A New Motif Necessary and Sufficient for Stable Localization of the δ2 Glutamate Receptors at Postsynaptic Spines*

The number of each subclass of ionotropic glutamate receptors (iGluRs) at the spines is differentially regulated either constitutively or in a neuronal activity-dependent manner. The δ2 glutamate receptor (GluRδ2) is abundantly expressed at the spines of Purkinje cell dendrites and controls synaptic plasticity in the cerebellum. To obtain clues to the trafficking mechanism of the iGluRs, we expressed wild-type or mutant GluRδ2 in cultured hippocampal and Purkinje neurons and analyzed their intracellular localization using immunocytochemical techniques. Quantitative analysis revealed that deletion of the 20 amino acids at the center of the C terminus (region E) significantly reduced the amount of GluRδ2 protein at the spines in both types of neurons. This effect was partially antagonized by the inhibition of endocytosis by high dose sucrose treatment or coexpression of dominant negative dynamin. In addition, mutant GluRδ2 lacking the E region (GluRδ2ΔE), but not wild-type GluRδ2, was found to colocalize with the endosomal markers Rab4 and Rab7. Moreover, the antibody-feeding assay revealed that GluRδ2ΔE was internalized more rapidly than GluRδ2wt. These results indicate that the E region (more specifically, a 12-amino-acid-long segment of the E2 region) is necessary for rendering GluRδ2 resistant to endocytosis from the cell surface at the spines. Furthermore, insertion of the E2 region alone into the C terminus of the GluR1 subtype of iGluRs was sufficient to increase the amount of GluR1 proteins in the spines. Therefore, we propose that the E2 region of GluRδ2 is necessary, and also sufficient, to inhibit endocytosis of the receptor from postsynaptic membranes.

The ␣-amino-3-hydroxy-5-methyl-4-isoxazolepropionate (AMPA) 2 subclass of ionotropic glutamate receptors (iGluRs) consisting of GluR1 through GluR4, which exist as heteromers (1), plays a major role in fast excitatory synaptic transmission at the dendritic spines in the vertebrate brain. It has become increasingly clear that neuronal, activity-driven changes in the number of AMPA receptors at the postsynaptic spines mediate synaptic plasticity, such as long term potentiation and long term depression (LTD), which is thought to underlie certain forms of memory in the brain. For example, GluR1 is selectively delivered to the spines where neuronal activity is high during synaptic long term potentiation, whereas GluR2 is constitutively delivered to the spines to replace existing synaptic AMPA receptors in the CA1 region of the hippocampus (2,3). In contrast, GluR2-containing AMPA receptors are selectively endocytosed during synaptic LTD in the hippocampus and cerebellum (4 -7). Interestingly, such distinct trafficking patterns of GluR1 or GluR2 are controlled by the respective C termini of the receptors. Furthermore, depending on the phosphorylation status of the C termini, the endocytosed GluR1 could be either reinserted into postsynaptic sites via recycling endosomes, or degraded via lysosomal pathways (8). Therefore, the number of postsynaptic AMPA receptors seems to be tightly regulated by mechanisms that recognize the C termini of the AMPA receptors at multiple checkpoints, including exocytosis, lateral diffusion, endocytosis, and degradation. However, the molecular mechanisms underlying such regulation are not well understood; a potential problem is the heteromerization of endogenous AMPA receptors with the exogenously expressed AMPA receptors, which could affect the trafficking patterns of the receptors in the neurons.
The ␦2 glutamate receptor (GluR␦2), which is predominantly expressed at the postsynaptic spines of parallel fiber-Purkinje cell synapses (9), is a member of the iGluR family. Unlike other iGluRs, GluR␦2 mainly exists as a homomer (10,11). In addition, although 50 -70% of the iGluRs are detected in the intracellular compartments of neurons (12,13), GluR␦2 is predominantly expressed at the cell surface (14). Interestingly, in the ataxic mutant mice hotfoot-4J, -7J, -11J, and -12J, GluR␦2 failed to be transported to the cell surface (14,15), which suggests that efficient trafficking of GluR␦2 to the cell surface is essential for its functioning in the cerebellum. Furthermore, GluR␦2 is not only transported to the Purkinje cell surface but also to spine regions where parallel fibers form synapses (16). Indeed, we found that GluR␦2 actively controls LTD by controlling endocytosis of the AMPA receptors at the postsynaptic spines of Purkinje cells (17). Therefore, characterization of the mechanisms responsible for GluR␦2 trafficking to the postsynaptic spines is not only necessary for understanding GluR␦2 signaling but also to obtain greater insight into the general pattern of iGluR signaling in the brain. In the study described here, we investigated the role of the C-terminal region of GluR␦2 to elucidate the mechanisms responsible for the efficient expression of this receptor at the postsynaptic spines. Although the most C-terminal region is often regarded as crucial for the anchoring of the iGluRs at postsynaptic sites (2), we found that 12 amino acids at the center of the C terminus of GluR␦2 are necessary, and also sufficient, for stable expression of the receptors at the spines in hippocampal and Purkinje cells.

