Characterization of Chemical Inhibitors of Brefeldin A-activated Mono-ADP-ribosylation*

Brefeldin A, a toxin inhibitor of vesicular traffic, induces the selective mono-ADP-ribosylation of two cytosolic proteins, glyceraldehyde-3-phosphate dehydrogenase and the novel GTP-binding protein BARS-50. Here, we have used a new quantitative assay for the characterization of this reaction and the development of specific pharmacological inhibitors. Mono-ADP-ribosylation is activated by brefeldin A with an EC50 of 17.0 ± 3.1 μg/ml, but not by biologically inactive analogs including a brefeldin A stereoisomer. Brefeldin A acts by increasing theV max of the reaction, whereas it does not influence the K m of the enzyme for NAD+(154 ± 13 μm). The enzyme is an integral membrane protein present in most tissues and is modulated by Zn2+, Cu2+, ATP (but not by other nucleotides), pH, temperature, and ionic strength. To identify inhibitors of the reaction, a large number of drugs previously tested as blockers of bacterial ADP-ribosyltransferases were screened. Two classes of molecules, one belonging to the coumarin group (dicumarol, coumermycin A1, and novobiocin) and the other to the quinone group (ilimaquinone, benzoquinone, and naphthoquinone), rather potently and specifically inhibited brefeldin A-dependent mono-ADP-ribosylation. When tested in living cells, these molecules antagonized the tubular reticular redistribution of the Golgi complex caused by brefeldin A at concentrations similar to those active in the mono-ADP-ribosylation assay in vitro, suggesting a role for mono-ADP-ribosylation in the cellular actions of brefeldin A.

Brefeldin A, a toxin inhibitor of vesicular traffic, induces the selective mono-ADP-ribosylation of two cytosolic proteins, glyceraldehyde-3-phosphate dehydrogenase and the novel GTP-binding protein BARS-50. Here, we have used a new quantitative assay for the characterization of this reaction and the development of specific pharmacological inhibitors. Mono-ADP-ribosylation is activated by brefeldin A with an EC 50 of 17.0 ؎ 3.1 g/ml, but not by biologically inactive analogs including a brefeldin A stereoisomer. Brefeldin A acts by increasing the V max of the reaction, whereas it does not influence the K m of the enzyme for NAD ؉ (154 ؎ 13 M). The enzyme is an integral membrane protein present in most tissues and is modulated by Zn 2؉ , Cu 2؉ , ATP (but not by other nucleotides), pH, temperature, and ionic strength. To identify inhibitors of the reaction, a large number of drugs previously tested as blockers of bacterial ADPribosyltransferases were screened. Two classes of molecules, one belonging to the coumarin group (dicumarol, coumermycin A 1 , and novobiocin) and the other to the quinone group (ilimaquinone, benzoquinone, and naphthoquinone), rather potently and specifically inhibited brefeldin A-dependent mono-ADP-ribosylation. When tested in living cells, these molecules antagonized the tubular reticular redistribution of the Golgi complex caused by brefeldin A at concentrations similar to those active in the mono-ADP-ribosylation assay in vitro, suggesting a role for mono-ADP-ribosylation in the cellular actions of brefeldin A.
Mono-ADP-ribosylation is a post-translational modification of proteins whereby the adenosine 5Ј-diphosphoribose moiety of NAD ϩ is transferred to one of a number of amino acid residues (1,2). Several well characterized mono-ADP-ribosylation reactions are catalyzed by bacterial toxins and result in the permanent modification of GTPases with key regulatory functions in cellular processes (1,2). Similar reactions are also catalyzed by eukaryotic enzymes, but their cellular significance, despite recent advances (3)(4)(5), is less well understood. Recently, we have reported that brefeldin A (BFA), 1 a fungal toxin metabolite of palmitic acid (6) with potent inhibitory effects on intracellular membrane traffic (7), stimulates the selective mono-ADP-ribosylation of two cytosolic proteins of 38 and 50 kDa in mammalian cells (8). The 50-kDa substrate (BARS-50; see Ref. 9) binds GTP and is regulated by ␤␥-subunits of trimeric G proteins; it has therefore been proposed to be a novel G protein involved in membrane transport (9). The p38 substrate is glyceraldehyde-3phosphate dehydrogenase (GAPDH), a multifunctional protein involved in several cellular processes (10 -16). These observations have raised the question as to whether some of the cellular effects of BFA might be mediated by mono-ADP-ribosylation.
The cellular actions of BFA are multiple and complex. BFA selectively blocks constitutive protein secretion and causes a rapid and extensive disruption of the Golgi apparatus consisting of the transformation of the Golgi stacks into a tubular reticular network followed by the redistribution of most of the Golgi resident proteins into the endoplasmic reticulum (ER) (17)(18)(19)(20). BFA also profoundly affects the morphology and function of the endocytic system (21)(22)(23)(24). Some of these effects are most probably due to an already well documented effect of the toxin, namely the release of a set of proteins (25)(26)(27)(28), in particular the coat proteins ADP-ribosylation factor (ARF) and coatomer (a major protein complex involved in COPI-coated vesicle formation), from Golgi membranes (7, 29 -32). It is far from clear, however, that all of the functional and structural effects of BFA are due to the inactivation of the above coat proteins. Moreover, there is evidence, albeit indirect, that suggests that also the ADP-ribosylation reaction may play a role in the cellular effects of BFA. First, BFA activates ADP-ribosylation both in intact and Triton-solubilized Golgi membranes through a site exhibiting a ligand selectivity identical to that involved in the BFA effects on the Golgi structure (9). This suggests that the same (or very similar) BFA-binding components may be involved in the two BFA-induced events. Second, nonspecific inhibitors of mono-ADP-ribosylation such as nicotinamide (33,34) can reduce the effects of the toxin on the morphology of the Golgi complex (35). It is therefore of interest to develop a set of specific inhibitors of BFA-induced mono-ADP-ribosylation as they would greatly help to define the cellular role of this reaction.
Here, we report the characterization and development of a series of chemical inhibitors of the BFA-dependent mono-ADPribosylation reaction. They belong to two chemical classes, containing either a coumarin (dicumarol, coumermycin A 1 , and novobiocin) or a quinone (ilimaquinones and analogs) group. These drugs, when tested in vivo at concentrations similar to those effective in vitro mono-ADP-ribosylation assays, antagonized the ability of BFA to induce the disassembly of the Golgi complex.

