Fragmentation and re-assembly of the Golgi apparatus in vitro. A requirement for phosphatidic acid and phosphatidylinositol 4,5-bisphosphate synthesis.

Recent work from our laboratory demonstrated that phosphatidic acid (PA) and phosphatidylinositol 4,5-bisphosphate (PtdIns(4,5)P(2)), are required to maintain the structural integrity of the Golgi apparatus. To investigate the role of these lipids in regulating Golgi structure and function, we developed a novel assay to follow the release of post-Golgi vesicles. Isolated rat liver Golgi membranes were incubated with [(3)H]CMP sialic acid to radiolabel endogenous soluble and membrane glycoproteins present in the late Golgi and trans-Golgi network. The release of post-Golgi secretory vesicles was determined by measuring incorporation of (3)H-labeled proteins into a medium speed supernatant. Vesicle budding was dependent on temperature, cytosol, energy and time. Electron microscopy of Golgi fractions prior to and after incubation demonstrated that the stacked Golgi cisternae generated a heterogeneous population of vesicles (50- to 350-nm diameter). Inhibition of phospholipase D-mediated PA synthesis, by incubation with 1-butanol, resulted in the complete fragmentation of the Golgi membranes in vitro into 50- to 100-nm vesicles; this correlated with diminished PtdIns(4,5)P(2) synthesis. Following alcohol washout, PA synthesis resumed and in the presence of cytosol PtdIns(4,5)P(2) synthesis was restored. Most significantly, under these conditions the fragmented Golgi elements reformed into flattened cisternae and the re-assembled Golgi supported vesicle release. These data demonstrate that inositol phospholipid synthesis is essential for the structure and function of the Golgi apparatus.

Intracellular protein transport through the secretory pathway is mediated by membrane-bound carrier vesicles (1). In the past decade significant progress has been made in characterizing different classes of coat proteins, which function at various steps of the pathway (2). Transport vesicles interact with at least two distinct types of protein complex; those involved in vesicle formation and budding from donor organelles, e.g. coatomer; COP-I and -II; and those facilitating vesicle target-ing to and fusion with the correct acceptor membranes (SNAREs) 1 (3). Several ancillary proteins, including members of the Rab family of small GTP-binding proteins interact with specific vesicles to regulate steps in the transfer process: targeting, docking, and fusion (4). Furthermore, recent theories have suggested that the Rab-related proteins in conjunction with SNAREs may play a key role in regulating the specificity of vesicle interactions (5). In addition to the Rabs, another family of GTP-binding proteins, ARFs (ADP-ribosylation factors) recruit COP-I and -II coat complexes as well as clathrin adaptors (6,7) and a novel class of clathrin-binding proteins to the cytoplasmic face of the trans-Golgi network (8 -11).
Considerable evidence has demonstrated that, in addition to the aforementioned components, inositol phospholipids play a key role in controlling vesicle transport from the Golgi apparatus and in regulating various steps in endocytosis in mammalian and yeast cells (12). Several laboratories have shown that ARFs may regulate vesicle trafficking, in part by effecting changes in phospholipid metabolism (6). Recently, it was demonstrated that ARF1 and -6 recruit phosphatidylinositol 4-kinase ␤ to Golgi membranes and Type-I␣ phosphatidylinositol-4-phosphate 5-kinase to the plasma membrane, respectively (13,55). It is likely that different phosphatidylinositol kinase isoforms function in regulating the level of phosphatidylinositol 4,5-bisphosphate (PtdIns(4,5)P 2 ) in the Golgi apparatus and plasma membrane as well as in other membranes. An initial link between phospholipid metabolism and ARFs came from earlier observations from several laboratories (6). It was shown that ARF1 activates a phospholipase D activity (14,15), which in some cells associates with membranes of the Golgi apparatus (16,17). Phospholipase D hydrolyzes phosphatidylcholine to generate phosphatidic acid (PA), which stimulates the synthesis of PtdIns(4,5)P 2 as well as promoting COP-1 binding to membranes (6) and release of nascent secretory vesicles from the TGN (18). Additionally, our data demonstrated that PA stimulation of PtdIns(4,5)P 2 synthesis is required for maintaining the structural integrity and function of the Golgi apparatus (19). Treatment of growth hormone-secreting rat pituitary GH3 cells with 1-BtOH led to the synthesis of phosphatidyl butanol rather than PA as a result of the transphosphatidylation activity of PLD. Under these conditions hormone secretion was inhibited, in part, due to diminished PtdIns(4,5)P 2 synthesis, which resulted in quantitative fragmentation of the Golgi ap-paratus. Upon alcohol removal, the Golgi apparatus structure was restored and hormone secretion resumed. In contrast to these results, other investigators have suggested that ARFactivated PLD does not function in mediating PtdIns(4,5)P 2 biosynthesis in the Golgi apparatus or in COP-I recruitment (13,20,21).
To understand further the mechanism by which polyphosphoinositide synthesis functions in regulating the structural organization of the Golgi apparatus, we have now utilized a novel assay system to follow the release of radiolabeled endogenous glycoproteins into post-Golgi vesicles in vitro. To this end, we exploited previous observations (22,23) that specific nucleotide sugar transporters and sialyl transferase enzymes, which localize to late Golgi cisternae and the TGN, can incorporate radiolabeled nucleotide sugars into endogenous cargo molecules efficiently (22,24). Our rationale was to use purified rat liver Golgi membranes labeled with radioactive sialic acid to follow the release of soluble and membrane glycoproteins from the TGN into post-Golgi vesicles in response to changes in PA and PtdIns(4,5)P 2 synthesis. In agreement with our earlier findings, vesicle budding was stimulated by exogenously added bacterial PLD and addition of 1-butanol resulted in complete fragmentation of Golgi cisternae (18,19). Strikingly, following alcohol washout, PA and PtdIns(4,5)P 2 synthesis resumed, and the fragmented Golgi reformed into its characteristic flattened cisternae in vitro. Most significantly, the re-assembled Golgi apparatus was able to support nascent vesicle release. Our results demonstrate that inositol phospholipid synthesis is essential for maintaining the structure and function of the Golgi apparatus.

