A Novel Mode of Action of an ArfGAP, AMAP2/PAG3/Papα, in Arf6 Function*

Previously we reported that AMAP2/PAG3/Papα/KIAA0400, a GTPase-activating protein (GAP), acts to antagonize Arf6 function when overexpressed, whereas it was shown to exhibit efficient GAP activities for other Arf isoforms in vitro. Here, we found that AMAP2, through its ArfGAP domain, binds to GTP-Arf6 but not to GDP-Arf6 or other Arfs irrespective of nucleotide status. The majority of AMAP2 was localized to intracellular tubulovesicular structures and redistributed to Arf6-enriched membrane areas upon Arf6 activation. In HeLa cells, Arf6 has been shown to be involved in the clathrin-independent endocytosis of Tac, but not the clathrin-dependent endocytosis of transferrin. We found that Arf6 silencing inhibited the internalization of Tac, but not transferrin, in HeLa cells. Internalization of Tac, but not transferrin, was also significantly inhibited by AMAP2 silencing and overexpression. AMAP2 was moreover found to bind to amphiphysin IIm, a component of the endocytic machinery, via its proline-rich domain. We propose that AMAP2 has dual mechanisms for its function; it exhibits efficient catalytic GAP activity for the class I and II Arfs and yet is involved in the cellular function of the class III Arf without immediate GAP activity. These dual mechanisms of AMAP2 may be important for the cellular function of GTP-Arf6.

Arfs belong to the Ras superfamily of small GTPases and play essential roles in intracellular membrane/vesicle trafficking (1,2). The family includes six isoforms of Arf and several Arf-like proteins in mammalian tissues. Arfs are subclassified into three classes by their structural similarities: class I (Arf1 to -3), class II (Arf4 and -5), and class III (Arf6). Arf1 primarily functions at perinuclear areas, mediating transport between the endoplasmic reticulum and Golgi by employing coat protein complexes, such as COP I (2). On the other hand, Arf6 primarily functions at the cell periphery, mediating membrane and cell surface receptor endocytosis and recycling (3)(4)(5)(6)(7)(8)(9). The function of the class II Arfs is largely unknown.
The structure and function of Arfs are significantly different from other members of the Ras superfamily GTPases. Arfs do not exhibit detectable intrinsic GTPase activity (10), unlike the other GTPases. Moreover, the binding sites for the GTPaseactivating proteins (GAPs) 1 and coatomers (effectors) do not overlap in Arf1, whereas the GAP binding domain and the effector binding domain largely overlap in the other GTPases such as Ras (11). Similarly, the primary structures of the GAP domain of ArfGAPs are significantly different from those of GAPs for other Ras-superfamily GTPases: the ArfGAP domains primarily consist of a zinc finger structure, which is not present in the other GAPs. The GTPase activity of other GTPases, such as Ras and Rho, is accelerated dramatically upon binding to their cognate GAPs. In contrast, ArfGAPs are found as components of GTP-Arf1-mediated priming complexes in the COP I system (12,13), and it has been suggested that GTP-Arf1 may act to recruit ArfGAP to the donor membrane during GTP-Arf1-mediated priming complex formation (11,14,15). Therefore, it has been hypothesized that some ArfGAPs can stay associated with the GTP-bound form of Arf at least for a while. In the case of the other small GTPases, the "arginine finger" model has been proposed, in which the corresponding GAP molecule supplies the critical arginine, which hydrogen-bonds to the ␥-phosphate of GTP bound to the GTPase and mediates the transition state (16). On the other hand, there are two contradictory models proposed for the Arf-ArfGAP complex. Goldberg has proposed that Arf1-ArfGAP1 constitutes only part of the usual protein binding interface and that the key residue(s) required for maximal rates of the GTP hydrolysis, like the "arginine finger," may not be provided by ArfGAP1, but further association of the coatomer with the Arf1-ArfGAP1 complex may be required for the maximal GTP hydrolysis activity (11). In contrast, structural analysis of Pap␤ suggests that the critical residues for hydrolysis of GTP bound to Arf1 may actually be provided by this ArfGAP molecule (17). This model is consistent with a result that Pap␤ exhibits efficient catalytic GAP activity for GTP-Arf1 in vitro (18). It should also be noted that there are several reports contradictory to Goldberg's model, suggesting that the coatomer may simply aid the productive association of Arf1 and ArfGAP1 (17,19,20). On the other hand, a study on the interface array of protein interactions of the small GTPases indicates that the protein-protein contacts in the Arf1-ArfGAP1 complex are significantly different from those found in other GAP and small GTPase complexes (21).
We have previously shown that PAG3/Pap␣/KIAA0400 (re-ferred to here as AMAP2) functions to antagonize Arf6 activity, which was determined by several different in vivo assay systems (22,23), including Fc␥ receptor-mediated phagocytosis in which the activity of Arf6, but not the other Arf isoforms, is essential (24). However, a biochemical assay revealed that AMAP2 exhibits phosphatidylinositol 4,5-bisphosphatedependent catalytic GAP activity for Arf1 (class I) and Arf5 (class II) but much less activity for Arf6 (class III) (18). DEF-1/ASAP1, a close isoform of AMAP2, has also been shown to exhibit GAP activity for Arf1, but not Arf6, in vivo, which was assessed by measuring cellular levels of GTP-loaded Arfs in DEF-1/ASAP1-overexpressing cells (25). AMAP2 has several protein interaction motifs other than the Src homology 3 (SH3) domain, to which Pyk2 and paxillin have been reported to bind (18,22). In this report, to further assess the isoform specificity and understand the mechanism of how AMAP2 is involved in cellular functions of Arfs, we identified other binding proteins to AMAP2. Consistent with our previous reports (22,23), our results described here again implicated AMAP2 in Arf6 function. Our results indicate that AMAP2 may function as an effector for Arf6 by stably binding to GTP-Arf6 rather than simply acting as its catalytic GAP enzyme. Such binding of AMAP2 to activated Arf6 may participate in priming complex formation required for Arf6-mediated endocytosis.
