AGAP1, an Endosome-associated, Phosphoinositide-dependent ADP-ribosylation Factor GTPase-activating Protein That Affects Actin Cytoskeleton*

We have identified three members of the AGAP subfamily of ASAP family ADP-ribosylation factor GTPase-activating proteins (Arf GAPs). In addition to the Arf GAP domain, these proteins contain GTP-binding protein-like, ankyrin repeat and pleckstrin homology domains. Here, we have characterized the ubiquitously expressed AGAP1/KIAA1099. AGAP1 had Arf GAP activity toward Arf1>Arf5>Arf6. Phosphatidylinositol 4,5-bisphosphate and phosphatidic acid synergistically stimulated GAP activity. As found for other ASAP family Arf GAPs, the pleckstrin homology domain was necessary for activity. Deletion of the GTP-binding protein-like domain affected lipid dependence of Arf GAP activity. In vivoeffects of AGAP1 were distinct from other ASAP family Arf GAPs. Overexpressed AGAP1 induced the formation of and was associated with punctate structures containing the endocytic markers transferrin and Rab4. AP1 was redistributed from the trans-Golgi to the punctate structures. Like other ASAP family members, AGAP1 overexpression inhibited the formation of PDGF-induced ruffles. However, distinct from other ASAP family members, AGAP1 also induced the loss of actin stress fibers. Thus, AGAP1 is a phosphoinositide-dependent Arf GAP that impacts both the endocytic compartment and actin.

erase chain reactions and subcloned into the EcoRI and NotI sites of pCI (Promega, Madison, WI) with a FLAG tag (DYKDDDDK) for mammalian expression or into pGEX4T-1 (Amersham Biosciences) for expression as a GST fusion protein in bacteria. pCDNA3-HA-Erk2 and v-Ras were gifts from Dr. Douglas Lowy at the NCI, National Institutes of Health (Bethesda, MD). GFP-Rab4, GFP-Rab5, and GFP-Rab11 were generously provided by Dr. Juan Bonifacino at the NICHD, National Institutes of Health (Bethesda, MD). Point mutations were introduced using the QuikChange site-directed mutagenesis kit from Stratagene (La Jolla, CA) and confirmed by DNA sequencing at the NCI, National Institutes of Health sequencing core facility.
Antibodies-Monoclonal M5 anti-FLAG epitope antibody and M2 anti-FLAG affinity matrix were from Sigma. Monoclonal anti-hemagglutinin antibody and anti-hemagglutinin affinity matrix were from Roche Molecular Biochemicals. Anti-Ras and anti-phosphorylated MAP kinase antibodies were from Upstate Biotechnology (Lake Placid, NY). Anti EEA1, Rab4, Rab5, and Rab11 were from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA). Anti-␤COP was from Affinity BioReagents (Golden, CO). Anti-␥-adaptin was from Transduction Laboratories (San Diego, CA). Rhodamine-conjugated phalloidin and rhodamine-conjugated transferrin, Alexa-488, and Alexa-594 secondary antibodies were from Molecular Probes (Eugene, OR). Alkaline phosphatase-conjugated donkey anti-mouse and alkaline phosphatase-conjugated donkey antirabbit antibodies were from Bio-Rad Laboratories (Hercules, CA). Rabbit antisera 1158 against residues 784 -804 of AGAP1 conjugated to a C-terminal cysteine residue were raised at Covance (Denver, PA). The antisera recognized the FLAG epitope-tagged AGAP1 or GST-AGAP1 by immunoblot, and the signals were blocked by the peptide to which the antisera were raised, whereas the peptide corresponding to residues 45 to 68 failed to compete for the signal.
Determination of Tissue Distribution of AGAP1 mRNA-Tissue cDNA panels were obtained from Clontech Laboratories (Palo Alto, CA). Specific primers were designed to amplify nucleotides 648 to 1245 (corresponding to codons 27 to 224) of the mouse cDNA and from 648 to 1369 of the human cDNA. The forward primer used for both the mouse and the human cDNAs was ACATCTACTCCATCTACGAGCTGC (nucleotides 648 to 671 of KIAA1099). The reverse primer for mouse cDNA was ACGTCCTGGAAGACCCGCTCCAC (corresponding to nucleotides FIG. 1. Primary structure of AGAP1. A, amino acid sequence of AGAP1. The structural domains were identified using Pfam. The GLD is underlined, PH domain is in bold in the box, Arf GAP domain is in bold, and ANK repeats are in bold and underlined. B, schematic structural domains of AGAPs. The percentages of identity of the GLDs to different G-proteins are indicated in parentheses. The starting and ending residues of each domain are numbered as identified by Pfam. C, comparison of AGAP1 GLD to K-Ras. Alignment was performed with ClustalW program. Consensus GTP binding motifs are in bold. Switch I and switch II regions are boxed. D, comparison of PH domains of AGAPs with that of human Bruton tyrosine kinase (BTK). Alignment was performed with ClustalW program and subsequent manual adjustment. Consensus motif for D3-phosphorylated phosphoinositol binding (30,31) is shown at the bottom, and the conservative mutation from arginine to lysine is indicated with an arrow. 1245 to 1223 of KIAA1099). The reverse primer for human cDNA was GCTGATGTGCACGGCAGACACC (nucleotides 1369 to 1347 of KIAA1099). Thermal cycling was performed under the conditions specified by the manual from Clontech Laboratories, Inc. PCR products were fractionated on a 1% agarose gel and stained with ethidium bromide.
