Ligand-binding Domain Determines Endoplasmic Reticulum Exit of AMPA Receptors*

AMPA receptors (AMPARs) are tetrameric ion channels that mediate rapid glutamate signaling in neurons and many non-neuronal cell types. Endoplasmic reticulum (ER) quality control mechanisms permit only correctly folded functional receptors to be delivered to the cell surface. We analyzed the biosynthetic maturation and transport of all 12 GluA1–4 subunit splice variants as homomeric receptors and observed robust isoform-dependent differences in ER exit competence and surface expression. In contrast to inefficient ER exit of both GluA3 splice forms and the flop variants of GluA1 and GluA4, prominent plasma membrane expression was observed for the other AMPAR isoforms. Surprisingly, deletion of the entire N-terminal domain did not alter the transport phenotype, nor did the different cytosolic C-terminal tail splice variants. Detailed analysis of mutant receptors led to the identification of distinct residues in the ligand-binding domain as primary determinants for isoform-specific maturation. Considered together with the essential role of bound agonist, our findings reveal the ligand-binding domain as the critical quality control target in AMPAR biogenesis.

AMPA receptors (AMPARs) 2 are key mediators of fast glutamate signaling in the vertebrate central nervous system. Combinatorial assembly from homologous subunits GluA1-4, each existing as multiple splice and/or RNA-editing variants, can produce a variety of tetrameric AMPARs, distinguished by their kinetic properties, ion conductivities, localization, and pharmacology (reviewed in Ref. 1). The majority of native AMPARs are believed to be GluA2-containing heteromers (2,3), which are impermeable to Ca 2ϩ and mediate basal neurotransmission (4,5). Homomeric receptors, typically formed by non-GluA2 subunits, compose another important class of AMPARs. They are characterized by permeability to Ca 2ϩ and by voltage-dependent polyamine block and are associated especially with use-dependent adjustments of synaptic strength (6,7).
Synaptic function is largely determined by the type and number of AMPARs present. Therefore, the mechanisms that regulate the biosynthesis of the subunits and the subsequent assembly, maturation, and cellular transport (biogenesis) of receptors have received much attention. AMPAR subunits have a modular architecture consisting of an extracellular N-terminal domain (NTD), a bilobed ligand-binding domain (LBD), a membrane-embedded channel core domain, and a cytosolic C-terminal domain (CTD) (1). The recent crystal structure of a tetrameric GluA2 receptor-antagonist complex (8) indicates that the most extensive subunit contacts involve the LBD and the transmembrane channel core segments. The structure shows clear 2-fold symmetry, consistent with previous studies indicating that AMPARs are assembled as dimers of dimers (9,10). There are indications that assembly of AMPARs may not be a simple stochastic process determined by the subunit pool expressed in the cell, but exactly to what extent or how the process is guided by intrinsic thermodynamic or kinetic mechanisms toward favored compositions is still unclear (11,12).
Cell biological investigations indicate that stringent quality control mechanisms acting at the level of endoplasmic reticulum (ER) exit allow only correctly folded and assembled receptors to be transported to the cell surface (13)(14)(15)(16). Interestingly, recent studies suggest that these mechanisms may also distinguish between isoforms of AMPAR subunits, thereby contributing to the propensity of subunits to build homomeric versus heteromeric receptors (11,12,17,18). To gain insight into the subunit-dependent mechanisms of AMPAR biogenesis, we analyzed the inherent ability to form homomeric receptors in the complete set of 12 AMPAR splice variants. The results demonstrate robust subunit-and splice form-dependent differences in the competence for ER exit and surface expression and identify the LBD as the critical sensor for correct assembly.
Cell Culture and Transfection-HEK293 and COS-7 cells were cultured and transfected as described (21). For coexpression, cDNAs were transfected at a 1:1 ratio. For patch-clamp experiments, the cells were cotransfected with pEGFP-C1 for visualization of GFP fluorescence.
