Effects of Brefeldin A and Accessory Proteins on Association of ADP-ribosylation Factors 1, 3, and 5 with Golgi*

ADP-ribosylation factors (ARFs) are -20-kDa guanine nucleotide-binding proteins initially identified by their ability to enhance in vitro cholera toxin-cata-lyzed ADP-ribosylation and subsequently shown to participate in vesicular transport in the Golgi and other cellular compartments. By cDNA and genomic cloning, at least six mammalian ARFs were identified. Brefeldin A (BFA) disrupts Golgi membranes and inhibits binding of soluble high molecular weight proteins to Golgi fractions. We examined the effects of BFA on binding of ARF1, -3, and -5 to a Golgi fraction in the presence of an ATP-regenerating system and a fraction of soluble, high molecular weight, accessory proteins (SAP), presumably containing complexes identified by others as coatomers that are involved in vesicular transport. ARF binding in all instances was dependent on guanosine 5’-0-(3-thiotriphosphate) and increased by the ATP-regenerating system. Binding of ARFl and -3, but not ARF5, was enhanced by SAP. BFA inhibited the SAP-dependent, but not the SAP-independent, binding of ARFl and -3. It had no effect on the increment in binding produced by an ATP-regenerating system. B36, an inactive derivative of BFA, did not inhibit SAP-dependent binding of ARFl and -3. Binding of ARF5, which was SAP-independent, was not affected by BFA. These observations are consistent

believed to participate in vesicular transport within the Golgi and other membrane compartments (2)(3)(4)(5)(6)(7). As components of the Golgi transport system, ARFs act in concert with a high molecular weight soluble complex, termed coatomer, to promote vesicle transport involving Golgi, a process believed to involve also a heterotrimeric G protein (5, €49).
At least six mammalian ARF members of a multigene family have been identified by cDNA and genomic cloning (10-14). Based on size, deduced amino acid sequence, phylogenetic analysis, and gene structure, they can be grouped into three classes (10-17). ARF1, -2, and -3 form Class I, ARF4 and -5 Class 11, and ARF6 Class 111 (14). ARFs from all three classes activate cholera toxin-catalyzed ADP-ribosylation (18,19). The stimulatory effects of certain phospholipids and detergents on this reaction are ARF-specific. Recombinant ARF2 (rARF2) was almost absolutely dependent on dimyristoylphosphatidylcholine plus cholate for activity, whereas rARF6 exhibited substantial activity in their absence (19).
ARFs are, at least in part, co-translationally modified by N-terminal myristoylation (20, 21). Activation of toxin by ARFs did not require myristoylation (18,19), although it has been proposed that the N-terminal amino acid sequence may be necessary for this action (22). In addition to being activated by phospholipids, purified native and recombinant ARFs synthesized in Escherichia coli bound to phospholipids or Golgi membranes in the presence of nonhydrolyzable GTP derivatives (e.g. GTP+) but not GDP (23-27). In contrast to toxin activation, binding of ARF to Golgi membranes did require N-terminal myristoylation (25).
ARFs have been localized by immunoreactivity to Golgi membranes in tissue sections (28), although most ARF activity in homogenates appears to be soluble (20). Recent reports have suggested that ARFs may be involved in a variety of cellular processes in addition to Golgi transport, including endocytosis and nuclear membrane assembly (3, 4). Since several mammalian ARFs have been described, it seemed reasonable that they should have different functions and, likely, different intracellular localizations. Consistent with this hypothesis, ARF1, -3, and -5 clearly differed in their binding to a Golgi fraction (27). Since association of ARF with Golgi involves other soluble proteins (5, 9) and since Golgi membranes are disrupted by BFA (29, 30), the studies reported here were initiated to determine whether BFA and soluble accessory proteins (SAP) might affect differently the association of ARF1, -3, and -5 with Golgi fraction membranes.

Materials
B36, a BFA derivative, was a gift from Dr. Julie G . Donaldson, National Institute of Child Health and Human Development, NIH, Bethesda, MD. BFA was purchased from Epicentre Technologies.
Sources of other materials are noted in earlier publications from this laboratory.

