al., 1985(Liebman et al., , 1987. 125 I-labeled Factor IX (1-47) was purified by the same method except that anti-Factor IX-Mg(II) antibodies were employed. These peptides were incubated with phospholipid vesicles (5 M) in 50 mM Tris, pH 7.4, 50 mM NaCl. CaCl 2 , MgCl 2 , or EDTA was added to a final concentration of 10 mM. The samples were irradiated simultaneously with a B-100 AP ultraviolet lamp (UVP) for 20 min at a distance of 5.5 cm. Non-covalent Ca 2ϩ -mediated proteinphospholipid binding was reversed by the addition of EDTA to a final concentration of 25 mM. Bound and free peptide were separated by gel filtration on a Superose 12 FPLC column equilibrated in 50 mM Tris, pH 7.4, 50 mM NaCl, 10 mM EDTA, 0.1% bovine serum albumin, 0.025% Tween 80. Fractions eluted from the column were assayed for 125 I in a ␥ scintillation spectrometer. The elution volume of the phospholipid vesicles was determined by analysis of phosphorus content (Chen et al., 1956). The amount of peptide cross-linked was determined by measuring the 125 I content associated with the phospholipid vesicles after treatment of the peptide-phospholipid complex with EDTA.

RESULTS
Fluorescence energy transfer studies have shown that the Factor IX (1-47) peptide behaves like full-length Factor IX in that it binds phospholipid vesicles in the presence of Ca 2ϩ , but not in the absence of metal ions (Jacobs et al., 1994). In separate studies, we have also determined that Factor IX does not bind phospholipid vesicles in the presence of magnesium ions (Liebman et al., 1987). Thus, we set out to verify whether the Factor IX (1-47) peptide had the same property. As shown in Fig. 1, the Factor IX (1-47) peptide binds to phospholipid in the presence of Ca 2ϩ . There is no evidence for Mg 2ϩ -dependent phospholipid binding, even at a concentration of peptide that results in saturation of the vesicles by the Ca 2ϩ -bound Factor IX (1-47) peptide. Therefore, the metal ion-dependent phospholipid binding characteristics of full-length Factor IX are reflected in the Factor IX (1-47) peptide. The differences between the three-dimensional structures of the Ca 2ϩ -stabilized Factor IX (1-47) peptide and the Mg 2ϩ -stabilized Factor IX (1-47) peptide should then define a region critical for phospholipid binding. For these reasons, we determined the solution structure of the Mg 2ϩ -stabilized Factor IX (1-47) peptide by using NMR spectroscopy.
The short, medium, and long range nuclear Overhauser effect (NOE) contacts for the carboxyl-terminal 36 residues were nearly identical to those of the Ca 2ϩ -bound peptide (Fig. 2).
Only a few cross-peaks were weak or obscured by spectral overlap. A comparison of the chemical shift values between the Ca 2ϩ -and Mg 2ϩ -bound structures showed that they were similar for residues 14 -47 ( Fig. 3), except for residues 21-23 (see below). In contrast, there was little similarity in the short, medium, and long range NOE contacts within the amino-terminal 13 residues between the spectra collected on the two metal ion-bound structures. All NOE contacts confined to this peptide region in Factor IX (1-47)-Mg 2ϩ were instead similar to those found in the apoFactor IX (1-47) spectra ( Fig. 2) (Freedman et al., 1995a). In addition, the chemical shift values for all residues in this region were similar for the apo and Mg 2ϩ -bound structures (Fig. 3). Many of the chemical shifts in the amino-terminal region were centered around random coil values (Wishart et al., 1991). For residues 21-23, the ␣ proton chemical shift differences between the Ca 2ϩ -and Mg 2ϩ -bound peptides are large, as indicated in Fig. 3. These may be accounted for by contacts between Tyr-1 and residues 21-23 in the Factor IX-Ca 2ϩ structure that do not occur in the Factor IX (1-47)-Mg 2ϩ structure (Freedman et al., 1995b).
