Effects of Membrane Charge and Order on Membrane Binding of the Retroviral Structural Protein Gag

ABSTRACT

The retroviral structural protein Gag binds to the inner leaflet of the plasma membrane (PM), and many cellular proteins do so as well. We used Rous sarcoma virus (RSV) Gag together with membrane sensors to study the principles governing peripheral protein membrane binding, including electrostatics, specific recognition of phospholipid headgroups, sensitivity to phospholipid acyl chain compositions, preference for membrane order, and protein multimerization. We used an in vitro liposome-pelleting assay to test protein membrane binding properties of Gag, the well-characterized MARCKS peptide, a series of fluorescent electrostatic sensor proteins (mNG-KRn), and the specific phosphatidylserine (PS) binding protein Evectin2. RSV Gag and mNG-KRn bound well to membranes with saturated and unsaturated acyl chains, whereas the MARCKS peptide and Evectin2 preferentially bound to membranes with unsaturated acyl chains. To further discriminate whether the primary driving force for Gag membrane binding is electrostatic interactions or preference for membrane order, we measured protein binding to giant unilamellar vesicles (GUVs) containing the same PS concentration in both disordered (Ld) and ordered (Lo) phases. RSV Gag and mNG-KRn membrane association followed membrane charge, independent of membrane order. Consistent with pelleting data, the MARCKS peptide showed preference for the Ld domain. Surprisingly, the PS sensor Evectin2 bound to the PS-rich Ld domain with 10-fold greater affinity than to the PS-rich Lo domain. In summary, we found that RSV Gag shows no preference for membrane order, while proteins with reported membrane-penetrating domains show preference for disordered membranes.

IMPORTANCE Retroviral particles assemble on the PM and bud from infected cells. Our understanding of how Gag interacts with the PM and how different membrane properties contribute to overall Gag assembly is incomplete. This study examined how membrane charge and membrane order influence Gag membrane association. Consistent with previous work on RSV Gag, we report here that electrostatic interactions provide the primary driving force for RSV Gag membrane association. Using phase-separated GUVs with known lipid composition of the Ld and Lo phases, we demonstrate for the first time that RSV Gag is sensitive to membrane charge but not membrane order. In contrast, the cellular protein domain MARCKS and the PS sensor Evectin2 show preference for disordered membranes. We also demonstrate how to define GUV phase composition, which could serve as a tool in future studies of protein membrane interactions.

INTRODUCTION

The retroviral structural protein Gag provides the primary driving force for virus assembly at the inner leaflet of the plasma membrane (PM), and Gag alone is sufficient to assemble into virus-like particles (VLPs) in cells (1) and also in vitro (2). Gag is synthesized in the cytosol and then is targeted to the PM by the N-terminal domain, MA. How Gag interacts with the inner leaflet of the PM and how the biophysical properties of the PM contribute to Gag assembly are not fully understood. Gag exploits one or more mechanisms governing binding of peripheral proteins to membranes, which include electrostatic interactions, hydrophobic interaction, recognition of specific lipid headgroups, protein multimerization, sensitivity to phospholipid acyl chain compositions, and preference for membrane order (3).

Prior work in the field has provided a number of important clues to how Gag binds to membranes. First, most retroviral Gag proteins have a cluster of basic residues in MA that interact with acidic lipids in the inner leaflet (4). Mutating these residues results in reduction of Gag membrane association and of virus release (5–7). In vitro, the negatively charged lipids phosphatidylserine (PS) or phosphatidylinositol 4,5-bisphosphate [PI(4,5)P₂] are crucial for Gag or MA membrane association (8–13). Second, the Gag proteins of most retroviruses, like HIV-1, are modified by addition of the 14-carbon fatty acid myristate at the N-terminal glycine residue. Mutating this residue blocks myristoylation and also membrane binding and virus budding (14–16). A few Gag proteins, like that of Rous sarcoma virus (RSV), are not fatty acylated. Third, in at least some retroviruses, like HIV-1, MA has a specific binding pocket for the PI(4,5)P₂ headgroup (17, 18), and both Gag-PM binding and virus budding are reduced in PI(4,5)P₂-depleted cells (8, 17). However, similar experiments for RSV Gag showed inconsistent results (12, 19, 20). For the PI(4,5)P₂ effect, the relative roles of electrostatic attraction and specific headgroup binding are uncertain. Fourth, Gag binds to membranes in a cooperative fashion, with multimerization enhancing this interaction (21). RSV and HIV-1 MA dimers (21–23) and RSV MA hexamers (21) bind to membranes much more avidly than monomeric MA, both in vivo (for HIV-1) and in vitro (RSV and HIV-1). Finally, HIV-1 Gag associates preferentially with membranes containing unsaturated acyl chains compared with saturated chains and also preferentially associates with membranes containing cholesterol (Chol) (24). The mechanisms underlying these effects are unknown.

The PM comprises less than 10% of all cellular membranes and exhibits complex compositional and biophysical properties (25). Lipids are distributed asymmetrically, with the outer leaflet composed primarily of sphingomyelin (SM), phosphatidylcholine (PC), and cholesterol and the inner leaflet rich in phosphatidylethanolamine (PE), PS, phosphatidylinositol (PI), and cholesterol (25, 26). Based on indirect evidence, it is sometimes said that retroviruses assemble and bud from PM microdomains called “rafts” (27–29). For example, several lipidomic studies report that the retroviral membrane is enriched in the outer leaflet lipid SM compared with the PM from which the virus buds (30–33). Some lipidomic studies (34) have reported an enrichment of cholesterol in virions compared to total cellular membranes. One previous study found a nonphysiological enrichment of cholesterol in virions compared to the cellular plasma membrane (33); however, a more thorough study found no significant enrichment in cholesterol (35). Additionally, in model membranes with coexisting raft and non-raft domains at a physiological cholesterol mole fraction of 0.35 to 0.4, cholesterol is present at similar levels in the coexisting domains (36–39). Also, under some conditions Gag is found in cold detergent-resistant membranes (DRMs) along with raft markers (40–43). However, DRMs are not a true reflection of proposed rafts in the PM, since detergent type, detergent concentration, temperature, and time affect the degree to which proteins and lipids are isolated in DRMs (44). Moreover, the difficulty in isolating pure PM limits our understanding of lipid and membrane behavior and of the contribution of membranes to retroviral assembly.

Possible phase separation must be taken into account in studies of protein binding to membranes, since lipid components will sort differently into any coexisting phases. In vitro, liquid-liquid phase separation can be observed with lipid-only mixtures of as few as three components (45, 46). Understanding biological membrane phases is facilitated by lipid compositional phase diagrams (36, 37, 47–50). Each lipid phase has well-defined and uniform properties, such as order and density, throughout the whole domain (25, 51). The liquid-ordered phase (Lo) is characterized by fast translational diffusion and high lipid acyl chain order, and it contains a large fraction of saturated acyl chains. The Lo phase has many of the properties attributed to lipid rafts in the PM (52). The liquid-disordered phase (Ld) is characterized by fast translational diffusion but low lipid acyl chain order and a high concentration of unsaturated acyl chains. The solid or gel phase (Lβ) is characterized by lipid diffusion thousands of times slower than that in Ld or Lo and by tightly packed saturated acyl chains. Of great biological relevance is the compositional region in which Ld and Lo phases coexist, which might reflect the phase behavior of the outer leaflet in the PM (45).

