Molecular View into Preferential Binding of the Factor VII Gla Domain to Phosphatidic Acid

Abstract

Factor VII (FVII) is a serine protease with a key role in initiating the coagulation cascade. It is part of a family of vitamin K-dependent clotting proteins, which require vitamin K for formation of their specialized membrane-binding domains (Gla domains). Membrane binding of the FVII Gla domain is critical to the activity of FVII, mediating the formation of its complex with other clotting factors. While Gla domains among coagulation factors are highly conserved in terms of amino acid sequence and structure, they demonstrate differential binding specificity toward anionic lipids. Although most Gla domain-containing clotting proteins display a strong preference for phosphatidylserine (PS), it has been demonstrated that FVII and protein C instead bind preferentially to phosphatidic acid (PA). We have developed the first model of the FVII Gla domain bound to PA lipids in membranes containing PA, the highly mobile membrane mimetic model, which accelerates slow diffusion of lipids in molecular dynamics simulations and therefore facilitates the membrane binding process and enhances sampling of lipid interactions. Simulations were performed using atomic level molecular dynamics, requiring a fixed charge to all atoms. The overall charge assigned to each PA lipid for this study was −1. We also developed an additional model of the FVII Gla domain bound to a 1:1 PS/PC membrane and compared the modes of binding of PS and PA lipids to FVII, allowing us to identify potential PA-specific binding sites.

Graphical Abstract

It is well established that the lipid composition of the platelet surface has a profound effect on blood clotting.^1–4 To initiate clotting, negatively charged lipids are exposed to the outer surface of the plasma membrane, allowing for recruitment of coagulation factors. Factor VII (FVII) is the catalytic subunit of the enzyme that triggers the extrinsic cascade of blood coagulation.^1,5 The formation of the extrinsic complex, consisting of tissue factor, FVIIa, and FX, is integral to generation of the explosive burst of thrombin that allows a clot to form.^5,6

FVII membrane binding depends on a specialized, post-translationally modified domain known as the Gla domain. Coagulation factors like FVII that contain Gla domains are termed vitamin K-dependent, as this vitamin is necessary to synthesize γ-carboxyglutamic acid (Gla) residues in these domains.⁸ Gla domain-containing coagulation factors are members of the extrinsic, intrinsic, and anticoagulant pathways.^5,9 The negatively charged Gla residues allow for coordination of multiple Ca²⁺ and Mg²⁺ ions, which are critical for the correct folding of the structure and its interaction with negatively charged surfaces.^10,11 The Gla domains of vitamin K-dependent coagulation factors are structurally and sequentially similar to each other (Figure 1), but they exhibit widely varying affinities for anionic lipids.^12–16 The FVII Gla domain (FVII-Gla), for example, is structurally similar to the Gla domains of factor X (FX-Gla) and prothrombin, but FVII has only weak affinity for phosphatidylserine (PS); FX and prothrombin show strong affinities for this lipid. FVII and protein C, on the contrary, have been found to have considerably higher affinities for membranes containing phosphatidic acid (PA) than for those containing PS. For example, FVIIa has been shown to exhibit increased enzymatic activity when bound to liposomes containing PA.¹⁷

Figure 1.

FVII-Gla, which is structurally and sequentially similar to other coagulation factor Gla domains, binds preferentially to PA-containing lipid bilayers. (A) Structure of the FVII Gla domain.⁷ Key elements shared by all Gla domains are highlighted. Specialized Gla residues each with a −2 charge are colored gold. The ω-loop is colored magenta. Hydrophobic “keel” residues, which penetrate into the membrane core, are colored light blue. Ca²⁺ ions are colored green. (B) PS and PA lipid structures. FVII shows preferential binding to PA-containing liposomes as compared to those with PS. Phosphorus atoms are colored gold, oxygen atoms red, carbon atoms beige, and nitrogen atoms blue. Hydrogen atoms have been omitted for the sake of simplicity. (C) Comparison of sequences of Gla domains from various coagulation factors. Gla residues (γ) in the sequences are highlighted in gold, keel residues in blue, and ω-loop residues in magenta.

PA is a rare anionic lipid, making up approximately 1–2% of the cell membrane, compared to 2–10% for PS.¹⁸ PA is the simplest anionic phospholipid and contains a single negatively charged phosphate group in its structure, although the overall charge can vary from −1 to −2 at physiologic pH (Figure 1). PS carries a net charge of −1 but contains multiple charged groups in its structure, a positively charged amino group, a negatively charged phosphate group, and a negatively charged carboxy group.