Cell Culture and Immunofluorescence-NIH 3T3, COS7, U87, and HEK293T cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% FBS at 37°C with 5% CO 2 . PC6 cells were maintained in RPMI 1640 medium supplemented with 10% heatinactivated horse serum and 5% FBS at 37°C with 5% CO 2 . Cells were transfected using LipofectAMINE 2000 (Invitrogen) and harvested 24 h FIG. 2. Arf GAP activity of AGAP1. A, activation of AGAP1 by phospholipids. Arf GAP activity of GST-AGAP1 was determined as described under "Experimental Procedures" using Arf1 as substrate. Data are the mean Ϯ S.E. of one experiment representative of three with similar results. *, p Ͻ 0.05 compared with no PA. An interaction between PA and PIP 2 was indicated as analyzed by two-way ANOVA. B, Potentiation of PIP 2 activation of AGAP1 by different acid phosphoinositides. Arf GAP activity of GST-AGAP1 was determined as in A. PA, phosphatidylserine (PS), phosphatidylinositol (PI), and phosphatidylcholine (PC) were all at 90 M, and PIP 2 was at 45 M. Data are the mean Ϯ S.E. of one experiment representative of three with similar results. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001 compared with PIP 2 alone; &&&, p Ͻ 0.001 compared with PIP 2 ϩ PA as analyzed by one-way ANOVA with Tukey post-test. C, Arf specificity of GST-AGAP1. GAP activity of GST-AGAP1 was determined using Arf1, 5, and 6 as substrates in the presence of 45 M PIP 2 and 360 M PA. Data are the mean Ϯ S.E. of one experiment representative of three with similar results. **, p Ͻ 0.01; ***, p Ͻ 0.001 compared with Arf1; &&, p Ͻ 0.01 compared with Arf5 as analyzed by one-way ANOVA with Tukey post-test. D, structural requirements of AGAP1 for Arf GAP activity. FLAG-tagged AGAP1, [R599K]AGAP1, [C594S]AGAP1, [⌬GLD]A-GAP1, and [⌬GLD,⌬PH]AGAP1 were expressed in HEK293T cells and immunoprecipitated with the anti-FLAG affinity-agarose and eluted with FLAG peptide. Equal amounts of soluble protein determined by Western blot were used for the Arf GAP assays. Data are the mean Ϯ S.E. of one experiment representative of three with similar results. ***, p Ͻ 0.001 compared with FLAG empty vector transfection control; &&&, p Ͻ 0.001 compared with AGAP1 as analyzed by one-way ANOVA with Tukey post-test. E, homology model of the [⌬GLD]AGAP1. The modeling was performed based on the crystal structure of the Arf GAP domain of PAP␤. The Arf GAP domain is shown in red, and the ANK repeats in green and blue. F, effect of GLD on PA potentiation of Arf GAP activity. The Arf GAP activity of GST-AGAP1 and GST-[⌬GLD]AGAP1 was compared in the presence or absence of 90 M PA with increasing concentrations of PIP 2 . The Arf GAP assay was performed as described under "Experimental Procedures." Data presented are the mean Ϯ S.E. from three experiments. later for reseeding, immunoblotting, or immunoprecipitation. Cells were deprived of serum for 12 h before growth factor stimulation. Cells were visualized by staining with rhodamine-phalloidin, and transfected cells were detected using an antibody to the epitope tag. Cells were examined by confocal microscopy using a Zeiss Pascal confocal mounted on an Axioplan 2 microscope equipped with an 100ϫ Plan-Neofluar oil immersion lens (Carl Zeiss). For PDGF-induced dorsal ruffle assays, cells were incubated for 6 h and then treated with 10 ng/ml PDGF for 4 min. For transferrin uptake, NIH 3T3 cells were incubated with regular medium for 6 h after reseeding. Then the cells were incubated with serum-free medium for 30 min before incubation with transferrin for another 30 min. Cells were then washed and fixed with 2% formaldehyde in phosphate-buffered saline for 10 min and stained with rhodamine-phalloidin and anti-FLAG antibody (25). At least 100 cells were counted from at least two separate coverslips to quantify differences in ruffling and stress fibers.