EXPERIMENTAL PROCEDURES
Antibodies-The following antibodies were used: monoclonal antibody 1D9 against ARF1 was provided by Dr. R. Kahn. A rabbit antibody to PLD1, designated P1-P4, was generated using four peptides specific to human PLD1 and has been described previously (17,25). A monoclonal antibody to TGN38 was purchased from Affinity Bioreagents, Golden, CO; monoclonal anti-Rab5 was obtained from Transduction Laboratories, Lexington, NY. Rabbit antibodies were as follows: to calnexin were from Santa Cruz Biotechnology, Santa Cruz, CA; to ␣ 1 -antitrypsin (from Zymed Laboratories Inc., San Francisco, CA); and to transferrin (from Cappel, Aurora, OH). Rabbit anti-connexin43 was a gift of Dr. Eliot Hertzberg, Albert Einstein College of Medicine; rabbit anti-Tom20 was a gift from Dr. Gordon Shore, McGill University. Bacterial PLD (Streptomyces chromofuscus) was purchased from Sigma Chemical Co.
Preparation of Rat Liver Golgi Membranes-Rat liver Golgi membranes were isolated using a modification of the procedure of Slusarewicz et al. (26). All steps were performed at 4°C; four to six livers from rats that had been fasted overnight were homogenized using a loose fitting Potter homogenizer in 3 volumes of 0.5 M sucrose containing 50 mM phosphate buffer, pH 6.8, 2.5 mM MgCl 2 and a mixture of protease inhibitors. A post-nuclear supernatant was obtained by centrifugation at 1000 ϫ g for 5 min; this fraction was diluted to 0.25 M sucrose and transferred to Beckman SW28 Ultracentrifuge tubes. 12 ml of the supernatant was underlayered with 7, 9, and 7 ml of 0.5, 0.86, and 1.3 M sucrose, respectively, in the same buffer and centrifuged at 28,000 rpm (100,000 ϫ g av ) for 90 min. Material floating at the 0.5 M/0.86 M sucrose interface, which was enriched in Golgi membranes (Fig. 1), was diluted to 0.25 M and centrifuged at 7000 rpm in the SW28 rotor for 30 min onto a 100-l cushion of 1.3 M sucrose. This step was repeated, and the membrane pellet was resuspended in 2-3 ml of 0.25 M sucrosecontaining buffer. Isolated Golgi membranes were stored at Ϫ80°C until use.
Preparation of Rat Brain Cytosol-All steps were preformed at 4°C. Four rat brains were homogenized using a loose fitting Potter homogenizer in 3 volumes of 0.25 M sucrose containing 20 mM Tris-HCl, pH 7.4, 50 mM KCl, 2 mM MgCl 2 , and a mixture of protease inhibitors. A post-nuclear supernatant was obtained by centrifugation at 1000 ϫ g for 10 min followed by centrifugation of the supernatant at 100,000 ϫ g for 3 h. The high speed supernatant was dialyzed overnight against 20 mM Tris-HCl, pH 7.4, 100 mM KCl, 2 mM MgCl 2 , and a mixture of protease inhibitors and concentrated using a Centriprep-10 concentra-tor (Amicon). Aliquots were frozen in liquid N 2 and stored at Ϫ80°C. Following incubation for 30 min at 37°C, the samples were spotted onto Whatman 3MM filter discs to determine the incorporation of radioactivity into ice-cold trichloroacetic acid-precipitable material. The filters were washed extensively, dried, and counted in a Beckman liquid scintillation counter.
Radiolabeling of Endogenous Golgi Glycoproteins, Vesicle Budding, and Immunoprecipitation-Golgi membranes (20 g) were incubated with [ 3 H]CMP sialic acid as above for 30 min at 37°C in a volume of 60 l. Following incubation, 100 M non-radioactive CMP-sialic acid was added in addition to 120 g of cytosol (2 mg/ml final concentration) and an energy regeneration system (EGS) containing (1 mM ATP, 0.02 mM GTP, 10 mM creatine phosphate, 80 g/ml creatine phosphate kinase). When cytosol alone was incubated in the absence of Golgi membranes, there was no incorporation of [ 3 H]CMP sialic acid. To assay for release of post-Golgi vesicles, samples were incubated at 37°C for 45 min and the vesicles were separated from residual Golgi membranes by centrifugation for 1 min at 12,000 ϫ g. The supernatant containing released vesicles and pellets (corresponding to residual Golgi membranes) were either precipitated with ice-cold 10% trichloroacetic acid or treated with 1% Triton X-100, 100 mM NaCl, 20 mM Hepes, pH 7.4, 0.1 mM EDTA, 1 mM phenylmethylsulfonyl fluoride and incubated overnight at 4°C with appropriate antibodies. The antibody-antigen complexes were incubated with Protein A-Sepharose beads for 1 h at 4°C and washed twice with 0.1% Triton X-100 in the preceding buffer followed by washing twice in phosphate-buffered saline, and the samples were analyzed by SDS-PAGE.
Alcohol Washout of Golgi Membranes-[ 3 H]CMP sialic acid-labeled Golgi membranes were treated with 1.0% 1-butanol or t-butanol for 30 min at 37°C or 4°C in the presence of cytosol and energy (EGS). Samples were then diluted 5-to 10-fold with buffer followed by centrifugation in a Beckman Airfuge at 100,000 ϫ g for 3 min. The supernatant was removed, and the pellet containing fragmented Golgi elements was resuspended in vesicle budding buffer. Aliquots of these membranes were incubated at 37°C for 45 min with rat brain cytosol (2 mg of protein/ml) in the absence or presence of EGS (above). Vesicles and reformed Golgi cisternae were separated by centrifugation at 12,000 ϫ g for 1 min. The supernatant and pelleted material were analyzed by SDS-PAGE and electron microscopy.
