Intracellular distribution of Arf proteins in mammalian cells. Arf6 is uniquely localized to the plasma membrane.

Subcellular distributions of the five human Arf proteins were examined, using a set of isoform-specific polyclonal and a pan-Arf monoclonal antibodies. Subcellular fractionation of cultured mammalian cells allowed the demonstration that Arf6 is uniquely localized to the plasma membranes of Chinese hamster ovary cells. The plasma membrane distrubution was unaffected by either GTPgammaS (guanosine 5'-O-(3-thio)triphosphate) or brefeldin A, an activator and inhibitor of Arf activities, respectively. In contrast, Arf proteins 1, 3, 4, and 5 were predominantly cytosolic but could be recruited to a variety of intracellular membranes, but not plasma membranes, upon incubation in the presence of GTPgammaS. The GTPgammaS-promoted binding of the cytosolic Arf proteins to membranes was blocked by brefeldin A. The stable association of Arf6 with plasma membranes and the insensitivity of its localization to either GTPgammaS or brefeldin A revealed a clear distinction between Arf6 and the other Arf isoforms. Localization of Arf6 to the plasma membrane suggests a unique cellular role for this isoform at the plasma membrane, but failure to find endogenous Arf6 on endocytic structures, including clathrin-coated vesicles, appears inconsistent with the proposed role of Arf6 in assembly of coat structures or endosomes in transfected fibroblasts (1,2).

Subcellular distributions of the five human Arf proteins were examined, using a set of isoform-specific polyclonal and a pan-Arf monoclonal antibodies. Subcellular fractionation of cultured mammalian cells allowed the demonstration that Arf6 is uniquely localized to the plasma membranes of Chinese hamster ovary cells. The plasma membrane distrubution was unaffected by either GTP␥S (guanosine 5-O-(3-thio)triphosphate) or brefeldin A, an activator and inhibitor of Arf activities, respectively. In contrast, Arf proteins 1, 3, 4, and 5 were predominantly cytosolic but could be recruited to a variety of intracellular membranes, but not plasma membranes, upon incubation in the presence of GTP␥S. The GTP␥S-promoted binding of the cytosolic Arf proteins to membranes was blocked by brefeldin A. The stable association of Arf6 with plasma membranes and the insensitivity of its localization to either GTP␥S or brefeldin A revealed a clear distinction between Arf6 and the other Arf isoforms. Localization of Arf6 to the plasma membrane suggests a unique cellular role for this isoform at the plasma membrane, but failure to find endogenous Arf6 on endocytic structures, including clathrin-coated vesicles, appears inconsistent with the proposed role of Arf6 in assembly of coat structures or endosomes in transfected fibroblasts (1, 2).
The ADP-ribosylation factor (Arf) 1 family is a group of structurally related proteins that form a subset of the Ras superfamily of regulatory GTP-binding proteins (for a recent review, see Ref. 3). In addition to serving as cofactors for cholera toxin-catalyzed ADP-ribosylation, Arf proteins have more recently been associated with a wide array of functions. These include acting as regulators of the binding of coat proteins and adaptins to intracellular membranes (4,5), activators of phospholipase D (6,7), regulators of ER and Golgi morphology and function (8,9), and cytosolic factors conferring sensitivity to GTP␥S in cell-free assays of intra-Golgi (10 -12) and ER-Golgi transport (13), and endosome (14) and nuclear membrane fusion (15,16).
The importance of Arf proteins in both membrane traffic and organelle organization was manifest due to the sensitivities of most Arf proteins to both GTP␥S, a slowly hydrolyzable GTP analog, and to brefeldin A (BFA), a fungal metabolite capable of inhibiting guanine nucleotide exchange on Arf in a crude system (17)(18)(19). The activation (GTP binding) and deactivation (GTP hydrolysis) cycle of Arf action in cells is thought to coincide with its binding and release, respectively, from intracellular membranes. In this model, activation of a soluble Arf protein results in its translocation to a membrane and the recruitment, through unknown mechanisms, of coat proteins or adaptor complexes to that membrane. It remains unclear how, or even if, Arf-mediated activation of phospholipase D or cholera toxin relate to mechanisms of regulation of membrane transport by Arf proteins.
Six mammalian Arf genes have been described, encoding proteins whose predicted sequences are highly conserved (3). Human, bovine, and murine Arf1, for example, are 100% identical. Indeed, the only differences in primary sequences of Arf isoforms between human and rodents appear in Arf4. There is also considerable conservation of Arf function across wide species boundaries. Saccharomyces cerevisiae has only two Arf genes, but each of the five known human Arf proteins can restore vegetative growth to yeast mutants harboring an otherwise lethal deletion of both yeast Arf proteins (20 -22).