pGEX-GGA3 VHS-GAT encoding the VHS and GAT domains (amino acids 1-313) of GGA3 fused to glutathione S-transferase (GST) was a gift from P. Randazzo. cDNAs for the AMAP2 ArfGAP domain (amino acids 421-541), and the amphiphysin II SH3 domain (amino acids 494 -588) were each ligated into pGEX-2TK (Amersham Biosciences) to be fused in frame to the C terminus of GST. GST-tagged recombinant proteins were produced in Escherichia coli and purified using glutathione-beads, as described previously (22).
Cells and Transfection-HeLa and COS-7 cells were maintained at 37°C in Dulbecco's modified Eagle's medium (DMEM) with 10% fetal calf serum (Hyclone). 5 ϫ 10 5 cells in a 100-mm culture dish were transfected with 4 g of cDNA using PolyFect (Qiagen), according to the manufacturer's instructions, and incubated for 24 h prior to analysis, unless otherwise indicated.
Yeast Two-hybrid Screening-A DNA fragment of the AMAP2 proline-rich domain (PRD) (amino acids 744 -1220) was cloned into pBTM116. A HeLa cDNA library (Clontech) was screened using this fragment as a bait in a two-hybrid system using Saccharomyces cerevisiae L40 as a host, as previously described (30).
Protein Binding and Immunoblotting-Cell lysates were prepared with 1% Nonidet P-40 buffer (1% Nonidet P-40, 150 mM NaCl, 20 mM Tris-HCl, pH 7.4, 5 mM EDTA, 1 mM Na 3 VO 4 , 1 mM phenylmethylsulfonyl fluoride, 5 g/ml aprotinin, 2 g/ml leupeptin, and 3 g/ml pep-statin A) unless otherwise described. Protein binding assays were performed by precipitating GST fusion proteins using glutathione-Sepharose beads (Amersham Biosciences), as previously described (22). Immunoprecipitation was done using 2 mg of cell lysate and the appropriate antibodies bound to Protein G-Sepharose beads (Amersham Biosciences), as described (22). Proteins retained on the beads were subjected to immunoblotting analysis after separation by SDS-PAGE and detected by an enzyme-linked chemiluminescence method (Amersham Biosciences). 20 g of total cell lysates were included when necessary.
Guanine Nucleotide Status of Arf6 -The guanine nucleotide binding status of Arf6 bound to the AMAP2 ArfGAP domain was analyzed using a similar method as described previously (25,31). Briefly, 5 ϫ 10 5 COS-7 cells in a 100-mm dish were transfected with 1 g of pcDNA-Arf6-HA. Twenty-four h later, cells were washed and radiolabeled with 325 Ci/ml [ 32 P]orthophosphate (PerkinElmer Life Sciences) for 16 h. Cell lysates were then prepared using a lysis buffer (20 mM Tris-HCl, pH 8.0, 100 mM NaCl, 1 mM MgCl 2 , 1% Triton X-100, 0.05% cholate, 0.005% SDS, 1 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride, 1 mM benzamidine, 5 g/ml aprotinin, 5 g/ml leupeptin, 1 mM NaF, and 1 mM Na 3 VO 4 ) and incubated with 5 g each of GST-ArfGAP or GST alone, coupled to glutathione-beads, for 2 h at 4°C. After washing the beads four times with the above buffer, the radioactive substance retained on the beads was eluted with 2 M formic acid at 70°C for 3 min and subjected to thin layer chromatography using polyethyleneimine-cellulose plates (Merck). As a control, Arf6-HA was immunoprecipitated using an anti-HA monoclonal antibody and analyzed similarly.
Peptide Binding-Proline-rich peptides were synthesized on derivatized cellulose membranes using the SPOTs KIT (Genosys) according to the manufacturer's instructions and examined for their binding to GSTamphiphysin II SH3 using a method previously described (32). After blocking, membranes were hybridized with 0.1 g/ml of 32 P-labeled GST proteins at 4°C overnight and then washed three times with Hyb 75 buffer (20 mM HEPES, pH 7.7, 75 mM KCl, 0.1 mM EDTA, 2.5 mM MgCl 2 , 1 mM dithiothreitol, 0.05% Nonidet P-40) (32) containing 5% bovine serum albumin. Radioactivities retained on filters were measured using FLA-2000 and its attached software, MacBAS, version 2.5 (Fujifilm). Phosphorylation of the GST-amphiphysin II SH3 or GST (a product of pGEX-2TK) was performed using the catalytic subunit of cAMP-dependent kinase (Sigma) and [␥-32 P]ATP, according to the manufacturer's instructions. The specific radioactivity of the probes was about 5 ϫ 10 6 cpm/g.
Microscopy-Subcellular localization of proteins were analyzed by fixing cells in 3.7% paraformaldehyde and being permeabilized with 0.2% saponin, as previously described (22), unless otherwise indicated, using laser confocal microscopy (LSM 510, version 2.5; Carl Zeiss) or fluorescence microscopy (Axiovert 200; Carl Zeiss). Each figure shows representative results that were observed in a majority of the transfected cells (more than 80%) in at least three independent experiments (more than 50 cells examined).