EXPERIMENTAL PROCEDURES
Construction and Transfection of the Expression Plasmids-We used the modified overlap extension method and Pfu DNA polymerase (Stratagene, La Jolla, CA) to delete portions of GluR␦2. Using PCR, we added a cDNA that encoded a hemagglutinin (HA) tag to the 5Ј end * This work was supported by a grant-in-aid for young scientists (to S. M. and K. M.), the Keio Gijuku academic development funds, Keio University grant-in-aid for encouragement of young medical scientists (to S. M.), the grant-in-aid for Scientific Research on Priority Areas, the national grant-in-aid for the establishment of a high tech research center in a private university (to S. M. and M. Y.), the Toray Science and Technology grant, and the Keio University Special grant-in-aid for Innovative Collaborative Research Projects (to M. Y.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. 1  (immediately following the signal sequence) or 3Ј end (immediately upstream of the stop codon) of the GluR␦2 cDNAs. We also added a FLAG tag to the 3Ј end (immediately upstream of the stop codon) of the dynamin1-K44E cDNA (provided by Dr. R. B. Vallee, Columbia University/) (18). Bidirectional sequencing confirmed the nucleotide sequence of the amplified open reading frame. cDNAs encoding Rab4a and Rab7a, tagged with green fluorescent protein (GFP), were provided by Dr. M. Fukuda (Tohoku University). After the cDNAs were cloned into the expression vector pCAGGS (provided by Dr. J. Miyazaki, Tohoku University), the constructs were transfected into hippocampal neurons using Effectene (Qiagen, Valencia, CA), as described previously (19). Immunohistochemical Analysis-To examine the subcellular localization of GluR␦2 in the hippocampal neurons, we used a pCAGGS vector expressing wild-type or mutant GluR␦2 tagged with HA. Cultured hippocampal neurons were transfected with these clones using effectene (Qiagen) and stained with an anti-HA antibody (Roche Applied Science), as described previously (19). To investigate the subcellular localization of GluR␦2 in the Purkinje cells, we used cells infected with a modified Sindbis virus. The GluR␦2 cDNA whose 3Ј end was linked to the sequence encoding the HA tag was cloned into a pSinRep vector (Invitrogen), and virus particles were generated according to the manufacturer's instructions. Cerebellar, dissociated cultures were prepared from embryonic day 18 mice, as described previously (6) and used at 14 -21 days in vitro. Sindbis virus encoding wild-type or mutant GluR␦2 was applied to cerebellar cultures. Eighteen hours later, the cells were fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS) for 2 h at 4°C and then washed three times with PBS. The cultures were first incubated with a blocking solution (1% bovine serum albumin, 0.4% Triton X-100, and 10% normal goat serum in PBS) and then with the anti-HA (Roche Applied Science) and anti-calbindin (Chemicon, Temecula, CA) antibodies. The bound primary antibodies were detected by secondary antibodies that were conjugated to either Alexa 488 or Alexa 546 (Molecular Probes, Eugene, OR).
Surface-labeling Assay-N-terminal HA-tagged wild-type or mutant GluR␦2 were expressed in cultured hippocampal neurons. Cells were fixed as described above, incubated with a blocking solution without Triton X-100, and then incubated with the anti-HA antibody (Covance, Richmond, CA). The primary antibodies bound on the cell surface were detected by secondary antibodies that were conjugated to Alexa 546 (Molecular Probes).
