Subtype-specific Assembly of α-Amino-3-hydroxy-5-methyl-4-isoxazole Propionic Acid Receptor Subunits Is Mediated by Their N-terminal Domains*

Glutamate receptors (GluR) are oligomeric protein complexes formed by the assembly of four or perhaps five subunits. The rules that govern the selectivity of this process are not well understood. Here, we expressed combinations of subunits from two related GluR subfamilies in COS7 cells, the α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid (AMPA) and kainate receptors. By co-immunoprecipitation experiments, we assessed the ability of AMPA receptor subunits to assemble into multimeric complexes. Subunits GluR1–4 associated with indistinguishable efficiency with each other, whereas the kainate receptor subunits GluR6 and 7 showed a much lower degree of association with GluR1. Using chimeric receptors and truncation fragments of subunits, we show that this assembly specificity is determined by N-terminal regions of these subunits and that the most N-terminal domain of GluR2 together with a membrane anchor efficiently associates with GluR1.

Fast excitatory synaptic transmission in the mammalian central nervous system is primarily mediated by multimeric ligand-gated ion channels, which are activated by the neurotransmitter glutamate. Based on pharmacological properties and sequence similarities, these receptors fall into three main classes: the glutamate receptor (GluR) 1 subunits 1-4 form the family of ␣-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid (AMPA) receptors, subunits GluR5-7 and KA1 and KA2 are the family of kainate receptors, and subunits NR1 and NR2A-D are components of N-methyl-D-aspartate receptors (1,2).
Functional receptors of each type are composed of multiple subunits. Recent experiments suggest that each receptor complex contains four subunits (3)(4)(5)(6), whereas previously a pentameric structure has been favored (7)(8)(9). Recombinant expression of an individual type of subunit in nonneuronal cells in many cases leads to the formation of functional homomeric ion channels. Coexpression of two or more subunits of the same class, however, mainly directs the formation of heteromeric channels (10 -13). Their electrophysiological characteristics are different from those of their homomeric counterparts and often are quite similar to channels present in neurons. In vivo, GluRs are thought to be predominantly or perhaps exclusively composed of heteromeric channels, based on their functional similarity with recombinant channels as well as co-immunoprecipitation experiments (7, 14 -16).
The subunit composition of GluRs determines the electrophysiological characteristics of these ligand-gated ion channels. For example, AMPA receptors lacking the GluR2 subunit are permeable for calcium, whereas channels containing at least one GluR2 subunit do not allow calcium entry into the cells (10,17). A number of studies have demonstrated that many neurons express several subunits from different subfamilies of GluRs (15, 18 -20). This diversity is further enhanced by alternative splicing of most subunits (21,22).
The rules guiding the assembly of GluRs from individual subunits are not known. One possibility is that association occurs randomly and the composition of the GluRs in each cell is simply determined by the availability of individual subunits. An alternative possibility is that subunits preferentially interact with one or a small group of other subunits and thereby favor the formation of GluRs with a distinct subunit composition. A classical example for such a stereotypic receptor assembly pathway is the muscle nicotinic acetylcholine receptor. N-terminal domains of individual subunits of this neurotransmitter receptor initiate association of specific subunits in an ordered process (23)(24)(25)(26).
Regions on GluR subunits mediating assembly of homomeric or heteromeric receptors have not been identified so far. All GluR subunits are thought to share a common transmembrane topology and domain structure (27)(28)(29)(30). The glutamate-binding region is formed by two extracellular domains, which are separated in the protein sequence by two transmembrane regions (31)(32)(33). A loop between these hydrophobic domains is part of the ion channel. A third transmembrane region defines the border of a C-terminal intracellular domain of variable size, which in many subunits is a target for binding of several PDZ-domain-containing proteins (34 -37). No function has been ascribed to date to the proximal N-terminal region. It forms a separate domain homologous to the bacterial periplasmic leucine-isoleucine-valine-binding protein (38).
Here, we addressed the question of which domains of AMPA receptor subunits mediate their assembly into receptor complexes. We expressed different pairs of subunits in COS7 cells and analyzed subunit association by immunoprecipitation. All AMPA receptor subunits but not kainate receptor subunits associated with the GluR1 subunit with high efficiency. Using chimeric receptors and truncation fragments of subunits, we show that this subfamily specificity of assembly is determined by N-terminal regions of subunits.

cDNA Clones
The cDNA clones GluR1 (flop), GluR2 (flip) GluR4 (flip), and GluR7 were kind gifts from R. Sprengel and P. Seeburg (Heidelberg); the clones of the AMPA receptor subunit GluR3 (flip) and the kainate receptor subunit GluR6 as well as the GluR1-and GluR6 derivatives (R1-PCS and R6-PCS) used for the construction of the GluR1-GluR6 chimeras were kind gifts from M. Hollmann (Göttingen).

