pH-dependent Fusion Induced by Vesicular Stomatitis Virus Glycoprotein Reconstituted into Phospholipid Vesicles*

Purified G-protein from vesicular stomatitis virus was reconstituted into egg phosphatidylcholine vesicles by detergent dialysis of octyl glucoside. A homo- geneous population of reconstituted vesicles could be obtained, provided the protein to lipid ratio was high (about 0.3 mol 70 protein) and the detergent removal was slow. The reconstituted vesicles were assayed for fusion activity using electron microscopy and fluorescence energy transfer. The fusion activity mediated by the viral envelope protein was dependent upon pH, temperature, and target membrane lipid composition. Incubation of reconstituted vesicles at low pH with small unilamellar vesicles containing negatively charged lipids resulted in the appearance of large coch- leate structures, as shown by electron microscopy using negative stain. This process did not cause leakage of a vesicle-encapsulated aqueous marker. The rate of fusion was pH-dependent with a pK of about 4 and the apparent energy of activation for the fusion was 16 f 1 kcal/mol. G-protein-mediated fusion showed a large preference for target membranes which contain phosphatidylserine or phosphatidic acid. Inclusion of 36% cholesterol in any of the lipid compositions had no effect on the rate of fusion. These reconstituted vesicles provide a system to study the mechanism of pH-de- pendent fusion induced

Purified G-protein from vesicular stomatitis virus was reconstituted into egg phosphatidylcholine vesicles by detergent dialysis of octyl glucoside. A homogeneous population of reconstituted vesicles could be obtained, provided the protein to lipid ratio was high (about 0.3 mol 70 protein) and the detergent removal was slow. The reconstituted vesicles were assayed for fusion activity using electron microscopy and fluorescence energy transfer. The fusion activity mediated by the viral envelope protein was dependent upon pH, temperature, and target membrane lipid composition. Incubation of reconstituted vesicles at low pH with small unilamellar vesicles containing negatively charged lipids resulted in the appearance of large cochleate structures, as shown by electron microscopy using negative stain. This process did not cause leakage of a vesicle-encapsulated aqueous marker. The rate of fusion was pH-dependent with a pK of about 4 and the apparent energy of activation for the fusion was 16 f 1 kcal/mol. G-protein-mediated fusion showed a large preference for target membranes which contain phosphatidylserine or phosphatidic acid. Inclusion of 36% cholesterol in any of the lipid compositions had no effect on the rate of fusion. These reconstituted vesicles provide a system to study the mechanism of pH-dependent fusion induced by a viral spike protein.

Enveloped viruses
often contain surface glycoproteins which participate in the attachment and penetration processes of viral infection (1)(2)(3)(4)(5). VSV' is a rhabdovirus containing one glycosylated protein, G-protein, which is vital for viral infectivity. Removal of G-protein by trypsinization (6)(7)(8) or reacting VSV with anti-G-protein antibodies (9) results in a marked decrease in viral infectivity. One of the functions of the G-protein is to facilitate attachment of the virus to the cell surface. Although the precise cellular attachment site for VSV is unknown, the finding that VSV attachment is unaf-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore he hereby marked "aduertisernent" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
$ Recipient of the Chaim Weizmann postdoctoral fellowship award.
VSV associates with "coated pit" regions of the plasma membrane (11)(12)(13) and is internalized into the cells via an endocytic pathway morphologically identical with that of receptor-bound ligands (13,14). Upon endocytosis, the virus is transrerred into cytoplasmic vacuoles termed "endosomes" or "receptosomes" (15, 16), which are rapidly acidified (17). This pH change apparently initiates VSV fusion with the endosome membrane and release of the viral nucleocapsid into the cell cytoplasm. Evidence exists that acidic pH initiates the second function of G-protein, that of a membrane fusogen: low pH can induce direct VSV-cell membrane fusion as well as cell-cell fusion of VSV-infected cells (18).
