GTP-dependent Binding of ADP-ribosylation Factor to Coatomer in Close Proximity to the Binding Site for Dilysine Retrieval Motifs and p23*

A site-directed photocross-linking approach was employed to determine components that act downstream of ADP-ribosylation factor (ARF). To this end, a photolabile phenylalanine analog was incorporated at various positions of the putative effector region of the ARF molecule. Depending on the position of incorporation, we find specific and GTP-dependent interactions of ARF with two subunits of the coatomer complex, β-COP and γ-COP, as well as an interaction with a cytosolic protein (∼185 kDa). In addition, we observe homodimer formation of ARF molecules at the Golgi membrane. These data suggest that the binding site of ARF to coatomer is at the interface of its β- and γ-subunits, and this is in close proximity to the second site of interaction of coatomer with the Golgi membrane, the binding site within γ-COP for cytosolic dibasic/diphenylalanine motifs.

COPI 1 -coated vesicles transport proteins and lipids between intracellular compartments along the secretory pathway (1,2). The initial step in the formation of these vesicles involves the recruitment of two factors from the cytosol to the Golgi membrane; ADP-ribosylation factor (ARF) binds to Golgi membranes in a GTP-dependent manner (3)(4)(5) with subsequent recruitment of coatomer, a soluble complex of seven subunits (COPs) (6,7).
Coatomer has a KKXX-COOH binding site (8), capable of binding both the dibasic ER retrieval motifs and the diphenylalanine motifs of members of the p24 family of proteins (9,10). Conflicting data have been obtained concerning the binding site(s) of these motifs within coatomer. Using an affinity column containing a KKXX sequence, a coatomer subcomplex of ␣-, ␤Ј-, and -COP could be isolated (8,11). The diphenylalanine motif was shown to bind to ␤-, ␥and -COP (9). However, using direct photolabeling of intact coatomer with various photolabile peptides containing the dibasic and diphenylalanine motifs, only the ␥-COP subunit was labeled, suggesting that coatomer has only one binding site for the various types of these motifs (12,13). The diphenylalanine sequence is present in the Cterminal tails of the members of p24 family of type I membrane proteins (9,10,14), and it has been proposed that one or more members of this family provide(s) a matrix for coatomer binding to Golgi membranes and subsequent formation of COPIcoated vesicles (10).
We previously used a site-directed photocross-linking approach to elucidate components that interact directly with ARF1 during coat assembly (15). This method is based on the replacement of an endogenous amino acid with a photolabile analog of phenylalanine, L-4Ј-(3-trifluoromenthyl-3H-diazirin-3-yl)phenyalanine ((Tmd)Phe) (16,17). It was found that this analog in position 82 of ARF1 directly and exclusively interacts with the ␤-COP subunit of coatomer during the formation of transport vesicles (15). Thus, it was inferred that a bivalent interaction of coatomer with Golgi membranes via ARF and via members of the p24 family of proteins is involved in the budding of a COPI-coated vesicle (15).
ARF, however, has been implicated in several other biochemical reactions such as activation of phospholipase D (18,19) and interaction with heterotrimeric G proteins (20,21) and biologically active phospholipids (22). To determine whether ARF interacts with additional effector molecules, we inserted the (Tmd)Phe analog at various positions of the putative effector loop of ARF, identified in its crystal structure and based on its similarity with Ras (23). We find that, depending on the site of insertion, the effector loop of ARF interacts with ␤-COP, ␥-COP, and an as yet unidentified protein of ϳ185 kDa. In addition, evidence is presented that ARF is capable of forming a homodimer on Golgi membranes, the function of which remains to be determined.

EXPERIMENTAL PROCEDURES
Synthesis of Site-specific Photolabile ARF Mutants-For details, refer to Zhao et al. (15). In short, the codon of the ARF cDNA corresponding to the amino acid position of interest was replaced with amber stop codon using the QuickChange site-directed mutagenesis kit from Stratagene. Amber suppressor tRNA, charged chemically with the photolabile amino acid (Tmd)Phe, was synthesized as described (16,17). In vitro transcription using T7 RNA polymerase was performed with linearized plasmids according to the manufacturer's protocol. In vitro translation using Flexi-lysate (Promega) was performed in the presence of [ 35 S]methionine and 5 M suppressor tRNA at 30°C for 2 h. The suppression efficiency was analyzed by SDS-PAGE (24) and subsequent autoradiography.