If all human Arf proteins are functionally redundant, why are there so many Arf genes? Are they expressed in different tissues, or at different times during development, or localized to discrete intracellular sites? Recent investigations of Arf function in transfected cells has begun to suggest that there may be at least some specificity in the activities of individual Arf proteins. Although Arf1, Arf3, and Arf4 appear to be functionally redundant, 2 results from cell lines overexpressing wild type or mutant Arf1 or Arf6 have led to the conclusion that Arf1 acts at the Golgi while Arf6 is found at the plasma membrane and in endosomes (1,2). Like Rab proteins, Arf proteins may act to control events on the organelles to which they are localized. On the other hand, the fact that a single Arf (Arf1) can mediate the in vitro recruitment of two distinct coat proteins (AP-1 adaptors, COP-I coat components) to three distinct compartments (trans-Golgi network, cis-Golgi/intermediate com-partment/ER, and endosomes (4, 5, 24 -27) suggests that a simple one Arf per organelle relationship may not exist. As overexpression or epitope tagging may alter the function or distribution of Arf proteins or even cell viability (28), 3 it is important to establish the distribution and membrane binding activities of endogenous Arf proteins when possible. Using isoform-specific antibodies, we have determined the tissue distribution, intracellular localization, and membrane binding activities of each of the five known human Arf proteins. Surprisingly, Arf6 was found to behave quite distinctly from the other four human Arf proteins, being present exclusively, and apparently permanently, on the plasma membrane.

MATERIALS AND METHODS
Peptide-directed Antibody Production-Peptides derived from human Arf proteins (see Fig. 1) were conjugated to keyhole limpet hemocyanin, as described previously (30). Each conjugated peptide was suspended in complete Freund's adjuvant and injected intradermally at multiple sites into three rabbits, with two boosts at 3-week intervals. The sera were tested for reactivity with the antigenic peptide by ELISA. All reactive sera were tested for specificity by immunoblotting against purified recombinant Arf proteins, and the most sensitive and specific antisera were chosen for more extensive characterization and subsequent work. None of the antibodies described herein were found to cross-react with the purified recombinant Arf-like proteins, Arl2 or Arl3 (data not shown), each of which share 40 -50% sequence identity with Arf proteins.
The antisera raised against Arf1 (R-1026), Arf5 (R-1525), and Arf6 (R-1471) were found to be highly specific for their respective proteins (Fig. 2). The antiserum raised against Arf3 (R-1023) cross-reacted with Arf1, but could be made specific for Arf3 by preincubating the antibody with the corresponding Arf1 peptide, P-34. The Arf4 antiserum, R-891, was found to cross-react with Arf5, but this reactivity was not specific as it was not blocked by preincubation with the immunizing peptide (P-19; see Fig. 2).
Monoclonal Antibody Production-Mouse monoclonal antibodies specific for Arf proteins were produced according to standard procedures (31) at Hazelton Laboratories. Purified recombinant human Arf1 was used to immunize mice and assay cell fusions and cultures by ELISA. Culture supernatants from hybridomas positive by ELISA were then assayed for the ability to recognize Arf1 in immunoblots. Four cell lines were positive in both assays and hybridomas were cloned by three rounds of limited dilution. Only one of these, termed 1D9, was used in the current study. Specific antibodies were purified from pooled ascites of 10 mice injected with hybridoma 1D9 by affinity chromatography on a protein G-agarose column. The purified antibody was dialyzed against phosphate-buffered saline and was stored at a concentration of 6.7 mg/ml. The use of monoclonal 1D9 has been reported previously, but its specificity and sensitivity are characterized for the first time below.
It is likely that 1D9 binds to a linear epitope found in Arf proteins, as it binds to proteins on nitrocellulose and immunoprecipitates them from detergent-containing solutions after denaturation of proteins in boiling SDS. The epitope recognized by 1D9 was investigated in two ways. A series of 88 peptide octamers (mimetopes) were synthesized by Chiron Mimetopes (Emeryville, CA), starting with the N-terminal octapeptide and moving two residues toward the C terminus to the end. ELISA assay of these peptides with 1D9 identified a number of potential epitopes. However, reactivity with secondary antibodies and difficulties with regenerating the peptides for consecutive assays limited our confidence in the results of this approach. However, one of the few potential epitopes identified in this manner included the only residue that was found by inspection to be absolutely conserved in Arf proteins 1, 3, 5, and 6 and different in (the less reactive) Arf4; specifically lysine 73 in the sequence Gln-Asp-Lys-Ile-Arg is Gln-Asp-Arg-Ile-Arg in Arf4 (see Fig. 1). To test the hypothesis that this lysine is a part of the epitope of 1D9, this residue was mutated to lysine in the Arf4 sequence. When recombinant [R73K]Arf4 was purified from bacteria the mutant protein was found to have increased (over that of wild type Arf4) immunoreactivity with 1D9 to a level comparable to that of Arf1. Thus, we conclude that this lysine is a critical part of the epitope for 1D9 and likely explains the lower sensitivity to Arf4 (see below). This region is highly divergent in all other GTP-binding proteins examined, including the Arf-like (Arl) proteins, and explains, at least in part, the specificity of 1D9 for Arf proteins. However, an 11-residue peptide, derived from the sequence of Arf1 with the key lysine in the middle, failed to compete for 1D9 immunoreactivity in immunoblots.
Arf Proteins-Recombinant Arf proteins were purified from BL21(DE3) cells as described previously for Arf1 (32,33). The recombinant proteins do not have identical electrophoretic mobilities with the same isoforms in cells or tissues. This is due, at least in part, to the lack of N-terminal myristoylation of the recombinant proteins. Thus, electrophoretic mobility alone is not a reliable indicator of Arf isoforms.
