Phosphorylation of Glutamate Receptor Interacting Protein 1 Regulates Surface Expression of Glutamate Receptors*

The number of synaptic α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid (AMPA)-type glutamate receptors (AMPARs) controls the strength of excitatory transmission. AMPARs cycle between internal endosomal compartments and the plasma membrane. Interactions between the AMPAR subunit GluR2, glutamate receptor interacting protein 1 (GRIP1), and the endosomal protein NEEP21 are essential for correct GluR2 recycling. Here we show that an about 85-kDa protein kinase phosphorylates GRIP1 on serine 917. This kinase is present in NEEP21 immunocomplexes and is activated in okadaic acid-treated neurons. Pulldown assays and atomic force microscopy indicate that phosphorylated GRIP shows reduced binding to NEEP21. AMPA or N-methyl-d-aspartate stimulation of hippocampal neurons induces delayed phosphorylation of the same serine 917. A wild type carboxy-terminal GRIP1 fragment expressed in hippocampal neurons interferes with GluR2 surface expression. On the contrary, a S917D mutant fragment does not interfere with GluR2 surface expression. Likewise, coexpression of GluR2 together with full-length wild type GRIP1 enhances GluR2 surface expression in fibroblasts, whereas full-length GRIP1-S917D had no effect. This indicates that this serine residue is implicated in AMPAR cycling. Our results identify an important regulatory mechanism in the trafficking of AMPAR subunits between internal compartments and the plasma membrane.

AMPA-type glutamate receptors mediate most of the fast excitatory synaptic transmission in the brain. Various forms of synaptic plasticity depend on changes in AMPA receptor (AMPAR) 2 transmission. Trafficking of AMPAR and their lat-eral diffusion to and from synapses are key mechanisms that govern synaptic strength (1)(2)(3)(4)(5). Induction of long-term depression in hippocampal cell cultures, hippocampal slice preparations, and cultured cerebellar Purkinje cells is associated with a reduction in synaptic AMPAR (6 -10). Likewise, the induction of long-term potentiation yields an increase in the number of AMPAR at synapses (11,12).
AMPAR are hetero-oligomers composed of subunits GluR1 to GluR4 (GluRA to GluRD). In the hippocampus GluR1/GluR2 or GluR2/GluR3 receptors predominate (13). Although the extracellular domain and the four membrane spanning domains are largely homologous between the different subunits, the cytoplasmic carboxyl termini differ, with long termini for GluR1 and GluR4 and short termini for GluR2 and GluR3. These cytoplasmic tails are responsible for subunit-specific AMPAR trafficking to synapses (14), which is mediated through binding of PDZ (postsynaptic density 95/disc large/ zonula occludens-1) domain proteins. GluR2/GluR3 receptors cycle constitutively between internal compartments and the plasma membrane (14,15), presumably at sites distant from the synapse (16). The interaction between GluR2 and type II PDZ domain proteins like protein interacting with C kinase 1 (PICK1), glutamate receptor interacting protein (GRIP), and AMPAR-binding protein (ABP) is necessary for this cycling (17). After internalization by clathrin-dependent endocytosis, AMPARs undergo selective sorting between recycling and degradative pathways, with AMPA stimulation inducing the former and NMDA stimulation inducing the latter pathway (18). In addition, phosphorylation and dephosphorylation of the different subunits play an important role in AMPAR trafficking (10, 19 -23). NMDA receptor-dependent internalization of AMPAR activates protein phosphatases and leads to a rapid reinsertion of AMPAR into the plasma membrane (24).
We have previously shown that NEEP21 interacts with a carboxyl-terminal site of GRIP1 and forms a regulated complex with GRIP1, GluR2, and the endosomal trafficking SNARE protein syntaxin 13 (38). NEEP21 is localized to endosomes in the soma and dendritic shafts and is essential for correct receptor recycling and synaptic transmission (39 -41). Previous studies proposed a possible phosphorylation of GRIP1 (28,31). Here we show that GRIP1a is indeed phosphorylated on serine 917 in the linker region between amino acids (aa) 881 and 933. This phosphorylation modulates the binding of GRIP1 to NEEP21 and alters surface expression of GluR2.