Vector Constructs
HV Vector with N-terminal Hexa-myc-tag (HV-N-myc)-The hexamyc-tag was amplified from the pCS2ϩMT vector (39) by PCR (primers: TCC CAT CGA TCT GCA GCT ATG GAG and GAG AGG CCT TGC ATG CAA GTC CTC TTC) and subcloned into the HV vector (40) between the PstI and SphI sites.
pCMV2 vector with C-terminal Hexa-myc-tag (pCMV2-C-myc)-The polylinker of the pCMV2-expression vector (41) was expanded by introducing a NotI site; the plasmid Sac-KiSS- (42) was digested with XbaI, and the resulting fragment containing the NotI site was inserted into the XbaI site of pCMV2. By digestion with NotI and re-ligation, the insert was eliminated resulting in a pCMV2 with a NotI site between two XbaI sites (pCMV2-Not). The PCR-amplified hexa-myc-tag (primers: CAT CGA TTT AAA GCG GCC GCT ATG GAG CAA and TAG TTC TAG AGT CTA GAG AGG CCT TGA) was introduced between the NotI and the second XbaI site, the TAG of XbaI forming the stop codon.

myc-tagged Constructs
Constructs that were to be tagged at their N terminus were first amplified by PCR (the 5Ј-primer contained an SphI site, the 3Ј-primer either an XbaI site or a NotI site), inserted into HV-N-myc between SphI and XbaI and then subcloned into pCMV2 using ClaI and XbaI except for GluR7N, where KpnI and XbaI was used. Constructs with a C-terminal myc-tag were PCR-amplified, inserted into the HV vector without myc-tag between SphI and NotI, and then subcloned into pCMV2-C-myc using ClaI and NotI.
Since ClaI and XbaI cut in the coding region of the AMPA subunit GluR4, the polylinker of HV-N-myc was modified by introducing a MluI site between SalI and ClaI. The PCR-amplified GluR4 sequence was inserted into the modified HV-N-myc vector with SphI and NotI and subcloned into the pCMV2-Not vector using MluI and NotI.
All full-length constructs start at position 1 behind the signal sequence (13). The position of the first and last amino acid of the deletion fragments are numbered according to Keinä nen et al.

Chimera with the Pore Loop Region of GluR6 (R1-PL6-R1)
GluR1-PCS and GluR6-PCS (44) were digested with NruI and EcoRI, and the fragment containing the pore loop region of GluR6 was inserted at the corresponding position of GluR1-PCS. The pore loop cassette was then excised using BglII and introduced into the GluR1-expression vector (pCDM8, Ref. 13).
Untagged GluR6-GluR1 Chimeras (R6-PL1-R1 and R6-PL6-R1) R1-PCS and R6-PCS were digested either by NotI and NruI (for R6-PL1-R1) or NotI and EcoRI (for R6-PL6-R1). The R6 fragment consisting of the N terminus with or without the pore loop was inserted into R1-PCS. Then these chimeras were digested with XhoI, the ends treated with the Klenow fragment of DNA polymerase I, and digested with NarI. The NarI/(XhoI) fragment was inserted between NarI and SmaI of untagged GluR6 (in pCMV2), thus yielding the final chimeric constructs in the pCMV2 expression vector.

GluR1-GluR6
Chimeras with an N-terminal myc-tag (R1N-PL6-R6 and R1N-PL1-R6) R1-PCS and R6-PCS were digested either by NotI and NruI (for R1N-PL6-R6) or NotI and EcoRI (for R1N-PL1-R6). The R1 fragment consisting of the N terminus with or without the pore loop was inserted into R6-PCS. These chimeras were digested with XmaCI and inserted into the N-terminally myc-tagged GluR1 (in the pCMV2) that had been digested with the same enzyme.

Deletion Fragments Fused to Transmembrane Domains (C1-TMD C, C2-TMD C, and C1-TMD A)
Fragments C1 and C2 as well as the C-terminal part containing the last putative transmembrane domain of GluR2 (TMD C) were amplified by PCR. The PCR products C1 and C2 were digested by SphI and NotI and inserted into HV-N-myc (see above).
PCR product TMD C was inserted into pCMV2 by exchanging the myc-tag of pCMV2-C-myc for TMD C using NotI and XbaI.
The N-terminally myc-tagged sequences for C1 and C2 were then subcloned from HV into pCMV2 (with TMD C) using ClaI and NotI yielding C1-TMD C and C2-TMD C.
C1-TMD A was constructed using two sense and two antisense oligonucleotides representing the first putative transmembrane domain of GluR2 (TMD A) containing a NotI site at the 5Ј-end and an XbaI site at the 3Ј-end as well as sticky ends corresponding to the mid-region of TMD A. Equal amounts of each oligonucleotide were used for ligation into pCMV2-Not between NotI and XbaI. C1 was then subcloned from HV-N-myc into pCMV2 containing the first transmembrane domain using ClaI and NotI.
The integrity of the constructs was verified by specific restriction analysis, expression in COS7 cells, and partial sequencing across regions connecting original sequences with the myc-tag or the transmembrane domains.

Antiserum Directed against the GluR1 Subunit
A cDNA fragment corresponding to the C terminus of GluR1 (amino acids 811-889, Ref. 13) was amplified by PCR (primers: CCT TAA TCG GAT CCT GCT ACA AAT and GGC ACT GCA GGG CTT GG) and inserted into the pQE30 vector (Qiagen) between BamHI and PstI. The fusion protein containing an N-terminal hexahistidine tag was overexpressed in Escherichia coli and purified as described (43). Rabbits were immunized with the fusion protein in Freund's adjuvant. After the fourth boost, the serum was purified over an affinity column with the fusion protein coupled to a mixture of Affi-Gel 10 and 15 (Bio-Rad).

