Tankyrase Is a Golgi-associated Mitogen-activated Protein Kinase Substrate That Interacts with IRAP in GLUT4 Vesicles*

The poly(ADP-ribose) polymerase tankyrase was originally described as a telomeric protein whose catalytic activity was proposed to regulate telomere function. Subsequent studies revealed that most tankyrase is ac-tually extranuclear, but a discordant pattern of cytoplasmic targeting was reported. Here we used fraction-ation and immunofluorescence to show in 3T3-L1 fibroblasts that tankyrase is a peripheral membrane protein associated with the Golgi. We further colocalized tankyrase with GLUT4 storage vesicles in the juxtanuclear region of adipocytes. Consistent with this colocalization, we found that tankyrase binds specifically to a resident protein of GLUT4 vesicles, IRAP (insulin- responsive amino peptidase). The binding of tankyrase to IRAP involves the ankyrin repeats of tankyrase and a defined sequence ( 96 RQSPDG 101 ) in the IRAP cytosolic domain (IRAP 1–109 ). Tankyrase is a novel signaling tar- get of mitogen-activated protein kinase (MAPK); it is stoichiometrically phosphorylated upon insulin stimulation. Phosphorylation enhances the poly(ADP-ribose) polymerase activity of tankyrase but apparently does not mediate the acute effect of insulin on GLUT4 targeting. Taken together, tankyrase is a novel target of MAPK

The poly(ADP-ribose) polymerase tankyrase was originally described as a telomeric protein whose catalytic activity was proposed to regulate telomere function. Subsequent studies revealed that most tankyrase is actually extranuclear, but a discordant pattern of cytoplasmic targeting was reported. Here we used fractionation and immunofluorescence to show in 3T3-L1 fibroblasts that tankyrase is a peripheral membrane protein associated with the Golgi. We further colocalized tankyrase with GLUT4 storage vesicles in the juxtanuclear region of adipocytes. Consistent with this colocalization, we found that tankyrase binds specifically to a resident protein of GLUT4 vesicles, IRAP (insulinresponsive amino peptidase). The binding of tankyrase to IRAP involves the ankyrin repeats of tankyrase and a defined sequence ( 96 RQSPDG 101 ) in the IRAP cytosolic domain (IRAP 1-109 ). Tankyrase is a novel signaling target of mitogen-activated protein kinase (MAPK); it is stoichiometrically phosphorylated upon insulin stimulation. Phosphorylation enhances the poly(ADP-ribose) polymerase activity of tankyrase but apparently does not mediate the acute effect of insulin on GLUT4 targeting. Taken together, tankyrase is a novel target of MAPK signaling in the Golgi, where it is tethered to GLUT4 vesicles by binding to IRAP. We speculate that tankyrase may be involved in the long term effect of the MAPK cascade on the metabolism of GLUT4 vesicles.
Tankyrase is an ADP-ribose transferase with certain features of both signaling and cytoskeletal proteins (1). Besides the PARP 1 domain, which catalyzes poly(ADP-ribosyl)ation of substrate proteins (1), the sterile ␣ module (SAM) in tankyrase is shared by signaling molecules such as the MAPK kinase kinase Byr2 (2), whereas its ANK domain containing 24 ankyrin repeats is homologous to the cytoskeletal protein ankyrin (3). Tankyrase was identified in a yeast two-hybrid screen where its ANK domain interacts with a telomeric protein, TRF-1 (telomere repeat binding factor-1) (1). Tankyrase was initially described as a telomeric protein, and its PARP activity was proposed to regulate telomere function (1). Subsequent data revealed that tankyrase targeting is cell cycle-dependent and, surprisingly, that only a minute fraction is found at the telomeres (4). In mitotic HeLa cells tankyrase is highly concentrated around the centrosomes. During interphase, the reported pattern varies dramatically with fixation methods (4). With formaldehyde fixation, tankyrase appears to decorate the cytoplasmic side of the nuclear envelope. With methanol fixation, however, a punctate pattern was observed in a cluster near the nucleus (4). Of note, the identity of this juxtanuclear localization was not addressed. Given the complex targeting pattern of tankyrase, we hypothesized that it might interact with proteins besides TRF-1 in extranuclear compartments.
