Activated ADP-ribosylation Factor Assembles Distinct Pools of Actin on Golgi Membranes*

The small GTP-binding protein ADP-ribosylation factor (ARF) has been shown to regulate the interaction of actin and actin-binding proteins with the Golgi apparatus. Here we report that ARF activation stimulates the assembly of distinct pools of actin on Golgi membranes. One pool of actin cofractionates with coatomer (COPI)-coated vesicles and is sensitive to salt extraction and the plus end actin-binding toxin cytochalasin D. A second ARF-dependent actin pool remains on the Golgi membranes following vesicle extraction and is insensitive to cytochalasin D. Isolation of the salt-extractable ARF-de-pendent actin from the Golgi reveals that it is bound to a distinct repertoire of actin-binding proteins. The two abundant actin-binding proteins of the ARF-dependent actin complex are identified as spectrin and drebrin. We show that drebrin is a specific component of the cytochalasin D-sensitive, ARF-dependent actin pool on the Golgi. Finally, we show that depolymerization of this actin pool with cytochalasin D increases the extent of the salt-dependent release of COPI-coated vesicles from the Golgi following cell-free budding reactions. Together these data suggest that regulation of the actin-based Polypeptide Sequence Determination and Mass Spectrometry— Iden- tification by mass spectrometry was done by cutting protein bands from SDS-PAGE gels and digesting the material with 0.2 m g of trypsin (24). The resulting peptide mixture was then loaded onto a 2- m l bed volume of Poros 50 R2 (PerSeptive Biosystems, Foster City, CA) reversed-phase beads (packed into an Eppendorf gel-loading tip) and eluted stepwise with 4 m l of 16% (and then with 4 m l of 30%) acetonitrile, 0.1% formic acid (25). The “16%” and “30%” peptide pools each analyzed twice by matrix-assisted laser desorption/ionization time of flight mass spec- trometry in the presence and absence of peptide calibrants (25) using a REFLEX III (Bru¨ker-Franzen, Bremen, Germany) instrument equipped with a gridless pulsed extraction ion source and a 2-GHz digitizer and operated in reflectron mode. Spectra were obtained by averaging multiple signals. After recalibration with internal standards, monoisotopic masses were assigned for the most prominent peaks, and a peptide mass list was generated to search a protein non-redundant data base (National Center for Biotechnology Information, Bethesda, MD) using the PeptideSearch (26) algorithm with an accuracy require-ment of 40 ppm. For the identification of the 280-kDa band, 35 of 63 experimental masses matched human a -spectrin, and 15 of the remaining 28 masses matched human b -spectrin (random matches, 12 of 63). For the identification of the 120-kDa band, 8 of 13 experimental masses matched human drebrin (random matches, 4 of 13).

Extensive biochemical and genetic studies have allowed a detailed description of the molecular events surrounding the assembly of coated transport vesicles (1,2). In particular, GTPases of the ARF 1 family act as molecular switches to trigger the assembly of the coat proteins. The ARF-dependent oligomerization of coat proteins has been shown to be sufficient to deform the membrane into a vesicle bud (3,4). Despite this progress, many details have yet to emerge regarding the mechanisms of cargo packaging, the regulation of vesicle formation, and the nature of interactions with molecular motors and other cytoskeletal proteins that allow directed movement.
Although ARF is best characterized for its role in vesicle formation, recent studies suggest it may serve additional functions in the cell. One role for ARF may be to regulate the activity of the phospholipid-modifying enzymes phospholipase D, phosphatidylinositol 4-kinase, and phosphatidylinositol 5-kinase (5)(6)(7). This regulation may play a role in vesicle release and/or be part of signaling pathways not directly related to transport. A second role for ARF may be in directing the assembly of the actin-based cytoskeleton. ARF has been shown to regulate the binding of spectrin, ankyrin, and actin to the Golgi membranes (8,9). ARF6 has been implicated in rearranging the actin cytoskeleton at the plasma membrane (7,10,11). ARF isoforms may also play a similar role in yeast (12).