Normal rat kidney cells, stably transfected with plasmids directing the inducible expression of each of the human Arf proteins, were obtained and grown as described previously (9). Cells were lysed by the addition of Laemmli's sample buffer either before or after induction with 1,000 units/ml interferon for 18 h. Total cellular proteins were resolved in 10 -20% gradient polyacrylamide gels, and immunoblotting was performed as described below. Protein concentrations were determined by the method of Schaffner and Weissman (34).
Human Tissue Proteins-Human tissue protein samples were purchased from Clontech Laboratories and were supplied as solutions of SDS-solubilized proteins in Laemmli's sample buffer. Each sample was prepared from tissue obtained from a single individual; thus, variations among the human population were not addressed.
Immunoblotting-Protein samples (15 g of tissue lysates) were separated on gradient 10 -20% SDS-polyacrylamide minigels (Daiichi Pure Chemicals Co., Ltd., Tokyo, Japan) before transfer to nitrocellulose membranes. The membranes were blocked in Blotto (50 mM Tris, pH 8.0, 5% nonfat dry milk, 80 mM NaCl, 2 mM CaCl 2 , 0.2% Nonidet P-40, and 0.02% sodium azide) for 1 h at room temperature. Primary antibodies were incubated with filters at 4°C overnight, or at room temperature for at least 2 h. The dilutions of rabbit polyclonal antisera used varied between 1/500 to 1/2,000, depending on sensitivities (see Table I). Incubations in secondary antibody, donkey anti-rabbit IgG conjugated to horseradish peroxidase (Amersham Corp.), were performed at room temperature for 1-2 h. Signal was detected using Amersham's ECL Western blotting detection reagents according to manufacturer's instructions. Most of the polyclonal antibodies interacted nonspecifically with proteins in the tissue extract other than Arf. Each of the isoform-specific antibodies were analyzed in pairs, one membrane incubated with the diluted antiserum, and the other with diluted antiserum after preincubation with 10 g/ml immunizing peptide. Only signals that were present on the first blot and absent on the second were considered specific. With only one or two minor exceptions, specific immunostaining was observed only in the 20-kDa region of the gels.
Slot Blots-Purified recombinant Arf proteins were diluted in PBS, deposited onto a nitrocellulose membrane held in a Schleicher & Schuell slot-blot manifold, and allowed to bind at room temperature for 1 h before being washed with PBS (31). The membranes were either immediately incubated with antibodies, as described above, or air-dried and stored at room temperature for later use.
Cell Fractionation-CHO cells were fractionated by sucrose density gradient centrifugation and free flow electrophoresis (FFE), a method that permits an effective one-step separation of endosomes from Golgi, endoplasmic reticulum, and plasma membranes as described previously (27,(35)(36)(37). Endosomes were first labeled by incubating the cells for 10 min in medium containing 2 mg/ml horseradish peroxidase (HRP; Sigma Type IV; Sigma). After washing in the cold, the cells were disrupted in TEAS (10 mM triethanolamine, 1 mM EDTA, 250 mM sucrose, pH 7.4) using a ball bearing homogenizer, and membranes were concentrated by flotation on a discontinuous sucrose density gradient. The membranes (1 mg/ml) were treated briefly with a low concentration of trypsin (1-2 g/mg of membrane protein, 4 min, 37°C), quenched with excess soybean trypsin inhibitor, and subjected to FFE. In control experiments, the trypsin treatment was found not to affect subsequent binding of Arf proteins; similarly, recovery of all endogenous markers (including Arf proteins) was quantitative throughout the procedure. Individual fractions were assayed for organelle-specific markers using both enzymatic assays and immunoblots.
Total CHO membranes and cytosol were prepared by centrifugation of a post-nuclear supernatant (250 ϫ g supernatant) at 150,000 ϫ g for 60 min. Pellets (membranes) and supernatants (cytosol) were collected separately and adjusted to equal volumes in TEAS. Triton X-100 was added to 0.1%, and proteins were precipitated using methanol and chloroform. Precipitates were resuspended in Laemmli's sample buffer and prepared for immunoblotting as described above.
Binding of Cytosolic Arf Proteins to Membranes-The binding of cytosolic Arf proteins to isolated membrane fractions was determined using a slight modification (27) of previously published methods (4). Briefly, membranes (30 g of membrane protein) were incubated with cytosol for 15 min in the presence of an ATP-regenerating system in the presence or absence of 200 M BFA and/or 25 M GTP␥S. Membranes were pelleted for 10 min at 14,000 ϫ g in a TLA 100.2 rotor, resuspended in Laemmli's sample buffer, and resolved by electrophoresis on 10 -20% SDS-PAGE gradient gels. After transfer to nitrocellulose, immunoblotting was performed using antibodies to Arf proteins and the coatomer component ␤-COP (38). Cytosol from CHO cells was prepared in 1 mM ATP as described (39).