Determination of Polyphosphoinositide Synthesis in Golgi Membranes-The synthesis of PtdIns(4)P and PtdIns(4,5)P 2 was determined as described by Siddhanta et al. (19). Rat liver Golgi membranes were incubated with or without 2.0 mg/ml rat brain cytosol in 20 mM Hepes, pH 7.3, 125 mM KCl, 2.5 mM MgCl 2 , 2 mM ATP, and [␥-32 P]ATP (final specific activity, 70 Ci/mmol) at 37°C for 15 min. Reactions were terminated by extraction with acidified chloroform/methanol. The chloroform phase was recovered and washed with methanol:1 N HCl (1:1), and the organic phase was dried, resuspended in chloroform:methanol:12 N HCl (200:100:1), and spotted on TLC plates. Radiolabeled phospholipids were detected by autoradiography and identified by comigration with nonradioactive standards.
SDS-PAGE and Western Blotting-Proteins were separated by SDS-PAGE on 10% gels and transferred to polyvinylidene difluoride membranes (Millipore). Before incubation with primary antibodies the membranes were blocked in 5% nonfat milk containing 0.2% Tween 20 in phosphate-buffered saline. Following incubation, horseradish peroxidase-conjugated goat anti-rabbit or anti-mouse antibodies diluted 1:2000 were used to detect primary antibodies; gel bands were visualized by Enhanced Chemiluminescence (ECL; Amersham Biosciences, Inc.). [ 3 H]Sialic acid-labeled proteins were resolved by SDS-PAGE, and the radioactive polypeptides were detected by fluorography. The fluorograms were scanned, and the band intensities were quantified using a computing densitometer (Molecular Dynamics, Sunnyvale, CA).
Electron Microscopy-Golgi membranes and samples from the in vitro membrane budding and vesiculation reactions were fixed with 2.5% glutaraldehyde in 0.1 M cacodylate buffer and post-fixed with 1% osmium tetroxide followed by 1% uranyl acetate. The samples were then dehydrated through a series of graded ethanol concentrations and embedded in LX112 resin (LADD Research Industries, Burlington, VT).
Ultra thin sections were cut on a Reichert Ultracut E ultramicrotome, stained with uranyl acetate followed by lead citrate, and viewed on a JEOL 1200EX transmission electron microscope at 80 kV.
Morphometry-To quantitate the relative numbers of Golgi stacks, individual cisternae, and vesicles, the method of Misteli (28). Four categories of membrane profiles were defined. Cisternae were identified as membranes with a length greater than four times the width; stacked Golgi profiles were those with two or more cisternae that were adjacent for more than half their length; small vesicles were those with a size of 50 -100 nm; and large vesicles were those of 100 -350 nm (28). Contaminating membranes such as plasma membrane or endoplasmic reticulum were not included in the analysis. The different membrane profiles were counted by randomly overlaying a transparent grid of 1-cm squares across each micrograph. Each membrane category that intersected a horizontal line and fell within a box that corresponded to a 2.5-m square (actual size) was counted. The proportion of each membrane profile to the total membrane was equal to the number of line intersections per category divided by total number of membrane intersections. Two random areas from at least 20 micrographs of each treatment were counted.

Characterization of Isolated Golgi Membranes-Purified
Golgi membranes were isolated from rat liver according to methods used previously (26). Following centrifugation, the gradient was fractionated, aliquots of alternate fractions were analyzed by SDS-PAGE and Western blotting using antibodies to Golgi, ER, mitochondria, endosomal, and plasma membrane marker proteins (Fig. 1A). TGN38 a trans-Golgi marker, cofractionated on the gradient with an opaque band of material at the interface of 0.5 and 0.86 M sucrose (fractions 23-28). Gradient fractions radiolabeled with both [ 3 H]CMP sialic acid or [ 3 H]UDP galactose had a peak of radioactivity at fraction 27 that co-sedimented with TGN38 (Fig. 1A). Furthermore, this material was highly enriched in Golgi stacks and cisternae, as evinced by electron microscopy (Fig. 1C). Calnexin, a marker for the ER (29), and Tom20, a mitochondrial outer membrane (30) protein, sedimented through the gradient (fractions 30 -39) and overlapped only minimally with Golgi marker proteins. A second band of membranous material floating between 0.86 and 1.3 M sucrose was highly enriched in the plasma membrane as indicated by the localization of connexin43, a gap junction protein (Fig. 1A) and electron microscopy (Fig. 1B). Rab5, a marker for early endosomes, was mostly cytosolic and present in fractions corresponding to the load zone (fractions 1-12). Our recent work demonstrated that in several different cell types PLD1 was detected in the perinuclear Golgi region and bound to Golgi membranes (17). Consequently, it was of interest to identify which fractions contained this enzyme: Most of the PLD1-immunoreactive material was present in two regions of the gradient; those corresponding to the Golgi apparatus (fractions 23-30) and plasma membrane (fractions 35-39; Fig.  1A). Fractions 17 and 18 were enriched in both ARF1 and PLD1, although most ARF1 was recovered in the cytosol. Surprisingly, little PLD2 immunoreactivity was detected in fractions corresponding to the plasma membrane (31). In contrast to PLD1, most of the PLD2-immunoreative material was present in the endosomal, Golgi, and lighter membrane fractions. Material collected from fractions 24 -29, corresponding to Golgi membranes, was pooled and used for subsequent experiments.