There is considerable evidence that the actin cytoskeleton and myosin motors are important for Golgi function in the cell (13,14). Recent studies show that actin is likely to be a component of Golgi-derived transport vesicles (15,16). In this study centractin, tropomyosin isoforms, and myosin isoforms were found to be selectively associated with distinct classes of Golgiderived vesicles. Some isoforms of spectrin and ankyrin localize to the Golgi apparatus, suggesting that this is a site of their action (9,17,18). The actin cytoskeleton may also play a role in Golgi morphology and positioning (19).
Here we present evidence that at least two distinct pools of actin are assembled on the Golgi membrane upon ARF activation. This actin affects the release of COPI-coated vesicles from the Golgi and regulates the association of spectrin and the actin-binding protein drebrin with the membrane. The implications of these findings for the ability of COPI-coated vesicles to mediate multiple trafficking steps and for the role of the cytoskeleton in Golgi function will be discussed.
Golgi Binding Assays-The final reaction conditions included 25 mM HEPES (pH 7.2), 2.5 mM magnesium acetate, 15 mM potassium chloride, 0.2 M sucrose, Golgi membranes (0.2 mg/ml), bovine brain cytosol (1.0 mg/ml), and an ATP-regenerating system. The incubations were carried out for 20 min at 37°C, and the final reaction volume was 1 ml. When specified, brefeldin A, cytochalasin D, or latrunculin A was added at the indicated concentration. For reactions containing brefeldin A, the membranes and cytosol were preincubated for 10 min at 37°C with the toxin or with a solvent control. For "float-up" binding assays, the membranes were isolated following the incubation by centrifugation at 15,000 ϫ g for 30 min at 4°C in a refrigerated microcentrifuge. The membranes were then resuspended in 50 l of 45% (w/w) sucrose in 25 mM HEPES (pH 7.2), 25 mM KCl and placed into a 7 ϫ 20-mm ultracentrifugation tube. The sample was overlaid with 125 l of 35% (w/w) sucrose in 25 mM HEPES (pH 7.2), 25 mM KCl and then 25 l of 15% (w/w) sucrose in 25 mM HEPES (pH 7.2), 25 mM KCl. The sample was spun at 100,000 rpm for 30 min in a TLA-100 rotor (Beckman Instruments). A 100-l sample containing the Golgi membranes was removed from the top of the step gradient, and the membrane proteins were precipitated by the addition of trichloroacetic acid to 10%. The pellet from the step gradient and the TCA precipitate were analyzed by SDS-PAGE and Western blotting. For two-stage incubations the conditions were just as described above except that following the first incubation, the membranes were reisolated by microcentrifugation and incubated with a second reaction mixture prior to the float-up centrifugation.
COPI-coated Vesicle Budding Reactions-Unless otherwise indicated, the final budding reaction conditions were identical to those used for binding reactions except that the reaction volume was 2.0 ml. Following the incubation, the Golgi membranes were isolated by centrifugation at 15,000 ϫ g for 15 min and washed one time with LSSB (25 mM HEPES (pH 7.2), 2.5 mM magnesium acetate, 50 mM potassium chloride, 0.2 M sucrose). The vesicles were stripped from the membrane by incubating in HSSB (25 mM HEPES (pH 7.2), 2.5 mM magnesium acetate, 250 mM potassium chloride, 0.2 M sucrose) for 10 min on ice. Following the incubation, the membranes were centrifuged at 15,000 ϫ g for 15 min. Where indicated, the pellet containing the Golgi remnants was resuspended in Laemmli sample buffer for Western blot analysis. The supernatant was loaded onto a 35% (w/w) sucrose cushion (25 mM HEPES (pH 7.2), 2.5 mM magnesium acetate, 250 mM potassium chloride, 35% (w/w) sucrose) and centrifuged at 350,000 ϫ g for 30 min. The 35% sucrose pellets were resuspended in Laemmli sample buffer and analyzed by Western blotting.
