The in vitro generation of post-Golgi vesicles carrying viral envelope glycoproteins requires an ARF-like GTP-binding protein and a protein kinase C associated with the Golgi apparatus.

We have developed a system that recreates in vitro the generation of post-Golgi vesicles from an isolated Golgi fraction prepared from vesicular stomatitis virus- or influenza virus-infected Madin-Darby canine kidney or HepG2 cells. In this system, vesicle generation is temperature- and ATP-dependent and requires a supply of cytosolic proteins, including an N-ethylmaleimide-sensitive factor distinct from NSF. Cytosolic proteins obtained from yeast were as effective as mammalian cytosolic proteins in supporting vesicle formation and had the same requirements. The vesicles produced (50-80 nm in diameter) are depleted of the trans Golgi marker sialyltransferase, contain the viral glycoprotein molecules with their cytoplasmic tails exposed, and do not show an easily recognizable protein coat. Vesicle generation was inhibited by brefeldin A, which indicates that it requires the activation of an Arf-like GTP-binding protein that promotes assembly of a vesicle coat. Vesicles formed in the presence of the nonhydrolyzable GTP analogue guanosine 5′-3-O-(thio)triphosphate retained a nonclathrin protein coat resembling that of COP-coated vesicles, and sedimented more rapidly in a sucrose gradient than the uncoated ones generated in its absence. This indicates that GTP hydrolysis is not required for vesicle generation but that it is for vesicle uncoating. The activity of a Golgi-associated protein kinase C (PKC) was found to be necessary for the release of post-Golgi vesicles, as indicated by the capacity of a variety of inhibitors and antibodies to PKC to suppress it, as well as by the stimulatory effect of the PKC activator 12-O-tetradecanoylphorbol-13-acetate.

Within the endomembrane system, proteins are transferred from one subcellular compartment to another by means of membrane vesicles that are formed in the donor compartment, move through the cytoplasm, and fuse with the membrane of the specific acceptor organelle (for reviews, see Refs. 1 and 2). The production of transport vesicles has been examined in greatest detail for the COPI-and COPII-coated vesicles that mediate intra-Golgi and endoplasmic reticulum to Golgi transport, respectively, and for the clathrin-coated vesicles that mediate endocytosis at the plasma membrane or are responsible for the transport of newly synthesized lysosomal hydrolases from the TGN 1 to incipient lysosomes (for reviews see Refs. [3][4][5][6]. These studies indicate that the first stage in vesicle formation is the assembly of a protein coat on the donor membrane from macromolecular complexes that are recruited from the cytosol, as a result of the activation of a small GTP-binding protein (i.e. Arf for the assembly of COPI and clathrin coats in the Golgi (7,8), and Sar1p for the assembly of the COPII coats in the endoplasmic reticulum (9,10)). In the case of clathrincoated vesicles, adaptor molecules that recognize the cytoplasmic tails of the proteins to be transported facilitate the assembly of the coat and provide a mechanism for the incorporation of specific cargo molecules into the vesicles (see Ref. 5). In all cases, the coat is thought to serve as a mechanochemical device that induces the membrane curvature necessary to generate a vesicle.
Much less is known about the formation in the TGN of the post-Golgi vesicles that carry membrane proteins or constitutively secreted proteins to the plasma membrane. Even the nature of the coat that generates the vesicles remains to be definitively established, and it seems likely that new kinds of adaptors or coats may be involved in selecting specific proteins for incorporation into their carrier vesicles (11,12). It is clear that progress in this field should benefit greatly from studies with cell-free systems for post-Golgi vesicle generation that should allow the identification of those molecules (proteins, lipids, and cofactors) that participate in or regulate the process of vesicle formation.
One cytosol-dependent cell-free system has been developed that effects the in vitro production of post-Golgi vesicles from a rat liver Golgi fraction bound to magnetic beads, from which the released vesicles that carry constitutively secreted proteins and membrane proteins could easily be separated (13,14). Using this system, it was possible to show that both types of proteins are incorporated into the same post-Golgi vesicles (13) and that specific transmembrane (TGN38) and cytosolic (p62) proteins participate in the process of vesicle generation. Moreover, the cyclic phosphorylation/dephosphorylation of p62 was suggested to be essential for its cyclic function. The in vitro production of post-Golgi vesicles of the constitutive and regulated secretory pathways has also been achieved using a post-nuclear supernatant from PC12 cells (15) or, more recently, a crude membrane preparation containing TGN membranes supplemented with a cytosolic fraction (16). This allowed the demonstration that the production of both types of vesicles is controlled by heterotrimeric G proteins of both the G i and G s classes, that a cytosolic phosphoprotein is likely to modulate the activity of the G proteins (16), and that phosphoinositides are likely to participate in vesicle formation (17).
Protein phosphorylation/dephosphorylation events have also been implicated in the control of several other steps of intracellular transport, although the precise roles of the proteins that serve as substrates for these reactions and the identity of the modifying enzymes is, in most cases, not yet known. Transport from the medial Golgi to the TGN also requires a protein kinase activity at an early stage, possibly in COP-coated vesicle formation (18), which may be fostered specifically by protein kinase C (PKC) (19). This is thought to explain the increase in constitutive secretion observed in various cell types, including MDCK, after stimulation of this enzyme (19). Stimulation of PKC has also been shown to selectively enhance the processing of the ␤-amyloid precursor along one of two alternative pathways (20,21), and recently, PKC was shown to stimulate the release of post-Golgi vesicles containing the ␤-amyloid precursor from a PC12 cell membrane fraction (22). On the other hand, protein dephosphorylation is required for endoplasmic reticulum to Golgi transport (23) and its inhibition may be related to the arrest of vesicular transport that takes place in mitotic cells. The fragmentation (vesiculation) of the Golgi apparatus that occurs in mitosis also requires the activity of a kinase, the Cdc2 kinase (24), and transport from the Golgi apparatus to the yeast vacuole requires a specific protein kinase that activates the phosphatidylinositol 3-kinase (25,26). Protein kinases have also been implicated in polarized secretion (27,28) and in transcytosis in epithelial cells (29,30).