The spectrum of Factor IX (1-47)-Mg 2ϩ taken on a sample containing 1.5 mM peptide in H 2 O, pH 5.8, had a small subclass of novel NOE contacts between residues 1-11 and residues 12-47 of the Mg 2ϩ -bound structure that could not be identified in either the apo or Ca 2ϩ -bound peptide spectra reported previously (Freedman et al., 1995a(Freedman et al., , 1995b (Fig. 2). These 22 NOE contacts represent intermolecular contacts since additional spectra collected using the solvent system previously used to determine the Factor IX (1-47)-Ca 2ϩ structure (Freedman et al., 1995b): 3 M urea and 2.5 M guanidine HCl, eliminated the novel NOE contacts. These NOE contacts were also eliminated at lower concentrations of peptide (0.5 mM) in the absence of guanidine/urea. Mg 2ϩ -bound peptide structures were generated by distance geometry methods using 710 distance restraints, of which 380 were intraresidue and sequential distances, and 330 were short, medium, and long range distances. Due to cross-peak broadening, coupling constants below approximately 7.5 Hz could not be measured. The NOE data indicate that the majority of residues are found in helices where ␣-NH coupling constants of less than 6 Hz occur. As a consequence, torsion angle information was not available for structure calculations. The Mg 2ϩ -bound peptide had defined structure from residues 12-46. For this region, an average backbone root-mean-square deviation value of 0.9 Ϯ 0.1 Å was determined by superimposition with the geometric average. A comparison with the analogous region of the Ca 2ϩ -bound structure shows that they contain the same secondary structural elements, predominantly helices, and the same tertiary structure (Fig. 4). The lowest energy Ca 2ϩ -bound structure (residues 12-46) is shown superimposed with a series of the Mg 2ϩ -bound structures in Fig. 4. The root-mean-square deviation upon superimposition of the average Mg 2ϩ -bound structure with the average Ca 2ϩbound structure is 1.2 Å for residues 12-46. In this region, both the Mg 2ϩ and Ca 2ϩ -bound forms are compact and approximately globular. In contrast, the amino-terminal 11 residues lacked defined structure in Factor IX (1-47)-Mg 2ϩ , with the exception of a short loop from residue 6 to 9. The same loop was found in the apoFactor IX (1-47) structure (Freedman et al., 1995a). The two-step conformational transition of Factor IX can now be defined in structural terms; the apoFactor IX lacks formal structure in residues 1-36, with the exception of resi-dues 6 -9 and 18 -23. Upon occupancy of specific metal bindings sites that interact with many divalent cations, including Mg 2ϩ , the Gla domain assumes formal structure except that residues 1-11 remain flexible and motile. With occupancy of another set of metal bindings, sites that can be occupied only by Ca 2ϩ , the entire polypeptide backbone of the Gla domain is defined, including the NH 2 -terminal loop (Fig. 5).
To compare the two binary complexes quantitatively, the structural model of Factor IX-Mg 2ϩ was overlaid on the structural model of Factor IX (1-47)-Ca 2ϩ . The molecular volume of each structure was determined, and the volumes that are shared by equivalent atoms within both structures were defined (Fig. 6A). Using only the structural model of Factor IX (1-47)-Ca 2ϩ , residues contained within the common volume are identified in white; these are the atoms whose positions are preserved in the two structures (Fig. 6B). Residues outside this volume represent the region from residue 1 to residue 11. Thus, the yellow atoms represent structural differences between the two metal-protein complexes, differences that correlate with phospholipid binding function, i.e. residues 1-11 represent the putative phospholipid binding site.
To prove that residues 1-11 of Factor IX are in close proximity of the phospholipid membrane in the Factor IX-phospholipid-Ca 2ϩ ternary complex, we used a photoactivatable crosslinker to covalently couple the Factor IX (1-47) peptide to phospholipid in the presence of Ca 2ϩ . Bpa is a photolabile amino acid analog composed of a side-chain benzophenone (Dorman and Prestwich, 1994) (Fig. 7A). Upon irradiation at  (Jacobs et al., 1994). It is shown here (closed circles) for comparison to binding studies performed with the Factor IX (1-47) peptide in the presence of 1 mM MgCl 2 (closed squares). The change in fluorescence was monitored by irradiating at 280 nm and recording dansyl emission at 520 nm as a function of increasing peptide concentration. I o , fluorescence (arbitrary units) in the absence of peptide; I, fluorescence at indicated peptide concentration.