A compartment of interest is defined by "GLUT4 vesicles", i.e. endocytic vesicles in myocytes and adipocytes that contain the glucose transporter GLUT4, and IRAP (the insulin-responsive amino peptidase) (5,6). In insulin-deprived adipocytes, most GLUT4 vesicles are sequestered intracellularly in the trans-Golgi reticulum and also in scattered cytosolic sites. Upon insulin stimulation, a major fraction of GLUT4 vesicles in the trans-Golgi reticulum, and to a smaller extent in cytosolic sites, translocate toward the cell surface (7,8). The translocation inserts GLUT4 and IRAP into plasma membrane, where GLUT4 can facilitate glucose uptake and IRAP can hydrolyze its extracellular substrates (9,10). Following insulin withdrawal, both GLUT4 and IRAP are internalized by endocytosis and resequestered in GLUT4 vesicles. This reversible translocation of GLUT4 allows insulin to regulate glucose utilization, but the physiological function of IRAP remains unclear (10). Because IRAP can hydrolyze vasopressin and other vasoactive peptides (9), impaired translocation of IRAP has been implicated in vascular complications caused by insulin resistance (11).
The targeting machinery for GLUT4 vesicles has not been completely identified, but it presumably interacts with domains in GLUT4 and IRAP that specify their insulin-responsive targeting. The relevant domain(s) within GLUT4 has not been mapped conclusively, presumably because of the multispanning membrane topology of GLUT4 (12)(13)(14). By contrast, IRAP is a type II transmembrane protein with a single cytosolic domain (aa 1-109) that clearly confers insulin responsiveness to its targeting (15). Curiously, overexpression of this IRAP domain causes GLUT4 translocation in adipocytes, suggesting that it can saturate a targeting machinery that also regulates GLUT4 exocytosis (16). The IRAP cytosolic domain is therefore an ideal handle for identifying components of the machinery that targets GLUT4 vesicles.
This study shows that the IRAP cytosolic domain binds to tankyrase in vitro and in vivo. Consistent with the binding, we also show that tankyrase colocalizes with a significant pool of GLUT4 vesicles in 3T3-L1 adipocytes. In contrast to previous studies, we consistently localize tankyrase to the Golgi in 3T3-L1 cells. Tankyrase is likely an important signaling molecule, because it is stoichiometrically phosphorylated by MAPK upon insulin stimulation. The phosphorylation of tankyrase by MAPK increases its PARP activity in vitro. However, the phosphorylation is apparently not involved mechanistically in the acute effect of insulin on GLUT4 translocation. Instead, we propose that tankyrase may mediate the long term regulation of GLUT4 vesicles by the MAPK cascade.

EXPERIMENTAL PROCEDURES
Expression Vectors-Human tankyrase cDNA in pTT20 (1) was FLAG-tagged N-terminally using standard techniques to yield pFLAGtankyrase.
Affinity Precipitation and Immunoprecipitation-GST fusion proteins were induced in Escherichia coli and purified on glutathione-Sepharose beads (Amersham Pharmacia Biotech) as described (18) except that no detergents were used. Tankyrase and ankyrin G 1-850 encoded by pTT20 and pAnkG-AR, respectively, were labeled with [ 35 S]methionine by coupled in vitro transcription/translation (Promega). Labeled proteins were incubated with GST fusion proteins (20 g) at 4°C for 3 h in 100 l of buffer A (200 mM NaCl, 50 mM Tris, pH 8.0, 1 mM EDTA, 10% glycerol, 1% Triton X-100, 1 mM dithiothreitol, 3 mM niacinamide (Sigma), 7.5 g/ml aprotinin, 0.5 mM phenylmethylsulfonyl fluoride, 1 mg/ml leupeptin, 200 M sodium orthovanadate, and 20 mM ␤-glycerophosphate). The precipitants were washed with buffer A (1 ml four times) and separated in 6.5% SDS gels for autoradiography. For co-immunoprecipitation, BOSC cells (19) were transfected with pFLAG-tankyrase, pIRAP-myc, or both vectors using a calcium phosphate kit as recommended by the manufacturer (Invitrogen). Cells were lysed at 48 h in buffer A and clarified at 13,000 ϫ g at 4°C for 10 min. The supernatant was incubated with resins (6 l/10-cm plate) containing anti-FLAG M2 antibody (Sigma) or anti-myc 9E10 antibody (BAbCo) at 4°C for 2 h. The immunoprecipitants were immunoblotted after washing in buffer A and buffer A containing 500 mM NaCl for 1 h. For antibody characterization, BOSC cells were similarly transfected with pFLAG-tankyrase or pFLAG-tankyrase-2, and the overexpressed proteins were purified using anti-FLAG affinity resin.