Point mutants and truncation mutants of AGAP1 were expressed in HEK293T and BL21 cells.  (20), we assume that the [⌬GLD,⌬PH]AGAP1 protein is folded properly. We cannot assert that the isolated GLD domain contains the native fold as present in the full-length protein. Tests for nucleotide binding were performed with both this isolated domain and the full-length protein with similar negative results.
Detection of Nucleotide by HPLC-Nucleotides copurified with proteins were detected by the method described (27) with slight modifications. Proteins were denatured by the addition of 1 M formic acid and separated from nucleotides by centrifugation. The supernatant was lyophilized and resuspended in a buffer containing 50 mM sodium phosphate, pH 6.5, 0.2 mM tetrabutylammonium hydrogen, 3% (v/v) acetonitril, and 0.2 mM NaN 3 . The nucleotides were then fractionated on a C18 column developed by isocratic elution with the resuspension buffer. Absorbance was monitored at 252 nm. Under this condition, the retention times for GDP, GTP, ADP, and ATP were 6.5, 10.2, 9.5, and 17 min, respectively. Arf was used as a positive control.
Equilibrium Binding-Nucleotide binding was measured using GST-GLD and GST-AGAP1 fusion proteins bound to 10 l of glutathione beads. The immobilized proteins were incubated with 0.25 M radiolabeled GTP, GTP␥S, or ATP in 20 mM Tris, pH 7.5, 100 mM NaCl, 1 mM EDTA, 1 mM dithiothreitol, either 0.5 (low Mg 2ϩ ) or 2 mM MgCl 2 (high Mg 2ϩ ), 0.1% Triton X-100, 360 M PA, and 45 M PIP 2 in a total volume of 50 l. After 5 min at room temperature, the reactions were centrifuged briefly at 500 ϫ g, and 40 l of the supernatant was removed. The radioactivity in the supernatant and the remaining pellet was quantified by scintillation spectroscopy.
Similarly, Arf⅐GTP␥S binding to AGAP1 was measured by immobilizing GST-AGAP1 on glutathione beads, incubating the protein under the same conditions as the GAP assay with Arf⅐[ 35 S]GTP␥S, and determining the amount of Arf⅐[ 35 S]GTP␥S associated with the beads by scintillation spectroscopy.
Miscellaneous-Nitrocellulose filter binding was performed as described (26). Briefly, the protein was incubated with radiolabeled nucleotide in a buffer containing 20 mM Tris, pH 7.4, 100 mM NaCl, 0.5 or 2.0 mM MgCl 2 , 1 mM EDTA, 1 mM dithiothreitol, and 0.1% Triton X-100. The reaction was terminated after 30 min at 30°C by addition of ice-cold binding buffer containing 20 mM Tris, pH 8.0, 100 mM NaCl, 10 mM MgCl 2 , 1 mM dithiothreitol. Bound nucleotide was determined by rapid filtration on nitrocellulose filters and quantified with a scintillation counter. Arf GAP activity was measured as described (28). Arf proteins were preloaded with [␣-32 P]GTP by incubation at 30°C for 30 min in 25 mM HEPES, pH 7.4, 100 mM NaCl, 2.5 mM MgCl 2 , 1 mM EDTA, 1 mM ATP, 25 mM KCl, 1.25 units/ml pyruvate kinase, and 3 mM phospho(enol)pyruvate. The preloaded Arf was diluted in a reaction buffer containing 25 mM HEPES, pH 7.4, 100 mM NaCl, 2 mM MgCl 2 , 1 mM GTP, 1 mM dithiothreitol. Reactions were terminated after 2 to 4 min by adding ice-cold buffer containing 10 mM Tris, pH 8.0, 100 mM NaCl, 10 mM MgCl 2 , 1 mM dithiothreitol. Protein-bound nucleotide was trapped on nitrocellulose filters, released into 2 M formic acid, fractionated by chromatography on polyethyleneimine cellulose plates, and quantified using a PhosphorImager (Molecular Dynamics). An equal amount of different forms of AGAP1 was used when comparing their activities. The quantity of immunoprecipitated proteins was compared by immunoblotting. Immunoblot signals were visualized with the Su-perSignal chemiluminescent substrate kit from Pierce (Rockford, IL). All lipids were presented as mixed micelles with Triton X-100. All experiments were performed at least three times with similar results. Statistical analyses were performed with GraphPad Prism.

