The Protein Cofactor Necessary for ADP-ribosylation of G , by Cholera Toxin is Itself a GTP Binding Protein*

A membrane-bound protein cofactor (ARF) is required for the cholera toxin-dependent ADP-ribosyla- tion of the stimulatory regulatory component (G,) of adenylate cyclase. Improved methods for the purifica- tion of ARF from bovine brain are described. ARF has a high-affinity binding site for guanine nucleotides. Binding of GTP or GTP-yS to ARF is necessary for the activity of the cofactor; GDP. ARF does not support ADP-ribosylation of G,. Although the protein as purified contains stoichiometric amounts of GDP, GTPase activity of isolated ARF was not detected. Cholera toxin-dependent activation of adenylate cyclase thus requires two guanine nucleotide binding proteins.

concluded that binding of GTP to G, was a prerequisite for its ADP-ribosylation. Gill and Meren (9) have suggested that two distinct guanine nucleotide binding sites are involved in the action of cholera toxin. These data and the development of new procedures for the purification of ARF, which allow production of milligram quantities of the factor in a more active form, prompted re-examination of the properties of the protein. ARF is in deed^ a guanine nucleotide-binding protein.
Although the physiological role of ARF remains to be defined, these observations suggest that at least three GTP binding proteins, ARF, G,, and Gi, may be involved in the regulation of adenylate cyclase activity.

MATERIALS AND METHODS
Purificatwn of ARF-Bovine brain membranes were prepared as described by Sternweis and Robishaw (lo), except that 0.5 mM phenylmethylsulfonyl fluoride was included. Extract was prepared from 12-15 g of membrane protein by incubation for 1 h at 4 "C in 2,000 ml of 20 mM Tris-C1 (pH 8.0), 1 mM EDTA, 1 mM dithiothreitol, 1% sodium cholate, and 0.1 mM phenylmethylsulfonyl fluoride. Extracted membranes were removed by centrifugation at 95,000 X g for 1 h. Supernatants were pooled and activators (AMF or 6 mM MgC12) were added at this point, as indicated. When activators were included they were added to all solutions used for chromatography on DEAE-Sephacel, Ultrogel AcA 44, and heptylamine-Sepharose.
The extract was applied to a DEAE-Sephacel column (5 X 60 cm), and proteins were eluted with a linear gradient (2 liters) of NaCl (0-250 mM) as described by Kahn and Gilman (8). The pooled DEAE peak was concentrated to approximately 30 ml by ultrafiltration using an Amicon PM 30 membrane and applied to an Ultrogel AcA 44 column as previously described (8). ARF eluted from Ultrogel AcA 44 with a KO of 0.5. The pool of ARF activity from the AcA 44 column was diluted with TED containing 100 mM NaCl to a final concentration of 0.25% sodium cholate and applied to a 130-ml column (27 X 2.5 em) of heptylamine-Sepharose that had been equilibrated with 600 ml of TED containing 100 mM NaCl and 0.25% sodium cholate. After loading, the column was washed with 100 ml of equilibration buffer before development with a 600-ml linear gradient of 0.5-1.5% sodium cholate in TED. The flow rate was 50-70 ml/h, and 8.5-ml fractions were collected. The peak of ARF activity eluted between 0.8 and 1.0% cholate. The pooled heptylamine-Sepharose peak was concentrated to 1 ml by ultrafiltration using an Amicon PM 30 membrane. Preparative sodium dodecyl sulfate-polyacrylamide gel electrophoresis and elution of activity from the gel were performed as described (8).
Assay of ARF-ARF was assayed as described by Kahn and Gilman (8). In brief, purified rabbit liver G, was incubated with activated cholera toxin, GTP, NAD, and ARF under conditions where the rate of ADP-ribosylation of G. was linearly related to the amount of ARF in the sample. Covalent modification of G. was then assessed by quantitation of the cholera toxin-dependent, GTP-stimulated adenylate cyclase activity after reconstitution of G, into cyc-(GB*deficient) membranes (11). The unit of ARF activity is expressed as pmol cyclic AMP/min.hg of G. (see Ref. 8).