"Antibody-feeding" Assay-N-terminal HA-tagged wild-type or mutant GluR␦2 were expressed in cultured hippocampal neurons. Mouse anti-HA monoclonal antibody (10 g/ml; Covance) was added to the culture medium for 30 min at 37°C to label GluR␦2 on the surface of live Amino acid residues are numbered beginning with the N-terminal residue of the mature subunit. N-or C-terminal HA-tagged wild-type (GluR␦2 wt ) or mutant GluR␦2 were expressed in cultured hippocampal neurons together with GFP, and their distribution was examined by immunocytochemical analysis using anti-HA antibody. B, endogenous GluR␦2 expression in cultured neurons. Representative images of the distribution patterns of endogenous GluR␦2 (red) and MAP2 (green) in hippocampal (upper and middle panels) and Purkinje (lower panel) neurons. C, representative images of the distribution patterns of N-terminal HA-tagged GluR␦2 wt (upper panels) and GluR␦2 ⌬E (lower panels) in cultured hippocampal neurons. Arrows and arrowheads indicate HA-positive and -negative spines, respectively. D, representative images of the distribution patterns of C-terminal HA-tagged GluR␦2 wt (upper panels) and GluR␦2 ⌬E (lower panels) in cultured hippocampal neurons. E, quantitative analysis of spine localization of C-terminal HA-tagged GluR␦2 in cultured hippocampal neurons. The HA-staining intensity in the spines was normalized to the GFP-staining intensity in the spines. The HA/GFP-staining intensity ratio of GluR␦2 wt was arbitrarily set at 100%. Each bar represents the mean Ϯ S.E., and significance was established in comparison with that measured in cells that expressed GluR␦2 wt (**, p Ͻ 0.01; n ϭ 7 cells).
hippocampal neurons. After washing with PBS, the neurons were treated with 0.5 M NaCl/0.2 M acetic acid (pH 3.5) for 4 min on ice to remove the remaining antibodies on the cell surface. The neurons were then rinsed and fixed with 4% paraformaldehyde with 4% sucrose. The cultures were permeabilized and incubated with a blocking solution, and total GluR␦2 proteins were stained by rabbit anti-GluR␦2 antibody (Chemicon), as described above. Anti-HA and anti-GluR␦2 antibodies were detected by secondary antibodies that were conjugated to Alexa 546 and Alexa 488, respectively (Molecular Probes).
Image Analysis-Image analysis was performed in a blind manner without knowing the identity of the samples during the analysis. Spines were defined as dendritic protrusions that have an enlargement at the tip. Spines within proximal 2-3 dendritic segments (ϳ50 m for each segment; total length of Ͼ100 m) were analyzed for individual neurons. The intensity of the HA immunoreactivity was normalized to the GFP fluorescence intensity in each spine by the IP-Lab imaging software (Scanalytics, Fairfax, VA). The mean normalized HA immunoreactivity, which represents the spine localization of the HA-tagged protein in each neuron, was calculated from at least 20 spine heads/neuron. These values were analyzed from a total of 5-9 neurons from at least two separate cultures, and the averages were compared. For the antibody-feeding assay, the fluorescence intensities of Alexa 546 (internalized GluR␦2) and Alexa 488 (total GluR␦2) were quantified in proximal 2-3 dendritic segments by the IP-Lab imaging software. The averages of seven neurons (from two separate cultures) were compared. For statistical analysis, we used the Student's t test.
Mice Treatment-All of the procedures related to the care and treatment of the experimental animals were conducted in accordance with the Guidelines for Animal Experiments at the Keio University. The animals were anesthetized with tribromoethanol before decapitation.

Identification of the Synaptic Localization Signal of GluR␦2-To
exclude the possibility of endogenous GluR␦2 modifying the trafficking of exogenously expressed GluR␦2, we expressed wild-type GluR␦2 with its N terminus tagged with HA (NT-HA-GluR␦2 wt (Fig. 1A) in cultured hippocampal neurons; no endogenous GluR␦2 is expressed in the hippocampus of adult rats (11). Indeed, we could not detect GluR␦2 expression in cultured hippocampal neurons by immunocytochemical analysis (Fig. 1B). Therefore, in these neurons, the majority of the exogenously expressed GluR␦2 should form homomer, although weak interaction between GluR␦2 and other iGluRs may occur (10, 11). Neurons were The HA-staining intensity in the spines was normalized to the calbindin-staining intensity in the spines. The HA/calbindin-staining intensity ratio of GluR␦2 wt was arbitrarily set at 100%. Each bar represents the mean Ϯ S.E., and significance was established in comparison with that measured in cells that expressed GluR␦2 wt (**, p Ͻ 0.01; n ϭ 7 cells).