Stripping of Arf6 and ␣-Adaptin from Isolated Membranes-CHO membranes (Ϸ25 g; prepared by flotation on sucrose density gradients) were incubated (in the presence of 0.2 M sucrose) in 0.5 M Tris (pH 7.0), 0.1 M sodium carbonate (pH 11), or 1 M NaCl for 30 min on ice. The membranes were then collected by centrifugation onto a 1 M sucrose cushion using a TLA 100.2 rotor as above. The cushion and interface were collected, brought to 0.2 M sucrose, and subjected to another round of stripping. Membranes were then pelleted in a microcentrifuge, and the pellets were subjected to SDS-PAGE. Proteins were transferred to nitrocellulose and immunoblotting performed using antibodies to Arf6 or to ␣-adaptin.
Coated Vesicle Preparation-Clathrin-coated vesicles used in these studies were prepared from cow brains by a modification of the method of Campbell et al. (40). Bovine brain and rabbit liver clathrin-coated vesicles were also kindly provided by Dr. Linton Traub, Washington University, St. Louis, MO.

Expression of Arf Isoforms in Tissues Examined with Arf
Isoform-specific Antisera-The subcellular distribution of each Arf isoform was determined using a battery of isoform-specific polyclonal antisera, raised against peptides derived from each protein. These antisera are described under "Materials and Methods" and shown in Table I and Fig. 1. In addition, a monoclonal antibody, termed 1D9, was produced and found to recognize all human Arf proteins, although its sensitivity toward Arf4 was relatively poor (Fig. 2).
Antisera R-1026 (␣Arf1), R-1023 (␣Arf3), and R-1471 (␣Arf6) had similar sensitivities, capable of detecting Ն5 ng of protein in a lane of a Western blot. R-891 (␣Arf4) and R-1525 (␣Arf5) were 2-5 fold more sensitive than the other sera. 1D9 had a lower limit of detection of Ͻ1 ng for Arf proteins 1, 3, 5, and 6 but about 10 ng for Arf4 (see Table I). The specificity and sensitivities of these antibodies make them generally useful for determining the expression of human Arf proteins in total cell or tissue homogenates.
These reagents were used to examine Arf expression in eight human tissues by immunoblotting, as described under "Materials and Methods." Results of this screen revealed a poor correlation between the levels of expression of Arf proteins and their messages (data not shown). In addition, each Arf protein appears to have a unique pattern of expression in these tissues (Fig. 3). Arf1 and Arf3 are usually expressed to the highest levels (typically 0.03-0.1% total cell protein). The greater sensitivity of the anti-Arf4 antibody can give the mistaken impression that Arf4 is expressed to similar levels (see Fig. 3). However, quantitative immunoblotting of many cell and tissue samples confirm that Arf4 -6 are usually expressed at levels 3-10-fold below those of Arf1 and Arf3. This finding likely explains the observation that Arf1 and Arf3 have been consistently identified in purified Arf preparations (6,7,11,41,42). Results of Northern and Western blotting indicate that all five Arf isoforms are expressed in all cells and tissues. The levels of Arf4 -6 are similar, although the higher sensitivity of the ␣Arf4 antiserum allows this isoform to be readily observed in human tissues (Fig. 3). Arf6 expression is inexplicably and uniquely very high in the ovary (estimated Ϸ0.2% total tissue protein).
Arf6 Is Stably Associated with Membranes in CHO Cells-The availability of isoform-specific antibodies allowed us to ask whether endogenous Arf proteins could be distinguished on the basis of their intracellular localizations. CHO cells were a good source for these studies as each of the five human Arf proteins have homologs in hamster that are expressed in levels detectable with isoform-specific antisera and 1D9. However, only the Arf4 antibody, R-891, and 1D9 proved useful for indirect immunofluorescence. 2 We therefore localized each Arf protein by subcellular fractionation.
CHO cells were first separated into membrane and soluble fractions to determine the amount of each Arf that would remain membrane-bound after homogenization. As shown in Fig. 4, Arf proteins 1, 3, 4, and 5 were found predominantly (Ͼ90%) in the high speed supernatant after ultracentrifugation of CHO cell homogenates. While low levels (Ͻ10% of cytosol) of Arf proteins 1-5 can usually be found on any membrane fraction, Arf6 was found exclusively in the high speed pellet. Even  1. Alignment of the human Arf proteins and derived peptides used to generate isoform-specific antisera. The five human Arf protein sequences are shown aligned, with identities marked in the line below by an asterisk (*) and homologies by a dot (⅐). The five proteins are 60% identical and 84% homologous. The peptides synthesized are indicated by bold type in the sequences and the name of the peptide is indicated on the line below. The underlined residue (aligning with lysine 73 in Arf1) is the residue included in the epitope for monoclonal 1D9, as described under "Materials and Methods." after long exposure times, Arf6 was not detectable in the soluble, cytosol fraction. Thus, Arf6 was distinguished from all other Arf proteins by its membrane association after cell homogenization. It should be noted that brain appears to be different from other tissues in yielding a high percentage of particulate Arf after homogenization (Ϸ50% total Arf), which facilitated its purification from this tissue in large amounts (43).