Gunnine Nucleotide Binding Assay-Binding of nucleotides to ARF was determined with a modification of the rapid filtration technique described by Northup et al. (12). Radioactivity retained by the filter was linearly related to the amount of protein applied. At least 70% (78 4 8%; n = 6) of bound nucleotide was retained by the filter under the conditions utilized, as compared to samples in which unbound ligand was removed by gel filtration on G-25 Sephadex. ARF was incubated at 30 "C in 20 mM Na Hepes (pH 81, 1 mM EDTA, 1 mM dithiothreitol, 800 mM NaC1,6 mM MgClz, 3 mM DMPC, 0.1% sodium cholate, and nucleotide, as described in the text. The binding reaction was stopped by the addition of 2 ml of ice-cold buffer (25 mM Tris-C1 (pH 8), 100 mM NaC1,lO mM MgC12, 1 mM dithiothreitol), followed by rapid filtration on 25-mm BA 85 nitrocellulose filters (Schleicher and Schuell). Filters were washed seven times with 2 ml of the same buffer, dried under a heat lamp, and counted in 8 ml of scintillation mixture. Standards for each assay were dried on filters and counted identically. Nonspecific binding was 0.05% or less of the filtered radioactivity under these conditions. GTPase Actiuity-The rate of GTP hydrolysis was estimated by a modification of the method of Brandt et al. (13). Proteins were incubated at 30 "C in either 20 mM Na Hepes (pH 8.0),1 mM EDTA, 10 mM MgCl,, 1 mM dithiothreitol, 0.1% Lubrol PX, and 1 pM [y-32P]GTP (10-50 cpm/fmol) or 20 mM Na Hepes (pH 8.0), 1 mM EDTA, 2 mM MgCl,, 1 mM dithiothreitol, 0.8 M NaCl, 3 mM DMPC, 0.1% sodium cholate, and 1 pM [y-32P]GTP (10-50 cpm/fmol). At various times, 50-pl aliquots were added to 750 pl of ice-cold 5% Norite in 50 mM NaH2P04. After vortexing, samples were clarified by centrifugation (1,000 X g for 15 min) and the supernatant was counted. The zero-time blank, caused by contamination of substrate with 32Pi, was less than 2% of the total.
Purification of ARF by HPLC-ARF was purified through the heptylamine-Sepharose stage, and the solution was concentrated to 1 ml or less as described above. The AFW was generally about 30% pure at this point. Concentrated ARF was applied to two 300 X 7.5 mm Bio-Si1 TSK-250 columns (Bio-Rad) in series, previously equilibrated in 50 mM Na Hepes (pkI 8.0), 100 mM NaC1, 1 mM dithiothreitol, and 6 mM MgCl,. The columns were eluted with the same buffer at a flow rate of 0.3 ml/min. Fractions (0.3 ml) were collected and assayed for protein, ARF, and guanine nucleotide binding.
Identification of Nucleotide Bound to ARF-Determination of the guanine nucleotide content of pure ARF was performed as described by Ferguson et al. (14). ARF was purified to homogeneity by HPLC and chromatographed on Sephadex G-25 in 10 mM potassium phosphate (pH 8.0), 100 mM NaCI, and 1 mM EDTA, to remove M$+ and dithiothreitol. ARF (100 pg in 0.26 mi) was then boiled for 5 min, and denatured protein was removed by ultrafiltration using a PM 10 membrane. The UV spectrum of the filtrate was obtained with a Perkin-Elmer Lambda 5 UV/Vis spectrophotometer and was characteristic of guanine. The nature of the presumed guanine nucleotide in the filtrate was then determined by chromatography on a Polyanion SI HR 5/S anion exchange column (containing polyethylenimine on a silica support) using a Pharmacia fast protein liquid chromatography system. The column was equilibrated in 10 mM potassium phosphate, pH 7.0, and a flow rate of 1.0 ml/min was used at all times. After application of the sample, the column was washed with one column volume of equilibration buffer. Nucleotides were eluted with the following program of washes and gradients: 10-300 mM potassium phosphate over 1 min, 300 D M potassium phosphate for 2 min; 300-650 mM potassium phosphate over 20 min, and 650-700 mM potassium phosphate over 4 min. The column eluate was monitored at 254 nm. Retention times of guanine nucleotide standards were: 5'-GMP, 5.2 min; GDP, 10.9 min; GTP, 18.8 min.
using a Spex Fluorolog 211 spectrophotometer (double monochro-Fluorescence Measurements-Fluorescence spectra were recorded meter for excitation; single monochrometer for emission). Temperature was kept constant at 30 "C. Measurements were performed in the ratio mode to correct for variations in intensity of the xenon lamp using a neutral density filter (optical density of 3) as an attenuator in the reference beam. The optical bandwidths on the excitation and emission monochrometers were 2.25 and 4.5 nm, respectively.