In this study, we used two sizes of lipid-only model membranes to study the membrane binding of Gag and other peripheral proteins: large unilamellar vesicles (LUVs) and giant unilamellar vesicles (GUVs). The LUVs we use are 100 nm in size, and LUV-bound protein can be quantitated by flotation in a density gradient (21) or centrifugation to a pellet (liposome pelleting) (53). GUVs we use are cell sized (≥10 μm) and thus easily visualized by fluorescence microscopy when the membrane lipid phases or a protein are marked with a fluorescent dye (54, 55). Both kinds of vesicles are prepared with defined lipid composition. In GUVs with compositions giving rise to Lo and Ld coexistence, the separate domains can be visualized directly by dyes that preferentially partition into Lo or Ld (56). The challenge for interpreting protein binding in phase-separated GUVs is that the composition of each phase can be inferred only if the phase diagram is known. Such diagrams have been developed only for a few compositions (36–39, 48, 57), and none are available for ternary mixtures that include PS in addition to PC-Chol.

A few studies with GUVs have provided insight into the membrane binding properties of HIV-1 Gag and MA (58–60). For example, purified ESCRT proteins were used to infer the assembly order of ESCRT proteins at HIV-1 Gag budding sites (58). In another study, in vitro-translated HIV-1 Gag from a wheat germ extract was used to demonstrate that saturated and unsaturated PI(4,5)P₂ species affect the membrane binding of HIV-1 Gag differently (59). In a different study, multimerized MA bound the Ld phase preferentially in two-phase GUVs (60). However, in those experiments the phase diagram was not known, hence the concentrations of negatively charged lipids in each phase were also unknown, making it difficult to ascribe the observed binding to phase preference as opposed to composition and charge preference.

It remains unclear how membrane order and membrane charge influence association of Gag with membranes. We previously reported that HIV-1 Gag binding to membranes is stimulated by high levels of cholesterol, which has the effect of ordering membranes, but also is stimulated by unsaturated lipids, which has the effect of disordering membranes (24). In a limited set of experiments in a follow-up study, we detected no significant sensitivity of purified RSV Gag to acyl chain saturation and membrane order (10). In the present study, in order to systematically probe the effect of membrane order and charge on RSV Gag membrane binding, we prepared a set of purified fluorescent electrostatic sensor proteins as well as phospholipid headgroup-specific binding proteins. Membrane binding of these control proteins in parallel with binding of RSV Gag and MA was assessed using two-phase GUVs with defined PS distribution, together with liposome-pelleting assays. The work described here represents the first systematic study of the effect of charge, phase, and lipid order on a purified retroviral Gag protein in two-phase GUVs with known PS concentrations in each phase.

MATERIALS AND METHODS

DNA vectors, protein purification, and tissue culture.

Purified proteins used in this study are pictured in Fig. 1. All DNA constructs used for protein purification were cloned into pSUMO (61) vector using standard subcloning techniques as previously described (53). Monomeric neon green (mNG) (62) was amplified from pHisII 6H-mNG, digested with BamHI and EagI, and ligated into pSUMO. pSUMO-mNG served as the vector to generate all other N-terminally tagged mNG proteins (Fig. 1). SUMO-mNG-KRn (n = 4 [RKKR], n = 8 [RKKRKKRK], or n = 12 [RKKRKKRKKRKK]) and SUMO-mNG-MARCKS (KKKKKRFSCKKSFKLSGFSFKKNKK) (63, 64) were made by inserting annealed primers into SUMO-mNG using EagI and XhoI. The linker sequence between the tandem Evectin2 domains is VDGT (65).

FIG 1.

Schematic representation of purified proteins. All charge sensors (top) were N-terminally tagged with the monomeric Neon Green (mNG) fluorescent protein; mNG-MARCKS contains the 26-residue effector domain of MARCKS. RSV Gag constructs (middle) are C-terminally mNG tagged. A nontagged version of RSV MA and Gag was also purified. MASP⁺⁵-mNG contains MA, CA, and SP plus the first 5 residues of the NC domain (53). Lipid sensors (bottom) are N-terminally tagged with mNG. The PS sensor mNG-(Ev2)₂ contains two tandem PH domains of human Evectin2. The PI(4,5)P₂ sensor mNG-PH contains the PH domain from human PLCδ1.

All SUMO-tagged proteins were purified using standard bacterial expression and affinity column techniques (10, 53). In brief, Escherichia coli BL21 cultures were grown at 37°C to an optical density at 600 nm of 0.6. Isopropyl β-d-1-thiogalactopyranoside (IPTG) was added to a final concentration of 0.5 mM, and induced cells were harvested 5 to 6 h postinduction. Pelleted cells were resuspended in lysis buffer [20 mM Tris, pH 8, 500 mM NaCl, 2 mM tris(2-carboxyethyl)-phosphine (TCEP), and 2 mM phenylmethylsulfonyl fluoride (PMSF)] and lysed by sonication. After ultracentrifugation in a TLA-110 Beckman rotor at 90,000 rpm for 45 min, the supernatant was collected. RSV SUMO-MA-(mNG), SUMO-Gag-(mNG), SUMO-MASP⁺⁵-mNG, SUMO-mNG-KR12, and SUMO-mNG-MARCKS supernatants were treated with polyethyleneimine (PEI) and spun at 10,000 rpm in a Sorvall 600 rotor at 4°C to remove nucleic acid. To the supernatant, ammonium sulfate was added to 20% to precipitate the protein of interest, followed by centrifugation. The pellet was resuspended in binding buffer (20 mM Tris-HCl, pH 8, 50 mM NaCl, and 2 mM TCEP) and further purified by cation exchange chromatography (HiTrap SP FF; GE Healthcare) and Ni²⁺ affinity chromatography. SUMO-mNG, SUMO-mNG-KR4, SUMO-mNG-(Ev2)₂, and SUMO-mNG-PH were only subjected to Ni²⁺ affinity chromatography (HisTrap HP; GE Healthcare). Following the first round of Ni²⁺ chromatography, eluted proteins were dialyzed against buffer (20 mM Tris-HCl, pH 8, 500 mM NaCl, and 1 mM TCEP) in the presence of ULP1 protease (66) to cleave off the SUMO tag. The ULP1 protease and SUMO tag were removed by Ni²⁺ affinity chromatography. Purified protein at 2 to 10 mg/ml was flash frozen in aliquots and stored at −80°C. The final protein preparation had an A₂₆₀/A₂₈₀ ratio of 0.58 to 0.59, indicating the absence of nucleic acid. Gag-mNG formed regular virus-like particles under standard in vitro assembly conditions screened by negative stain electron microscopy (10). All proteins had a purity of approximately 90% after affinity column purification, as judged from stained gels.

A subset of constructs (Gag, KR4, KR8, KR12, and MARCKS) was cloned into a tissue culture plasmid for expression of mNG fusion proteins in avian cells. Cells were maintained, transfected, and prepared for imaging as previously described (53). Gag-mNG was localized in puncta at the PM, consistent with reports for Gag-green fluorescent protein (GFP) (21, 53). KR4 was mostly cytoplasmic and KR8 was in the nucleus, while KR12 and MARCKS were mostly in endosome-like vesicles (data not shown).

Phospholipids and fluorescent probes.