Indirect Immunofluorescence-Cells grown on coverslips were rinsed in PBS and fixed in methanol (5 min) and acetone (2 min) at Ϫ20°C.
Fibroblast Lysates-Confluent 3T3-L1 fibroblasts (20) grown in DMEM containing 10% fetal bovine serum were serum-starved in DMEM for 2 h and stimulated when indicated with insulin (1 g/ml, Sigma) for 10 min at 37°C. Cells were lysed in buffer A, mixed with SDS-PAGE sample buffer, and passed through QIAshredder (Qiagen) to obtain total lysates. For subcellular fractionation, fibroblasts grown in DMEM containing 10% fetal bovine serum were scraped at 4°C in 250 mM sucrose, 10 mM Tris 7.5, 0.5 mM EDTA, and the same protease inhibitors as in buffer A (1 ml/15-cm plate). Cells were homogenized in a tight-fitting Dounce grinder and clarified at 15,000 ϫ g for 10 min. The supernant was centrifuged at 50,000 ϫ g for 20 min to recover high density microsomes. The supernatant was again centrifuged at 160,000 ϫ g for 70 min to separate low density microsomes from soluble proteins. For solubility studies, fibroblasts grown in DMEM containing 10% fetal bovine serum were Dounce-homogenized in 100 mM Na 2 CO 3 (pH 11.5) or in 1 M NaCl, 1 mM EDTA, and 50 mM Tris, pH 8 (750 l/6-cm plate) at 4°C, and soluble proteins were separated from the insoluble by centrifugation at 300,000 ϫ g for 30 min. For PARP assays, serum-starved and insulin-stimulated fibroblasts were washed in PBS and lysed in buffer A (0.5 ml/15 ml plate). After preclearing by incubating with 40 g of GST in buffer A at 4°C for 2 h, tankyrase was affinity precipitated overnight using 40 g of wild-type GST-IRAP 2-109 or mutant GST-IRAP 2-109 containing a G101A substitution. The precipitant was washed in buffer N (150 mM NaCl, 50 mM Tris, pH 8, 5 mM MgCl 2 , 10% glycerol, 1 mM dithiothreitol, and the same protease inhibitors as in buffer A but without benzamide) and incubated with 1 M [adenylate-32 P]NAD (4 Ci/mmol) in 100 l of buffer N at 37°C for 90 min. The products were resolved in 6.5%/10% discontinuous gradient gels and detected by autoradiography and phosphorimaging (Fuji). Some products were also immunoblotted with antibodies against polymers of ADP-ribose (SA216, 1:1500, BioMol) and against tankyrase (T1S).

RESULTS
Tankyrase Binds to the Cytosolic Domain of IRAP-To identify novel components of the targeting machinery for GLUT4 vesicles, we screened a skeletal muscle cDNA library for proteins that interact with the IRAP cytosolic domain (aa 1-109) GLUT4 Vesicles Associate with a MAPK Substrate in the Golgi in a yeast two-hybrid system. This revealed a specific interaction of IRAP with the ANK repeat domain (aa 1-810) of tankyrase-2, a novel tankyrase homologue to be described elsewhere. 2 The ANK domain of tankyrase-2 shows 95% similarity to tankyrase, 3 suggesting that tankyrase may also interact with the IRAP cytosolic domain.