Preparation of Guanine Nucleotide-liganded ARF-Guanine nucleotides were bound to ARF as described above. Samples were chilled on ice, and sodium cholate was added to a final concentration of 1%. Bound and free nucleotides were then separated by the method of Fleming and Ross (15). Sephadex G-25 was swollen and equilibrated in 20 mM Na Hepes (pH 8.0), 1 mM EDTA, 1 mM dithiothreitol, 2 mM MgC12, 100 mM NaCl, and 1% sodium cholate. Polypropylene columns (48 X 8 mm; Isolab, Inc.) were poured with a bed volume of 3 ml. Interstitial solvent was removed by centrifugation at 1000 rpm for 5 min in a Beckman model TJ-6 centrifuge. Samples (100 pl) were then applied, and voided material was collected by centrifugation at 1000 rpm for 5 min. Greater than 80% of protein was routinely recovered in a volume of 140-175 pl. Less than 0.03% of the unbound nucleotide was recovered in the voided fraction.

RESULTS
Purification of ARF-Purification of ARF from bovine brain membranes was achieved with essentially the same fourstep procedure described for rabbit liver (8). Starting with 15 g of brain membranes, one can obtain 2-5 mg of pure ARF with an overall recovery of 10% (Table I). This is about 10 times the yield from a comparable rabbit liver preparation; this results from the greater abundance of ARF in brain and the greater stability of the protein during purification. Only 300-fold purification was necessary to obtain pure ARF from brain membranes, compared to the 1800-fold purification required previously. Like ARF from rabbit liver, the purified material from bovine brain appears as a doublet (Mr -21,000) in sodium dodecyl sulfate-polyacrylamide gels. Partial resolution of these two forms of ARF revealed very similar specific activities. The addition of a mixture of protease inhibitors (phenylmethylsulfonyl fluoride, leupeptin, N-tosyl-L-phenylalanine chloromethyl ketone, and bacitracin) appeared t o increase the relative amount of the larger band; however, the relationship of these bands remains open to question.
The presence of AMF was required for recovery of any ARF activity from the DEAE-Sephacel column using rabbit liver as the source of membranes (8): A single peak of ARF activity was observed when AMF or Mg2+ alone was included during the chromatography of brain ARF on DEAE-Sephacel (Fig. lA). Omission of Mg2+ or AMF caused ARF to elute in two peaks; the latter was coincidental with G, (Fig. 1B).
The lability of ARF from rabbit liver during chromatography on heptylamine-Sepharose made it necessary to elute activity with a step gradient (8). This was not the case with ARF from brain, and a linear gradient of sodium cholate resulted in elution of ARF that was between 20 and 60% pure.
T h e predominant contaminants after such chromatography are the a subunit of Go and an unknown protein (Mr = 26,000). The inclusion of AMF or MgC12 during chromatography on heptylamine-Sepharose was necessary for the elution of ARF in a single peak. Omission of AMF or Mg2+ at this step caused extreme broadening of the peak of ARF activity.
Studies on the effects of AMF on the chromatographic behavior of ARF led to the realization that AMF or Mg2+ alone markedly stabilizes ARF activity in cholate extracts of brain membranes. The half-life of ARF activity in cholate extracts was 2.5 days at 4 "C in the absence of Mg2+ (and presence of 1 mM EDTA) and 14 days in 5 mM Mg2+. Mn2+, but not Ca2+, could also stabilize ARF activity in such extracts.
Results of purification in which all buffers contained 5 mM free Mg2+ (but no added A13+ or F-) were the same or better (recoveries) than those done in AMF, suggesting that both the stability and the chromatographic behavior of ARF are dependent on the divalent cation.