In this paper we describe a cell-free in vitro system that employs purified Golgi fractions isolated from virus-infected MDCK or HepG2 cells and a cytosolic protein fraction to reproduce post-Golgi vesicle production. We show that the formation of post-Golgi vesicles containing the sialylated VSV-G protein, a molecule widely used as a paradigm for studies of intracellular protein transport (4), involves the assembly of a protein coat in the TGN membrane, in a process that is promoted by GTP and requires the participation of an Arf-like protein, as revealed by the inhibitory effect of brefeldin A. GTP hydrolysis, however, is not required for post-Golgi vesicle generation, but it is for vesicle uncoating, since in the presence of GTP␥S all the vesicles found in a fraction that contained the released VSV-G protein remained coated. In addition, a Golgi-associated PKC activity is shown to play an important role in vesicle generation, since this could be suppressed by various specific PKC inhibitors and enhanced by the PKC stimulator 12-O-tetradecanoylphorbol-13-acetate (TPA).

Preparation of a Golgi Fraction
Containing Labeled Viral Glycoproteins-Cultures of MDCK or HepG2 cells were infected with vesicular stomatitis (VSV, Indiana Serotype) or influenza (strain A, PR8) virus, and pulse-labeled and chased as described (31). A Golgi fraction was prepared (32), and aliquots (0.1 ml, ϳ100 g of protein; 3 ϫ 10 5 cpm/ml) were stored in liquid nitrogen. Frozen Golgi membranes were as efficient as fresh membranes and had the same requirements for post-Golgi vesicle production. Densitometry of fluorographs obtained after gel (10% polyacrylamide) electrophoretic analysis indicated that greater than 95% of the radioactivity in Golgi fractions from VSVinfected cells corresponded to the VSV-G protein (see Fig. 3A, inset). For studies using yeast cytosolic proteins, the Golgi fraction was prepared using maltose instead of sucrose (33).
Preparation of a Cytosolic Protein Fraction-Sprague-Dawley (male) rats (200 -250 g) were starved overnight and, after sacrifice, the livers were excised and homogenized (1:10, w/v) with a type B Dounce homogenizer in a cold (4°C) buffer containing 0.2 M sucrose, 20 mM Hepes-KOH, pH 7.3, 1 mM dithiothreitol, 10 units/ml trasylol, and 1 mM phenylmethylsulfonyl fluoride. The homogenate was centrifuged for 30 min at 20,000 ϫ g and the resulting supernatant for 2 h at 150,000 ϫ g in a Beckman Tl60 rotor to obtain the high speed supernatant from which total cytosolic proteins were precipitated by the addition of solid ammonium sulfate to 100% saturation. The proteins were resuspended in 1 ⁄10 volume of homogenization buffer lacking sucrose, dialyzed against this buffer, clarified by centrifugation, concentrated to 60 mg/ml protein by ultrafiltration in a Centriprep-10 apparatus (Amicon, Beverly, MA), and stored at Ϫ80°C in small aliquots.
A yeast cytosolic protein fraction was prepared from Saccharomyces cerevisiae (strain X2180 -1A), essentially as described (34). Membranes and organelles were removed by high speed centrifugation (150,000 ϫ g for 60 min in a Ti60 rotor), and the cytosolic proteins were precipitated with 100% saturated ammonium sulfate and processed as described above for the rat liver cytosolic protein fractions.
Vesicle Release Assay-The standard incubation mixture contains (in 40 l) 5 l of Golgi fraction, 20 l of a 2-fold concentrated assay buffer containing 120 mM potassium aspartate, 1 mM EDTA, 6 mM MgCl 2 , 20 mM Hepes-KOH, pH 7.3, and an energy regenerating system (4 mM ATP, 2 mM GTP, 2 mM UTP, 50 mM creatine phosphate, and 2 mg/ml creatine phosphokinase), 10 l of rat liver cytosolic proteins, and 5 l of a solution containing 0.8 M sucrose, 20 mM Hepes-KOH, pH 7.3. The sucrose concentration in the initial Golgi fraction was close to 0.8 M, so that the final sucrose concentration in the assay mixture was approximately 0.2 M. The mixtures were incubated at 37°C for times ranging from 0 -2 h, and the reaction was terminated by cooling in ice. The Golgi membranes were then removed by sedimentation at 10,000 ϫ g for 10 min, and the release of labeled viral glycoprotein in the supernatant was measured by liquid scintillation counting. Release is expressed as the percentage of total labeled protein initially present in the Golgi fraction that appeared in the supernatant. Temperature-dependent release is the difference between the radioactivity released at 37°C and that released by an equivalent sample incubated for the same time at 4°C.