The interaction was verified in vitro, where resins containing GST fused to the entire IRAP cytosolic domain (IRAP 2-109 ) efficiently precipitated 35 S-labeled tankyrase (Fig. 1A, lane 5). Fig. 1A also shows that GST fused to the juxtamembranous hexapeptide RQSPDG (IRAP 96 -101 ) bound tankyrase as efficiently as GST fused to the entire IRAP cytosolic domain (lane 4), whereas the binding was abolished when this hexapeptide was shortened from either end (data not shown). We also examined the binding of tankyrase to the telomere repeat binding factor, TRF-1, because TRF-1 1-68 has been shown to interact with tankyrase in the yeast two-hybrid system (1). This interaction, however, has not been verified biochemically. Fig. 1A shows that tankyrase indeed bound efficiently to GST-TRF-1 1-68 (lane 6). The observed binding was specific, as tankyrase did not bind to GST, GST fused to IRAP cytosolic fragments (IRAP 2-52 and IRAP 55-82 ) lacking the 96 RQSPDG 101 hexapeptide (Fig. 1A, lanes 1-3), or GST fused to GLUT4 C-terminal cytosolic tail (aa 465-509; data not shown). Conversely, neither GST-IRAP nor GST-TRF-1 bound to the 24 ANK repeats in the ANK domain of ankyrin (Fig. 1B), which is homologous to the ANK domain of tankyrase.
To show binding of tankyrase to IRAP in vivo, FLAG-tagged tankyrase and myc-tagged IRAP were transiently expressed either alone or in combination in BOSC cells. The expression of full-length tankyrase and IRAP was confirmed in Fig. 2 (A and  B, respectively). Fig. 2C shows that IRAP co-immunoprecipitated with tankyrase only when both proteins were expressed simultaneously (lane 3).
Tankyrase Colocalizes with a Golgi Marker-The tankyrase antibody used in previous immunofluorescence studies revealed a discordant pattern of tankyrase targeting that varied drastically with fixatives (4). Moreover, the antibody was raised against a region in tankyrase (aa 973-1149) (1) that shows 68% identity to tankyrase-2 (aa 822-996) 3 and thus can potentially cross-react with tankyrase-2. We therefore used the tankyrase-specific antiserum, T1S, to re-examine tankyrase localization by indirect immunofluorescence.
In formaldehyde-fixed 3T3-L1 fibroblasts, wide field microscopy showed that most tankyrase was excluded from the nucleus and formed a Golgi-like pattern near the nucleus (Fig.  4A). This juxtanuclear pattern was tankyrase-specific, because it was not observed with control immunoglobulin (Fig. 4B). To show that tankyrase is indeed localized to the Golgi regardless of fixation methods, cells were also methanol-fixed and costained for tankyrase and the Golgi marker FTCD (21). Fig. 4C shows that methanol preserved the juxtanuclear pattern of tankyrase (left panel), which overlapped extensively with FTCD (right panel). Therefore, our data consistently localized tankyrase to the Golgi.
Tankyrase Is Dispersed in Vivo by Golgi-disrupting Agents-The Golgi apparatus is disintegrated in cells treated with brefeldin A or nocodazole (22). These agents were therefore used to confirm the Golgi localization of tankyrase. In brefeldin-treated 3T3-L1 fibroblasts, tankyrase was dispersed into a fine, punctate pattern throughout the cytosol (Fig. 4D). By comparison, tankyrase was dispersed into a coarser granular pattern with nocodazole treatment (Fig. 4E), presumably because nocodazole disrupted the Golgi using a different mecha-

FIG. 2. Tankyrase co-immunoprecipitates with IRAP in transfected cells.
BOSC cells were transfected with pFLAG-tankyrase, pIRAP-myc, or both vectors (lanes 1-3, respectively) as described under "Experimental Procedures." Lysates were immunoprecipitated with an anti-FLAG affinity resin and immunoblotted with anti-FLAG (A) or anti-myc antibody (C). Lysates were also immunoprecipitated with an anti-myc affinity resin and immunoblotted with an anti-myc antibody (B).
nism (22). Of note, even with the disruption of the Golgi, tankyrase remained virtually excluded from the nucleus (Fig. 4, D and E).
Tankyrase Cofractionates with a Golgi Marker-Golgi membranes cofractionate with the low density microsomes when cellular components are resolved by differential centrifugation (23). To confirm the Golgi localization of tankyrase, we fractionated 3T3-L1 fibroblasts and recovered the majority of tankyrase in the low density microsome fraction (Fig. 5A, lane   3, upper panel). The soluble fraction also contained a significant amount of tankyrase, suggesting that tankyrase was not completely membrane-associated (Fig. 5A, lane 4, upper panel). Like tankyrase, the Golgi marker FTCD was recovered in both the low density microsome and soluble fractions, consistent with FTCD being a peripheral membrane protein in the Golgi (Fig. 5A, lanes 3 and 4,

lower panel, and Ref. 21).