FIG. 2. The aromatic methyl side-chain region of a NOESY spectrum illustrating the presence of long range NOEs in the Factor IX (1-47)-Mg 2؉ binary complex. The sample contained 1.5 mM Factor IX (1-47) peptide, 12 mM MgCl 2 , and 10% D 2 O, pH 5.8, and the spectrum was recorded with a mixing time of 300 ms. All crosspeaks are labeled by residue number and proton type for the two protons making the NOE contact. The cross-peaks enclosed in the rectangular box are also found in the metal-free and Ca 2ϩ -bound peptide spectra. The cross-peaks indicated by an asterisk are found only in the Factor IX (1-47)-Mg 2ϩ spectrum at high peptide concentration in the absence of guanidine/urea. Otherwise unmarked cross-peaks are common to the Factor IX (1-47)-Ca 2ϩ NOESY spectrum. The analogous region of the Ca 2ϩ -bound peptide is presented in Freedman et al. (1995b). 350 nm, the benzophenone decomposes and forms a covalent bond with carbon-hydrogen bonds within 3 Å. By substituting Bpa for specific hydrophobic residues in Factor IX (1-47), we evaluated the ability of Bpa incorporated into specific regions of the peptide to mediate cross-linking to phospholipid. In order to avoid disruption of the tertiary structure of the Gla domain, we designed peptides that placed Bpa in positions of hydrophobic residues whose side chains are oriented toward solvent (Freedman et al., 1995b). Leucine 6, phenylalanine 9, and valine 46 meet these criteria. Three additional peptides were prepared by solid phase synthesis: Factor IX (1-47)/Bpa 6, Factor IX (1-47)/Bpa 9, and Factor IX (1-47)/Bpa 46 (Fig. 7B). These peptides were homogeneous by high performance liquid chromatography and by SDS-gel electrophoresis.
To demonstrate that these Bpa-containing peptides exhibited the membrane binding properties characteristic of Factor IX (1-47), we used fluorescence energy transfer to study peptide-membrane interaction. Using fluorescence energy transfer with phospholipid vesicles composed of phosphatidylserine: phosphatidylcholine:dansyl-phosphatidylethanolamine (40:50: 10), we observed binding of the Bpa-containing peptides to phospholipid membranes in the presence of Ca 2ϩ (Fig. 8). The K d for Factor IX (1-47)/Bpa 6 binding to phospholipid was 1.0 M; Factor IX (1-47)/Bpa 9 was 1.5 M; Factor IX (1-47)/Bpa 46 was 6.0 M. These values are equivalent within experimental error to the K d of 2.4 M measured for Factor IX (1-47). The interaction was Ca 2ϩ -dependent; the addition of EDTA inhibited binding. The results indicate that, in the absence of conditions that induce covalent cross-linking, the Factor IX/Bpacontaining peptides interact non-covalently with phospholipid membranes with characteristics indistinguishable from Factor IX (1-47).