Vesicle Budding from Isolated Golgi Membranes-Our initial goal was to radiolabel endogenous glycoproteins in the late FIG. 1. Isolation of rat liver Golgi membranes. Rat liver Golgi membranes were isolated using a sucrose density gradient ("Experimental Procedures"), and 1-ml fractions were collected from the top of the gradient. A, 20 g of protein from alternate gradient fractions was resolved by SDS-PAGE using a 10% gel, transferred to polyvinylidene difluoride membranes, and immunoblotted for the indicated marker proteins. 30 g of protein from the same gradient fractions was assayed for sialyl transferase activity (Ⅺ), galactosyl transferase activity (q), and total protein (f). The ordinate corresponds to the radioactivity in each fraction expressed as a percentage of the total of all fractions. B and C, gradient fractions 32 and 27 were prepared for electron microscopy ("Experimental Procedures"); C, inset, higher magnification of Golgi cisternae. The gradient is representative of nine separate preparations; the coincidence of ARF1 and PLD1 (fraction 17) was evident in five such preparations. The arrowhead indicates the peak of Golgi membrane material.
Golgi/TGN and follow their release into post-Golgi vesicles in vitro. Isolated Golgi membranes were incubated with [ 3 H]CMP sialic acid, and its incorporation into N-linked glycoproteins was determined (Fig. 2). Consistent with earlier observations (22) the translocation of [ 3 H]CMP sialic acid across Golgi membranes and its incorporation into endogenous glycoprotein acceptors was cytosol-and energy-independent ( Fig. 2A). Measurement of the trichloroacetic acid precipitable radioactivity across the sucrose gradient ([ 3 H]sialic acid-labeled or 3 H-galactosylated polypeptides; Fig. 1A) revealed only background radioactivity in fractions corresponding to the cytosol (fractions 1-12) demonstrating incorporation was dependent on the presence of Golgi membranes. Analysis of the total radiolabeled material by SDS-PAGE revealed a heterogeneous distribution of polypeptides ranging in size from ϳ200 to 40 kDa (Fig. 2B,  lanes 7 and 8). To determine the efficiency of vesicle budding from the TGN, we measured the release of [ 3 H]sialic acidlabeled cargo molecules into a medium speed supernatant fraction (Fig. 2, B and C). Samples were incubated in the presence and absence of an energy regeneration system (EGS) and/or cytosol. Following incubation, the putative post-Golgi vesicles (supernatant, S) were separated from residual Golgi membranes (pellet, P) by brief centrifugation (12,000 ϫ g for 1 min). Similar to data obtained using permeabilized cells (27), vesicle release was energy-, cytosol-, and temperature-dependent (Figs. 2B, lanes 3-8, and 2C).

Identification of [ 3 H]Sialic
Acid-labeled Polypeptides-Hepatocytes synthesize numerous specific proteins. We therefore used antibodies to two such soluble secreted glycoproteins, ␣ 1 -antitrypsin and transferrin (Fig. 3A, lanes 5, 6, 9, 10; M r 51,000 and 78,000, respectively) as well as to the Golgi integral membrane protein TGN38 (ϳ90 kDa) to identify these molecules by immunoprecipitation following vesicle release from the TGN (Fig. 3, A and B). All three polypeptides were evident, although their relative distribution in the vesicle (S) and residual Golgi (P) fractions was significantly different. In particular, the release of transferrin-containing vesicles was higher (ϳ20%) than that of TGN38 (ϳ14%; Fig. 3, B and C) and more efficient than release of the total acid-precipitable radioactivity, suggesting selective packaging of cargo molecules in vitro. Whether this represents formation of separate vesicle populations containing different cargo molecules or the differential "filling" of transferrin and TGN38 into the same vesicle remains to be determined. Additionally, two major transferrin- . Aliquots were removed and spotted onto Whatman 3MM discs, and the radioactivity was determined by liquid scintillation counting. B, isolated Golgi membranes were incubated at 20°C or 37°C for 45 min in the presence of an energy generating system (EGS) with and without 2 mg/ml rat brain cytosol as indicated. Following incubation, samples were centrifuged at 12,000 ϫ g for 1 min and the supernatant (S) and pellet (P) were analyzed by SDS-PAGE. C, quantitation of vesicle release from B; % vesicle release ϭ S/(S ϩ P) ϫ 100. Data are the average of two separate experiments.  1-4). The remainder of the samples was treated sequentially (see "Experimental Procedures") with rabbit anti-␣ 1 -antitrypsin (lanes 5, 6), rabbit anti-TGN38 (lanes 7, 8), and rabbit antitransferrin (TF) (lanes 9, 10) and the immunoprecipitates analyzed by SDS-PAGE. B, time course of vesicle budding: [ 3 H]CMP sialic acidlabeled Golgi membranes were incubated under budding conditions for the indicated times. Following incubation, samples were centrifuged briefly, and the pellets and supernatants were incubated with rabbit antibodies to transferrin or TGN38 ("Experimental Procedures"), and the immunoprecipitates were resolved SDS-PAGE. C, quantitation of vesicle release: the intensity of bands corresponding to transferrin (f), TGN38 (Ⅺ) and total trichloroacetic acid-precipitable radiolabeled protein (q) was quantitated by densitometry. The ordinate corresponds to the percentage of the total 3 H-labeled material released into the supernatant fraction; S/(S ϩ P) ϫ 100. Data are the average of two separate experiments.
immunoreactive bands were evident in the residual Golgi (P) fraction (Figs. 3A, lane 10, and 3B) whereas a faster migrating species was released into the vesicle fraction (S) (Fig. 3A, lanes  9 and 10). The latter transferrin molecule may correspond to post-Golgi processing of transferrin in the released vesicles.