Quantification of protein colocalization was done on unprocessed images using Metamorph Imaging System Software (version 4.6; Universal Imaging), as described previously (33,34). Briefly, the average grayscale pixel intensity of the nucleus region was measured in the AMAP2, amphiphysin IIm, and Arf channels and defined as background. To subtract background, the threshold of each channel was set at the value obtained for background. The average pixel intensity was measured for the thresholded images, and the threshold was then set at this new value. The percentage of area of overlap between AMAP2 pixels over amphiphysin IIm or Arf pixels or Arf pixels over amphiphysin IIm pixels was calculated. Ten randomly selected cells were examined in each assay.
Immunoelectron microscopy was carried out using the silver enhancement technique as described previously (26,35), with a slight modification. Briefly, cells were fixed in 2% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4) at room temperature for 4 h, rinsed once with 0.1 M phosphate buffer, and blocked with 0.1 M phosphate buffer containing 5% bovine serum albumin for 10 min. Cells were then incubated with primary antibodies in 0.1 M phosphate buffer containing 0.2% bovine serum albumin and 0.02% saponin overnight, followed by a wash with 0.005% saponin in 0.1 M phosphate buffer, and incubated with secondary antibodies coupled with 1.4-nm gold particles (Nanoprobes) for 3 h. After washing, gold particles were enhanced by incubation with silver developer (Nanoprobes). The sections were postfixed with 0.5% OsO 4 , dehydrated with ethanol, and embedded in Wpon. Ultrathin sections were then made and analyzed using an electron microscope (JEM-1010; JEOL). Representative results are shown from more than 10 sections examined.
Endocytosis-Internalization of Tac molecules expressed in HeLa cells by cDNA transfection was measured using an anti-Tac antibody, as described previously (36). Briefly, 0.4 ϫ 10 5 cells in a 35-mm dish, transfected with cDNAs or siRNA duplexes, were incubated with 10 g/ml mouse anti-Tac antibody (clone 7G7) for 30 min at 4°C. Cells were then washed twice with ice-cold DMEM containing 10% fetal calf serum and then incubated with the same medium for 30 min at 37°C. To remove Tac antibodies remaining on the cell surface, cells were incubated with a low pH solution (0.5 M NaCl and 0.5% acetic acid, pH 3.0) at an ambient temperature for 15 s and rinsed once with the medium before fixation with 4% paraformaldehyde. Cells were then permeabilized with 0.2% saponin, and labeled using Cy3-conjugated anti-mouse IgG antibodies to detect internalized Tac antibodies. Internalization of transferrin (Tfn) was measured according to a method described previously (37), using biotinylated human Tfn (Sigma). Briefly, 0.4 ϫ 10 5 cells in a 35-mm dish, transfected with cDNAs or siRNA duplexes, were washed twice with DMEM, incubated with DMEM for 1 h at 37°C, and then incubated with 25 g/ml Tfn for 1 h at 37°C in DMEM containing 10% fetal calf serum. After washing three times with DMEM at an ambient temperature to reduce surface and nonspecific labeling, cells were fixed, permeabilized, and labeled with Cy5-conjugated streptavidin to detect internalized biotinylated Tfn. Amounts of internalized Tac or Tfn molecules were evaluated by quan-tifying the intensities of the fluorescent signals using computer software associated with the confocal laser-scanning microscope (LSM 510 version 2.5). Transfection-positive cells were identified by immunolabeling of transfected proteins by their tags or by the autofluorescence from co-transfected enhanced green fluorescent protein (EGFP). Endocytosis blocked cells were defined as transfected cells in which Tac or Tfn uptake was less than 20% of that seen in untransfected cells in the same fields, as described previously (38). Endocytic blockage was then expressed as a percentage of the untransfected control (37). Results are shown as means Ϯ S.E. in three independent experiments (more than 50 cells were examined in each experiment).

RESULTS
AMAP2 Binds to GTP-Arf6 via Its ArfGAP Domain-It has been shown by protein pull-down assays that guanine nucleotide exchange factor domains can selectively and stably bind to their specific substrate GTPases in the GDP-bound forms (39,40). As one of the possible ways to assess the isoform specificity of ArfGAPs, we examined whether the ArfGAP domain of AMAP2 binds stably to Arf. For this, we prepared purified the GST fusion form of the AMAP2 ArfGAP domain (GST-ArfGAP) on glutathione-beads and incubated them with COS-7 cell lysates, expressing HA-tagged Arf proteins by cDNA transfection. We used Arf1, Arf5, and Arf6 as representing each class of Arfs. As shown in Fig. 1A, GST-ArfGAP pulled down the GTP hydrolysis-deficient mutant of Arf6, Arf6Q67L, but only marginally the GTP binding-deficient mutant, Arf6T27N, and wild type Arf6. Pull-down of the other classes of Arfs was marginal, irrespective of their nucleotide binding status (Fig. 1A). As a control, we used the GST fusion form of the VHS-GAT domain of GGA3 (GST-GGA3), which is known to bind to GTP-bound Arfs (41). This protein pulled down all of the Arf isoforms bound to GTP, Arf1Q71L, Arf5Q71L, and Arf6Q67L (Fig. 1B). We also examined whether GST-ArfGAP can pull down wild , and bound radioactivities were analyzed by thin layer chromatography, as described under "Experimental Procedures." As a control, HA-Arf6 was immunoprecipitated using an anti-HA antibody and similarly analyzed (lane 3). A representative result is shown from four independent experiments. D, 0.3 g of pcDNA-Arf-HA was co-transfected with 4 g of pEBG-AMAP2 or with 2 g of pEBG vector (for GST alone) plus 2 g of pcDNA empty vector in COS-7 cells, as indicated. GST proteins were then pulled down from 500 g of the cell lysates, prepared in 1% Nonidet P-40 buffer in the absence or presence of 10 mM Mg 2ϩ , and analyzed for the co-precipitation of Arf isoforms. In A, B, and D, tagged proteins were detected by antibodies against tags, as indicated. The total includes 20 g of total cell lysate.