cotransfected with cDNA encoding GFP to identify the morphology of the dendritic spines, and the distribution of GluR␦2 wt was examined by immunocytochemical analysis using anti-HA antibody. As reported for endogenous GluR␦2 in the cerebellar Purkinje cells (16), exogenous NT-HA-GluR␦2 wt was also efficiently transported to the spines in the cultured hippocampal neurons (Fig. 1C). Because GluR␦2 mainly exists as a homomer (10), these findings indicate that exogenous GluR␦2 wt contains sufficient information for efficient trafficking to spines by a mechanism common to both hippocampal and Purkinje neurons.
To identify the region important for the synaptic localization of GluR␦2, a series of deletions were introduced in the C-terminal intracellular domain of GluR␦2 and an HA tag was added to the extreme C terminus of each clone (Fig. 1A). These mutant or wild-type receptors were expressed in cultured hippocampal neurons together with GFP. Similar to NT-HA-GluR␦2 wt , GluR␦2 wt with its C terminus tagged with HA was effectively transported to the spines of the cultured hippocampal neurons (Fig. 1D). Several postsynaptic density-95/Discs large/zona occludens-1 (PDZ) domain-containing proteins, such as PSD-93, PTPMEG, delphilin, and nPIST, have been reported to bind to the C terminus of GluR␦2 receptors, providing them anchorage at the postsynaptic spines (21). However, because an HA tag attached to the C terminus would completely interrupt such interactions, it is unlikely that these anchoring proteins are involved in the postsynaptic localization of GluR␦2.
Interestingly, although all mutant GluR␦2 proteins were normally trafficked to the cell surface in human embryonic kidney cells (22), the mutant lacking amino acids 895-915 (GluR␦2 ⌬E ) was mostly excluded from the spines of cultured hippocampal neurons, regardless of the position of the HA tag (Fig. 1, C and D). Quantitative analysis of the HA-staining intensity in the spines revealed that the deletion of the "E region" significantly reduced the amount of GluR␦2 protein at the spines (p Ͻ 0.01; n ϭ 7) (Fig. 1E), indicating that the E region is necessary for the synaptic localization of GluR␦2.
We examined whether the E region is also necessary for the synaptic localization of GluR␦2 in Purkinje cells, the type of cell in which GluR␦2 is predominantly expressed. GluR␦2 wt or GluR␦2 ⌬E , with the C terminus tagged with HA, was expressed in cultured Purkinje cells using the Sindbis virus. The localization of the receptor proteins and the morphology of the infected cells were examined by immunocytochemical analysis using anti-HA and anti-calbindin antibody, respectively (Fig.   FIGURE 3. Effects of sucrose treatment on GluR␦2 ⌬E distribution in cultured hippocampal neurons. GluR␦2 wt and GluR␦2 ⌬E , with their C termini tagged with HA, were expressed with GFP in cultured hippocampal neurons. Neurons were treated with sucrose at 350 mM for 15 or 30 min, and the distribution of GluR␦2 was examined by immunocytochemical analysis. A, representative images of the distribution patterns of C-terminal HA-tagged GluR␦2 ⌬E in the absence of sucrose treatment (top panels), of GluR␦2 ⌬E after 30 min of sucrose treatment (middle panels), and GluR␦2 wt in the absence of sucrose treatment (bottom panels) in cultured hippocampal neurons. Arrows and arrowheads indicate HA-positive and -negative spines, respectively. B, quantitative analysis of the spinal localization of GluR␦2 in cultured hippocampal neurons. The HA-staining intensity in the spines was normalized to the GFP-staining intensity in the spines. The HA/GFP-staining intensity ratio of GluR␦2 wt was arbitrarily set at 100%. Each bar represents the mean Ϯ S.E., and significance was established in comparison with that measured in cells that expressed GluR␦2 ⌬E at time 0 (**, p Ͻ 0.01; n ϭ 5 cells). C, detection of GluR␦2 proteins on the cell surface. Surface GluR␦2 proteins were labeled by applying anti-HA antibody to hippocampal neurons expressing N-terminal HA-tagged GluR␦2 wt (left panels) or GluR␦2 ⌬E (right panels) under non-permeabilizing conditions and visualized by Alexa 546-conjugated secondary antibodies.