Purification of the Released Vesicles by Flotation in Sucrose Density Gradients-Pooled supernatant fractions (a total of 12 ml) obtained after the in vitro incubation for post-Golgi vesicle formation were mixed with an equivalent volume of 2.6 M sucrose, 20 mM Hepes-KOH, pH 7.3, 1 mM dithiothreitol, 2 mM EDTA and transferred to six SW41 centrifuge tubes. The samples (4 ml each) were then overlaid with two layers of 6 and 1.5 ml of 1.2 and 0.6 M sucrose, respectively, both in 20 mM Hepes-KOH, 1 mM dithiothreitol. After centrifugation for 40 h at 150,000 ϫ g at 4°C, the material accumulated at the 0.6/1.2 M sucrose interface was collected in 1.5 ml for each tube. The sucrose concentration was measured by refractometry and adjusted to 0.2 M by the addition of cold PBS. The vesicles were then sedimented (150,000 ϫ g for 20 h in a SW60 rotor), and the pellets were resuspended in a small amount (250 l) of residual supernatant. After dilution to 1 ml each with cold PBS, the suspensions were transferred to high speed polyallomer microfuge tubes and recentrifuged for 5 h at 150,000 ϫ g in a TLA-100.3 rotor (Beckman Instruments, Inc., Palo Alto, CA) to obtain the purified vesicles. These were used for measurements of the sialyltransferase specific activity and for electron microscopy.
Sialyltransferase Assays-These were carried out with asialofetuin as exogenous acceptor (35). Each reaction mixture (100 l) contained 10 l of a Golgi membrane fraction, vesicles, or control buffer; 10 l each of 10% Triton X-100, 1 M cacodylate-NaOH pH 6.6, and 100 mg/ml asialofetuin; 40 l of distilled water; and 20 l (0.1 Ci) of CMP-NAN (sialic-9-3 H-labeled; specific activity, 21.3 Ci/mmol). The mixtures were incubated for 2 h at 37°C. Under these conditions, the amount of [ 3 H]sialic acid incorporated into asialofetuin proceeds linearly with respect to the protein input. The reactions were stopped by the addition of 20% trichloracetic acid and 1% phosphotungstic acid in 0.5 N HCl and processed for liquid scintillation counting, as described (35).
Sialyltransferase released during the in vitro incubation was measured in the vesicles recovered from a supernatant obtained from a incubation mixture containing 10 l of a Golgi fraction and is expressed as a percentage of the activity in this fraction. Sialyltransferase-specific activities were measured in purified vesicles and in a sample of Golgi membranes submitted to the same procedure used to purify the vesicles. Protein concentrations were measured (36) after the addition of 0.1 N NaOH. Specific activities are expressed as IU/mg of protein. An IU corresponds to 1 mol of [ 3 H]sialic acid incorporated in asialofetuin/h. Electron Microscopy-Golgi fractions undergoing the process of vesicle generation were obtained by centrifugation (10 min at 20,000 ϫ g in a TLS-55 swinging bucket rotor) from mixtures (200 l) that were incubated for only 5 min at 37°C. The pellets were fixed with 2% glutaraldehyde, followed by 2% of OsO 4 (both in 0.1 M cacodylate buffer, pH 7.5) and were treated with 1% tannic acid (37), before embedding in epon. Thin sections were doubly stained with uranyl acetate and lead citrate. Control Golgi membranes (50 l) were fixed in suspension with glutaraldehyde and sedimented and processed, as above, with omission of the tannic acid treatment.
For thin section electron microscopy, the purified vesicles, or peak fractions from the velocity gradients (200 l), were fixed in suspension for 1 h at 4°C with 2% glutaraldehyde, sedimented (60 min at 100,000 ϫ g in the TLS-55 rotor), and processed by standard procedures, including tannic acid staining.
For immunogold labeling, aliquots (1 l) of unfixed purified vesicles were spread on carbon-coated Formvar grids, which were then incubated twice, for 60 min each at 4°C: first with an affinity-purified rabbit antibody prepared against the last 17 amino acids of the cytoplasmic tail of the VSV-G protein and then, after washing with PBS, with protein A-gold conjugates 10 nm in diameter (EY laboratories, Inc., San Mateo, CA). Following further washing, the specimens were fixed for 15 min with 2% glutaraldehyde and negatively stained with 1% uranyl acetate.
Protein Kinase Inhibitors-H-89, N-acetylsphingosine, N,N-dimethylsphingosine, ML-7, and the protein kinase C inhibitor peptide (amino acids 19 -36) were obtained from Calbiochem, and Calphostin C, staurosporine, and TPA were from Sigma. A monoclonal antibody to protein kinase C (clone 1.9) that recognizes a common epitope on the catalytic domain of all known protein kinase C isoforms was obtained from Boehringer-Mannhein GmbH (Mannhein, Germany). Stock solutions of calphostin C (1 mM), staurosporine (10 mM), N-acetylsphingosine, and N,N-dimethylsphingosine (10 mM) were prepared in dimethysulfoxide (Me 2 SO) and that of ML-7 (5 mM) in 50% Me 2 SO. H-89 (10 mM) was dissolved in water and the pseudosubstrate PKC inhibitor peptide (1 mM) was dissolved in 5% acetic acid before neutralization with NaOH to pH 7.0. Controls and samples containing different concentrations of inhibitors all contained the same amounts of the drug solvent.

Characterization of the Donor Golgi Fraction-
To study the production of post-Golgi vesicles in vitro, donor Golgi fractions were prepared from HepG2 or MDCK cells that were infected with VSV or influenza virus and in which the exit of the pulse-labeled, newly synthesized, viral envelope glycoproteins G or HA from the Golgi apparatus had been prevented by incubating the cells under chase conditions for a period of 2 h at 20°C (38). Analysis of the Golgi fractions showed that, after this treatment, nearly all the labeled viral glycoprotein molecules (illustrated in Fig. 1A for the VSV-G protein in HepG2 cells) present in the Golgi apparatus bear Endo H-resistant oligosaccharides and, hence, had traversed the medial cisternae. Moreover, a vast majority (Ͼ95%) of the Golgi-associated VSV-G glycoprotein molecules had acquired sialic acid residues, which are known (39,40) to be added in the trans Golgi region and/or TGN, since neuraminidase treatment shifted the isoelectric points of the various charged species of VSV-G protein toward more basic pH values (illustrated for MDCK cells in Fig. 1B).