Immunoprecipitation, Surface Biotinylation, and Cell Fractionation-Preparation of brain membrane extracts from postnatal day 10 rats and covalent cross-linking of antibodies to protein A or G beads were done as previously described (38). Antibody beads (10 l) were incubated with membrane extract (2 mg) for 4 h and washed 3 times with buffer B (20 mM HEPES/ KOH, pH 7.4, 2 mM EDTA, 2 mM EGTA, 0.1 mM dithiothreitol, 0.1 M KCl, 1% Triton X-100). Bound proteins were either glycine-eluted and neutralized by Tris-buffer for in vitro kinase assays, or sample buffer was added for Western blotting. For GST pulldown experiments 4 dishes of neurons were treated with the indicated drugs at 37°C, washed in phosphate-buffered saline, and lysed in buffer B. The extract was incubated for 2 h at 4°C with GST fusion protein beads. After washing bound proteins were analyzed by Western blots. Surface biotinylation was done as previously described (47,48). Briefly, cells were biotinylated with 1 mg/ml sulfo-NHS-biotin (4°C, 30 min). Cells were then lysed, and biotinylated proteins were precipitated with streptavidin-linked beads. For fractionation of hippocampal neurons, post-nuclear supernatants were prepared by cell lysis in buffer B without detergent by 10 strokes through a G25 needle followed by a 1000 ϫ g centrifugation. It was further separated into a cytosolic fraction by centrifugation for 60 min at 100,000 ϫ g. To recover the membrane fraction, the pellet was rehomogenized, solubilized by 1% Triton X-100, and submitted to another 100,000 ϫ g centrifugation.
In Vitro Kinase Assay-Fusion proteins were immobilized on glutathione-agarose beads and incubated for 8 min at 30°C in the presence of [␥-32 P]ATP (10 Ci, Amersham Biosciences) in phosphorylation buffer with either 10 g of neuron extract or eluted proteins from IgG or anti-NEEP21-immunoprecipitations. When neuron extracts were used cells were either untreated or incubated with 100 nM okadaic acid (OA) or 1 M calphostin C or 1 M PMA for 1 h. Stimulation with 100 M AMPA or 50 M NMDA was for 3 min followed by further incubation for 7 or 17 min. When immune pellets were used, PMA (1 M) was added to the phosphorylation reaction where indicated. Beads were then washed once with phosphorylation buffer, and bound proteins were separated on SDS-PAGE and stained with Coomassie Blue (Sigma-Aldrich), and the dried gel was exposed to Kodak X-Omat film (Amersham Biosciences).
Interaction Force Measurement by Atomic Force Microscopy (AFM)-AFM tips and substrates were prepared as previously described (42). Briefly, GST fusion proteins (200 -300 ng/l) were covalently cross-linked by glutaraldehyde (0.5%) to the AFM tip (Digital Instruments, Santa Barbara; nominal spring constant 0.06 newton/m, calibrated by thermal noise analysis) and to a freshly cleaved mica sheet that had been functionalized by aminopropyltriethoxysilane. This method had previously been verified to keep the functionality of proteins intact (43).
Experiments were performed at room temperature on a Nanoscope III (Digital Instruments) with force volume mode operating in Tris-buffered saline or injected neuron cell extract in phosphorylation buffer containing 300 M ATP at a constant retraction speed of 355 nm/s. Force curves were analyzed offline by a fuzzy logic algorithm developed in our laboratory (44).
Immunocytochemistry-Okadaic acid (OA)-treated or transfected neurons were fixed in 4% paraformaldehyde, 4% sucrose for 5 min for surface staining or for 12 min for intracellular staining. For surface staining an extracellularly binding anti-GluR2 or anti-GluR1 antibody was added in the absence of detergent. For intracellular staining antibodies were added in presence of 0.3% Triton X-100 for 2 h. Neurons were washed in 66 mM NaCl, 150 mM Tris HCl, pH 7.4, and secondary antibodies were added for 1 h. After washing coverslips were mounted in 50% glycerol containing Mowiol (Fluka) and DABCO (1,4-diazabicyclo(2.2.2)octane; Sigma-Aldrich) to retard photobleaching.
Quantitative Image Analysis-For quantification of surface GluR1 and GluR2 labeling, 10 confocal sections were acquired with identical parameters; an average projection was applied resulting in the average fluorescence intensity for each pixel. The fluorescence intensity was quantified on the cell body. A threshold of 80 was applied, and the red pixels summed using Metamorph software. The sum of the red pixels was divided by the size of the perimeter of the cells to normalize for different cell sizes. Colocalization was defined as the pixels that are positive for both GluR2 (see Fig.  1E, red) and NEEP21 (green). 100% in Fig. 1E corresponds to the sum of pixels that are red or green. The following numbers of cells n were analyzed in Fig. 1 Site-directed Mutagenesis-In vitro site-directed mutagenesis was performed using the QuikChange TM site-directed mutagenesis kit (Stratagene, La Jolla, CA) according to the manufacturer's protocol. Template DNA was either full-length rat GRIP1 or GFP-GRIP1-(810 -1112). The antisense primers for the different mutated sites were as follows:  Corresponding DIC images are shown on the lower row. C, fluorescence was quantified on confocal images. 100% corresponds to integrated fluorescence intensity of untreated cells. After OA treatment, GluR2 surface expression (black bars) decreased significantly (**, p Ͻ 0.01; n, please refer to the "Experimental Procedures"), whereas GluR1 surface expression (gray bars) did not show a significant change. D, total extracts of hippocampal neurons treated as above were analyzed for GluR2 by Western blotting. Reprobing for actin verified equal loading. The total amount of GluR2 (surface plus internal) remained unchanged after OA treatment. E, treatment of hippocampal neurons with 100 nM OA for 1 h shifts some GluR2 to endosomal compartments. Colocalization of GluR2 and the endosomal NEEP21 was calculated as the ratio of the pixels positive for both NEEP21 and GluR2 divided by all pixels, i.e. those positives for either NEEP21 or GluR2. There is a significant increase in colocalization (**, p Ͻ 0.01; n ϭ 63). Typical cells are shown on the left. Bars, 20 m.