2A-C). Although GluR␦2 wt was abundantly localized in a punctate pattern in the spines of the cultured Purkinje cells, GluR␦2 ⌬E was more prominently localized on the dendritic shafts (Fig. 2C). Quantitative analysis of the HA-staining intensity in the spines revealed that the amount of GluR␦2 ⌬E protein in the Purkinje cell spines was significantly lower than that of GluR␦2 wt (p Ͻ 0.01; n ϭ 7) (Fig. 2D). These results indicate that the E region is essential for the synaptic localization of GluR␦2 in both the hippocampal and Purkinje neurons.
Effect of Blockade of Endocytosis on GluR␦2 ⌬E Localization in the Spines-There are two plausible mechanisms by which the E region might control the synaptic localization of GluR␦2, 1) it may facilitate the delivery of GluR␦2 to the dendritic spines by lateral diffusion from extrasynaptic sites, or 2) it may stabilize GluR␦2 localization by inhibiting its removal from the spines. To examine the latter possibility, we treated hippocampal neurons expressing GluR␦2 wt or GluR␦2 ⌬E with sucrose at a concentration of 350 mM, which is known to inhibit endocytosis (23). Treatment with this concentration of sucrose for 30 min, but not for 15 min, significantly increased the amount of the GluR␦2 ⌬E protein in the spines (p Ͻ 0.01; n ϭ 5), whereas it had no effect on the spinal localization of GluR␦2 wt (Fig. 3, A and B). Although we could not challenge the cells for longer periods of time with sucrose because of its detrimental effects on the morphology of the neurons, our results suggested that GluR␦2 ⌬E was removed more rapidly from the spines than GluR␦2 wt by endocytosis.
To confirm that GluR␦2 in spines was located on the cell surface, we performed the surface labeling assay. NT-HA-GluR␦2 wt or NT-HA-GluR␦2 ⌬E was expressed in cultured hippocampal neurons, and surface receptors were labeled with anti-HA antibody under non-permeabilizing conditions. Although NT-HA-GluR␦2 wt was highly expressed in spines, weak NT-HA-GluR␦2 ⌬E immunoreactivities were observed diffusely throughout the dendrites (Fig. 3C). These results indicated that spine localization of GluR␦2 indeed represented receptors on the cell surface and that the E region stabilized GluR␦2 on the postsynaptic membrane. A, representative images of GluR␦2 ⌬E localization without (upper panels) or with (lower panels) dominant negative dynamin in cultured hippocampal neurons. Arrows and arrowheads indicate GluR␦2-containing and non-GluR␦2-containing spines, respectively. B, quantitative analysis of the spinal localization of GluR␦2 ⌬E in cultured hippocampal neurons. The HA-staining intensity in the spines was normalized to the GFP-staining intensity in the spines. The HA/GFP-staining intensity of non-dynamin-expressing neurons was arbitrarily set at 100%. Each bar represents the mean Ϯ S.E., and significance was established in comparison with that measured in cells that expressed GluR␦2 ⌬E alone (*, p Ͻ 0.05; n ϭ 5 cells). C, representative images of N-terminal HA-tagged GluR␦2 ⌬E localization without (upper panels) or with (lower panels) dominant negative dynamin in cultured hippocampal neurons. Arrows and arrowheads indicate GluR␦2-containing and non-GluR␦2-containing spines, respectively.
Dynamin has been shown to be essential for clathrin-mediated endocytosis, whereas its mutant dynamin-K44E, in which the lysine at position 44 is replaced with glutamine, blocks endocytosis in a dominant negative manner (18). When dynamin-K44E was coexpressed with GluR␦2 ⌬E , it significantly increased the amount of GluR␦2 ⌬E protein in the spines (p Ͻ 0.05; n ϭ 5) (Fig. 4, A and B). Identical results were obtained from the experiment using NT-HA-GluR␦2 ⌬E (Fig. 4C). Taken together, these results indicate that the E region might be necessary for rendering GluR␦2 resistant to endocytosis at the postsynaptic membrane.
