Arfaptin 1, a putative cytosolic target protein of ADP-ribosylation factor, is recruited to Golgi membranes.

ADP-ribosylation factors (ARFs) have been implicated in vesicle transport in the Golgi complex. Employing yeast two-hybrid screening of an HL60 cDNA library using a constitutively active mutant of ARF3 (ARF3·;Q71L), as a probe, we have identified a cDNA encoding a novel protein with a calculated molecular mass of 38.6 kDa, which we have named arfaptin 1. The mRNA of arfaptin 1 was ubiquitously expressed, and recombinant arfaptin 1 bound preferentially to class I ARFs, especially ARF1, but only in the GTP-bound form. The interactions were independent of myristoylation of ARF. Arfaptin 1 in cytosol was recruited to Golgi membranes by ARF in a guanosine 5′-O-(3-thiotriphosphate)-dependent and brefeldin A-sensitive manner. When expressed in COS cells, arfaptin 1 was localized to the Golgi complex. The yeast two-hybrid system yielded another clone, which encoded a putative protein, which we have named arfaptin 2. This consisted of the same number of amino acids as arfaptin 1 and was 60% identical to it. Arfaptin 2 was also ubiquitously expressed and bound to the GTP-, but not GDP-liganded form of class I ARFs, especially ARF1. These results suggest that arfaptins 1 and 2 may be direct target proteins of class 1 ARFs. Arfaptin 1 may be involved in Golgi function along with ARF1.

ADP-ribosylation factors (ARFs), 1 which were originally identified and purified by their ability to enhance the ADPribosyltransferase activity of cholera toxin, comprise a distinct subfamily of Ras-related small GTP-binding proteins and have been found in all eukaryotic cells from yeast to human (1). To date, at least six mammalian ARF genes have been cloned. They can be divided into three classes based on deduced amino acid sequence, protein size, phylogenetic analysis, and gene structure: class I, including ARF1, -2, and -3; class II, including ARF4 and -5; and class III, including ARF6 (2,3). ARFs have recently been recognized as regulators of intracellular vesicular transport. They are essential molecules for coated vesicle formation in the Golgi complex (4,5) and have also been im-plicated in vesicle transport between endoplasmic reticulum and Golgi (6,7) and in nuclear vesicle fusion (8). More recently, ARFs have been shown to activate phospholipase D (9,10).
Like other members of the Ras superfamily, ARF proteins transmit signals to downstream effectors in a cyclical and guanine nucleotide-dependent manner. Conformational differences between their GDP-and GTP-bound forms determine their interaction with regulatory proteins, namely a guanine nucleotide exchange protein (GEP) and a GTPase-activating protein (GAP). Class I ARF-directed GEP has been purified (11), and ARF1-directed GAP has also been purified and cloned (12,13). In contrast, the downstream effectors of ARF remain to be clarified. An ARF-responsive phospholipase D has been purified to a high degree from porcine brain (14). More recently, a form of mammalian phospholipase D has been cloned, and the recombinant enzyme expressed in Sf9 cells has been shown to be activated by recombinant ARF1 (15). Although this does not prove direct interaction between ARF and phospholipase D, it strongly suggests that it is a downstream effector. Nevertheless, it is still unclear if phospholipase D mediates ARF signals to initiate coated vesicle formation.
A point mutation of ARF1 at Gln 71 with Leu or Ile (Q71L or Q71I) slows the rate of GTP hydrolysis and therefore makes the mutant constitutively active (16). In an attempt to identify a direct downstream effector of ARF, we have employed the Q71L mutant of ARF3, which is 96% identical to ARF1, to screen a cDNA library using the yeast two-hybrid system. We have identified two novel proteins, termed arfaptins 2 1 and 2, which interact with class I ARFs only in their GTP-bound conformation. Arfaptin 1 was shown to be recruited to Golgi membranes by GTP-bound ARF.
Plasmids and cDNA Library-The ARF cDNAs used in the twohybrid screening and interaction assays were polymerase chain reaction-amplified from human ARF3 cDNA as a template (kindly provided by J. Moss, National Institutes of Health). Full-length wild type ARF3 cDNA was generated using 5Ј (sense) oligonucleotide primer A (5Ј-CG GAA TTC ATG GGC AAT ATC TTT GGA AAC CTT CTC) and 3Ј (antisense) oligonucleotide primer B (3Ј-CG GGA TCC TCA CTT CTT GTT TTT GAG CTG ATT GGC C). Substitution of Gln 71 with Leu (Q71L) was introduced by synthesizing the N-terminal half (corresponding to amino acids 1-71) and C-terminal half (corresponding to amino acids 71-181). The N-terminal half was generated using 5Ј * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The nucleotide sequence(s) reported in this paper has been submitted to the GenBank TM /EBI Data Bank with accession number(s) U52521 (arfaptin 1) and U52522 (arfaptin 2).