RESULTS
The Phosphatase Inhibitor Okadaic Acid Decreases Surface GluR2 in Hippocampal Neurons-Endocytosis of AMPAR and subsequent sorting to either degradation or recycling back to the plasma membrane depends on differential stimuli and protein phosphatase cascades (18,24). To analyze the implication of phosphorylation in AMPAR surface expression, we treated hippocampal neurons at days in vitro 14 with the protein phosphatase 1 and 2A inhibitor OA. Then we performed surface labeling using extracellularly binding antibodies for the AMPAR subunits GluR1 or GluR2 (Fig. 1, A and B, respectively, upper row; corresponding DIC images are on the lower row). After quantitative image analysis, we found that treatment for 1 h with 100 nM OA resulted in a strong reduction of surface GluR2 (62.6%) compared with control untreated neurons (100%; p Ͻ 0.01). This lower GluR2 level persisted upon prolonged treatments of 2 or 3 h (Fig. 1C, black bars). Interestingly, the same OA treatments had no significant effect on the surface expression of GluR1 (Fig. 1C, gray bars). The total GluR2 content of the neurons was not altered, as verified by Western blotting of OA-treated neurons (Fig.  1D), excluding an effect on gene expression or degradation. Instead, this result indicated a change in GluR2 trafficking between the neuron surface and intracellular compartments. Costaining for the presynaptic marker synaptophysin indicated that GluR2 was localized to synaptic as well as extra-synaptic sites, and this distribution did not change upon OA treatment (data not shown).
To examine the site to which GluR2 redistributed after OA treatment, we performed double-immunolabeling for GluR2 and NEEP21. The latter protein of neuronal early endosomes associates with the endosomal SNARE protein syntaxin 13, the scaffolding protein GRIP, and GluR2, and it regulates GluR2 sorting (38,39). OA treatment significantly increased the colocalization between GluR2 and NEEP21 compared with control neurons (Fig. 1E; from 5.4 to 12.6%; p Ͼ 0.01). These results indicate that OA treatment causes a shift from surface GluR2 to intracellular GluR2. It also suggests that protein dephosphorylation is involved in trafficking of GluR2-containing AMPAR.
The Interaction between GRIP and NEEP21 Is Altered by OA Treatment of Neurons-It has been suggested previously that phosphorylation of GRIP might regulate targeting of surface receptors (28,31). To assess whether OA treatment leads to a modification of GRIP correlating with the change in surface GluR2 localization, we analyzed the electrophoretic mobility of GRIP in total extract of OA-treated hippocampal neurons by Western blot. After 1 h of exposure, OA induced a dose-dependent retardation of GRIP migration consistent with a potential GRIP phosphorylation ( Fig. 2A). We then analyzed by GST pulldown assays whether this OA-induced, slower-migrating form of GRIP has altered binding to immobilized fusion protein between GST and the cytosolic domain of NEEP21 spanning amino acids 104 -185 (GST-NEEP21ct). GRIP was efficiently pulled down by GST-NEEP21ct beads from extracts of non-treated neurons and extracts of neurons treated with 10 nM OA (Fig. 2B). In contrast, treatment of neurons with 100 nM or 1 M OA strongly decreased the binding of GRIP to NEEP21ct FIGURE 2. Binding between GRIP and NEEP21ct is decreased by the phosphorylation of GRIP. A and B, total extracts of hippocampal neurons were either nontreated or treated with the indicated concentrations of OA and subjected to pull-down assays on beads with immobilized GST or GST-NEEP21ct (the cytosolic domain of NEEP21 spanning aa 104 -185). Bound proteins were analyzed for GRIP on Western blots. Total extracts before binding are shown in A that indicate a shift of the GRIP band to higher molecular weights with increasing OA concentration. Concomitant with appearance of this shift, the binding of GRIP to NEEP21ct (B) is strongly decreased. C, GRIP-(810 -1112) was covalently attached to the AFM tip and NEEP21ct to the mica (both as GST fusion proteins). Shown are binding events during continuous contact-retraction cycles in the presence of Tris-buffered saline (0 -40 min) after injection (white arrow) of extract from nontreated hippocampal neuron (48 -84 min) and after injection (black arrow) of extract from 100 nM OA-treated hippocampal neurons (96 -132 min). Note that, although control extract had no effect, OA-neuron extract caused a drastic drop in detected binding events. D, quantification of binding events from three independent experiments. **, p Ͻ 0.01. TBS, Tris-buffered saline. ctr, control, E, hippocampal neurons were treated with OA or left untreated. The cells were lysed by passage through a G25 needle and fractionated by differential centrifugation of post-nuclear supernatant. The distribution of GRIP was assessed by Western blotting. In OA-treated neurons GRIP was shifted to the cytoplasmic fraction. F, the bands in the cytosolic and membrane fraction of five independent experiments were quantified, and the ratio was displayed. *, p ϭ 0.029. in this in vitro assay (Fig. 2B). No binding was observed to glutathione beads bound to a negative control of GST alone. These results suggest that the mechanism underlying the effect of OA treatment on GluR2 surface expression includes altered association of GRIP to NEEP21.