Materials
BFA, NAD ϩ , nucleotides, DL-dithiothreitol, and fluorescein 5-isothiocyanate (FITC)-conjugated Helix pomatia lectin were from Sigma. Coenzymes Q 2 , Q 4 , and Q 6 were from Fluka. All the other inhibitors, unless otherwise specified, were from Sigma. BFA analogs were prepared by Dr. A. Greene (J. Fourier University, Grenoble, France) as described previously (36,37); ilimaquinone was a gift from Dr. V. Malhotra (University of California, San Diego, CA); and AA861 was from Dr. T. Shimizu (University of Tokyo, Tokyo, Japan). [ 32 P]NAD ϩ was from DuPont NEN or Amersham Corp. Rabbit skeletal muscle GAPDH was from Calbiochem or Sigma. The monoclonal antibody against ARF (1D9) was a gift of Dr. R. Kahn (Emory School of Medicine, Atlanta, GA). The 35 S-labeled rabbit antibody against mouse IgG was from Amersham Corp. Polyclonal antibodies against the calcium-binding protein (CaBP1) and calreticulin were gifts of Dr. J. Fü llekrug (University of Göttingen, Göttingen, Germany). The polyclonal antibody against mannosidase II was a gift of Dr. K. Moremen (University of Georgia, Athens, GA). Monoclonal antibodies against Rab5 and Rab7 were gifts from Dr. M. Zerial (EMBL, Heidelberg, Germany); the monoclonal antibody against TGN38 was donated by Dr. G. Banting (School of Medical Science, Bristol, United Kingdom). FITC-and tetramethylrhodamine 5-isothiocyanate-conjugated secondary antibodies were from Sigma or Cappel. The materials for cell culture, including plasticware, chamber slides (Nunc, Roskilde, Denmark), and medium (Life Technologies, Inc.), were purchased from Mascia Brunelli S. p. A. (Milano, Italy).

Cell Culture
Rat basophilic leukemia 2H3 cells were grown in Eagle's minimal essential medium supplemented with 16% fetal calf serum and 1 mM L-glutamine as described (38).