Western Blotting-Proteins were fractionated using SDS-PAGE and blotted onto polyvinylidene difluoride membranes using standard protocol for the Bio-Rad minigel and blotting apparatuses. Following the transfer, the membranes were dried and incubated with appropriate dilutions of the indicated primary antibodies. The signal was visualized using horseradish peroxidase-conjugated secondary antibodies (Bio-Rad) and ECL (Amersham Pharmacia Biotech). Where indicated, the signals were quantitated using densitometry.
Polypeptide Sequence Determination and Mass Spectrometry-Identification by mass spectrometry was done by cutting protein bands from SDS-PAGE gels and digesting the material with 0.2 g of trypsin (24). The resulting peptide mixture was then loaded onto a 2-l bed volume of Poros 50 R2 (PerSeptive Biosystems, Foster City, CA) reversed-phase beads (packed into an Eppendorf gel-loading tip) and eluted stepwise with 4 l of 16% (and then with 4 l of 30%) acetonitrile, 0.1% formic acid (25). The "16%" and "30%" peptide pools were each analyzed twice by matrix-assisted laser desorption/ionization time of flight mass spectrometry in the presence and absence of peptide calibrants (25) using a REFLEX III (Brü ker-Franzen, Bremen, Germany) instrument equipped with a gridless pulsed extraction ion source and a 2-GHz digitizer and operated in reflectron mode. Spectra were obtained by averaging multiple signals. After recalibration with internal standards, monoisotopic masses were assigned for the most prominent peaks, and a peptide mass list was generated to search a protein non-redundant data base (National Center for Biotechnology Information, Bethesda, MD) using the PeptideSearch (26)

RESULTS
While characterizing the protein components of Golgi-derived transport vesicles, we observed that non-muscle actin is a major GTP␥S-dependent protein constituent of both Golgi membranes and the Golgi-derived vesicles following incubations with cytosol (data not shown). This finding, together with recent reports indicating that ARF regulates the levels of actin, spectrin, and ankyrin on Golgi membranes (8) and that actinbinding proteins associate with vesicles (15,16), led us to begin dissecting the nature and role of actin on the Golgi apparatus. As a first step, cell-free Golgi binding assays were used to characterize the conditions for GTP␥S/ARF-dependent actin association with the Golgi membranes. We developed a float-up binding assay that allowed actin polymerization to be assayed independently of Golgi binding (Fig. 1, see "Experimental Procedures"). In short, Golgi membranes were incubated with cytosol under conditions known to promote coated vesicle formation. Following the incubation, the membranes were resuspended in 45% sucrose and then overlaid with 35% su-crose in a centrifuge tube. Because Golgi membranes have a buoyant density equivalent to about 30% sucrose, they float to the top of the gradient during ultracentrifugation (Fig. 1A, G fraction). By contrast, large cytoskeletal polymers pelleted at the bottom of the centrifuge tube ( Fig. 1A, P fraction). The presence of the Golgi membranes in the G fraction is confirmed by blotting for the cis-Golgi marker, KDEL receptor (Fig. 1A).
In the absence of GTP␥S, the Golgi membranes bound neither to G-actin, present in the cytosol (Fig. 1A, lanes 1 and 2), nor to F-actin, assembled by the addition of the actin-polymerizing toxin phalloidin (Fig. 1A, lanes 5 and 6). In the presence of phalloidin, the F-actin was completely resolved from the Golgi membranes on the step gradient. By contrast, when GTP␥S was included in the incubation (Fig. 1A, lanes 3 and 4) 50 -80% of the actin was present in the G fraction. The GTP␥Sdependent binding of actin to the Golgi was completely inhibited by the addition of brefeldin A, indicating that the GTP␥S effect is mediated by ARF (Fig. 1B). Brefeldin A inhibits nucleotide exchange for ARF1 but not for ARF6 or the rho family of GTPases (27,28), indicating that the brefeldin A-sensitive actin binding we observe is distinct from the modifications of the actin cytoskeleton involving these GTPases. If either cytosol or the membranes are omitted from the reaction, no actin appears in the G fraction (Fig. 1C). These findings suggest that actin is assembled directly by Golgi membranes upon activation of ARF with GTP␥S.