Here, we use molecular dynamics simulations to study the membrane binding of FVII-Gla to PA-and PS-containing lipid bilayers [with the balance being phosphatidylcholine (PC)]. A number of MD studies have previously probed Gla domain lipid interactions using full lipid models^10,15,19 but did not sufficiently reproduce spontaneous membrane binding of the studied domains. We use a specialized membrane model, termed the highly mobile membrane mimetic (HMMM), in which lipid tails are truncated and replaced with a solvent, thus allowing for substantially improved lipid diffusion, enhancing the sampling of lipid–protein interactions, and capturing their spontaneous binding.²⁰ This method has been used previously in a wide variety of membrane systems,^20–40 including FX-Gla and FVII-Gla membrane binding³² to 100% PS bilayers.²⁰ To the best of our knowledge, this is the first simulation study to examine binding of FVII-Gla to PA lipids. As the atomic level MD model requires a fixed charge to be assigned to all atoms, PA lipids were assigned a −1 charge as the most likely at physiological pH. We generate 48 independent systems of FVII-Gla over PA/PC and PS/PC membranes and allow FVII-Gla to spontaneously bind from solution. Each membrane patch is either 1:1 PA/PC or 1:1 PS/PC, consistent with concentrations used in experimental studies.¹⁷ While this is much larger than the concentrations typically found in physiological membranes, a 50% concentration of anionic lipid is used to reproduce the experimental system while also increasing the frequency of sampling of interactions of the anionic lipid with FVII-Gla. This allows us to characterize the membrane binding mode of FVII-Gla and to assess possible molecular level sources of its differential affinity for PA- and PS-containing membranes.

METHODS

Spontaneous Membrane Binding Simulations.

A crystal structure of FVII-Gla from Protein Data Bank entry 1DAN⁷ was used as the starting structure for all Gla domain simulations. Using the PSFGEN plugin of VMD,⁴¹ an N-terminal ammonium group and a C-terminal N-methylamide C-terminal group were added, along with hydrogen atoms. The neutral N-methylamide group was chosen because the C-terminus of FVII-Gla in the full-length FVII protein would be connected to the first epidermal growth factor (EGF) domain, and a C-terminal patch with a default charge would introduce an artificial charge into this region. FVII-Gla was next placed in a 60 Å × 60 Å × 60 Å water box using the SOLVATE plugin of VMD and neutralized with Na⁺ ions added using the AUTOIONIZE plugin of VMD. This solution system was then equilibrated for 1 ns after 2000 steps of energy minimization.

Our HMMM membrane construct was used in the binding simulations to increase lipid dynamics and allow for more frequent sampling and insertion of the Gla domain into the membrane. Three independent 1:1 PA/PC HMMM lipid patches as well as three independent 1:1 PS/PC HMMM patches were generated. PA lipid headgroups were assigned a charge of −1. Each patch was 50 Å × 50 Å × 114 Å in size and was constructed and solvated using the HMMM Builder³³ in CHARMM-GUI.^42,43

Nine independent orientations of FVII-Gla (a rotation of 15° in the xz or yz plane) were generated for each of the three PA/PC and three PS/PC membrane patches. In the beginning of each simulation, FVII-Gla was placed at least 3 Å above the membrane patch. This resulted in generation of 27 unique FVII-Gla systems for each lipid composition. Each system was neutralized with Na⁺ ions using the AUTOIONIZE plugin of VMD. The final systems each included approximately 24 000 atoms. The z-position of HMMM short-tail lipids was restrained harmonically (k = 0.1 kcal mol⁻¹ Å⁻² applied to C21 and C31 lipid-tail atoms) throughout the simulations to more closely reproduce the natural distribution of the headgroup atoms. Each system was then simulated for 100 ns collecting 2.7 μs of sampling for both FVII-Gla PS/PC and FVII-Gla PA/PC systems.

Conversion of HMMM Membranes into Conventional Full Membranes.