Preparation of the Post-nuclear Supernatant, Membranes, and Cytosol from Rat Tissues
Harlan Sprague Dawley male rats (150 -250 g) were killed by decapitation. Tissues were immediately removed and placed in ice-cold homogenization buffer (0.32 M sucrose, 4 mM HEPES, and 1 mM EDTA, pH 7.3), washed four to five times with the same buffer, minced with scissors, and homogenized in a Teflon-glass Potter-Elvehjem homogenizer (15-20 strokes). All procedures were carried out at 4°C. The homogenate was centrifuged at 700 ϫ g for 10 min, and the post-nuclear supernatant was collected and ultracentrifuged at 150,000 ϫ g for 90 min. The supernatant was discarded, and the pellet (suspended in 3 M KCl and incubated for 30 min with stirring) was centrifuged at 12,700 ϫ g in a microcentrifuge. The pellet obtained, termed "total membranes," was washed and suspended in phosphate-buffered saline (PBS), pH 7.4; aliquoted; frozen in liquid nitrogen; and then stored at Ϫ80°C until needed. The reproducibility of these preparations was satisfactory. Each experiment was performed with at least two preparations, with similar results. Rat brain cytosol was prepared as described (39).

Subcellular Fractionation
Subcellular fractions were prepared as described (39,40). In some experiments, mitochondria were further purified as described by Rusinol et al. (41). To check the composition of each fraction (postnuclear supernatant, rough ER, smooth ER, Golgi membranes, mitochondria, and cytosol), equal amounts of protein (80 g) were separated by SDS-PAGE, blotted onto nitrocellulose, and probed with polyclonal antibodies against specific markers of intracellular organelles by Western blotting. The purity of the mitochondrial fraction was also tested by electron microscopy. Mitochondria were pelleted and processed as described previously (42) and then photographed with a Zeiss 109 transmission electron microscope. They were Ͼ90% pure.

ADP-ribosylation
ADP-ribosylation of the Cytosol-Unless otherwise specified, 10 g of total membranes and 50 g of cytosol, both from rat brain (or alterna-tively, 50 g of post-nuclear supernatant), were incubated at 37°C for 1 h with 30 g/ml BFA in 50 mM phosphate buffer, pH 7.4, 30 M NAD ϩ , 0.01 Ci/l [ 32 P]NAD ϩ , 2.5 mM MgCl 2 , 5 mM DL-dithiothreitol, and 10 mM thymidine in a final volume of 50 l. The reaction was stopped by the addition of Laemmli sample buffer (51). Samples were boiled for 5 min and analyzed by 10% SDS-PAGE. Proteins were then transferred to a nitrocellulose sheet by electroblotting, and the radiolabeled proteins were quantified by autoradiography or by an Instant Imager (Canberra Packard).
ADP-ribosylation of Purified GAPDH-A rapid filtration assay was set up to measure the labeling of purified commercial GAPDH (8,9). Unless otherwise specified, 10 g of GAPDH were incubated with 10 g of total membranes from rat brain as described above for the ADPribosylation assay. Samples were then pelleted at 12,000 ϫ g for 10 min, and the supernatant was mixed with 100 l of ice-cold 1 mM NAD ϩ , 1 mM AMP, and 0.6 mM sodium pyrophosphate (final concentrations) and then filtered on a 96-well Bio-Rad dot-blot apparatus through a nitrocellulose sheet. Each well was filled with ice-cold 20% trichloroacetic acid (200 l/well) prior to the addition of the sample. GAPDH was allowed to adhere to the nitrocellulose by a gentle filtration and washed twice by filtering ice-cold 20% trichloroacetic acid (100 l/well) and four times with low salt Tris-buffered saline (20 mM Tris-HCl, pH 7.5, and 500 mM NaCl). The nitrocellulose sheet was then boiled in distilled water for 10 min and stained with Ponceau S. Radiolabeled GAPDH was quantified by an Instant Imager. GAPDH was recovered quantitatively on the filters. This was established by processing in parallel a variety of identical samples by both the filter and SDS-PAGE-based assays (8,9). None of the treatments described under "Results and Discussion" affected the amount of protein recovered on the filter. As previously reported (15,16), GAPDH is nonenzymatically mono-ADPribosylated in the absence of BFA. Thus, for each condition tested in this study, control experiments using the BFA vehicle (dimethyl sulfoxide) were carried out to evaluate the basal nonenzymatic ADP-ribosylation of GAPDH. The pmol of ADP-ribosylated GAPDH reported in this study were calculated by subtracting the values of nonenzymatically ADP-ribosylated GAPDH from those obtained in the presence of BFA.