type Arf6 bound to GTP. For this, COS-7 cells overexpressing HA-tagged wild type Arf6 was radiolabeled with [ 32 P]orthophosphate, and the cell lysate was incubated with GST-ArfGAP bound to beads. Thin layer chromatography indicated the preferential existence of GTP, but not GDP, in the pulled down fraction of GST-ArfGAP, despite the fact that most of the wild type Arf6-HA, immunoprecipitated by an anti-HA antibody, binds to GDP rather than GTP in COS-7 cells (Fig. 1C).
We then examined possible binding of full-length AMAP2 with wild type Arfs in vivo. To do this, GST-tagged full-length wild type AMAP2 (GST-AMAP2) was co-overexpressed with HA-tagged wild type Arf in COS-7 cells. As shown in Fig. 1D, GST-AMAP2 selectively and efficiently pulled down Arf6, but not the other Arfs. The same result was obtained with Arf cDNAs without any tags (data not shown). This pull-down of Arf6 by GST-AMAP2 was not affected by the presence of 10 mM MgCl 2 in the lysis buffer (Fig. 1D).
AMAP2 Binds to Components of the Endocytic Machinery via Its PRD-AMAP2 also contains a PRD. To obtain further clues regarding the cellular function of AMAP2, we next searched for proteins that bind to the AMAP2 PRD by yeast two-hybrid screening using the PRD as the bait. cDNAs encoding amphiphysin IIm (42) and intersectin-Is (40) were then identified from a HeLa cDNA library. These proteins both have SH3 domains and are isoforms of amphiphysin II and intersectin-I respectively, both of which are well known endocytic components (43,44). Amphiphysin IIm shares a common SH3 domain with other amphiphysin II isoforms at the C terminus but lacks the clathrin-binding domain that the other amphiphysin II isoforms possess (42,45). Binding of AMAP2 with these two proteins was confirmed in vivo, by co-expressing GST-AMAP2 with Xpress-tagged amphiphysin IIm or intersectin-Is in COS-7 cells (Fig. 2A). Anti-amphiphysin II immunoprecipitants prepared from HeLa cell lysate also contained endogenous AMAP2 (Fig. 2B). We hereafter used amphiphysin IIm for further analysis of AMAP2 function.
The AMAP2 PRD contains eight repeats of proline-rich sequences, with significant variations (see Fig. 2C). We then identified the proline-rich sequence responsible for binding to the amphiphysin II SH3 domain. Each of the proline-rich sequences was synthesized on a filter membrane and incubated with radiolabeled GST-amphiphysin II SH3. The amphiphysin II SH3 domain is also known to bind to dynamin. The prolinerich sequence of dynamin-2, essential for binding to the amphiphysin II SH3 domain (38), was also synthesized as a control (spot 9 in Fig. 2C). We found that the eighth proline-rich sequence, AMVLQPPAPMPRKSQ, is most potent in binding to FIG. 2. Binding of AMAP2 to amphiphysin IIm. A, 2 g of amphiphysin IIm or intersectin-Is cDNAs, each tagged with Xpress tag, were co-transfected with 2 g of pEBG-AMAP2 or with 1 g of pEBG vector plus 1 g of pcDNA empty vector in COS-7 cells. GST proteins were then pulled down from 500 g of the cell lysates, and co-precipitating proteins were analyzed by immunoblotting. B, HeLa cell lysate was subjected to immunoprecipitation using an anti-amphiphysin II antibody (Amph) or control mouse IgG1 (MOPC21; IgG), coupled with protein G-Sepharose beads, and analyzed by immunoblotting using an anti-AMAP2 antibody. C, each proline-rich sequence of the AMAP2 PRD (see upper panel) was synthesized on membranes and probed with radiolabeled GST-amphiphysin II SH3 or GST alone, as indicated. Spots 1-8 correspond to the first through eighth proline-rich sequences of the AMAP2 PRD, respectively. Spot 9 contains the QVPSRPNRAP sequence of the dynamin-2 PRD, and spot 10 is a blank. Radioactivities retained on each spot were expressed as a value with spot 9 normalized as 1.0 (bottom right). D, the eighth proline-rich sequence of the AMAP2 PRD (sequence 1) and its alanine-scanning mutants (sequences 2-7) were probed with GST-amphiphysin II SH3 (bottom left), and quantitated with spot 1 normalized as 1.0 (bottom right). Spot 8 is a blank. E, 1 g of pEBG-amphiphysin IIm (encoding GST-amphiphysin IIm) was co-transfected with each 3 g of pEGFP-AMAP2 (WT), the 2M mutant, or the 8M mutant in COS-7 cells. GST-amphiphysin IIm was pulled down from 500 g of cell lysates, and the amounts of protein coprecipitated were determined by immunoblotting, as indicated. F, 1 g of pEBG-amphiphysin IIm (GST-Amph) was co-transfected with 3 g each of pEGFP-AMAP2 (EGFP-AMAP2 (WT)) or a mutant of the eighth proline-rich sequence (EGFP-AMAP2 (8M)) or 2 g of pEGFP-dynamin-2 (EGFP-Dyn) in COS-7 cells, as indicated. GST-amphiphysin IIm was pulled down from 500 g of cell lysates, and the amounts of protein co-precipitated were determined by immunoblotting, as indicated. G, COS-7 cells were transfected with 0.3 g of pcDNA-Arf6-HA and 2 g of pEBG-AMAP2, together with 2 g of pcDNA empty vector (lane 1) or increasing amounts of pcDNA-HA-amphiphysin IIm (0.1, 0.3, 1.0, and 3.0 g in lanes 2-5, respectively). GST-AMAP2 was then pulled down from 500 g of the cell lysates, prepared with 1% Nonidet P-40 buffer in the presence of 10 mM Mg 2ϩ , and co-precipitating proteins were analyzed. In A, E, F, and G, tagged proteins were detected by antibodies against tags, as indicated. The total includes 20 g of total cell lysate.