Cell Culture and Transient Transfections
One day prior to transfection, COS7 cells were seeded at a density of 700,000 cells/10-cm culture dish in COS medium (10% (v/v) fetal calf serum in Dulbecco's modified Eagle's medium). COS7 cells were transfected transiently as described previously (45) with some minor variations that were found to improve transfection efficiency; the transfection mix contained 40 g of DNA/10-cm culture dish, and the incubation time was reduced to 6 h, thereby increasing cell survival. Usually, cells were harvested on the third day after transfection in phosphate-buffered saline containing 5 mM EDTA, 5 mM EGTA, 1 mM phenylmethylsulfonyl fluoride, and 1 mM benzamidine.

Cell Lysis
COS7 cells were lysed with 500 l of ice-cold solubilization buffer (SB: 1% (w/v) Triton X-100, 500 mM NaCl, 5 mM EGTA, 5 mM EDTA in phosphate-buffered saline, pH 7.1) containing freshly added protease inhibitors (1 mM phenylmethylsulfonyl fluoride, 1 mM benzamidine, 1 g/ml leupeptin, 1 g/ml pepstatin, 1 g/ml aprotinin) for 2 h at 4°C in a head-over-head shaker. Then, the lysate was centrifuged at 100,000 ϫ g for 1 h at 4°C and the protein concentration of the supernatant was determined using bicinchoninic acid (BCA, Pierce) with bovine serum albumin as standard. In the solubilization experiments shown in Table  I, proteins from 200 l of the detergent extract were precipitated and together with a corresponding aliquot of the insoluble fraction analyzed by SDS-PAGE and Western blotting.

Co-immunoprecipitation Assay with Streptavidin-Agarose
25 l of streptavidin-agarose were incubated with 1 g of biotinylated anti-rabbit F(abЈ) 2 fragment in SB for 1 h at 4°C in a head-overhead shaker. After three washes with SB, the beads were incubated with anti-GluR1 pAb in SB (1 h, 4°C). The beads were washed three times in SB before adding 500 -600 l of cleared lysate containing 0.5 mg/ml protein. After incubation overnight at 4°C followed by three washes with SB, the beads were incubated with anti-myc mAb in SB for 1 h, followed by another three washes with SB. Finally, peroxidaselinked anti-mouse Ig was added in SB and incubated for 1 h at 4°C. The beads were then washed three times with SB and once with peroxidase reaction buffer (100 mM sodium acetate, 50 mM NaH 2 PO 4 , pH 6). 300 l of peroxidase reaction buffer containing 2 mM 2,2Ј-azinobis(3-ethylbenzthiazolinesulfonic acid) and 0.016% H 2 O 2 were added and incubated for approximately 10 min at room temperature. The beads were pelleted by centrifugation, and the absorbance of the supernatant measured at 405 nm.

Normalization of Co-immunoprecipitation Data
The amount of co-immunoprecipitated myc-tagged protein was normalized with respect to the amount of the same protein present in the cell lysate. The data were then corrected for the amount of either GluR1 or the chimeras R1-PL6-R1, R6-PL1-R1, or R6-PL6-R1 by using the calibration curve given in Fig. 2. For better comparison between single results, the association efficiency was expressed as percentage of the association obtained for the combination of GluR1 and myc-tagged GluR2 ("R1 ϩ R2N").

Determination of the Extent of Co-immunoprecipitation of GluR2
with GluR1 25 l of monomeric avidin-beads were incubated with 2 mM D-biotin in solubilization buffer at 4°C (three times, 20 min each) to block biotin-binding sites of residual oligomeric avidin. The biotin was then removed from monomeric avidin by washing the beads three times (5 min each) with elution buffer (100 mM glycine, pH 2.5, 1% (w/v) Triton X-100). The beads were incubated first with biotinylated anti-rabbit F(abЈ) 2 fragment in SB, then with anti-GluR1 pAb (see above) before 400 l of detergent extract (protein concentration 0.5 mg/ml) were added and incubated overnight. The beads were washed three times with SB, and then the precipitated proteins were eluted by incubating the beads twice with 100 l of elution buffer for 15 min at room temperature. The two eluates were combined, and the amount of coprecipitated myc-tagged protein as well as its amount present in the solubilized fraction was determined by dot blotting (see below).

Western Blotting
Proteins were precipitated from detergent extracts (46) and separated by SDS-PAGE (47). After transfer to nitrocellulose (48) using the buffer system described by Eckerskorn et al. (49), incubation with the primary antibodies as indicated in the figure legends and the corresponding secondary antibodies, immunoreactive bands were visualized by chemiluminescence using SuperSignal (Pierce) as a substrate.