2 Apto (␣) is Greek for "I bind to." primer A and mutagenic 3Ј primer C (3Ј-G CTC TAG ACC ACC CAC ATC CCA CAC TGT AAA GCT; Leu 71 is underlined). The C-terminal half was generated using mutagenic 5Ј primer D (5Ј-GCT CTA GAC AAG ATT CGA CCC CTC TGG AGA CA; Leu 71 is underlined) and 3Ј primer B. The fragments were digested with XbaI and ligated to produce ARF3⅐Q71L. Double mutant ARF3⅐G2A⅐Q71L was also generated using N-terminal (amino acids 1-71) and C-terminal (amino acids 71-181) fragments. The N-terminal half was synthesized using mutagenic 5Ј primer E (5Ј-CG GAA TTC ATG GCC AAT ATC TTT GGA AAC CTT CTC; Ala 2 is underlined) and 3Ј primer C. The C-terminal half was the same as described above. The fragments were digested with XbaI and ligated. Wild type and mutated ARFs were subcloned into the EcoRI/ BamHI sites of pGBT9 yeast/Escherichia coli shuttle vector (Clontech) to generate fusions with the GAL4 DNA binding domain (pGBT9-ARF3, pGBT9-ARF3⅐Q71L, and pGBT9-ARF3⅐G2A⅐Q71L). All constructs were confirmed by DNA sequencing. For the preparation of the HL60 cDNA library (custom made by Clontech), total RNA was isolated from undifferentiated HL60 cells by the acid guanidium thiocyanate-phenol-chloroform extraction method (18), and poly(A) ϩ RNA was purified through an oligo(dT)-cellulose column (19). Double-stranded cDNA was made using both oligo(dT) and random primers and introduced into pGAD10 yeast/E. coli shuttle vector using EcoRI-NotI-SalI adaptor to generate fusions with GAL4 activation domain.
Two-hybrid Screening-Two-hybrid screening was performed using a Matchmaker Two-Hybrid System (Clontech) according to the instructions provided by the manufacturer. The yeast reporter strain HF7c was transformed sequentially with pGBT9-ARF3⅐Q71L and then with the HL60 cDNA library using a lithium acetate-based method. The double transformants were grown on SD agar medium lacking Trp, Leu, and His for 5 days at 30°C before positive colonies were picked, restreaked onto triple minus plates, and assayed for the LacZ phenotype. Plasmids containing GAL4-activation domain library fusion were isolated from HIS3 and LacZ positive colonies and used for further investigations.
Northern Blot Analysis-HL60 poly(A) ϩ RNA (2 g) isolated as described above was separated by a denaturing formaldehyde 1% agarose gel, transferred to a nitrocellulose membrane (Durarose-UV, Stratagene), and hybridized with a 32 P-labeled probe using a standard procedure (19). To examine the expression in various human tissues, a Northern blot filter with 2 g of poly(A) ϩ RNA from different human tissues (Clontech) was hybridized with a 32 P-labeled probe according to the manufacturer's recommendations. A 1.2-kb cDNA insert of arfaptin 1 and a 1.6-kb cDNA insert of arfaptin 2 were 32 P-labeled using a random primer DNA labeling kit (Bio-Rad) and used as probes.
Recombinant Arfaptin 1 and Arfaptin 2-Recombinant arfaptin 1 and arfaptin 2 were produced as glutathione S-transferase (GST) fusion proteins. For GST-arfaptin 1, a short 5Ј-fragment from nucleotide 130 to 180 (internal SpeI site) with a SmaI site at the 5Ј-end was synthesized by annealing sense and antisense oligonucleotides in 50 mM Tris-HCl (pH 7.5), 10 mM MgCl 2 , 5 mM dithiothreitol, and 0.1 mM EDTA at 37°C for 1 h. The fragment was ligated to a SpeI/EcoRI fragment (nucleotides 177-1234) at the SpeI site. The resulting SmaI/EcoRI fragment was subcloned into the same sites of GST fusion vector pGEX-2T (Pharmacia Biotech Inc.) to generate pGEX-arfaptin 1. For GST-arfaptin 2, a short 5Ј-fragment stretching from the predicted translation initiation codon (nucleotide 68) to the internal BamHI site (nucleotide 82) with an EcoRI site at the 5Ј-end was synthesized by annealing sense and antisense oligonucleotides as described above and ligated to a BamHI/NotI fragment (nucleotides 78 -1654) at the BamHI site. The resulting EcoRI/NotI fragment was subcloned into the same sites of pGEX-4T vector (Pharmacia) to generate pGEX-arfaptin 2.
E. coli strain BL21 was transformed with pGEX-arfaptin 1 or pGEXarfaptin 2. Transformed cells were grown at 37°C to A 600 ϭ 0.8, and protein expression was induced with 0.1 mM isopropyl-1-thio-␤-D-galactopyranoside for 3 h at 27°C. The cells were resuspended in phosphatebuffered saline containing 0.5 mM EGTA, 0.5 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, 10 g/ml leupeptin, and 10 g/ml antipain and lysed by sonication. The lysate was incubated with glutathione-Sepharose beads (Pharmacia) for 30 min at room temperature. The beads were then washed with phosphate-buffered saline, and the GST fusion protein was eluted by 10 mM glutathione in 50 mM Tris-HCl (pH 8.0). Recombinant arfaptin 1 without the GST moiety was obtained by treating GST-arfaptin 1-immobilized glutathione-Sepharose beads with thrombin (4 g/ml) in the buffer consisting of 50 mM Tris-HCl (pH 8.0), 150 mM NaCl, 2.5 mM CaCl 2 , and 14.2 mM ␤-mercaptoethanol for 40 min at room temperature. The supernatant was later subjected to benzamidine-Sepharose to remove thrombin.
Recombinant ARFs-Human ARF3 cDNA was polymerase chain reaction-amplified with the NcoI site upstream of the initiator methionine codon and the HindIII site downstream of the termination codon and subcloned into pKK233-2 expression vector (Clontech) to generate pKK233-2-ARF3. E. coli strain JM109 was transformed with pKK233-2-ARF3, and protein expression was induced with 2 mM isopropyl-1thio-␤-D-galactopyranoside for 5 h at 37°C. Recombinant ARF3 was purified by successive column chromatography on DEAE-Sephacel and then Sephadex 75 as described by Weiss et al. (20) with slight modifications.