Using gel filtration to assess covalent binding of 125 I-labeled Bpa-containing peptides to phospholipid, we investigated the nature of the peptide-phospholipid complex following photoactivation and after dissociation of non-covalently bound peptide to phospholipid by the removal of Ca 2ϩ with EDTA. The extent of covalent cross-linking using a photoactivatable reagent is related to the conditions of the reaction, the geometry of the labile group and the target, and the chemical reactivity of the photolabile compound and the target bonds. Therefore, the direct comparison of the relative amount of cross-linking with each peptide to phospholipid vesicles is critical for analysis. Factor IX (1-47)/Bpa 6 in the presence of phospholipid and Ca 2ϩ was irradiated at 350 nm for 20 min. EDTA was then added to remove non-covalently bound peptide from the vesicles. The mixture was then analyzed by gel filtration in the FIG. 3. The absolute proton chemical shift differences for ␣ protons between Factor IX (1-47) peptide structures. Top graph, the chemical shift differences between apoFactor IX (1-47) and Factor IX (1-47)-Mg 2ϩ . Bottom graph, the chemical shift differences between Factor IX (1-47)-Ca 2ϩ and Factor IX (1-47)-Mg 2ϩ . The chemical shift values reflect the structural environment of the protons. The bars below the abscissa indicate that the chemical shift differences are small for ␣ protons in the designated peptide region (apo/Mg 2ϩ residues 1-13: ␣H, 0.1 Ϯ 0.1 ppm; Ca 2ϩ /Mg 2ϩ residues 14 -47: ␣H, 0.1 Ϯ 0.2 ppm). The peptide regions designated by the bar are predicted to be structurally similar since the backbone ␣ and amide protons of amino acids undergo an average chemical shift change of approximately 0.4 ppm from the random coil value when they are incorporated in different secondary structures (Wishart et al., 1991). The arrows indicate large chemical shift differences for the ␣ protons in a region where all other ␣ protons show small chemical shift differences. Similar findings were observed for the backbone amide protons (apo/Mg 2ϩ residues 1-13: NH, 0.1 Ϯ 0.1 ppm; Ca 2ϩ /Mg 2ϩ residues 14 -47: NH, 0.2 Ϯ 0.2 ppm).

FIG. 4. Superimposition of the Factor IX (1-47)-Mg 2؉ structures with the Factor IX (1-47)-Ca 2؉ structure. Seven Factor IX
(1-47)-Mg 2ϩ structures (gray), arbitrarily chosen from the 15 calculated, are superimposed. For comparison, the lowest energy Factor IX (1-47)-Ca 2ϩ structure (black) is overlaid. Only the backbone atoms are displayed. The amino-terminal 11 residues of the Factor IX (1-47)-Mg 2ϩ peptide are not structurally defined, in contrast to the well defined loop for this region in the Factor IX (1-47)-Ca 2ϩ structure. Possibly due to intermolecular association, the Mg 2ϩ -induced linewidth broadening also blunts the magnitude and resolution of the NOE crosspeaks, reducing the number and accuracy of NOE distance measurements. Therefore, the resolution of the Mg 2ϩ -bound structure is not as high as that of the Ca 2ϩ -bound structure. During data analysis, we identified some NOE cross-peaks that were due to intermolecular interactions. These NOEs were not present when data were collected at lower peptide concentration or when the peptide was dissolved in aqueous solution including guanidine and urea. These additional NOE crosspeaks were not employed in defining the Factor IX (1-47)-Mg 2ϩ model. presence of EDTA for covalent peptide-phospholipid vesicle complexes. Under the conditions employed, about 10% of the added peptide was covalently cross-linked to phospholipid; this fraction of the peptide eluted in the void volume with phospholipid (Fig. 9A). If the ternary complex of peptide, phospholipid, and Ca 2ϩ was not irradiated, no covalent peptide-phospholipid complex was formed (Fig. 9D). If the peptide and phospholipid were irradiated but in the presence of either Mg 2ϩ or EDTA instead of Ca 2ϩ , no covalent peptide-phospholipid complex was formed (Fig. 9, B and C). These results indicate specific crosslinking of Factor IX (1-47)/Bpa 6 to phospholipid only under conditions in which the ternary peptide-phospholipid-Ca 2ϩ complex is formed. The Bpa at residue 6 is within or in close proximity (Ͻ3 Å) of the phospholipid bilayer. This result directly demonstrates that leucine 6 in Factor IX is a component of the phospholipid binding site.