Dot-Blot Analysis
Dot-blot analysis was performed as described (50) with slight modifications: 25-30 l of cleared lysate (protein concentration: 1 mg/ml) were immobilized on a nitrocellulose membrane (diameter: 15 mm). The air-dried membrane was incubated with either the anti-myc mAb or the anti-GluR1 pAb and secondary antibodies conjugated to horseradish. After the last washing step, membrane dots were punched out and incubated with 250 l of reaction buffer (100 mM NaH 2 PO 4 , pH 4.5, containing 0.4 mg/ml o-phenylene diamine and 0.01% H 2 O 2 ) for 5-20 min. The reaction was stopped by adding 105 l of 30% H 2 SO 4 and the absorbance measured at 492 nm.

Surface Staining of myc-tagged Proteins
COS7 cells were transfected with cDNAs coding for various myctagged proteins. On the third day after transfection, anti-myc mAb was added to the culture medium and incubated for 2 h at room temperature. After three washes with phosphate-buffered saline containing 1 mM Ca 2ϩ and 0.5 mM Mg 2ϩ (PBS Ca/Mg), the cells were incubated for 30 min with blocking buffer (5% goat serum and 1.5% bovine serum albumin in PBS Ca/Mg). The cells were then exposed for 1 h to Cy3conjugated secondary antibody in blocking buffer and washed extensively with PBS Ca/Mg.