As a more definitive test of the model that ARF activation leads to changes on Golgi membranes that subsequently allow them to assemble actin, we used two-stage binding reactions (Fig. 2). In the first stage, cytosol, GTP␥S, and membranes were incubated either with or without the actin monomerbinding toxin latrunculin A to block actin assembly. Fig. 2 shows that upon reisolation of the membranes following the first stage, actin bound in a GTP␥S-dependent manner and that this binding was completely blocked by the presence of latrunculin A. If the reisolated Golgi membranes were then incubated with cytosol in a second stage without GTP␥S and latrunculin A, actin was found to be present on the membranes provided that GTP␥S had been included in stage 1. Actin assembled on the Golgi without GTP␥S in stage 2 even though actin assembly had been blocked by latrunculin A in stage 1. This result supports the model that actin assembles directly on the Golgi in the presence of membrane-bound activated ARF, as opposed to a model whereby the Golgi membranes associate with actin that had been previously assembled through a membrane-independent but GTP␥S-dependent mechanism. Because the 21-kDa ARF proteins are the major small GTPases to associate with the Golgi membranes when incubated with GTP␥S (29) and no additional G proteins are activated in stage 2, this result also provides strong additional evidence that the observed GTP␥S effects on actin assembly are ARF-mediated.
Because ARF triggers coated vesicle assembly and the conditions used above for ARF-dependent actin binding are known to bud COPI-coated vesicles from the Golgi, we wished to examine whether the ARF-dependent actin pool was involved in vesicle assembly or release. Following cell-free budding reactions, COPI-coated vesicles are typically stripped from the membrane with 250 mM KCl and then purified using an isopycnic sucrose gradient (20) or by pelleting through a sucrose cushion (30). We find that much of the ARF-dependent actin is extracted from the Golgi with 250 mM KCl (as are COPI-coated vesicles) and pellets through a 35% cushion during ultracentrifugation (Fig. 3). This finding could indicate that the ARFdependent actin is directly associated with the vesicles.
In addition to the actin that cofractionated with the vesicles, we observed that some actin is resistant to the salt extraction and remained on the Golgi membranes left behind after removing the vesicles (Fig. 3, Golgi remnants). Surprisingly, although the appearance of actin in the vesicle-enriched 35% sucrose pellet was completely blocked by the addition of the plus-endbinding toxin cytochalasin D, the levels of actin on the Golgi remnants were largely unaffected by the presence of this toxin (Fig. 3). Blotting for the vesicle marker ⑀-COP revealed that the levels of Golgi membrane and vesicles were unaffected by the toxin treatment. Tubulin levels, both on the Golgi remnants and in the 35% sucrose pellet, were also unaffected by GTP␥S or cytochalasin D, indicating that these reagents did not have effects on microtubule assembly or disassembly. The actin monomer-binding toxin latrunculin A reduced the levels of actin in both the 35% sucrose pellet and the Golgi remnants (Fig. 3B). These results indicate the presence of distinct pools of actin with unique toxin sensitivities on the Golgi membranes following incubation with ARF.