An examination of the Golgi fraction by electron microscopy (Fig. 2) showed that it primarily consisted of Golgi cisternae that remained in their characteristic stacked configuration, with frequent tubular elements appearing to emerge from the most peripheral cisternae within the stacks. Because of this organization, the Golgi membranes could be sedimented by centrifugation at relatively low g forces (e.g. at 10,000 ϫ g for 10 min), which, after an in vitro incubation, permitted a simple separation of the residual Golgi membranes from any released vesicles (see below).
The in Vitro Release of Viral Glycoproteins from the Golgi Fraction Is a Temperature-and ATP-dependent Process That Can Be Supported by Mammalian or Yeast Cytosolic Proteins-When the Golgi fraction from VSV-infected cells was incubated at 37°C with cytosolic proteins and an energy supply, a sub-stantial proportion (ϳ20 -40%) of the labeled VSV-G protein was released from the sedimentable cisternae and was recovered in the supernatant obtained after centrifugation for 10 min at 10,000 ϫ g (Fig. 3A). A similar release of labeled HA was observed with Golgi fractions from influenza-infected cells (not shown), but we have chosen to illustrate the properties of the system mainly with experiments employing Golgi fractions from VSV-infected cells. The release of viral glycoproteins did not occur at 4°C (Fig. 3A), was greatly reduced at 20°C (Fig.  3B), and was totally dependent on the presence of the cytosolic protein fraction, with similar requirements for VSV-G and HA  (31), and analyzed by SDS-polyacrylamide (10%) gel electrophoresis and fluorography. G and dG indicate the positions of the bands corresponding to the glycosylated and deglycosylated forms of VSV-G, respectively. B, a Golgi fraction from MDCK cells containing labeled VSV-G protein accumulated in the TGN was incubated for 90 min at 37°C in a complete assay mixture. Golgi remnants were removed by sedimentation (10 min at 10,000 ϫ g), and the supernatant, diluted (12.5 times) with PBS, was centrifuged (1 h at 100,000 ϫ g) to recover sedimentable material containing the VSV-G. This fraction (lanes c and d), as well as the initial Golgi (lanes a and b), were dissolved with detergent, incubated with (lanes b and d) or without (lanes a and c) neuraminidase for 20 h at 37°C and analyzed by one-dimensional isoelectric focusing (87). Numbers 1-7 indicate charged species of the VSV-G molecule. Essentially all labeled VSV-G molecules were shifted toward the cathode by the neuraminidase treatment.

FIG. 2. Golgi fraction from VSV-infected MDCK cells used for vesicle generation.
The starting Golgi fraction was processed for conventional electron microscopy (see "Materials and Methods"). This electron micrograph shows that most Golgi cisternae retain their characteristic stacked configuration (GA). When sectioned, peripheral fenestrated regions of the cisternae appear as tubular elements (t) at the periphery of the stacks. Bar, 0.5 m. release (Fig. 3B). Cytosolic protein fractions with comparable activities could be obtained from rat liver (Fig. 3B), bovine brain, or MDCK cells (not shown). The in vitro release of the glycoproteins requires ATP hydrolysis, since it did not occur when the system was depleted of ATP ( Fig. 4A) and could not be sustained when this was replaced by ATP␥S (Fig. 4A).
During the standard incubation, VSV-G protein release was essentially completed by 60 min (Fig. 3A). This reflected an exhaustion of the capacity of the Golgi fraction to effect the release of the labeled protein and not a depletion of the energy supply or an inactivation of components in the cytosolic fraction, since used medium was effective in supporting VSV-G protein release from a fresh Golgi fraction, but the addition of fresh cytosolic proteins, and/or a new energy supply, to a preincubated Golgi fraction did not lead to any significant additional release of VSV-G protein (not shown).
Previous studies have shown that some of the mammalian and yeast factors required for intracellular transport are functionally equivalent and can be interchanged in in vitro systems or in transfected cells (see Ref. 3). We, therefore, tested the capacity of a yeast cytosolic protein fraction to support vesicle formation in our system and found it to be as active (Fig. 4C) as mammalian cytosolic proteins. In addition, vesicle generation supported by the yeast fraction had the same requirements as that supported by rat liver cytosolic proteins (Fig. 4C).
Vesicle Release Can Be Inhibited by Primaquine and Requires an N-Ethylmaleimide (NEM)-sensitive Cytosolic Factor-To further characterize the in vitro system, we examined the effect of the lysosomotropic amine primaquine, which inhibits secretion in vivo (41) and prevents the formation of the transport vesicles that mediate intra Golgi transport in vitro (42,43). We found that this drug almost completely prevented the release of VSV-G protein (not shown). On the other hand, monensin, an ionophore that promotes the exchange of K ϩ for H ϩ ions across membranes, had no effect (not shown). This drug inhibits the secretion of many proteins (44) as well as the passage of viral glycoproteins from the medial to the trans region of the Golgi apparatus (45, 46) but does not block the in vitro formation of intra-Golgi transport vesicles (42,43). The inhibitory effect of primaquine can be taken as evidence that vesicles are generated by a budding process rather than by cisternal fragmentation.
The SH-alkylating reagent NEM is known to inhibit in vitro vesicular transport between various organelles (47)(48)(49), and this is due, at least in part, to the inactivation of NSF, the NEM-sensitive factor that is thought to be necessary for the fusion of a vesicle with its acceptor membrane (see Ref. 4). In our system, NEM treatment of the mammalian (Fig. 4B) or yeast-derived (Fig. 4C) cytosolic protein fractions greatly reduced VSV-G protein release. On the other hand, preincubation of the Golgi membranes with NEM had no effect on the release of the viral glycoprotein (Fig. 4B).