Methods
Preparation of Golgi Fraction, ARF Fraction, and SAP from Rat Brain Homogenutes-As described in detail previously (27, 31), fresh rat brain was homogenized (Teflon Dounce homogenizer) in 2 volumes (w/v) of buffer A (0.25 M sucrose, 1 mM MgCI,, 1 mM dithiothreitol, 0.5 mM phenylmethylsulfonyl fluoride, soybean and lima bean trypsin inhibitors, leupeptin, and aprotinin, each 1 pglml, 10 mM Hepes buffer pH 7.5). The homogenate was diluted with another 2 volumes per weight of buffer A and centrifuged (850 X g, 10 min, Sorvall SS-34). The supernatant was further centrifuged (35 min, 175,000 X g, SW 41, 37,000 rpm), and the pellet was dispersed in buffer A with 1.4 M instead of 0.25 M sucrose (volume equal to original tissue weight). A 4-ml sample of the suspension was transferred to the bottom of a centrifuge tube which contained layers (each 2 ml) of 0.25,0.6,0.85, and 1.15 M sucrose in buffer A without sucrose. After centrifugation (100 min, 175,000 X g), the fraction at the interface of 0.85 and 1.15 M sucrose, which contained the highest specific activity of galactosyltransferase, a Golgi resident enzyme (32), was collected and stored in small portions at -70 "C.
To prepare partially purified rat brain ARFs, the supernatant after centrifugation at 175,000 X g for 35 min was concentrated (YM 10 membrane, Amicon) to half its original volume. An 18-ml sample (-270 mg of protein) was applied to a column (2 X 89 cm, 279 ml) of Ultrogel AcA 54 equilibrated and eluted with buffer B (0.25 M sucrose, 100 mM NaCl, 1 mM NaN,, 5 mM MgCI,, 1 mM EDTA, 2 mM dithiothreitol, 20 mM Tris, pH 8.0). Fractions (3 ml) were collected and assayed for ARF activity based on stimulation of CTA ADPribosyltransferase activity (33). Fractions containing ARF activity were pooled, concentrated (Amicon Centriprep lo), and stored in small portions at -70 "C. Based on ARF activity, mobility on SDS-PAGE, and immunoreactivity with rARF5 polyclonal antibodies, a native ARF5 was partially purified from an enriched AcA 54 column fraction of rat brain by sequential chromatography on DEAE-Sephacel and CM-Sepharose. From the DEAE-Sephacel column, ARF5 was eluted with 107-116 mM NaCl and myristoylated rARF5 (mrARF5) with 90-110 mM NaCl. ARF5 active fractions, which were separated from most of ARFl and -3 (based on reaction with sARF I1 and rARF5 antibodies), were titrated to pH 5.3; the supernatant fraction after centrifugation at 10,000 X g for 30 min was applied to a CM-Sepharose column. ARF5 did not bind and emerged in the void as did mrARF5. After SDS-PAGE, the ARF5 band stained poorly with Coomassie Blue but well with silver stain. Based on silver staining ARF5 represented almost 10% of total protein.
Interaction of ARFs with Golgi Membranes-For most experiments, Golgi membranes (6.25 pg of protein) were incubated with ARF1, ARF3, mrARF5, or crude partially purified rat brain ARFs, 0.2 mM GTPyS or GDPPS, SAP, BSA, and/or ATP-regenerating system (1 mM ATP, 5 mM creatinephosphate, 1 unit of creatine phosphokinase) as indicated in a total volume of 100 p1 containing 60-70 mM NaCI, 4.4 m M MgC12, 0.8 mM EDTA, 10 mM DTT, 0.25 M sucrose, 15 mM Tris (pH 8.0), and protease inhibitors. After 40 min at 37 "C (or as indicated), samples were centrifuged (30 min, 14,000 X g, Eppendorf microcentrifuge) and pellets were rinsed and suspended in buffer A (27). Samples of each suspension were assayed for stimulation of CTA ADP-ribosyltransferase activity. Data are presented as total ARF activity associated with the Golgi pellets. ADP-ribosyltransferase activity of CTA was assayed as described (27, 34) except that Cibacron Blue concentration was 20 p~. All data reported are means of duplicate assays. All experiments have been replicated more than twice.
Zmmumdetection of ARFl and -3 and mrARF5"Two polyclonal antibodies, one raised against sARF I1 (ARF3) and the other against recombinant ARF5, were used (27). The former is relatively specific for class I ARFs (ARF1, -2, and -3), and the latter reacts well with class 11, but not class I, ARFs. ARFl and -3 differ in electrophoretic mobility. ARF5 differs from ARF4 (not detected in these experiments) as well as from ARFl and -3 in electrophoretic mobility.
After precipitation with 7.5% trichloroacetic acid, proteins were separated by SDS-PAGE in 15% gels and transferred to nitrocellulose, which was incubated overnight with a mixture of the two antibodies (1:lOOO dilution of rARF5 antibodies and IgG from anti-sARF I1 antiserum, 1 pg/ml) as described (35). The blot was then incubated with horseradish peroxidase-conjugated goat anti-rabbit IgG followed by color development with 4-chloro-1-naphthol and H,O,. After incubation of Golgi and purified ARF in the presence of GDPPS, a small amount of immunoreactive ARF was associated with the membranes, although no ARF activity was detected. This was perhaps a result of the presence of some ARF that had been denatured during the purification procedure.
Other Methods-Protein was measured by Bio-Rad assay using BSA as a standard.
For silver staining, proteins were precipitated with trichloroacetic acid, separated by SDS-PAGE in a 14% Tris/glycine gel (Novex, San Diego, CA), and stained using a Rapid-Ag-Stain (ICN). The intensity of ARF bands after silver staining or immunoreaction was measured by densitometry (Molecular Dynamics 300B computing densitometer). Readings were normalized as described in figure legends.