Phospholipids (Table 1) were purchased from Avanti Polar Lipids (Alabaster, AL). Cholesterol was obtained from Nu-Chek Prep (Elysian, MN). Cholesterol stock solution was prepared by standard gravimetric procedures to ∼0.2% error (67). The working stocks of the fluorescent lipid analogs dehydroergosterol (DHE; Sigma-Aldrich, St. Louis, MO), 3,3′-dilinoleyloxacarbocyanine perchlorate (Fast DiO; ThermoFisher, NY), naphtho[2,3-a]pyrene (naphthopyrene; Sigma-Aldrich, MO), and lissamine rhodamine 18:1,18:1-PE (LR-DOPE; Avanti, AL) were prepared in methanol. Probe concentrations were determined in methanol by absorption spectroscopy using an HP 8452A spectrophotometer (Hewlett-Packard, Palo Alto, CA). Fluorescent probe extinction coefficients were obtained from lot certificates of analysis: 12,900 M⁻¹ cm⁻¹ at 324 nm for DHE, 138,000 M⁻¹ cm⁻¹ at 499 nm for Fast DiO, 23,700 M⁻¹ cm⁻¹ at 460 nm for naphthopyrene, and 75,000 M⁻¹ cm⁻¹ at 560 nm for LR-DOPE. Concentrations of phospholipid stocks were determined to <1% error by inorganic phosphate assay (67). Purity of >99.5% was confirmed by thin-layer chromatography (TLC) on washed, activated silica gel plates (Alltech, Deerfield, IL), developed with chloroform-methanol-water (65:25:4) for all phospholipids, petroleum ether-diethyl ether-chloroform (7:3:3) for cholesterol, DHE, and naphthopyrene, chloroform-methanol (2:1) for Fast DiO, and chloroform-methanol (9:2) for LR-DOPE. TLC plates for brain PI(4,5)P₂ were activated with K₂ oxalate and developed with 1-propanol–2 M acetic acid (65:35).

TABLE 1.

Lipids used to form unilamellar vesicles

Lipid name	Abbreviation	Acyl chains	Head group charge
1-Palmitoyl,2-oleoyl-sn-glycero-3-phosphocholine	POPC	16:0, 18:1	0
1,2-Dioleoyl-sn-glycero-3-phosphocholine	DOPC	18:1, 18:1	0
1,2-Distearoyl-sn-glycero-3-phosphocholine	DSPC	18:0, 18:0	0
1-Palmitoyl-2-oleoyl-sn-glycero-3-phospho-l-serine	POPS	16:0, 18:1	−1
1,2-Dioleoyl-sn-glycero-3-phospho-l-serine	DOPS	18:1, 18:1	−1
1,2-Dihexadecanoyl-sn-glycero-3-phospho-l-serine	DPPS	16:0, 16:0	−1
Brain l-α-phosphatidylinositol-4,5-bisphosphate	PI(4,5)P2	18:0, 20:4a	−3 to −4

^aPredominant species.

LUV preparation and liposome-pelleting assay.

LUVs with diameters of 100 nm in 20 mM Tris-HCl, pH 8.0, were prepared by rapid solvent exchange (RSE) as described by Buboltz and Feigenson (68), followed by extrusion through polycarbonate filters 31 times. In the liposome-pelleting assay (53), 15 μg of purified protein and 50 μg of LUVs were added to binding buffer (20 mM Tris-HCl, pH 8, with varied salt concentrations) to a final volume of 200 μl binding reaction mixture at a final NaCl concentration of 50 mM, 150 mM, or 300 mM. The mixture was incubated at room temperature for 10 min. Tubes were ultracentrifuged at 90,000 rpm in a TLA-100 (Beckman) rotor for 45 min at 4°C. Supernatant was removed and the pellet was resuspended and subjected to SDS-PAGE analysis. Gels were Coomassie blue stained overnight and destained, and band intensity was determined by densitometry analysis using ImageQuant software 5.2.

ESR.

Multilamellar vesicles (MLVs) were prepared as described previously (69) in buffer (20 mM Tris-HCl, pH 8, 150 mM NaCl). Each sample contained 2 μmol lipid and 0.2 mol% of the spin label probe 16-DOXYL-stearic acid (Sigma-Aldrich). MLVs were centrifuged, excess buffer removed, and samples placed in thin glass capillary tubes. Electron spin resonance (ESR) spectra were collected for all samples on a 9.4-GHz Bruker continuous wave (cw)-ESR spectrometer (EMS) at room temperature (22°C). At least five scans were averaged for each sample. The maximum and minimum values of tensor A (A_max and A_min) were determined from each spectrum, and the order parameter was calculated according to Schorn and Marsh (70) using the hyperfine tensor (A_xx, A_yy, A_zz) = (5, 5, 32.8 G).

FRET.

The phase boundaries at the ends of the tieline for composition DSPC/DOPC/Chol (30/45/25) [1,2-distearoyl-sn-glycero-3-phosphocholine–1,2-dioleoyl-sn-glycero-3-phosphocholine–cholesterol (30/45/25)] were determined by fluorescence resonance energy transfer (FRET) as previously described (36, 48). FRET samples were prepared at 2% compositional increments except near the phase boundary, where 1% increments were used (37, 48). DHE and Fast DiO were chosen as a FRET pair, at 1/100 and 1/700 fractions in total lipid. Light scattering and fluorescence background were controlled as described previously (71, 72). Lipids were mixed in chloroform to 120 nmol, and the organic solvent was replaced by buffer using rapid solvent exchange (RSE) (68) and then sealed under argon. Equilibration was achieved by first heating samples to 55°C for 2 h followed by gradually cooling to 23°C at a rate of 2°C/h and then incubation at 23°C for 48 h. A 1.8-ml volume of buffer [200 mM KCl, 5 mM piperazine-N,N′-bis(2-ethanesulfonic acid), 1 mM EDTA, pH 7] was added to 0.2 ml of RSE sample to yield approximately 30 μM lipid vesicles in the cuvette. Data were collected on a Hitachi F-7000 FL spectrofluorimeter (Hitachi High Technologies America, Inc., Schaumburg, IL) at 23°C. Using 5-nm bandpasses for excitation and emission slits and a 10-s integration time, intensity was measured in 4 channels (excitation/emission): DHE (327/393 nm), Fast DiO sensitized emission (327/503 nm), Fast DiO direct fluorescence (477/503 nm), and light scattering (440/420 nm). Briefly, corrections account for non-FRET contributions of direct fluorescence emission from donor and acceptor and scattering of excitation light by the vesicle suspension (71). Phase boundaries were determined from FRET trajectories as the intersection of two straight lines drawn on either side of the boundary, as described in the relevant figure legends. Analysis of unsmoothed data gave similar boundary locations but with slightly larger uncertainties.

GUV preparation, imaging, and protein binding.

The lipid compositions of each phase on the three types of GUVs used are shown in the figures. The same starting composition was chosen, DSPC/DOPC/Chol (30/45/25), marked by a black arrow in the figures. For GUVs having PS enriched in Ld, 33% of DOPC was replaced with 1,2-dioleoyl-sn-glycero-3-phospho-l-serine (DOPS), and the resulting lipid mixture, DSPC/DOPC/DOPS/Chol (30/30/15/25), contains approximately 24 mol% PS in Ld and 2 mol% PS in Lo. For GUVs having PS enriched in Lo, 41% of the DSPC was replaced with 1,2-dihexadecanoyl-sn-glycero-3-phospho-l-serine (DPPS), and the resulting lipid mixture, DSPC/DPPS/DOPC/Chol (18/12/45/25), contains approximately 24 mol% PS in Lo and 2 mol% PS in Ld. The lipid composition of each phase of GUVs corresponds to that of the LUV Ld or Lo phase with 24 mol% PS in purple or red, as well as the LUV Ld or Lo phase with 2 mol% PS in blue or green used in the liposome-pelleting assay. The third type of GUV has PS in both Ld and Lo phases, with 33% of DOPC replaced with DOPS and 41 mol% of DSPC replaced with DPPS, yielding the lipid mixture DSPC/DPPS/DOPC/DOPS/Chol (18/12/30/15/25), containing approximately 24 mol% PS in Ld and 24 mol% PS in Lo.