GluR␦2 ⌬E Is Internalized by Endocytosis-Endocytosed membrane proteins are either recycled via recycling endosomes or degraded via late endosomes and lysosomes (8). To examine the intracellular localization of GluR␦2 ⌬E , HA-tagged GluR␦2 ⌬E was coexpressed with GFP-tagged Rab4 (an early endosome marker) or Rab7 (a late endosome marker) (24) in cultured hippocampal neurons. In neurons expressing HA-GluR␦2 wt , the HA immunoreactivities were observed in the spines, and there was no evidence of colocalization with Rab4 or Rab7, confirming that GluR␦2 wt was predominantly localized at the cell surface (Fig. 5A). In contrast, in neurons expressing HA-GluR␦2 ⌬E , the HA immunoreactivities were predominantly colocalized with Rab4 or Rab7 (Fig. 5B). The colocalization of GluR␦2 ⌬E with early or late endosomal markers could be attributable to the role of the E region in promoting recycling of endocytosed GluR␦2 to the spine. However, because inhibition of endocytosis by high dose sucrose treatment or dynamin-K44E resulted in an increased presence of GluR␦2 ⌬E at the spines (Figs. 3 and 4), it is more plausible that the E region inhibited endocytosis and stabilized the localization of GluR␦2 at the spines.
To further confirm that the E region inhibited the endocytosis of GluR␦2, we carried out an antibody-feeding immunofluorescence internalization assay in hippocampal neurons expressing NT-HA-GluR␦2 wt or NT-HA-GluR␦2 ⌬E . A significantly higher degree of internalization was observed with GluR␦2 ⌬E than with GluR␦2 wt (p Ͻ 0.05; n ϭ 7) (Fig.  5, C and D). Therefore, the E region played an essential role in preventing endocytosis of GluR␦2 from the postsynaptic membrane.
Further Characterization of the E Region-The amino acid sequence around the E region is highly conserved among several species (Fig. 6A), suggesting that this region probably plays an essential role in GluR␦2 function. To further narrow down the important region for stable localization at the spines, smaller deletions were introduced in the E region, GluR␦2 ⌬E1 lacking the former half and GluR␦2 ⌬E2 lacking the latter half of the E region (Fig. 6B). When expressed in cultured hippocampal neurons, GluR␦2 ⌬E1 proteins were found to be localized in the spines, FIGURE 5. Endocytosis of GluR␦2 ⌬E in cultured hippocampal neurons. A and B, colocalization of GluR␦2 wt and GluR␦2 ⌬E with endosome markers. C-terminal HA-tagged GluR␦2 wt (A) or GluR␦2 ⌬E (B) was expressed with GFP-tagged Rab4 (early endosome marker) or GFP-tagged Rab7 (late endosome marker), and their distribution pattern was examined by immunocytochemical analysis using anti-HA antibody. C and D, the antibody-feeding assay to monitor endocytosis of surface GluR␦2 proteins. Anti-HA antibody was added to live hippocampal neurons expressing N-terminal HA-tagged GluR␦2 wt or GluR␦2 ⌬E to label GluR␦2 on the surface. Thirty minutes later, internalized and total GluR␦2 were detected by anti-HA (red) and anti-GluR␦2 (green) antibodies, respectively. Representative images of overall (C ) and dendritic (D) regions are shown. E, quantitation of GluR␦2 internalization, measured as the ratio of internalized (red)/total (green) fluorescence. Each bar represents the mean Ϯ S.E., and significance was established in comparison with that measured in cells that expressed GluR␦2 wt (*, p Ͻ 0.05; n ϭ 7 cells).
whereas GluR␦2 ⌬E2 proteins were mostly excluded from the spines (Fig.  6C), suggesting that the E2 region probably contains a sequence necessary for stable expression of the receptor at the spines. Interestingly, GluR␦2 was recently shown to interact with Shank (a multifunctional anchoring protein for metabotropic glutamate receptors at the spines) via a region containing the E2 region (25). Therefore, we examined whether the spine localization of GluR␦2 was mediated by Shank by replacing the serine at position 905 with alanine (GluR␦2 S905A ), a mutation previously shown to block the binding ability of the receptor to Shank in vitro (25). However, GluR␦2 S905A was found to be abundantly localized at the spines in the same manner as GluR␦2 wt (Fig. 6C). Similarly, GluR␦2 S905A,T915A,F917A , which included additional mutations at positions 915 and 917 outside the E2 region and which completely blocked the GluR␦2 binding to Shank (25), were also found to be localized abundantly at the spines (data not shown). From these results, it is considered unlikely that Shank is involved in the spine localization of GluR␦2.