For production of myristoylated ARF3, JM109 bacteria were cotransformed with pKK233-2-ARF3 and pBB131 (yeast N-myristoyltransferase, kindly provided by Dr. J. I. Gordon, Washington University; see Ref. 21) and selected for both ampicillin and kanamycin resistance. Transformed cells were grown at 37°C to A 600 ϭ 0.6, and protein expression was induced with 1 mM isopropyl-1-thio-␤-D-galactopyranoside in the presence of 200 M sodium myristate for 3 h at 27°C. The lower temperature (27°C) was employed to increase the efficiency of N-myristoylation (22). The myristoylated ARF3 was partially purified (ϳ20%) employing the same procedure as described for nonmyristoylated ARF3. Efficiency of myristoylation was approximately 50% as judged by the change in mobility on SDS-PAGE.
Recombinant nonmyristoylated human ARF1, ARF5, and ARF6 were kindly provided by W. Patton and J. Moss (National Institutes of Health).
Antisera-Recombinant arfaptin 1 obtained by thrombin cleavage of GST-arfaptin 1 was used as an immunogen to raise polyclonal antisera in rabbits. In Western blotting, anti-arfaptin 1 was used at a 1:2000 dilution. Anti-sARFII, which recognizes ARF1 and ARF3 equally (23), was a kind gift of Dr. J. Moss and was used at a 1:1000 dilution in Western blotting.
In Vitro Binding Studies-Nucleotide-free ARF was prepared by dialysis of recombinant ARF against 7 M urea as described (20) and incubated at 0.5 M with or without guanosine 5Ј-O-(3-thiotriphosphate) (GTP␥S) (5 M) in a reaction mixture consisting of 25 mM Hepes (pH 7.4), 100 mM NaCl, 25 mM KCl, 2.5 mM MgCl 2 , 1 mM EDTA, 0.1% Triton X-100, 1 mM dithiothreitol, and 0.3 mg/ml bovine serum albumin at 30°C for 30 min. Triton X-100 was included to increase GTP␥S loading and to stabilize the GTP␥S-ARF formed (25). GDP-bound ARF was prepared by incubating ARF (not treated with urea) with GDP (5 M) in the same reaction mixture. GST-, GST-arfaptin 1-, and GSTarfaptin 2-immobilized glutathione-Sepharose beads prepared as described above were equilibrated with washing buffer A (25 mM Tris-HCl, pH 8.0, 100 mM NaCl, 10 mM MgCl 2 , 1 mM dithiothreitol). Fifty l of nucleotide-preloaded ARF mixture was then incubated with 25 l of GST-, GST-arfaptin 1-, or GST-arfaptin 2-immobilized beads for 20 min at 4°C with gentle rocking. The beads were then collected by pulse centrifugation in a microcentrifuge, washed four times with 0.5 ml of washing buffer A, and resuspended in SDS-sample buffer. ARF associated with the beads was detected by 14% SDS-PAGE followed by immunoblotting using anti-sARFII antibody.
To determine the affinity of arfaptin 1 for ARF-GTP␥S, ARF3 was loaded with 10 M [ 35 S]GTP␥S (4 Ci/nmol) as described above. The concentration of ARF3-[ 35 S]GTP␥S was determined by a nitrocellulose filter binding assay, based on the specific activity of [ 35 S]GTP␥S. GSTarfaptin 1 was then incubated with ARF-[ 35 S]GTP␥S at different concentrations at 30°C for 30 min. Anti-arfaptin 1 antiserum and protein A-Sepharose beads were then added, and the mixture was incubated at 4°C for 60 min. The beads were then washed three times in washing buffer A and counted to determine ARF-[ 35 S]GTP␥S binding. The data were analyzed by Eadie-Hofstee plot to yield K d values.
When the relative affinity of each ARF isozyme to arfaptin 1 or arfaptin 2 was examined, nucleotide-free ARF was preloaded with 5 M [ 35 S]GTP␥S (4 Ci/nmol) as described above. Thereafter, aliquots were determined for [ 35 S]GTP␥S binding by the nitrocellulose filter assay. GTP␥S-bound ARF was stable on ice for at least 5 h. Each mixture was adjusted to contain the same amount of [ 35 S]GTP␥S-ARF and used for the interaction experiment described above. After washing, the beads were mixed with 10 ml of Ready Safe liquid scintillation mixture (Beckman), and the radioactivity was counted. The stoichiometry of [ 35 S]GTP␥S binding for ARF1, ARF3, ARF5, and ARF6 was 0.26, 0.28, 0.08, and 0.21, respectively (means of the three determinations).
Arfaptin 1 Translocation Studies-Highly concentrated cytosol was prepared from HL60 cells as described by Malhotra et al. (26) except that the cells were homogenized by sonication. Prior to use, the cytosol was stored at Ϫ80°C and thawed and centrifuged at 100,000 ϫ g for 1 h at 4°C to remove any aggregated proteins. Golgi-enriched membrane fractions were prepared from rat liver by sucrose gradient centrifugation according to Malhotra et al. (26). Membranes were collected at the 0.5 M sucrose/1 M sucrose interface and stored at Ϫ80°C. Prior to use, membranes were washed with 10 mM Tris-HCl (pH 7.4).
