SAP97 Is Associated with the α-Amino-3-hydroxy-5-methylisoxazole-4-propionic Acid Receptor GluR1 Subunit*

Rapid glutamatergic synaptic transmission is mediated by ionotropic glutamate receptors and depends on their precise localization at postsynaptic membranes opposing the presynaptic neurotransmitter release sites. Postsynaptic localization ofN-methyl-d-aspartate-type glutamate receptors may be mediated by the synapse-associated proteins (SAPs) SAP90, SAP102, and chapsyn-110. SAPs contain three PDZ domains that can interact with the C termini of proteins such asN-methyl-d-aspartate receptor subunits that carry a serine or threonine at the -2 position and a valine, isoleucine, or leucine at the very C terminus (position 0). We now show that SAP97, a SAP whose function at the synapse has been unclear, is associated with α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid (AMPA)-type glutamate receptors. AMPA receptors are probably tetramers and are formed by two or more of the four AMPA receptor subunits GluR1–4. GluR1 possesses a C-terminal consensus sequence for interactions with PDZ domains of SAPs. SAP97 was present in AMPA receptor complexes immunoprecipitated from detergent extracts of rat brain. After treatment of rat brain membrane fractions with the cross-linker dithiobis(succinimidylpropionate) and solubilization with sodium dodecylsulfate, SAP97 was associated with GluR1 but not GluR2 or GluR3. In vitro experiments with recombinant proteins indicate that SAP97 specifically associates with the C terminus of GluR1 but not other AMPA receptor subunits. Our findings suggest that SAP97 may be involved in localizing AMPA receptors at postsynaptic sites through its interaction with the GluR1 subunit.

The prevailing excitatory neurotransmitter in the mammalian brain is glutamate (1,2). Upon its release from presynaptic sites, this neurotransmitter binds to ionotropic glutamate receptors that mediate rapid excitatory synaptic transmission in the mammalian brain (1,2). Several immuno-electron microscopic studies have demonstrated that ionotropic glutamate receptors are clustered at postsynaptic sites of excitatory synapses (3)(4)(5). Two major glutamate receptor families exist, namely N-methyl-D-aspartate (NMDA) 1 receptors, which mediate Ca 2ϩ influx, and non-NMDA receptors, which are usually not Ca 2ϩ -permeable (1,2,6). Non-NMDA receptors are further divided into ␣-amino-3-hydroxy-5-methylisoxazole-4-propionic acid (AMPA) receptors and kainate receptors. At low frequency, synaptic transmission normally depends nearly exclusively upon AMPA receptors. On the other hand, kainate and NMDA receptors require higher frequencies for activation. NMDA receptor-mediated Ca 2ϩ influx is necessary for different forms of synaptic plasticity, such as long term potentiation (1,7,8). At different synapses in the hippocampus and in other brain areas, a few bursts of high frequency electric stimulation that activate NMDA receptors induce a long lasting increase in synaptic transmission, the hallmark of long term potentiation.
Glutamate receptors are thought to be heterotetramers consisting of homologous subunits (39). AMPA receptors are formed by two or more of the four AMPA receptor subunits designated GluR1-4 (9). NMDA receptors are heterooligomers composed of one or two NR1 and two or three NR2 subunits. One NR1 subunit and four different NR2 subunits (NR2A-D) have been identified (6,10,11). Hydrophobicity plots indicated the presence of four hydrophobic regions termed M1-M4 in all glutamate receptor subunits. The N terminus is extracellular and is followed by the transmembrane region M1. There is now strong evidence that the hydrophobic region M2 of glutamate receptors loops only partially into the plasma membrane and then back into the cytosol. M3 and M4 are transmembrane regions. The C terminus is localized on the intracellular side of the plasma membrane (12,13).