RESULTS
The AGAP Subfamily of ASAP-type Arf GAPs-We have defined ASAP-type Arf GAPs to be proteins containing a core of a PH, Arf GAP, and ANK repeat domains. Database searches revealed three cDNA-encoding proteins containing these structures, KIAA1099 (accession number AB029022), KIAA0167 (accession number D79989), and M-RIP (accession number AF359283). These three proteins are distinguished from other ASAP family proteins in two ways. First, they contain an N-terminal GLD (GTP-binding protein-like domain). Second, the PH domain is split (Fig. 1A, bold in box). We propose calling these proteins AGAPs for Arf GAP with GTP-binding proteinlike, ANK repeat and PH domains. In this nomenclature, KIAA1099 and KIAA0167 are AGAP1 and AGAP2.
The GLDs of the AGAPs, though highly similar to Ras family proteins, have some critical differences in primary structure. The GLD of AGAP1 is 51% similar and 27% identical to K-Ras, AGAP2 is 42% similar and 28% identical to Rab9, and M-RIP is 47% similar and 27% identical to H-Ras (Fig. 1B). Three GTP binding motif consensus sequences are well conserved, GXXXXGK(S/T) (residues 78 -85 in AGAP1; see Fig. 1C, bold), DXXG (residues 124 -127 in AGAP1; see Fig. 1C, bold), and TCAT (residues 210 -213 in AGAP1; see Fig. 1C, bold). However, glutamine 61 of Ras (threonine 61 in Rap) is replaced with a proline in AGAPs, and there is no identifiable NKXD motif. In AGAP1, the stretches over switch I and II (Fig. 1C, box) have 25 and 17% identity to K-Ras (29). Residues in switch I critical for FIG. 3. Detection of AGAP1 in cultured cells. A, endogenous AGAP1 in different cell lines. Twenty-five g of NIH 3T3 lysates, 50 g of lysates from U87, HeLa, and HEK293T cells were loaded for SDS-PAGE. FLAG-AGAP1 immunoprecipitated from HEK293T cells and GST-[⌬GLD]AGAP1 purified from BL21 cells were included as positive controls. The AGAP1 was detected using a polyclonal antibody against the C-terminal peptide sequence of AGAP1. B, blockade of AGAP1 signal by the peptide. The peptide (1 mg/ml) to which the antisera were raised was incubated with the antisera for 1 h at room temperature. The antisera were used subsequently for Western blot.
The PH domains of the AGAPs are distinct from those in other ASAP family Arf GAPs. They are split and, therefore, extend over 189 residues (Fig. 1A, bold in box). AGAP1 contains nine of the ten residues identified in the phosphatidylinositol 3,4,5-trisphosphate (PIP 3 ) binding consensus sequence (30,31) including a tyrosine that is present in all identified PIP 3 binding PH domains (Fig. 1D). The difference is conservative, with a lysine substituting for an arginine (Fig. 1D, arrow).
We have characterized the product of the AGAP1 gene, a protein with 804 amino acids (Fig. 1A) and a calculated molecular mass of 89 kDa. Expressed sequence tag database search revealed a mouse homologue of AGAP1 (accession BF322239), which is 96% identical (179/187) to AGAP1 at the amino acid level and 87% (489/559) identical to AGAP1 at the nucleic acid level for the 187 amino acids encoded by the expressed se-quence tag. Based on reverse transcriptase-PCR using cDNA from human heart, brain, placenta, lung, liver, skeletal muscle, kidney, and pancreas, AGAP1 is expressed in all the tissues tested (data not shown).