Expression of Recombinant GluR Subunits and Generation of
a Subunit-specific Antiserum-We wanted to monitor the assembly of AMPA and other receptor subunits into heteromeric receptors by co-expressing epitope-tagged and untagged subunits in COS7 cells and immunoprecipitating heteromeric receptors with a subunit-specific antibody. To this end, we cloned the following cDNAs into expression vectors appropriate for transient transfection in COS7 cells: the four AMPA receptor subunits (GluR1-4), two kainate receptor subunits (GluR6 and 7), and a nonrelated membrane protein of comparable size, the muscle-specific kinase (MuSK). These subunits were either expressed untagged or contained six repeats of a peptide derived from the Myc protein. This epitope tag was fused either to the C-terminal end of the proteins or to the N terminus behind a signal sequence and allowed us to detect all expressed receptor proteins with a single antibody. Thus, we avoided quantification problems originating from the use of multiple antibodies with different affinities. Transfection of all constructs into COS7 cells yielded proteins of the expected size, as seen on Western blots probed with the myc-tag-specific antibody (Fig. 1A).
For specific immunoprecipitation of homomeric and heteromeric GluRs, we generated an antiserum recognizing selectively the AMPA receptor subunit GluR1. The serum was directed against a fusion protein that consisted of an N-terminal hexahistidine tag followed by the C-terminal sequence of GluR1. This part of the subunit is localized intracellularly and represents the most divergent region between different GluR subunits. Antibodies affinity-purified from this serum recognized a band of the expected size in COS7 cells that had been transfected with expression constructs encoding both untagged and myc-tagged GluR1 subunits. No cross-reactivity with any of the three other AMPA receptor subunits: the kainate receptor subunits GluR6 and 7 or the control protein MuSK, was detected (Fig. 1B). The serum was able to immunoprecipitate FIG. 1. Specificity of anti-GluR1 antibody and expression of myc-tagged proteins. COS7 cells were transfected with various fulllength constructs of either AMPA receptor subunits (R1-R4), the kainate receptor subunits GluR6 and 7 (R6, R7), or the control protein MuSK. Cells were solubilized, and the same amounts of specific protein were subjected to SDS-PAGE. After transfer to a nitrocellulose membrane, the blots were probed with either the anti-myc (9E10)-mAb (A) or the anti-GluR1-pAb (B). All constructs but one (R1, first lane) contained a myc-tag at either the N-or C terminus. more than 90% of GluR1 subunit from detergent extracts of transfected COS7 cells, as determined by Western blotting and dot blotting (data not shown).
Heteromeric Assembly of AMPA Receptors-In order to assess the influence of subunit concentration on heteromeric receptor assembly, we expressed a constant amount of epitopetagged GluR2 subunit with variable amounts of untagged GluR1 subunit. Receptors containing at least one GluR1 subunit were immunoprecipitated, and the fraction of GluR2 subunits present in heteromeric complexes was measured by a quantitative immunoassay. In the presence of a considerable excess of GluR1 subunits, about 40% of total GluR2 subunits were immunoprecipitated by the GluR1-specific antiserum and therefore had been assembled into heteromeric complexes (Fig.  2). This fraction only slightly decreased over a wide range of expression levels of the GluR1 subunit. Only below a threshold when GluR1 presumably was present in limiting amounts, the fraction of GluR2 subunit in heteromeric receptors strongly decreased with lower GluR1 concentrations. In all of the following co-transfection experiments, the GluR1 concentration was above the critical concentration of 50% (Fig. 2).
To estimate the size of receptor complexes present in COS cells, we analyzed detergent extracts containing either GluR2N alone or GluR2N and GluR1 by sucrose gradient centrifugation. In both cases, two peaks sedimentating at about 5.2 and 9.8 S were found, the first probably representing monomeric subunits, the second partially or fully assembled complexes. 2 Apparently, about half of GluR2N did not assemble into homomeric or heteromeric complexes.
Next, we addressed the question whether individual GluR subunits differ in their ability to co-assemble with the GluR1 subunit. The AMPA receptor subunits GluR2, 3, and 4, the kainate receptor subunits GluR6 and 7, as well as the MuSK as control protein were each transfected into COS7 cells together with the GluR1 subunit. The extent of heteromeric assembly was analyzed by co-immunoprecipitation.
Each of the AMPA receptor subunits co-assembled with the GluR1 subunit with comparable efficiency (Fig. 3). We found no evidence for a preference of the GluR1 subunit for assembly with one of the other AMPA receptor subunits. In contrast, the kainate receptor subunits GluR6 and 7 reached only about 40% of this level of association. Nevertheless, both subunits had a higher tendency to assemble with AMPA receptor subunits as compared with the control protein MuSK. This non-GluR-related protein displayed approximately 20% of the association seen with GluR2, which as our standard represented 100% assembly in all of the following experiments.
The great majority of protein association took place within cells; only a little occurred during the immunoprecipitation procedure. When extracts of cells expressing GluR1 or another GluR subunit or MuSK separately were mixed after solubilization, at most 10% of the level observed upon co-expression of AMPA receptor subunits was co-precipitated. Furthermore, no GluR2 immunoreactivity was detected in precipitates from cells expressing GluR2 but not GluR1. Control experiments were performed in parallel to all our co-immunoprecipitation experiments with other constructs with similar results (data not shown).
To identify any effects of epitope tags on the association of subunits in our assay, we expressed GluR2 with a tag added either to the N terminus (R2N) or to the C terminus (R2C). Both proteins were co-precipitated with similar efficiency, suggesting that the tags did not interfere with receptor assembly (Fig. 3). Addition of a different N-terminal tag to GluR1 has no effect on functional properties and synaptic targeting of this subunit (51).
Assembly of Chimeric AMPA Receptor/Kainate Receptor Subunits-In our co-immunoprecipitation experiments, AMPA receptor subunits preferentially assembled with each other into heteromeric receptors. To determine which regions of the GluRs were responsible for selective oligomerization, we constructed a number of chimeric receptor subunits consisting of parts of the AMPA receptor subunit GluR1 fused to complementary regions of the kainate receptor subunit GluR6. Chimeras containing the epitope for the GluR1-specific antibody were co-expressed with tagged GluR2 or GluR6 subunits, and their ability to assemble with one of these subunits was determined by co-immunoprecipitation (Fig. 4). The chimeric subunit R1-PL6-R1, which contains only the channel-lining segment of GluR6, assembled with GluR2 with an efficiency similar to that for the GluR1 subunit (Fig. 4C) FIG. 3. Co-immunoprecipitation of myc-tagged proteins with GluR1. COS7 cells were co-transfected with cDNAs coding for GluR1 and myc-tagged GluR subunits or the control protein MuSK. After solubilization of the cells, the GluR1 subunit was immunoprecipitated and the amount of co-precipitated myc-tagged protein was determined. All values were corrected for expression levels of both GluR1 and the myc-tagged protein, using the dot-blot procedure described under "Experimental Procedures" and the curve shown in Fig. 2. Relative association of myc-tagged proteins with GluR1 is shown as percentage of the association obtained for the combination of GluR1 with GluR2 (R1 ϩ R2N, 100%). R2N indicates the association obtained from cells transfected with GluR2N alone. R1 mixed with MuSK, R6N, or R2N describes the association obtained from cells transfected separately with cDNAs coding for GluR1 or the myc-tagged protein; before precipitation with the anti-GluR1 pAb, the solubilized fractions of the singly transfected COS7 cells were mixed at a ratio of 1:1. The data represent the mean Ϯ S.E. of three different transfection experiments except R1ϩR7N and the mixing experiments, which were duplicates. co-precipitated GluR6 subunits (Fig. 4C), whereas chimeras including the N-terminal half of GluR1 did not associate efficiently with this subunit.
The complementary constructs containing C-terminal GluR6 sequence could not be assessed by the same type of co-immunoprecipitation assay, since they do not contain the epitope for the antibody used for precipitation. To study assembly of these constructs, we cloned the myc-epitope tag between the signal sequence and the N terminus of the chimeric receptors (Fig.  5A). After Western blot analysis (Fig. 5B), these constructs were co-transfected with unlabeled GluR1 into COS7 cells and the expressed proteins were immunoprecipitated using the GluR1 antiserum. Chimeras containing N-terminal GluR1 sequence either with or without the pore loop and C-terminal GluR6 sequence co-immunoprecipitated with the GluR1 subunit with an efficiency indistinguishable from that of GluR2 (Fig. 5C).
Taken together, these data show that the regions responsible for the subclass-specific assembly of GluRs largely reside in the N-terminal half of their subunits.
Recombinant Expression and Co-immunoprecipitation of Fragments of the GluR2 Subunit-In order to further characterize regions in AMPA receptor subunits mediating receptor assembly, we expressed different epitope-tagged fragments of the GluR2 subunit together with the GluR1 subunit and measured their degree of association. We constructed a series of truncation fragments of increasing length, starting from the C terminus (Fig. 6A). All deletion fragments could be detected on Western blots as bands migrating at positions expected from their calculated molecular weights (Fig. 6B). In some lanes, an additional band at much higher apparent molecular weight was present, which most likely represented nonreduced multimeric forms.
Deletion of the very C-terminal cytoplasmic region resulted in fragment C5, whose association with the GluR1 subunit was virtually indistinguishable from the association of the unaltered GluR2 subunit (Fig. 6C). Deletion of the extracellular region between transmembrane domains 2 and 3 as in construct C4 and deletion of the small loop thought to be part of the ion channel pore as in fragment C3 only slightly reduced the association with GluR1. These results demonstrate that in agreement with the data obtained with the chimeric receptors the C-terminal half of the receptor is not required for specific assembly. In contrast, deletion of the most N-terminal domain X as in fragment N1 reduced association with GluR to a value slightly above the level observed for kainate receptor subunits.
Removal of the first transmembrane domain resulted in soluble fragment C2, which did not specifically associate with GluR1. In fragment C1, which also co-precipitated at a low level, additionally the adjacent extracellular region S1 involved in the formation of the glutamate binding region was deleted.
These data implied an important role for the first putative transmembrane region in subunit assembly of AMPA receptors since its deletion strongly reduced the assembly of GluR2 fragments with GluR1. This domain might contain sequence stretches mediating specific association with other transmembrane domains. Alternatively, the presence of a transmembrane region might be necessary to keep the N-terminal fragments close to the membrane in an orientation favorable for assembly. In this case, the sequence of this domain would not be expected to be of importance for assembly. To distinguish between these possibilities, we replaced the transmembrane  Fig. 6) fused to the C terminus of GluR1 including TMD B; R1-PL6-R1, pore loop of GluR1 exchanged for the same region of GluR6. B, COS7 cells were transfected with cDNA coding for GluR1 or the chimeric constructs, lysed, and the cell extract subjected to SDS-PAGE. After transfer to a nitrocellulose membrane, the blot was probed with the anti-GluR1 pAb. C, COS7 cells were co-transfected with various combinations of GluR2 or GluR6 and the chimeras. The extent of association was determined as described previously. The result for the association of GluR6 with GluR1 was taken from Fig. 3. The bars represent the mean Ϯ S.E. of three different transfection experiments, except R6-PL6-R1ϩR6N, which was done in duplicate. GluR2 was co-precipitated efficiently with the chimera containing the N-terminal half of GluR1, whereas GluR6 was co-precipitated with the chimera containing the N-terminal half of GluR6. The pore loop region did not contribute to subfamily-specific association. domain 1 of fragment C3 by transmembrane domain 3 including the C-terminal intracellular domain and assessed the capacity of the resulting transmembrane protein (C2-TMD C) to assemble with GluR1 (Fig. 7). Fragment C2-TMD C co-immunoprecipitated with GluR1 with nearly the same efficiency as fragment C3 or the unaltered GluR2 subunit indicating that the second alternative is correct, i.e. the presence of a transmembrane region irrespective of its sequence appears to be required for efficient subunit assembly. This was corroborated by control experiments, which showed that transmembrane domain C expressed in the absence of fused sequence did not co-precipitate above the level of the control protein MuSK. A similar low level of association was found for a very hydrophobic fragment comprising the transmembrane domains A and B and the pore loop (Fig. 7).
These results prompted us to reassess the assembly of the most N-terminal domain C1 in the presence of a transmembrane domain. We fused either transmembrane domain A or C to this fragment and transfected the fusion fragments (C1-TMD A and C1-TMD C) together with GluR1 into COS7 cells. Both fragments had the expected size when analyzed by SDS-PAGE. They co-precipitated with the GluR1 subunit with only slightly reduced efficiency (about 80%) as compared with the GluR2 subunit (Fig. 7). These results indicate that the most important domain directing subunit assembly is localized in the more N-terminal part of the extracellular region, which is not part of the glutamate binding region. The still lower capability of fragments C1-TMD A and C in comparison with the complete subunit could suggest that other parts of the Nterminal domain also participate in subunit assembly. Alternatively, the slightly reduced assembly could be due to distorted conformations of these constructs caused by our splicing of normally separated domains.
Structural instability of proteins often causes their nonspecific aggregation and retention in the endoplasmic reticulum by the quality control system located there (52). To obtain an indication of how much our constructs were affected by these processes, we investigated which portion of each fragment could be solubilized in a Triton X-100-containing buffer and which constructs were expressed on the cell surface (Table I). Upon detergent extraction, more than 40% of the full-length subunits and the chimeric proteins were found in the soluble fraction. Many of our deletion fragments could be solubilized to a similar or even higher extent. However, fragments C1 and C2, which do not contain transmembrane domains, as well as these fragments artificially fused to transmembrane domains were found to a higher degree in the insoluble fraction. The apparent structural instability of these fragments could be caused by the presence of only one half of the glutamate binding region, which is normally formed by an association of domains S1 and S2 (31)(32)(33). Even these fragments, however, could be solubilized to at least one third of the level of the entire subunit. Only association of the solubilized fraction was measured in our co-immunoprecipitation assay.
While the nature of the transmembrane domain did not influence assembly with GluR1, it modified surface expression of fusion proteins. In nonpermeabilized COS7 cells, fragments C1-TMD C and C2-TMD C were accessible to antibodies added into the medium and therefore were expressed on the cell surface (Table I). In contrast, fragment C1-TMD A could not be detected in the plasma membrane and most likely was retained within the cell. Similarly, the deletion fragment C4 also was not detected on the surface, although its association with GluR1 was comparable to that of construct C3.