Golgi membranes were incubated with HL60 cytosol or comparable amounts of gel-filtered cytosol under conditions previously defined (27) with slight modifications. Briefly, Golgi membranes (7.5 g of protein) and a saturating concentration of cytosol protein (300 g) were incubated with or without GTP␥S (25 M) at 37°C for 10 min in the reaction mixture (100 l) consisting of 25 mM Hepes-KOH (pH 7.0), 125 mM KCl, 2.5 mM MgCl 2 , 1 mM dithiothreitol, 0.2 M sucrose, 1 mM ATP, 5 mM creatine phosphate, and 10 units/ml creatine kinase. When the effect of brefeldin A was examined, Golgi membranes were incubated with brefeldin A for 10 min at 37°C prior to the addition of cytosol and GTP␥S. After incubation, 90 l of the mixture was layered on 330 l of 25% (w/v) sucrose in washing buffer B, consisting of 25 mM Hepes-KOH (pH 7.0), 125 mM KCI, 2.5 mM MgCl 2 , and 1 mM dithiothreitol, and Golgi membranes were collected as pellets by centrifugation for 30 min at 14,000 rpm at 4°C in a microcentrifuge as described (28). The sucrose cushion greatly reduced the amounts of ARF and arfaptin 1 that were nonspecifically precipitated by centrifugation when Golgi membranes were omitted from the reaction mixture. The pellet was washed with the washing buffer B containing 0.2 M sucrose and resuspended in SDS-sample buffer. ARF and arfaptin 1 associated with the membranes were separated on 14% SDS-PAGE and detected by immunoblotting using anti-sARFII and anti-arfaptin 1, respectively. In some experiments, the immunoblots were analyzed by densitometry (Apple Macintosh, One Scanner).
Expression of Epitope-tagged arfaptin 1 and Indirect Immunofluorescent Microscopy-An epitope sequence corresponding to FLAG octapeptide (Asp-Tyr-Lys-Asp-Asp-Asp-Asp-Lys), which is specifically recognized by anti-FLAG M2 monoclonal antibody (Eastman Kodak Co.), was introduced at the N terminus of arfaptin 1. A 5Ј-fragment from nucleotides 131-180 was synthesized with a HindIII site upstream and the FLAG encoding sequence downstream of the translation initiation codon by annealing sense and antisense oligonucleotides as described above. This fragment was ligated to the SpeI/EcoRI fragment (base pairs 177-1234) at the SpeI site. The resulting HindIII/EcoRI fragment was subcloned into the same sites of the mammalian expression vector pcDNA3 (Invitrogen) to generate pcDNA3-FLAG⅐arfaptin 1.
COS7 cells grown to 50 -80% confluency in 100-mm dishes were transiently transfected by electroporation at 220 V and 960 microfarads using 15 g of pcDNA3-FLAG⅐arfaptin 1 and grown on glass coverslips. Two days after transfection, cells were fixed using 3.7% formaldehyde and then incubated with mouse anti-FLAG M2 monoclonal antibody (20 g/ml) followed by Texas red-conjugated horse anti-mouse IgG (1:150 dilution, Vector) according to standard procedures (29). Expression of exogenous arfaptin 1 was detected by fluorescent microscopy. Transfection efficiency was 10 -20%.
Miscellaneous Procedures-SDS-PAGE and Western blotting were performed as described (17) except that an ECL kit (Amersham Corp.) was used for detection. Protein concentrations were determined using the Coomassie Plus protein assay reagent (Pierce) with bovine serum albumin as a standard. DNA sequencing was performed using a Sequenase II DNA sequencing kit (U.S. Biochemical Corp.). HL60 cytosol and crude membrane fractions used in Fig. 2 were prepared as follows. HL60 cells were homogenized by sonication in a buffer consisting of 50 mM Hepes (pH 7.2), 100 mM KCl, 5 mM NaCl, 0.5 mM EGTA, 3.5 mM MgCl 2 , 1 mM phenylmethylsulfonyl fluoride, 10 g/ml leupeptin, and 10 g/ml antipain. After unbroken cells and nuclei were removed by centrifugation at 500 ϫ g for 10 min, crude membranes and cytosol were obtained by centrifugation at 100,000 ϫ g for 90 min. Membranes were washed once with the sonication buffer.

Identification of an ARF3-interacting Protein by Yeast Two-
hybrid Screening-To identify molecules that act as downstream effectors of ARF, we employed the yeast two-hybrid system and screened an HL60 cDNA library using a constitutively active mutant of ARF3 (ARF3⅐Q71L) (16, 29) as a target. When approximately 2 ϫ 10 6 double transformants were screened and selected for histidine prototrophy, three positive colonies were obtained. These colonies were assayed for activation of second reporter gene LacZ by ␤-galactosidase filter assay, and all were revealed to be positive. Partial sequencing and restriction mapping of the cDNA inserts of these library clones revealed that two clones (clones 1 and 3) had the same cDNA insert (1.2 kb), while another clone (number 2) had a insert (1.6 kb) distinct from clones 1 and 3. To eliminate false positives, further genetic assays were conducted (Table I). Neither clone 1 nor clone 2 by itself was capable of activating ␤-galactosidase activity, indicating that they do not contain a latent transcriptional activator. The GAL4-binding domain alone or human lamin C, which is a protein unrelated to ARF, did not interact with either clone 1 or 2. These observations rule out the possibility that clone 1 and clone 2 nonspecifically interact with other proteins.
ARFs are modified by myristoylation at Gly 2 through the action of an N-myristoyltransferase in vivo (30). However, ARF3 expressed as a fusion protein with a GAL4-binding domain cannot be modified, because N-myristoyltransferase can only act at the N terminus (21). Therefore, it is presumed that the interaction of ARF3⅐Q71L with clone 1 or clone 2 in yeast is independent of myristoylation. This was confirmed by studies with the double mutant ARF3⅐G2A⅐Q71L, which has an additional mutation at Gly 2 to abolish myristoylation, which showed that this also interacted with clones 1 and 2 with apparently the same affinity as ARF3⅐Q71L (Table I). Interestingly, the two-hybrid assay indicated that neither clone 1 nor 2 interacted with wild-type ARF3, suggesting that both clones interact only with the active, GTP-bound form of ARF (Table I).