GUV samples described here were prepared by gentle hydration (38, 73). A total of 300 nmol total lipid containing 0.02 mol% LR-DOPE was mixed in chloroform, partially dried to a thin film in a culture tube using a rotary evaporator, and then dried with heating at 55°C under high vacuum for 1.5 h. The thin dry film was then hydrated with wet N₂ gas at 55°C for 30 min. Lipid films were further hydrated with prewarmed sucrose buffer (125 mM sucrose, pH 8) and incubated at 55°C for 30 min. GUVs formed as the sample was cooled over 10 h to room temperature (∼23°C). GUVs were harvested into buffer (50 mM NaCl, 20 mM Tris, pH 8). All buffers were osmotically balanced, confirmed by measurements using an osmometer (Precision Systems Inc., Natick, MA).

GUVs were imaged on a Nikon Eclipse Ti microscope (Nikon Instruments, Melville, NY) at 23°C, using a 60×, 1.2-numeric-aperture water immersion objective. Images were taken with an Andor Zyla VSC-01037 camera (Oxford Instruments, South Windsor, CT). Imaging used a low exposure time of 50 ms to minimize light-induced artifacts (74). LR-DOPE was excited at 555 nm, and emission was collected at 600 to 675 nm; naphthopyrene was excited at 430 nm, and emission was collected at 460 to 500 nm. When performing protein-membrane assays, only LR-DOPE was added as an Ld marker in GUVs. Protein was added to a final concentration of 1 μM to preformed GUVs and incubated for 30 s before imaging. mNG was excited at 470 nm, and emission was collected at 495 to 525 nm. GUV images containing multiple colors were color merged using NIS software: mNG is green, LR-DOPE is red, and naphthopyrene is blue. The contrast of entire images was enhanced with NIS Elements Basic Research Software (MVI, Inc.). Quantification of fluorescence signals in microscopy images was determined using the line scan function of NIS Elements Basic Research Software. For the two-phase GUVs, line scans were performed on no fewer than 80 GUVs for each protein. Quantification of Ld/Lo-bound protein was performed on GUVs with PS in Ld and Lo. On the merged GUV images, the background fluorescence intensity of unbound protein was subtracted from that of GUV-bound proteins. The Ld/Lo protein binding ratio was determined and expressed as a box plot, where the bottom, middle, and top lines of the box represent the first, second (median), and third quartiles, whiskers are the minimum and maximum values, and circles are outliers. For PI(4,5)P₂ containing GUVs, two line scans per GUV and no fewer than 30 GUVs were measured. The relative fluorescence intensity of protein binding above background is shown as (membrane-bound fluorescence − background)/background, expressed as a percentage.

RESULTS

While electrostatic interaction is known to play a role in Gag-membrane binding, there have been no systematic studies using purified proteins. While at least some retroviruses have raft-like lipid compositions, no controlled experiments have probed the possible role of lipid-phase properties in membrane interactions. We address both of these topics for RSV Gag. We built and purified a series of fluorescent proteins that act as charge sensors, based on mNG with an added C-terminal tail of 4, 8, and 12 alternating lysine and arginine residues (Fig. 1). To bridge this study to other cellular proteins, mNG-MARCKS was included, since the MARCKS protein contains a well-characterized, 26-residue peripheral membrane interacting peptide (63, 64, 75, 76). This peptide has a cluster of 13 basic amino acids but also includes 5 hydrophobic residues that can contribute to membrane interaction. As further controls we created fusions of mNG with two lipid headgroup-binding proteins, Evectin2 and the PH domain of phospholipase C δ1. Evectin2, which we used as a tandem duplication (65), interacts specifically with PS (77), while the PH domain interacts specifically with PI(4,5)P₂ (78).

Membrane surface potential is strongly affected by ions, which shield the negatively charged lipid headgroups and positively charged amino acid side chains and thus reduce electrostatic attraction to basic residues on proteins (22). Previous studies based on liposome flotation reported that RSV MA membrane binding decreases markedly with increasing NaCl concentration (22). To confirm and extend these observations, we used a centrifugation assay to determine how ionic strength influences interaction of viral and control proteins with LUVs. The percentage of purified protein in a membrane pellet was quantified at 50, 150, and 300 mM NaCl (Fig. 2A, lanes 1 to 3), with centrifugation in the absence of LUVs serving as the background control (background, lanes 4 to 6). mNG itself did not detectably pellet with LUVs at any NaCl concentration (Fig. 2B), confirming the expected lack of contribution of the mNG moiety to membrane binding. As predicted, the charge sensors were very sensitive to increasing salt, with a longer stretch of basic residues leading to more extensive LUV binding at a fixed NaCl concentration (79). mNG-MARCKS behaved similarly to mNG-KR12. For the viral proteins, approximately 70% of MA-(mNG) and Gag-(mNG) bound to LUVs at 50 mM, almost 10-fold more than at 300 mM (Fig. 2B), consistent with an electrostatic mode of membrane interaction (22). On the other hand, compared with the retroviral proteins and the charge sensors, the PS sensor mNG-(Ev2)₂ was much less sensitive to ionic strength, showing a less than 2-fold change when NaCl was decreased from 300 mM to 50 mM. The PI(4,5)P₂ sensor mNG-PH was somewhat sensitive to ionic strength, pelleting with LUVs 8-fold less at 300 mM than at 50 mM. While PH is accepted to be specific for the PI(4,5)P₂ headgroup (78, 80), we found that mNG-PH showed significant membrane binding at 50 mM NaCl even in the absence of PI(4,5)P₂. This interaction may be mediated by several basic residues in the β1/β2 loop of the PH domain, which also form part of the specific binding pocket for PI(4,5)P₂ (81).

FIG 2.

Effects of sodium chloride on protein-membrane binding. (A) LUVs were composed of POPC/POPS/Chol (34/30/36). A stained SDS gel of RSV MA is shown as an example of the liposome-pelleting assay at 50, 150, and 300 mM NaCl. Gel bands represent the amount of protein associated with pelleted LUVs at the three NaCl concentrations; background bands represent the protein pelleted in the absence of LUVs at the corresponding NaCl concentrations. Total is the total amount of protein used in each binding reaction. (B) White, light gray, and dark gray bars represent the average percentage of no fewer than three independent pelleting reactions at 50, 150, and 300 mM NaCl. Error bars represent standard deviations from the means.

We also examined whether elevating PS concentration in membranes enhances protein binding (Fig. 3). As expected, RSV MA, mNG-KR8, mNG-KR12, and mNG-MARCKS had almost undetectable membrane binding at 0 mol% PS, with their LUV binding increasing as the PS concentration was raised from 10 to 50 mol%. Note that because of the absence of cholesterol, the membrane binding was weaker in this set of experiments than in those in Fig. 2. Overall, the results shown in Fig. 2 and 3 are evidence that electrostatic attraction is the driving force for membrane association of RSV MA and Gag, as it is for the charge sensors.

FIG 3.

Protein binding to POPC/POPS LUVs with increasing PS concentration. LUVs were prepared with POPC-POPS, with 10 mol% POPS incremental increases from 0 to 50 mol%. The curves show the percentage of protein bound to LUVs at the six POPS concentrations. Binding levels of mNG-KR8, mNG-KR12, mNG-MARCKS, and RSV MA are shown in purple, green, red, and blue, respectively. All data points represent the averages from no fewer than three independent liposome-pelleting assays at 150 mM NaCl; error bars represent standard deviations from the means.

RSV Gag prefers membranes with mixed acyl chain compositions.