In an effort to identify factors that may confer these distinct properties to Golgi-associated actin, we examined the protein composition of the ARF-dependent salt-extractable actin pool. Fig. 4 shows that when GTP␥S-assembled actin is extracted from the Golgi and pelleted through a cushion, a remarkably simple set of abundant proteins is present (Fig. 4, lane 2). Besides actin, two proteins of 120 and 220 kDa disappeared when cytochalasin D was added to the reaction (lane 1), indicating that these proteins are actin-associated. This was an unexpectedly small number of actin-bound proteins, given that these incubations were carried out with a whole cytosol preparation. To rule out that our cytosol was devoid of additional actin-binding proteins, we incubated cytosol alone under the identical conditions used in the budding reaction in the presence (lane 3) or absence (lane 4) of phalloidin to polymerize the actin. The phalloidin-polymerized actin was extracted with the 250 mM high salt stripping buffer using the same conditions as for the standard budding reaction. The phalloidin-polymerized actin contained several additional major actin-binding proteins not found on the ARF-dependent actin. Whether or not the 250 mM KCl wash was carried out on the phalloidin-polymerized actin had no effect on the levels of actin-binding proteins, ruling out that actin-binding proteins are lost during the vesicle/actin extractions (data not shown). Thus, based on the salt extractability, the toxin sensitivity, and the protein composi-  tion, the ARF-dependent actin in the 35% sucrose pellet is distinct.
To identify the putative actin-binding proteins, the tryptic fragments of the 120-and 220-kDa polypeptides were analyzed by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry. Using mass fingerprint analysis, the identity of these proteins was determined to be drebrin and a mixture of ␣and ␤-spectrin, respectively. Spectrin has been previously identified as part of an ARF-regulated cytoskeleton on the Golgi and Golgi-derived vesicles (8,9). Drebrin has mostly been characterized as a protein involved in neuronal differentiation (31,32). Interestingly, the yeast homolog of drebrin, ABP1, has been implicated in endocytosis, suggesting that this family of actin-binding proteins could play a role in trafficking (33).
We confirmed that drebrin binding to the Golgi membranes is GTP␥S-and ARF-dependent using a Western blot analysis of the vesicle-budding reaction. Drebrin appeared in the 35% sucrose pellet and on Golgi remnants in a GTP␥S-dependent manner (Fig. 5). The addition of brefeldin A completely blocked the appearance of drebrin in both fractions, indicating that drebrin is ARF-dependent. When cytochalasin D was included in the incubation, actin and drebrin completely disappeared from the 35% sucrose pellet. The cytochalasin D resistant pool of actin is observed on the Golgi remnants (lane 7) and is found to be ARF-dependent. Interestingly, drebrin is not associated with this pool of actin. Together these data suggest that ARF activation leads to the assembly of at least two pools of actin on the Golgi. The first pool is cytochalasin D-sensitive, salt-extractable, and contains drebrin. The second pool is less extractable with salt, is cytochalasin D-resistant, and does not contain drebrin.
Because the same small GTPase, ARF, triggered vesicle formation and the assembly of actin on the Golgi and because actin may be directly associated with the Golgi-derived vesicles (Fig. 1) (15, 16), we reasoned that the ARF-dependent actin assembly could play a role in vesicle assembly or release. Therefore, we examined the effects of adding the actin-binding toxin cytochalasin D on COPI vesicle formation using the cellfree budding reaction. Addition of the toxin had no effect on the binding of ARF and coatomer to Golgi membranes, suggesting that actin does not play a role in these early events of COPIcoated vesicle assembly (data not shown). Effects were observed, however, on the release of vesicles following the incubation. Typically, vesicles are extracted from the Golgi membranes following cell-free budding reactions by a high salt wash (20). At the standard salt concentration (250 mM KCl), cytochalasin D had little effect on the release of COPI-coated vesicles (Fig. 6). Studies on the giantin/GM130/p115-mediated docking of COPI-coated vesicles revealed that the effects of disruption of this complex on vesicle release could be observed when lower salt concentrations were used for extracting vesicles (30). Because some aspects of vesicle docking and release are salt-sensitive, we examined the effects of cytochalasin D on COPI-coated vesicle release at lower salt concentrations. Fig. 6 shows that when vesicles were stripped at the lower salt concentrations, 150 and 75 mM KCl, significantly more ␤-COP and KDEL receptor were found in the 35% sucrose pellet when cytochalasin D was included in the incubations. Therefore, the salt-dependent vesicle release appears