Irradiation of the Photolabile ARF Mutants-20 l of the in vitro translation samples (performed as described above) was incubated in 25 mM Hepes/KOH (pH 7.2), 2.5 mM Mg(OAc) 2 , 20 mM KCl, 1 mg/ml ovalbumin, 1 mM dithiothreitol, 0.2 M sucrose, and 50 M nucleotide (GDP␤S or GTP␥S) in the presence or absence of isolated CHO Golgi membranes, of purified rabbit liver coatomer, of recombinant mARF1, and of isolated bovine brain cytosol as described in the figure legends in a total volume of 500 l for 15 min at 37°C. After incubation the Golgi membranes were pelleted in a microcentrifuge (30 min at 14,000 rpm at 4°C) and resuspended in 20 l of 25 mM Hepes/KOH (pH 7.2), 20 mM KCl, 2.5 mM Mg(OAc) 2 , and 0.2 M sucrose. Samples were irradiated at 366 nm for 2 min on ice, and thereafter the membranes were pelleted in a microcentrifuge and analyzed by SDS-PAGE and sequent autoradiography. When no membranes were present during the incubation, the total volume of incubation was 50 l and these samples were not subjected to centrifugation but rather the complete incubation subjected to irradiation. Golgi membranes from Chinese hamster ovary cells and bovine brain cytosol were isolated as described by Beckers et al. (25), coatomer was purified as described by Waters et al. (26) with modifications according to Pavel et al. (27), and recombinant mARF1 was produced and purified as described by Weiss et al. (28) and modified according to Helms et al. (29).
Antibodies and Immunoprecipitation-In case the proteins were denatured prior to immunoprecipitation, SDS was added to a final concentration of 1% and the samples (20 l) were incubated for 3 min at 95°C. For immunoprecipitation, the samples were incubated with the indicated antibodies in 200 l of 20 mM Tris-HCl (pH 7.5), 2 mM EDTA, 0.15 M NaCl, and 0.5% Triton X-100 (immunoprecipitation buffer) for 2 h at 4°C by head-over-head rotation. If SDS was present in the sample, the amount of Triton X-100 was increased to 0.9% to have a 10-fold excess of Triton X-100 over SDS. Subsequently the samples were incubated with protein A-Sepharose (Amersham Pharmacia Biotech) for 2 h at 4°C by head-over-head rotation. The beads were washed five times in immunoprecipitation buffer and once in phosphate-buffered saline. The precipitated material was solubilized in sample buffer and analyzed by Western blotting and autoradiography. For immunoprecipitation of native coatomer complex, anti-␤Ј-COP antibodies were used (30), and for immunoprecipitation of dissociated coatomer, antibodies against the respective individual subunits were used (against ␣-COP (31), ␤-COP (32), ␤Ј-COP (30), and ␥-COP (12)). The same antibodies were also used for analysis by Western blotting. For visualization the enhanced chemiluminescence system (Amersham Pharmacia Biotech) was used.

RESULTS
In order to analyze proteins or molecules that act downstream of ARF (effector molecules), we focused on ␤-sheet ␤2 (amino acids [41][42][43][44][45][46] and loop L4 (amino acids 47-50) (23). These structures constitute the putative effector region in ARF, based on its similarity with the Ras proteins, as revealed by a comparison of the crystal structure of ARF and Ras (23). The following criteria were applied for selection of amino acids to be replaced with the photolabile (Tmd)Phe: (i) insertion of the photolabile residue should not change the net charge of the ARF molecule and (ii) the amino acid to be replaced should have an optimal orientation (i.e. facing outward from the ARF molecule) for optimal cross-link efficiencies. Based on these criteria, we choose Val-43, Ile-46, and Ile-49 to be replaced with (Tmd)Phe, resulting in ARF mutants designated ARF-(Tmd)-Phe-43, ARF-(Tmd)Phe-46, and ARF-(Tmd)Phe-49, respectively.