Previously, we showed that HIV-1 Gag binds more strongly to membranes with unsaturated lipids than to membranes with saturated lipids (24). HIV-1 Gag may also be sensitive to the acyl chain content of PI(4,5)P₂ in membranes (59). Sensitivity to acyl chain type is specific for the charged lipid [PS or PI(4,5)P₂], whereas the acyl chains on PC have little influence on binding (10). In contrast to HIV-1 Gag, in previous studies RSV Gag did not respond to the level of unsaturation, either of PS or PC (10). Here, we systematically tested RSV Gag binding to five types of LUVs at the physiologically relevant concentration of 36 mol% cholesterol. The fluorescent protein charge sensors and lipid headgroup sensors were tested in parallel. Membrane lipid compositions were confirmed by GUV imaging analysis to be uniform, not phase separated (data not shown).

RSV Gag, RSV MA, and mNG-KR8 shared the same membrane-binding patterns (Fig. 4). More specifically, the two types of LUVs having one phospholipid with two unsaturated acyl chains (18:1, 18:1) and the other phospholipid with two saturated acyl chains (16:0, 16:0 or 18:0, 18:0) consistently supported approximately 15% more binding than the other three types of LUVs in which all of the acyl chains in both phospholipids were either saturated or unsaturated. The origin of this difference is unclear, but it is known that in PC-PS membranes where each lipid has different acyl chains, the lipid mixing is nonideal (82) and the PS has a higher thermodynamic activity (83).

FIG 4.

Effects of saturated and unsaturated lipid acyl chains on protein membrane association. The five LUV compositions are shown on the right side. Cartoons represent different types of lipid acyl chains and headgroups. Black and white ovals represent PC and PS, respectively. Straight lines represent saturated acyl chains, and kinked lines represent unsaturated acyl chains. The oval symbol represents cholesterol. All graphs show the percentage of total protein associated with LUVs at 150 mM NaCl. Bars represent the averages from no fewer than three independent liposome-pelleting assays; error bars represent standard deviations from the means.

In contrast, mNG-MARCKS and mNG-(Ev2)₂ had different membrane binding properties than RSV MA, Gag, and mNG-KR8 (Fig. 4). Both mNG-MARCKS and mNG-(Ev2)₂ showed the strongest binding to LUVs with unsaturated PS (for example, DOPC/DOPS/Chol), regardless of the acyl chain saturation of PC, and had the weakest binding to LUVs with both PC and PS having saturated chains (DSPC/DPPS/Chol). In addition, for the two types of LUVs with different saturated and unsaturated lipids, the one with unsaturated PS and saturated PC (DSPC/DOPS/Chol) supported higher binding than the one with saturated PS and unsaturated PC (DOPC/DPPS/Chol). One model to account for these results holds that mNG-MARCKS and mNG-(Ev2)₂ prefer unsaturated PS, which may be more favorable for protein hydrophobic amino acid side chain insertion. A similar effect of the HIV-1 Gag myristate modification might explain why this Gag protein strongly prefers to bind to membranes with unsaturated PS (24). To further test the effects of PS acyl chain saturation, we prepared two series of four-component LUVs, systematically varying the PS concentrations (Fig. 5A). Starting with LUVs rich in unsaturated PC [DSPC/DOPC/Chol (8/56/36)], DOPC was replaced with unsaturated DOPS in increments of 10 mol%. Similarly, starting with LUVs rich in saturated PC [DSPC/DOPC/Chol (56/8/36)], DSPC was replaced with the saturated DPPS in increments of 10 mol%. As expected, in both cases for all proteins tested, higher concentrations of PS led to increased binding (Fig. 5B). For MA, mNG-KR12, and mNG-MARCKS, the LUVs with unsaturated PS (0 to 50% DOPS) and LUVs with saturated PS (0 to 50% DPPS) supported protein binding to a similar level at the same PS concentration. However, mNG-(Ev2)₂ had a very strong preference for LUVs with unsaturated PS, consistent with the results shown in Fig. 4. In summary, RSV MA and Gag do not show a strong preference for membranes with either saturated or unsaturated lipids, while Evectin2 and possibly MARCKS show preference for membranes with unsaturated PS.

FIG 5.

Effect of increasing PS concentration of Ld and Lo membranes on protein membrane association. (A) Compositions of LUVs studied. The left side shows LUVs composed of DOPC/DSPC/DOPS/Chol (Ld) with fixed DSPC and Chol mol% but increasing DOPS (in boldface) and decreasing DOPC mol%. The right side shows LUVs composed of DOPC-DSPC-DPPS-Chol (Lo) with fixed DOPC and Chol mol% but increasing DPPS (in boldface) and decreasing DSPC mol%. All LUVs are single phase (uniform), confirmed by GUV image analysis (data not shown). (B) Dashed lines represent protein binding to Ld LUVs composed of DOPC/DSPC/DOPS/Chol (DOPS 0 to 50 mol%); solid lines represent protein binding with Lo LUVs composed of DOPC/DSPC/DPPS/Chol (DPPS 0 to 50 mol%). All data points represent averages from no fewer than three independent liposome-pelleting assays at 150 mM NaCl; error bars represent standard deviations from the means.

Gag binding follows membrane charge while Evectin2 binding follows membrane order.

At certain concentrations, mixing high-melting-temperature (T_m) lipids, low T_m lipids, and cholesterol results in phase separation (45, 51). Numerous FRET and/or ESR measurements and GUV imaging studies are needed to create phase diagrams to represent the lipid compositions of such membranes (49, 84). On a ternary phase diagram, the phase boundaries and samples prepared along a tieline enable calculating the exact lipid compositions of each phase in the membrane. On any point of total sample composition within the Ld+Lo coexistence region, the tieline through this point (dashed line in Fig. 6B) intercepts the phase boundaries, connecting the compositions of the two coexisting phases that are in equilibrium (46). Along a tieline the fraction of each coexisting phase is known. Currently, the majority of published phase diagrams describe neutral lipid mixtures (36–39, 48). No published data are available with ternary phase diagrams containing PC/PS/Chol.

FIG 6.

Protein binding to LUVs with PS in Ld or Lo phases. (A) The effect on phase boundaries of adding PS to DSPC/DOPC/Chol. Shown is stimulated acceptor emission (FRET; arbitrary units) versus DSPC mole fraction (χDSPC) for three compositional trajectories, measured along straight lines as shown in panel B. Each trajectory passes through the mixture composition DSPC/DOPC/Chol (30/45/25) (black circle) in the phase diagrams in panel B. The top graph in panel A shows the measured FRET along a single trajectory in which 33% DOPC has been replaced with DOPS. The middle graph shows the FRET trajectory of membranes of DSPC/DOPC/Chol, with no PS replacement. The bottom graph shows the FRET trajectory of membranes in which 41% of the DSPC is replaced with DPPS. Each trajectory contains 75 data points. Data were locally smoothed, and phase boundaries were determined by fitting a subset of data near the boundary (colored open circles) to a piecewise linear function. Vertical dashed lines indicate the left and right phase boundaries, and vertical colored bars indicate the confidence intervals. (B, left) Phase diagram of DSPC/DOPC/Chol with 33% of total DOPC replaced with DOPS. (Right) Phase diagram of DSPC/DOPC/Chol with 41% of total DSPC replaced with DPPS. The locations of the four LUV compositions studied are marked on the corresponding phase diagram. Colors correspond to LUV compositions just below the phase diagrams, and their representative levels of protein binding are pictured in the charts. High PS means 24 mol% PS, and low PS means 2 mol% PS. (C) The amount of protein binding to the four types of LUVs shown in panel B. Liposome-pelleting assays for mNG-KR12, mNG-MARCKS, and mNG-(EV2)₂ at 150 mM NaCl are shown on the left; the liposome binding for Gag-mNG at 50 mM NaCl is shown on the right. All bars represent the averages from no fewer than three independent liposome-pelleting assays; error bars represent standard deviations from the means. The order parameter, S, determined by ESR (not shown) of each membrane composition was 0.14 for Ld (purple and blue) and 0.31 for Lo (green and red).