The altered binding between NEEP21 and GRIP1 upon OA treatment of neurons was further analyzed by on-line AFM recordings. AFM allows a direct measurement of the number of binding events and the relative interaction forces between a protein on the tip of the AFM cantilever and another protein on the mica surface (42,45). We have previously shown that NEEP21ct binds to a site located in the carboxyl-terminal aa 810 -1112 of GRIP1, and our previous AFM measurements yielded an interaction force of around 111 piconewtons (38). To monitor online the changes in their binding due to a potential phosphorylation of GRIP1-(810 -1112) by extracts from OAtreated neurons, GRIP1-(810 -1112) was covalently attached to the AFM tip, and NEEP21ct was covalently attached to the mica (all DNA constructs were based on rat GRIP1a cDNA and are denominated herein GRIP). Binding events during continuous contact-retraction cycles were recorded in the presence of Tris-buffered saline during 40 min ( Fig.  2C; 0 -40 min). This yielded a total binding probability of 7.9 (defined as the ratio of number of binding events divided by the number of contact-retraction cycles) and an average interaction force of 110 piconewtons, in agreement with our previous recordings. Then, a control extract from nontreated hippocampal neurons was injected into the AFM chamber. No change in binding probability was observed during subsequent recordings (Fig.  2C, 50 -86 min). Finally, extract from OA-treated neurons was injected. Interestingly, this extract caused a drastic decrease in the binding probability (Fig. 2C, 96 -132 min). Quantification of three independent experiments showed that the total binding probability dropped from 7.9% with Tris-buffered saline and 7.61% with control extract to 4.29% (p Ͻ 0.01) with extract from OAtreated neurons (Fig. 2D). These data further show that OA treatment leads to altered binding between NEEP21ct and GRIP-(810 -1112) in a manner consistent with a phosphorylation of GRIP.
GRIP associates with cellular membranes via protein-protein interactions and palmitoylation. To investigate whether this association is altered upon OA treatment, we determined GRIP in cytosol and membrane fractions of untreated or OA-treated neurons. In untreated neurons the majority of GRIP is associated with the membrane fraction. In contrast, in neurons that had been treated with OA, GRIP translocates to the cytosol (Fig. 2, E and  F). This is in agreement with a previous study (28) and suggests that phosphorylation detaches GRIP from membranes.
Serine 917 of GRIP Is Phosphorylated by a NEEP21-associated Kinase-The PDZ-domain scaffolding protein GRIP, a crucial component of AMPAR trafficking, can occur as phosphoprotein (28). Therefore, we analyzed whether OA treatment of neurons induces phosphorylation of GRIP. We generated recombinant fusion proteins between GST and fragments that span different regions of GRIP (Fig. 3A). These fusion proteins, immobilized on beads, were incubated with [␥-32 P]ATP and extracts of either untreated neurons or neurons treated with 100 nM OA. First we analyzed three large fragments: aa 1-342 (containing PDZ domains 1-3), aa 343-809 (containing the GluR2/3-binding PDZ 4 -6 and part of the linker region LII), and aa 810 -1112 (containing LII and PDZ 7). Although aa 1-342 and 343-809 yielded only very weak signals, the carboxyl-terminal fragment aa 810 -1112 was strongly phosphoryla- showed robust phosphorylation by the extract from OA-treated neurons. C, as in B, but applying shorter GST fusion proteins of the indicated fragments. Strong phosphorylation was detected only with aa 881-933. D, as in B, except that immune pellets of rabbit IgG (rIgG) immunoprecipitations (IP) or specific anti-NEEP21 immunoprecipitations from rat brain extracts were applied instead of neuron extracts. As with extracts from OAtreated neurons, aa 881-933 was strongly phosphorylated. The two arrows indicate nonspecific phosphoproteins that were present with all fusion proteins including GST alone. ted in the presence of the extract from OA-treated neurons ( Fig. 3B; arrowheads indicate the sizes of the different fusion proteins). We then generated shorter GST fusion proteins of the region aa 810 -1112. These showed phosphorylation of fragment aa 881-982 (Fig. 3C). Slight phosphorylation of fragment aa 810 -880 was also observed. Finally, two small fragments of aa 881-982 were constructed (aa 881-933 and 933-982) showing that the strong phosphorylation signal could be attributed to aa 881-933 (Fig. 3C).