If vesicles were released from the Golgi membranes, then cargo molecules should be resistant to exogenously added proteases. Conversely, if the cargo molecules had leaked into the incubation medium from ruptured membranes or were present as a result of nonspecific effects on membranes, the radiolabeled polypeptides would be protease-sensitive. To distinguish between these possibilities, supernatant and pellet fractions were digested with proteinase K in the absence and presence of detergent (Fig. 4). In the absence of detergent, virtually all the polypeptides were protease resistant (Fig. 4A, lanes 3 and 4). In contrast, in the presence of Triton X-100, there was quantitative proteolysis of all polypeptides (lanes 5 and 6). Additional evidence that the radiolabeled cargo molecules were present in vesicles was obtained by use of high speed centrifugation, 100,000 ϫ g for 5 min (Fig. 4; lanes 7-12). If the cargo molecules were present in vesicles, they would be expected to sediment upon high speed centrifugation. Alternatively, if these glycoproteins had leaked into the 12,000 ϫ g supernatant fractions (lanes 2 and 4), they would be unlikely to sediment at 100,000 ϫ g. Consistent with the protease protection data, all the radiolabeled glycoproteins were present in the pellet fraction following centrifugation at 100,000 ϫ g (Fig. 4, lanes 7, 8). Furthermore, solubilization of the membranes with Triton X-100 released all the radiolabeled cargo molecules into the medium and high speed supernatant fractions (Fig. 4, lanes  9 -12), a result that is consistent with these molecules being present in membrane-bound vesicles. To confirm that specific cargo molecules were present in vesicles, aliquots of the pellets and supernatant fractions were incubated with or without proteinase K followed by immunoprecipitation with antibodies to TGN38 (Fig. 4B). Using this assay, ϳ60% of the TGN38-immunoreactive material was protease resistant; it is likely that this was a low estimate, because partial removal of the cytosol tail may have rendered TGN38 less antigenic than the untreated control samples. Together these data support our idea that the radiolabeled cargo molecules were packaged into nascent post-Golgi vesicles.
Phospholipase D Increases Release of Golgi-derived Vesicles-Previous work from our laboratory showed that endogenous or exogenous plant or human PLD stimulated the release of nascent secretory vesicles from the late Golgi apparatus in vitro (18). We therefore determined if exogenous PLD would enhance vesicle release from isolated Golgi membranes; addition of bacterial PLD stimulated vesicle budding ϳ2to 3-fold ( Fig. 4A; lanes 13-16). In agreement with foregoing data, most of the radiolabeled polypeptides in the medium speed supernatant fraction were resistant to proteinase K digestion ( Fig. 4A;  lanes 15, 16). These data further demonstrate that release of vesicles from isolated Golgi membranes did not result from leakage of endogenous cargo molecules from the organelle (see below).
By exploiting the transphosphatidylation activity of PLD, we showed previously that primary but not secondary or tertiary alcohols inhibited vesicle budding from the Golgi apparatus due to lack of PA synthesis (18,19). To determine if treatment of isolated Golgi membranes with alcohol might inhibit Golgi function, isolated membranes were incubated with [ 3 H]sialic acid in the presence of 1-BtOH ( Fig. 2A, diamond). Under these conditions 1-BtOH had no effect on [ 3 H]sialic acid incorpora-

B, budding of TGN38-containing vesicles. [ 3 H]CMP sialic acid-labeled Golgi membranes were incubated under vesicle budding conditions ("Experimental
Procedures") and centrifuged at 12,000 ϫ g for 1 min. The pellets and supernatants were digested with proteinase K after which the protease was inactivated with 1 mM phenylmethylsulfonyl fluoride; the samples were then treated with 1% Triton X-100 and incubated with rabbit anti-TGN38 antiserum ("Experimental Procedures"). The immunoprecipitable material was analyzed by SDS-PAGE followed by fluorography. tion into radiolabeled cargo molecules suggesting that the alcohol per se did not disrupt Golgi function nonspecifically. Based on our earlier results (18,19), we expected that vesicle release from purified Golgi membranes would be inhibited by 1-BtOH; surprisingly, this was not the case. Instead, apparent vesicle release from the Golgi membranes increased in the presence of 1-BtOH (Fig. 4A, lanes 17 and 18); however, this resulted from fragmentation of the organelle into small vesicles (see below). Treatment of the membranes with t-BtOH, which does not participate in the transphosphatidylation reaction, had little effect on vesicle budding (Fig. 4A; compare lanes 17,  18 with 19, 20). In contrast to incubation with PLD, in the presence of 1-BtOH significantly higher levels of vesicles containing [ 3 H]sialic acid-labeled polypeptides were present in the supernatant fraction; furthermore, under these conditions release of cargo material was energy-independent. These results suggested that cargo release was a consequence of fragmentation of the Golgi cisternae rather than vesicle budding.
Morphology of Golgi Membranes-To characterize the Golgi elements during in vitro incubation, we compared the morphology of the Golgi membranes and post-Golgi vesicles following vesicle budding (Fig. 5, A, C, and D). Golgi membranes were incubated under vesicle budding conditions and separated into supernatant and pellet fractions that were prepared for trans-mission electron microscopy. The starting material (see Fig.  1C) contained numerous Golgi stacks with 2 or 3 characteristic flattened cisternae (arrows) as well as cisternal elements and multivesicular bodies (MVBs; arrowheads, Figs. 1C, 5A, and 5C). Following incubation under budding conditions total membranes were sedimented at 100,000 ϫ g, and their morphology was examined (Fig. 5A). Under these conditions abundant individual Golgi cisternae and MVBs were evident, although there were correspondingly fewer Golgi stacks and cisternae present than in the starting material (compare Fig. 1C with Fig. 5, A, C, and D). Furthermore, upon medium speed centrifugation (12,000 ϫ g for 1 min) the residual Golgi cisternae and MVBs were present predominantly in the pellet fraction (Fig.  5C), consistent with the data from the budding assay (Figs. 2  and 3). In contrast, the supernatant fraction contained mostly a heterogeneous population of vesicles ranging in diameter from ϳ50 to 350 nm (Fig. 5D). Although individual Golgi cisternae were present following incubation under vesicle budding conditions (Fig. 5, A and C), no further changes in morphology were seen with longer incubation times (data not shown). Together with the vesicle budding results, our morphological data further demonstrated that release of material into the supernatant fraction was not a consequence of nonspecific fragmentation of the Golgi apparatus. Recent data from our laboratory using rat pituitary GH3 cells demonstrated that, in the absence of PA synthesis, induced by treatment with primary alcohols, the Golgi apparatus became fragmented in vivo (19). To determine if this also occurred in the isolated organelle, control and 1-BtOH-treated samples were analyzed using electron microscopy (cf. Figs. 5A and 6B). Our previous data showed that 1% t-BtOH did not affect Golgi structure or secretion in whole cells; however, to control for possible nonspecific effects of the alcohol, we examined the effect of t-BtOH on the morphology of the isolated membranes (Fig. 5B). In contrast to control post-Golgi vesicles or those treated with t-BtOH in which some cisternae were still present, samples incubated with 1-BtOH consisted of a relatively uniform population of small vesicles (ϳ50-to 100-nm diameter); most significantly, no Golgi cisternae were evident (compare Fig. 5, A and B, with Fig. 6B). The absence of individual cisternae suggested complete fragmentation of the Golgi apparatus (Fig. 6B).