Synapse-associated proteins (SAPs) constitute a family of closely related proteins that has been implicated in the process of clustering NMDA receptors at postsynaptic sites (14 -17). A prototypic SAP consists of three PDZ domains in the N-terminal part followed by an SH3 domain and a guanylate kinaselike domain in the C-terminal region. The guanylate kinaselike domain does not appear to be catalytically active (18) but interacts with another family of structural proteins known as guanylate kinase-associated proteins or SAP90/PSD-95-associated proteins (19,20). SAP90/PSD-95 (21,22), chapsyn-110/ PSD-93 (15,23), and SAP102 (24,25) directly bind to the very C termini of NMDA receptor subunits (14,15,24,25). These interactions are mediated by the first and second PDZ domain of these SAPs and require the presence of a valine, isoleucine, or leucine at the very C-terminal position (designated as the 0 position) of the NMDA receptor subunits (17,26). In addition, serine or threonine has to be present at the Ϫ2 position, which is two amino acids upstream of the C-terminal 0 position (17,26). GluR1 possesses a threonine at the C-terminal Ϫ2 position and leucine at the 0 position. The C terminus of GluR1, thereby, constitutes a consensus site for SAP binding via interaction with PDZ domains (17,26). SAP97, another member of the SAP family (27), does not coimmunoprecipitate with solubilized NMDA receptors, in contrast to SAP90/PSD-95, chapsyn-110/PSD-93, and SAP102 (28). In addition, the synaptic function of SAP97 has been unclear. Here we show that SAP97 is associated with AMPA receptors in vivo. It binds to the GluR1 subunit and does not appear to directly interact with the GluR2 or GluR3 subunits. Our findings raise the possibility that SAP97 may be important for AMPA receptor localization at synaptic junctions.

EXPERIMENTAL PROCEDURES
Immunoprecipitation and Immunoblotting-Crude membrane fractions were prepared from rat brains in the presence of protease inhibitors, washed by resuspension and re-centrifugation in buffer A (150 mM NaCl, 10 mM Tris-Cl, pH 7.4, 10 mM EDTA, 10 mM EGTA), solubilized in buffer A containing protease inhibitors and 1% of either deoxycholate or Triton X-100 or SDS, cleared by ultracentrifugation, and used for immunoprecipitation and subsequent immunoblotting as described (29). For immunoprecipitation after SDS solubilization, a 7-fold volume of a 1% Triton X-100 solution in buffer A was added. This procedure converts the pure SDS micelles into mixed micelles containing mainly Triton X-100 and a much smaller fraction of SDS. The result is a significant reduction in the potential of SDS to dissolve and denature proteins, allowing the addition of antibodies for immunoprecipitation (30,31). For chemical cross-linking, membrane fractions were washed by resuspension in 10 mM HEPES-NaOH, pH 7.0, followed by centrifugation and then incubated for 10 min on ice with 200 M of the cross-linking agent dithiobis(succinimidylpropionate) (DSP). The reaction was stopped by adding Tris-Cl, pH 7.4, to a final concentration of 100 mM before solubilization with SDS. Subsequent immunoprecipitation and immunoblotting were performed as described earlier (29).
NMDA receptors were precipitated using antibody 54.2 (␣NR1), which specifically binds to NR1 (29). NR2A and NR2B were detected by immunoblotting with an anti-peptide antibody (␣NR2A/B) that specifically associates with both subunits (32). Antibody ␣GluR1 specifically recognizes GluR1 (Ab7 in Wenthold et al. (9)), and ␣GluR2/3 specifically binds to GluR2 and GluR3 (Ab25 in Wenthold et al. (9)). SAP90/PSD-95 was specifically detected with antibody RA2d (␣SAP90 (22)) and SAP97 and SAP102 with the corresponding affinity-purified rabbit antibodies ␣SAP97 and ␣SAP102 characterized in Muller et al. (Refs. 27 and 24,respectively). The anti-glutathione S-transferase (GST) antibody ␣GST had been produced by immunizing rabbits with recombinant GST that had been expressed from the pGEX2T vector (Amersham Pharmacia Biotech) in Escherichia coli and had been affinity-purified on glutathione-agarose. For control immunoprecipitations, chromatographically purified nonspecific rabbit IgG antibodies were obtained from Zymed Laboratories Inc. These control precipitations were performed to exclude the possibility that SAPs interacted nonspecifically with either rabbit IgG or the resin during the immunoprecipitation procedures rather than with the respective glutamate receptor subunits. The specificities of all antibodies have been carefully characterized as described earlier (9,22,24,27,29,32).