We have recently shown that GRIP interacts with GluR2 and NEEP21 and that this complex is necessary for AMPAR trafficking and changes in synaptic transmission (38). Therefore, we hypothesized that a kinase that phosphorylates GRIP might be associated with this complex. To test this hypothesis, we incubated the GRIP fragments with glycine-eluted immune pellets from either IgG control immunoprecipitations or from specific immunoprecipitations using an anti-NEEP21 antibody (39). Strikingly, the anti-NEEP21 immune pellet phosphorylated the same GRIP fragments (Fig. 3D) as the extracts of OA-treated neurons (Fig. 3C). Although some phosphorylation was also detected with the negative control IgG immune pellet, suggesting some nonspecific precipitation of the kinase activity, the activity is much stronger in the specific anti-NEEP21 immune pellet. Further nonspecific protein phosphorylations, independent of the GRIP fusion proteins, were observed under all conditions (Fig. 3D, arrows). These results indicate that both the extract from OA-treated neurons and the anti-NEEP21 immune pellet contain protein kinase activities that phosphorylate sites in the same domain of GRIP.
To analyze functionally the observed phosphorylation between aa 881 and 933, we introduced point mutations into the fusion protein GST-GRIP-(881-933) that exchanged serines or threonines to alanines. We then performed in vitro phosphorylations of these different recombinant proteins by anti-NEEP21 immune pellets. The mutation that strongly reduced radioactive phosphate incorporation was the exchange at position 917 (Fig. 4A). This weakly phosphorylated S917A mutant migrates lower on the electrophoresis, presumably due to the reduced phosphate incorporation. Exchange at serine 920 also slightly reduced the intensity of the radioactive band. Equal loading of fusion proteins was verified by the bands on the Coomassie Blue-stained gel before autoradiography (Fig. 4A, CB), which represent mainly nonphosphorylated molecules since the substrate is in excess in these phosphorylation reactions. We then repeated the phosphorylation reaction of the wild type or S917A fusion proteins, but instead of anti-NEEP21 immune pellets, we applied extracts from control neurons or OA-treated neurons. Again, whereas the wild type protein was strongly phosphorylated by the OA-neuron extract, the mutant S917A was not phosphorylated above background level with control extract (Fig. 4B). These results indicate that serine 917 is the primary site of the observed phosphorylation. They further suggest that it is the same kinase activity detected in extracts from OA-treated neurons and in anti-NEEP21 immune pellets that target this common site.
Phosphorylation of Serine 917 Is Involved in AMPAR Trafficking-Next, we asked whether glutamatergic signals that enhance AMPAR trafficking also induce phosphorylation of serine 917. We stimulated hippocampal neurons for 3 min with 100 M AMPA and continued the incubation for 17 min without the agonist. This stimulus induces internalization of surface AMPAR and subsequent recycling (18). We then tested whether extracts from such neurons phosphorylate either wild type GST-GRIP-(881-933)-wt or the mutant GST-GRIP-(881-933)-S917A. As before and as positive control here, extracts from OA-treated Serine and threonine residues were exchanged to alanines by site-directed mutagenesis on the plasmid encoding the fusion protein GST-GRIPaa881-933. A, in vitro phosphorylations of these fusion proteins were performed as described in Fig. 3D using an immune pellet from an anti-NEEP21 immunoprecipitation. Phosphorylation is almost abolished with mutant S917A as shown on the autoradiography (auto). The Coomassie Blue-stained gel (CB) before autoradiography indicates equal loading of the different fusion proteins. B, in vitro phosphorylation of the wt or S917A version of GST-GRIP-(881-933) was performed as described in Fig. 3, B and C, using extracts of untreated or OA-treated hippocampal neurons. The wild type fragment was strongly phosphorylated by the OA-neuron extract, whereas the mutant fragment S917A remains unchanged. neurons phosphorylated the wild type protein (Fig. 5A; compare lane 3 with lane 1), but not the mutant (Fig. 5A, compare lane 4  with lane 2). Interestingly, the wild type protein was clearly phosphorylated by extracts from neurons that were AMPAstimulated (Fig. 5A, compare lane 5 with lane 1). In contrast, the S917A mutant was not phosphorylated above control level by the extract of AMPA-stimulated neurons (Fig. 5A, compare  lane 6 with lane 2).