Our previous data (19) showed that, in vivo, wash out of 1-BtOH led to reformation of the Golgi apparatus and restoration of secretion. To determine if alcohol-induced fragmentation of the Golgi apparatus could be reversed in vitro, 1-BtOHtreated Golgi membranes were isolated by centrifugation and resuspended in fresh buffer lacking 1-BtOH (Fig. 6A); this was repeated to remove traces of the alcohol. The washed vesicles were incubated with cytosol and energy after which the samples were prepared for electron microscopy (Fig. 6C). Numerous structures containing one or two putative elongated cisternae were evident, suggesting partial re-assembly of the Golgi apparatus, although fewer bone fide Golgi stacks were present than in the starting material (Fig. 6C, inset). Little, if any Golgi re-assembly was observed in the absence of both ATP and GTP or cytosol.
Morphometry of Golgi Membranes-To quantitate the relative distribution of Golgi stacks, individual cisternae, and vesicles, the method of Misteli was used (28). Analysis of the starting Golgi material showed that it consisted of predominantly stacked cisternae (70% of the total membrane) with only ϳ20% of the membrane present as vesicles (Fig. 5E, "Start"). Following incubation under budding conditions the amount of total membranes present in vesicles increased ϳ3-fold (ϳ17% to 50%) compared with the starting Golgi material, and there was a concomitant decrease in cisternae to ϳ35% of total (Fig.  5E, "Budding"). These data were consistent with the aforementioned protease protection experiments (Fig. 4) and suggested that nascent vesicles were released into the supernatant fraction. As evident from the electron micrographs, after 1-BtOH treatment only membrane vesicles were present; no cisternae or stacks were detected. Indeed, quantitation demonstrated virtually complete vesiculation of the Golgi apparatus (Fig. 5E, "1-butanol"). Most significantly, upon removal of the alcohol, the amount of membrane present in vesicles decreased to ϳ50% of the total membrane population and ϳ45% of the membranes were now in Golgi cisternae and stacks (Fig. 5E, "Washout").
Phosphoinositide Synthesis on Golgi Membranes-Our previous work also showed that the fragmentation of the Golgi apparatus correlated with decreased PtdIns(4,5)P 2 synthesis (19). We expected that Golgi membrane vesiculation would be a consequence of diminished PtdIns(4,5)P 2 synthesis. To test this idea, Golgi membranes were incubated with and without cytosol (as a source of phosphatidylinositol-4-phosphate 5-kinases), [␥-32 P]ATP, in the absence or presence of either 1-or t-BtOH and the radiolabeled inositol phospholipids analyzed (Fig. 7). Inclusion of cytosol resulted in robust PtdIns(4,5)P 2 and PA synthesis (Fig. 7A, lane 2), whereas when 1-BtOH was FIG. 6. Morphology of Golgi vesicles following washout of 1-BtOH. Golgi membranes were treated with 1% 1-BtOH for 15 min in the presence of cytosol and energy (as outlined in A) and an aliquot prepared for electron microscopy (B). The remainder of the sample was diluted 10-fold with buffer and centrifuged at 100,000 ϫ g for 5 min (alcohol washout). The washout, pelleted material was reincubated with cytosol and energy ("Experimental Procedures") for 45 min at 37°C, and an aliquot was prepared for electron microscopy (C). Arrows indicate reformed Golgi cisternae. Insets: B, higher magnification of Golgi-derived vesicles prior to alcohol washout; C, higher magnification of re-assembled Golgi cisternae following alcohol washout. B and C arrowheads, multivesicular bodies; C arrows, cisternae. present, the synthesis of both lipids, particularly PtdIns(4,5)P 2 , was inhibited dramatically (Fig. 7A, lane 4). As expected, t-BtOH had no effect on the synthesis of either lipid and their levels were indistinguishable from controls ( lanes 5, 6). These results further demonstrate a strong correlation between the synthesis of PtdIns(4,5)P 2 and the structural integrity of the Golgi apparatus.
We argued that, if 1-BtOH treatment led to fragmentation of the Golgi apparatus via diminished PtdIns(4,5)P 2 synthesis, then alcohol removal (washout) should allow re-synthesis of this lipid. Following alcohol washout (Fig. 6A) the 1-BtOHderived vesicles were incubated in the presence of cytosol, ATP, and GTP and analyzed: (i) using TLC to assay the synthesis of PtdIns(4,5)P 2 (Fig. 7B) or (ii) for their ability to support vesicle release (Fig. 8). Significantly, following removal of the alcohol, PtdIns(4,5)P 2 synthesis was restored (Fig. 7B, lanes 4 and 5). This result, which was consistent with our earlier observations (19), demonstrated a strong correlation between the synthesis of PtdIns(4,5)P 2 and maintenance of Golgi structure.