The photolabile ARF molecules were generated by mutating the corresponding codons of ARF1 cDNA to amber and subsequent in vitro transcription and translation with suppression of this stop codon by addition of suppressor tRNA chemically loaded with the photolabile amino acid analog (see "Experimental Procedures"). Translation in the presence of [ 35 S]methionine allows visualization of the corresponding photolabile fulllength protein by SDS-PAGE and autoradiography. Each mutated and radiolabeled ARF showed a GTP-dependent binding to Golgi membranes, excluding the possibility that replacement of the endogenous amino acid for (Tmd)Phe had severely affected ARF activity (data not shown).
Incubation of [ 35 S]ARF-(Tmd)Phe-46 with Golgi membranes in the presence of GTP␥S and subsequent irradiation of the membrane-bound ARF resulted in three cross-link products with estimated molecular masses of between 120 and 140 kDa (Fig. 1A, lane 2). The molecular masses of the cross-link products are in the range of several coatomer subunits (reviewed in Ref. 33), and therefore we determined whether addition of  [3][4][5][6] or absence (all other lanes) of purified coatomer, in the presence of 20 g of myristoylated ARF1 (lane 6) or in the presence of 2 mg/ml bovine brain cytosol (lanes 7-9). After incubation, the samples were centrifuged and the membranes were irradiated (except lanes 4 and 8), and membrane-bound material was analyzed by 12% SDS-PAGE (24) and autoradiography. When no membranes were present during the incubation (lane 1), 1/10th of the complete incubation was irradiated (without prior centrifugation) and was analyzed as described above. Before addition of cytosol, Golgi membranes were preincubated for 5 min at 37°C with [ 35 S]ARF-(Tmd)Phe-46 to prevent competition of photolabile ARF with ARF1 from cytosol. B, immunoprecipitation of membrane-bound coatomer and its individual subunits. In vitro translated ARF-(Tmd)Phe46 was bound to CHO Golgi membranes in the presence of GTP␥S and coatomer as described in panel A, lane 5. After irradiation of membrane-bound ARF, the membranes were solubilized in detergent-buffer and the coatomer complex was immunoprecipitated using an anti-␤Ј-COP antibody (lane 2), or the individual coatomer subunits were immunoprecipitated after denaturation of the complex, presence of GDP␤S (Fig. 1A, lane 3) or in the absence of irradiation (Fig. 1A, lane 4) or in the presence of an excess of wtARF (Fig. 1A, lane 6). ARF has to be bound to Golgi membranes for the cross-linking to occur, as an incubation in the absence of membranes lead to a complete absence of cross-link products (Fig. 1A, lane 1).
The specific cross-link products are also observed in the presence of high concentrations of bovine brain cytosol (a high speed supernatant from bovine brain homogenate), demonstrating the highly specific interaction between ARF and these proteins (Fig. 1, lanes 7-9). Under these conditions, Golgi membranes were preincubated for 5 min at 37°C with [ 35 S]ARF-(Tmd)Phe-46 before addition of cytosol in order to prevent competition of photolabile ARF1 with ARF1 from cytosol. Surprisingly, another cross-link product of ϳ205 kDa appears in the presence of cytosol (Fig. 1, lane 9). Subtracting the molecular mass of ARF, this protein has an apparent molecular mass in the range of about 180 -185 kDa. We considered the possibility that under these circumstances, ARF might also interact with clathrin heavy chain (CHC), a protein with similar molecular mass, or with AP180, a clathrin assembly-promoting phosphoprotein, or with subunits of the adaptor protein (AP) complex, all of which are recruited to the membrane to form a clathrin coated vesicle (34 -37). Immunoprecipitation with an ␣-CHC antibody did immunoprecipitate CHC from cytosol but failed to immunoprecipitate this radioactive band (data not shown). Likewise, antibodies against AP180, ␥-adaptin (a subunit of the AP-1 complex), and ␤3B (a subunit of the AP-3 complex) did immunoprecipitate their respective antigen but failed to immunoprecipitate the cross-linked product. Although we cannot exclude the possibility that the cross-link occurs at the epitope of the antibody (making it inaccessible for immunoprecipitation), these data make it unlikely that the crosslink-product is due to an interaction of ARF involved in the formation of clathrin-coated vesicles.