Vesicles Are Produced by Budding from Golgi Cisternae: Characterization of the Released Vesicles Containing the VSV-G Protein-Golgi fractions recovered during the course of incubation with cytosolic proteins and an ATP generating system were examined by electron microscopy to characterize the process that leads to the release of the labeled viral glycoprotein. This revealed that, as early as 5 min after the beginning of the incubation, numerous non-clathrin-coated buds and vesicles emerged from the Golgi cisternae, particularly from their rims and on the concave side of the curved stacks (Fig. 5, A and  B). Clathrin-coated vesicles were also formed during the incubation, but almost invariably, these were produced from dilated cisternae that were adjacent to one side of the Golgi stack and were not involved in the production of non-clathrin-coated buds and vesicles (Fig. 5B). ) of the temperature-dependent releases of VSV-G protein obtained with three different cytosolic protein and Golgi preparations. C, a yeast cytosolic protein fraction supports the energy-dependent and NEM-sensitive release of VSV-G-containing vesicles from the mammalian Golgi fraction. The standard assay was carried out with (b, c, and d) or without (a) a yeast cytosolic protein fraction (10 mg/ml), either in a complete system containing an ATP-regenerating system (a, b, and d) or in one containing ATP␥S (c). In one case (d), the complete standard reaction mixture included cytosolic proteins that had been pretreated with NEM, as described above.
The VSV-G protein released in vitro was found to be contained in sealed membrane structures in which the luminal domain of the protein was protected from digestion by exogenous proteases (Fig. 6A). Thus, when the supernatant obtained after a 1.5-h in vitro incubation was treated with bromelain, this protease was able to remove only the small C-terminal segment of the VSV-G polypeptide (Fig. 6A), which is known to be exposed on the cytoplasmic side of the membrane. On the other hand, protease digestion after detergent solubilization of membranes led to further degradation of the VSV-G protein (Fig. 6A).
The membrane vesicles containing labeled VSV-G protein could be quantitatively recovered by sedimentation for 1 h at 100,000 ϫ g, and their isopycnic density, measured by banding in a sucrose density gradient (not shown), was found to be 1.10 -1.12 g/cm 3 . The released vesicles, purified by flotation as described under "Materials and Methods," were also examined by electron microscopy (Fig. 6B). This revealed that they ranged in size from 50 to 80 nm in diameter and that they lacked a discernible coat structure. This was particularly apparent after treatment with tannic acid (37), a mordant that enhances the electron density of the coats of COPI (4) and COPII (9) -coated vesicles (Fig. 6B, bottom). This treatment also revealed the presence in the lumen of many of the vesicles of a content, whose electron density was greatly enhanced in the tannic acid-treated samples. Immunolabeling of the purified vesicles with an antibody to the cytoplasmic tail of VSV-G also indicated that the C-terminal tail of the viral glycoprotein was exposed and accessible on the outer surface of many of the vesicles (Fig. 6C). In these samples, vesicles labeled with at least three gold particles were frequently found (52 Ϯ 7 vesicles per square window in a 300 mesh electron microscopy grid), whereas such vesicles were absent from control samples of purified vesicles that were incubated with the antibody in the presence of immunogenic peptide, incubated with preimmune serum, or treated with protein-A gold alone.
To establish that the vesicular structures released in vitro do not simply result from the random fragmentation of the Golgi membranes or the TGN, the amount of sialyltransferase, a TGN marker, released from the Golgi was measured and com-pared with the amount of labeled VSV-G protein released. As shown in Fig. 7, whereas ϳ30% of the labeled VSV-G protein in the initial Golgi fraction was released in an ATP-and temperature-dependent fashion, very low levels of sialyltransferase (less than 4%), were released during a parallel incubation. The depletion of sialyltransferase from the released vesicles relative to the starting Golgi fraction was also indicated by the much lower specific activity of the enzyme in the purified released vesicles (0.74 Ϯ 0.08 ϫ 10 Ϫ12 IU/mg of protein, n ϭ 3) than in the original Golgi (7.9 Ϯ 0.3 ϫ 10 Ϫ12 IU/mg of protein, n ϭ 3). In addition, the labeled VSV-G protein molecules released were clearly derived from the TGN, since they carried sialic acid residues that could be removed by neuraminidase treatment (see Fig. 1B).
An Arf-like GTP-binding Protein Is Required for Post-Golgi Vesicle Formation-Arf proteins, of which several forms have been identified (50), are low molecular weight, amino-terminally myristoylated, GTP-binding proteins that in their activated state promote the formation of coated vesicles that mediate transport between various intracellular compartments (for review see Ref. 51). The fungal metabolite brefeldin A (BFA) has been found to inhibit transport events by blocking the activation of Arf through the inhibition, probably indirect, of its guanine nucleotide exchange factor (52,53).
BFA had a powerful inhibitory effect on our in vitro system, reducing vesicle generation by 60% (Fig. 8). Since this implicated the active form of an Arf protein in vesicle generation we next examined the effect of GTP␥S on this process. Sucrose density gradient analysis of the reaction mixture (Fig. 9A) showed that the GTP analogue did not affect the amount of vesicles released. It revealed, however, that the vesicles generated in the presence of the GTP␥S sedimented more rapidly than those produced in its absence. Electron microscopy showed that this was due to the retention on these vesicles of a nonclathrin protein coat (Fig. 9C), which morphologically resembled the coats of COPI (4) and COPII (9) -coated vesicles and was absent from the vesicles released when GTP␥S was omitted (Fig. 9B).