DISCUSSION
Transient transfection of subunits into eukaryotic cell lines of nonneuronal origin has been used in many studies to identify different regions of neurotransmitter receptors directing assembly into functional receptors (10,12,53,54). In this study, we applied this approach to identify regions that mediate specific association of AMPA receptor subunits.
Subfamily Specificity of AMPA Receptor Assembly-Upon heterologous expression in COS7 cells, antibodies against the GluR1 subunit co-precipitated each of the co-expressed AMPA receptor subunits GluR2, 3, or 4 with indistinguishable effi-FIG. 5. The first extracellular domain together with the first transmembrane domain of GluR1 mediates association. A, schematic representation of chimeras consisting of an N-terminal part of GluR1 (white) and a C-terminal part of GluR6 (black); transmembrane domains are shown as vertical bars: R1N-PL6-R6, R6 fused directly to the end of TMD A of GluR1 (with N-terminal hexamyc-tag, black ball); R1N-PL1-R6, N-terminal part of GluR1 containing the pore loop fused to the C terminus of GluR6 including TMD C. B, COS7 cells were transfected with cDNA coding for myc-tagged GluR1 or the myc-tagged chimeric constructs, lysed, and the cell extract subjected to SDS-PAGE. After transfer to a nitrocellulose membrane, the blot was probed with the anti-myc mAb. C, COS7 cells were co-transfected with various combinations of GluR1 and the chimeras or GluR2. The extent of association was determined as described above. The bars represent the mean Ϯ S.E. of three different transfection experiments. ciency. In contrast, a kainate receptor subunit was co-precipitated to a much lower extent, which, however, was still above the level of our control protein. Thus the subfamily specificity of GluR assembly was conserved in COS7 cells, an important feature known from functional analysis of heterologously expressed GluRs. Immunoprecipitation experiments from detergent extract of neurons show a similar segregation of subunits between subfamilies (14 -16, 55). Whether a small amount of receptor complexes containing both AMPA and kainate receptor subunits is present in neurons is still a matter of debate (15,55). The number of mixed complexes in neurons might be reduced not only by preferential subunit assembly, but also by differential targeting and stabilization of receptor complexes.
In order to quantitatively compare assembly of subunits by co-immunoprecipitation, we first assessed its dependence on the expression level of transfected receptors. Above a critical concentration, co-precipitation only slightly increased with higher subunit levels. By investigating subunit assembly under conditions when expression was above this concentration and carefully monitoring and correcting for individual expression levels of subunits, it was possible to determine subunit association with considerable accuracy. Despite the high sensitivity of our method, we were unable to detect differences in assembly between the subunit GluR1 and GluR2, 3, or 4.
Electrophysiological studies have shown that upon heterologous expression of combinations of two or more AMPA receptor subunits glutamate-gated channels are formed, which differ in their characteristics from homomeric channels (10,11). It has been difficult to directly compare the efficiency of subunit assembly in these functional studies because the expression levels of individual subunits are not known. In addition, the abundance of ion channels with low opening probabilities and ion conductivities tend to be underestimated in whole cell recordings. Similar difficulties were avoided by our experimental approach, in which we controlled expression levels and did not rely on functional properties for detection of receptor complexes. Our study indicated that AMPA receptor subunits do not carry intrinsic signals directing a preferential assembly of distinct subunit combinations. A similar random association of closely related subunits was shown for different ␣-subunits of the glycine receptor, which can freely substitute for each other (56).