Parallel experiments were performed using Factor IX (1-47)/ Bpa 9. The results are similar to those of Factor IX (1-47)/Bpa 6, and approximately 10% of Factor IX (1-47)/Bpa 9 was specifically cross-linked to phospholipid following photoactivation (Fig. 9). Thus, phenylalanine 9 is within or in close proximity of the phospholipid bilayer in the Factor IX-phospholipid complex. In contrast, Factor IX (1-47)/Bpa 46 showed no specific cross-linking to phospholipid vesicles (Fig. 9, A-D). This indicates that although this peptide bound normally to phospholipid in the presence of Ca 2ϩ , the site of the Bpa at residue 46 is sufficiently removed from the phospholipid bilayer that no cross-linking, and thus no covalent peptide-phospholipid com-plex, can occur.
As an additional control, Factor IX (1-47) lacking the Bpa amino acid analog did not cross-link to phospholipid membranes upon irradiation (Fig. 9, A-D). Furthermore, none of the Factor (1-47) Bpa peptides cross-linked to phospholipid composed exclusively of phosphatidylcholine when irradiated in the presence of calcium ions (data not shown).

DISCUSSION
The vitamin K-dependent proteins are a unique class of membrane-binding proteins, interacting with membrane surfaces through the Gla domain in the presence of calcium ions (Furie and Furie, 1988). This domain exhibits multiple metal binding sites for calcium ions. Protein-membrane interaction is reversible. The removal of calcium ions leads to rapid separation of the protein from the membrane surface (Nelsestuen et al., 1976). Furthermore, this class of proteins has a requirement for acidic phospholipid, such as phosphatidylserine, as a component of the membrane surface. Speculations on the molecular details of the nature of the protein-membrane interface over the past 20 years have resulted in conflicting models. Some groups have proposed that the vitamin K-dependent blood clotting proteins bind to acidic lipid membranes through calcium ions that bridge ␥-carboxyglutamic acid residues on the protein to the phosphate head groups on the lipid membrane (Lim et al., 1977;Mann et al., 1982;Schwalbe et al., 1989). This paradigm requires that the side chains of the ␥-carboxyglutamic acid residues extend into the solvent and are available on the surface to link to the membrane through calcium ion bridges. A salient feature of this model is that ␥-carboxyglutamic acid participates as a contact residue in the protein-membrane interface. However, three significant experimental observations have argued against this interpretation. 1) We had proposed from NMR studies of a peptide fragment of prothrombin that the high affinity metal binding sites were internal and formed by intramolecular ␥-carboxyglutamic acid residues, both bound to a common metal ion (Furie et al., 1979); 2) Rhee et al. (1982) have shown that the hydration of metal ions bound by ␥-carboxyglutamic acids in prothrombin is not altered by the addition of phospholipid membranes, thus demonstrating the absence of metal-phosphate interaction in the ternary complex; 3) many groups have observed that excess concentrations of calcium ions do not disrupt protein-membrane interaction. Borowski et al. (1986) put forth an alternative model in which the metal ion binding sites are created by the ␥-carboxyglutamic acid residue side chains oriented toward the interior of the protein, thus with internal bound Ca 2ϩ ; this Ca 2ϩ -stabilized conformer expresses a phospholipid binding site that does not include ␥-carboxyglutamic acid at the protein-membrane interface. From the prothrombin fragment 1 crystal structure (Soriano-Garcia, 1992), a definitive structure of the protein-Ca 2ϩ complex has emerged. In this structure, the calcium ions are coordinated by oxygens of the carboxylate groups of ␥-carboxyglutamic acid. These calcium ions are, for the most part, not exposed to solvent, but rather form an internal array that stabilizes the folding of the NH 2 terminus of the prothrombin. Only a few of the ␥-carboxyglutamic acid side chains are available on the surface of the molecule. Our solution structure of the Factor IX Gla domain is consistent with this structure (Freedman et al., 1995b). However, the location of the phospholipid binding site of the vitamin K-dependent proteins remains speculative and has been based on the comparison of the Ca 2ϩ -bound structures and the metal-free structures (Freedman et al., 1995a(Freedman et al., , 1995bSunnerhagen et al., 1995).