Affinity Chromatography Using GST Fusion Proteins-For in vitro interaction assays, SAPs and the C termini of GluR1, GluR2, and GluR4 were expressed as recombinant GST fusion proteins in E. coli. Expression constructs for the GST-SAP90, GST-SAP97, and GST-SAP102 fusion proteins are described in Muller et al. (24). SAP97 was also expressed in E. coli as recombinant protein without a tag fused to it using the pRK174 vector (27). To express the final 10 C-terminal amino acids of GluR1, GluR2, or GluR4 fused to the C terminus of GST, complimentary sense and antisense oligonucleotides encoding the respective sequences with engineered endonuclease restriction site overhangs for BamHI added on the 5Ј end and EcoRI on the 3Ј end were annealed and ligated into a BamHI-and EcoRI-digested expression . Please note that stop codons are underlined and that the ten codons immediately upstream of these stop codons encode the very C-terminal amino acids of each of the corresponding AMPA receptor subunits. All constructs were verified by DNA sequence analysis and transformed into the proteasedeficient E. coli strain BL21 (Novagen) for protein production.
GST fusion proteins were expressed and purified according to the protocols suggested by the manufacturer (Amersham). Briefly, overnight cultures from single colonies were grown in 50 ml of LB medium containing 100 g/ml ampicillin at 37°C with aeration to saturation, diluted 1:10 with LB medium, and incubated under the same conditions for 2-4 h until the culture reached an A 600 of about 1.0. For induction, isopropyl-␤-D-thiogalactopyranoside (0.1-0.5 mM) was then added, and the bacteria were incubated for another 4 -5 h and harvested by centrifugation. Bacteria were resuspended in 50 ml of 150 mM NaCl, 15 mM Tris-Cl, pH 7.4, (TBS) containing 0.1 mg/ml lysozyme and gently sonicated on ice with a tip sonicator, applying three bursts for 10 s on a low intensity setting. Triton X-100 was added to a final concentration of 1%, and after 20 min, insoluble material was removed by centrifugation (30,000 rpm, 60 min, 4°C, 45Ti rotor). The supernatant was incubated with 1 ml glutathione-Sepharose (Amersham) at 4°C for 1-2 h, and the resins were extensively washed with TBS.
To measure in vitro interactions between dissociated AMPA receptor subunits and recombinant GST-SAP fusion proteins, 20-l resin samples loaded with 1 g of GST-SAP90, GST-SAP97, or GST-SAP102 fusion proteins (see preceding paragraph) were incubated with 2 ml of Triton X-100-neutralized SDS extracts of brain membranes (see previous section), washed three times with 1% Triton X-100, 150 mM NaCl, 10 mM Tris-Cl, pH 7.4, once with 10 mM Tris-Cl, pH 7.4, and then extracted with SDS sample buffer for immunoblotting as described (29).
To investigate the role of the C termini of AMPA receptor subunits with respect to the interaction with SAP97, GST fusion proteins carrying the C-terminal amino acids of GluR1, GluR2, or GluR4 were expressed and adsorbed to glutathione-Sepharose as described above. After thorough washing with TBS, the GST fusion proteins were eluted with 15 mM glutathione, 150 mM NaCl, 50 mM Tris-Cl, pH 8.2. Recombinant SAP97 was expressed from the pRK174-SAP97 vector without a tag (see above) in the E. coli strain BL21DE3 (Novagen) as described above with the following variation. Cultures were grown until an A 600 of 0.6 was reached and induced with 1 mM isopropyl-␤-D-thiogalactopyranoside for 3 h. Cultures were harvested, the bacteria were lysed, and the recombinant proteins were solubilized as above. The cleared lysates were frozen at Ϫ80°C for storage. SAP97 was immunoprecipitated from these lysates using ␣SAP97 and protein G-Sepharose, and the precipitates were washed with TBS and then incubated with 1 g of the GST fusion proteins carrying the C-terminal amino acids of GluR1, GluR2, or GluR4. After further washing, the pellets were extracted for immunoblotting with ␣GST.
In contrast to NMDA receptors, AMPA receptors are completely solubilized by Triton X-100 (Fig. 1). Immunoblotting of either a crude membrane fraction from rat brain or a Triton X-100 extract or a deoxycholate extract of the membrane fraction with ␣GluR1 yielded equally strong signals in all three fractions. Because identical amounts of the crude membrane fraction had been used as starting material for all three samples, these results show that not only deoxycholate but also Triton X-100 can efficiently solubilize AMPA receptors. After extraction with deoxycholate, SAP102 is not associated with AMPA receptor ␣GluR1 immunocomplexes (Fig. 2, lane 1). Taken together, these findings indicate that AMPA receptors do not form a deoxycholate-resistant complex with SAP102 or the other two SAPs, which are inefficiently extracted with Triton X-100 and are usually found associated with NMDA receptors.