GST-amphiphysin II SH3 under this hybridization condition, whereas the second and third proline-rich sequences also showed significant affinity (Fig. 2C). These three peptides exhibited significantly higher affinities than that of the dynamin-2 peptide (Fig. 2C). Regarding the eighth proline-rich sequence, by changing each residue to alanine, we furthermore determined that Pro-932, Arg-937, and Lys-938 are critical for the binding to GST-amphiphysin II SH3 (Fig. 2D). We also similarly examined amino acids in the second proline-rich sequence and found that Pro-785, Pro-788, and Arg-790 are critical for the binding to GST-amphiphysin II SH3 (data not shown). We then made two different mutants of AMAP2, 2M (mutation of Pro-785, Pro-788, and Arg-790 into alanines: P785A/P788A/R790A) and 8M (P932A/R937A/K938A) to determine which domain is responsible for the binding of AMAP2 to amphiphysin IIm in vivo. These AMAP2 mutants tagged with EGFP were co-expressed with GST-amphiphysin IIm in COS-7 cells. As shown in Fig. 2E, we found that mutations in the eighth proline-rich sequence abolished the binding, whereas mutations in the second proline-rich sequence did not. Then, EGFP-tagged AMAP2 and dynamin-2 were co-expressed with GST-amphiphysin IIm in COS-7 cells to compare the strength of their binding to amphiphysin IIm in vivo. Under a condition in which EGFP-AMAP2 and EGFP-dynamin-2 were expressed almost equally, only EGFP-AMAP2 was detected to bind to GST-amphiphysin IIm (Fig. 2F). The 8M mutant of AMAP2, on the other hand, could no longer compete with EGFP-dynamin-2 for binding to GST-amphiphysin IIm (Fig. 2F). Collectively, the eighth proline-rich sequence in the AMAP2 PRD appears to be the bona fide interface of AMAP2 binding to the amphiphysin II SH3 domain, and the binding of AMAP2 to amphiphysin IIm appears to be much stronger than that of dynamin-2 to amphiphysin IIm.
We also investigated whether AMAP2 can bind to both Arf6 and amphiphysin IIm simultaneously. GST-AMAP2 and wild type Arf6-HA were co-overexpressed with different amounts of HA-tagged amphiphysin IIm in COS-7 cells. As shown in Fig.  2G, HA-tagged amphiphysin IIm was pulled down together with Arf6-HA by GST-AMAP2, in which increasing amounts of HA-amphiphysin IIm did not affect the amounts of precipitated Arf6-HA.
Recruitment of AMAP2, Together with Amphiphysin IIm, to Arf6 Q67L-enriched Membranes-We next examined the subcellular localization of AMAP2 in HeLa cells. Endogenous AMAP2, as labeled by its antibodies, was found predominantly localized to intracellular tubulovesicular structures in unstimulated HeLa cells (Fig. 3A, a). We also expressed EGFP-AMAP2 and detected the autofluorescence from the EGFP tag. The expression levels of EGFP-AMAP2 were tuned carefully within a 2-3-fold excess of that of endogenous AMAP2 (data not shown) to avoid the possible effects caused by its overexpression, as we did previously (23). We found that this EGFP-AMAP2, expressed at a low level, exhibits a similar subcellular localization as seen with endogenous AMAP2 (Fig. 3A, b). It has been shown in HeLa cells that the majority of Arf6Q67L localizes to the plasma membrane, whereas the majority of Arf6T27N localizes to intracellular tubulovesicular structures, which probably represent the recycling compartment (6,7). In Arf6Q67L-expressing HeLa cells, we found that EGFP-AMAP2 was predominantly redistributed to the Arf6Q67L-positive plasma membrane areas (Fig. 3B, a-d). In these cells, amphiphysin IIm tagged with Xpress-tag also became enriched at the Arf6Q67L-positive membrane areas (Fig. 3B, a-d). At the plasma membrane, a significant fraction of Arf6Q67L is known to localize to the invaginated pits (7). We examined whether a fraction of EGFP-AMAP2 also localizes to the invaginated pits, and this was confirmed by electron microscopy (Fig. 3D). It has been reported in HeLa cells that Arf6Q67L can also be enriched in large intracellular vacuoles, whereas in such cells localization of Arf6Q67L to the plasma membrane is instead barely observed (9). In such HeLa cells bearing Arf6Q67L-positive large vacuoles, we found that both EGFP-AMAP2 and Xpressamphiphysin IIm were also enriched to the Arf6Q67L-positive vacuoles, but not to the plasma membrane (Fig. 3B, e-h). On the other hand, much lesser amounts of EGFP-AMAP2 or Xpress-amphiphysin IIm were colocalized with wild type Arf6-HA (Fig. 3B, m-p). Although we used tagged cDNAs for amphiphysin IIm and AMAP2 to show their colocalization clearly, colocalization of endogenous AMAP2 and endogenous amphiphysin IIm could be seen by immunostaining them in HeLa cells expressing Arf6Q67L or wild type Arf6 (Fig. 3E). Moreover, although both Arf6T27N-HA and EGFP-AMAP2 predominantly localized to intracellular tubulovesicular structures, most of their localization did not seem to be overlapped (Fig. 3B, i-l). Co-localization of EGFP-AMAP2 or Xpress-amphiphysin IIm with Arf1Q71L (Fig. 3B, q-t), Arf1N126I (Fig.  3B, u-x), Arf5Q71L or Arf5N126I (data not shown) was also very limited. To help assessment of results, we quantitated the extent of areas where EGFP-AMAP2, Xpress-amphiphysin IIm, and HA-tagged Arfs were colocalized (Fig. 3C).