The purpose of this study was to construct an in uitro system for analyzing the role of G-protein in membrane fusion. The first step was to reconstitute the G-protein in phospholipid vesicles. Petri and Wagner (19) have shown that VSV G-protein can be reconstituted in egg phosphatidylcholine vesicles by detergent dialysis using octyl glucoside. The insertion and orientation of G-protein in those reconstituted vesicles is the same as in intact virions (19). Similar reconstituted vesicles can inhibit VSV infectivity (21).
In this study we found that a homogeneous population of reconstituted vesicles (called virosome (22)) could be obtained by dialysis of octyl glucoside, provided the protein:lipid ratio was high (about 0.3 mol % protein) and the detergent removal was slow. Incubating virosomes with target vesicles containing negatively charged lipids and lowering the pH resulted in massive fusion as shown by electron microscopy. We examined the mixing of membrane lipid resulting from membrane fusion by monitoring the efficiency of resonant energy transfer between two fluorescent lipid probes (23) incorporated in the same virosome membrane.

Preparation of VSV G-protein
Reconstituted Vesicles (Virosomes)-Reconstitution of purified G-protein into lipid vesicles was performed by mixing purified G-protein with dried lipids as described under "Experimental Procedures" and dialyzing out the detergent. The reconstituted preparation was evaluated by two criteria: (a) homogeneity of vesicle size and G-protein distribution and (b) fusogenic activity at low pH. The first criterion was emphasized because a homogeneous population is required for a valid interpretation and analysis of fusion data.
Preliminary attempts to prepare virosomes by dialysis against large volumes (>lOOO:l) of buffer produced populations of proteolipid vesicles which were inhomogeneous, as judged from negative staining electron microscopy. Vesicle sizes ranged from 20 nm to more than 200 nm and the distribution of G-proteins was not uniform: there were vesicles containing no protein while other vesicles contained dense patches of G-protein.
We therefore developed a protocol of slow dialysis against small volumes with about 10 changes of buffer. The rationale for this method is that the slow removal of octyl glucoside allows ample time for interactions of detergent-lipid-protein micelles to occur before the structures close to form vesicles upon the final elimination of octyl glucoside (29-32). Slow dialysis produced vesicles with a fairly uniform size distribution, but the distribution of G-protein in the vesicles was dependent upon the starting G-protein to lipid ratio. At low G-protein to lipid ratio, there was a subpopulation of vesicles containing large amounts of spikes, while other vesicles lacked spikes observable by negative staining (data not shown). This was similar to the behavior found by Petri and Wagner (19) in the reconstitution of the VSV Gprotein and by Rivnay and Metzger (32) in the reconstitution of IgE receptor. It seems to indicate that the proteolipid vesicles are formed with a defined ratio of G-protein to phospholipid, while the excess lipid forms protein-free vesicles. Homogeneous preparations were obtained when the Gprotein to lipid ratio was similar to that of the native virus envelope (about 1 mol %).* The virosomes obtained are shown in Fig. 1A. The vesicles had a fairly uniform size and a homogeneous distribution of G-protein in their membranes. The average area of a virosome was 5.74 x 10e3 pm2 as measured using a Hewlett-Packard digitizer and computer. pH-dependent Association of Virosomes and Vesicles-Incubating virosomes with small unilamellar PS:PC vesicles and lowering the pH caused massive vesicle-vesicle association as indicated by changes in light scattering (not shown) and in sedimentation rate of vesicles as shown by Table I and centrifuged at 120,000 x g,, for 15 min. When the incubation was done at pH 7.4, less than 2% of the i4C label were pelleted. However, at pH 3.0, over 90% of the vesicles were pelleted. There was no pelleting of vesicles in the absence of virosomes at either pH value. Under these centrifugation conditions, about 55% of the 3H-virosomes were pelleted by themselves at pH 7.4 either with or without added vesicles and at pH 3.0 without added vesicles (see Table  I), whereas at pH 3.0 about 97% of the 3H label were pelleted (together with the 'Y?-vesicles). This indicates that the G- protein was part of an enlarged virosome-vesicle structure. The virosome surface could only accommodate 3-4% of the total amount of small unilamellar vesicles initially present. Therefore, those large structures probably represent fused rather than aggregated structures. We used ultrastructural examinations to study this process in more detail.