RESULTS AND DISCUSSION
To investigate the interaction of ARFs with Golgi, a fraction enriched in ARFs from gel filtration of rat brain supernatant was used with or without an ARF-free fraction of SAP and/or an ATP-regenerating system. In the presence of GTPrS, ARF activity associated with Golgi membranes increased rapidly for 20 min and much more slowly thereafter (Fig. 1). The rate of binding was increased by addition of an ATP-regenerating system or SAP, or both (Fig. 1). The ATPregenerating system increased binding of ARF1, -3, and -5 to Golgi membranes whether or not SAP was present, and SAP alone enhanced binding of ARFl and -3 (data not shown), but not ARF5 (25).

Effects of Brefeldin A on
Binding of ARFs to Golgi BFA (25 or 100 pg/ml) did not inhibit basal or ATPregenerating system-dependent binding of ARF activity to membranes (Fig. 2 A ) . It did, however, inhibit the SAP-dependent binding of ARF activity regardless of whether an ATP-regenerating system was present (Fig. 2 4 ) . On immunoblots and silver staining, it appeared that BFA decreased the binding of ARFl and -3 when SAP was present but had little effect on ARF5 (Fig. 2, B and C).
The effects of BFA, ATP-regenerating system, and SAP on binding of purified native or recombinant ARFs were also evaluated. With the native ARFl (Fig. 3) or ARF3 (Fig. 4) gene products purified from bovine brain cytosol, as noted with the crude ARF fraction, GTPyS, but not GDPBS, enhanced binding of ARF activity to Golgi following apparently a slight delay (Figs. 3 and 4). In the presence of GTPyS, either an ATP-regenerating system or SAP enhanced association of ARFl (Fig. 3) or ARF3 (Fig. 4) with Golgi and the former perhaps diminished the delay time.
In the presence of GTPyS and an ATP-regenerating system, the effect of increasing SAP on binding of ARFl (Fig. 5) or ARF3 (Fig. 6) to membranes was biphasic. As the enhanced ARF binding observed with smaller amounts of SAP was not reproduced by the addition of BSA, it was presumably not a nonspecific "stabilizing" effect. Without the regenerating sys-  tem, increasing SAP increased ARFl and ARF3 binding to a maximum that was not decreased with more SAP (Figs. 5 and  6). SAP alone had no effect on Golgi binding of mrARF5, but in the presence of an ATP-regenerating system, it was inhibitory (Fig. 7). Described another way, the effect of the ATPregenerating system on binding of ARFl and -3 was increased by relatively small amounts of SAP but progressively declined with larger amounts of SAP, whereas its effect on ARF5 binding was monotonically decreased by increasing SAP. The presence of the inhibitory activity (or activities) in the SAP fraction contributes to some of the variability between exper-

TABLE I Effect of brefetclin A on association of ARF1, ARF3, or mrARF5 with
Golgi membranes in the presence of AR and SAP Samples of Golgi fraction (6.25 pg), ARF1 (2.5 pg), ARF3 (2 pg) or mrARF5 (3.6 pg) with AR or SAP (99 pg), or both, and other components described under "Methods" (total volume 95 pl) were incubated at 21 "C for 10 min with or without BFA (4 pg) before addition of 5 p1 of 4 mM GTPyS. Samples were then incubated at 37 "C for 40 min before assay of ARF activity bound to Golgi. ARFl and ARF3 samples contained 20 pg of BSA and those with mrARF5, 5 @E. SamDles with or without BFA contained 1% ethanol. iments, because concentrations of SAP were not always the same (nor were amounts relative to amounts of Golgi), and different preparations of SAP may well have differed somewhat in protein compositions. The effects of BFA on binding of purified native or recombinant ARF proteins were assessed in the presence of SAP and an ATP-regenerating system (Table I, Fig. 8). BFA (4 pg) inhibited the SAP-stimulated binding of ARFl and ARF3 to Golgi membranes but did not inhibit binding in the absence of additions or in the presence of the ATP-regenerating system ( Table I). As in Fig. 7, mrARF5 binding was inhibited by SAP in the presence of the ATP-regenerating system, but BFA did not inhibit mrARF5 binding under any conditions (Table I). As shown in Fig. 8, inhibitory effects of BFA on binding of ARFl and ARF3 were concentration-dependent. B36, an inactive derivative of BFA (30), did not inhibit the binding of ARFl or ARF3 to Golgi in the presence of SAP (Table 11).
Prior studies had demonstrated an inhibitory effect of BFA on binding of a coatomer protein to Golgi (8,9). Based on an apparent requirement for ARF for coatomer binding, it was proposed that ARF associates first with the Golgi (26). In the studies reported here, binding of ARFs to Golgi was determined directly using either an enriched soluble ARF fraction or individual purified native or recombinant ARFs. As noted previously, GTPyS was essential for binding of each of the three ARFs to Golgi, and an ATP-regenerating system enhanced binding of all three (27). The individual ARFs were affected differently, however, by SAP. The binding of ARFl