Vesicle Release Is Inhibited by Protein Kinase Inhibitors-As noted above, protein phosphorylation/dephosphorylation FIG. 5. In vitro production of buds and vesicles from Golgi cisternae. After 5 min of incubation at 37°C with cytosolic proteins and an ATP-generating system, the Golgi fraction was recovered by centrifugation and processed for electron microscopy, as described under "Materials and Methods." A and B, two Golgi stacks on which numerous vesicles (arrows) are seen budding from the Golgi cisternae, particularly from their rims and on the concave side of the curved stacks. In B, note that clathrin-coated vesicles (arrowheads) bud from a distinct dilated cisternae that is adjacent to one side of the stack. Bar, 100 nm. events have been implicated in various steps of intracellular transport. We therefore examined the effect of various protein kinase and phosphatase inhibitors on post-Golgi vesicle formation. Staurosporine, a broad spectrum inhibitor of protein kinases C and A and of tyrosine protein kinases that acts on the catalytic sites of these enzymes (54), markedly inhibited (Ͼ70%) the release of VSV-G in a dose-dependent fashion (Fig.  10A). Calphostin C, which is a specific PKC inhibitor (55), was found to nearly completely suppress vesicle generation in vitro (Fig. 9A). This inhibitory effect only occurred after illumination of the drug, which is required for its activation (56). The light dependence of calphostin C activity permitted the demonstration that its PKC target is located in the Golgi membrane rather than in the cytosolic protein fraction (Fig. 11). Whereas a preincubation of the Golgi fraction in the light with calphostin C markedly reduced its capacity to generate vesicles during a subsequent incubation with cytosolic proteins in the dark, a similar preincubation of the cytosolic proteins did not affect its activity. These observations implicated a PKC in vesicle generation and led us to examine the effect of another specific PKC inhibitor, the 18-amino acid pseudosubstrate synthetic peptide (57) that is identical to an autoinhibitory portion of the regulatory domain of the enzyme (see Ref. 58). This peptide suppressed vesicle release by more than 80% (Fig. 10A) and exerted its half-maximal effect at 15 M, a concentration that corresponds to its reported K i with the purified enzyme (57). A confirmation of the involvement of a PKC in vesicle release was also provided by the stimulatory effect of TPA (Fig. 10D), a phorbol ester that activates the enzyme by binding to the diacylglycerol site in its regulatory domain (see Ref. 59). TPA FIG. 6. A, the released VSV-G protein is contained in membrane vesicles with its cytoplasmic domain exposed on the outer surface. Aliquots (10 l) of supernatant fractions obtained after incubation for in vitro vesicle formation were incubated for 1 h at 4°C in the absence (a and b) or presence (c and d) of 10 l of bromelain (50 g/ml), without (a and c) or with (b and d) the addition of 0.1% Triton X-100, as indicated. The samples were then incubated for 10 min at 0°C with 1 mM of the protease inhibitor N␣-p-tosyl-L-lysine chloromethyl ketone and processed for SDS gel electrophoresis and fluorography. G and GЈ indicate the positions of the bands corresponding to the intact VSV-G protein and to the protein from which the exposed cytoplasmic domain was removed by proteolysis, respectively. B, the released VSV-G protein is contained in membrane-bound vesicles that vary in size from 50 to 80 nm in diameter. The vesicles were purified and processed for conventional electron microscopy, as described under "Materials and Methods." The gallery of micrographs at the bottom clearly shows that the membranes surrounding the vesicles do not bear a recognizable coat structure. C, the C-terminal tail of the VSV-G protein is exposed on the surface of the purified vesicles. Aliquots of the vesicles were processed for immunogold labeling and negative staining using an antibody to the VSV-G tail, as described under "Materials and Methods." Vesicles labeled with three or more gold particles, like those shown here, were frequently found but were absent from controls. Bars, 50 nm. FIG. 7. The release of VSV-G in post-Golgi vesicles is not accompanied by a significant release of sialyltransferase, a resident TGN marker. Golgi fractions containing labeled or unlabeled VSV-G accumulated in the TGN were incubated in parallel for 90 min at 4 or 37°C in complete medium or at 37°C in medium lacking ATP and the ATP-regenerating system. The supernatants obtained after removing the remnant Golgi fraction were diluted (12.5 times) with PBS, and the released vesicles were recovered by sedimentation (1 h at 100,000 ϫ g). The sialyltransferase activity (ST) was measured (35) in the nonradioactive samples, and the release of VSV-G from the radioactivity was recovered with the vesicles. Both are expressed as a percentage of the amounts present in the original Golgi fraction. Each point represents the average (ϮS.D.) from five independent experiments with two different Golgi and cytosolic protein preparations.