We also examined the solubility of tankyrase to see whether it associated with membranes as a peripheral or an integral membrane protein. Peripheral, but not integral, membrane proteins can be extracted by alkaline solutions of Na 2 CO 3 (24). Furthermore, only peripheral membrane proteins are soluble in the absence of detergents (24). Fig. 5B shows that tankyrase was mostly soluble in a high salt buffer that contained no detergent (lanes 1 and 2), and it was quantitatively extracted by Na 2 CO 3 (lanes 3 and 4). By contrast, the integral membrane protein IRAP was completely insoluble in both conditions (Fig.  5B). Our data therefore indicate that tankyrase is peripherally associated with Golgi membranes.
Tankyrase Colocalizes with Juxtanuclear GLUT4 in Adipocytes-Previous studies have shown that a significant pool of GLUT4 vesicles are closely associated with the trans-Golgi network (7,8). Because tankyrase is a Golgi-associated protein that binds to IRAP (a resident of GLUT4 vesicles), we expected it to colocalize with GLUT4 near the Golgi. 3T3-L1 adipocytes were therefore co-stained for tankyrase and GLUT4, and 0.8-m optical sections were obtained using confocal micros-

FIG. 3. Specificity of anti-tankyrase antibody.
A, BOSC cells were mock transfected or transfected with pFLAGtankyrase or tankyrase-2 as indicated, and recombinant proteins were immunoblotted with T1S, T2S, T12, or anti-FLAG antiserum as described under "Experimental Procedures." B, total lysates of insulin-stimulated 3T3-L1 fibroblasts (lane 1) and affinity purified FLAG-tagged tankyrase (lane 2) as prepared in A were immunoblotted with T1S as described under "Experimental Procedures."

FIG. 4. Tankyrase is a Golgi protein in fibroblasts.
3T3-L1 fibroblasts in A and B were fixed in formaldehyde and stained with T1S (A, red) or nonimmune rabbit ␥ globulin (B, red) and counterstained with DAPI (blue) as described under "Experimental Procedures." Cells in C were fixed in methanol and stained for both tankyrase (red) and FTCD (green). Cells in D and E were pretreated with brefeldin A (10 g/ml; 10 min) and nocodazole (33 M; 1 h), respectively, and stained for tankyrase (red) after methanol fixation. copy. Fig. 6A shows that in serum-starved adipocytes, tankyrase colocalized with the juxtanuclear pool of GLUT4 but not with the GLUT4 scattered in peripheral regions of the cytoplasm. Upon insulin stimulation, some GLUT4 was recruited from intracellular sites toward the cell surface (arrows in Fig. 6B). However, insulin did not affect the Golgi localization of tankyrase or its colocalization with the GLUT4 that remained near the nucleus (Fig. 6B). The observed colocalization was GLUT4-specific, because GLUT1 did not colocalize with tankyrase as shown in a wide field micrograph of adipocytes (Fig. 6C). We could not address the colocalization of tankyrase with IRAP by indirect immunofluorescence, because available antibodies for both proteins were all raised in rabbits.
Tankyrase Is a MAPK Substrate-Given that tankyrase binds to IRAP and colocalizes with a pool of GLUT4 vesicles, we hypothesized that tankyrase might constitute the insulin-regulated targeting machinery for GLUT4. We therefore examined the effect of insulin on endogenous tankyrase in 3T3-L1 cells. Fig. 7A shows that tankyrase obtained from serum-starved fibroblasts migrated as a 165-kDa protein (lanes 1). Interestingly, the apparent M r shifted to 175 kDa within 10 min of insulin stimulation (lanes 2). Similar experiments in adipocytes showed that insulin, PDGF, and EGF elicited an indistinguishable M r shift in tankyrase (Fig. 7B, lanes 1, 2, 6, and 7). The completeness of the shift indicated a stoichiometric modification on tankyrase upon growth factor stimulation. These factors (insulin, PDGF, and EGF) activate two major kinase pathways: the phosphatidylinositol 3-kinase pathway, which can be blocked by LY294002 or wortmannin, and the MAPK cascade, where certain MAPK kinases (e.g. MAPK kinases 1 and 2) can be inhibited by U0126 (25,26). Among these inhibitors, only U0126 blocked the insulin-induced M r shift of tankyrase (Fig. 7B, lanes 3-5), suggesting that tankyrase is a target of the MAPK cascade.