ARF Binding Assay
Membranes (10 g of protein) were incubated for 10 min with 100 g of rat brain cytosol in a final volume of 100 l containing 25 mM HEPES-KOH, pH 7, 25 mM KCl, 2.5 mM MgCl 2 , 1 mM ATP, 1 mM DL-dithiothreitol, and 40 g/ml BFA where indicated; 10 M GTP␥S was then added, and incubation was continued for 5 min. The reaction was stopped by centrifugation at 14,000 ϫ g at 4°C, and the pellets were boiled for 5 min in Laemmli sample buffer (51). Membrane proteins were separated by 12% SDS-PAGE, blotted onto nitrocellulose, and probed with a mouse monoclonal antibody against ARF (1D9) and a 35 S-labeled rabbit antibody against mouse IgG. Radioactivity was quantified by an Instant Imager.

Immunofluorescence Microscopy
Rat basophilic leukemia cells (grown on glass chamber slides) were fixed in 4% paraformaldehyde in PBS at room temperature for 10 min; quenched with 10 mM NH 4 Cl for 10 min; washed with PBS; and permeabilized with 0.05% saponin and 0.2% bovine serum albumin in PBS for 30 min at room temperature. The cells were stained with FITCconjugated H. pomatia lectin (100 g/ml in PBS containing 0.2% bovine serum albumin) for 45 min or were incubated with primary antibody for 1 h at room temperature, washed thoroughly with PBS, and incubated with specific FITC-or tetramethylrhodamine 5-isothiocyanate-conjugated secondary antibody for 30 min at room temperature. After thorough washing, the slides were mounted in Mowiol 4-88 (Calbiochem) and examined using a Zeiss Axiophot microscope equipped with a Plan-Neofluar 100ϫ objective.

Kinetic Features of the BFA-dependent Mono-ADP-ribosylation Reaction
So far, work on BFA-stimulated mono-ADP-ribosylation has been carried out by labeling the cytosolic substrates with radioactive NAD ϩ , separation by SDS-PAGE, and autoradiography (8,9). In this study, to characterize the reaction in a quantitative fashion, we have developed a novel simplified assay based on the use of salt-washed membranes as the source of enzyme, pure commercial GAPDH as the substrate, a rapid filtration method to separate free from GAPDH-bound radioactivity, and counting of the radiolabeled GAPDH by an Instant Imager (see "Experimental Procedures"). This approach afforded a high sample output and reproducibility as well as the possibility to study modulatory effects of added cytosolic factors. Fig. 1 shows that BFA stimulated the reaction with an EC 50 of 17 Ϯ 3 g/ml. The apparent K m of the reaction with respect to NAD ϩ was 154 Ϯ 13 M (Fig. 1, inset). The V max was 510 Ϯ 150 pmol/h/mg of membrane protein (at 30 g/ml BFA). BFA regulated the reaction velocity by increasing the V max , whereas the K m was not detectably modified by varying the toxin concentration (Fig. 1, inset). Of note, the EC 50 of BFA was similar to that found by us in assays of ARF binding to Golgi membranes, but higher than the BFA EC 50 in vitro. The difference between the in vivo and in vitro potencies of BFA has been noticed and discussed before by several authors (31,43). A comparison of the results of the new assay ( Fig. 1) with data obtained using the old method (where the cytosol was the source of substrate; see Ref. 8) did not reveal evident differences, suggesting that cytosolic factors have no major effects on the reaction kinetics.

Influence of pH, Temperature, Ions, and Nucleotides
The BFA-stimulated ADP-ribosylation of GAPDH was maximal near pH 7.6 at temperatures between 37 and 40°C, with a second smaller peak at 30°C, and was inhibited by higher than physiological levels of NaCl (Fig. 2). The double peak in the temperature profile is unusual. At present, we do not have a simple explanation for it. Of the major physiological divalent ions that were tested, Ca 2ϩ , Mg 2ϩ , and Mn 2ϩ were inactive, whereas Zn 2ϩ and Cu 2ϩ were inhibitory (with IC 50 values of ϳ75 and 500 M, respectively). Several nucleotides (GTP, GDP, GTP␥S, GDP␤S, ATP␥S, and cAMP and different combinations of these molecules) had no effect on GAPDH mono-ADP-ribosylation at concentrations up to 1 mM, whereas ATP was inhibitory (with an IC 50 of ϳ1 mM).