AGAP1 Is a PA/PIP 2 -dependent Arf1 GAP-AGAP1 expressed in bacteria as a GST fusion protein and in mammalian cells as an epitope-tagged protein was purified and examined as an Arf GAP. Both bacterially and mammalian-expressed AGAP1 contained Arf GAP activity. Activity was dependent on phospholipids. In the absence of other phospholipids, 10 and 45 M PIP 3 stimulated GAP activity to a greater extent than did PIP 2 . However, phosphatidic acid potentiated PIP 2 stimulation, which was not observed for PIP 3 (Fig. 2A). An interaction between PA and PIP 2 was indicated by two-way ANOVA analysis. The synergy was specific for PA. Phosphatidylserine (PS), phosphatidylinositol (PI), and phosphatidylcholine (PC) had less of an effect on PIP 2 -stimulated activity (Fig. 2B). Repre- FIG. 4. Colocalization of AGAP1 with endocytic markers. A-C, colocalization of AGAP1 with Rab4. NIH 3T3 cells were co-transfected with AGAP1 and GFP-Rab4 at a ratio of 5 to 1 with a total of 3 g cDNA/35-mm well for 24 h. Cells were replated on fibronectin-coated coverslips in Opti-MEM for 6 h before fixing and staining with M5 antibody. Representative areas of AGAP1 and GFP-Rab4 colocalization are indicated by arrows. D-F, distribution of GFP-Rab5 and AGAP1 in NIH 3T3 cells. Cells were cotransfected with FLAG-AGAP1 and GFP-Rab5 at a ratio of 5 to 1 with a total of 3 g cDNA/35-mm well for 24 h. Cells were replated on fibronectin-coated coverslips in Opti-MEM for 6 h before fixing and staining with M5 antibody. No colocalization was observed between AGAP1 and Rab5 under the experimental conditions. GϪI, distribution of AGAP1 and Rab11 in NIH 3T3 cells. Cells were co-transfected with FLAG-AGAP1 and GFP-Rab11 at a ratio of 5 to 1 with a total of 3 g cDNA/35-mm well for 24 h. Cells were replated on fibronectin-coated coverslips for 6 h in Opti-MEM before fixing and staining with M5 antibody. Limited colocalization of AGAP1 and Rab11 was observed as marked by arrows. J-L, colocalization of AGAP1 with endogenous AP1. NIH 3T3 cells were transfected with FLAG-AGAP1 at 3 g of cDNA/35-mm well for 24 h. Cells were replated on fibronectin-coated coverslips in Opti-MEM for 6 h. Cells were then fixed and stained with M5 for transfected cells and AP1 antibody (Transduction Laboratories) for endogenous AP1. Representative areas of AGAP1 and AP1 colocalization are indicated by arrows. M, distribution of GFP-Rab4 in NIH 3T3 cells. NIH 3T3 cells were transfected with FLAG empty vector and GFP-Rab4 at a ratio of 5 to 1 with a total of 3 g/35-mm well for 24 h. Cells were replated on fibronectin-coated coverslips in Opti-MEM for 6 h before fixing. N, distribution of GFP-Rab11 in NIH 3T3 cells. NIH 3T3 cells were transfected with FLAG empty vector and GFP-Rab11 at a ratio of 5 to 1 with a total of 3 g/35-mm well for 24 h. Cells were replated on fibronectin-coated coverslips in Opti-MEM for 6 h before fixing. O, AP1 in untransfected cells. Untransfected NIH 3T3 cells were replated on fibronectin-coated coverslips in Opti-MEM for 6 h. The cells were then fixed and stained with AP1 antibody. sentative proteins from each class of Arf, Arf1, Arf5, and Arf6, were used as substrates for AGAP1 under optimal conditions. Both bacterially expressed GST-AGAP1 (Fig. 2C) and mammalian-expressed FLAG-AGAP1 (not shown) used Arf1 as the preferred substrate.
We next examined the structural requirements for Arf GAP activity as defined for other ASAP family members. Different AGAP1 constructs were expressed in HEK293T cells, immunoprecipitated with M2 gel, and eluted with FLAG peptide. Equal amounts of protein were used in these assays. We found that AGAP1 was similar to other ASAP family members in three structural requirements as follows. (i) A zinc finger was necessary. The zinc finger motif could be disrupted by substituting a serine for a cysteine. This mutation resulted in a protein with no activity (Fig. 2D). (ii) The arginine five residues C-terminal to the last cysteine defining the zinc finger was also required for activity. A conservative substitution of a lysine resulted in a protein with no activity (Fig. 2D). This mutation was unlikely to cause the protein to misfold. Homology modeling (Fig. 2E) was carried out against the crystal structure of the Arf GAP domain of PAP␤ (32), using Homology/Discover module within Insight II (Accerlys, San Diego, CA). Based on this result, a conservative mutation of arginine 599 to lysine is expected to have minimal effect on the stability and structure of the protein. Furthermore, mutation of arginine 599 to lysine did not affect the ability of GST-[⌬GLD]AGAP1 to bind to Arf⅐GTP␥S. tained with the GST-tagged proteins purified from bacteria, except for [C594S]AGAP1, which was insoluble (not shown), consistent with the suspected unfolding of the zinc fingercontaining domain.
GLD Affects the Phospholipid Dependence of Arf GAP Activity-Because of the similarity of the GLD to K-Ras, we examined AGAP1 for analogous function. First we tested for nucleotide binding under a number of conditions, including the presence or absence of detergent, and/or phosphoinositides, in the presence of 1 mM MgCl 2 or Mg 2ϩ buffered to 1 M with EDTA. No high affinity binding of GDP, GTP, or GTP␥S was detected by a nitrocellulose filter binding assay under any of these conditions. The presence of nucleotide on purified AGAP1 was examined by denaturing the protein in formic acid and detecting released nucleotide by HPLC. No GDP, GTP, ADP, or ATP was detected on purified GLD. We also could not detect any low affinity ATP, GTP, or GTP␥S binding in equilibrium dialysis experiments (not shown). Finally, no GTPase activity was detected in full-length AGAP1 or the isolated GLD (not shown). Similarly, we could not detect any effect on the Ras phosphorylation cascade. Overexpression of AGAP1 in PC6 or COS7 cells had no effect on MAP kinase phosphorylation observed in cells maintained in serum-free medium, either in the presence or absence of epidermal growth factor (34) or nerve growth factor (35). AGAP1 also had no effect on MAP kinase phosphorylation induced by coexpressing v-Ras (36) (data not shown).