Immunoprecipitation studies in the absence of cytosol confirmed the presence of the cross-link products (ϳ120 -140 kDa) in the coatomer complex (Fig. 1B). An anti-coatomer antibody that immunoprecipitates native coatomer complex did immunoprecipitate all three cross-link products (Fig. 1B, lane 2). Coatomer has four subunits (␣-, ␤-, ␤Ј-, and ␥-COP) with a molecular weight in the range in the cross-link product. Antibodies directed against each of these subunits were used to determine which subunit(s) is bound to ARF-(Tmd)Phe-46 ( Fig.  1B, lanes 3-6). To this end, the samples were treated with SDS in order to dissociate coatomer before immunoprecipitation (for details see "Experimental Procedures"). Antibodies against ␣-COP (lane 3) and ␤Ј-COP (lane 5) did not immunoprecipitate the cross-link products, whereas antibodies against ␤-COP (lane 4) and ␥-COP (lane 6) efficiently immunoprecipitated two different cross-link products. The third band might be derived from ␤-COP, as it is faintly seen with the ␤-COP antibody (Fig.  1B, lane 4). One possibility is that more than one ARF binds to ␤-COP under these conditions. Alternatively, it could be that a small fraction of ARF cross-links at a different position within ␤-COP, affecting its migration.
Incubation of [ 35 S]ARF-(Tmd)Phe-49 with isolated CHO Golgi membranes in the presence of GTP␥S and subsequent irradiation of the membrane-bound [ 35 S]ARF-(Tmd)Phe-49 gave a relatively complex cross-link pattern (Fig. 2A, lane 4). However, most of these cross-links are unspecific, as they also appear in the presence of GDP␤S or in the absence of irradiation ( Fig. 2A, lanes 2 and 3, respectively). Two bands at ϳ40 kDa are specific according to these criteria and only appear in the complete incubation ( Fig. 2A, lane 4 versus lanes 2 and 3). These bands are dependent on the orientation of ARF in the membrane as they do not appear without membranes (Fig. 2A,  lane 1). Several potential candidates could be involved in the formation of the cross-link product of ϳ40 kDa, including an interaction of ARF with the p24 family of proteins (10,14). However, antibodies against p23 and p24 did not immunoprecipitate the cross-link products, although they did immunoprecipitate their respective antigens (data not shown). Alternatively, the cross-link products at ϳ40 kDa could be due to dimerization of ARF itself. This possibility cannot, of course, be tested by immunoprecipitation with ␣-ARF antibodies, as they would also immunoprecipitate ARF-(Tmd)Phe-49 together with its cross-link product. Therefore, we added increasing amounts of wtARF to the incubation of ARF-(Tmd)Phe-49 with Golgi membranes and GTP␥S (Fig. 2B). If a cross-link exists with a protein other than ARF, this would be expected to reduce the efficiency of cross-linking as the specific activity of [ 35 S]ARF-(Tmd)Phe-49 is reduced (see Fig. 1 and Ref. 15). However, the efficiency of the cross-linking was increased with increasing amounts of wtARF (Fig. 2B). This strongly suggests that ARF-(Tmd)Phe49 forms a dimer with ARF present on the membranes, because according to the law of mass effect, only in this case an increase of the cross-link signal is expected.
When ARF-(Tmd)Phe-49 is incubated under standard conditions but in the presence of coatomer, an additional cross-link product appears at ϳ120 kDa ( Fig. 2A, lane 7). This band is not observed in the presence of GDP␤S or in the absence of irradiation ( Fig. 2A, lane 7 versus lanes 5 and 6). Addition of cytosol to the complete incubation (under conditions as described for ARF(Tmd)Phe-46) also yielded the specific cross-link products at ϳ40 kDa and at ϳ120 kDa ( Fig. 2A, lane 10 versus lanes 8  and 9). The dependence on the presence of coatomer or cytosol for the ϳ120-kDa products to appear suggested a cross-linking of ARF-(Tmd)Phe-49 to a coatomer subunit. Indeed, immunoprecipitation of coatomer with an anti-coatomer antibody immunoprecipitated the 120-kDa cross-link product (Fig. 2C, lane  2). After dissociation with SDS, antibodies against ␣-COP (lane 3) and ␤Ј-COP (lane 5) did not immunoprecipitate of the crosslink product, whereas antibodies against ␤-COP (lane 4), and, to a much lesser extent, also ␥-COP, immunoprecipitated the cross-link product. Interestingly, both with ARF-(Tmd)Phe-46 and with ARF-(Tmd)Phe-49, in fact two bands (migrating at ϳ120 and 140 kDa) are immunoprecipitated with the ␥-COP antibody (Fig. 1B, lane 6; Fig. 2C, lane 6). Whereas the upper band predominates with ARF-(Tmd)Phe-46 (Fig. 1B), the lower band predominates with ARF-(Tmd)Phe-49 (Fig. 2C). As the immunoprecipitation of coatomer-subunits after dissociation of the coatomer complex is subunit-specific and no other subunits co-immunoprecipitate (15), it is likely that ARF can cross-link to two different positions of ␥-COP, affecting its migration by SDS-PAGE. Depending on the location of the photolabile analogue in ARF, one cross-link of ␥-COP prevails over the other.