Electron Microscopy Studies of Virosome-Vesicle Fwion-Virosomes (Fig. 1A) were incubated with small unilamellar PS/PC (1:l) vesicles (Fig. 1B). Lowering the pH to 3.3 resulted in the appearance of large, irregularly sized, cochleate structures (Fig. IC). The   dipalmitoylphosphatidylcholine, or both, were suspended in airfuge tubes containing 150 pl of saline (145 mM NaCI, 10 mM Hepes, 10 mM Mes) set to the indicated pH value. An aliquot of the suspension was taken for counting of the radioactivity of the two labels (total). The amount of label in the pellet was calculated from the difference between the total counts and the counts in the supernatant after centrifugation of the samples (120,000 X g, 15 min). There was no fusion of virosomes alone incubated at pH 3.3 (data not shown). In order to follow the process of fusion more closely, we incubated virosomes with large vesicles which were clearly distinguishable from virosomes (Fig. 2). Lowering the pH resulted in massive fusion, with little resolution of the structures formed. Incubating the virosomes with target vesicles a t neutral pH, however, led to fusion at a very slow rate (see below). Fig. 2, A-C shows the fusion steps between virosome and large vesicle. In Fig. 2 A , they are separate, in Fig. 2 Fig. 2C the virosome and target vesicle appear to be in the early fused state.

adherent, and in
The observed fusion process was studied quantitatively by examining the mixing of virosome and target liposome membrane lipids (which results from fusion) by resonance energy transfer between two lipid probes incorporated into the virosome.
Kinetics of Virosome-Vesicle Fusion-Quantitation of virosome-vesicle fusion was performed by continuously monitoring the efficiency of energy transfer between two fluorescent lipid probes incorporated into virosomes: N-NBD-PE (donor) and N-Rho-PE (acceptor) (see "Experimental Procedures"). The fluorescently labeled virosomes were then mixed with various amounts of PC:PS (4:l) vesicles and the intensity of NBD fluorescence was measured. The recordings of NBD fluorescence intensity with time are shown in Fig. 3. pg of lipid) were injected into a fluorimeter cuvette containing 2.5 ml of buffer (145 mM NaCI, 10 mM Hepes, pH 7.4) and the indicated amounts of PS:PC (4:l) small unilamellar vesicles. Fluorescence intensity was measured after acidification of the medium to pH 3.3 as described under "Experimental Procedures." The shutter of the photomultiplier compartment was closed for the first few minutes during the mixing of HCI with medium. The traces show an initial rise which is background fluorescence. The fluorescence intensity after an hour is indicated at the left and in the inset. a. u., arbitrary unit.
After addition of HCl to adjust the pH to 3.3, the fluorescence intensity increased, indicating that the fluorescent labels had been diluted by fusing with unlabeled liposomes. Both the rate of increase and the total change in fluorescence was dependent on the amount of target lipid present. The inset shows the fluorescence intensity change after 1 h when most of the fusion reaction has gone to completion. As a quantitative measure of fusion rates, we used in subsequent experiments the parameter tH, the time needed to attain the halfmaximum fluorescence change (see "Experimental Proce-dures"). Under the experimental conditions described in Fig.  3, the half-maximum fluorescence change corresponded to the dilution of virosome lipids with about 40 pg of target lipid, reducing the probe surface density from 1% to 0.3%.
pH Dependence of Virosome-Vesicle Fusion- Fig. 4 shows the rates of fusion between virosome and PC/PS (1:l) vesicles incubated together at various pH values. There was no significant change in fusion rate when the pH was lowered to pH 5.5. However, there was a marked increase as the pH was lowered further. From Fig. 4, we deduce that the apparent pK for G-protein-mediated fusion is lower than pH 4. A similar pH profile was noted with PC/PA (1:l) vesicles (not shown). Fig. 5 shows that the G-protein was not inactivated by exposure to low pH: adding HCl to a final pH of 3.3 initiated fusion and titration back to pH 7.0 by NaOH stopped the fusion reaction. Subsequent additions of HC1 and NaOH caused the fusion to start again or to stop, respectively.