Immunofluorescence Microscopy-To analyze the surface expression levels of AMPAR subunits, transfected cells were fixed and immunostained as described (21). Images were obtained and quantified as described previously (17,20). To analyze the colocalization of receptor subunits with ER exit sites, transfected COS-7 cells were incubated at either 15°C for 2 h to prevent ER exit or at 20°C for 4 h to prevent Golgi exit (22). Cells were then fixed, permeabilized, and costained for the receptor subunit and Sec23 as described previously (17). Images were obtained with a Leica TCS SP5 confocal microscope using an HCX APO 63ϫ/1.30 corr (glycerol immersion) CS21 objective and Leica Application Suite Advanced Fluorescence software. Micrographs were processed using Image Pro-Plus 5.0 software.
Biochemical Analyses-Cell-surface biotinylation, endoglycosidase H treatment, and immunoblotting were carried out essentially as described previously (15,17). For immunoblotting, the ECL signal was detected and measured by either exposure to Hyperfilm TM (GE Healthcare) and analyzed using the Image ProPlus software as described (17) or by the Bio-Rad ChemiDoc XRS system and Quantity One software.
Electrophysiology-Whole-cell patch-clamp recording from transfected HEK293 cells was carried out as described previously (17).
Statistical Analysis-All data are presented as the mean Ϯ S.E. For electrophysiology data, n is number of cells recorded from; for all other data, n is the number of independent transfections. Electrophysiology data were examined by unpaired Student's t test. All other data were examined by one-way anal-ysis of variance, followed by the Bonferroni multiple comparison test. Significance was considered p Ͻ 0.05.
Molecular Modeling-The GluA3-flop LBD dimer was modeled using MODELLER Version 9.7 (23) with the crystal structure of the rat GluA2-flop LBD with bound L-glutamate (Protein Data Bank code 1ftj (24)) as a template. The full-length tetramer model of GRIA3-flop was modeled with MODELLER using the crystal structure of rat GluA2 (Protein Data Bank code 3kg2 (8)) as a template. Figures were prepared with BODIL (25), MolScript Version 2.1 (26), and Raster3D (27).

RESULTS
To analyze whether AMPAR subunits and splice variants have intrinsic differences in forming transport-competent homomeric receptors, the steady-state plasma membrane levels of all 12 AMPAR subunit variants in transfected COS-7 cells were determined. Examination by immunofluorescence microscopy at the time of peak expression, 40 h post-transfection, employing a FLAG epitope engineered at the mature N terminus allowed direct comparison between receptor isoforms. The immunostainings revealed remarkable subunit/ splice isoform-specific differences (Fig. 1, A and B). Consistent with previous findings (17), the flip isoforms of GluA1 and GluA4 were expressed at high levels on the cell surface, whereas the corresponding flop isoforms were largely retained inside the cells. Flip/flop splicing appeared to have little influence on the plasma membrane levels of GluA2 and GluA3 receptors, which, however, showed a striking subunit-dependent difference: all GluA2 isoforms had a high and GluA3 isoforms a very low surface expression. Of note, the short-and long-tailed variants of GluA2 and GluA4 showed similar levels of surface staining, indicating that the cytosolic tail does not play an important role in the process. The total expression levels of all the subunit variants were in the same range, except for the flip isoforms of GluA2, which had a slightly higher expression level (supplemental Fig. S1, A and B). Next, we used a cell-surface biotinylation assay to obtain a population level average of the surface-exposed fraction of each subunit. However, because of differing antibodies (targeted to cytosolic epitopes), this method does not allow comparison across all subunits. The results agreed well with those obtained by immunofluorescence microscopy and demonstrated that, also with GluA2 and GluA3, the flip isoforms were expressed at slightly higher levels on the cell surface than the respective flop variants (Fig. 1C). Based on the above analysis, differential maturation of homomeric AMPARs is determined by inherent subunit-and flip/flop isoform-specific mechanisms.