A number of approaches have been taken to localize the phospholipid binding site on the vitamin K-dependent proteins.  Liebman et al. (1987) on Factor IX, and the structural data on the Factor IX (1-47) peptide obtained by NMR methods. According to the functional model, the apoGla domain binds a variety of divalent metal ions (closed triangles), including Mg 2ϩ , to produce a conformation that contains a metal ion-nonspecific structural determinant. This epitope is recognized by anti-Factor IX-Mg 2ϩ antibodies. Upon saturation with calcium ions (shaded circles), a second conformational transition occurs that results in the expression of a Ca 2ϩ -specific structural determinant (convex element) recognized by anti-Factor IX-Ca 2ϩ specific antibodies. Below each conformational state is the representative Factor IX Gla domain ribbon structure. In the absence of metal ions, the Factor IX (1-47) peptide is generally unstructured except for the presence of three well ordered regions, residues 6 -9, 18 -23, and 37-46. The intermediate structure contains additional helical regions that are separated by structured loops and turns. However, the amino-terminal region (residues 1-11) is generally unstructured, except for residues 6 -9, and phospholipid binding cannot occur. The second conformational transition is associated with the formation of a large amino-terminal loop that terminates in close proximity to the disulfide loop and contains a structured hydrophobic surface patch at residues 6 -10. Fragment 1, the NH 2 -terminal third of prothrombin, contains the phospholipid binding site of prothrombin (Gitel et al., 1973). The region between the Gla-aromatic amino acid stack domains and the serine protease domain, including epidermal growth factor and kringle domains, does not appear to play a significant role in mediating phospholipid binding in the vitamin K-dependent proteins (Kotkow et al., 1993). Proteolytic fragments of Factor X and prothrombin that contain the Gla domain and the aromatic amino acid stack domain bind to phospholipid vesicles (Schwalbe et al., 1989;Pollock et al., 1988), and synthetic fragments based on the Gla-aromatic amino acid stack domain of protein C and Factor IX bind to acidic phospholipid vesicles (Jacobs et al., 1994;Colpitts and Castellino, 1994). These results have led us and others to conclude that the Gla-aromatic amino acid stack domains constitute the unit for phospholipid binding. Zhang and Castellino (1994) have used site-specific mutagenesis to demonstrate that mutation of leucine 5 in protein C interferes with lipid binding, and Christianson et al. (1995) showed that the mutation of leucine 8 and to a lesser extent, phenylalanine 4, also affected lipid binding. We have mutated Arg-15 in prothrombin and observed significant decreases in binding affinity when a glycine is substituted at position 15 (Dietcher et al., 1994). Furthermore, mutation of  to aspartic acid abolishes high affinity phospholipid binding in human prothrombin (Ratcliffe et al., 1993). None of these experiments allow interpretation that a specific amino acid is located at the protein-membrane interface since alteration of side chains of amino acids that are located at the protein-membrane interface and mutation of amino acids that lead to a significant change in the tertiary structure of the Gla domain structure can both disrupt membrane binding. More recently, we have compared the structures of the Factor IX Gla domain in the presence and absence of Ca 2ϩ (Freedman et al., 1995a(Freedman et al., , 1995b. This analysis concluded that Ca 2ϩ -induced folding of polypeptide backbone linking the three well structured regions in the apo form correlate with the induction of phospholipid binding properties. The apo structure of the Factor X proteolytic fragment including the Gla, aromatic amino acid stack, and the first epidermal growth factor domains revealed somewhat more stable structure than the apo form of the Factor IX Gla domain (Sunnerhagen et al., 1995). Comparison of this structure to a homology model based upon the prothrombin fragment 1-Ca 2ϩ crystal structure (Soriano-Garcia et al., 1992) also suggested that Ca 2ϩ induces further folding of the NH 2 terminus of the Gla domain. Although a hydrophobic patch near the NH 2 terminus has been noted previously (Soriano-Garcia et al., 1992), there has been no direct evidence for the interaction of this region with acidic phospholipid vesicles.