FIG. 8. BFA inhibits the in vitro release of VSV-G from the TGN. Incubations with or without BFA were carried out at 4 or 37°C for the indicated times. Control samples received 2.5% methanol, the vehicle in which the BFA was dissolved. Each point represents the mean (ϮS.D.) from three independent experiments using three different Golgi preparations and two different cytosolic protein fractions. stimulated vesicle release by 60 -100% (Fig. 8D), and its effect was eliminated by N,N-dimethylsphingosine (Fig. 10D), an inhibitor of the kinase that competes with TPA for the diacyl-glycerol binding site. Indeed, N,N-dimethylsphingosine alone (Fig. 10D), but not its inactive analogue N-acetylsphingosine (not shown), was a potent inhibitor of vesicle release. Finally, the participation of a PKC in the process of post-Golgi vesicle release was also indicated by the inhibitory effect of a monoclonal antibody that recognizes all isozymes of PKC (Fig. 10C). Other protein kinase inhibitors, such as H89 and ML-7, which at low concentrations specifically inhibit protein kinase A (60), and myosin light chain kinase (61), respectively, were able to suppress post-Golgi vesicle release by 50 -60% only at the higher concentrations at which they are known to also inhibit PKC (Fig. 10B). Although these experiments can be taken to indicate that certain proteins must be in a phosphorylated state in order for vesicle generation to proceed optimally, the phosphatase inhibitors microcystin LR and okadaic acid did not stimulate vesicle release (not shown). DISCUSSION We have developed an in vitro system for the generation of post-Golgi vesicles from an isolated Golgi fraction containing labeled viral glycoproteins that had acquired sialic acid residues and had been accumulated in vivo in the TGN. By monitoring the release of these molecules from the sedimentable membranes under different conditions we were able to establish that the system recreates the process of vesicle formation that occurs in vivo; VSV-G protein release did not occur when the Golgi fraction was incubated at 4°C or at 20°C, a temper- FIG. 9. Vesicles generated in the presence of GTP␥S retain a protein coat. Golgi fractions containing the labeled VSV-G protein were incubated for 90 min at either 37 or 4°C in reaction mixtures containing a rat liver cytosolic protein fraction supplemented with an ATP-generating system, in the presence or absence of 100 M GTP␥S, as indicated. Following incubation, the mixtures were chilled on ice and loaded on a continuous sucrose gradient (10 ml; 0.4 -0.8 M sucrose over a 1-ml 2.0 M sucrose cushion, in 20 mM Hepes-KOH, pH 7.3), which was centrifuged for 1 h at 125,000 ϫ g in an SW41 rotor. A, the radioactivity distribution in the gradient fractions (0.5 ml) and in the loading zone (S), and resuspended pellet (P) is expressed as percentage of the total VSV-G radioactivity recovered in the gradient. B and C, aliquots of the radioactive peak fractions in A, produced without (B) or with (C) GTP␥S, were analyzed by conventional electron microscopy. This revealed in B the presence of a population of 50 -80-nm naked vesicles and in C a population of non-clathrin-coated vesicles. Larger magnification images of the naked and coated vesicles are shown below panels B and C, respectively. Bars, 50 nm. ature at which exit from the trans Golgi in vivo is blocked, and it required the presence of cytosolic proteins as well as the hydrolysis of ATP. In addition, the released viral glycoproteins were contained in vesicles of relatively uniform size that contained very low levels of sialyltransferase, a TGN marker of which insignificant amounts were released concomitantly with the viral glycoproteins. We can therefore conclude that the vesicles generated in this system, containing the sialylated labeled viral glycoproteins and depleted of sialyltransferase, are the result of a sorting process that occurs in the Golgi apparatus and not of a random fragmentation of this organelle or of a dispersion of its elements that mimics the one that takes place in mitotic cells (62) or that which is induced by microtubule depolymerization (63).
Several observations lead us to conclude that the formation of post-Golgi vesicles requires the activation of an Arf-like GTP-binding protein that is incorporated into the coat of the vesicles during their formation in the TGN and that hydrolysis of the bound nucleotide on this protein is necessary for vesicle uncoating to occur. Thus, 1) vesicle generation was inhibited by BFA, and 2) when the nonhydrolyzable analogue GTP␥S was present during the incubation, only coated vesicles were recovered, in contrast to the uncoated ones that accumulated in the presence of GTP. The inhibitory effect of BFA observed in our system directly demonstrates that the previously reported inhibitory effect of this drug on the transport of the VSV-G protein to the cell surface in BHK cells (64) is due to a block in the formation of the carrier vesicles in the TGN. In this regard, it is noteworthy that a putative coat component for post-Golgi vesicles (p200) has been identified (12) that dissociates from Golgi membranes when cells are treated with BFA and, in vitro, binds more effectively to them when GTP␥S is present.
In its sensitivity to BFA, the in vitro budding from the TGN of vesicles containing the VSV-G protein resembles the processes by which constitutive secretory vesicles and immature secretory granules are formed in an in vitro system that utilizes a postnuclear supernatant from the pheochromocytomaderived PC12 cells (15,65). On the other hand, the fact that GTP␥S does not inhibit post-Golgi vesicle generation in our in vitro system contrasts with the 40 -60% inhibition in vesicle production that this analogue caused in the PC12 cell-free system (16,66) or in permeabilized GH 3 cells (67). Although it was originally proposed that in the PC12 system the GTP␥S inhibitory effect reflected a requirement for GTP hydrolysis during vesicle formation (66), subsequent work (68) led to the suggestion that the analogue acts by activating Golgi-associated heterotrimeric G proteins of both the G ␣i and G ␣s classes (69) and that the effect of the inhibitory G ␣i protein is dominant over that of the G ␣s . If this is the case, the differences observed in the effect of GTP␥S on vesicle production in the different cell systems could also reflect different set points in the balance of the opposite actions of both types of G proteins. It is possible, of course, that the formation of post-Golgi vesicles containing secretory proteins differs significantly from that of vesicles containing plasma membrane proteins. In this regard it may be noted that the vesicular flow that transfers secretory proteins to the cell surface may be distinct from that which carries membrane proteins (70,71), although a membrane protein, the polymeric Ig receptor (pIgR), was found together with a secretory protein in rat liver post-Golgi vesicles generated in vitro (13).