To examine whether the E region was sufficient for spine localization of GluR␦2, we expressed a truncated version of GluR␦2 (GluR␦2 E-) in cultured hippocampal neurons in which the C terminus immediately after the E region was removed (Fig. 7A). Similar to GluR␦2 wt , but unlike GluR␦2 ⌬E , GluR␦2 Ewas highly localized to spines (Fig. 7B). To further examine whether the E region was sufficient to localize other membrane proteins, we inserted the E2 region in the corresponding C-terminal region of the AMPA receptor GluR1 (Fig. 7C). Consistent with earlier reports (26), wild-type GluR1 was not effectively transported to the spines under basal conditions; however, the insertion of the E2 region significantly increased the amount of GluR1 protein detected in the spines (p Ͻ 0.05; n ϭ 9) (Fig. 7, D and E). These results indicated that the E2 region of GluR␦2 was sufficient for stable localization of GluR1 at the spines in a context-independent manner.

DISCUSSION
GluR␦2 is abundantly expressed at the postsynaptic membranes of Purkinje cells and plays crucial roles in cerebellar motor learning and synaptic plasticity, including LTD. In our previous study, we have shown that the juxtamembrane "A region" of GluR␦2 played an essential role in the efficient surface transport of GluR␦2 (22). However, the molecular mechanism that enables enrichment of GluR␦2 at dendritic spines remains unclear. In the present study, we demonstrated that a region at the center of the C terminus consisting of 12 amino acids (E2 region) is necessary for the efficient localization of GluR␦2 at the spines in hippocampal and Purkinje neurons. Inhibition of endocytosis by treatment with sucrose at 350 mM (Fig. 3) or expression of dominant negative dynamin (Fig. 4) increased the amount of GluR␦2 ⌬E protein in the spines. In addition, GluR␦2 ⌬E , but not GluR␦2 wt , colocalized with the endosomal proteins Rab4 and Rab7 (Fig. 5, A and B). The antibodyfeeding assay also revealed that GluR␦2 ⌬E was internalized more rapidly than GluR␦2 wt (Fig. 5, C and D). These results strongly indicate that the E2 region is necessary for rendering GluR␦2 resistant to endocytosis from the cell surface at the spines. Truncation of the C terminus immediately after the E region did not affect spine localization of GluR␦2 (Fig.  7B). Furthermore, insertion of the E2 region alone at the C terminus of GluR1 was sufficient to increase the amount of GluR1 proteins in the spines (Fig. 7, D and E). Therefore, we propose that the E2 region of GluR␦2 is necessary and also sufficient to inhibit endocytosis from the postsynaptic membranes by a mechanism common to hippocampal and Purkinje neurons.
Wild-type GluR1 was not effectively transported to the spines under basal conditions, as reported previously (Fig. 7D) (2). Thus, in addition to inhibiting endocytosis from the postsynaptic membranes, the E2 region may also enhance the delivery of GluR1 to the spines. Alternatively, because neurons show low levels of spontaneous activities in culture preparations (27), only small amounts of GluR1 may be delivered to the spines, and by inhibiting endocytosis, the E2 region may enhance the retention of GluR1 delivered to the spines.
Many proteins containing the PDZ domain are thought to be involved in the stabilization of postsynaptic iGluRs at the dendritic spines. For example, localization of the AMPA receptor subunit GluR2 at the spines requires the interaction of its C terminus with GRIP, a PDZ domain-containing protein (28). Neuronal activity is thought to induce phosphorylation of the serine at position 880 in the C terminus of GluR2 (29,30); GluR2 is then released from GRIP and endocytosed from the cell surface, resulting in LTD (6,20). Similarly, several other PDZ domain-containing proteins, such as PSD-93, PTPMEG, delphilin, and nPIST, have been shown to bind with the C terminus of GluR␦2 (21), although such binding depends on the C-terminal end itself and not specifically on the E2 region of GluR␦2. In contrast, structural analysis of the PDZ domain revealed that it is architecturally designed to allow binding to consensus sequences located at the C-terminal end of the peptide. Indeed, tagging of the C-terminal end of GluR␦2 with HA, which would completely block its interaction with these PDZ domaincontaining proteins, did not affect the localization of GluR␦2 (Fig. 1). Therefore, it is unlikely that these PDZ domain-containing proteins mediate the preferential localization of GluR␦2 at the spines.