ARF-(Tmd)Phe-43 was bound to Golgi membranes under the same conditions as described above for ARF-(Tmd)Phe-46 and ARF-(Tmd)Phe-49, but did not result in any detectable crosslinking to an interacting protein. This does not exclude the possibility that this site is important for interaction with other proteins, as the distance of the interaction partner with the photolabile residue is critical for cross-link efficiency (17). DISCUSSION Using a site-directed photocross-linking approach, we previously identified a direct interaction between ARF and ␤-COP (15). With the photolabile group at a different positions within the ARF molecule, we now find that ARF also interacts with the ␥-subunit of coatomer. Because single amino acids at position 46 as well as position 49 can interact both with ␤-COP and with ␥-COP (Figs. 1 and 2), this implies that these two subunits are close neighbors within the coatomer complex. Several coatomer subcomplexes have previously been described (8,11,27,31), including a ␤/␥/ subcomplex (9). It is important to note, however, that the cross-linking approach used here is qualitative and does not allow a quantitative determination of, e.g., the stoichiometry between ARF and coatomer. The main reason is that, in an aqueous environment, the generated carbene will react predominantly with water molecules, making crosslink yields Յ1% quite typical (38). This high reactivity, however, renders this photoprobe so specific, because a very close proximity of a protein is needed to compete with the ubiquitously abundant water molecules. Under these conditions, it can be calculated (based on the specific activity of [ 35 S]methionine incorporated into in vitro translated ARF) that low picogram amounts of coatomer subunits are covalently bound to suppressed ARF, which does not permit quantitation of the amount of coatomer, bound to suppressed ARF (by bandshift on SDS-PAGE).
A schematic drawing of how the coatomer complex is bound to Golgi membranes is shown in Fig. 3. This model is based on the following experimental data. (i) The cytoplasmic tail of the p24 family of proteins specifically binds to coatomer (9,10,39). This tail contains a diphenylalanine motif and in some instances, like p23, an additional dilysine motif. The diphenylalanine motif sets these proteins apart from ER-resident proteins that contain only the dilysine ER retrieval signal (10,40). A single binding site exists within coatomer for both dilysine retrieval motifs and the cytoplasmic domain of p23, and this site resides in ␥-COP (13). (ii) By use of site-directed photocross-linking, we have shown a direct and GTP-dependent interaction of ARF with ␤-COP, which persists in COPI-coated vesicles and thus is not restricted to the budding process (15). These interactions indicate a bivalent interaction of coatomer with the Golgi membrane during vesicle formation, i.e. an anchoring of coatomer to Golgi membranes by ARF as well as by the C-terminal tails of the p24 family of proteins. (iii) The results described here demonstrate that ARF interacts with coatomer at the ␤/␥-COP interface. Thus, ␥-COP interacts both with ARF and the p24 family of proteins, indicating that the bivalent interaction occurs in close proximity of one to another. In the absence of ARF, coatomer cannot bind to Golgi membranes (6,7), and therefore it is tempting to speculate that the initial recruitment of coatomer to Golgi membranes by ARF affects the ␥-subunit so that it subsequently allows the C-   5-7) or absence (all other lanes) of purified coatomer, or in the presence of ARF-depleted cytosol (lanes 8 -10). After incubation, the samples were centrifuged and the membranes were irradiated (except lanes 4 and 8) and membrane-bound material was analyzed by 12% SDS-PAGE (24) and autoradiography. When no membranes were present during the incubation (lane 1), the complete incubation was irradiated (without prior centrifugation) and was analyzed as described in the legend to Fig. 1. Cytosol was added as described in the legend to Fig. 1.  3 and 4, respectively). The incubations were performed and analyzed as described in panel A. C, immunoprecipitation of membrane-bound coatomer and its individual subunits. In vitro translated ARF-(Tmd)Phe49 was bound to CHO Golgi membranes in the presence of GTP␥S and coatomer as described in terminal tail of p24 protein(s) to bind. This binding then would lead to a conformational change of coatomer, leading to polymerization of the complex and subsequent shaping of the membrane to form a bud (41). The uncoating of COPI-coated vesicles would simply be the reversal of this process: hydrolysis of ARF-bound GTP by a GTPase-activating protein results in uncoating of COPI-coated vesicles (42). Thus, hydrolysis of ARF-bound GTP reduces the affinity of coatomer for ARF and ARF dissociates from COPI-coated vesicles. This in turn could reverse the conformational change in ␥-COP to reduce the affinity for the p24 family of proteins, resulting in dissociation from transport vesicles of the coatomer complex.