Tissue and Subcellular Distribution of BFA-dependent
Mono-ADP-ribosylation Rat brain, liver, heart, spleen, lung, skeletal muscle, and kidney were tested for their content of enzymatic mono-ADPribosylating activity (using commercial GAPDH as a substrate) and of endogenous cytosolic substrates (GAPDH and BARS-50) (using rat brain membranes as the enzyme source). Fig. 3A shows that the enzymatic activity was present in all tissues examined, with the highest levels in the brain and the lowest (nearly undetectable) in the kidney. It is unclear whether this reflects a real lack of expression of the enzyme in kidney tissue rather than, for instance, the presence of inhibitory factors or of degradative enzymes. The protein substrates were ubiquitous, with the highest levels in the brain, spleen, and heart (data not shown).
Among the subcellular fractions, one fraction enriched in Golgi membranes and one containing the smooth ER as well as Golgi membranes were highly active; a lower but significant activity was found in mitochondria and the lowest in the rough ER (Fig. 3B), whereas the cytosol was completely devoid of enzyme (data not shown). Since the ability of BFA to prevent the activation and binding of ARF to intracellular membranes is a well known in vitro activity of the toxin and is considered responsible for at least some of the cellular BFA effects, it was interesting to compare the subcellular distribution of this activity with the distribution of BFA-dependent mono-ADP-ribosylation. Fig. 3C shows that ARF binding was enriched in the Golgi membrane-containing subcellular fractions and was present (at lower levels) in the ER and mitochondria. This distribution profile was quite similar to that of BFA-dependent mono-ADP-ribosylation and is therefore compatible with the idea that the two activities of BFA (ARF binding and mono-ADP-ribosylation) might be associated (this point is discussed further below). The meaning of the presence of both activities in mitochondria is not clear at present. It is very unlikely to be due to contamination by Golgi or smooth ER membranes, as indicated by a comparison between their distribution with that of marker enzymes (CaBP1 for the endoplasmic reticulum, mannosidase II for Golgi membranes, and Rab5 for early endosomes; see Fig. 3D). The binding of ARF to mitochondria and other subcellular fractions has been observed by others (44). Since mitochondria are capable of fission and fusion (45), it is possible that ARF binding and mono-ADP-ribosylation might reflect the presence of a membrane transport machinery in this organelle.

Structure-Activity Relationships of BFA Analogs in the Mono-ADP-ribosylation Assay
As noted above, BFA is nearly equipotent in in vitro mono-ADP-ribosylation and ARF binding assays. Moreover, as previously reported, two synthetic analogs of BFA known to be inactive in preventing ARF binding to isolated Golgi membranes, B5 and B36, are also devoid of mono-ADP-ribosylationstimulating activity (8). Based on these observations, it has been suggested that the BFA-binding sites involved in the two activities are similar or identical (8). We wanted to perform a more stringent test of parallelism between the two activities by using a large variety of BFA-derived analogs, including the biologically inactive stereoisomer of BFA (7). Fig. 4 shows that the BFA analogs B5, B17, B18, B23, B27, and B36, reported to lack BFA activity in living cells and in ARF binding assays (7,30), also did not stimulate BFA-dependent mono-ADP-ribosylation, whereas compound B30, reported to possess weak BFAlike activity, stimulated BFA-dependent mono-ADP-ribosylation with a correspondingly low potency. Among the inactive analogs, of particular interest is the stereoisomer of BFA (B27): its low activity obviously indicates a remarkable selectivity of the binding site (in some experiments, B27 had weak stimulatory effects probably to the presence of contaminating quantities of the active isomer). Thus, these results confirm and expand the previous conclusion that the BFA-binding components mediating mono-ADP-ribosylation, ARF binding, and the effects of BFA in vivo possess a closely similar ligand specificity (8), suggesting that they may belong to the same protein family.