The meager plasma membrane levels of GluA3 indicate a limited capacity to form properly maturing homomeric receptors, possibly relating to a strong preference of GluA3 for heteromeric assembly. In support of this notion, coexpression of GluA3 with GluA2 led to a prominent (ϳ20-fold) increase in GluA3 surface expression as measured by biotinylation assay (see Fig. 3A). To gain insight into the mechanisms underlying the poor maturational efficiency of GluA3, we first wished to identify the responsible structural determinants. GluA3 shows an overall sequence identity of 66 -72% to the other subunits. Most of this variation arises from the NTD (52-61% identity), whereas the "channel core domain" (the rest of the polypeptide) of GluA3 shares 79 -86% sequence identity with the other subunits. Therefore, we initially investigated whether deletion of the NTD would alter the relative plasma membrane levels of GluA3 and other AMPAR variants. However, NTD-deleted receptors showed surface levels that were remarkably similar to those of the cognate full-length subunits (Fig. 2). This finding strongly indicates that the structural determinants responsible for the poor maturation of homomeric GluA3 are all located within the channel core domain. To identify these structures, we substituted parts of GluA2 with the corresponding GluA3derived sequences ( Fig. 3B and supplemental Fig. S2). A chimera that contained the CTD from GluA3 residue 765 (comprising the flip cassette, third transmembrane segment, and cytosolic tail) showed high surface expression similar to wildtype GluA2. Similar behavior was observed with a further chimera that contained the entire C-terminal half of GluA3, starting from residue 542 just before the first transmembrane segment. In conjunction with the NTD deletion data, these findings implicate the N-terminal segment of the LBD ("S1") preceding the first transmembrane domain in the surface expression difference between GluA3 and GluA2. A further chimera representing full-length GluA3-flip carrying only the S1 segment from GluA2 showed robust GluA2-like surface expression, thus confirming the critical role of the S1 segment (Fig. 3B).
Inspection of the S1 segments of AMPAR subunits revealed only two positions in which GluA3 showed consistent nonconservative differences from the other subunits (Fig. 4A). We reasoned that these differences, Tyr-454 in GluA3 versus Ala in GluA1/A2 and Ser in GluA4 and Arg-461 in GluA3 versus Gly  in all other subunits, may contribute to the poor plasma expression of GluA3. Accordingly, these residues were mutated either individually or in combination to the corresponding residue in GluA2, and the effects on surface expression were analyzed (Fig. 4, B-D, and supplemental Fig. S3). Similarly, the corresponding GluA2 residues, Ala-451 and Gly-458, were replaced with their GluA3 counterparts. In agreement with expectations, the single mutations led to partial reciprocal increases or decreases in the surface levels of GluA3 and GluA2, respectively. The relative effects of mutations at the R/G site were noticeably stronger than the mutations at the Y/A/S site for both GluA2 and GluA3. Remarkably, the double mutation GluA3(Y454A/R461G) rescued the surface expression of GluA3 to almost the same level as WT GluA2 as measured by the intensity of anti-FLAG immunofluorescence (GluA3-(Y454A/R461G), 4.23 Ϯ 0.15; WT GluA2, 6.25 Ϯ 0.27; arbitrary units Ϯ S.E.; n ϭ 3). Surface expression of the GluA2 double mutant (A451Y/G458R) decreased significantly but still remained higher than WT GluA3 (GluA2(A451Y/G458R), 2.51 Ϯ 0.28; WT GluA3, 0.71 Ϯ 0.11; n ϭ 3). The mutually reciprocal nature and the relative magnitude of the mutational effects indicate that Tyr-454 and Arg-461 together impart poor maturational efficiency to homomeric GluA3 receptors. In the crystal structure of the GluA3 LBD (28), these residues occupy a peripheral location, clearly separate from the agonist-binding cleft and the main dimer interface (Fig. 5A). To examine the possibility that mutations at these positions have functional effects on GluA3 channels, we analyzed glutamate responses from whole-cell patches of HEK293 cells expressing either wild-type GluA3 or GluA3(Y454A/R461G) (Fig. 4E). In agreement with the plasma membrane levels, the current responses of the double mutant were much higher than those of those of WT GluA3, but no significant differences in the apparent rates of desensitization ( des values: WT GluA3, 4.5 Ϯ 0.93 ms (n ϭ 8); GluA3(Y454A/R461G), 6.78 Ϯ 0.64 ms (n ϭ 7); mean Ϯ S.E.; p ϭ 0.080) or in the relative steady-state-to-peak ratios (WT GluA3, 0.028 Ϯ 0.020 (n ϭ 8); GluA3(Y454A/R461G), 0.035 Ϯ 0.019 (n ϭ 7); p ϭ 0.463) were observed (Fig. 4E). Moreover, wild-type and double mutant GluA3 produced similar glutamate dose-response curves, with EC 50 values of 1.55 and 1.23 mM, respectively, consistent with the lack of major functional effects of the GluA3(Y454A/ R461G) mutation (Fig. 4F).