to be facilitated by interfering with actin assembly on the Golgi membranes. These results indicate that ARF-mediated changes in the cytoskeleton could play an important role in regulating vesicular trafficking through the Golgi. DISCUSSION We describe an ARF-dependent salt extractable actin pool assembled from whole cytosol with a remarkably simple protein composition. One of the two major actin-binding proteins of this complex was identified as a mixture of ␣and ␤-spectrin, confirming previous studies showing an ARF-mediated interaction between spectrin and the Golgi (8). The second protein was identified as drebrin. Previously, drebrin had been characterized as a neuron-specific protein that may play a role in neuronal development. In particular, it has been localized to dendrites and may play a role in dendrite outgrowth. If drebrin were restricted to neuronal dendrites, it would be unlikely that it plays a general role in Golgi transport or other vesicular trafficking steps. Several findings indicate that this is not the case. First, we find by Western blot analysis that a protein identical in size to brain cytosol drebrin is recognized by the drebrin antibody in Chinese hamster ovary cells (data not shown), indicating that drebrin or a closely related isoform is present in non-neuronal cells. Second, other recent studies have found drebrin to be present in several non-neuronal tissues (34,35). Finally, a homolog of drebrin, ABP1, is present in the yeast Saccharomyces cerivisiae. ABP1 interacts genetically with genes encoding proteins required for endocytosis (33). This finding, together with the results we report here, raises the possibility that drebrin is involved in vesicular trafficking. Additional studies will be required to determine the precise role of drebrin or drebrin-related proteins in these processes.
The effects of cytochalasin D on the salt-dependent extraction of COPI-coated vesicles could indicate a role for actin in the fission reaction or vesicle release. It is, however, equally likely that actin plays a role in a vesicle tethering or docking event that is not directly related to the fission reaction. In this respect, it is interesting that similar results have been obtained when the function of the GM130/giantin complex is compromised. This complex is thought to tether vesicles to the Golgi after they are formed (36) or to dock vesicles to a target organelle prior to SNARE-mediated fusion (37,38). Although additional studies will be required to establish the precise role of actin in these trafficking steps, the cytochalasin D effects indicate that the ARF-dependent actin plays a role in the interaction of COPI vesicles with the Golgi and therefore is likely to function during COPI-mediated trafficking to, from, or within the Golgi.
It is of interest to speculate why pools of actin with distinct properties are assembled on the Golgi. One possibility is that the different pools of actin reflect different classes of assembling vesicles. ARF triggers the assembly of both COPI-coated and AP1/clathrin-coated vesicles on the Golgi apparatus (39,40). It is possible that ARF directs the assembly of distinct actin pools for different classes of vesicles. Additional evidence supporting this model comes from the recent finding that different classes of Golgi-derived vesicles associate with distinct actin-binding proteins (15). A second possibility is that ARF directs the assembly of different actin pools on different types of membranes. For example, actin assembled on cis-Golgi membranes may have different properties from actin assembled on trans-Golgi membranes. Assembling actin with specific properties on each type of membrane or at each class of vesicle assembly site could direct the overall organization, positioning, or morphology of these organelles. Alternatively or in addition, it could be involved in conferring cytoskeleton-dependent directionality and targeting to the transport process.
Targeting proteins in the cell almost certainly requires specific interactions with molecular motor proteins and cytoskeletal elements to move them in the proper direction to the correct location in the cell. In some instances, the same class of vesicle coat can be used for multiple transport steps; for example, COPI-coated vesicles appear to be used for anterograde and retrograde trafficking from Golgi cisternae as well as in the endocytic pathway (41)(42)(43)(44)(45). Therefore, it is highly probable that there is cellular machinery in addition to the coat proteins that specify interactions with the cytoskeleton. The ARF-mediated changes in the actin cytoskeleton may be part of this machinery. Thus ARF-regulated vesicle assembly may not only involve the directed oligomerization of coat proteins but also additional complex regulation of cytoskeletal components to ensure that the assembled vesicle moves along a specific cytoskeletal filament with the correct molecular motor and thus arrives at the proper destination in the cell.