To ensure that the final FVII-Gla membrane-bound model was not unduly affected by the use of the HMMM as opposed to a full-tail lipid construct, all bound systems were converted to conventional (full-tail) membranes, using a PDB generated from the final frame of each 100 ns membrane binding HMMM simulation. We used 1-palmitoyl-2-oleoyl (PO) tails to produce complete POPC, POPA, and POPS lipids. Conversion to a full-tail membrane was performed only for the systems in which the keel penetrated below the level of phosphorus atoms in the membrane. A full-tail lipid was overlaid onto each short-tail lipid, and coordinates of the missing atoms were then copied from the full-tail lipid, thus allowing headgroup interactions obtained from the HMMM simulations to be preserved. The layer of 1,1-dichloroethane (DCLE) solvent molecules was then removed from the center of the HMMM bilayer. After a 2500-step energy minimization, each membrane-bound FVII-Gla system was simulated in a full membrane for an additional 100 ns.

Simulation Protocols.

All simulations were performed using NAMD^44,45 and the CHARMM36⁴⁶ force-field parameters. Parameters of Gla residues were those previously developed by our lab¹⁰ with atom types renamed as appropriate for the conversion from CHARMM27. The NPT conditions and TIP3P water model were used for all simulations. Constant pressure was maintained at a target of 1 atm using the Nosé–Hoover–Langevin piston method.⁴⁷ Constant temperature was maintained by Langevin dynamics with a damping coefficient, γ, of 1 ps⁻¹ applied to all atoms.⁴⁸ All simulations used a time step of 2 fs and a temperature of 310 K. The particle mesh Ewald (PME) method⁴⁹ was used for long-range electrostatic calculations with a grid density of >1 Å⁻³. Nonbonded interactions were cut after 12 Å with a smoothing function after 10 Å. The system setup, visualization, and analysis were carried out using VMD.⁴¹

Analysis.

The depth of insertion of the Gla domain into the membrane was assessed by calculating the z-components of the centers of mass (COM) of these groups, individually, relative to the midplane of phosphate groups (phosphorus atoms and four bonded oxygen atoms) of lipids: (1) backbone atoms in residues 4, 5, and 8 in the keel of FVII-Gla and (2) Ca²⁺ ions 1–7 (Ca1–Ca7, respectively). Similarly, the z-components of lipid carboxy groups (carbon and two bonded oxygen atoms) and lipid amino groups (nitrogen and three bonded hydrogen atoms) were also calculated relative to the phosphate plane.

The orientation of the Gla domain relative to the membrane plane was assessed using two angles defined in two perpendicular planes, termed “port-to-starboard” and “bow-to-stern” (Figure 2). The bow-to-stern angle is the angle between the membrane normal (z-axis) and the vector connecting the COM of Ca1–Ca3 and Ca4–Ca6. These are the same measures as previously used to analyze the orientation of the domain on the membrane in our previous study of FX-Gla membrane binding.³² The port-to-starboard angle is the angle between the membrane normal and the vector connecting the COM of backbone atoms in residues 13, 14, 29, and 30 and the COM of backbone atoms in residues 19 and 22–24. These two sets of residues represent opposite ends of the two helices directly above the level of the coordinated Ca²⁺ ions. Thus, the bow-to-stern angle measures the angle between the stern and bow ends of FVII-Gla and the membrane normal, while the port-to-starboard angle represents the tilt of the port and starboard ends relative to the membrane normal.

Figure 2.

Illustration of the “bow-to-stern” (left) and “port-to-starboard” (right) angles used to compare the orientation of FVII Gla domains upon binding to lipid bilayers. In the left panel , the bow atoms, Ca1–Ca3, are colored pink and the stern ones, Ca4–Ca6, purple. The bow-to-stern angle (θ) is the angle between the bow-to-stern vector and membrane normal (z-axis). In the right panel , the port atoms are the backbone atoms of protein residues 19 and 22–24, which are colored pink, while the starboard atoms, backbone atoms of protein residues 13, 14, 29, and 30, are colored purple. The port-to-starboard angle (θ) is the angle between the port-to-starboard vector and membrane normal.

Contacts between lipid headgroup charged moieties and FVII-Gla (bound Ca²⁺ or protein residues) were calculated on the basis of distance cutoffs that were determined using radial pair distribution function calculations, as in our previous study of the FX Gla domain.³² The number of contacts here is the number of atoms on one group (e.g., a carboxy group) within a specified cutoff of the second group (e.g., a specific protein residue). The distances between all protein and Ca²⁺ atoms and phosphate, amino, and carboxy atoms were calculated, and distances equal to or less than the cutoff were determined to be a contact. The radial pair distribution function was calculated separately for distances between the residue type of interest (protein or Ca²⁺ ions) and the lipid headgroup moiety of interest (carboxy, amino, or phosphate). The calculation sampled every 10th frame of the 100 ns full-membrane simulations. For protein residues and Ca²⁺ ions, all atoms were included in the contact calculation. Cutoffs determined were 2.50 Å for phosphate–Ca²⁺, 3.00 Å for carboxy–Ca²⁺, 2.25 Å for phosphate–protein, 2.50 Å for carboxy–protein, and 2.25 Å for amino–protein contacts.