SAP97 Coimmunoprecipitates with the AMPA Receptor Complex Extracted with Triton X-100 -The synaptic function of SAP97, the fourth SAP family member, has not been elucidated. Like AMPA receptors, SAP97 is readily extracted by Triton X-100 (Fig. 1). Furthermore, one of the AMPA receptor subunits possesses an established C-terminal consensus se-quence for PDZ domain binding. We, therefore, tested whether SAP97 is associated with AMPA receptors. Triton X-100 solubilizes AMPA receptors as a complex containing several different AMPA receptor subunits (9). Immunoprecipitation of Triton X-100-solubilized AMPA receptor complexes with either ␣GluR1 or ␣GluR2/3 antibodies resulted in coprecipitation of a protein immunoreactive to SAP97 antibodies (Fig. 2, lanes 4  and 5). Immunoprecipitation with control antibodies (see "Experimental Procedures") did not yield SAP97-immunoreactive bands (Fig. 2, lane 6), thus excluding the possibility that SAP97 directly bound nonspecifically either to the resin or to antibodies not directed against the AMPA receptor-SAP97 complex.
It is possible that, after initial solubilization, SAP97 would then be free to bind to GluR1 or other proteins carrying a C-terminal PDZ binding consensus sequence. Like all the NR2 subunits, four of the eight NR1 splice isoforms possess a Cterminal consensus sequence for PDZ domain binding (33). Therefore, we tested whether SAP97 would bind to NR1 subunits solubilized with Triton X-100. A significant portion of NR1 can be extracted by Triton X-100 from crude brain membrane fractions. These NR1 subunits are not associated with NR2 subunits. They may derive from microsomes that originate from the endoplasmic reticulum and may correspond to an immature NR1 form that is not assembled with NR2 subunits (11). We did not detect the presence of a SAP97-immunoreactive protein in NR1 immunoprecipitates from these Triton X-100 extracts (Fig. 2, lane 7). These observations make it unlikely that SAP97 associates after solubilization with potential but physiologically irrelevant binding partners such as NR1 subunits containing a C-terminal PDZ binding sequence. Accordingly, the association of SAP97 with AMPA receptor immunocomplexes appears to be specific.
SAP97 Coprecipitates with GluR1 but Not GluR2 or GluR3 after Cross-linking and Solubilization with SDS-After solubilization of brain extracts with SDS, ␣GluR1 immunoprecipitated GluR1 (Fig. 3A, lane 1) but not GluR2/3 (lane 4). Similarly, precipitation with ␣GluR2/3 yielded GluR2/3 but no GluR1 immunoreactivity (Fig. 3A, lanes 5 and 2, respectively). Accordingly, SDS dissociates the different AMPA receptor subunits from each other. To investigate whether SAP97 may interact with GluR1, membrane fractions were treated with DSP. DSP is a homobifunctional cross-linker containing an internal disulfide bridge. The disulfide bridge allows cleavage by reducing agents and subsequent detection of the DSP-linked proteins at their normal apparent molecular masses observed without cross-linking. After cross-linking with DSP, the membrane fractions were extracted with SDS to dissociate GluR1 from GluR2 and GluR3. Under these conditions, SAP97 immunoreactivity coprecipitated with GluR1 but not with GluR2/3 nor with NMDA receptor immunocomplexes or with control antibodies (Fig. 3B). These findings indicate that DSP specifically linked SAP97 to the GluR1 subunit, suggesting that GluR1 directly interacts with SAP97 in vivo. In contrast, SAP102 immunoreactivity that coprecipitated with the NMDA receptor complex was not associated with immunoprecipitated GluR1, GluR2, or GluR3 subunits following the same protocol (Fig. 3C). Accordingly, DSP does not cross-link AMPA receptor subunits to those SAPs that are known to be associated with NMDA receptors rather than AMPA receptors. This control corroborates the specificity of GluR1-SAP97 cross-linking by DSP.
SDS does not dissociate NMDA receptor-SAP complexes at 4°C (Refs. 25 and 28 and data not shown); accordingly, the presence of SAP102 in NMDA receptor complexes immunoisolated from rat brain membranes that had been treated with DSP and solubilized with SDS does not necessarily indicate Proteins indicated on the left side were detected with corresponding antibodies described under "Experimental Procedures." Note that SAP97 and SAP102 immunoreactivity migrates at around 120 and 100 kDa, respectively, and can often be separated into doublets that may arise from differential splicing (24,27). Positions of marker proteins are indicated on the right, together with their molecular masses.