We next examined the possible recruitment of AMAP2 by wild type Arf6 when Arf6 is activated. Epidermal growth factor stimulation of HeLa cells has been shown to activate Arf6 and recruit it to the plasma membrane ruffles (46,47). Upon epidermal growth factor stimulation of HeLa cells, we observed the recruitment of a significant population of endogenous AMAP2 to the wild type Arf6-HA-positive plasma membrane (Fig. 3F), which is consistent with the above results.
Involvement of AMAP2 in Arf6-mediated Endocytosis-From experiments overexpressing Arf6 or its mutant cDNAs, it has been proposed that Arf6 is involved in clathrin-independent endocytosis and recycling in HeLa cells, such as for Tac and major histocompatibility complex class I molecules, both of which apparently do not contain cytoplasmic tails conferring clathrin/AP-2 localization (5,6,9). On the other hand, Arf6 may not be involved in the clathrin-dependent endocytosis of Tfn in HeLa cells (5,6,9), whereas in Chinese hamster ovary cells, Arf6 has been proposed to be involved in the endocytosis and recycling of Tfn (3).
To obtain further evidence supporting that AMAP2 is involved in the functions of Arf6 in vivo, we then examined the effects of siRNA-mediated down-regulation of AMAP2 as well as Arf6 on receptor endocytosis in HeLa cells (Fig. 4A). For this purpose, Tac cDNA was expressed in the siRNA-treated cells, and internalization of Tac proteins upon their binding to an anti-Tac antibody was examined. Besides an siRNA duplex with an irrelevant sequence (irr), we also used an siRNA for APAP2/Git2, which we have demonstrated to act as a GAP for Arf1 (26), whereas others have suggested that it acts as a GAP for Arf6, as another control (48). As shown in Fig. 4, B and C, siRNA-mediated knockdown of AMAP2 protein expression significantly inhibited the uptake of Tac, whereas it did not affect the uptake of Tfn. Similarly, suppression of Arf6 also inhibited Tac uptake, but not Tfn uptake (Fig. 4, B and C). In contrast, siRNA-mediated knockdown of APAP2 did not affect the uptake of Tac or Tfn (Fig. 4, B and C). In these experiments, the protein levels of Arf6, AMAP2, and APAP2 were all suppressed to less than 10% of that in control cells (Fig. 4A). It should be noted, however, that about 30 or 50% of Tac uptake activity remained in these Arf6-or AMAP2-siRNA treated cells, respectively. Thus, although Arf6 and AMAP2 both participate in the endocytosis of Tac, it is likely that proteins other than Arf6 and AMAP2 are also independently involved in regulating the endocytotic activity of Tac in HeLa cells.
We have previously shown that AMAP2 is involved in Fc␥-receptor mediated phagocytosis, and this phagocytosis is blocked by the overexpression of AMAP2 but not by APAP2 (23). Arf6 plays a crucial role in Fc␥-receptor mediated phago-

FIG. 3. Subcellular localization of AMAP2 in HeLa cells.
A, cells were fixed in methanol for 5 min at Ϫ20°C and labeled with rabbit AMAP2 polyclonal antibodies coupled with a Cy2-conjugated anti-rabbit IgG antibody (a), or live cells expressing EGFP-AMAP2 at a low level were analyzed by detecting the autofluorescence from the tag (b). B, 1 g each of pcDNA-Xpress-amphiphysin IIm and pEGFP-AMAP2 were co-transfected with 0.5 g of pcDNA-Arf6-HA and its mutants (a-p) or pcDNA-Arf1-HA and its mutants (q-x), as indicated. HA-Arfs and Xpress-amphiphysin IIm were detected using an HA polyclonal antibody coupled with a Cy5-conjugated anti-rabbit IgG antibody and an anti-Xpress monoclonal antibody coupled with a Cy3-conjugated anti-mouse IgG antibody, respectively. EGFP-AMAP2 was detected by its autofluorescence. The right panels show merged views of the left three images. C, protein colocalization was quantified as indicated, according to a method described under "Experimental Procedures." Results were shown as mean values Ϯ S.E. from three independent experiments. D, cells transfected with 4 g of pEGFP-AMAP2 and 2 g of pcDNA-Arf6Q67L-HA were subjected to immunoelectron microscopy analysis using with anti-HA (a) or anti-GFP antibodies (b). E, cells were transfected with 0.5 g of pcDNA-Arf6Q67L-HA (a-d) or pcDNA-wild type Arf6-HA (e-h), as indicated. Twenty-four h later, cells were fixed with paraformaldehyde. Endogenous AMAP2 and amphiphysin IIm were detected by an anti-AMAP2 rabbit polyclonal antibody coupled with a Cy2-conjugated anti-rabbit IgG antibody or an anti-amphiphysin IIm rat monoclonal antibody coupled with a Cy3-conjugated anti-rat IgM antibody. Arf6-HA proteins were detected using an HA monoclonal antibody coupled with a Cy5-conjugated anti-mouse IgG antibody. The right panels show merged views of the left three images. F, cells were transfected with 0.5 g of pcDNA-Arf6-HA. Twenty-four h after transfection, cells were starved of serum for another 24 h (a-c) and then stimulated with 10 ng/ml epidermal growth factor for 7 min (d-f) before fixation with paraformaldehyde. Endogenous AMAP2 was detected by an AMAP2 rabbit polyclonal antibody coupled with a Cy2-conjugated anti-rabbit IgG antibody. Arf6-HA proteins were detected using an HA monoclonal antibody coupled with a Cy5-conjugated anti-mouse IgG antibody. The right panels show merged views of the left two images. Antibodies that can clearly label endogenous Arf6 in HeLa cells were not available. In B, D, E, and F, images were acquired using laser confocal microscopy by adjusting the focus 3.0 m above the surface of the glass chamber plate, which was near the center of the nucleus of the majority of cells. Bars, 10 m in A, B, E, and F; 200 nm in D.