RESULTS

Binding of FVII-Gla to PS- and PA-Containing Lipid Bilayers.

We performed 54 independent HMMM simulations to capture spontaneous binding of FVII-Gla to anionic membranes, 27 copies of PA/PC (1:1) membranes and 27 PS/PC (1:1) membranes. The FVII Gla domains bound successfully to 24 of 27 PA/PC bilayers, while in only 17 of 27 PS/PC systems did FVII-Gla bind successfully to the membrane. Despite the multiple copies used, the number of simulations is not large enough for this result to be statistically significant. As described in Methods, binding was deemed successful if the keel residues of FVII-Gla penetrated beneath the phosphate level in the lipid headgroups. Figure 3 shows examples of poses for binding of FVII-Gla to PA/PC and PS/PC bilayers, as well as a graphical representation of the membrane depth of each FVII-Gla over the course of both HMMM and full-membrane simulations. As the simulation system can fluctuate during full-membrane simulations, before the analysis, the system was shifted to cancel any drift along the membrane normal (adjusting the COM of the phosphorus atoms to z = 25 Å) to allow ease of comparison. As shown in Figure 3, once the keel of FVII-Gla penetrated below the phosphate level, it remained stably bound for the remainder of the simulation.

Figure 3.

FVII-Gla binds spontaneously to PS/PC and PA/PC membranes. (A) Representative FVII-Gla bound to a PS/PC full membrane. Ca²⁺ ions are colored green, and Gla residues orange. The keel backbone is colored dark blue, and keel residues are colored light blue. For lipids, carbon is colored beige, oxygen red, nitrogen blue, and phosphorus gold. (B) Representative FVII-Gla bound to a PA/PC full membrane. Same color scheme as in panel A. (C) Membrane depth of insertion of FVII-Gla measured by the z-position of the COM of Gla Ca²⁺ ions (top) and the COM of keel CA atoms (bottom) relative to the membrane. Different shades of blue (for PS/PC simulations) and green (for PA/PC simulations) are used to show different traced COMs of the FVI-Gla components. Orange shows the position of the phosphate layer used as the reference (z = 25 Å), and maroon shows the z-component of the positions of the choline nitrogens in lipids. The dashed vertical line indicates the transition from HMMM simulations to full-membrane simulations.

Interestingly, we found a deeper penetration of FVII-Gla into PA/PC membranes tnan into PS/PC membranes. When FVII-Gla was bound to the PA/PC membranes, the keel height (averaged over the last 10 ns of the full-membrane simulations) was 17.7 ± 1.0 Å and the Gla Ca²⁺ ions were at 27.2 ± 0.8 Å, where the average height of phosphate atoms in the membrane was 25 Å. When FVII-Gla was bound to the PS/PC lipids, the average keel height was 19.7 ± 1.7 Å and the Ca²⁺ ions were at 29.3 ± 1.7 Å. FVII-Gla is thus binding with approximately 2.0 Å deeper penetration into PA/PC bilayers than into PS/PC bilayers, as well as to a more converged depth. This is particularly notable as there was more sampling for FVII-Gla when it was bound to PA/PC.

The orientation of the FVII Gla domain relative to the membrane was assessed using the bow-to-stern and port-to-starboard tilt angles previously developed in our study of FX-Gla membrane binding.³² As described in Methods, the bow-to-stern tilt assesses the angle between the membrane normal vector and a vector through FVII-Gla Ca²⁺ ions 1–6. The port-to-starboard tilt, on the contrary, is the angle between the membrane normal and a vector that runs parallel to the α-helices in the Gla domain closest to the keel. Figure 4 shows the time series of these angles throughout the HMMM and full-membrane simulations.

Figure 4.