Effects of brefekfin A and B36 on ARFl or ARF3 binding to
Golgi in the presence of SAP Golgi (6.25 pg), SAP (107 pg), ARFl (2.5 pg) or ARFB (2 pg), and other additions with either BFA or B36, a derivative of BFA (2 or 5 pg in 5 p1 of 10% ethanol), were incubated at 21 "C for 10 min (95 PI total volume) as in Fig. 8. After addition of 5 pI of 4 mM GTPyS, incubation was continued at 37 "C for 40 min. Pellets were assayed for ARF activitv. and ARF3 was increased, whereas the binding of ARF5 was either unaffected, or when an ATP-regenerating system was present, was inhibited. Consistent with this observation, BFA inhibited only the binding of ARFl and -3, specifically the SAP-dependent increment in binding ARFl and -3, and not ARF5. Initially, we had some concern that the lack of an effect of SAP and BFA on mrARF5 binding might reflect a defect in the recombinant proteins not detected by the cholera toxin activation assay. The observation that ARF5 partially purified from rat brain behaved similarly (data not shown) supports the conclusion that the mrARF5, for intrinsic rather than artifactual reasons, behaved differently from ARFl and ARF3 in response to BFA and SAP.
The SAP fraction was used in our experiments as a potential source of coatomers (and perhaps other proteins) that might influence ARF binding and/or vesicle formation, recognizing that its heterogeneity could result in multiple effects. It seems quite possible, for example, that the inhibition of ARF binding that occurred with increasing amounts of SAP in the presence of the ATP-regenerating system is a result of inhibition or destruction of the source of ATP. In addition, it is unclear just what ARF binding means in functional terms in this system. We believe that some, possibly large, fraction of ARF binding, although GTP-dependent, is "nonspecific," perhaps analogous to the GTP-dependent binding of ARF to phospholipids (23, 26). The BFA inhibition of binding of ARFl and -3 would appear to define one relatively specific component, i.e. that which is dependent on something(s) in the SAP fraction. These observations, which are very clearcut and reproducible, seem to be consistent with the view that ARF binding is an early step in Golgi vesicle formation, and it is this step that is inhibited by BFA (8,9,36).
While this paper was being completed, there were two more reports of effects of BFA that are believed to be due to its interference with ARF binding (37,38). In these studies, ARF binding, evaluated for the most part as an exchange of proteinbound guanyl nucleotide produced by incubation of myristoylated recombinant ARFl with Golgi, was essentially completely inhibited by BFA. This BFA inhibition corresponds to a specific functional step, the result of interaction of an ARF molecule with a specific nucleotide exchange protein, which had been earlier postulated as a mechanism for triggering vesicle formation when an ARF molecule with bound GDP interacts with a specific membrane protein that activates it, i.e. catalyzes the replacement of GDP with GTP. This is a very plausible, in fact appealing, mechanism, but it leaves us with the question of why BFA-inhibited binding in our studies was dependent on SAP. We think it quite possible that the Golgi preparations used by other workers, which are obtained from different tissue sources, contain the ARF-binding protein(s) that is missing from our rat brain Golgi fractions but is supplied by SAP.
BFA is believed to affect several types of membranes, in addition to the Golgi (29). Thus, it might be expected that its effects on ARF binding would include ARFs that interact with membranes other than Golgi. Current studies are directed toward identifying the membranes involved in the ARFspecific interactions. In any case, the observation that ARF5 binding was unaffected by BFA or SAP is consistent with the view that it does not bind to the same membranes as ARFl or ARFS.