The GLD did influence the Arf GAP activity of AGAP1. As noted above, the protein lacking the GLD, [⌬GLD]AGAP1, had GAP activity. The activity was stimulated by phospholipids. However, stimulation by PIP 2 did not require, nor was it synergistic with, PA. At concentration range from 11.25 to 90 M, PIP 2 activated [⌬GLD]AGAP1 GAP activity to a similar extent regardless of whether PA was present (Fig. 2F).
In Vivo Analysis of AGAP1 Function-To examine the in vivo function of AGAP1, we first examined the expression of endogenous AGAP1 by Western blot using an antibody raised to a peptide corresponding to the C terminus of AGAP1. In lysates from U87, HeLa, and HEK293T cells, two forms of the protein were found. One band was detected at the predicted molecular mass and one additional band was at a slightly greater molec-ular mass than predicted. This difference could represent either a covalent modification or splicing variation. In NIH 3T3 cells, the molecular mass was slightly greater than predicted for AGAP1 (Fig. 3A). A species of the same molecular mass was noted in skeletal muscle (data not shown). The difference in molecular mass is most likely the result of different splicing of the message. 2 In addition, immunoprecipitated FLAG-AGAP1 and purified GST-[⌬GLD]AGAP1 were also recognized by the antisera. These specific bands were blocked with addition of the peptide to which the antisera were raised (Fig. 3B). These results indicated that the bands around 98 kDa and above as recognized by the antisera represent the endogenous AGAP1.
AGAP1 associated with an endocytic compartment. U87 and NIH 3T3 cells expressing FLAG-epitope tagged AGAP1 were fixed and stained with an antibody to the FLAG epitope. Signal was detected in punctate structures in the periphery of the cell (see Fig. 7A and Fig. 8B, arrows). At higher expression levels the signal was more diffuse. To determine whether the punctate structures represented a membrane-bound organelle, we co-stained cells for a number of known organellar markers. Overexpression of AGAP1 caused redistribution of GFP-Rab4 (Fig. 4, A-C and M) and AP1 (Fig. 4, J-L and O) to the punctate structures that contained AGAP1. GFP-Rab11 showed predominant perinuclear localization when coexpressed with AGAP1, with limited colocalization with AGAP1 (Fig. 4, G-I and N). The punctate structures did not contain, nor did AGAP1 affect, the distribution of Rab5 (Fig. 4, D-F), AP2, EEA1, or ␤-COP (not shown). The colocalization of AGAP1 with Rab4 and Rab11, but not Rab5, was also observed when the cells were stained with antibodies against the different Rab proteins (not shown). The effect on Rab4 and AP1 depended on GAP activity.
[R599K]AGAP1, which lacks GAP activity, neither associated with the punctate structures nor colocalized with AP1 and Rab4. The effect on AP1 and Rab4 did not depend on the GLD. To further test for a role of AGAP1 in endocytic traffic, we looked at its effect on transferrin uptake. NIH 3T3 cells were loaded with rhodamine-conjugated transferrin for 30 min. In untransfected NIH 3T3 cells, transferrin was in punctate structures distributed evenly throughout the cytosol (Fig. 5M). Overexpression of AGAP1 resulted in the accumulation of transferrin into larger punctate structures in the cell periphery, similar to those containing Rab4 and AP1 (Fig. 5, A-C). This effect of AGAP1 was dependent on its Arf GAP activity.
[R599K]AGAP1, which is deficient in GAP activity, did not Twenty-four h after transfection, cells were re-seeded on fibronectin-coated coverslips in Opti-MEM for another 6 h before fixing. B, AGAP1 induced loss of stress fibers specifically. NIH 3T3 cells were transfected with empty vector or FLAG-tagged AGAP1, ASAP1, and ACAP1 at 3 g DNA/35-mm well for 24 h, re-seeded on fibronectin-coated coverslips in Opti-MEM for 6 h, fixed, and stained as described in A. Data are the mean Ϯ S.E. of three experiments. *, p Ͻ 0.05; ***, p Ͻ 0.001 compared with FLAG empty vector transfection control; &&&, p Ͻ 0.001 compared with AGAP1 as analyzed by one-way ANOVA with Tukey post-test. Inset, ectopic expression level of AGAP1, ASAP1, and ACAP1 in NIH 3T3 cells. C, structural requirements for AGAP1-induced loss of actin stress fibers. NIH 3T3 cells were transfected with AGAP1 constructs at 3 g DNA/35-mm well for 24 h, re-seeded on fibronectin-coated coverslips for 6 h in Opti-MEM, and stained as described. Data are the mean Ϯ S.E. of three experiments. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001 compared with FLAG empty vector transfection control; &&&, p Ͻ 0.001 compared with AGAP1 as analyzed by one-way ANOVA with Tukey post-test. D, expression level of different AGAP1 constructs in NIH 3T3 cells. Cells were transfected as in C and were harvested 24 h after transfection. Cleared cell lysates were used for SDS-PAGE and Western blot and detected by M5 antibody against the FLAG tag.