Clone 1 was revealed to contain a full-length open reading frame encoding a protein of 341 amino acids with a calculated molecular mass of 38,596 Da (Fig. 1). The translation initiation codon (nucleotides 131-133) was surrounded by a consensus Kozak sequence (31). Furthermore, an antibody raised against a recombinant protein encoded by this open reading frame immunoprecipitated a protein from HL60 cell lysate whose migration on SDS-PAGE was similar to that of the recombinant protein (ϳ44 kDa) (Fig. 2). Because the recombinant protein was obtained by thrombin cleavage of the GST fusion protein, it had four additional amino acids at the N terminus. This could partly explain the slight difference in migration. We have named the protein encoded by clone 1 arfaptin 1. 2 When cytosol and crude membranes from HL60 cells were subjected to immunoblotting, arfaptin 1 was detected predominantly in cytosol (Fig. 2).
Interestingly, sequence analysis of clone 2 revealed the presence of a potential open reading frame whose size was exactly the same as that of arfaptin 1. The nucleotide sequence of this open reading frame was 57% identical to that of arfaptin 1, and the deduced amino acid sequence was 60% identical and 81% homologous when consensus substitutions were included. The homology was pronounced at the C-terminal half of the molecule (Fig. 3). Thus we have tentatively named the protein encoded by the putative open reading frame of clone 2 arfaptin 2.
The predicted amino acid sequences of arfaptins 1 and 2 were used to search a protein data base (the BLAST network service at the National Center for Biotechnology Information (NCBI)). However, no homologous proteins of known function were found. 3 The highest identity score was found with a Caenorhabditis elegans hypothetical 35-kDa protein of unknown function (CEF54C8.7, EMBL/GenBank TM accession number Z22178). The primary structure analysis indicated that arfaptin 1 and arfaptin 2 were hydrophilic proteins with a few minor hydrophobic regions. Both proteins had several potential phosphorylation sites for protein kinase C, 4 protein kinase A, and casein kinase II. Interestingly, arfaptin 2 had a periodic repe- 3 After submission of this paper, a re-search revealed POR1, a Racbinding protein that is identical to arfaptin 2, but lacks the first 38 amino acids. It has been implicated in Rac-induced membrane ruffling in fibroblasts (54,55). 4 Preliminary experiments indicated that both GST-arfaptin 1 and GST-arfaptin 2 but not GST alone were phosphorylated by purified protein kinase C (Promega) but only with poor stoichiometry (25-35% for GST-arfaptin 1, 15-25% for GST-arfaptin 2). tition of five leucine residues at every seventh position (leucine zipper) (residues 150, 157, 164, 171, and 178 in Fig. 3). Such a motif, with a highly positively charged region immediately adjacent to its N-terminal side, forms a DNA-binding domain (33). Arfaptin 2, however, lacks such a positively charged region. Arfaptin 1 also has a leucine zipper motif, but this is interrupted at residue 157 by replacement with methionine (Fig. 3).
Northern blot analysis indicated that arfaptins 1 and 2 were ubiquitously expressed as mRNAs of approximately 3.4 and 2.1 kb, respectively, in various human tissues as well as HL60 cells, ruling out the possibility that they are cancer cell-specific products. Both arfaptins 1 and 2 were expressed at a relatively high level in liver, pancreas, and placenta. In addition, arfaptin 1 was expressed at a relatively high level in skeletal muscle and heart (Fig. 4).
Arfaptins 1 and 2 Bind Directly to the GTP-bound Form of Class 1 ARFs in Vitro-As described above, the two-hybrid interaction assay suggested that arfaptins 1 and 2 bind ARF3 only in its GTP-bound conformation and that the binding was independent of myristoylation of ARF. To confirm this bio-chemically, in vitro interactions of nonmyristoylated ARF3 with arfaptins 1 and 2 were examined using recombinant proteins (Fig. 5). Since recombinant ARF binds GTP or GTP␥S with very low stoichiometry due to its tightly bound GDP, we utilized nucleotide-free ARF to prepare GTP␥S-bound ARF. Under the conditions employed, the stoichiometry of GTP␥S binding of ARF3 was 0.28. When GST-arfaptin 1-immobilized glutathione-Sepharose beads were incubated with GTP␥S-ARF3, GDP-ARF3, or nucleotide-free ARF3, only GTP␥S-ARF was associated with the beads, while very little nucleotide-free ARF and no GDP-ARF were detected. In addition, GTP␥S-ARF3 did not bind to GST-immobilized beads, indicating that the binding of GTP␥S-ARF3 to the beads was through arfaptin 1. The binding, however, was not quantitative (see below). These results indicated that arfaptin 1 was capable of interacting directly with nonmyristoylated GTP-ARF3 but not GDP-ARF3, consistent with the results of the two-hybrid interaction assay. Qualitatively similar results were obtained with arfaptin 2 and ARF3 (Fig. 5). The affinity of arfaptin 1 for GTP␥S-ARF3 was measured as described under "Materials and Methods." The mean K d from two experiments was 1.4 ϫ 10 Ϫ7 M.
The binding of arfaptins 1 and 2 to other ARF isoforms was examined using recombinant nonmyristoylated ARFs preloaded with [ 35 S]GTP␥S (Table II). ARF1, which is 96% identical to ARF3 in amino acid sequence, was used as another representative of class I ARF, and ARF5 and ARF6 were used as representatives of class II and III, respectively. After preloading of nucleotide-free ARF with [ 35 S]GTP␥S, equal amounts of each GTP␥S-bound ARF were incubated with GSTarfaptin 1-or GST-arfaptin 2-immobilized beads. Both arfaptins 1 and 2 bound ARF1 with the highest affinity. On the other hand, they bound much less ARF5 and very little ARF6. It was confirmed that both arfaptins 1 and 2 bound ARF1 only in its GTP␥S-bound form by employing the same experiments described for ARF3 in Fig. 5 (not shown). These results indicate that arfaptins 1 and 2 interact preferentially with class I ARFs, especially ARF1.