FIG. 2. Coimmunoprecipitation of SAP97 and AMPA receptors.
Crude membrane fractions were solubilized with deoxycholate (lanes 1-3) or Triton X-100 (lanes 4 -7), and the proteins indicated at the bottom were immunoprecipitated (IP) with the respective antibodies before immunoblotting with ␣SAP102 (lanes 1-3) or ␣SAP97 antibodies (lanes 4 -7). SAP102 coprecipitated with deoxycholate-extracted NMDA but not AMPA receptor complexes (lanes 1 and 2) and SAP97 with Triton X-100-solubilized AMPA receptors (lanes 4 and 5) but not control antibodies or NR1 subunits (lanes 6 and 7). Positions of marker proteins are given on the left. that SAP102 had been cross-linked by DSP to NMDA receptor subunits. To investigate this point, SDS-polyacrylamide gel electrophoresis of cross-linked NMDA receptor-SAP complexes was performed under nonreducing conditions. Under these conditions, the disulfide bridge in the middle of the cross-linker is not cleaved, and cross-linked polypeptides will still be covalently bound to each other during the gel electrophoresis, often migrating as a very broad band with a high apparent molecular mass. Immunoblotting of cross-linked NMDA receptor-SAP complexes showed SAP102 immunoreactivity from the top of the SDS-polyacrylamide gel to about 150 kDa (not shown); such immunostaining was not detectable under reducing conditions (Fig. 3C). These observations indicate that DSP can also cross-link SAP102 to NMDA receptor subunits, but as expected, the linker is cleaved by reducing agents. Although we did not analyze in detail to which of the NMDA receptor subunits SAP102 is covalently linked by DSP, these results confirm that SAP102 can be cross-linked by DSP to its natural binding partner. Accordingly, the absence of SAP102 immunoreactivity in ␣GluR1 immunoprecipitates argues against puta-tive nonspecific cross-linking of various SAPs with GluR1.
GluR1 Directly Associates with SAP97-To test whether GluR1 can directly bind to SAP97, AMPA receptor subunits were dissociated with SDS, SDS was diluted with Triton X-100, and the extracts were applied to affinity chromatography on glutathione-Sepharose loaded with either GST-SAP90, GST-SAP97, or GST-SAP102. Similar amounts of these three fusion proteins were present on the affinity resins as confirmed by SDS-polyacrylamide gel electrophoresis and Coomassie staining (Fig. 4B). Only GluR1 but not GluR2 or GluR3 bound to GST-SAP97 (Fig. 4A, lanes 3 and 9, respectively). None of these AMPA receptor subunits associated with GST-SAP90 or GST-SAP102, indicating that the interaction between GluR1 and GST-SAP97 is specific with respect to the AMPA receptor subunit as well as the SAP. Of note, when equal amounts of the same extract were probed by immunoblotting with ␣GluR1 and ␣GluR2/3 antibodies, immunostaining of similar intensity was observed on the same exposures (Fig. 4C). Thus, the detection sensitivity for GluR1 and for GluR2/3 was very similar, and the lack of GluR2/3 immunoreactivity after affinity chromatography with the GST-SAP fusion proteins was not due to a difference in sensitivity of the respective antibodies.  ␣GluR1 (lanes 1 and 4) or ␣GluR2/3 (lanes 2 and 5) or control antibodies (lanes 3 and 6) before immunoblotting with ␣GluR1 (lanes 1-3) or ␣GluR2/3 (lanes 4 -6). GluR1 and GluR2/3 immunoreactivity was only detectable after immunoprecipitation with the corresponding antibodies, demonstrating that SDS dissociates GluR1 and GluR2/3 subunits from each other. B and C, membrane fractions were treated with DSP before solubilization with SDS, immunoprecipitation of AMPA receptor subunits with ␣GluR1 (lane 1) or ␣GluR2/3 (lane 2), immunoprecipitation of NMDA receptor complexes with ␣NR1 (lane 3) or with control antibodies (lane 4), and immunoblotting with ␣SAP97 (B) or ␣SAP102 (C). SAP97 coprecipitated only with GluR1 but not GluR2/3 or NMDA receptor complexes. SAP102 was associated with NMDA receptor complexes that are (in contrast to AMPA receptor complexes) not dissociated by SDS under these conditions but did not coprecipitate with GluR1 or GluR2/3. Positions of marker proteins are given on the left.