cytosis, whereas other Arf isoforms may not be directly involved in this process (23,24). Moreover, Fc␥ receptor-mediated phagocytosis is believed not to directly employ clathrins for the endocytosis, although clathrin-mediated endosomal recycling may be indirectly required for the phagocytic activity (49). We finally examined the effects of AMAP2 overexpression on Tac and Tfn uptakes in HeLa cells. As shown in Fig. 4D, we found that internalization of Tac was significantly blocked in HeLa cells overexpressing EGFP-AMAP2, whereas overexpression of EGFP-APAP2 did not. A mutant of AMAP2, in which the ArfGAP domain was deleted (⌬ArfGAP), did not exert such blockage (Fig. 4D). Co-overexpression of wild type Arf6-HA together with EGFP-AMAP2 was able to restore Tac internalization (Fig.  4D), similar to what we have previously shown in the case of Fc␥ receptor phagocytosis (23). On the other hand, AMAP2 overexpression did not inhibit Tfn uptake (Fig. 4E). Overexpression of EGFP-APAP2 neither blocked Tfn uptake (Fig. 4E). In these experiments, amounts of each of the exogenous proteins were almost equal (data not shown). As a control, we used the K44A mutant of dynamin-2 (Dyn K44A) (44) to show the blockage of Tac and Tfn uptake in HeLa cells (Fig. 4, D and E).
These experiments using siRNA-mediated protein knockdown and cDNA-mediated protein overexpression both support our notion that AMAP2 is involved in Arf6-mediated endocytosis. DISCUSSION Isoform specificity of the GAP activity of AMAP2/PAG3/Pap␣ has been controversial. Biochemical assays have revealed that AMAP2 exhibits catalytic GAP activity for Arf1 (class I) and Arf5 (class II) but much less activity for Arf6 (class III) (18,25). In contrast, our previous studies have indicated that AMAP2 can antagonize Arf6 activities in vivo (22,23). In this paper, we show that AMAP2, via its ArfGAP domain, can stably bind to Arf6Q67L, a GTP hydrolysis-deficient mutant of Arf6. In contrast, binding of AMAP2 to GDP-Arf6 or other Arf isoforms, irrespective of their nucleotide binding status, was only marginal or almost undetectable. We also provide evidence indicating that AMAP2 binds to the GTP-bound form of wild type Arf6 and that this complex does not immediately dissociate even in the presence of a high concentration of Mg 2ϩ . Consistent with these results obtained in vitro, we found that AMAP2 colocalizes well with Arf6Q67L. We also show that AMAP2 is recruited to Arf6-positive plasma membrane areas upon activation of Arf6 by epidermal growth factor. Further evidence supporting the relationship between AMAP2 and Arf6 was obtained from analysis of the effects of protein overexpression as well as siRNA-mediated protein silencing in HeLa cells. As already mentioned, in HeLa cells, Arf6 has been shown to be involved in the clathrin-independent endocytosis of Tac but not the clathrin-dependent endocytosis of Tfn. We found that AMAP2 silencing caused a significant inhibition of the internalization of Tac but not Tfn. Silencing of Arf6 also caused significant inhibition of the internalization of Tac but not Tfn. Moreover, we showed that overexpression of AMAP2 blocks the internalization of Tac but not Tfn, and this blockage of Tac internalization could be restored by co-overexpression of Arf6. These results are all consistent with a notion that AMAP2 plays a role in Arf6 function. AMAP2 may act as an effector of Arf6 by binding to GTP-Arf6 rather than by simply acting as its catalytic GAP enzyme (also see below). We speculate that the property of AMAP2 being able to stably bind to GTP-Arf6 may prevent AMAP2 from exhibiting efficient catalytic GAP activity against GTP-Arf6 in biochemical assays, but AMAP2 exhibits blockage of cellular functions of Arf6 when overexpressed.