Velocity Sedimentation and Purification in the Absence of Detergent-Electrophoretically purified ARF sediments as a mono-disperse species with an S value of 2.10 (identical t o ARF from liver; 8) when applied to a linear sucrose gradient AMF was originally added to preparations from rabbit liver to stabilize G, activity and to facilitate purification of both G. and ARF from the same extract. It is likely that only Mg2+ is required to recover ARF activity after chromatography of rabbit liver extracts on DEAE-Sephacel, but this has not been tested.

TABLE I Purification of ARF from bovine brain membranes
The extract was prepared from a crude membrane preparation obtained from two bovine brains as described under "Materials and Methods." Recoveries shown are cumulative through the purification. Alternative final steps in the purification scheme are shown for comparison, and values are normalized to reflect yields from the entire preparation. (SDS-PAGE, sodium dodecyl sulfate-polyacrylamide gel electrophoresis.) Step  containing 1% sodium cholate. Recovery of activity is nearly complete. In the absence of added detergent, recovery is poor (8-20%), and activity is found throughout the gradient (Fig.   2). When ARF from the heptylamine-Sepharose column (-30% pure) is applied to sucrose gradients with or without cholate, a single peak of activity is observed and recovery of activity is high (SO-90%). However, the activity profile from gradients centrifuged in the absence of added detergent is consistently broader than those that contain cholate. Thus, electrophoresis of ARF in the presence of sodium dodecyl sulfate appears to change its behavior during sucrose density gradient centrifugation in the absence of detergent.
These results suggested that alternative methods of purification should be explored, and it was found that ARF can be purified to homogeneity with good recovery by gel filtration (Bio-Si1 TSK) chromatography in the absence of detergents (Fig. 3). The main contaminants in the preparation after chromatography on heptylamine-Sepharose seem to aggregate in the absence of detergent and appear in the voided fractions of the TSK column. ARF elutes at the position expected for a 21,000-dalton protein. The specific activity of ARF (up to 400,000 units/mg) purified in this way is 50-400% greater than comparable samples prepared by gel electrophoresis in the presence of sodium dodecyl sulfate. HPLC purified ARF was used in all subsequent experiments.
Binding of Guanine Nucleotides-ARF binds a number of guanine nucleotides, including GTP, GDP, and GTP+. Such  (6.6 mg, about 20-30% pure) was applied to two Bio-Si1 TSK 250 (300 X 7.5 mm) columns, connected in series, which were equilibrated and developed in 50 mM Na Hepes (pH 7.5), 1 mM dithiothreitol, 5 mM MgCl,, and 100 mM NaCl at a flow rate of 0.3 ml/min as described under "Materials and Methods." Fractions (0.3 ml) were collected and assayed as described under "Materials and Methods." The peak of the excluded volume was in fraction 27; the peak of the column volume was in fraction 73. ARF activity (O), GTP+ binding ( X ) , and protein (A) were determined as described under "Materials and Methods." Guanine nucleotide-binding activity in the excluded peak was due to Go,.

FIG. 3. Purification of ARF on Bio-Si1 TSK columns. ARF
binding is observed in the presence of lipid, M P , and high ionic strength ( Fig. 3; Table 11). Rapid dissociation of bound nucleotide is observed at 30 "C upon removal of M e (not shown). Binding is optimal between pH 8.0 and 8.5. The maximal binding of guanine nucleotides was routinely between 0.15 and 0.3 mol of nucleotide/mol of protein using electrophoretically purified ARF from bovine brain or rabbit liver. This stoichiometry is improved with brain ARF purified on the Bio-Si1 TSK column (0.45-0.70 mol of nucleotide bound/mol of 21,000-dalton protein), consistent with the greater specific activity of this preparation (see above and Fig. 4). If a correction is made for the recovery of binding sites in the filter trapping assay (78%; see "Materials and Methods"), the maximal stoichiometry observed is nearly 0. for the binding of GTPyS with apparent dissociation constants of 90 and 40 nM, respectively (Fig. 5). Binding of ATP, ADP, and GMP was not detectable at ligand concentrations as high as 0.1 mM.
The possibility of hydrolysis of GTP by ARF was investigated using conditions optimal for binding of guanine nucleotides to ARF, as well as with conditions optimal for GTP hydrolysis by G proteins (13). GTPase activity was not found in either case (turnover <0.0015 min"). In the same experiments purified Goa hydrolyzed GTP at a rate of about 0.3-0.4 min". ARF did not hydrolyze measurable amounts of GTP even when the concentration of MgClz was increased to as high as 0.1 M. Consistent with the lack of measurable GTPase activity was the observation that the extents of binding of [a-32P]GTP and [y-32P]GTP to ARF were identical.