Arfaptin 1 Is Recruited by ARF from Cytosol to Golgi Membranes in a GTP␥S-dependent
Manner-Among the subfamilies of Ras low molecular weight GTP-binding proteins, ARF is the clearest example of a GTPase whose GTP binding and hydrolysis cycle appears to be strictly coupled to a membranecytosol localization cycle. Inactive GDP-bound ARF is cytosolic, whereas active GTP-bound ARF is associated with membranes, especially with the Golgi apparatus (24). As described above, arfaptin 1 is present in the cytosolic fraction and specifically binds GTP-bound ARF. Therefore, we examined the possibility that arfaptin 1 was translocated from cytosol to Golgi membranes in association with GTP-bound ARF (Fig. 6). Golgienriched membranes from rat liver were incubated with HL60 cytosol in the presence or absence of GTP␥S and then the membranes were collected by centrifugation. In the presence of GTP␥S, arfaptin 1 and ARF were detected in the pellet (lane 3), whereas in the absence of GTP␥S very little arfaptin 1 or ARF was detected (lane 1). When Golgi membranes were omitted from the reaction mixture, arfaptin 1 and ARF were barely detectable in the pellet (lane 2), ruling out the possibility that arfaptin 1 and ARF were precipitated due to nonspecific aggregation. Brefeldin A has been shown to disrupt Golgi membranes and inhibit the binding of ARF to Golgi membranes (34). When Golgi membranes were treated with brefeldin A (40 g/ml) prior to the addition of cytosol and GTP␥S, translocation of arfaptin 1 and ARF was inhibited (lane 4), suggesting that the association of arfaptin 1 with Golgi membranes depends on ARF. When the results of three experiments were analyzed by densitometry, the inhibition of arfaptin translocation (32 Ϯ 2%) was similar to that of ARF (38 Ϯ 7%).
To confirm that translocation of arfaptin 1 is ARF-dependent, we utilized ARF-depleted cytosol (Fig. 7B). To obtain this, HL60 cytosol was subjected to Sephacryl S-200 gel filtration column. Since arfaptin 1 was present in a high molecular weight complex, it could easily be separated from ARF (Fig.  7A). When Golgi membranes were incubated with ARF-depleted cytosol (fractions 22-28) alone, only a trace amount of arfaptin 1 was translocated to membranes even in the presence of GTP␥S. This was probably caused by the tiny amount of aggregated ARF that was co-eluted with arfaptin 1. On the other hand, when fractions containing native ARF (fractions 35-40 in Fig. 7A) were included in the above mixture, GTP␥Sdependent translocation of arfaptin 1 to the Golgi membranes was restored, and this was well correlated with translocation of ARF. When partially purified recombinant myristoylated ARF3 was used instead of native ARF fraction, the same results were observed. These results indicate that translocation of arfaptin 1 to Golgi membranes requires GTP-bound ARF. It should be noted that although HL60 cytosol contained both ARF1 (lower band of doublet, Fig. 7A) and ARF3 (upper band of doublet, Fig. 7A), translocation of native ARF3 was much less than that of native ARF1. This differential interaction of ARF1 and ARF3 with Golgi membranes is consistent with the previous observations (23). Thus, it appears that translocation of arfaptin 1 to Golgi membranes is mediated mainly by ARF1.
Arfaptin 1 Is Not a Component of Coatomer-One of the well characterized functions of ARF is the recruitment of cytosolic coat proteins to Golgi membranes, which is thought to play an important role in vesicle trafficking (35). Coat proteins are preassembled in a cytosolic complex, coatomer, which has an apparent molecular mass of ϳ800 kDa (36). As described above, arfaptin 1 was present in cytosol as a high molecular weight complex that was recruited to Golgi membranes by activated ARF (Fig. 7). These findings raised the possibility that arfaptin 1 was a constituent of coatomer. To examine this, HL60 cytosol was fractionated through a Sephacryl S-400 gel filtration column. Arfaptin 1, however, was clearly separated from ␤-COP, one of the components of coatomer, since this was eluted at around 800 kDa, whereas arfaptin 1 was eluted at around 450  1, 4, and 5) or without GTP␥S (lanes 3  and 7), and untreated ARF3 (0.5 g) was incubated with 5 M GDP (lanes 2 and 6) in 50-l reaction mixtures as described under "Materials and Methods." Subsequently, these mixtures were incubated with GSTarfaptin 1-, or GST-arfaptin 2-immobilized glutathione-Sepharose beads. After washing, ARF3 associated with the beads was detected by SDS-PAGE followed by immunoblotting. The beads (25 l) contained 12 g of GST-arfaptin 1, 12 g of GST-arfaptin 2, and 7 g of GST. NF⅐ARF3, nucleotide-free ARF3.  kDa (not shown). These results indicate that arfaptin 1 is not a constituent of coatomer. It is, however, currently uncertain whether arfaptin 1 is present in such an 800-kDa complex in vivo or if this complex is a result of nonspecific aggregation during the preparation of HL60 cytosol. Arfaptin 1 Is Localized to Golgi in Transfected Cells-To investigate the in vivo localization of arfaptin 1, epitope-tagged arfaptin 1 was transiently expressed in COS7 cells and detected by immunostaining (Fig. 8). Indirect immunofluorescent microscopy showed perinuclear staining of arfaptin 1. Treatment of cells with brefeldin A caused the redistribution of arfaptin 1 into the cytoplasm, suggesting that arfaptin 1 is localized to the Golgi complex in intact cells. DISCUSSION In the present study, we have identified a novel protein arfaptin 1, whose mRNA is ubiquitously expressed in human tissues. Arfaptin 1 is a hydrophilic protein with a few minor hydrophobic regions and is present in cytosol. The calculated mass is 38,596 Da, and it migrates on SDS-PAGE at approximately 44 kDa. Recombinant arfaptin 1 binds preferentially class I ARFs, especially ARF1, in their GTP-bound state and not at all in their GDP-bound form ( Fig. 5 and Table II). Thus, arfaptin 1 is likely to be a direct target protein for class I ARFs.