Our results, together with results in literature, indicate that AMAP2 may have dual mechanisms for its function; it exhibits efficient catalytic GAP activity for the class I and class II Arfs by itself, as long as phospholipids such as phosphatidylinositol 4,5-bisphosphate exist at the required levels (18), yet it is involved in the cellular function of the class III Arf, probably through binding to GTP-Arf6 without immediate GAP activity. In this regard, it is noteworthy that unlike other members of the Ras superfamily of GTPases, Arfs themselves do not have intrinsic GTPase activity (10) and are hence thought to be unable to terminate their own activity. Therefore, some Arf-GAPs should possess "nonspecific" GAP activity toward certain isoforms of Arfs, which is used to silence Arf activities being inappropriately activated and/or mislocalized. Such dual mechanisms of AMAP2 function may also be necessary in excluding the inappropriate activities of the class I and class II Arfs from areas where only Arf6 is required to function (50). In this regard, it is interesting to note that significant amounts of cellular phosphatidylinositol 4,5-bisphosphate, an activator of the AMAP2 GAP activity against the class I and II Arfs (18), are produced by an activity of phosphatidylinositol 4-phosphate 5-kinases (51), and phosphatidylinositol 4-phosphate 5-kinases can be activated by GTP-Arf6 (46).
Our results indicated that when AMAP2 is recruited to plasma membrane areas by GTP-Arf6, it can also bring its binding partner, such as amphiphysin IIm. This mode of interaction appears to be analogous to that implicated for the interaction of GTP-Arf1 and its GAP molecules in GTP-Arf1-mediated priming complex formation at the donor membrane (14,15). In this model, the GAP molecules are also thought to be FIG. 4. Involvement of AMAP2 in Tac internalization in HeLa cells. A-C, cells were treated with 50 nM of siRNA oligonucleotide duplexes, each designed for Arf6, AMAP2, or APAP2 silencing, as indicated. Twelve h later, the cells were transfected with 0.4 g of the pKCR-Tac plasmid and 0.1 g of pEGFP-C1 and incubated for a further 24 h before being subjected to endocytosis assays using an anti-Tac antibody (B) or biotinylated Tfn (C) as probes. Oligonucleotides with an irrelevant sequence (irr) were included as a negative control. Protein knockdown was assessed by immunoblotting, as indicated (A). siRNA treatment did not affect protein expression levels of Tac (data not shown). D and E, cells were co-transfected with 0.4 g of pKCR-Tac and 1.6 g of plasmids encoding EGFP-AMAP2 (AMAP2), its ArfGAP domain deletion mutant (⌬ArfGAP), EGFP-APAP2 (APAP2), K44A mutant of EGFP-dynamin-2 (Dyn (K44A)), or EGFP alone, as indicated. They were then subjected to endocytosis assays of Tac (D) and Tfn (E). For co-overexpression of wild type Arf6-HA together with EGFP-AMAP2, 0.3 g of pcDNA-Arf6-HA was included (AMAP2 ϩ Arf6). Endocytic blockage in transfection-positive cells was defined as described under "Experimental Procedures." Results are means Ϯ S.E. from three independent experiments. *, p Ͻ 0.01 against values of the control. recruited by GTP-Arf1 to the donor membrane without immediately exhibiting GAP activity. Moreover, it has been proposed that the hydrolysis of GTP-Arf6 is necessary for Tac internalization in HeLa cells (6). Amphiphysin is thought to play a role in recruiting dynamin to the neck of deeply invaginated pits at the late stage of endocytosis, and dynamin may then contribute to narrow the neck and thus to the final fission reaction of the invaginatedpits (44).Althoughthismodelisproposedforclathrindependent endocytosis, it is well documented that dynamin-2 is also involved in clathrin-independent endocytosis (52,53). We showed that AMAP2 binding to amphiphysin IIm acts to block dynamin-2 binding to amphiphysin IIm. Therefore, one possible scenario for the biological significance of AMAP2-mediated protein complex formation with GTP-Arf6 and amphiphysin IIm is as follows. Upon Arf6 activation, AMAP2 acts to recruit a subset of endocytic components, such as amphiphysin IIm, to GTP-Arf6-positive plasma membrane sites by binding to GTP-Arf6. This process is for the formation of priming complexes at the early phase of Arf6-mediated endocytosis. As long as amphiphysin IIm is bound to AMAP2, the subsequent steps of endocytosis are blocked. At some point during endocytosis, AMAP2 dissociates from amphiphysin IIm and allows amphiphysin IIm to then recruit dynamin, which is necessary for the late phase of endocytosis. This process may be coupled with the hydrolysis or some further processing of GTP-Arf6. AMAP2 can be tyrosine-phosphorylated by Pyk2 (18). We have found that tyrosine-phosphorylated AMAP2 does not bind to Arf6 and amphiphysin IIm, 2 suggesting that the Pyk2-mediated phosphorylation may be involved in processing the AMAP2-mediated protein complex. Further analysis on the fate of the AMAP2 protein complex with GTP-Arf6 is necessary to understand the precise regulation of Arf6, whose activity is involved in the higher orders of cellular functions (22).
In conclusion, we provide evidence that AMAP2 has the property to act as an adaptor protein accumulating several endocytic proteins to the sites of Arf6 activation, besides its proposed role acting as a catalytic GTPase-activating protein for other Arfs. It should be noted that the AMAP2 PRD consists of several different proline-rich sequences and that amphiphysin IIm is not the only protein that binds to AMAP2, and other proteins like intersectin-I also bind to the AMAP2 PRD. A number of different proteins, each bearing SH3 domains, are involved in endocytosis (43,44), in addition to amphiphysin II and intersectin-I. Although we used Tac molecules as a model to show the possible involvement of AMAP2 in Arf6-mediated endocytosis, we have yet to analyze which AMAP2-binding SH3 protein(s) is involved in Tac uptake in HeLa cells. Further studies will be necessary for each specific type of Arf6-mediated endocytosis, together with identification of the SH3-containing proteins acting with AMAP2, for the precise understanding of their molecular mechanisms.