induce redistribution of transferrin (Fig. 5, G-I). The GLD domain does not have a direct role in this effect of AGAP1. Overexpression of GLD had no observable effect on the distribution of transferrin (Fig. 5, J-L).
All ASAP family members examined to date affect the actin cytoskeleton when overexpressed. Here, we examined whether AGAP1 also had effects on actin. We used NIH 3T3 cells as a model, because they are well characterized in this regard. When overexpressed in NIH 3T3 cells, AGAP1 inhib-ited PDGF-induced ruffling effectively (Fig. 6A). This effect was dependent on Arf GAP activity.
[⌬GLD]AGAP1 was as effective as the full-length AGAP1. [⌬GLD,⌬PH]AGAP1, which also lost GAP activity, had only a small effect on ruffling.
[R599K]AGAP1, which also lacked GAP activity, was about 50% as effective as wild type protein indicating that interactions, in addition to the GAP activity, might also be involved in inhibiting ruffling. These differences in the effects of the protein were not a result of differences in [R599K]AGAP1, [⌬GLD,⌬PH]AGAP1, and GLD, which did not block ruffling, were expressed at a similar or higher level to AGAP1 and [⌬GLD]AGAP1, which did block ruffling (Fig. 6B).
Examination of NIH 3T3 cells also revealed a specific effect of AGAP1 on the actin cytoskeleton. Cells had fewer stress fibers and a thickened cortical actin around the cell periphery (Fig. 7A, medium arrow). This effect on stress fibers was specific for AGAP1. Two other members of this family, ASAP1 and ACAP1, had only a small effect on stress fibers in NIH 3T3 cells (Fig. 7B). In the case of AGAP1, although there was a loss of stress fibers, the cells remained flat. Further analysis of the structural requirements of AGAP1 to induce this effect on actin revealed several characteristics of the protein. First, multiple domains of AGAP1 contributed to this effect. The maximum effect was seen with full-length protein.
[⌬GLD]AGAP1, a functional Arf GAP, was less effective than wild type protein (Fig.  7C). The isolated GLD, though to a lesser extent, also affected stress fibers. Therefore, both the GLD and the Arf GAP domain contributed to this effect on actin stress fibers. Second, Arf GAP activity contributed to the effect of the protein on stress fibers.
[R599K]AGAP1, which lacked GAP activity, had less of an effect. Expression levels are unlikely to cause these differences (Fig. 7D). The GLD also induced morphological changes that were distinct from the effects on stress fibers. Cells expressing the isolated GLD were more likely to be elongated with filopodia (Fig. 7A, arrowhead). This effect of the GLD was attenuated in the context of the full-length protein.
The in vivo effects of AGAP1 were similar in U87 cells. FLAG-tagged AGAP1 constructs were expressed in U87 cells (Fig. 8A). First, as in NIH 3T3 fibroblasts, both AGAP1 and [⌬GLD]AGAP1 induced the accumulation of the punctate structures presumed to be endocytic intermediates (Fig. 8B,  arrow). Second, overexpression of AGAP1 in U87 cells also resulted in loss of actin stress fibers (Fig. 8C). Although all examined proteins had an effect, the full-length wild type AGAP1 was most effective. Third, when GLD was overexpressed in U87, the cells were more likely to appear in an elongated form, and about 50% cells had pseudopodia or long processes (Fig. 8B, arrowhead). This effect of GLD was attenuated or blocked in the context of the full-length protein. DISCUSSION Here we describe a fourth subgroup of the ASAP family of Arf GAPs, the AGAPs. These proteins have a G-protein like domain, in addition to the core of PH, Arf GAP, and ANK repeat domains seen in other ASAP family members. We characterized one gene product, AGAP1. AGAP1 had PA-and PIP 2 -dependent Arf GAP activity. The GLD imparted the PA dependence. AGAP1 induced the formation of punctate structures that contained Rab4 and AP1 and were an intermediate in transferrin trafficking in cells. AGAP1 also has specific effects on the actin cytoskeleton, causing the reduction of stress fibers when overexpressed. The membrane trafficking and cytoskeletal effects of AGAP1 were separable; the former was not dependent on the GLD, and the latter was dependent on the GLD. Both effects required Arf GAP activity. These results raise the possibility that AGAP1 could coordinate changes in actin with endocytic trafficking.