FIG. 3. Tissue and subcellular distribution of the BFA-dependent mono-ADP-ribosylating activity.
A, GAPDH (10 g) was incubated with 10 g of membranes from the indicated rat tissues as described under "Experimental Procedures." B, 50 g of rat brain cytosol were incubated with 8 g of subcellular fractions enriched in Golgi membranes, smooth ER, rough ER, and mitochondria as described under "Experimental Procedures." Samples were then processed by SDS-PAGE, and the labeling of GAPDH was analyzed by electronic autoradiography (Instant Imager). C, 10 g of each subcellular fraction were incubated and processed in the ARF binding assay as described under "Experimental Procedures." Data are means Ϯ S.D. of triplicate determinations. Similar results were obtained in at least three independent experiments. D, 80 g of each subcellular fraction were separated by SDS-PAGE, blotted onto nitrocellulose, and probed with antibodies against specific markers of intracellular organelles (CaBP1 for the endoplasmic reticulum, mannosidase II for Golgi membranes, and Rab5 for early endosomes). ADPr, adenosine 5Ј-diphosphoribose.

Chemical Inhibitors of in Vitro BFA-dependent Mono-ADP-ribosylation
Coumarin-containing Molecules-A large number of molecules chosen mostly among proposed inhibitors of characterized eukaryotic ADP-ribosyltransferases (33,34) were screened for inhibitory activity in the BFA-dependent mono-ADP-ribosylation assay. In an initial survey, the antibiotic novobiocin (34) was found to be the most potent, with an effect on the BFA-dependent mono-ADP-ribosylation of GAPDH at 350 M. Novobiocin contains the coumarin moiety also present in many inhibitors of the NADH-binding enzyme diaphorase (46). A series of structurally similar diaphorase inhibitors was thus tested. Among them, coumermycin A 1 and dicumarol were found to be more potent BFA-dependent mono-ADP-ribosylation blockers than novobiocin; however, other molecules with coumarin-containing structures, including potent inhibitors of diaphorase such as warfarin (47), were inactive in the mono-ADP-ribosylation assay (Fig. 5A). Moreover, even the active molecules (coumermycin, dicumarol, and novobiocin) inhibited diaphorase and ADP-ribosylation with very different potencies. Thus, while the site(s) involved in BFA-dependent mono-ADPribosylation recognizes certain coumarinic ligands, it does so with a selectivity completely different from that of diaphorase. The type of inhibition caused by these agents was noncompetitive with respect to both NAD ϩ and BFA (Fig. 5 (B and C) shows a characterization of the effects of dicumarol) and was apparently not modulated by cytosolic factors or by nucleotides or divalent ions (data not shown).
Ilimaquinone and Related Compounds-Certain quinones have been reported to inhibit eukaryotic mono-ADP-ribosyltransferases (34). An interesting quinone in the context of studies on membrane traffic is ilimaquinone, a sea sponge metabolite recently reported to have potent effects on membrane transport and on the Golgi structure (48 -50). Strikingly, when ilimaquinone was tested in the mono-ADP-ribosylation assay, it inhibited the BFA-induced reaction with an IC 50 of ϳ30 M (Table I). Like dicumarol, ilimaquinone acted in a noncompetitive manner with respect to both BFA and NAD ϩ (data not shown). A number of other quinones were then tested in the mono-ADP-ribosylation assay. Benzoquinone, naphthoquinone, menadione, and coenzyme Q 0 had significant inhibitory activity, suggesting a role of the quinone moiety in the action of these agents (Table I). Inactive quinones included coenzymes Q 2 , Q 4 , Q 6 , Q 9 , and Q 10 and vitamins K 1 and K 2 ( Table I).
Nicotinamide and Analogs-Nicotinamide and aminobenzamides are known mono-ADP-ribosylation inhibitors that function by competing with the nicotinamide moiety of NAD ϩ at several NAD ϩ -binding sites. They inhibit nuclear poly-ADPribosyltransferase (33,34) as well as, and more weakly, eukaryotic (33,34) and prokaryotic mono-ADP-ribosyltransferases (with IC 50 values in the mM range), including cholera toxin, pertussis toxin, and other bacterial enzymes. As expected, nicotinamide and aminobenzamides inhibited also BFAdependent mono-ADP-ribosylation, albeit with a low potency (IC 50 ϳ 20 mM; see Table I), and acted in an apparently competitive fashion with respect to NAD ϩ (data not shown).
Ligand Selectivity of the ADP-ribosylation Inhibitor (ARI) Site-Functionally, the ARI site is defined by the rank order of potency exhibited by the ADP-ribosylation inhibitors. We wanted to know whether the ARI site(s) resembles sites present on known ADP-ribosyltransferases or NADases. To address this point, we examined whether dicumarol, coumermycin A 1 , novobiocin, and ilimaquinone are able to interfere with the in vitro activity of ADP-ribosylating enzymes, such as cholera and pertussis toxins, or of rat brain NAD ϩ glycohydrolases. None of the above blockers exerted any effect on these enzymatic activities at concentrations up to 1 mM (data not shown). Moreover, an inspection of previously published lists of inhibitors of eukaryotic ADP-ribosyltransferases (33,34) indicated that there is little overlap between the groups of inhibitory agents active on these ADP-ribosyltransferases and those acting on the ARI site. Altogether, the data therefore suggest that the ligand selectivity of the ARI site(s) is unique among ADP-ribosyltransferases and NADases.
The ARI site(s) might be located on the enzyme or on the mono-ADP-ribosylation substrates. If the latter is the case, then the two substrates should have very similar binding sites since all the inhibitors tested so far inhibit ribosylation in the same fashion. More work is needed to distinguish between these two possibilities.