Restoration of Vesicle Budding-If Golgi cisternae had reassembled following removal of the alcohol, then, upon incubation under vesicle budding conditions, radiolabeled cargo molecules should be released into the supernatant fraction in an energy-and cytosol-dependent reaction. To test this prediction, Golgi membranes radiolabeled with [ 3 H]sialic acid were treated with 1-BtOH, the alcohol was removed, and the residual membranes were incubated under vesicle budding conditions (Fig. 8). As above (Fig. 2), addition of 1-BtOH resulted in elevated levels of cargo molecules released into the 12,000 ϫ g supernatant fraction, presumably resulting from fragmentation of the organelle (Fig. 8A, lane 3, 4). Similar to control Golgi membranes, upon removal of the alcohol little material was released into the supernatant in the absence of either cytosol and/or energy (lanes 5-10). Strikingly, in the presence of cytosol and energy vesicle budding was restored albeit at ϳ50% of the efficiency of control untreated Golgi membranes ( Fig. 8A; lanes 11, 12; cf. Fig. 2). Together, these results (Figs. 6 -8) suggest that upon alcohol washout functional Golgi cisternae reformed that were able to support vesicle release. DISCUSSION Numerous factors cause disruption of the Golgi apparatus; these include pathological conditions, overexpression of Golgiassociated proteins, and pharmacological agents (32). In several neurological diseases, including amylotrophic lateral sclerosis, the Golgi apparatus in spinal cord neurons is fragmented (33) and resembles that of cells treated with nocodazole, which disrupts microtubule organization. During mitosis the Golgi stacks fragment into vesicle clusters and tubules, which partition into daughter cells, and several factors mediating this process have been characterized in detail previously (34 -37).  3,5) or with (lanes 2, 4, 6) 2 mg/ml rat brain cytosol in the presence of either 1% 1- BtOH (lanes 3, 4) or 1% t-BtOH (lanes 5, 6). Following incubation, total phospholipids were extracted, and the radiolabeled lipids were resolved by TLC followed by autoradiography. B, Golgi membranes were preincubated in the absence (lane 1) or presence (lane 2, 3) of cytosol and energy or with 1% 1-BtOH for 15 min at 37°C (lane 3). The samples were then diluted 10-fold with buffer and centrifuged at 100,000 ϫ g to remove the alcohol (washout; Fig. 6A). The pellets were resuspended and reincubated in the presence of cytosol and energy with [␥-32 P]ATP for 10 min after which the lipids were extracted and analyzed by TLC. PA, phosphatidic acid; PIP, PtdIns(4)P; PIP 2 , PtdIns(4,5)P 2 . The identity of the PtdIns(4,5)P 2 material was confirmed by high performance liquid chromatography (19).  (lanes 3, 4), followed by centrifugation and alcohol washout ( Fig. 7A; "Experimental Procedures"). The washed, fragmented Golgi vesicles were resuspended in buffer and incubated at 37°C for 1 h with buffer alone (lanes 5, 6), energy alone (lanes 7, 8), cytosol alone (lanes 9, 10), or with energy and cytosol (lanes 11,12). Following incubation, samples were centrifuged (12,000 ϫ g for 1 min), and the pellets (P) and supernatants (S) were resolved by SDS-PAGE followed by fluorography. B, the gel in A was scanned by densitometry. Percent release corresponds to the percentage of the total material released into the supernatant fraction. Data are the average of three separate experiments. Error bars correspond to the standard error of the mean. The asterisk denotes values significantly different from control (minus cytosol or minus energy; p Ͻ 0.05), as assessed using the two-tailed paired Student's t test.
Treatment of cells with the sponge metabolite ilumiquinone induces a similar Golgi phenotype to that observed during mitosis (38 -40). More recently, it has been shown that, during apoptosis, the Golgi apparatus fragments into vesicles that are reminiscent of those present during mitotic breakdown (54); in this case fragmentation, which is irreversible, involves the action of several caspases that cleave a Golgi-associated matrix protein (41).
Treatment of cells with the fungal metabolite brefeldin A (BFA) causes collapse of the Golgi apparatus into tubulo-vesicular clusters that fuse with the ER (42). BFA prevents GTP-GDP exchange on several high molecular weight ARF guanine nucleotide exchange factors (GEF) (7). By maintaining ARF in a GDP-bound state in a complex with its GEF, GTP exchange is inhibited resulting in coat (COP-I) dissociation from the membrane and collapse of the Golgi structure (43). In this context, overexpression of a GDP mutant form of ARF1 (44) resulted in a BFA-like phenotype with respect to Golgi morphology. Similarly, overexpression of the low molecular weight ARF GEFs, ARNO-1 or -3 (7), which are relatively BFA-insensitive, also caused fragmentation of the Golgi apparatus and redistribution of molecules to the ER (45). Because ARF1 has been shown to regulate the activity and or recruitment of several Golgiassociated enzymes involved in phospholipid metabolism (6,13), it is possible that the above-mentioned effects resulted from changes in Golgi phospholipid metabolism. In this context, overexpression of a dominant negative mutant form of phosphatidylinositol-4-phosphate kinase ␤, whose Golgi association appears to be ARF-dependent, caused fragmentation of the Golgi apparatus (13). Treatment of cells with inhibitors of phospholipase A2 or overexpression of the enzyme led to marked disruption of the Golgi architecture further suggesting a role for phospholipid-modifying enzymes in maintaining Golgi structure (46,47).