The pronounced inhibitory effect of NEM that we observed is in accord with a previous report (13) using a rat liver Golgi fraction and rat liver cytosol in which the alkylating agent was added to the complete reaction mixture. We showed, however, that the effect of NEM is due to the inactivation of one or more essential cytosolic components and that the reagent had no effect when applied only to the donor Golgi fraction. Although it is well known that the cytosolic NEM-sensitive factor, NSF, plays an essential role in vesicular transport between several organelles of the endomembrane system (see Refs. 4 and 72), the factor(s) necessary for post-Golgi vesicle formation must be distinct from NSF, since the latter functions in vesicle fusion at the acceptor membrane rather than in vesicle generation. Moreover, we found that the NEM-sensitive factor(s) that participates in vesicle formation is not inactivated under conditions (incubation at 37°C in the absence of ATP) that abolish the activity of NSF in a cytosolic fraction (73). Finally, the Golgi fraction itself is known to be able to provide enough NSF to sustain intra-Golgi transport in vitro (73,74), and yet in our system the untreated Golgi fraction could not compensate for the NEM inactivation of the cytosolic protein fraction. In one other case, that of the vesicular transport of the mannose 6-phosphate receptor from endosomes to the TGN, an NEMsensitive factor distinct from NSF was found to act at an early stage, probably during vesicle formation (49). It is possible that this factor is the same required for post-Golgi vesicle budding in our assay, but this cannot be ascertained until both fractions are more fully characterized.
We have found that several reagents that inhibit protein kinase C markedly suppressed post-Golgi vesicle production, and that the PKC activator TPA substantially promoted it. Vesicle generation was inhibited by agents that act on the PKC catalytic domain, such as the pseudosubstrate inhibitor peptide and, at high concentrations, the general protein kinase inhibitor staurosporine, as well as by agents that act on the regulatory diacylglycerol-binding site of the PKC, such as calphostin C and N,N-dimethylsphingosine. Moreover, a monoclonal antibody that recognizes all known isoforms of PKC and binds to an epitope in the catalytic domain, also substantially inhibited post-Golgi vesicle production. An effect of calphostin C in suppressing the transfer of VSV-G from the TGN to the cell surface has been observed in intact cells (75). Moreover, the PKC stimulator TPA has been shown to promote the constitutive release of 35 SO 4 -glycosaminoglycan molecules from rat basophilic leukemic and MDCK cells (19). All this evidence strongly FIG. 11. The protein kinase C is associated with Golgi membranes. Golgi (A) or cytosolic protein (B) fractions were preincubated for 10 min on ice with 0.5 and 2.0 M calphostin C, respectively, under white light or in the dark, as indicated in the figure. The calphostin C-treated Golgi membranes and cytosolic proteins were then mixed with untreated complementary components and assay buffer containing an energy-generating system. (The final concentrations of calphostin C in the reaction mixtures containing treated Golgi or cytosolic proteins were 0.125 and 0.5 M, respectively; IC 50 ϭ 0.5 M). The mixtures were then incubated for 90 min at either 4 or 37°C under light or dark conditions, as indicated. Following incubation, the percentage of radiolabeled VSV-G protein released was determined. Calphostin C is active and capable of inhibiting PKC only during the period of illumination. In this experiment, the concentration of calphostin C was selected to be enough to cause only a 50% inhibition of vesicle release, so that in panel A, after dilution resulting from the addition of cytosolic proteins and other components, the residual calphostin C would no longer be inhibitory.
implicates the action of a PKC as essential for post-Golgi vesicle generation. In fact, a recent report using a membrane fraction obtained from the postnuclear supernatant of PC12 cells demonstrated directly that the addition of purified PKC stimulated the release of vesicles containing the ␤-amyloid precursor protein, even in the absence of added cytosol (22). Several of the protein kinase modulators we employed have also been utilized to study endoplasmic reticulum to Golgi and intra-Golgi transport (75). This work led to the conclusion that a diacylglycerol/phorbol ester binding protein controls passage from the endoplasmic reticulum to the Golgi apparatus but is not involved in transport between Golgi cisternae. The diacylglycerol/phorbol ester binding protein, however, appears to be distinct from PKC, since it was only affected by agents that act at the regulatory site of this enzyme, but not by those that act at the catalytic site, including the pseudosubstrate peptide and staurosporine, which we found inhibited post-Golgi vesicle formation.
Our results, together with the reports just discussed, indicate that both Arf and PKC play essential roles in the production in the TGN of post-Golgi vesicles that carry to the cell surface constitutively secreted proteins and/or newly synthesized plasma membrane polypeptides. A link between Arf and PKC seems to be established by the observation that activation of the latter leads to enhanced binding of Arf and ␤-COP to Golgi membranes (19). We suggest that a Golgi-associated PKC may control the activation of Arf. In fact, an exchange factor that activates Arf appears to be under the control of heterotrimeric G proteins (see Ref. 76), which themselves are known to be substrates for PKC (see Ref. 77).
Although the binding of activated Arf to TGN membranes would be expected to stimulate vesicle formation directly by promoting the assembly of a coat (78), it has also become clear that activation of Arf may lead to the modification of membrane lipids through the Arf-dependent activation of a phospholipase D (79,80) that is associated with Golgi membranes (81). Phospholipase D generates phosphatidic acid from phosphatidylcholine, and it has been proposed that the activation of a phosphatidylinositol 4-P 5-kinase mediated by phosphatidic acid, would further modify the composition of the membrane by stimulating the production of phosphatidylinositol 4,5-P 2 , which also further stimulates phospholipase D activity (see Ref. 82). The final result of these processes would be the creation of membrane microdomains enriched in negatively charged lipids in the vicinity of Arf molecules. Although these lipid modifications have been proposed to occur during the docking stage of a vesicle on an acceptor membrane and foster membrane fusion (82), it seems equally reasonable that they could promote the membrane fission events that lead to vesicle release from a donor membrane. In addition to this potential Arf-related activation of phospholipase D by PKC, it is also known that phospholipase D can be activated directly by PKC (83)(84)(85) or by signaling pathways (86) that may not depend on the participation of Arf. In either case, it appears that PKC may serve as a link through which extracellular signals affect constitutive traffic out of the Golgi through the control of post-Golgi vesicle formation.