Together with our previous study that shows an interaction of Phe-82 in ARF with ␤-COP, we have now mapped the interaction site of ARF and coatomer to a surface area of about 400 Å 2 , typical for protein-protein interactions. Interestingly, Ile-46, Ile-49, and Phe-82 are located on the switch 1 and 2 regions of ARF, respectively, and exactly these regions have been predicted to interact with the Sec7 domain of Arno (43), a guanine nucleotide exchange factor for ARF (44). It thus appears that one interface on ARF protein is used both for the interaction with the Sec7 domain (likely in the ARF GDP-bound form) and with the coatomer complex (in the GTP-bound form). It has been shown that N-terminally deleted ARF in its GDP-bound form can functionally interact with the Sec7 domain (45). It will be interesting to determine whether corresponding N-terminally deleted photolabile ARF proteins, mutated at the positions described in this study, can also cross-link to the Sec7 domain under these conditions.
With the photolabile amino acid residue at positions 46 and 49 of the ARF molecule, we have also found interactions with proteins other than coatomer. In the presence of cytosol, ARF-(Tmd)Phe-46 cross-links to a protein with an apparent molecular mass of ϳ185 kDa. Due to steric hindrance, it seems unlikely that, in addition to coatomer, a third protein binds to ARF at the same molecular interface. We consider it more likely that, under these conditions, part of the membranebound ARF is bound to coatomer and the other part of ARF is associated with p185. As ARF is also involved in the budding of clathrin-coated vesicles from the trans-Golgi network (34,35), we considered the possibility that CHC, a protein with a molecular mass of ϳ180 kDa, was bound to ARF under these conditions. However, immunoprecipitation studies with antibodies against CHC failed to immunoprecipitate the cross-link product. Similarly, antibodies against AP180 and various subunits of the adaptor-protein complex that are recruited to the Golgi complex via ARF did not immunoprecipitate the crosslink product of ϳ205 kDa. As both coatomer and the AP-1 complex are recruited by ARF to the Golgi complex and the trans-Golgi network, respectively (in the presence of GTP␥S), it is remarkable that so far we have not cross-linked one of the subunits of the adaptor protein complex. With respect to ARF involved, there are notable differences between COPI-coated vesicles and clathrin coated vesicles that bud from the Golgi. In the presence of GTP␥S, only COPI-coated vesicles are observed, the reason for which is not clear. Possibly this guanine analogue simultaneously inhibits dynamin, a putative pinchase for clathrin-coated vesicles, which is a GTP-binding protein itself (46,47). In addition, it has recently been reported that ARF1 only transiently activates high affinity adaptor protein complex AP-1 binding sites and that hydrolysis of ARF bound GTP can occur even before AP-1 binds (48). Finally, although ARF is involved in the recruitment of coat proteins for the formation of clathrin-coated vesicles, in contrast to COPI-coated vesicles, ARF1 is not present on coathrin-coated vesicles (48). Thus, whereas ARF is a stoichiometric component in COPI-coated vesicles, ARF seems only transiently involved in the formation of clathrin-coated vesicles and this might be the reason for the fact that we have only detected interactions with components of the COPI coat.