Dependence of Fusion on Lipid Composition- Fig. 6 shows the fusion rate as a function of PS content in mixed PC:PS target vesicles. Addition of up to 10% PS had little effect on the rate of fusion. Further increases caused a progressive increase in fusion rate. For target vesicles containing 50% PS and 50% PC, the rate of virosome fusion was found to be 22-25-fold increased relative to the background level of fusion with pure PC vesicles. Table 11 shows the fusion rates for vesicles with different lipid compositions at a given ratio of phospholipid to PC. Fusion rate was enhanced 120-fold by substituting 50% of the phosphatidylcholine in the vesicle membranes with phosphatidic acid and 25-fold with 50% phosphatidylserine. Phosphatidylinositol appeared to enhance the rate slightly; replacement with 50% phosphatidylethanolamine had no effect. Inclusion of 36% cholesterol in any of these vesicle compositions also had no effect.
Fusion Is Dependent upon G-protein-Several experiments were designed to verify that the fusion of virosomes to vesicles was dependent upon the presence of functional G-protein.

TABLE I1
Target membrane phospholipid selectivity for G-protein-mediated fusion The fusion of virosomes and sonicated vesicles of the indicated composition was monitored by following the increase of NBD fluorescence as described under "Experimedntal Procedures." The measured half-times for fluorescence increase are shown for vesicles either with or without added cholesterol. sulted in formation of PC vesicles which did not fuse with PS/PC vesicles at any pH (data not shown). In fact, if the initial protein/lipid ratio was too low or the initial detergent removal too fast, the resulting virosomes were fusion-incompetent. ( b ) When fusion-competent virosomes were mildly treated with trypsin (50 Fg/ml, 30 min, room temperature), the rate of fusion was slowed down to about 70% of the original value (data not shown). (c) Virosomes treated with a 1:50 dilution of anti-G-protein anti-serum (20) (see "Experimental Procedures") were inhibited by more than 90% in their fusion ability (Fig. 7). Temperature Dependence-The fusion of PC/PS (1:l) vesicles with virosomes was also studied at various temperatures. An Arrhenius plot of the dependence of fusion rate on temperature (Fig. 8) showed a single linear slope between 0 and 30°C. This indicates that there was no change in membrane structure in that temperature range that affects fusion. The calculated energy of activation was 16 & 1 kcal/mol.
Retention of Vesicle Contents during Fusion-To test whether the fusion process was "leaky," we prepared target vesicles (PS:PC, 1:1) containing self-quenching concentrations of lucifer yellow. The method to monitor release of contents was similar to that used for vesicles containing carboxyfluorescein (36,37). Upon release of the contents, the dye is diluted into a larger volume and consequently an increase in fluorescence is measured due to relief of selfquenching. We used lucifer yellow for these experiments since this compound did not leak spontaneously upon lowering the pH. Incubating virosomes with those lucifer yellow-containing vesicles under the same conditions as shown in Fig. 3 resulted in no increased fluorescence (data not shown). This indicated that virosome-vesicle fusion was non-leaky.

DISCUSSION
The results of the present study indicate that VSV Gprotein can be functionally reconstituted into lipid vesicles. Although reconstitution of G-protein into lipid vesicles had been reported previously (19,21), no functional (fusion) studies were performed. In order to obtain a functional reconstitution product, appropriate reconstitution conditions and techniques must be chosen. The underlying processes which lead to the formation of a desired reconstitution product are not well understood. However, a number of detailed reconstitution studies published recently (29-32) have guided us in these experiments. Jackson and Litman (30) have shown that in a micellar mixture of phospholipid, octyl glucoside, and membrane protein, the octyl glucoside-phospholipid micelles are less stable than the protein-lipid-detergent micelles. Therefore, lowering the octyl glucoside concentration rapidly by dialysis against a large volume of buffer results in the formation of protein-depleted phospholipid vesicles. As the detergent concentration is further decreased, the micelles containing protein become unstable and reconstituted vesicles are formed. Petri and Wagner (38) in fact have shown that G-protein can partition spontaneously from glycoprotein micelles into preformed sonicated vesicles. This model for asymmetric membrane reconstitution had also been proposed by Helenius et al. (31). However, we found in this study that virosomes reconstituted in this fashion (i.e. fast dialysis) were not fusogenic.