FIG. 4. GluR1 interacts directly with SAP97 in vitro.
A, glutathione-Sepharose (20 l) was charged with about 2 g of GST-SAP90, GST-SAP97, or GST-SAP102 fusion proteins, incubated with (ϩ) or without (Ϫ) Triton X-100-neutralized SDS extracts containing dissociated AMPA receptors, washed, and used for immunoblotting with ␣GluR1 (A, lanes 1-6) or ␣GluR2/3 (A, lanes 7-12). GluR1 immunoreactivity was only detectable after affinity purification on GST-SAP97, which did not bind GluR2/3. Note that no immunoreactivity appeared when extract was absent. B, each SAP-loaded resin (20 l) was also analyzed by SDS-polyacrylamide gel electrophoresis and subsequent staining with Coomassie Brillant Blue, confirming that equal amounts of each SAP fusion protein were present in these experiments. Please note that the GST-SAP102 fusion protein is missing the first 133 amino acids at its N terminus and, therefore, migrates slightly faster than GST-SAP90 fusion protein. The truncation does not involve the first PDZ domain and does not prevent the interaction of this GST-SAP102 fusion protein with NMDA receptors (data not shown). The GST-SAP90 and GST-SAP97 vectors encode the whole SAP90/PSD-95 and SAP97 amino acid sequence, respectively. The identities of the GST-SAP90, GST-SAP97, and GST-SAP102 fusion proteins were confirmed by immunoblotting with ␣SAP90, ␣SAP97, and ␣SAP102 (data not shown). C, identical amounts of extracts were also directly used for immunoblotting with ␣GluR1 (lane 1) or ␣GluR2/3 (lane 2). The immunosignals were of similar strength, indicating that immunodetection was of comparable sensitivity for both antibodies. Positions of marker proteins are given on the left.
The recombinant GST-SAP fusion proteins bound by the glutathione-Sepharose for these affinity interaction tests were present in large quantities on the immunoblots (about 1-2 g/sample) and migrate at apparent molecular masses similar to those of the AMPA receptor subunits. To exclude the possibility that the GluR1 immunoreactivity observed after affinity chromatography on GST-SAP97 was due to any cross-reactivity of ␣GluR1 toward the GST-SAP97 fusion protein, affinity resins were also mock-treated in the absence of brain extracts and prepared in parallel with extract-treated affinity resins for immunoblotting. Neither ␣GluR1 nor ␣GluR2/3 cross-reacted with any of the three fusion proteins (Fig. 4A, even-numbered  lanes).
SAP97 Specifically Binds to the C-terminal End of GluR1-GST fusion proteins carrying the sequence of the 10 C-terminal amino acids of GluR1, GluR2, or GluR4 at their C termini were produced and purified (Fig. 5A). Note that GluR1 possesses a threonine in the Ϫ2 position and a leucine in the 0 position, a consensus sequence for interaction with SAPs. GluR2 carries a serine in the Ϫ3 rather than the Ϫ2 position and in vitro also binds to PDZ domains, namely to PDZ domain 4 and 5 of glutamate receptor interacting protein (GRIP) (Ref. 34; see "Discussion"). The C terminus of GluR2 was included in our studies to test for the specificity of SAP97 association with C-terminal sequences capable of binding to PDZ domains. The very C-terminal amino acid of GluR4 is proline, and the Cterminal GluR4 sequence is not expected to interact with PDZ domains. Therefore, the C terminus of GluR4 was chosen as a likely nonspecific negative control.
Recombinant SAP97 was expressed in E. coli, solubilized with detergent, and immunoprecipitated using protein A-Sepharose and ␣SAP97. This antibody is directed against the N terminus of SAP97 and should not occlude binding of the three PDZ domains of the immuno-isolated SAP97 to potential target proteins such as GluR1. The immunoprecipitates were incubated with the fusion proteins consisting of GST and the C termini of GluR1, GluR2, or GluR4 followed by extensive washing. The association of these fusion proteins with recombinant SAP97 was analyzed by immunoblotting with ␣GST. A significant amount of GST-GluR1 bound to SAP97 (Fig. 5D, lane 1). In contrast, GST-GluR4 was not detectable (Fig. 5D, lane 3), demonstrating that the interaction between GST-GluR1 and SAP97 was specific. A minor amount of GST-GluR2 bound to SAP97 (Fig. 5, lane 2), but GST-GluR2 binding was hardly detectable. This result is in agreement with earlier observations that in vitro binding assays between PDZ domains and target consensus sequences can result in interactions not seen in intact cells. However, the apparent affinity of those in vitro interactions is much lower than of those that occur in vivo (Ref. 35 and data not shown). Accordingly, the binding between SAP97 and GST-GluR1 appears in this in vitro assay to be much stronger than the binding between SAP97 and GST-GluR2. Therefore, these findings corroborate that SAP97 specifically associates with GluR1 and indicate that this interaction is likely to be mediated by the C terminus of GluR1.