Effects of the Mono-ADP-ribosylation Inhibitors on the BFA-induced Disassembly of the Golgi Complex
Having developed a series of specific inhibitors of the ARI site(s), we wanted to know if they are able to antagonize any of the in vivo effects of BFA and, more important, if the rank order of in vivo potency of these inhibitors corresponds to that found in the in vitro mono-ADP-ribosylation assay. If this were the case, it would strongly suggest that the ARI site(s) is involved in the cellular effects of BFA. One such effect is the disassembly and redistribution of the Golgi complex into the ER. The ability of the ADP-ribosylation inhibitors to prevent this effect of BFA was tested. Indeed, dicumarol, an inhibitor of the coumarin class, markedly inhibited the BFA-induced Golgi disassembly (as revealed by staining the Golgi complex with H. pomatia lectin or a mannosidase II antibody) (Fig. 6) at concentrations (IC 50 ϳ 200 M) very similar to those effective in blocking ADP-ribosylation in vitro. The other two active coumarin-containing mono-ADP-ribosylation blockers, novobiocin and coumermycin, also inhibited the Golgi disassembly by BFA with IC 50 values of ϳ450 and 60 M, respectively. Again, these concentrations are very similar to those effective in inhibiting in vitro mono-ADP-ribosylation. In contrast, coumarin analogs that were inactive in the mono-ADP-ribosylation test were all inactive as BFA antagonists in vivo, although they had structures and chemical-physical properties similar to those of dicumarol. Thus, there was an excellent agreement between the in vitro and in vitro potencies of the inhibitors of the coumarin class. It is known that BFA not only acts on the Golgi complex, but also causes the mixing and disorganization of the trans-Golgi network and endosomes (21)(22)(23)(24). To examine whether these BFA effects were inhibited by ADP-ribosylation inhibitors, we stained these structures with antibodies against TGN38 and Rab7 (markers of the trans-Golgi network and late endosomes, respectively). As expected, dicumarol (200 M) partially but clearly prevented the BFA-induced diffusion and mixing of these structures, whereas it had no effect on the distribution of calreticulin, a marker of the endoplasmic reticulum (data not shown). Thus, it appears that ADP-ribosylation inhibitors might also act on the trans-Golgi network-endosomal system.
Dicumarol was then characterized further with respect to whether its inhibitory action in vivo might be due to general toxicity. For instance, an inhibition of the BFA-induced redistribution of the Golgi complex might be due to a decrease in the   cytosolic levels of ATP or to microtubule disruption (7). Dicumarol affected neither ATP levels nor microtubules 2 ; moreover, its effects were reversible (data not shown), and viability tests (trypan blue and cellular morphology) carried out for up to several hours of exposure to dicumarol did not reveal any apparent toxicity. Finally, the inhibition by dicumarol could be overcome by increasing the concentration of BFA: at 5 g/ml, the toxin induced an apparently normal BFA-dependent Golgi redistribution despite the presence of the inhibitor (data not shown; a complete description of the cellular effects of the ADP-ribosylation inhibitors will be published separately). 2 These data indicate that dicumarol does not impair the complex machinery that must underlie Golgi redistribution. They suggest, rather, that the drug is likely to act by antagonizing specifically the action of BFA. When the other two classes of ARI, namely quinones and aminobenzamides, were tested for their ability to inhibit BFA in vivo, a good correlation was again found between their inhibitory potency in in vitro mono-ADP-ribosylation assays and their potency in antagonizing BFA in vivo. Ilimaquinone, in agreement with previous studies, caused the breakdown of the Golgi complex into initially large and then (after 30 min or more) progressively smaller fragments; it also, however, inhibited the typical, rapid (5 