We have previously reported that complex formation between GluR2, GRIP, and NEEP21 is increased at 2 min after brief NMDA/tetrodoxin treatment, a stimulus known to induce AMPAR internalization (38). To investigate whether the observed GRIP phosphorylation coincides with this early time point, we stimulated hippocampal neurons with 100 M AMPA for 3 min as before and continued the incubation for 7 or 17 min without the agonist (Fig. 5B). We then used extracts from unstimulated or AMPA-stimulated extracts to phosphorylate the wild type fragment GST-GRIPaa881-933-wt (wt). Although this fragment was clearly phosphorylated by the 17-min extracts as before, no increased signal could be detected using the 7-min extract. Interestingly, NMDA stimulation induced the same retarded phosphorylation at 17 min but not at 7 min (Fig. 5B). This result indicates that serine 917 phosphorylation acts at a later step than GluR2-GRIP-NEEP21 complex formation in AMPAR trafficking, possibly complex dissociation.
We then tested whether the observed phosphorylation of GRIP regulates AMPAR surface expression. We transfected cDNAs encoding fusion proteins between green fluorescent protein (GFP) and aa 810 -1112 of GRIP. We transfected either the wt protein or a mutant in which serine 917 was replaced by an alanine (mut S917A) or a mutant in which serine 917 was replaced by an aspartate to mimic phosphorylation (mut S917D). We then analyzed steady-state levels of surface GluR2 in transfected cells. With the wild type and the mutant S917A we observed a significant reduction in endogenous GluR2 surface expression compared with cells transfected for GFP alone (Fig. 6, A and B; wt, 58.4%; S917A, 47.2%; p Ͻ 0.01). This is in agreement with our previous result that expression of this GRIP fragment reduces surface GluR2 (38). Interestingly, the fragment carrying the serine-to-aspartate point mutation had no effect on GluR2 surface expression (Fig.  6, A and B). This suggests that the wild type fragment and the non-phosphorylatable S917A mutant interfered with the trafficking machinery necessary for GluR2 surface expression, whereas the phosphorylation-mimicking S917D mutant did not.
To further strengthen the role of serine 917 in the trafficking of AMPAR, we cotransfected HEK293T fibroblast cells with cDNAs encoding GluR2 and either full-length GRIP-wt or fulllength GRIP-S917D or GFP as control. Then surface expression of GluR2 was assessed by surface biotinylation. Full-length GRIP-wt increased GluR2 surface expression by a factor of 1.46 compared with cells transfected for GFP. In contrast, fulllength mutant GRIP-S917D had no effect compared with control (Fig. 6, C and D; wt, 146.6%; S917D, 95.3%; GFP, 100% p Ͻ 0.038). Taken together these results suggest that a dynamic phosphorylation/dephosphorylation of serine 917 is involved in AMPAR trafficking.
An 85-kDa Protein Kinase Phosphorylates GRIP-Protein kinase C (PKC) is a prominent kinase implicated in synaptic transmission and plasticity, and it is known to regulate the interaction between GluR2 and GRIP by phosphorylation of serine 880 of GluR2 (1)(2)(3)(4)(5). When GRIP-(881-933)-wt was incubated with purified PKC, we did not detect any phosphate incorporation (data not shown). This is not surprising given the non-matching consensus site of conventional PKC isoforms around serine 917. Nevertheless, we analyzed whether an inhibitor (calphostin C) or an activator (PMA) of PKC affects phosphorylation of serine 917. Extracts from AMPA-stimulated neurons strongly phosphorylated GST-GRIP-(881-933)-wt but not the S917A form compared with extracts of non-stimulated neurons (Fig. 7A), as shown above. Interestingly, extracts from neurons that were preincubated with the PKC inhibitor calphostin C before AMPA stimulation yielded clearly less phosphorylation (Fig. 7A). In addition, incubation of neurons with PMA alone was sufficient to increase phosphorylation. The fragment carrying the exchange S917A was barely phosphorylated under all conditions (Fig. 7A), as before in Fig. 5A. We then analyzed whether the kinase activity in immune pellets from anti-NEEP21 precipitations could also be stimulated by PMA. Indeed, the phosphorylation of GST-GRIP-(881-933)-wt was increased after incubation in the presence of PMA (Fig. 7B). These results strongly indicate that PKC is indirectly involved in the phosphorylation of serine 917 of GRIP.
In an attempt to identify the protein kinase in extracts from OA-treated neurons or in the anti-NEEP21 immune pellet that phosphorylated the GRIP fragments, we conducted an in-gel kinase assay. As substrate contained in the 8% acrylamide gel, we used the fusion protein GST-GRIP-(881-982). We separated extracts of neurons either treated for 1 h with 100 nM OA or of control neurons (Fig. 7C, right panel). In addition, we loaded immune pellets either from control IgG precipitations or from specific precipitations using an anti-NEEP21 antibody (Fig. 7C, left panel). After renaturation, the gel was incubated with [␥-32 P]ATP, washed, dried, and exposed to x-ray film. In both the extract of OA-treated neurons and the specific anti-NEEP21 precipitation, we detected a major band at around 85 kDa on the resulting autoradiography. This indicates that the major kinase responsible for the observed GRIP phosphorylation has an apparent molecular mass of around 85 kDa.