The M r of tankyrase can be increased by auto-poly(ADPribosyl)ation (1). However, the insulin-induced 165 kDa to 175 kDa shift was entirely due to phosphorylation by MAPK for several reasons. First, the 175-kDa form of tankyrase affinity purified from insulin-stimulated cells was quantitatively converted to 165 kDa by in vitro phosphatase treatment (Fig. 7C). Conversely, the 165-kDa form of tankyrase affinity purified from serum-starved cells was shifted to 175 kDa in vitro in the presence of activated MAPK and [ 32 P]ATP. In this kinase reaction, tankyrase became 32 P-labeled as expected (data not shown). Interestingly, both the in vitro phosphatase and kinase reactions involved a transitional species of tankyrase with a distinct M r (arrowheads in Fig. 7, C and D). This suggests that MAPK phosphorylated the same residues of tankyrase in vitro as were phosphorylated in vivo upon insulin stimulation. Of note, equivalent amounts of tankyrase were affinity precipitated from serum-starved versus insulin-stimulated cells by GST-IRAP 78 -109 (Fig. 7B, lanes 1 and 2, and data not shown), suggesting that tankyrase phosphorylation did not affect its binding to GST-IRAP 78 -109 . Whether binding to endogenous IRAP is affected by tankyrase phosphorylation remains to be addressed, because we have been unable to consistently coimmunoprecipitate endogenous IRAP with tankyrase.
Tankyrase phosphorylation was further characterized in adipocytes labeled with [ 32 P] in vivo, again using GST-IRAP 78 -109 to affinity purify tankyrase. Fig. 8A shows that insulin in-FIG. 6. Tankyrase colocalizes with GLUT4 near the nucleus in adipocytes. Serum-starved (A) and insulinstimulated 3T3-L1 adipocytes (B) were costained for tankyrase (red) and GLUT4 (green) and examined under a confocal microscope as described under "Experimental Procedures." Arrows in B indicate the plasma membrane. In C, 3T3-L1 adipocytes grown in DMEM/10% fetal bovine serum were co-stained for tankyrase (red) and GLUT1 (green) and examined under a wide field microscope as described under "Experimental Procedures." The merged images show overlapping red and green pixels in yellow.
creased the phosphorylation state of tankyrase (lanes 3 and 4), and the increase was abolished by the MAPK kinase inhibitor U0126 (lane 5), consistent with tankyrase being a MAPK substrate. In keeping with the serine/threonine specificity of MAPKs (27), phosphoamino acid analysis of tankyrase revealed only phosphoserine residues (Fig. 8B). Taken together, our data show that tankyrase is quantitatively phosphorylated on certain serine residues by MAPK upon stimulation with insulin, PDGF, and EGF.