To allow accurate calculations of lipid compositions that include PS and to determine phase boundaries, we employed FRET assays (72) using the two fluorescent dyes DHE and Fast DiO to measure 75 lipid mixtures prepared along a tieline. DHE preferentially partitions into the Lo phase (48), and Fast DiO preferentially partitions into the Ld phase (56). The stimulated acceptor emission was measured as a FRET signal along the tieline on the phase diagram (Fig. 6A and B). First, we measured the FRET signal of the 75 samples composed of DSPC/DOPC/Chol without any PS replacement (Fig. 6A, middle). Consistent with the published phase diagram (36, 48), the original phase boundaries are at χ_DSPC of 0.05 (Ld) and χ_DSPC of 0.58 (Lo). We next measured two sets of 75 samples with partial PS replacement. In each set PC was replaced with PS that contained the same acyl chain type, so that the percentage of high T_m and low T_m lipids remained the same. In the first set of experiments, 33% of the total DOPC was replaced with DOPS for all samples along the tieline (Fig. 6A, top). This replacement shifted the Ld phase boundary to χ_DSPC of 0.09, i.e., 0.04 to the right compared with the original boundary along the tieline. The Lo phase boundary remained close to χ_DSPC of 0.59. In the second set of experiments, 41% of the total DSPC was replaced with DPPS for all samples along the tieline (Fig. 6A, bottom). We found the Ld phase boundary shifted to χ_DSPC of 0.09 along the tieline, while the Lo phase boundary stayed at χ_DSPC of 0.58. In summary, these results imply that replacing up to 41% of PC with PS in the membrane changes the phase boundaries along the chosen tieline only slightly. Importantly for our experiments, replacing 33% of the total DOPC with DOPS, or replacing 41% of the total DSPC with DPPS, resulted in the same PS concentration in the Ld and Lo phases, respectively, at our chosen two-phase GUV compositions.

The previous figures show that Gag has no preference for membranes composed either of fully saturated lipids (DSPC/DPPS/Chol) or fully unsaturated lipids (DOPC/DOPS/Chol) (Fig. 4). To directly test the hypothesis that Gag does not recognize membrane order, we compared membrane binding using two sets of LUVs having the same PS concentration but with different membrane orders (Fig. 6B). In the case of partially replacing DOPC with DOPS (Fig. 6B, left), the two points at the phase boundaries correspond to compositions of the coexisting Ld (purple circle) and Lo (green circle) phases. The corresponding Ld composition contains 24 mol% PS, while Lo composition contains 2 mol% PS. In the case of partially replacing DSPC with DPPS (Fig. 6B, right), the two points at the phase boundaries correspond to compositions of the coexisting Ld (blue circle) and Lo (red circle) phases. The corresponding Ld composition has 2 mol% PS, while the Lo composition has 24 mol% PS. In liposome-pelleting assays shown in Fig. 6C, Gag-mNG and mNG-KR12 behaved similarly, with binding correlated only with PS concentration, independent of membrane order. Both proteins bound to high-PS LUVs strongly and to low-PS LUVs weakly. In contrast, mNG-MARCKS and mNG-(Ev2)₂ bound more strongly to LUVs with high PS in Ld, showing a 2-fold and 10-fold preference, respectively.

We next sought to confirm the liposome-pelleting results with a GUV assay that enables direct visualization of protein binding to membranes of controlled lipid composition. An advantage of GUVs having coexisting phases is that they allow competition for binding to portions of the membrane with distinct properties, such as membrane order. Three types of GUVs were prepared with lipid compositions similar to those for the LUVs shown in Fig. 6. The first type (PS in Ld; Fig. 7A, top) contained 24 mol% PS in Ld and 2 mol% PS in Lo. The second type (PS in Lo; Fig. 7A, middle) contained 2 mol% PS in Ld and 24 mol% PS in Lo. The third type of GUV (PS in Ld+Lo; Fig. 7A, bottom), contained 24 mol% PS in both Ld and Lo phases. LR-DOPE, a red fluorescent marker for the Ld phase, was added for all three types of GUVs (56) (see Materials and Methods for details).

FIG 7.

Protein binding to Ld+Lo GUVs. (A) DSPC/DOPC/Chol phase diagram. The dotted line drawn between the dashed circle (Ld) and solid circle (Lo) is the tieline that connects the compositions of the Ld and Lo phases that coexist. The circle at the center of the two-phase region of the phase diagram represents the overall composition that has split into the two coexisting phases. GUV lipid compositions are shown on the right, and the diagram depicts the location of PS enrichment in blue: top, PS in the Ld phase; middle, PS in the Lo phase; bottom, PS in both Ld and Lo phases. (B) Box plot shows the ratio of protein fluorescence in Ld phase to that in Lo phase when PS is in both phases. (Inset) Example of line-scan and intensity plot for mNG signal (green) and LR-DOPE (Ld phase marker; red). Box plots represent no fewer than 80 individual GUVs quantified for each protein-GUV binding assay at 50 mM NaCl, as described in Materials and Methods. (C) Representative fluorescence from wide-field microscopy images of protein (final concentration, ∼1 μM) binding to phase-separated GUVs at 50 mM NaCl. GUV binding assays were performed on the three GUV compositions depicted in panel A. For each panel, the first column is LR-DOPE labeling of Ld phase (red channel), the middle column is mNG-tagged protein (green channel), and the right column shows LR-DOPE and mNG merged channels. Scale bar, 20 μm.

In protein binding assays with GUVs, both mNG-KR12 and Gag-mNG bound similarly to the phase with high PS (24 mol%), whether it was Ld or Lo (Fig. 7C). For GUVs with high PS in both Ld and Lo, the ratio of protein bound to Ld compared with Lo was approximately 1, as quantified for over 80 GUVs (Fig. 7B). mNG-MARCKS also bound to GUVs with high PS in Ld and to GUVs with high PS in Lo. However, for the GUVs with high PS in both phases, the fluorescence intensity of mNG-MARCKS in the Ld phase was twice that of the Lo phase (Fig. 7B). Thus, consistent with experiments using liposome pelleting, mNG-MARCKS has a modest preference for disordered membranes at the same PS concentration. Remarkably, the fluorescent PS sensor mNG-(Ev2)₂ showed an even stronger preference for the Ld phase. It bound detectably only to the GUVs with a PS-rich Ld phase; in GUVs with the same high PS in both phases, it bound 10-fold more strongly to Ld than to Lo. In summary, the results both from liposome pelleting and GUVs demonstrate that RSV Gag-mNG behaves like the charge sensor mNG-KR12, independent of the membrane order. In contrast, mNG-(Ev2)₂ and, to a lesser extent, mNG-MARCKS also recognize membrane order, preferring the disordered phase.

Effects of PI(4,5)P₂, cholesterol, and multimerization on membrane binding.

PI(4,5)P₂ comprises approximately 2 mol% of total inner leaflet lipid (85) and contributes to the negative charge of the PM. This phosphoinositide is found at higher levels in viral membranes of HIV-1 and murine leukemia virus (MuLV) than in the PM (33), suggesting a role in retroviral assembly and budding (8). While HIV-1 MA has a specific PI(4,5)P₂ binding pocket (3, 17), RSV MA has no known PI(4,5)P₂ binding pocket. When PI(4,5)P₂ was depleted in HIV-1 Gag-expressing cells by cotransfecting 5-phosphatase, most Gag was no longer PM localized and virus budding was reduced (8). In parallel experiments with RSV Gag and HIV-1 Gag, depleting PI(4,5)P₂ affected PM association and viral budding much less for RSV than for HIV-1 (19). However, other researchers observed reduced RSV Gag PM localization and viral release in PI(4,5)P₂-depleted cells but without a parallel assay with HIV-1 Gag (20). Thus, the nature of the interactions in vivo between RSV Gag and PI(4,5)P₂ remains unclear.