Certain PDZ domain-containing proteins could interact with non-Cterminal, internal regions of proteins. Indeed, Shank was shown to interact with a region within the C terminus of GluR␦2, which con- The gray box and the bold letters indicate the inserted GluR␦2 E2 region and its amino acid sequence, respectively. D, representative images of the distribution patterns of C-terminal HA-tagged GluR1 wt (upper panels) and GluR1 E2 (lower panels) in cultured Purkinje neurons. GluR1 wt or GluR1 E2 was expressed in cultured hippocampal neurons and their spinal localization was examined by immunocytochemical analysis using anti-HA antibody. Arrows and arrowheads indicate HA-positive and -negative spines, respectively. E, quantitative analysis of the spinal localization of GluR1 in cultured hippocampal neurons. The HA-staining intensity in the spines was normalized to the GFP-staining intensity in the spines. The HA/GFP-staining intensity ratio of GluR1 wt was arbitrarily set at 100%. Each bar represents the mean Ϯ S.E., and significance was established in comparison with that measured in cells that expressed GluR1 (*, p Ͻ 0.05; n ϭ 9 cells). tained the E2 region (25). However, we found that a mutation that was known to dissociate GluR␦2 from Shank did not affect the spinal localization of GluR␦2 (Fig. 6). Similarly, PSD-95, a PDZ protein that binds to N-methyl-D-aspartate receptors and mediates important postsynaptic signaling, is not essential for postsynaptic targeting of N-methyl-D-aspartate receptors (31). Furthermore, mutant mice lacking delphilin, a PDZ protein that binds to GluR␦2, showed normal postsynaptic localization of GluR␦2 (32). Therefore, although PDZ domain-containing proteins are important for postsynaptic signaling, they are unlikely to mediate the postsynaptic localization of iGluRs.
The actin cytoskeleton has also been found to be critical for the stabilization of iGluR localization at the spines (33). For example, the C terminus of N-methyl-D-aspartate receptors binds to spectrin (34) and actinin (35), both of which bind to the F-actin present in abundance in the spines. Similarly, GluR␦2 also binds to the actin cytoskeleton via spectrin (36). It has been suggested that the neuronal activity-induced increase in the Ca 2ϩ in Purkinje cells may promote dissociation of GluR␦2 from spectrin, leading to endocytosis of GluR␦2 from the postsynaptic membrane (23). Thus, the stabilization of GluR␦2 localization at the spines may be mediated by the binding of the E2 region with spectrin and actin. However, the AMPA receptor GluR1 has also been shown to interact with spectrin via adapter proteins 4.1N (37) and RIL (38) in hippocampal neurons. Because insertion of the E2 region of GluR␦2 into the C terminus of GluR1 promoted the synaptic accumulation of the receptors, it seems unlikely that the role of the E2 region is simply confined to its association with the spectrin-actin cytoskeleton. In addition, our preliminary analysis also indicated that spectrin does not bind, at least directly, to the E2 region of GluR␦2 in vitro. 3 However, because many actin-binding proteins, such as calponin (39), spinophilin (40), actinin (41), myosin Va (42), myosin VI (43), and drebrin (44), are known to be localized at the spines, we postulate that, similar to the N-methyl-D-aspartate receptors that are closely associated with F-actin via multiple proteins (including actinin and spectrin), GluR␦2 localization at the spines may also be stabilized by several actin-binding proteins, one of which may bind to the E2 region.
Hotfoot mice are spontaneous ataxic mouse mutants resulting from various mutations in the gene encoding GluR␦2. Interestingly, of the 20 alleles known so far, most mutants retain mutant GluR␦2 in the endoplasmic reticulum, indicating that GluR␦2 must be transported to the Purkinje cell surface for it to function properly (14,15,21). In addition, we recently demonstrated that the application of an antibody against the extracellular domain of GluR␦2 induced endocytosis of the AMPA receptor GluR2 and inhibited further induction of LTD (17). These findings indicate the essential roles of GluR␦2 at the postsynaptic spine surface. Therefore, further studies are warranted to identify specific proteins that bind to the E2 region and regulate stabilization of GluR␦2 in the dendritic spines.