The underlying assumption in the above analyses has been that the poor surface expression of GluA3 is due to limited forward trafficking from the ER. To confirm this, the maturation of GluA3 N-glycans from the ER-specific high-mannose state to Golgi-modified complex state was analyzed using an endoglycosidase H sensitivity assay. In agreement with deficient ER exit of GluA3, the electrophoretic mobility shifts observed after endoglycosidase H treatment indicated that almost all of WT GluA3 is in an immature high-mannose state. In contrast, the double mutant GluA3(Y454A/R461G) showed only partial sensitivity to endoglycosidase H, indicative of ER exit (Fig. 6A). Analysis of GluA2 revealed reciprocal patterns in the wild-type and double mutant receptors. A substantial fraction of WT GluA2 was insensitive to endoglycosidase H, whereas the double mutant GluA2(A451Y/G458R) was almost entirely in an immature state (Fig. 6B). To further confirm that the poor plasma membrane expression of GluA3 is due to impaired exit from the ER, we utilized a 15°C temperature block, which promotes accumulation of proteins in ER exit sites/pre-Golgi compartment (22), and costained wild-type and mutant receptors with pSec23, a marker of ER exit sites (29). Under confocal microscopy, both GluA2 and GluA3(Y454A/ R461G) showed colocalization with pSec23, whereas WT GluA3 clearly did not (Fig. 6C). Next, we used a 20°C temperature block to allow accumulation of proteins in the Golgi compartment (22). Under these conditions, GluA2 and GluA3(Y454A/R461G) showed rosette-like staining typical of the cis-Golgi in COS-7 cells (17), whereas WT GluA3 exhibited the reticular staining pattern of the ER (supplemental Fig. S4). The strong correlation observed between mature-type glycosylation and the surface expression levels of wild-type and mutant GluA2 and GluA3 receptors, together with the subcellular localization data ( Fig. 6 and supplemental Fig. S4), is consistent with the idea that the steady-state plasma membrane levels reflect the efficiency of exit of AMPARs from the ER.

DISCUSSION
Our analysis revealed robust subunit-and splice form-dependent differences in the maturation of homomeric AMPARs. These were caused by only a few sequence differences in the LBD. Considered together with the essential role of bound agonist in facilitating ER exit of AMPARs, our findings identify the LBD as a key sensor of the secretory competence of AMPARs (15,17). The molecular logic of such a role is obvious: the bipartite LBD encloses the ion channel domain and can thus serve as a folding reporter for the entire functional core of the receptor.
In earlier studies on AMPARs, correlations between desensitization kinetics and receptor maturation were found, leading to suggestions that the ability of a receptor to undergo transient gating motions typical of AMPAR activity cycle is monitored in the ER (15,18,30). Similar findings have been reported for kainate receptors (31). However, the near-total ER retention of GluA3, GluA1-flop, and GluA4-flop receptors is not readily explained by such activity differences. The double mutation that rescued the surface expression of GluA3 had little qualitative effect on glutamate responses. Moreover, if all subunits are considered, differential maturation of flip and flop forms shows little correlation with desensitization/resensitization kinetics. The flip/flop trafficking phenotype is equally strong in GluA1 and GluA4 and much more pronounced than in GluA2, yet these subunits exhibit quite different (and partly opposite) associations between the flip/flop isoform and desensitization kinetics (supplemental Table S1) (32)(33)(34). Therefore, we favor the notion that ER quality control monitors LBD-dependent structural features that are largely independent of activity. This interpretation is also more consistent with the independence of the transport from channel activity and with the findings that the secretion of separately expressed LBDs closely mimics the transport of mutant and wild-type subunits (15)(16)(17).