We are currently trying to elucidate the identity of the 205-kDa cross-linked protein by fractionation of cytosol and subsequent cross-link studies. One interesting alternative possibility is a spectrin homologue (ϳ220 kDa), recently shown to be recruited to the Golgi membranes by ARF1 (49,50).
Surprisingly, we have found evidence for the formation of ARF dimers on the Golgi membrane. ARF also crystallizes as a dimer (23), but it remains to be determined whether ARF dimers on Golgi membranes have the same orientation to one another as in the crystal structure. The N termini of both ARF molecules, thought to be involved in membrane anchorage of the ARF molecule, are on one interface of the crystallized dimer, theoretically allowing both N termini to interact with the membrane in this orientation. Another argument for the TABLE I Summary of the mutated positions in ARF1 and their respective photocross-linked proteins The indicated amino acid residues, secondary structures, and structural features of the positions chosen to be mutated, as shown in the table, are derived from the crystal structure of ARF1 (23). The interaction of ARF1 with ␤-COP at position 82 and a possible intramolecular photocross-linking at position 13 have been described previously (15). crystallized dimer orientation also occurring on the Golgi membranes is that according to the crystal structure, amino acid 49 in ARF is in close proximity to the other ARF molecule, i.e. in an optimal orientation for cross-linking. In this case, (Tmd)-Phe-49 would result in a cross-link between ARF-(Tmd)Phe-49 (in L4) with amino acid residues 159 -164 (in L12) in the second ARF molecule. If the cross-link represents an ARF homodimer, then the presence of a double band (cf. Fig. 2B) is not as easily explained, unless the two ARF molecules are different in, e.g., their myristoylation, known to affect the migration of ARF on SDS-PAGE (51). The purified recombinant myristoylated ARF added in excess is in fact a mixture of non-myristoylated and myristoylated ARF (29). Alternatively, intramolecular crosslinking may affect the migration of ARF on SDS-PAGE and might explain this phenomenon (see below). Finally, the two cross-link products at ϳ40 kDa might represent dimerization of different ARF isoforms. At least three different isoforms of ARF have been shown to bind to Golgi membranes (52).
In the course of these site-directed photocross-link studies, we have tried several additional positions for cross-link products. An overview of all the point mutations that we have analyzed is given in Table I. Positions 5 and 13 are present in two structural features that sets ARF apart from other GTPbinding proteins and are unique to ARF: the N-terminal ␣-helix (amino acid 2-11) and the connecting loop L1 (amino acid 10 -17). These two elements are not present in small GTPbinding proteins, but are present in the G␣ subunit of heterotrimeric G proteins (23). Mutations in both these elements (Phe-5 and Phe-13) did not cross-link a protein but caused a small shift (Յ1 kDa) of the apparent molecular mass of ARF-(Tmd)Phe after irradiation. Compared with the input (membrane-bound, myristoylated ARF), it was observed that after cross-linking, the cross-link product migrated like non-myristoylated ARF, which can be separated from myristoylated ARF by high resolution SDS-PAGE. This shift is not likely due to a cross-link with membrane lipids, as treatment with several lipases did not affect the migration of the cross-link product (data not shown). It is possible that cross-linking at these positions reflects an intramolecular cross-linking with e.g. the myristoyl moiety (which lies in close proximity), causing myristoylated ARF to migrate like non-myristoylated ARF.
Likewise, introduction of (Tmd)Phe at positions Lys-142 in ␣-helix E and Tyr-154 in ␤-sheet 7 of the ARF molecule did not yield protein-cross-links.
In summary, depending on the position of (Tmd)Phe in the ARF molecule, strikingly different cross-link products are obtained. The strict discrimination underlines the high specificity of the interactions observed with site-directed photocross-linking. We have identified the interface of ARF that interacts with coatomer as it is covered by Ile-46, Ile-49, and Phe-82, all on one site of the ARF molecule. With the many additional functions attributed to ARF, generation of additional photolabile ARF proteins with photocross-link residues at positions distal from the ones described here will allow one to pinpoint other partners than coatomer of this remarkable small GTP binding protein.