We have also obtained another clone encoding a putative protein, arfaptin 2, which consists of the same number of amino acids as arfaptin 1. The mRNA of arfaptin 2 is also ubiquitously expressed, and the deduced amino acid sequence of arfaptin 2 is 60% identical and 81% homologous to that of arfaptin 1. Furthermore, recombinant arfaptin 2 has very similar characteristics for ARF binding as arfaptin 1; it binds preferentially class I ARFs, especially ARF1, but only in their GTP-bound conformation, implicating arfaptin 2 as another target protein for class I ARFs.
To date, six distinct but highly homologous mammalian ARF proteins have been identified (2,3). Our data concerning the relative affinity of arfaptin 1 for different ARF isoforms suggest that ARF domains that may be important for the interaction with arfaptin 1. Among the isoforms examined, arfaptin 1 binds ARF1 with the highest affinity and ARF3 with less affinity. It binds much less ARF5 and very little ARF6. The order of affinity for different ARF isoforms is correlated with their structural homology; ARF3, -5, and -6 differ from ARF1 in amino acid sequence by 4, 20, and 32%, respectively (2). It is surprising, however, that the very small difference in the amino acid sequences of ARF1 and ARF3 (7 of 181 amino acids) results in an approximately 3-fold difference in binding affinity for arfaptin 1. The differences between ARF1 and ARF3 are attributed exclusively to their N and C termini; 4 amino acids are different in the N-terminal 13 amino acids and 3 amino acids in the C-terminal 8 amino acids (37). These facts suggest that the N and/or C termini may be important domains of ARF1 for interaction with arfaptin 1.
Several lines of evidence have pointed out the significance of the N terminus of ARF for its functions. A synthetic peptide derived from the N terminus of ARF1 inhibits the cofactor activity of ARF for cholera toxin ADP-ribosyltransferase activity and ARF-dependent intra-Golgi protein transport in vitro (38). Experiments using a N-terminal deletion mutant of ARF1 have also shown that the N terminus of ARF1 is essential for FIG. 6. Recruitment of arfaptin 1 to Golgi membranes by GTP␥S and its inhibition by brefeldin A. Golgi-enriched membranes (7.5 g) were preincubated for 10 min at 37°C in the absence (lanes 1 and 3) or in the presence of brefeldin A (40 g/ml, lane 4). Subsequently, cytosol (300 g) and/or GTP␥S (25 M) were added, and the incubation was continued for additional 10 min. Golgi membranes were collected by centrifugation through a 25% sucrose cushion, washed, and resuspended in SDS-sample buffer. Arfaptin 1 and ARF associated with the membranes were detected by SDS-PAGE and immunoblotting using anti-arfaptin 1 (upper panel) and anti-sARFII (lower panel), respectively. To examine if there was nonspecific precipitation of arfaptin 1 or ARF, cytosol alone was incubated with GTP␥S (lane 2).

FIG. 7. Recruitment of arfaptin 1 to Golgi membranes requires ARF.
A, arfaptin 1 in HL60 cytosol was separated from ARF by gel filtration. HL60 cytosol (0.75 ml, 13 mg of protein) was applied to a Sephacryl S-200 column (0.9 ϫ 39 cm) and eluted with a buffer consisting of 25 mM Tris-HCl (pH 8.0), 50 mM KCl, 1 g/ml leupeptin, and 1 g/ml antipain. Fractions (0.5 ml) were collected and analyzed by SDS-PAGE and immunoblotting using anti-arfaptin 1 and anti-sARFII. Arrowheads indicate the elution positions of molecular size markers, namely blue dextran, aldolase (158 kDa), bovine serum albumin (67 kDa), and chymotrypsinogen A (25 kDa). Fractions 22-28 and fractions 35-40 were collected, concentrated, and used as ARF-depleted cytosol and native ARF fraction, respectively, B, Golgi membranes (7.5 g) were mixed with ARF-depleted cytosol alone (225 g, lanes 1 and 2), ARF-depleted cytosol (225 g) plus native ARF fraction (17 g, lanes 3  and 4), or ARF-depleted cytosol (225 g) plus partially purified myristoylated ARF3 (8 g, lanes 5 and 6) and then incubated for 10 min at 37°C in the presence or absence of GTP␥S (25 M). Arfaptin 1 and ARF associated with membranes were detected as described in Fig. 6. the interaction with GAP and for the cofactor activity (39). More recently, analysis of the GTP-dependent conformational change of ARF1 has led to the proposal that the N terminus of ARF1 is an effector domain (40). Thus the N terminus rather than C terminus of ARF might be more important for interaction with arfaptin 1. In addition to the N terminus itself, modification of the N terminus by myristoylation is also critical for ARF functions, because myristoylation confers an ability to associate with membranes (41) and is critical for the GTP-dependent conformational change (40). Although the present study indicates that arfaptin 1 interaction with ARF does not require myristoylation (Table I), the possibility that the affinity may be increased by myristoylation remains to be examined.