To further explore whether RSV Gag binds to PI(4,5)P₂ through a specific lipid headgroup pocket or through electrostatic interactions, we compared RSV Gag with mNG-PH, which has a structured PI(4,5)P₂ binding pocket (78, 80), and also with the mNG-based charge sensors, which have only unstructured basic tails (Fig. 8). Membrane binding was tested for mNG-KR8, Gag-mNG, and mNG-PH at 30 mol% PS and for mNG-KR12 and mNG-MARCKS at 20 mol% PS. The membrane binding of mNG-KR12 and mNG-MARCKS at 30 mol% PS without Chol and without PI(4,5)P₂ was already high and thus masked the Chol and PI(4,5)P₂ effect. Therefore, we performed the assay at 20 mol% PS. At these PS concentrations the effects of cholesterol and PI(4,5)P₂ were most pronounced (Fig. 8B). The additional PI(4,5)P₂ increased membrane binding of mNG-KR8 slightly, and as expected it enhanced mNG-PH binding very strongly. Membrane association of mNG-KR12, mNG-MARCKS, and RSV Gag-mNG was significantly increased (Fig. 8), which implies that a protein being highly responsive to PI(4,5)P₂ does not require a PI(4,5)P₂ binding pocket. A similar inference was made previously in studies of the MARCKS peptide (63, 76).

FIG 8.

Effects of cholesterol and PI(4,5)P₂ on protein-membrane binding. (A and B) Compositions of LUVs are shown at the top. PIP2 is short for PI(4,5)P₂. (A) LUVs contained 30 mol% POPS; liposome-pelleting assays were at 150 mM NaCl. (B) LUVs contained 20 mol% POPS; liposome-pelleting assays were at 150 mM NaCl. Data are averages from no fewer than three independent liposome-pelleting assays; error bars represent standard deviations from the means.

Cholesterol increases HIV-1 Gag and RSV Gag binding to LUVs (10, 24). To address if this is a general effect for all peripheral membrane binding proteins, we tested Gag-mNG and, in parallel, the charge and lipid headgroup sensors (Fig. 8). Cholesterol added to 36 mol% increased the membrane binding of Gag-mNG and mNG-KR12 significantly, mNG-MARCKS mildly, but mNG-KR8 and mNG-PH not at all. These data suggest that cholesterol enhances membrane interactions only for strong membrane binders such as RSV Gag-mNG and mNG-KR12 but has little effect on weak binders such as mNG-KR8.

To build upon the liposome-pelleting results, we used GUVs that did not have Ld+Lo phase separation to examine the effects of PI(4,5)P₂ on membrane binding of RSV MA-mNG, Gag-mNG, and the charge sensors. GUVs were prepared with POPC/POPS/Chol in the presence or absence of 2 mol% PI(4,5)P₂. All of the proteins tested bound to GUVs uniformly but with different intensities (Fig. 9A). Protein-membrane binding was semiquantified by performing line scan analysis on at least 30 GUVs for each fluorescent protein, with subtraction of background fluorescence measured as far as possible from the GUV itself (Fig. 9B). Of note, all of the GUV binding assays were performed at 50 mM NaCl, which is lower than the level in the liposome-pelleting assays. As expected, mNG-PH bound weakly to GUVs with no PI(4,5)P₂, similarly to mNG-KR4, while it bound 16-fold more strongly to GUVs with 2 mol% PI(4,5)P₂ (Fig. 9B, far right column). PI(4,5)P₂ increased GUV binding of mNG-KR12 about 3-fold above background and increased Gag-mNG binding by about 50%. These results are qualitatively consistent with those based on liposome pelleting (Fig. 8). In summary, these results show that PI(4,5)P₂ stimulates membrane binding significantly not only for mNG-PH, which has a structured PI(4,5)P₂ binding pocket, but also for the charge sensors. Thus, robust responses to PI(4,5)P₂ can be entirely electrostatic.

FIG 9.

PI(4,5)P₂ enhances binding of proteins to uniform POPC/POPS/Chol GUVs. (A) Representative images of GUV protein binding assays at 50 mM NaCl. For each protein a GUV binding assay was performed at least three times independently. Top row, POPC/POPS/Chol (34/30/36); bottom row, POPC/POPS/Chol/PI(4,5)P₂ (32/30/36/2). All samples were imaged using the same microscope settings. Brightness and contrast for each image were not adjusted. The fluorescence that appears to be inside the GUVs is due to out-of-focus light. Scale bar, 20 μm. (B) Relative fluorescence intensity of membrane-bound protein above background. At least 30 individual GUVs and 2 line scans per GUV were quantified for each protein, as described in Materials and Methods. The relative percentage of protein binding above background is shown. The averages from at least 60 measurements are displayed ± standard deviations.

Previous experiments suggested that the stronger membrane interaction of RSV Gag compared to that of MA is due to the capacity of Gag to multimerize (21) and not to the presence of the basic NC domain, which also has the capacity to interact with anionic lipids (86). At least in part, multimerization is induced by the formation of a six-helix bundle comprising the SP segment and short adjoining sequences of six Gag molecules (2, 87). Using the GUV assay, we tested the effects of PI(4,5)P₂ on binding of a Gag protein missing NC but including the entire SP domain, MASP⁺⁵-mNG (21). In the absence of PI(4,5)P₂, MASP⁺⁵-mNG bound to GUVs weakly, similar to MA-mNG. However, in the presence of PI(4,5)P₂, the binding of MASP⁺⁵-mNG was almost double that of MA-mNG, although it was lower than that of Gag-mNG, possibly due to a lack of positively charged NC domain. These results are consistent with the model in which membrane interaction of proteins lacking a specific binding pocket is stimulated by PI(4,5)P₂ only for proteins that bind tightly (e.g., can multimerize).

DISCUSSION

The mechanism by which the retroviral Gag protein is targeted to the inner leaflet of the PM is complex and incompletely defined. At least for some retroviruses, the viral lipid composition differs from the composition of the plasma membrane, including an enrichment of the outer leaflet lipid SM and the inner leaflet lipids PS and plasmalogen-PE (30–32, 34, 35, 88). Both because high SM content is considered a characteristic of lipid rafts (27) and because, under some conditions, the viral Gag and Env proteins fractionate with raft markers in lysed cells (89, 90), retroviruses sometimes have been said to “bud from rafts” (27–29, 89). However, since rafts in living cells are poorly defined and nanoscopic (3) and since lipid mixtures like that found in the inner leaflet do not form an Lo phase, the meaning of a putative raft association is unclear. The mechanism of Gag-membrane association probably is simplest for RSV, since this protein is unusual in not carrying a critical hydrophobic N-terminal myristate modification. From in vitro experiments, the isolated RSV MA domain previously was shown to interact with membranes electrostatically (22). A similar conclusion was reached for RSV Gag in vivo based on mutational analyses and quantitative budding assays (5). However, systematic analyses of the effect of lipid composition on the membrane binding properties of any purified Gag protein have not been reported previously. Lipid composition is known to dramatically influence HIV-1 Gag-membrane interaction in vitro, but the mechanism remains to be clarified (24).