The precise nature of the activityindependent molecular features that act as cues to promote the exit from or retention in the ER is presently unclear. The leucine residue causing ER block of flop isoforms (17) is located close to the LBD dimer interface but is not directly involved in subunit interactions. The GluA3-specific arginine and tyrosine residues are even more clearly separated from the known subunit or interdomain contacts and face the periphery of the receptor (28). Such a localization is consistent with the idea that the subunit differences in maturation are largely determined at the postoligomerization stage but cannot exclude the possibility that oligomerization may be involved in an indirect manner. It is conceivable that the large polar side chains of GluA3 may influence interactions with ER luminal chaperones by either promoting retentive interactions or inhibiting interactions that improve maturation. Alternatively, the outwardly pointing side chains could affect the ability of the receptor to engage in homophilic associations that may occur between tetramers on the ER membrane. In a crystal structure of the GluA2 LBD in complex with glutamate (Protein Data Bank code 1ftj (28)) with three chains per asymmetric unit, the contact between chains A and B may correspond to such an intermolecular interface. Although it may represent a crystal artifact, it is intriguing that the LBD residues critical for receptor maturation are localized to this interface, suggesting that they may affect lateral packing of AMPAR tetramers (Fig. 5 and supplemental Fig. S5). In either scenario, i.e. chaperones or homophilic interactions, entering of the receptor into ER exit sites would be obstructed or diminished by the critical residues in GluA3 and in the flop isoforms. The rescued transport of GluA3 to the plasma membrane upon coexpression with GluA2 indicates that the inhibitory capacity of Tyr-454 and Arg-461 is strongly dependent on the stoichiometry and is masked in heteromeric assembly. Interestingly, comparison of AMPAR subunits across vertebrate lineages indicates that Tyr-454 and Arg-461 emerged in GluA3 at an early stage of mammalian evolution (supplemental Fig. S6). Both tyrosine and arginine are conserved in GluA3 orthologs of marsupials and placental mammals, whereas monotremes (egg-laying mammals likely to represent the earliest mammals) harbor tyrosine at position 454, whereas position 461 is occupied by glycine as in other subunits. In contrast, GluA3 orthologs of non-mammalian species have serine and glycine at these positions (like the other subunits). The conservation pattern is consistent with the functional importance of the residues in mammalian species and possibly related to increased diversity of AMPAR subunit composition in mammals. The negligible role of the NTD as a determinant of subunit differences in receptor maturation is worthy of mention. This domain encompasses almost half of the receptor polypeptide and shows higher sequence variation than the LBD and the channel core. Previously, the NTD has been shown to participate in the assembly by preventing subunit associations across ionotropic glutamate receptor subclasses, e.g. the coassembly of AMPAR and the kainate receptor (35,36). It remains possible that the NTD may play a modulatory role in the heteromeric assembly of AMPARs, e.g. by favoring certain subunit pairings over others, but a primary role in the assembly is difficult to reconcile with the present findings. A notable finding of the study relates to the GluA2 subunit, which produced homomeric receptors quite efficiently, a result that is somewhat counterintuitive regarding the reports of an ER transport block of GluA2 due to the edited arginine in the channel pore (37). However, comparable mutations in other subunits exert little or no effect on the transport efficiency (17). Although, in neurons, most GluA2 subunits form heteromeric receptors with other subunits, recent evidence actually suggests that, in the absence of other subunits, homomeric GluA2 receptors may also contribute to synaptic responses (3).
This study was performed in non-neuronal cells to distinguish the intrinsic differences between subunits. Assembly and early trafficking of AMPARs take place in a highly similar manner in neurons and cell lines, although the presence of transmembrane AMPAR regulatory proteins and possibly other accessory factors may modulate the processes in neurons. Structure-function studies indicate that transmembrane AMPAR regulatory proteins interact with AMPARs via extracellular, intramembrane, and cytosolic C-terminal interactions to regulate receptor transport to the cell surface, targeting to synapses, and modulation of channel properties and pharmacology (reviewed in Refs. 38 and 39). Previously, we reported that ER exit of GluA4-flop (but not GluA4-flip) is strongly improved by coexpression with stargazin (17), suggesting that the ER chaperone function of stargazin is dependent on its interactions with the LBD. On the basis of this and the present findings, we propose that the LBD is the major target of all ER quality control mechanisms that regulate AMPAR maturation and transport to the cell surface, whether caused by intrinsic (i.e. subunit composition) or extrinsic (transmembrane AMPAR regulatory proteins and cornichons) factors.
In conclusion, this study demonstrates that ER quality control mechanisms use structural features in the LBD as a readout of the transport competence of AMPARs. These mechanisms are likely to contribute to the determination of AMPAR subunit stoichiometry, a process of central importance for synaptic function.