In this present study, we used isolated rat liver Golgi membranes and a novel vesicle budding assay to further understand the fragmentation and reassembly of the Golgi apparatus in response to changes in phosphoinositide synthesis. By utilizing the Golgi sialyl transferase activity to radiolabel endogenous glycoprotein cargo molecules, we demonstrated that release of post-Golgi vesicles was dependent on energy, cytosol, temperature, and time ( Figs. 2 and 3). The release of radiolabeled cargo did not result from leakage of proteins from the Golgi membranes, because the molecules were resistant to exogenously added proteases and released into a high speed supernatant fraction only upon solubilization of the membranes with detergent suggesting their presence in sealed vesicles (Fig. 4). Furthermore, release of cargo into nascent vesicles was selective as evinced by the differential kinetics of transferrin and TGN38 release (Fig. 3).
Phospholipase D and Post-Golgi Vesicle Release-The data presented here confirm earlier studies from our and other laboratories, which showed that a pool of PLD1 is associated with Golgi membranes (16,17). In this present study, addition of bacterial PLD enhanced the selective release of glycoprotein cargo molecules into post-Golgi vesicles (Fig. 4). This observation was consistent with our previous studies in which plant PLD added to permeabilized cells stimulated the budding of nascent secretory granules from the TGN (18). It is noteworthy that the pattern of 3 H-labeled sialic acid-containing proteins released into the nascent vesicle fraction was different from that found upon fragmentation of the Golgi in response to alcohol treatment (Fig. 4) further demonstrating the specificity of vesicle release. At present the mechanism whereby PLD enhances the packaging of specific hepatocyte molecules into post-Golgi vesicles is unknown.
Data from our laboratory demonstrated that the PLD activity associated with Golgi membranes regulates PtdIns(4,5)P 2 in the Golgi apparatus (19). In part, PtdIns(4,5)P 2 functions to release nascent secretory vesicles and in maintaining the structure of the Golgi apparatus, although the mechanism of how this occurs remains to be determined. Protein kinase C regulates PLD activity and stimulates nascent vesicle budding (48). In this context, incubation of isolated Golgi membranes with calphostin C, an inhibitor of protein kinase C, (49) and PLD (50) also led to fragmentation of the Golgi apparatus similar to that observed following butanol treatment. Most significantly, the calphostin C-induced Golgi fragmentation occurred in the absence of its effect on protein kinase C, because other inhibitors of the enzyme had no effect on Golgi morphology (49). We speculate that Golgi fragmentation in response to calphostin C occurred via its inhibition of PLD activity.
PtdIns(4,5)P 2 and Formation of Golgi Cisternae-In contrast to other studies (20), our data showed that inhibition of PA synthesis by treatment with a primary alcohol (1-BtOH) but not t-BtOH led to a dramatic decrease in Golgi PtdIns(4,5)P 2 synthesis with little effect on the synthesis of PtdIns(4)P. This correlated with the in vitro fragmentation of the Golgi cisternae into a relatively uniform population of 50-to 100-nm diameter vesicles (Fig. 6). A likely explanation for the discrepancy between our present results and those reports may be due to differences in cell fractionation that could lead to diminished recovery of Golgi membranes. It is possible that the PtdIns(4,5)P 2 synthesis and its inhibition by 1-BtOH that we observed in isolated Golgi membranes (Fig. 7) resulted from contamination by plasma membrane-derived vesicles. However, given the low level of plasma membrane proteins in our Golgi fractions (Fig. 1), this seems unlikely. The present data further support our and other earlier observations that PA stimulates type I phosphatidylinositol-4-phosphate 5-kinase, the final enzyme involved in PtdIns(4,5)P 2 synthesis (19,51,56). Together, these data suggest that PA and PtdIns(4,5)P 2 synthesis are required for maintaining the structure and function of the Golgi.
Strikingly, following washout of 1-BtOH from Golgi membranes in vitro PtdIns(4,5)P 2 synthesis was restored as was the formation of Golgi cisternae, albeit inefficiently (Figs. 6 and 7). Most significantly, the reassembled Golgi cisternae were capable of supporting limited vesicle release (Fig. 8). Together, these data strengthen the correlation of PtdIns(4,5)P 2 synthesis and the maintenance of Golgi structure and function. It might be argued that the effects of 1-BtOH on Golgi morphology were independent of phospholipid synthesis. Although such an explanation is possible, several observations suggest it is unlikely. First, treatment of Golgi membranes with t-BtOH (up to 3.5%) had no effect on Golgi structure in vitro (Fig. 5) or in vivo (19). Second, Golgi fragmentation was rapidly reversible in vivo (19) and in vitro following incubation with cytosol and energy, and this correlated with renewed PtdIns(4,5)P 2 synthesis (Figs. 6 and 7). If the alcohol were extracting lipids or protein from the membrane nonspecifically, it is unlikely this would be a readily reversible reaction. Finally, only primary but not secondary or tertiary alcohols inhibit the production of phosphatidic acid (31).
Although re-assembly of mitotic Golgi membrane fragments has been observed following incubation with an interphase cytosolic extract (35), this is the first observation that fragmentation of the Golgi apparatus in response to decreased phosphoinositide synthesis, was also reversible in vitro. This observation will enable us to isolate factors that restore both PtdIns(4,5)P 2 synthesis and Golgi reassembly in vitro. At present it is unclear how decreased PtdIns(4,5)P 2 synthesis results in fragmentation of the Golgi apparatus. Our working hypothesis is that specific structural proteins possessing a pleckstrin homology domain interact with this lipid thereby enhancing their recruitment and binding to the Golgi apparatus. A candidate for such a PtdIns(4,5)P 2 -binding protein is a Golgispecific isoform of spectrin, designated ␤-III spectrin (52). Indeed, Godi et al. (53) demonstrated that the binding of ␤-III spectrin to Golgi membranes is inhibited in vitro in the absence of PtdIns(4,5)P 2 . We speculate that, in part, the fragmentation of the Golgi apparatus and its re-assembly in vitro may be a consequence of ␤-III spectrin binding to the membranes. Currently, we are testing this model.