DISCUSSION
Endocytosis of AMPAR and subsequent sorting to the degradation pathway or to recycling back to the cell surface are fundamental to regulate the strength of synaptic transmission at excitatory synapses of the nervous system. PDZ domain scaffolding proteins at the postsynaptic density and on intracellular organelles interact with receptor subunits and are essential for their proper trafficking. GRIP is reported to sort, transport (25,37), and stabilize AMPAR at synapses and intracellular subsynaptic compartments (30). Nevertheless, regulatory mechanisms that control these divergent transport pathways have been elusive.
We have previously shown that the neuronal trafficking protein NEEP21 associates with GRIP and GluR2 (38). We report here that GRIP can be phosphorylated at serine 917 by a 85-kDa kinase present in neurons treated with the phosphatase inhibitor OA or in immune pellets from anti-NEEP21 immunoprecipitations. Likewise, neurons that were stimulated with AMPA or NMDA and further incubated for another 17 min could also phosphorylate serine 917. At an earlier time point (7 min) after the stimulations, however, serine 917 was not phosphorylated over background levels. Interestingly, serine 917 phosphorylation involves indirectly activation of PKC. Serine 917 is implicated in GluR2 trafficking. Overexpression of a wild type GRIP  S917A. B, as in A, but using pellets from either IgG or anti-NEEP21 immunoprecipitations from rat brain extracts. Incubation with GST fusion proteins was done in the absence or presence of PMA. C, an in-gel kinase assay was performed using an 8% acrylamide gel that contained as substrate purified recombinant GST-GRIP-(881-982). Extracts of untreated or OA-treated hippocampal neurons (right panel) or glycine-eluted immunoprecipitations (IP) from brain extracts using either nonspecific rabbit IgG (rIgG) immunoprecipitations or specific anti-NEEP21 immunoprecipitations (left panel) were separated on this gel. After renaturation the gel was incubated with [␥-32 P]ATP, dried, and exposed to x-ray film. A major kinase activity was detected at around 85 kDa from the anti-NEEP21 immune pellet and from the extract of OA-treated neurons. fragment reduces GluR2 surface expression, whereas the fragment carrying a S917D mutation had no effect. Likewise, expression of full-length wild type GRIP, but not S917D-mutated GRIP, enhanced surface expression of coexpressed GluR2, suggesting that only the nonphosphorylated form actively promotes GluR2 surface trafficking.
A previous study indicated that GRIP recruits a serine/threonine protein kinase which phosphorylates sites in the linker region between PDZ 6 and 7 of GRIP (28). In line with our results, Bruckner et al. (28) suggested mainly phosphoserines. Serine 917, which we found here as a major phosphorylation site, is highly conserved in GRIP throughout vertebrates. It is located in the linker region LII between PDZ 6 and 7. Although LII shows an overall low identity between ABP/GRIP and GRIP (31.8%) (46), the regions adjacent to serine 917 are well conserved. ABP/GRIP has a threonine residue at the corresponding position of serine 917 in GRIP. A region upstream of this corresponding site in ABP/GRIP binds to PICK1 (46), another PDZ-domain protein implicated in the synaptic targeting of GluR2 subunits of AMPAR (20). GRIP also interacts through the linker region LII with the motor protein KIF5 (also known as kinesin-1). Phosphorylation close to the interaction domains, as we describe here, might regulate this binding.
GluR2 binds to GRIP via the central PDZ domains 5 and 6, which are rather distant from serine 917 (35). Expression of aa (810 -1112)-wt or aa (810 -1112)-S917A versus aa (810 -1112)-S917D in neurons and expression of full-length GRIP-wt versus full-length GRIP-S917D in fibroblasts indicated an implication of this serine residue in GluR2 surface expression that might indirectly affect the trafficking of GluR2. Although the large carboxyl-terminal fragment (aa 810 -1112) was strongly phosphorylated, we also observed a weak phosphate incorporation with the central GRIP fragments aa 342-809. In addition, when aa 810 -1112 was analyzed in more detail, we found weak phosphorylation also with aa 810 -880. This suggests that there are further minor sites on the GRIP polypeptide chain that might be used by several kinases in this in vitro assay. Multiple phosphorylations could be responsible for the shift in gel migration that we observed for endogenous GRIP in neurons upon OA treatment. Interestingly, this treatment only affected surface GluR2, whereas GluR1 was not decreased, which points to a subunit-specific regulation.