It is reported that different ARF isoforms have different affinities for Golgi membranes and other cellular membranes. Among the isoforms, it has been shown that ARF1 as well as ARF5 associate with Golgi membranes with high affinity and specificity in vitro, while ARF3 associates with the membranes with lower affinity and is distributed to other cellular membranes (23). Localization of ARF1 to the Golgi complex was confirmed in overexpression experiments, whereas ARF6, to which arfaptin 1 does not bind, was localized to the endosomal/ plasma membrane system (42). Another study has shown that ARF6 is uniquely localized to the plasma membrane of Chinese hamster ovary cells (43). Therefore, the preferential binding of arfaptin 1 to ARF1 shown by in vitro interaction experiments is in good agreement with the localization of arfaptin 1 in the Golgi complex shown by overexpression in COS cells in vivo (Fig. 8).
Accumulating evidence suggests that there is a cyclic localization of ARF between cytosol and membrane fractions. Inactive GDP-bound ARF present in the cytosol is activated by GEP in Golgi membranes or cytosol in a brefeldin A-sensitive manner and is translocated to Golgi membranes. Subsequently, GTP bound to ARF is hydrolyzed in the presence of GAP in Golgi membranes, and the resulting GDP-ARF is released to cytosol (13, 27, 44 -47). Our in vitro translocation data and in vivo transient expression experiments suggest that the GTPbound form of ARF serves as a membrane anchor for arfaptin 1 and that arfaptin 1 cycles between cytosol and Golgi membranes depending on the activity status of ARF. Such a relationship between a GTP-binding protein and its target, i.e. a GTP-binding protein that serves as regulatable membrane anchor for its target protein, has also been observed for other small GTP-binding proteins and their functional target proteins. Ras, when it is activated, translocates its cytosolic target Raf-1 protein kinase to the plasma membrane, where Raf-1 is activated and initiates a phosphorylation cascade (48). RhoA, which is involved in morphological events involving the actin cytoskeleton, recruits its target serine/threonine kinase ROK␣ to peripheral membranes (49). Rab3A and Rab3C have been suggested to recruit rabphilin-3A, a Rab3A target protein, to synaptic vesicle membranes (50). Rab5, a potent regulator of endocytic transport, recruits its target protein rabaptin-5 to early endosomes in a GTP-dependent manner (51). Therefore, it is possible that ARF, which belongs to another subfamily of Ras-related proteins, also has a cytosolic target protein (arfaptin) and recruits it to membrane fractions.
Arfaptin 2 has a leucine zipper structure in the middle of the molecule (Fig. 3). Arfaptin 1 also has a similar repetition of leucine residues, although one of them is replaced with methionine (Fig. 3). The leucine zipper structure provides hydrophobic faces through which zipper proteins interact to form dimers, which can be homodimers or heterodimers. The leucine zipper structure is often found in transcription factors, since the dimer can interact with DNA through a domain enriched with positively charged amino acids immediately adjacent to the N terminus of the zipper (33). Since arfaptin 1 and arfaptin 2 lack such a domain, they are unlikely to bind DNA, but they may still be able to form homo-or heterodimers. In fact, when HL60 cytosol was separated by gel filtration chromatography, arfaptin 1 was found as a high molecular weight complex (Fig. 6), raising the possibility that it may exist as an oligomer and/or as a complex with other proteins.
A major question of the present study is the physiological function of arfaptin 1. This is being explored in in vitro studies carried out in collaboration with the group of Dr. J. Moss (National Institutes of Health). These results will be reported elsewhere when complete, but initial findings are that recombinant arfaptin acts as an inhibitor of the in vitro action of ARF on phospholipase D and cholera toxin-catalyzed ADP-ribosyltransferase activity. However, it does not alter the binding of GTP␥S or GDP to ARF in the presence of GEP (11) or alter GTPase activity. Irrespective of these observations, the fact that arfaptin 1 only interacts with ARF liganded to GTP␥S, but not GDP, suggests that it may be an effector i.e. a physiological target of ARF whose function is presently unknown. An additional possibility is that arfaptin 1 is an adaptor protein that may require a third component before its function can be observed.
The relative affinity of arfaptin 2 for different ARF isoforms is strikingly similar to that of arfaptin 1 (Table II). Tissue Immunolocalization of arfaptin 1 in transfected COS7 cells. COS7 cells were transiently transfected with pcDNA3-FLAG⅐arfaptin 1. Two days after transfection, cells were treated without (panels 1 and 2) or with (panels 3 and 4) brefeldin A (5 g/ml) for 10 min at 37°C and stained with a monoclonal antibody specific for the FLAG epitope. Fluorescence microscopy (lanes 1 and 3) and phase contrast microscopy (lanes 2 and 4) are shown. distribution of arfaptin 2 is also similar to that of arfaptin 1 except for muscle (Fig. 4). Furthermore, arfaptin 2 may form a heterodimer with arfaptin 1 as discussed above, raising the possibility that arfaptin 2 might act cooperatively with arfaptin 1. Alternatively, arfaptin 1 may require other protein(s) beside arfaptin 2. As mentioned above, estimation of the size of arfaptin 1 using gel filtration (Fig. 6) suggests that it may associate with other proteins of unknown identity. GTP-bound ARF recruits not only arfaptin 1 to Golgi membranes, but also coat proteins including coatomer (27,35), p200 protein (52), and AP1 adaptins (53). Recently, ARF1-GAP has also been shown to be recruited to the Golgi complex by ARF (13), suggesting complex interactions among the various molecules. Thus, experimental conditions that are more physiological may be required to explore the function of arfaptin 1, e.g. those involving overexpression or "knockout" of arfaptin 1.