DISCUSSION
Evidence for the Specific Interaction between GluR1 and SAP97 in Vivo-Our studies show that AMPA receptor complexes solubilized with Triton X-100 contain SAP97. After dissociation of AMPA receptor subunits with SDS, GluR1, but not GluR2 or GluR3, binds to recombinant SAP97 rather than to SAP90/PSD-95 or to SAP102, two SAPs that are known to interact with NMDA receptor subunits. Similar to most other proteins that have been shown to associate with PDZ domains, the interaction between GluR1 and SAP97 is mediated by the C terminus of GluR1; recombinant SAP97 specifically associated with a GST construct carrying the C terminus of GluR1. The presence of SAP97 immunoreactivity in the Triton X-100 solubilized AMPA receptor complex and the high degree of specificity of the in vitro interactions between GluR1 and SAP97 strongly argue that the interaction between GluR1 and SAP97 occurs in vivo. This interaction is corroborated by the crosslinking experiments that demonstrate that SAP97 is closely associated with GluR1 in the plasma membrane ( Fig. 3; see below). Furthermore, immunohistochemical characterization of SAP97 suggests that it may be co-localized with AMPA receptors at the postsynaptic site (27). Immuno-electron microscopy with ␣SAP97 in combination with a secondary antibody coupled to horseradish peroxidase showed that the electron-dense reaction product of horseradish peroxidase was not only associated with presynaptic nerve terminals but also to a significant extent with postsynaptic sites. Accordingly, SAP97 may be present not only at presynaptic but also at postsynaptic sites (27).
Both SAP97, a membrane-associated cortical cytoskeletal protein, and the AMPA receptor, an integral membrane protein complex, require detergent for solubilization. To test whether SAP97 is closely associated with GluR1 in the plasma mem- FIG. 5. Interaction between the C terminus of GluR1 and Recombinant SAP97. A, this panel shows the sequences of the C-terminal ends of the GST fusion proteins used in the interaction assays with SAP97. These sequences correspond to the 10 amino acids at the Cterminal ends of GluR1, GluR2, and GluR4 and were cloned in frame with GST. B, recombinant SAP97 was expressed in E. coli without a tag, immunoprecipitated with ␣SAP97, and incubated with 1 g of the affinity-purified GST fusion proteins (as determined from A 280 values). The protein A-Sepharose-bound protein complexes were washed and used for immunoblotting. The upper parts of the blots were probed with ␣SAP97, showing that equal amounts of recombinant SAP97 were present in all samples. C, to ensure that comparable amounts of GST fusion proteins were used in each SAP97 binding assay, amounts similar to those applied for these binding assays were also directly subjected to immunoblotting with ␣GST, which was performed in parallel with the immunoblots of the binding assays. D, the lower part of the immunoblots described in B was probed with ␣GST, demonstrating that the C-terminal GluR1 sequence specifically binds to SAP97. Positions of marker proteins are given on the left. brane and to exclude the possibility that SAP97 associated with AMPA receptors after solubilization, we incubated membrane fractions with the cross-linker DSP before extraction with SDS. SDS dissociated GluR1 from other AMPA receptor subunits as well as from SAP97. However, our cross-linking procedure resulted in coprecipitation of SAP97 with GluR1, but not with GluR2 or GluR3, even after extraction with SDS, suggesting that GluR1 directly interacts with SAP97 in vivo. The possibility that one or more additional proteins mediate the interaction between GluR1 and SAP97 and are also cross-linked to GluR1 and SAP97 at the same time cannot be excluded but is unlikely. Such a scenario would require at least two cross-linking reactions within one complex. However, cross-linking reactions are in general very inefficient and our cross-linking conditions were quite mild; they did not result in cross-linking of GluR1 with GluR2 or GluR3 as happens when DSP is used at much higher concentrations (9). In any case, our experiments indicate that SAP97 is either associated with or in close proximity to GluR1 in synaptic junctions (3).