Given that recombinant Arf1 has previously been shown to bind to membranes or phospholipid vesicles upon activation, or binding of GTP (3, 4, 44 -46), we next determined whether each of the soluble Arf proteins could be recruited onto membranes in vitro. Crude, Golgi-enriched membranes were prepared by sucrose density gradient centrifugation and incubated at 37°C with CHO cytosol in the presence or absence of GTP␥S. Membranes were then washed and Arf proteins and ␤-COP (a component and marker for the coatomer or COP-I) detected by immunoblot, using the Arf isoform-specific or ␤-COP antibodies. The binding of each of the cytosolic Arf proteins and ␤-COP to membranes was markedly stimulated by GTP␥S (Fig. 5, lane  5). The amount of Arf6 on membranes was unchanged by incubation at 37°C with GTP␥S, but there was no detectable cytosolic Arf6 to be recruited. Addition of BFA (200 M) to the reaction prevented the binding of Arf proteins 1, 3, 4, or 5 and ␤-COP to the membranes, but had no effect on Arf6 (Fig. 5, lane  4). Thus, the behaviors of Arf proteins 1, 3, 4, and 5 were similar in that each bound to membranes upon activation by GTP␥S, and membrane binding was sensitive to BFA. Each of these characteristics again distinguished Arf1, 3, 4, and 5 from Arf6. These results suggest that if Arf6 is binding GTP␥S under these conditions that this isoform is incapable of recruiting ␤-COP to the membrane, as has been reported previously for Arf1 (4).
BFA is thought to prevent association of soluble Arf proteins with membranes by interfering with GDP-GTP exchange (17)(18)(19). Thus, the failure of BFA to dissociate prebound Arf6 does not necessarily mean that Arf6 binding is BFA-resistant. Assuming that Arf6, like other Arf proteins, dissociates from membranes upon GTP hydrolysis during its normal activity cycle, we asked whether incubation of intact cells with BFA would result in the translocation of Arf6 from a membranebound to a soluble pool. CHO cells were treated with BFA for 90 min, conditions that resulted in the complete dissociation of COP-I proteins from membranes and the subsequent tubulation of the Golgi complex and endosomes (47,48). Upon homogenization and separation of membrane and soluble fractions, however, Arf6 was found to remain membrane-associated (data not shown). Thus, either Arf6 does not cycle on and off intracellular membranes like the other Arf proteins, or the reassociation of any cytosolic Arf6 generated in intact cells is insensitive to BFA. In any event, Arf6 is distinct from all other Arf proteins in either or both of these features.
Endogenous Arf6 Is Localized Exclusively to the Plasma Membrane-We took advantage of the stable association of Arf6 with membranes to determine its subcellular localization by FFE, a technique that resolves intracellular membranes on the basis of their net surface charge. For example, endosomes migrate toward the anode, well removed from the plasma mem-  Table I. FIG. 4. Arf6 fractionates with the membranes, while other Arf proteins remain cytosolic. CHO cells were fractionated into crude membrane and cytosol fractions and were tested for immunoreactivity with the isoform-specific antisera, as described under "Materials and Methods." In the top panel, R-1023 was used without blocking peptide to allow reactivity with both Arf1 and Arf3.

FIG. 2. Specificity of the Arf antibodies is illustrated in slot blots of purified recombinant Arf proteins with isoform-specific and monoclonal anti-Arf antibodies.
The five human Arf proteins used are shown on the top in a Coomassie Blue-stained SDS-polyacrylamide gel loaded with 0.3 g/lane. Each antibody was tested for immunoreactivity against each of the different Arf proteins in slot blots, as described under "Materials and Methods." Antisera were diluted as indicated in Table I. Note that the cross-reactivity of the Arf4 antiserum (R-891) with Arf5 is nonspecific as it is not competed with the immunizing peptide. brane, Golgi, and ER membranes, all of which are deflected less toward the anode. Since transfected Arf6 has recently been suggested to mediate the assembly of a novel coat involved in receptor recycling (1,2), it was of particular interest to determine whether any endogenous Arf6 is associated with endosomal membranes. To provide a marker for endosomes, CHO cells were first incubated with HRP for 10 min at 37°C, homogenized, and then fractionated by flotation on a discontinuous sucrose density gradient to generate an enriched endosome-Golgi fraction, as described above. This material was then subjected to FFE, and the resulting fractions were analyzed for markers of different intracellular compartments as well as for Arf6 proteins (by immunoblot). As shown in Fig. 6A, Arf6 was detected in a single peak associated with the least anodally deflected membranes, largely containing ER and plasma membranes. No Arf6 was found co-migrating with endosomes or Golgi membranes, even after long exposures of the chemiluminescence detection system used to visualize the immunoblots. This finding was surprising given the abundant presence of overexpressed wild type and mutant Arf6 on intracellular membranes, in addition to the plasma membrane (2). Although all of the Arf6 injected into the FFE chamber could be recovered in fractions containing plasma membrane and ER markers, no Arf6 was detected on endosomal membranes.