N-terminal Domains of AMPA Receptor Subunits as Well as a Membrane
Anchor Are Important for Subfamily-specific Receptor Assembly-Using chimeric receptors and truncation fragments of GluR subunits, we identified domains of AMPA receptors directing specific assembly. Both approaches indicated that regions important for association are localized in the N-terminal half of these subunits. Chimeric receptors, in which the N-terminal part of an AMPA receptor was combined with a C-terminal half derived from a kainate receptor subunit, efficiently co-precipitated with the GluR1 subunit. Similarly, truncation fragments of an AMPA receptor subunit retaining the N-terminal X domain as well as at least one transmembrane region associated efficiently with GluR1.
The specificity of co-precipitation of the chimeric receptors is highlighted by the fact that the replacement of the N-terminal half of subunit GluR1 with the corresponding region of GluR6 had opposite effects on the assembly with AMPA and kainate receptors; this exchange increased the co-precipitation with GluR6, whereas it decreased the co-precipitation with GluR2 to the level normally observed with kainate receptor subunits. Such an association pattern could not easily be explained by abnormal folding of the chimeric protein. Rather, the reduced association with GluR2 and increased association with GluR6 had to be ascribed to the removal or addition of regions mediating subfamily-specific subunit binding. Furthermore, these observations imply that, for the assembly of kainate receptors, N-terminal domains of the corresponding subunits are instrumental as well, although this aspect was not further analyzed.
The specificity of association is more difficult to assess for deletion fragments. Co-immunoprecipitation of fragments with GluR1 might be lost either because specific assembly domains were deleted or because of misfolding of fragments induced by truncations. Differences in detergent solubility and surface expression between our fragments indeed suggest that the FIG. 7. Association of C-terminal GluR2 deletion fragments fused to transmembrane domains with GluR1. A, schematic representation of the C-terminal deletion fragments fused to TMD A or C (compare with legend to Fig. 6A). B, Western blot analysis of expression of the deletion fragments. Detergent extracts of COS7 cells transfected with cDNAs coding for C2-TMD C, C1-TMD C, or C1-TMD A were subjected to SDS-PAGE, transferred to nitrocellulose membranes, and probed with the anti-myc mAb. C, association of the deletion fragments with GluR1. COS7 cells were co-transfected with cDNAs of GluR1 and one of the deletion fragments. The extent of association was determined as described above. The bars represent the mean Ϯ S.E. of three different transfection experiments. chimeras, and deletion fragments of GluR2 COS7 cells were transfected with cDNA coding for the full-length GluR subunits, deletion fragments, or the chimeric proteins R1N-PL6-R6 and R1N-PL1-R6. For the determination of detergent solubility, cells were treated for 1 h with a Triton X-100-containing extraction buffer. Proteins from the supernatant and pellet were analyzed by SDS-PAGE and Western blotting. Indicated is the mean percentage of total immunoreactivity recovered in the supernatant from two experiments. For the assessment of surface localization, nonpermeabilized cells were stained with the anti-myc mAb. ND, not determined; surface localization of fragment C5 could not be analyzed since its epitope tag is located in an intracellular region. A fraction of the fragments C1 and C2, which do not contain a transmembrane region, most likely precipitated upon secretion or unspecifically attached to the cell surface and was therefore detectable by our staining procedure. Both fragments were also detected in the medium.