Slow binding of guanine nucleotides to G proteins has recently been shown to result in large part from the presence of GDP on the purified proteins (14). The slow release of bound GDP from Gi, or Go, is the rate-limiting step for the binding of labeled nucleotide. In view of similar kinetics of guanine nucleotide binding to ARF (see Fig. 7), the presence of noncovalently bound nucleotide on ARF was investigated as described under "Materials and Methods." ARF was purified by HPLC and filtered on Sephadex G-25 to remove Mg2+ and any free nucleotide. After heat denaturation and removal of protein by ultrafiltration, the UV spectrum of the filtrate was recorded. The spectrum between 240 and 300 nm was characteristic of a guanine nucleotide, and the optical density was consistent with the presence of 1 mol of guanine nucleotide/mol of ARF. The nature of the putative guanine nucleotide in this sample was then determined by anion exchange chromatography as described under "Materials and Methods." The filtrate produced a single peak of UV absorbance that migrated at the position corresponding to GDP. Control experiments indicated that GTP was not hydrolyzed during the preparation of samples. Quantitation of the GDP content (by optical density) yielded a value of 0.9 mol of GDP/mol of ARF. Thus, purified ARF contains essentially equimolar amounts of noncovalently bound GDP.
Effects of Guanine Nucleotides on the Intrinsic Fluorescence of ARF-The fluorescence of tryptophan residues of ARF containing bound GDP, GTP, or GTPyS was examined by excitation at 290 nm. When GTP or GTPyS was bound to ARF, the intrinsic fluorescence was increased by about 110% over that observed with the GDP form of the protein (Fig. 6). The spectra for GTP-and GTPyS-liganded ARF were superimposable, consistent with the failure to observe GTPase activity in the preparation. The addition of AMF had no effect on fluorescence, and the spectrum of ARF in AMF was identical to that of GDP . ARF.
The intimate association between the binding of GTP+ and the increase in intrinsic fluorescence as a function of time is demonstrated in Fig. 7. When GDP was added instead of GTPyS, the intrinsic fluorescence did not change during the 2-h incubation at 30 "C.
Guanine Nucleotide Requirement for ARF Activity-With the findings that ARF binds guanine nucleotides and that conditions required for binding are very similar to those previously described for the expression of ARF activity, it was possible to reexamine the site of action of GTP required for cholera toxin catalyzed ADP-ribosylation of G,. Purified ARF was first incubated with either no guanine nucleotide, GDP, GTP, or GTPyS for 2 h at 30 "C. Cholate (1%) was added at  is a GTP Binding Protein 0 "C to disrupt vesicles, and each sample was then applied to a G-25 column to separate bound and free nucleotides, as described under "Materials and Methods." The various forms of ARF were then assayed for their ability to support cholera toxin-dependent activation of G,, either in the presence or absence of added GTP. Carryover of free nucleotides was less than 0.03%, sufficient to bind to less than 1% of the G, in the reaction mixture. ARF that had not been preincubated (not shown) or that had been incubated with no guanine nucleotide or with GDP was unable to support ADP-ribosylation unless GTP was present during the incubation with toxin and G, (Fig. 8). When GTP was added to these samples, there was a lag of up to 5 min before the rate of activation became linear. ARF containing bound GTP or GTPyS was able to support activation of up to 30% of the G, present, even when no GTP was added during the second incubation (Fig. 8). There was no lag with GTP-ARF or GTPyS-ARF. Activation was maximal when GTP was included in the second incubation. The decreased rate and extent of activation seen when samples first incubated with GTP or GTPyS were assayed without GTP in the second incubation results in part from the dissociation of nucleotides from ARF. In addition, ADP-ribosylated G, is very unstable in the absence of GTP, and the assay utilized depends on the ability of G, to stimulate adenylate cyclase activity. These results demonstrate that binding of GTP to G, is not required for ADP-ribosylation catalyzed by cholera toxin, as previously believed. Rather, the requirement for GTP in this reaction is ascribable to the nucleotide binding site on ARF. The lag noted previously (8) is due to the ratelimiting binding of GTP to ARF.