As mentioned above, the region of the FFE profile where Arf6 was found contained the bulk of the plasma membrane and ER membranes. Fig. 6B shows an expanded view of this region of the profile. Although the patterns for each marker overlap, the Arf6 immunoreactivity clearly co-migrated with the plasma membrane marker (␣-adaptin) and not with markers for the ER (calnexin; data not shown) or the ER-Golgi intermediate compartment (ERGIC58). Fig. 6C shows a typical immunoblot comparison of the FFE profiles of ERGIC58, ␣-adaptin, and Arf6 that was used for quantitation, shown in panel B. Identical results were obtained when a total membrane fraction was used for FFE rather than a membrane fraction enriched by sucrose gradient centrifugation, demonstrating that there was not a specific loss of other Arf6-containing membranes during fractionation. Moreover, the brief trypsin treatment (1-2 g/mg of membrane protein for 4 min at 37°C, see "Materials and Methods") used prior to FFE did not strip any Arf6 from CHO membranes. We conclude that Arf6 was uniquely distributed on the plasma membrane of CHO cells and was not detected on endosomes or other intracellular membranes.
Specificity of Arf Recruitment to Organelle Membranes in Vitro-It was also of interest to determine whether the soluble pool of Arf proteins (Arf proteins 1, 3, 4, and 5) were selectively or non-selectively recruited to the different membrane fractions. This was accomplished by collecting FFE fractions into three pools, with pool I enriched in plasma membrane (roughly equivalent to fractions 40 -42 in Fig. 6A), pool II enriched in ER and Golgi membranes (comparable to fractions [43][44][45][46][47][48][49], and pool III enriched in endosomes (comparable to fractions 55-61). Equal amounts of membrane protein from each pool were incubated together with CHO cytosol in the presence or absence of GTP␥S, and Arf proteins analyzed by immunoblot. All soluble Arf proteins could be recruited to pools II and III, and this recruitment was greatly enhanced by the addition of GTP␥S, as shown above for crude membranes. However, the four soluble Arf species could not be recruited, even in the presence of GTP␥S, to plasma membranes (pool I; see Fig. 7). Although Arf proteins 1, 3, 4, and 5 could be effectively recruited to ER, Golgi, and endosome membranes (Fig. 7) (27), they exhibited striking specificity in their inability to bind to plasma membranes in vitro.
The stable association of Arf6 with the plasma membrane is in marked contrast to the behavior of all other Arf proteins in these and previous studies. We compared the elution of Arf6 to the elution of ␣-adaptin, a subunit of AP-2 clathrin adaptor complex, because like Arf6 the association of AP-2 is restricted to the plasma membrane and is BFA-resistant (24). We therefore sought conditions that could release Arf6 from membranes in vitro. Interestingly, Arf6 was found to be eluted in a fashion that reflected the stripping of ␣-adaptin-containing AP-2 adaptor complexes from the plasma membrane. As shown in Fig. 8, treatment with 1 M NaCl, 0.5 M Tris, or 0.1 M sodium carbonate variably released Arf6 and ␣-adaptin from membranes to very similar extents. These results are in marked contrast to the binding of ␥-adaptin containing HA-1 adaptor complexes to the TGN, an event that is BFA-sensitive much like the binding of Arf1, which can serve to recruit HA-1 adaptors to Golgi membranes in vitro (5,25).
Arf6 Is Not Found on Clathrin-coated Vesicles-The finding that Arf6 binding to the plasma membrane was at least superficially similar to the binding of the AP-2 adaptor complexes suggested that Arf6 may play a role in the assembly of clathrincoated pits. Arf6 might then be expected to be internalized in coated vesicles, perhaps resulting in the delivery of a small fraction to endosomes. It was therefore important to determine whether Arf6 was present on purified clathrin-coated vesicles. Bovine brain or rabbit liver coated vesicles were analyzed by immunoblotting using the pan-reactive monoclonal antibody 1D9 (Fig. 9, lanes 1-4) as well as each of the isoform-specific Arf antibodies. As reported previously (14) and shown in Fig. 9, 1D9 detected a single band of immunoreactivity in the coated vesicle preparations (lanes 3 and 4). However, none of the Arf isoform-specific antibodies, including that for Arf6 (lanes 6 and  7), reacted with the clathrin-coated vesicle proteins, even after overexposure of the film (Fig. 9, lane 7). The nature of the cross-reactive species is under investigation and is noted to be more abundant (Ϸ5-fold) on liver than brain coated vesicles. Although these findings do not eliminate the possibility that Arf6 is involved in clathrin assembly, they indicate that Arf6 is not a component of clathrin-coated vesicles. Arf6 is thus unlikely to be efficiently internalized during endocytosis but rather to remain stably associated with the plasma membrane. DISCUSSION The very complexity of the Arf family of regulatory proteins suggests that Arf proteins may exhibit a corresponding diversity in function. At the same time, the extraordinarily high degree of structural conservation (e.g. three isoforms Ͼ96% identical) has made it difficult to examine this issue in any detail. Although the Arf isoforms were found to be ubiquitously expressed, the development of isoform-specific antibodies has allowed us to demonstrate that the different Arf proteins do not have identical intracellular distributions. Unlike all other family members, Arf6 was found to be stably membrane-bound and, moreover, to be restricted to the plasma membrane of CHO cells. Other Arf proteins could not even be recruited to CHO plasma membranes in vitro, although their binding to Golgi, ER, and endosome membranes could be stimulated by GTP␥S. Arf6 was also distinguished by its insensitivity to both GTP␥S and BFA. These features suggest that Arf6 does not undergo repeated transitions between membrane-bound and soluble states during its activity cycle, a property that had previously been thought to be intimately related to the mechanism of action of all Arf proteins.