Construct
Detergentsoluble Surface localization structural integrity of some constructs was affected. Therefore, we based our conclusions mainly on fragments retaining the capability of association with GluR1 at a level comparable to that of the parental subunit. Structural alterations of fragments might give rise to unspecific hydrophobic interactions.
To exclude this possibility, we measured the co-precipitation of fragments in the absence of GluR1, which was very low for all fragments presented here. Upon heterologous expression in mammalian cells, about 30 -40% of homomeric or heteromeric AMPA receptors are found on the cell surface (57). Many of our chimeras and subunit fragments also were targeted to the plasma membrane, but cell surface expression clearly was no prerequisite for association: For example, fragments C3 and C4 co-immunoprecipitated with similar efficiencies with GluR1, although only C3 was expressed on the cell surface. Subunits of many receptors associate to larger complexes immediately after synthesis in the endoplasmic reticulum (58). Our observation that fragments that were retained within the cell did not lose their ability to interact with other subunits suggests that folding of the assembly region is to some extent independent from other regions.
Deletion of the C-terminal half of GluR2 up to the first transmembrane domain did not strongly affect its association with GluR1. These data were in agreement with our observations obtained from chimeric receptors. They indicate that this half of the protein does not play an important role in subunit assembly. Deletion of the first transmembrane region strongly reduced association, as has been shown for the assembly of nicotinic acetylcholine receptors and potassium channels (26,59,60). The association of two fusion proteins in which the first transmembrane domain was replaced by the third showed that the sequence of this domain was not important for assembly.
Membrane domains have been shown to direct the multimerization of a number of proteins, for example the T cell receptor, the major histocompatibility complex class II complex, and glycophorin (61)(62)(63). Apparently, these domains do not play a comparable role for GluR subunits. Our data suggest that a transmembrane domain is simply required for efficient assembly because it provides a membrane anchor leading to an enrichment of the attached domains in close vicinity to the membrane where assembly occurs. In addition, such a domain may support subunit association by orienting subunits in a way that favors their assembly into larger complexes.
Considerable evidence identifies the proximal N-terminal X domain that is homologous to the bacterial leucine/isoleucine/ valine-binding protein (38) as the major determinant of subfamily-specific subunit association. This region fused to a transmembrane domain directed an association with GluR1, almost reaching the level of the entire GluR2 subunit. Additional chimeric proteins and fragments should in future experiments allow a more detailed analysis of this assembly domain. The key role of N-terminal domains for subunit assembly found here mirrors a similar importance of these regions for assembly of other ion channels, for example nicotinic acetylcholine, GABA A , and glycine receptors, as well as potassium channels (26,56,59,60,64,65). The advantages of such an arrangement are presently unclear.
It remains to be seen whether assembly of GluRs in neurons follows a similar pattern as the one described here for COS7 cells or if subunit association is modified by neuron-specific factors. A number of proteins have recently been identified that can associate with the cytoplasmic domain of individual GluR subunits (34,35,66,67). Binding of these components could shift the assembly pattern in favor of specific subunit combinations while preventing others. In addition, neurons might be able to selectively direct distinct GluR subtypes to particular synapses, either by local synthesis (68) and assembly of AMPA receptor subunits in dendrites or by selective targeting of AMPA receptors to synaptic regions based on their subunit composition. One of these mechanisms or their combination could be responsible for an unequal distribution of certain types of AMPA receptors at different synapses (16,69,70).