These results also clarify another ambiguity. We previously reported that G,. GTPyS is not a substrate for cholera toxin, yet GTPyS can partially support the toxin-catalyzed modification of G, when added to reaction mixtures containing basal G, (19). It is now clear that, indeed, G,. GTPyS is not a substrate for cholera toxin; when GTPyS is added to basal G, and ARF in the presence of cholera toxin, the ADP-ribosylation that occurs is the result of the interaction of ARF. GTPyS with basal G,. discussed previously suggests that ARF interacts with G,, rather than the toxin (8). The data presented above indicate that ARF binds guanine nucleotides with high affinity and, as purified, contains 1 mol of GDP/mol of protein. Although the physiological role of ARF is unknown and effects of ARF on adenylate cyclase activity have not yet been detected, these observations raise the possibility that two guanine nucleotidebinding regulatory proteins are involved in the pathway responsible for stimulation of adenylate cyclase activity.
Binding of GTP (or GTPyS) to ARF is necessary and sufficient for the ADP-ribosylation of G,. Nucleotide hydrolysis is not necessary, and GDP.ARF does not support the covalent modification. Although hydrolysis of GTP by ARF has not been detected under the conditions tested, the finding that ARF is purified as a complex with GDP and analogies with other proteins suggest that ARF may in fact be a GTPase. A component necessary for GTPase activity may have been resolved from ARF during purification. Precedent for such comes from bacterial elongation factor Tu, which binds GTP but hydrolyzes the nucleotide only when bound to ribosomes in the presence of mRNA and aminoacyl-tRNA (20). A systematic search t o detect an additional hypothetical factor will be necessary. Although G, is an obvious candidate, the rate of GTP hydrolysis observed in the presence of G, was unchanged by the addition of equimolar amounts of ARF, under conditions where ARF binds guanine nucleotide^.^ These negative data serve only to set an upper limit to the rate of any hydrolysis of GTP by ARF under these conditions. This limit (0.0015 min") is close to the rate of hydrolysis of GTP by the transforming viral ras proteins (21).
There are similarities between the guanine nucleotide-binding properties of ARF and those of G proteins. These include high affinity and specificity for guanine nucleotides and retention of bound GDP during purification. Differences are also apparent, including the lack of intrinsic GTPase activity of ARF, the lack of effect of A13+ and F-,4 and the absolute requirement for Mg2' for nucleotide binding to ARF. The failure of ARF to bind guanine nucleotides in the presence of detergent and the requirement for high ionic strength for such binding are also notable. These characteristics, in addition to partial denaturation of ARF during purification by sodium dodecyl sulfate-polyacrylamide gel electrophoresis and the very limited amount of protein previously available, account for our earlier failure to detect the guanine nucleotide-binding capacity of ARF.
It is more tempting to compare ARF and the ras gene products, although this exercise is clearly compromised by ignorance. ARF and the p21 products of the ras genes are membrane-bound, GTP-binding proteins of similar mass. Both are found in high relative abundance in brain. Of interest, both have been implicated as consorts of adenylate cyclase. The yeast ras gene products, which are approximately 40,000-dalton proteins, and the smaller p21 proteins of viral and human origin clearly stimulate yeast adenylate cyclase activity (22) but cannot replace G, as an activator of adenylate cyclases of higher organisms (23). It is not known if there is a direct interaction between the ras proteins and adenylate cyclase or if the GTP binding site on ras proteins accounts for the sensitivity of yeast adenylate cyclase to guanine nucleotides. In yeast it is possible that ras influences adenylate cyclase via the intermediacy of a distinct G,-like protein. If R. A. Kahn and A. G. Gilman, unpublished observations. A13+ and F-(probably acting as AlF;) cause dissociation of the and B-y subunits of G proteins and greatly enhance the ability of G, to stimulate adenylate cyclase. so, the analogy with ARF may become obvious. Unfortunately, our simple attempts to test this relationship have yet to yield positive results. These tests have utilized viral ras proteins expressed as fusidn proteins in Escherichia coli and monoclonal antibodies to the viral proteins (24, 25). p21 has no ARF activity. Antibodies to the viral ras proteins do not appear to interact with ARF. Limited amino acid sequence information on ARF reveals no striking homology with ras proteins. More detailed characterization of ARF will be necessary to allow further comparison.