The PARP Activity of Tankyrase Is Enhanced by Phosphorylation-Having identified tankyrase as a novel MAPK substrate, we wondered whether tankyrase phosphorylation might serve to regulate its PARP activity. This would allow tankyrase to transduce MAPK signaling into poly(ADP-ribosyl)ation of effector proteins. The physiological substrates of tankyrase remain unidentified. However, tankyrase can use NAD as a cofactor in vitro to poly(ADP-ribosyl)ate itself (i.e. automodification) and its interacting protein, TRF-1 (1). Another candidate substrate besides TRF-1 is IRAP, because IRAP also interacts with tankyrase ( Figs. 1 and 2). To determine whether phosphorylation affects the PARP activity of tankyrase, endogenous tankyrase was affinity purified from 3T3-L1 fibroblasts using GST-IRAP 2-109 . Equivalent amounts of tankyrase were recovered from serum-starved versus insulin-stimulated cells, and tankyrase from neither source was automodified as judged by anti-tankyrase and anti-poly(ADP-ribose) immunoblots (data not shown). However, when the GST-IRAP⅐tankyrase complex was incubated with 32 P-labeled NAD, both proteins in the complex became poly(ADP-ribosyl)ated and labeled with 32 P (Fig. 9, lanes 1 and 2, and data not shown). The radiolabeling was on average 40% more intense with tankyrase obtained from insulin-stimulated cells (n ϭ 6) (Fig. 9, lane 1  versus lane 2). Fig. 9 also shows that the PARP inhibitor niacinamide (28) effectively abolished the radiolabeling of both tankyrase and GST-IRAP (lanes 3 and 4), confirming that the labeling resulted from poly(ADP-ribosyl)ation. To show that the PARP reaction is tankyrase-dependent, we introduced a Gly 101 -to-Ala mutation into the tankyrase-binding site ( 96 RQSPDG 101 ) of GST-IRAP 2-109 . This mutant GST-IRAP 2-109 failed to precipitate any tankyrase protein or PARP activity (Fig. 9, lanes 5 and 6, and data not shown). Our data therefore indicate that the PARP activity of tankyrase toward itself (automodification) and GST-IRAP was enhanced by MAPK phosphorylation. Similar results were obtained when tankyrase was affinity purified using GST-TRF-1 1-68 instead of GST-IRAP (data not shown).

DISCUSSION
In an effort to identify proteins that interact with GLUT4 vesicles, we discovered that tankyrase binds to the cytosolic domain of IRAP, a resident protein of GLUT4 vesicles. Our data suggest that tankyrase is a peripheral membrane protein in the Golgi, and it colocalizes with a subpopulation of GLUT4 vesicles. We also show that tankyrase is a novel target of MAPK cascade, and its PARP activity is enhanced by MAPK phosphorylation following insulin stimulation.
Previous studies revealed that tankyrase is overwhelmingly extranuclear but did not provide a congruent pattern of targeting (4). Using an antibody that does not cross-react with tankyrase-2, we consistently localized tankyrase to the Golgi region in 3T3-L1 fibroblasts and adipocytes. The Golgi localization as seen by immunofluorescence is also collaborated by subcellular fractionation. Consistent with the Golgi targeting, tankyrase colocalizes with the juxtanuclear pool of GLUT4 vesicles, which are known to closely associate with the trans-Golgi network (7). We suspect that the colocalization of tankyrase with GLUT4 reflects tankyrase binding to IRAP that resides in GLUT4 vesicles.
Tankyrase interacts with a membrane-proximal hexapeptide (aa 96 -101) in the IRAP cytosolic domain (IRAP 1-109 ). This  7), tankyrase was affinity purified using GST-IRAP 78 -109 , resolved by SDS-PAGE, and immunoblotted with T12 as described under "Experimental Procedures." C, in vitro dephosphorylation. Tankyrase affinity purified from insulin-stimulated adipocytes using GST-IRAP 78 -109 was incubated with protein phosphatase-1 for 0, 15, or 70 min (lanes 1-3, respectively) as described under "Experimental Procedures." The products were immunoblotted with T12 along with tankyrase purified from serum-starved adipocytes (lane 4). Arrowhead indicates a transitional species of tankyrase. D, in vitro phosphorylation. Tankyrase affinity purified from serum-starved adipocytes using GST-IRAP 78 -109 was incubated with 0, 50, or 250 units of p42 MAPK (lanes 1-3, respectively) for 1 h as described under "Experimental Procedures." The products were immunoblotted with T12 along with tankyrase purified from insulin-stimulated adipocytes (lane 4). Arrowhead indicates a transitional species of tankyrase. hexapeptide is apparently not the only binding motif for tankyrase, because tankyrase also binds to TRF-1 1-68 , which does not contain a similar motif. 3 Tankyrase therefore resembles ankyrin in that the ANK domain in ankyrin interacts with a membrane-proximal heptapeptide in ␣-Na,K-ATPase and also binds to other unrelated proteins (3,29,30). Ankyrin is a peripheral membrane protein that tethers membrane channels and receptors to the submembranous cytoskeleton (31), whereas its Golgi isoforms are implicated in vesicular trafficking by tethering Golgi membranes to motor proteins (32). By analogy, IRAP binding may enable tankyrase to tether GLUT4 vesicles to cytoskeletal or motor proteins. Given that ANK domains can bind to diverse proteins (29), tankyrase may provide a tethered scaffold where GLUT4 vesicles interact with signaling molecules or targeting machinery.