The two structural characteristics that distinguish AGAPs from other ASAP-type Arf GAPs are the GLD and the split PH domain. Although the GLD of AGAP1 is highly similar to K-Ras at 27% identity, it lacks NKXD, the motif involved in binding the nucleotide ring (37). Consistent with this structural deficiency, our data indicate that AGAP1 is not a GTPbinding protein. We cannot exclude the possibility that under a specific set of conditions, the equivalent of NKXD is contrib-uted by another part of AGAP1 or from an interacting protein.
With the high level of homology with K-Ras, we also tested for a role of AGAP1 in the Ras-dependent MAP kinase pathway. However, we did not find any effects.
Although we could not detect Ras-like function (38,39), we did find that GLD affected lipid dependence. Like ASAP1 (40), AGAP1 was activated synergistically by PA and PIP 2 . The presence of the GLD contributed to the effect of PA. With the deletion of the GLD domain, PA was no longer required, nor did it potentiate the effect of PIP 2 for GAP activity. One explanation is that GLD is conferring an inhibitory effect on the Arf GAP activity. Interaction of PA with GLD could remove this inhibition. Another possibility is that the GLD interacts with the PH domain, affecting the lipid binding properties of the latter. PH domains have been found to fold together with adjacent domains in other proteins (41). We have found some indication that this is true for ASAP1 (42). The insert in the PH domain of AGAP1 could mediate this or other interactions and confer another level of regulation of GAP activity.
The association of AGAP1 with Rab4-containing endosomes distinguishes AGAP1 from other Arf GAPs. The localization was dependent on a functional Arf GAP domain. However, Arf binding to AGAP1 is not likely involved in targeting, because [R599K]AGAP1 also binds Arf but does not associate with endosomes. We speculate that the split PH domain of AGAP1 has a role in targeting to the endosomal structures. The colocalization with Rab4 suggests that AGAP1 functions in transport between early and recycling endosomes (43). The colocalization of AP1 also suggests a role in endosome function; however, in this case, the movement would be from the trans-Golgi network (44). The coincidence of AP1 with Rab4 may indicate that the overexpression of AGAP1 slows transport out of the recycling endosome or from some intermediate, causing the accumulation of material in this particular compartment. This "trapping" of an intermediate would explain the colocalization of AP1 and Rab4, which is not normally observed. Furthermore, the distribution of endogenous AGAP1 is still unknown and may not be identical to that of the overexpressed AGAP1 in the endocytic compartment containing Rab4 and AP1. A difference in the distribution of endogenous and ectopically expressed protein is not unprecedented. Different distribution patterns have been observed with endogenous and overexpressed ASAP1. Although ASAP1 is known to associate with and impact the turnover of focal adhesions, overexpressed ASAP1 is not detected in these structures (19,24). Thus, although our data do implicate AGAP1 as a regulator of the endocytic compartment, we need to investigate the cellular distribution of endogenous AGAP1 and its relationship to the observed effects of overexpressed protein.
The other distinguishing effect of AGAP1 is reduction of stress fibers. ARAP1 will induce a loss in stress fiber but only when serum is excluded from the cell culture medium. ASAP1 and ACAP1 had much less of an effect. Like the effects on membrane traffic, the effect on actin required Arf GAP activity. Distinct from the membrane traffic effects, this effect of AGAP1 was dependent on both the GLD and Arf GAP activity. The common requirement for Arf GAP activity could result from a number of mechanisms. For instance, the effect on actin may be secondary to an effect on membrane traffic. However, the difference in GLD dependence indicates a more complex mechanism. Possibly, the common requirement ensures that the membranes and actin are affected coordinately, but a distinct machinery actually remodels actin. The effects of the isolated GLD on cell morphology indicate that this may be the case. We are identifying associated proteins in an effort to further define the molecular mechanisms underlying the cellular effects of AGAP1.
Among the effects of expression of the isolated GLD was the induction of filopodia and cell elongation. These effects raise the possibility that AGAP1 may influence activity of Rho family GTP-binding proteins. Filopodia are also induced by activation of Cdc42, and overexpression of CEP, a CDC42 effector, induces a phenotype that is similar to that seen on overexpression of AGAP1 (45). It is conceivable that the effects of the GLD are mediated by CEP, Cdc42, or another protein in that pathway.
In summary, we have characterized a phosphoinositide-dependent Arf GAP that is associated with Rab4-containing endosomes and affects a specific aspect of actin remodeling. We speculate that this protein coordinates changes in membranes and actin.