Within the last two years, several interactions between NMDA receptor subunits and structural proteins such as SAPs and ␣-actinin have been described and proposed to be involved in localizing NMDA receptors at synaptic junctions (16,17,28). However, NMDA and AMPA receptors behave quite different with respect to their solubilization properties and probably depend on different sets of structural proteins for synaptic localization. NMDA receptor subunits can be solubilized with Triton X-100 from a membrane fraction that is probably derived from the endoplasmic reticulum. NR1 subunits in this fraction are not yet associated with other components of the NMDA receptor complex (11). However, the mature NMDA receptors present in synaptic membrane fractions require the ionic detergents deoxycholate or SDS for extraction as do their partner SAPs. Furthermore, these two detergents do not disrupt the interactions between NMDA receptors and SAPs. It is, therefore, possible that the interactions between NMDA receptor subunits and SAPs are part of a large postsynaptic matrix that can only be broken up by strong ionic detergents, possibly due to the presence of additional interactions with other proteins such as guanylate kinase-associated proteins/ SAP90/ PSD-95-associated proteins, ␣-actinin, or with the cortical cytoskeleton. AMPA receptors may not be so tightly integrated into this matrix; alternatively, the interactions between components of the AMPA receptor complex and the NMDA receptor complex may be dissociated by SDS. It is tempting to speculate whether the interactions between AMPA receptor subunits and structural proteins such as SAP97 or GRIP are less rigid than those between NMDA receptors and SAP90/PSD-95, SAP102, chapsyn-110/PSD-93 or ␣-actinin, thus allowing a larger degree of flexibility and plasticity for AMPA receptor anchoring than for NMDA receptor anchoring.
Physiological Relevance of GluR1-SAP97 Interaction-Similar to those SAPs that are associated with NMDA receptors, SAP97 may help to localize GluR1-containing AMPA receptors at the synapse. GRIP has very recently been identified as another structural protein that interacts with AMPA receptors (34) 2 . GRIP contains seven PDZ domains but is otherwise quite different from the SAPs described above. It binds to the very C-terminal ends of GluR2 and GluR3 subunits, which are nearly identical to each other (34). This binding is mediated by the fourth and fifth PDZ domain of GRIP. The interaction between GRIP and GluR2 or GluR3 requires that a serine or threonine is in the Ϫ3 position (Fig. 5A) (34). However, interactions between GRIP and GluR2 or GluR3 have only been described in vitro or in heterologous cell lines overexpressing these proteins (34). Therefore, it is unclear whether GRIP is associated with AMPA receptors in vivo.
Synaptic plasticity, which is a change in the efficacy of synaptic transmission, is thought to be the physiological correlate of learning and memory (36). Long lasting forms of synaptic plasticity, such as long term potentiation in the hippocampus, probably require restructuring of the synapse (37). The identification of different sets of structural proteins interacting with AMPA receptors and NMDA receptors suggests that the functional postsynaptic localization of these two glutamate receptors may be regulated in different ways. At fully functional synapses, AMPA receptors can be activated by relatively mild electrical stimulation, whereas NMDA receptors mediating the Ca 2ϩ influx necessary for long term potentiation require stronger stimulation. Many so-called silent synapses exist in the hippocampus that do not show AMPA receptor-mediated responses but exhibit NMDA receptor-mediated responses to a strong stimulus. However, functional AMPA receptors appear within minutes in these synapses when a strong stimulus paradigm is applied that initially activates NMDA receptors (38). AMPA receptors may be present at silent synapses but may reside in a compartment at or near the postsynaptic site where they cannot respond to glutamate. NMDA receptor-mediated Ca 2ϩ influx may directly or indirectly induce a change in the interactions between AMPA receptors and structural proteins, leading to a relocation of AMPA receptors and thereby making them functionally available for synaptic transmission. It is also possible that interactions with certain proteins regulate the activity of AMPA receptors and that changes in protein-protein interactions are necessary to render silent AMPA receptors functional. Defining the interactions between glutamate receptors and structural proteins will not only help to elucidate the architecture of a synapse but will also be indispensable for a molecular understanding of synaptic plasticity.