To circumvent the problems of interpretation associated with site-directed mutagenesis and large changes in structure in the presence and absence of calcium ions, we determined the threedimensional structures of the Gla domain in the presence of calcium ions and in the presence of magnesium ions to identify the phospholipid binding site. Calcium and magnesium ions induce a conformational change that can be monitored by fluorescence quenching and the expression of neoantigens. Calcium ions induce unique neoantigens in addition to those present in the protein-Mg 2ϩ complex (Borowski et al., 1986;Liebman et al., 1987). Furthermore, despite the common metalinduced conformational change, the Ca 2ϩ -stabilized conformer binds to phospholipids but the Mg 2ϩ -stabilized conformer does not. Thus, we anticipated that the structural comparison of the Factor IX (1-47)-Ca 2ϩ and the Factor IX (1-47)-Mg 2ϩ complexes would reveal both regions of structure common to these two forms and regions of structure that were unique to each liganded form, allowing identification of the site on Factor IX that is required for phospholipid binding. We have now shown that the region of difference, required for phospholipid binding, includes the NH 2 terminus of Factor IX, from residues 1 to 11.
Chemical modification of the NH 2 terminus of prothrombin is blocked in the presence of calcium ions but not in the presence of magnesium ions or in the absence of metal ions (Welsch and Nelsestuen, 1988;Schwalbe et al., 1989). Chemical modification of the metal-free forms of the vitamin K-dependent proteins preclude phospholipid binding in the presence of Ca 2ϩ . These results suggest that the NH 2 terminus is buried in the   FIG. 7. A, chemical structure of Bpa. B, chemical synthesis of Factor IX (1-47) peptides containing benzoyl-L-phenylalanine. Bpa, a photoreactive homolog of phenylalanine, was incorporated into peptides using solid phase synthesis and Fmoc chemistry. Bpa was substituted for leucine 6 (Factor IX (1-47)/Bpa 6), for phenylalanine 9 (Factor IX (1-47)/Bpa 9), and for valine 46 (Factor IX (1-47)/Bpa 46). Bpa, q-q. In A-C, the sample was irradiated at 350 nm for 20 min; the sample in D was not irradiated. Non-covalent Ca 2ϩ -mediated proteinphospholipid binding was reversed by the addition of EDTA after irradiation. Bound and free peptide were separated by gel filtration. The percentage of total peptide that was cross-linked to phospholipid vesicles is shown on the y axis. Factor IX (1-47)/Bpa 6 (black); Factor IX (1-47)/Bpa 9 (gray); Factor IX (1-47)/Bpa 46 (striped); Factor IX (1-47) (white). presence of calcium ions, but is exposed in the absence of metal ions or in the presence of Mg 2ϩ . Indeed, the crystal structure of prothrombin fragment 1 and the NMR structure of Factor IX support this model for the Ca 2ϩ -bound form (Soriano-Garcia et al., 1992;Freedman et al., 1995b), and our current study supports this model for the Mg 2ϩ -bound form.
We have tested the hypothesis that residues 1-11 define the phospholipid contact site by localizing the lipid binding site on the peptide with a photocross-linking reagent. Residues 1-11 include a hydrophobic patch, noted by ourselves (Freedman et al., 1995b) and others (Soriano-Garcia et al., 1992;Colpitts and Castellino, 1994;Sunnerhagen et al., 1995), that might represent a component of the phospholipid binding site that inserts into the phospholipid bilayer. By introducing a photoactivatable cross-linker into this patch, we demonstrated the close proximity of the patch with the phospholipid bilayer.
p-Benzoyl-L-phenylalanine is an amino acid analog that has ideal properties as a cross-linking agent Williams and Shoelson, 1993). First, it allows facile incorporation directly into chemically synthesized peptides using standard Fmoc chemistry. Second, it is conveniently manipulated in ambient light, photolyzable upon irradiation at 350 nm, and causes minimal damage to the polypeptide. Third, the photo-activation of p-benzoyl-L-phenylalanine involves highly reactive carbene chemistry, thus allowing attack of weak C-H bonds with high selectivity and yield. Fourth, the sweep of the Bpa is limited to less than 3 Å, thus permitting identification of a specific site on the peptide that is proximal to the lipid membrane. Peptides containing the amino acid analog have been used previously by ourselves (Yamada et al., 1995) and others (Schoelson et al., 1993;Williams and Schoelson, 1993) to localize peptide binding sites on proteins. In the current work, we show that Bpa-containing peptides can be used to cross-link peptides to phospholipids as well as proteins.