One important feature which was necessary for successful reconstitution of G-protein was the slow removal of octyl glucoside. We performed an initial 1:3 dilution of the detergent followed by a number of changes of the dialyzing solution gradually decreasing the concentration of octyl glucoside in the dialysate over a period of 3-4 days. Since the initial concentration of octyl glucoside was 60 mM, the initial dilution yields 20 mM, which is about the critical micellar concentration of octyl glucoside (39). The protein to lipid ratio was also important for proper reconstitution. The ratio of Gprotein to phospholipid in the virosomes was about equal to that in the original VSV membrane,' suggesting that proteinprotein interactions have an important role in functional reconstitution.
In this study, we demonstrated pH-dependent, non-leaky fusion of the virosomes with negatively charged lipid vesicles. Fig. 1C G-protein we believe that the structures are not sheets but flattened bags which role up into spirals. The electron micrographs also show closed structures at the ends of the swirls (Fig. IC). The exact nature of the fusion product needs to be more closely examined by freeze-cleavage and thin section electron microscopy.
The fusion of virosomes with negatively charged lipids was pH-dependent with a pK around 4.0. This pattern differs from that found by White et al. (18) for pH dependence of VSV-induced cell fusion (pK around 6.0). In those experiments, fusion of virus with cells is induced by lowering the pH of the medium for 30-60 s. The cells subsequently fuse into giant polykaryons. The difference in pH profile might be due to the fact that we work with a reconstituted system. Moreover, White et al. (18) consider extent of fusion, whereas we monitor rates of fusion. An interesting feature of the fusion reaction was that it could be turned on and off by lowering and raising the pH (Fig. 5). Although the fused particles (the product of the reaction) were irreversible structures in that they did not fall apart upon raising the pH, the activation of fusion was reversible. This indicates that the G-protein was not denatured by exposure to the low pH.
G-protein-mediated fusion showed a large preference toward target membranes containing lipids which are negatively charged at neutral pH or uncharged at the pH where fusion occurs ( Table 11). Although phosphatidylethanolamine facilitates Caz+-induced fusion of phospholipid vesicles (40), consistent with the greater ability to remove the water of hydration from its head group, it did not enhance G-proteinmediated fusion. Probably this was due to the positive charge on the head group at the low pH. The greater fusion capacity of PA over PS and PI (Table 11) is consistent with data on Ca2+-induced fusion of vesicles containing those head groups (41) and on the relative hydration energies of those head groups. Table I1 also shows that cholesterol was not required for fusion. This is in contrast with the absolute requirement for cholesterol in the target membrane for Semliki forest virus fusion (42). On the other hand, influenza virus does not require cholesterol in its target membrane (43, 44). The specificity for fusion of the negatively charged lipids in the target membrane contrasted with the data on binding of lipids to the virus, which shows specificity for PS over all other negatively charged lipids (10). However, binding experiments were done at neutral pH. At low pH, the G-protein appears to bind (and fuse) to lipids which are uncharged at that pH.
Several mechanisms can explain the pH-dependent activtion step. One is a pH-dependent conformational change in the protein leading to the exposure of a "fusion peptide" as in the case of influenza virus (45). Other possibilities include the titration of amino acid residues on the protein and/or head groups on the target phospholipid resulting in association of virosome and target membranes. The pK of about 4 is consistent with the titration of carboxylic groups of glutamic and aspartic acids. We hope to follow possible structural changes in the G-protein with pH and correlate this with its fusogenic activity.