min) redistribution of Golgi markers into an ER-like pattern induced by BFA (with an IC 50 of ϳ50 M). Also other quinone inhibitors of mono-ADP-ribosylation appeared to have dicumarol-like effects. However, they also caused evident cellular toxicity, rendering their effects in vivo as antagonists of BFA difficult to assess. Finally, nicotinamide and aminobenzamides, the third chemical group of mono-ADPribosylation inhibitors, inhibited the effects of BFA on the Golgi structure at very high concentrations (ϳ20 mM), in agreement with their low potency (IC 50 ϳ 20 mM) in the in vitro mono-ADP-ribosylation assay.
Altogether, the rank order of potency of all of the abovementioned drugs as BFA antagonists in vivo therefore appeared to be very similar to that defined in mono-ADP-ribosylation assays in vitro. Table I shows a list of the drugs tested in the mono-ADP-ribosylation assay and in the test of in vivo antagonism of BFA. The correlation between the potencies shown by the inhibitors in the two assays ( Fig. 7) is striking and strongly suggests that these inhibitors prevent the BFAinduced Golgi disassembly via their action on the ARI site. Clearly, the possibility that some of the ARI inhibitors may be partially toxic or exert cellular effects unrelated to mono-ADPribosylation cannot be ruled out; however, the above correlation indicates that nonspecific effects are very unlikely to contribute to the ability of these drugs to antagonize the Golgi disassembly induced by BFA. Not only the rank orders of potency, but also the actual active concentrations of these drugs in vitro and in vivo are quite similar, suggesting that factors often found to affect drug behavior in vivo, such as membrane permeability, transport, and metabolism, do not seem to play a major role in the case of mono-ADP-ribosylation inhibitors. This is somewhat surprising in the case of nicotinamide and related compounds, whose limited ability to permeate membranes has been proposed by other authors to be responsible for their higher potencies in in vitro than in in vivo assays (33). The significance of these differences is unclear at present.
In conclusion, a series of specific inhibitors of BFA-dependent mono-ADP-ribosylation has been developed, characterized, and shown to prevent the effects of BFA on the structure of the Golgi complex by acting, apparently, on the same site through which they block mono-ADP-ribosylation in vitro. This result is To evaluate the IC 50 of the inhibitors in the ADP-ribosylation assay, 10 g of commercial GAPDH were incubated with 10 g of rat brain membranes for 1 h at 37°C in the presence of 30 g/ml BFA, 30 M [ 32 P]NAD, and different concentrations of the inhibitors. The IC 50 is the drug concentration inhibiting 50% of the BFA-dependent signal. The IC 50 in the Golgi redistribution assay was evaluated in rat basophilic leukemia cells treated with 1 g/ml BFA and is defined as the concentration of inhibitor that prevented the BFA-induced redistribution of H. pomatia lectin staining in 50% of the BFA-sensitive cells (90% of the cells responded to BFA under control conditions). The percentage of cells exhibiting BFA-induced redistribution was evaluated by an expert observer in a blind fashion by examining 50 randomly chosen cells/ sample. Experiments were repeated three times in duplicate with similar results.
in itself of interest in that it suggests that this BFA-dependent reaction and its protein substrates GAPDH and/or BARS-50 play a critical role in the maintenance of the structure/function of the Golgi complex. The precise role and mechanism of ADPribosylation in this organelle remain, however, to be determined. Clearly, the use of pharmacological inhibitors in vivo has limitations. However, the ease with which these inhibitors can be used and the current lack of molecular approaches for studies of BFA-dependent ADP-ribosylation may render these drugs important tools in future investigations of the role of ADP-ribosylation in the cellular effects of BFA and membrane traffic.