Extracts from AMPA-stimulated neurons could phosphorylate serine 917 of GRIP in a calphostin C-dependent manner. Likewise, PMA-treated anti-NEEP21 immune pellets phosphorylated serine 917. This suggests that agonist stimulation activates PKC, which in turn activates a serine 917-phosphorylating 85-kDa kinase. The sequence around serine 917 does not match a consensus sequence of conventional PKC isoforms ((S/ T)X(K/R)), since it is followed by a threonine and a methionine. It is, therefore, not surprising that purified PKC does not phosphorylate the GRIP fragment containing this serine. Interestingly, the serine 917-phosphorylating kinase in OA-treated neurons and the serine 917-phosphorylating kinase in anti-NEEP21 immune pellets phosphorylate the same serine 917. In addition, in-gel kinase assays indicate the same molecular mass of ϳ85 kDa of these activities. This suggests that the two activities originate from the same protein kinase. Treatment of neu-rons with activators or inhibitors of calcium-calmodulin-dependent kinase II, protein kinase A, or phosphatidylinositol 3-kinase did not alter electrophoretic mobility of GRIP (data nor shown), making it unlikely that these kinases phosphorylate directly serine 917. Further proteomics work might clarify this point and allow us to identify the serine 917-phosphorylating kinase.
Phosphorylation of serine 880 of GluR2 by PKC regulates GluR2 surface expression by modulating the binding between GluR2, PICK1, and GRIP, which implicates PKC in long-term depression (1). PKC-dependent activation of the serine 917phosphorylating kinase followed by temporary phosphorylation of GRIP on serine 917 might be another way to affect trafficking of GluR2 during synaptic transmission.
We have previously shown that GRIP interacts with NEEP21 via a binding site located in the carboxyl-terminal 302 amino acids (aa 810 -1112; Ref. 38). Pulldown assays and AFM recordings using extracts from OA-treated neurons suggest a reduced binding between phosphorylated GRIP and NEEP21. In line with a decreased NEEP21-GRIP interaction, we find that in extracts of OA-treated neurons GRIP shows less association to the membrane fraction. A shift from membrane to cytosol of GRIP has previously been observed after 12-O-tetradecanoylphorbol-13-acetate treatment (28). The authors suggested phosphorylation of either GRIP or ephrin to cause this redistribution.
The wild type fragment aa 810 -1112 caused reduced GluR2 surface expression, whereas the fragment with an aspartate at position 917 had no effect. Expression in fibroblasts of fulllength wild type GRIP enhanced GluR2 surface expression, whereas full-length GRIP-S917D had no effect. Together, these two functional results, obtained by quantitative microscopy and biochemical biotinylation assays, indicated that the nonphosphorylated GRIP is the "active" form, whereas the serine 917-phosphorylated GRIP is the "inactive" form. This further indicates that the wild type aa 810 -1112 fragment can act in a dominant-negative manner to interfere with GluR2 surface expression, whereas the mutant fragment carrying the phosphorylation-mimicking S917D exchange has no effect. Therefore, a likely possibility is that the nonphosphorylated fragment sequesters GRIP-interacting proteins necessary for receptor surface targeting or stabilization. Western blot analysis did not detect NEEP21 in HEK293T fibroblast (data not shown). This suggests that regulation of GluR2 trafficking by serine 917 phosphorylation includes additional non-neuronal factors in HEK293T cells.
Importantly, kinetic analysis of serine 917 phosphorylation after brief stimulation suggests that it occurs at late times after stimulus, which does not coincide with previously shown increased GluR2-GRIP-NEEP21 complex formation (38). It rather suggests that it corresponds to a later event in endosomal trafficking. This might be complex dissociation to allow for release of GluR2 and further trafficking steps. NMDA stimulation caused the same delayed phosphorylation at 17 min, but not at 7 min, as AMPA stimulation. Because AMPA and NMDA stimulation have been suggested to induce differential sorting (18) and the observed phosphorylation occurred at a time point presumably beyond this routing decision, it is apparently not directly involved in initiating differential sorting.
We propose that after endocytosis GluR2 arrives at early endosomes positive for NEEP21 and that at that time point GRIP needs to be nonphosphorylated to allow for protein-protein interactions and endosomal trafficking. The OA-induced phosphorylation of GRIP reduces these interactions, blocking GluR2 at early endosomes. This is in agreement with our previous data (38) after overexpression of the dominant-negative GRIP-binding site of NEEP21; this also reduced GluR2 surface levels at steady state and caused accumulation of GluR2 on EEA1-positive endosomes. Nevertheless stimulus-induced endocytosis of GluR2, i.e. its removal from the plasma membrane, was not affected. At later trafficking steps GRIP becomes phosphorylated to dissociate again interactions with endosomal proteins. In this sense, phosphorylation of serine 917 might act as a temporary off-switch for such interactions. The serineto-aspartate exchange would keep the protein "off," and in consequence transfection of either the S917D fragment or the S917D full-length protein does not affect GluR2 trafficking.
Our results indicate that the described phosphorylation is a potential mechanism to regulate proper routing of GluR2-containing AMPAR along intracellular compartments and to the cell surface.