Although two recent publications also reported that overexpressed, transfected Arf6 was associated largely with the plasma membrane, a significant amount of both the wild type and mutant forms of the protein were also detected on internal membranes, presumably endosomes (1,2). In contrast, we did not detect endogenous Arf6 on highly purified endosomes, nor could we find Arf6 in cytosol or on clathrin-coated vesicles. Thus, it seems unlikely that Arf6 normally shuttles between endosomes and the plasma membrane either by internalization via coated vesicles or via a soluble, cytosolic intermediate. The appearance of Arf6 on internal structures (1, 2) may be a consequence of the degree of overexpression in these experimental systems, estimated at several hundredfold over endogenous Arf6.
Purification of Arf proteins has been achieved by a number of investigators using a variety of biochemical assays (6,7,11,49,FIG. 7. Arf6 is associated with plasma membrane, and other Arf proteins are not recruited to plasma membrane by GTP␥S. CHO cells were fractionated using FFE and pooled into three fractions containing mostly plasma membrane (PMϾER), mostly ER and Golgi membranes (ER/Golgi), and mostly endosomes (endosomes). Membranes were then incubated with cytosol in the presence or absence of GTP␥S and analyzed as in Fig. 7. Arf1, 3, and 4 immunoreactivity was qualitatively the same as that shown in the top panel for Arf5.  50). 4 In at least five of these instances, only Arf1 and/or Arf3 were identified by microsequencing of the final preparations. A simple explanation for this finding emerges from the analysis of Arf proteins in human tissues; Arf1 and Arf3 are usually expressed to higher levels than the other three isoforms. Indeed, the levels of Arf4 -6 are often only about 10% those of Arf1 and/or Arf3. The availability of isoform-specific antibodies provides a sensitive method for the detection of Arf isoforms in purified protein preparations or crude tissue or cell lysates and should allow for further tests of isoform specificity.
Membrane Association and Intracellular Distribution of Arf Proteins-After lysis of CHO cells by homogenization, Arf1, 3, 4, and 5 were found primarily in cytosol. Previous estimates, based on Arf activity, revealed the percentage of Arf that was membrane-bound varied from Ϸ50% in bovine brain to Ͻ10% in most other tissues (52). This picture appears at odds with images obtained by indirect immunofluorescence of cultured cells stained with Arf antibodies, which reveal intense staining of the Golgi region as well as some poorly defined punctate staining in the cytosol (9,54,55). Several potential explanations are possible, including inefficient fixation of soluble Arf proteins and redistribution of Arf proteins from membranes to cytosol upon cell lysis. The latter is known to occur in the case of ␤-COP, a component of the COP-I-containing coatomer complex whose membrane binding is Arf-dependent (4,55). Thus, the percentage of the cytosolic Arf proteins that are membranebound in live cells is not known with any precision.
Incubation of cytosol, a ready source of the soluble Arf proteins, with different membranes and GTP␥S resulted in the recruitment of Arf proteins 1-5 to purified fractions enriched in ER, Golgi, or endosome membranes. Indeed, it has recently been reported that coatomer proteins can also be recruited to each of these fractions (27). Although activated Arf1 has been shown to bind phospholipid vesicles in the absence of other proteins, it is unlikely that the membrane association observed here entirely reflected simple lipid partitioning because no Arf recruitment to enriched plasma membranes was observed, even in the presence of GTP␥S. Some membrane specificity in Arf recruitment was apparent. This specificity may, however, reflect the different lipid content of the plasma membrane versus intracellular membranes. That the plasma membrane was capable of harboring Arf proteins was amply demonstrated by the stable association with Arf6. The insensitivity to GTP␥S and BFA of the Arf6 binding to plasmalemma indicates a distinct mechanism of membrane association is likely.
Prevention of Arf activation, and the consequent failure to bind membranes, result from exposure of cells or membranes to BFA. Sensitivity to BFA was also a property shared by the four cytosolic Arf proteins: 1, 3, 4, and 5. This is in contrast to the previous conclusion that the association of recombinant Arf5 with membranes is insensitive to BFA (56). These different conclusions may reflect the use of relatively high concentration of GTP␥S used in the previous study. High concentrations of GTP␥S have been observed to overcome the inhibition of Arf activation and membrane binding by BFA. 5,6 Membrane Association and Possible Functions of Arf6 -It is unclear why the association of Arf6 with the plasma membrane differs so radically in its sensitivity to BFA from all other Arf proteins. Each member of the family is presumably myristoylated in vivo at the conserved glycine residue at position 2, although this has only been directly demonstrated for Arf1 and Arf3, purified from mammalian tissues (53). As myristoylation is required for both membrane association and Arf1 function and Arf6 is a substrate for N-myristoyl transferase when coexpressed in Escherichia coli (51), we presume that N-myristoylation also plays a role in the binding of Arf6 to membranes. Based on biochemical studies and the three dimensional structure of Arf1, the myristic acid forms a critical part of the phospholipid and GTP-sensitive switch region (23,57). The role for N-myristoylation of Arf6 in its interaction with the plasma membrane and the explanation(s) for the stable, GTP-and BFA-insensitive, association of Arf6 with plasma membranes must now be sought.