Orientation of the membrane-bound FVII-Gla relative to the lipid bilayer. (A) Orientation over the course of the HMMM and conventional membrane simulations. The bow-to-stern orientation is shown in the top panels, and the port-to-starboard orientation in the bottom panels. Traces for binding of FVII-Gla to PS/PC bilayers are shown in different shades of blue, and those for binding to PA/PC bilayers are shown in different shades of green. The vertical dashed line shows the transition from the HMMM simulations to full-membrane simulations. (B) Histograms of bow-to-stern (top) and port-to-starboard (bottom) angles calculated for the last 10 ns of simulations with full membranes. All properly bound trajectories are included in the data. Histograms for FVII-Gla bound to PS/PC membranes are colored blue, and histograms for FVII-Gla bound to PA/PC membranes green.

In addition to the difference in the depth of membrane insertion, the orientational flexibility of FVII-Gla upon binding to the membrane is also different between the PA/PC and PS/PC membranes (Figure 4). We found average “port-to-starboard” tilt angles (averaged over the last 10 ns of the full-membrane simulations) of 109 ± 15° for the FVII Gla domain when bound to PA/PC membranes and 105 ± 14° when bound to PS/PC bilayers. The bow-to stern tilt angle for the PA/PC membranes was 88 ± 5°, while for FVII-Gla bound to the PS/PC membranes, it was 91 ± 13°. While the average orientation between the PA-bound and PS-bound Gla domains is similar, PS-bound Gla domains took on a wider range of bow-to-stern tilts in the same 10 ns of simulation. When bound to PA-containing bilayers, the standard deviations of both port-to-starboard and bow-to-stern tilts are similar to those found for the FX Gla domain described previously.³² Although the units may not be directly compared, the orientation of the protein on the surface of the membrane represents a softer degree of freedom that is expected to fluctuate more widely compared to the depth of penetration into the membrane.

FVII-Gla Lipid Interactions in PA/PC and PS/PC Membranes.

To further compare the interactions of the FVII Gla domain with the two studied anionic membranes, we quantified individual interactions between the residues in the FVII Gla domain and any charged phospholipid moieties during the full-membrane simulations. We assessed contacts between either protein residues or bound Ca²⁺ ions and the phosphate, carboxy, or amino moieties of lipids. The average contact numbers between these lipid moieties and individual FVII-Gla residues are shown in Figure 5.

Figure 5.

Specific contacts between lipids and FVII-Gla. (A) FVII-Gla residues most frequently contacting PA phosphates are colored yellow. (B) The FVII-Gla residues with most frequent contacts to PS lipid groups are highlighted, with red indicating the most frequent contacts with PS carboxy groups, yellow the most frequent contacts with phosphate groups, and blue the most frequent contacts with PS amino groups. (C–F) Histograms of average contacts with lipid charged groups. Average contact values are calculated for the entire 100 ns of the full-membrane simulations. (C) Histogram of contacts of FVII-Gla/Ca²⁺ with lipid charged groups in 1:1 PA/PC membranes. Orange indicates contact with the phosphate group of either PA or PC lipid. (D) Histogram of contacts of Ca²⁺ residue with lipid charged groups in 1:1 PS/PC membranes. Orange indicates contact with PS or PC phosphate groups, and red contact with PS carboxy groups. (E) Histogram of contacts of protein residue with lipid charged groups in 1:1 PA/PC membranes. Orange indicates contact with a phosphate group of either PA or PC lipid. (F) Histogram of contacts of protein residue with lipid charged groups in 1:1 PS/PC membranes. Orange indicates contact with a PS or PC phosphate group, red contact with a PS carboxy group, and blue contact with a PS amino group.

As the only charged group that can form strong electrostatic interactions with the FVII Gla domain in the PA/PC bilayers is the lipid phosphate, significant interactions are found between both of these groups and the positive protein residues or the bound Ca²⁺ ions of the FVII Gla domain in the PA/PC simulations. In particular, Ca1 and Ca6 (individual Ca²⁺ ions that are tightly bound to the FVII Gla domain) interact frequently with the phosphate groups, with average contact numbers of 1.4 and 1.2, respectively (Figure 5), indicating that they interact with one or more phosphate groups for the majority of the simulation. In the simulations with PS/PC membranes, the positive sites of the protein instead interact largely with the carboxy groups of lipids, which can be attributed to the larger exposure of the carboxy groups in PS lipids, compared to the buried position of their phosphate groups. The overall numbers of contacts for Ca1 and Ca6 with PS and PC phosphate are also generally decreased to 1.0 and 0.8, respectively (combined for both phosphate and carboxy contacts).