In this study, we examined how membrane charge, acyl chain type, and membrane order influence the membrane binding of purified RSV Gag. Key to interpretation of the results were comparisons with purified nonviral control proteins. The latter included fluorescent proteins with added stretches of 4, 8, or 12 basic residues (charge sensors) or with added domains known to bind to specific lipid headgroups (lipid sensors). Membrane binding for all of the proteins was quantified both by pelleting assay with LUVs and by fluorescence assay with GUVs. Overall, the data showed that membrane interaction of RSV MA, Gag, and the charge sensors was sensitive to ionic strength, to PS concentration, and, for the charge sensors, also to the number of basic residues (5). Gag bound more strongly to acidic membranes than did MA, as shown previously (21, 53), presumably due to its ability to multimerize. Overall, these results confirm that electrostatics provides the primary driving force for RSV Gag membrane association. We also found that Gag binding was enhanced by cholesterol. Gag preferred membranes with phospholipids containing different acyl chain saturations, possibly due to nonideal mixing (e.g., DSPC/DOPS/Chol or DOPC/DPPS/Chol). In contrast, the control proteins carrying the MARCKS peptide (mNG-MARCKS) or the PS headgroup-specific binding domain [mNG-(Ev2)₂] preferred membranes with unsaturated lipids. Finally, Gag and the charge sensors showed no preference for liquid-ordered (Lo, i.e., raft-like) versus liquid-disordered (Ld) phases when these phases had the same PS composition, in striking contrast to mNG-(Ev2)₂, for which Ld binding was at least 10-fold higher than for Lo.

Specially constructed fluorescent protein charge sensors have been used previously for in vivo and in vitro studies of PM binding proteins (79, 91). The mNG-KR4, mNG-KR8, and mNG-KR12 proteins obviously differ from MA or Gag in that the basic residues are sequential in the primary amino acid sequence and thus presumably are unstructured. In contrast, the basic residues in the membrane interacting face of MA domains of retroviruses are largely fixed and mostly not contiguous. Thus, detailed comparisons of the behavior of MA and the charge sensors may not be warranted. Nevertheless, of the three charge sensors that we constructed, the one with eight basic residues, mNG-KR8, in most assays behaved most similarly to MA-(mNG), consistent with the presence of up to 6 membrane-interacting lysine residues on the surface that is inferred to interact with the bilayer (5) (M. Doktorava, personal communication). For the purposes of this study, our first-order assumption was that K and R residues would be functionally similar. However, in fact the two proteins with lysine or arginine homopolymeric basic residues, mNG-K8 and mNG-R8, were not identical in their binding properties, with mNG-R8 interacting more tightly with LUVs (data not shown). The mechanism underlying these differences is unclear.

Previous studies that sought to compare retroviral protein binding to Lo and Ld phases did not take into account the possibly different PS concentrations in each phase (60). For proteins with basic patches that bind to acidic lipids like PS, it is not possible to interpret preference for one phase over another without knowing how much PS is in each phase. It is not trivial to establish conditions in which Lo and Ld have the same PS concentration. In our studies, we started with a published phase diagram without PS (36, 48) and then performed FRET measurements to determine the new phase boundaries when PC was partially replaced with PS along a chosen tieline. The resulting phase boundaries then were used to find lipid compositions in which the PS concentration was identical in Ld and Lo. To our knowledge, this is the first study to systematically and rigorously compare the roles of membrane charge and membrane order on the interaction of a purified retroviral Gag protein with membranes.

Why does mNG(Ev2)₂ strongly prefer Ld over Lo phases, even when the concentration of the PS headgroups to which the protein binds is the same? From the crystal structure of Ev2 bound to PS (77), we speculate that hydrophobic residues in Ev2 that are close to the membrane become inserted in a shallow manner into the inner core of the bilayer, and that this insertion is disfavored in the Lo phase. Alternatively, perhaps the headgroups of PS in Ld have a slightly different conformation than those in Lo, for example, with different conformation around the glycerol backbone, which somehow disfavors binding. The former model seems more plausible, since mNG-MARCKS also shows preference for Ld, although this preference is less extreme. The MARCKS peptide comprises not only 13 basic residues, which presumably are unstructured, but also 5 phenylalanine residues that might insert into the membrane (63, 85, 92). Mutational studies could address the mechanism of Ld preference for these two proteins.

The effects of cholesterol and acyl chains on binding of RSV Gag and the control proteins are difficult to interpret in a simple mechanistic way. While no detailed model fully accounts for the distinctive acyl chain preferences of Gag and the control proteins, we do have such a model for the cholesterol effect on MA binding, based on all-atom molecular dynamics simulations of membrane with bound protein and on calculations of surface potentials (M. Doktorova, personal communication). Cholesterol condenses the phospholipids in a bilayer, leading to an increase in charge density in the plane of the headgroups, thereby increasing negative surface potential.

Consistent with previous reports that PI(4,5)P₂ enhanced HIV-1 Gag membrane binding (24), PI(4,5)P₂ also enhanced the membrane binding of all proteins tested. There is a known PI(4,5)P₂ binding pocket in HIV-1 Gag but no known PI(4,5)P₂ binding pocket in RSV Gag. Here, we found that 2 mol% PI(4,5)P₂ not only enhanced the membrane binding of the PI(4,5)P₂ sensor PH that has a specific binding pocket but also enhanced the membrane binding of the charge sensors, such as mNG-KR12 and mNG-MARCKS, which have no structured binding pocket. The finding for MARCKS is also consistent with the previous finding that the MARCKS peptide itself responds to PI(4,5)P₂ significantly (63, 75, 76, 85, 93). PI(4,5)P₂ has 3 to 4 negative charges, which creates higher local charge density than PS. PI(4,5)P₂ can enhance protein membrane binding through electrostatic interactions or through specific binding pockets. Even though it is still not clear whether RSV Gag has a PI(4,5)P₂ binding pocket, we showed that a specific binding pocket is not required for a protein to be highly responsive to the presence of PI(4,5)P₂. We also observed that MA-mNG and MASP⁺⁵-mNG bound to membranes to a similar level in the absence of PI(4,5)P₂, whereas MASP⁺⁵-mNG and hexameric MA showed higher membrane binding than monomeric MA in the presence of PI(4,5)P₂, suggesting multimerization enhances membrane binding or PI(4,5)P₂ enhances multimerization.

In our study, we used both pelleting assays with LUVs and fluorescence assay with GUVs to study protein membrane interactions. The majority of both results were comparable. For example, when examining whether protein binding follows membrane charge or membrane order, we tested protein binding using Ld and Lo coexisting GUVs and uniform LUVs that mimic each phase. Both assays gave the same readout of a high PS Ld/high PS Lo protein-bound ratio. We also employed both assays for probing the effect of PI(4,5)P₂, in which PI(4,5)P₂ was shown to increase membrane binding of all tested proteins. The similarity in results between GUVs and LUVs implies that membrane curvature is not a major factor affecting protein binding. However, the two assays also have a few minor discrepancies. Most liposome-pelleting assays were done at 150 mM NaCl, whereas for technical reasons all of the protein GUV binding assays were done at 50 mM NaCl. This difference in ionic strength probably explains why PI(4,5)P₂ enhanced Gag-mNG to different levels. The two membrane binding assays complement each other. On the one hand, LUV pelleting has the advantage of being more efficient, flexible, and rapid at testing various protein-membrane combinations. LUVs also are less likely to have composition variation, since LUVs are prepared through RSE, which is designed to bypass the dry lipid film state to avoid lipid demixing (68, 94). On the other hand, GUV image analysis not only enables direct visualization of protein membrane binding but also enables quantification of simultaneous and competitive protein binding.

In summary, we employed biochemical and biophysical approaches to elucidate that membrane charges provide RSV Gag the primary driving force for associating with membranes, while membrane order barely affects RSV membrane binding. However, membrane order clearly does influence the membrane association of the cellular protein domain MARCKS and the PS sensor Evectin2. The results shown here for RSV will serve as a platform for further studies with other retroviral proteins like HIV-1 Gag, as well as cellular proteins that interact with the inner leaflet of the PM.