Factor IX (1-47) containing Bpa exhibited phospholipid binding properties that were indistinguishable from Factor IX (1-47). Prior to irradiation and cross-linking, these peptides were shown to be similar to Factor IX (1-47) in their phospholipid binding properties. In the presence of Ca 2ϩ , they bound with the same K d , within experimental error; in the presence of EDTA or Mg 2ϩ , they did not bind to phospholipid vesicles. Thus, these peptides represent functional analogs of Factor IX (1-47), and the Bpa does not interfere with any of the measured properties. These observations justified the use of the Bpa-containing peptides as photoactivatable cross-linking reagents to localize the sites on Factor IX (1-47) that bind to phospholipid.
We conclude from this study that, in the presence of calcium ions, the NH 2 -terminal loop of Factor IX (and presumably the other vitamin K-dependent proteins) is a critical component of the phospholipid binding site. A space-filling model of the Glaaromatic amino acid stack domains is shown in Fig. 10. This phospholipid binding region extends from residues 1 to 11 (yellow). Since the three-dimensional structure of residues 6 -9 is conserved in the absence of metal ions (Freedman et al., 1995a), in the presence of calcium ions (Freedman et al., 1995b), and in the presence of magnesium ions (this study), either the conformation of residues from 1 to 5 and from 9 to 11 FIG. 10. The phospholipid binding site of Factor IX. The atomic coordinates of this binary complex were derived from two-dimensional NMR spectroscopy (Freedman et al., 1995b). The positions of atoms of residues 12-47 (white) are nearly identical in the Factor IX (1-47)-Mg 2ϩ and Factor IX (1-47)-Ca 2ϩ structures. Residues 1-11 (yellow and shades thereof) include the regions required for phospholipid binding. Hydrophobic residues (dark yellow), including leucine 6, phenylalanine 9, and valine 10, define a hydrophobic patch on the exterior of the protein that likely buries inside the phospholipid bilayer. The NH 2terminal tyrosine 1 (orange-yellow) interacts with Lys-22. Tyrosine 1 (Y1), leucine 6 (L6), phenylalanine 9 (F9), and valine 10 (V10) are indicated.
FIG. 11. Hypothetical model of the interaction of Factor IX and phospholipid membranes. In this model, the hydrophobic residues (black) that form the hydrophobic patch in the phospholipid binding site of Factor IX are buried in the phospholipid bilayer. Specific residues in the Gla domain interact with the phospholipid head groups and are responsible for the requirement for anionic phospholipids for effective binding. may be critical for phospholipid binding or, and more likely, the fold of the NH 2 -terminal amino acids 1-11, including the interaction of tyrosine 1 with Gla-21 and Lys-22, define the critical phospholipid binding surface. Based upon the crosslinking analysis, we propose that residues 6 -10, including leucine 6, phenylalanine 9, and valine 10, form a hydrophobic patch on the exterior of the protein that likely buries inside the lipid bilayer (Fig. 11). This patch represents at least part of the contact site, but electrostatic interactions involving charged amino acids may also be important (Atkins and Ganz, 1992). The interaction with unsaturated fatty acid chains of phospholipids has been described (Govers-Riemslag et al., 1994).
Membrane proteins, including families of integral membrane proteins and extrinsic membrane proteins, fall into structural classes that characterize their functional properties. The vitamin K-dependent, ␥-carboxyglutamic acid-containing proteins interact reversibly in the presence of calcium ions with acidic phospholipids on membranes following cell activation. The regulation of blood coagulation on cell surfaces involves the formation of membrane-bound protein complexes, including protein cofactors and the enzymes formed following zymogen activation (Furie and Furie, 1988). The structural motif responsible for Factor IX interaction with membrane surfaces is a general feature of the homologous Gla regions in this family of proteins.