A possible explanation for the fewer number of interactions in PS-containing bilayers may lie in an examination of representative binding modes. Figure 6 illustrates representative lipid–Ca²⁺ interactions for PA and PS lipids at the end of 100 ns full-membrane simulations. Representative structures used had average binding depths, as well as average port-to-starboard and bow-to-stern orientations. The interactions shown were stable for the entirety of the 100 ns full-membrane simulation. For the PA/PC membrane, three PA lipids and the PC lipid contribute to Ca²⁺ binding, with two lipids interacting with each of Ca²⁺ residues 1 and 6. In contrast, three PS lipids interact with Ca²⁺ ions, one each with Ca²⁺ 6 and 7, and the third coordination with both Ca²⁺ residues 1 and 2. While multiple phosphate groups coordinate with Ca²⁺ ions in the PA/PC binding simulation, no more than one PS lipid carboxy group interacts simultaneously with each Ca²⁺ ion. It is possible that the truncated PA structure allows both PA and PC phosphate to pack more closely around the Ca²⁺ ions, thus increasing the average number of total lipid contacts for Ca²⁺ ions in PA/PC membrane simulations.

Figure 6.

Characteristic lipid–Ca²⁺ interactions from full-membrane simulations for FVII-Gla. See Methods for distance cutoffs. Ca²⁺ ions are colored green, oxygen atoms red, nitrogen atoms blue, and phosphorus atoms gold. Lipid carbon atoms are colored beige. The ω-loop of the Gla domain is colored magenta, and keel residues are colored light blue. Remainder of FVII-Gla shown in silver. (A) Lipids interacting with FVII-Gla Ca²⁺ in the PA/PC membrane binding simulation. One PA and one PC interact with Ca6 through their phosphate groups, while two PA lipids interact with Ca1. (B) Lipids interacting with FVII-Gla Ca²⁺ in the PS/PC membrane binding simulation. One PS lipid each interacts with Ca6 and Ca7, while one PS lipid interacts with Ca1 and Ca2 simultaneously. All PS lipids interact through their phosphate groups.

For protein–lipid interactions, residues placed farther from the membrane are found to have a higher number of carboxy interactions when in PS/PC membranes than do those on the ω-loop. In the PA/PC membranes, the protein residues that have significant interactions with the lipid phosphate groups include Arg9, Arg15, Arg28, and Lys32. In the PS/PC membranes, only Arg9 and Arg15 have a majority of lipid interactions that are charged group interactions with phosphate, while Arg28 and Lys32 have a majority of interactions with carboxy groups. This may be due to the shallower binding depth of FVII-Gla in PS/PC membranes than in PA/PC membranes, which would make these two relatively membrane-distant residues less accessible to phosphate groups.

Representative membrane-bound configurations of FVII-Gla are shown in Figure 7, where important lipid–protein interactions are highlighted. In the snapshot shown in Figure 7 for FVII-Gla bound to PA/PC membranes, the basic side chain of Arg28 directly interacts with the phosphate group of a PC while the Lys32 residue is in contact with a PA phosphate group. In the representative snapshot shown in Figure 7A, PS carboxy groups interact with both Arg28 and Lys32. In panels C and D of Figure 7, Arg9 is shown to bind to PA phosphate and PS phosphate, respectively.

Figure 7.

Characteristic lipid–protein interactions from simulations with full-length fatty acyl chains. Ca²⁺ ions are colored green, oxygen atoms red, nitrogen atoms blue, and phosphorus atoms gold. Lipid carbon atoms are colored beige. The ω-loop is colored magenta, and keel residues are colored light blue. The remainder of FVII-Gla is colored silver. Four simulation snapshots are shown in panels A–D. (A) In PA/PC membranes, R28 and K32 interact with PC and PA lipid phosphate groups. (B) In PS/PC membranes, R28 and K32 interact with PS lipids through their carboxy groups. (C) In PA/PC membranes, K9 interacts with PA lipids through the phosphate group. (D) In PS/PC membranes, R9 interacts with the phosphate of PS while the carboxy group of the same lipid interacts with Ca6.