One hint concerning the nature of Arf6:plasma membrane interaction also provides a provocative suggestion concerning a possible function of Arf6. Incubation of membranes with high salt, chaotropic agents, or alkaline pH was equally effective at releasing Arf6 from membranes as ␣-adaptin, a component of the clathrin adaptor complex HA-2. The binding of HA-2 to the plasma membrane as well as to the membrane of coated vesicles has previously been shown to be insensitive to BFA and to GTP␥S (24), establishing HA-2 as the only membrane-associated coat protein complex whose binding is not obviously Arfdependent. Binding of the closely related HA-1 adaptor complex to the TGN is BFA-sensitive and Arf-dependent in vitro and in intact cells (5). Even the recruitment of COP-II coats to the ER of yeast cells involves the Arf-like protein Sar1 (29). Taken together, these considerations suggest that an Arf protein may function in the recruitment of HA-2 adaptors to the plasma membrane, regulating the formation of clathrin-coated pits, and the residence of Arf6 at this locale makes it a likely candidate. The BFA and GTP␥S insensitivity of HA-2 binding could then be understood because its affiliated Arf protein is already stably membrane-bound in a BFA-insensitive fashion. Arf6 might still engage in a cycle of GTP binding and hydrolysis concomitant with coated pit assembly, but nucleotide hydrolysis would not result in Arf6 dissociation from the membrane.
The findings that Arf6 was absent from preparations of clathrin-coated vesicles (see Fig. 9), and is neither preferentially included within nor excluded from plasma membranecoated pits (2), are not necessarily inconsistent with a role for Arf6 in HA-2 binding. For example, the association of Arf protein with COP-I-containing vesicles is reversible with GTP hydrolysis and allows the Arf to recycle. Arf1 need not interact directly with COP-I components and may not be an integral component of the coat itself. Although a role for Arf6 in clathrin-coated vesicle formation remains speculative, paradigms derived for Arf proteins that dissociate from membranes during their activity cycle cannot be precisely applicable to understanding the mechanism of Arf6 action. Thus, it is possible that Arf6 acts catalytically to recruit HA-2 to coated pits and is rapidly inactivated, diffusing away from the coated pit but remaining attached to the plasma membrane. Alternatively, it is also possible that Arf6 is not involved in regulating recruitment of HA2 to coated pits.
Recent work in which wild type or mutant forms of Arf6 were overexpressed in fibroblasts implicates Arf6 in one or more roles in receptor-mediated endocytosis (1,2). This conclusion was based in part on the observations that expression of wild type or a presumptive GTP hydrolysis mutant (Q67L) of Arf6 decreased the initial rate of transferrin receptor endocytosis as well as recycling. Conceivably, overexpression of these proteins interfered with coated pit function or the recruitment of receptors to otherwise normal coated pits. In contrast, a mutant presumably deficient in nucleotide binding (T27N) seemed to have little effect on endocytosis but did result in a redistribution of receptors from the plasma membrane to internal mem-branes. The nature or origin of these internal membranes remains unknown. It is difficult to interpret these results precisely, in part due to the known potential for toxic effects of Arf overexpression (20), the magnitude of the excess Arf6 produced, and to the possible interference of normal Arf function by the introduction of the epitope tags. Based on the structure of Arf1, such a tag is predicted to lie very close to and could interfere with the essential GTP-and lipid-sensitive N-terminal switch (23). Indeed, a similar epitope tag, added to the C terminus of S. cerevisiae Arf1, caused a failure to fully complement an arf1 Ϫ mutant. 3 The apparent absence of endogenous Arf6 from endosomes makes it difficult to understand how it could directly regulate endosome function. Conceivably, the relocalization of the Arf6 T27N mutant may reflect the irreversible clearance of a mutant and possibly inactive Arf6 or Arf6-containing protein complex from the plasma membrane. Endosomes are capable of recruiting cytosolic proteins, including at least some COP-I components, clathrin, and Arf proteins (27,52). The fact that this recruitment is GTP␥S and BFA-sensitive makes it unlikely to be an event that is under the control of Arf6.
Further complicating interpretation of these, or any other experiments involving Arf6, is the fact that its properties with respect to membrane association are so different from the other Arf proteins. As reversible association of Arf proteins with membranes is such an integral part of their presumptive mechanism of action, it is difficult to predict how tightly Arf6 will adhere to the paradigm developed from studies of (mostly) Arf1. Nevertheless, these differences provide a unique opportunity to further analyze the functions and mechanisms of Arf protein action. They also clearly illustrate that not all of the multiple Arf proteins expressed in animal cells are completely interchangeable. We suspect that as the Arf family is studied in greater detail, other such differences will become evident.