To explain how insulin induces GLUT4 translocation, an intracellular retention system has been postulated that sequesters GLUT4 vesicles in the basal state and releases them upon insulin stimulation (5). The retention can apparently be saturated by both the IRAP cytosolic domain and the GLUT4 Cterminal cytosolic tail, because either domain when overexpressed can cause GLUT4 translocation (16,33). Our results do not implicate tankyrase as a component of this retention machinery, because tankyrase binds to the IRAP cytosolic domain in a region (aa 96 -101) distinct from those that cause GLUT4 translocation when overexpressed (aa 1-52 and 55-82) (16). Secondly, tankyrase does not bind to the GLUT4 C-terminal cytosolic tail in vitro. Lastly, despite causing tankyrase phosphorylation, insulin does not regulate binding of tankyrase to GST-IRAP in vitro. It remains possible that endogenous IRAP binding to tankyrase is regulated by insulin, perhaps through reversible modifications on IRAP. However, this seems unlikely given that insulin does not abolish the colocalization of tankyrase with GLUT4 in vivo (Fig. 6B).
The phosphorylation of tankyrase upon insulin stimulation is stoichiometric, suggesting that tankyrase is an important insulin signaling target. However, despite the proximity of tankyrase to GLUT4, tankyrase phosphorylation and GLUT4 translocation are apparently regulated by divergent signaling cascades: tankyrase phosphorylation depends on MAPK, whereas GLUT4 translocation depends on phosphatidylinositol 3-kinase (10,34). Moreover, PDGF and EGF stimulate tankyrase phosphorylation but do not cause GLUT4 translocation (10). Therefore, tankyrase phosphorylation by MAPK alone does not explain how insulin stimulates GLUT4 translocation.
Although the ras-MAPK cascade does not mediate the acute effect of insulin on GLUT4 targeting, it apparently affects long term regulation of GLUT4 metabolism. When the cascade is constitutively activated by mutant alleles of ras or MAPK kinase, adipocytes recruit more GLUT4 to their cell surface despite a lowered cellular GLUT4 content (Refs. 35 and 36, but also see Refs. 37 and 38). No molecular mechanism has been proposed to link the MAPK cascade directly with GLUT4 vesicles. Given that tankyrase is a MAPK target in close proximity to GLUT4, its PARP activity may conceivably couple the MAPK cascade to GLUT4 vesicles. Our data show that phosphorylation by MAPK enhances the PARP activity of tankyrase as judged by in vitro automodification. The biological significance of this enhancement is unclear, because endogenous tankyrase is not detectably automodified. It is possible that the PARP activity of tankyrase on its physiological substrates is also regulated by MAPK phosphorylation. This would allow MAPK to modulate through tankyrase the poly(ADP-ribosyl)ation of Golgi-associated proteins that affect the sorting or stability of GLUT4 vesicles.
A precedent for ADP-ribosylation to regulate Golgi dynamics involves the G protein BARS (brefeldin ADP-ribosylation substrate). BARS is a multifunctional protein that can acylate lysophosphatidic acid in the Golgi membranes to promote FIG. 8. Tankyrase is phosphorylated on serine residues. A, in vivo 32 P labeling of tankyrase. Serum-starved 3T3-L1 adipocytes were labeled in vivo with 32 P, pretreated with U0126 (lane 5), and stimulated with insulin (lanes 2, 4, and 5) when indicated as described under "Experimental Procedures." Lysates were affinity purified using GST (lanes 1 and 2) or GST-IRAP 78 -109 (lanes [3][4][5], and bound proteins were resolved by SDS-PAGE and autoradiographed. B, phosphoamino acid analysis. Bracketed regions in A representing [ 32 P]tankyrase from serum-starved and insulin-stimulated cells (lanes 3 and 4, respectively) were excised for phosphoamino acid analysis as described (41). The positions of phosphoamino acids, P i , and an unidentified species (*) that was not consistently detected are marked.