In examining sequence conservation in Gla domains among these residues, we find a high degree of conservation for residue 28, which is Arg for FVII, FX, and FIX and Lys for protein C. Thus, the residue is a basic amino acid for both PS-specific FX and FIX and PA-specific FVII and protein C. This appears to be a fairly nonspecific interaction that does not seem to be an element of the lipid-specific affinity to PA for FVII-Gla. Similarly, residue 9 is positively charged Lys for FX and Arg for prothrombin, FVII, and protein C. In our previous study of FX, Lys9 interacted like Arg9 for FVII does here, providing a phosphate anchor. Residue 15 is a conserved Arg for all five Gla domains we assessed (Figure 1). For residue 32, however, sequences are clearly different between PA-specific and PS-specific Gla domains. In FX, FIX, and prothrombin, residue 32 is a negatively charged Gla residue. In PA-specific FVII and protein C, however, it is either a positively charged Lys or a polar Gln residue. While electrostatically, the positive charge of a lysine can interact favorably with both lipid phosphate (present in both PA and PS) and lipid carboxy groups (present in only PS), judged by the sequence, the Lys in this location in the FVII Gla domain is a strong candidate for mediating the preferential binding to PA.

DISCUSSION AND CONCLUDING REMARKS

Experimental measurements have established improved binding of FVII to PA-containing membranes relative to PS-containing membranes.¹⁷ To investigate the molecular basis of this lipid effect, we have used molecular dynamics simulations to capture spontaneous binding of the FVII Gla domain to PA/PC (1:1) membranes and compared the process to that in PS/PC (1:1) membranes. Our analysis suggests that FVII-Gla binds the membrane with deeper penetration into PA/PC lipid bilayers than into PS/PC ones. FVII-Gla keel residues penetrated ~2.0 Å deeper below the level of the lipid phosphates when bound to PA-containing bilayers as compared to PS-containing bilayers. The binding of the FVII Gla domain to the PA-containing membranes was also more homogeneous than in PS-containing membranes. The standard deviation in the z-position (membrane normal) was narrower for the PA-associated FVII Gla domains than for the PS-associated ones. In addition, the standard deviations of the tilt angles (orientation) exhibited by the FVII Gla domain were also narrower for the PA/PC simulations for the “bow-to-stern” angle. FVII-Gla also interacts with different phospholipid moieties in the PA/PC bilayers and in the PS/PC bilayers. In the PA-containing bilayers, FVII-Gla residues interacted primarily with the phosphate groups and with a higher average number of lipid groups. This was particularly notable when examining contacts between the lipid charged groups and FVII-Gla Ca²⁺ions.

A possible explanation for the obtained findings may be the differential steric hindrance of the two bilayers. PA, with its truncated phosphate headgroup, is quite short compared to PS, which is more comparable in headgroup size to PC. The bulky choline headgroup is likely to sterically hinder the access of charged FVII-Gla residues to available phosphate moieties. In the PA/PC membranes, the steric hindrance incurred by the choline groups may be reduced, as the truncated PA headgroup allows it more space to move within the headgroup layer. FVII-Gla Ca²⁺ ions are thus able to penetrate more deeply into the membrane and coordinate with neighboring PC phosphate groups. In the PS/PC bilayers, the carboxy groups of the PS lipids are more available to coordinate the Ca²⁺ ions but may be unable to coordinate charged FVII-Gla residues as effectively. This would explain the narrower range of bow-to-stern tilts visited in the PA-containing bilayers and why the tilt is dependent on the relative position of the Ca²⁺ ions.

The positions of the tightly bound Ca²⁺ ions in Gla domains are structurally similar, however. What differences could account for the differential affinity between FVII and FX? One notable sequence difference between the FVII Gla domain and other Gla domains, which are generally more PS-specific, lies in the keel residues that penetrate into the core of the membrane (see Figure 1). At residue 8, both FVII and protein C, which have high PA affinities, have a Leu. In FX and prothrombin, the equivalent residues are Met and Val, respectively, which are bulkier hydrophobic residues. The other two side chains that make up the keel are otherwise identical (Phe and Leu) in these three proteins. The PA-preferential Gla domains thus have a somewhat less bulky keel residue than PS-preferential ones, which may necessitate a greater binding depth for equivalently strong hydrophobic interactions.

A limitation in our study is the fact that PA lipids are modeled with a fixed single negative charge. As one of the deprotonation pK_a values for PA is close to physiological pH,^50,51 it is possible that in vivo PA exhibits a mix of −1 and −2 charged states, especially upon binding to the FVII Gla domain (or other largely charged species). This aspect has not been taken into account in our fixed-charge simulation system. Our conclusions are also limited by use of a simplified two-lipid bilayer system. Despite the limitations imposed by the model, we believe the molecular insights into lipid–protein interactions provided by the study can be used as a basis for further experimental exploration of protein–lipid modulation of FVII and other clotting proteins.