Aggregation Limits Surface Expression of Homomeric GluA3 Receptors*

AMPA receptors are glutamate-gated cation channels assembled from GluA1–4 subunits and have properties that are strongly dependent on the subunit composition. The subunits have different propensities to form homomeric or various heteromeric receptors expressed on cell surface, but the underlying mechanisms are still poorly understood. Here, we examined the biochemical basis for the poor ability of GluA3 subunits to form homomeric receptors, linked previously to two amino acid residues, Tyr-454 and Arg-461, in its ligand binding domain (LBD). Surface expression of GluA3 was improved by co-assembly with GluA2 but not with stargazin, a trafficking chaperone and modulator of AMPA receptors. The secretion efficiency of GluA2 and GluA3 LBDs paralleled the transport difference between the respective full-length receptors and was similarly dependent on Tyr-454/Arg-461 but not on LBD stability. In comparison to GluA2, GluA3 homomeric receptors showed a strong and Tyr-454/Arg-461-dependent tendency to aggregate both in the macroscopic scale measured as lower solubility in nonionic detergent and in the microscopic scale evident as the preponderance of hydrodynamically large structures in density gradient centrifugation and native gel electrophoresis. We conclude that the impaired surface expression of homomeric GluA3 receptors is caused by nonproductive assembly and aggregation to which LBD residues Tyr-454 and Arg-461 strongly contribute. This aggregation inhibits the entry of newly synthesized GluA3 receptors to the secretory pathway.

␣-Amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors are tetrameric ligand-gated ion channels that mediate fast excitatory neurotransmission in vertebrate brain (1,2). The functional properties of AMPA receptors, including channel kinetics, ion permeability, ligand pharmacology, and regulation, are determined by the subunit composition. AMPA receptors are built from four subunit types (GluA1-4), each expressed as multiple alternatively spliced and/or RNA-edited variants. Studies with native and recombinant receptors indicate that receptor assembly is not random but strongly favors certain subunit combinations (3,4). The majority of native AMPA receptors are heteromers of edited GluA2 subunits with GluA1 or GluA3 subunits, forming channels that are impermeable to Ca 2ϩ and showing a linear current-voltage (I-V) relation (3,5,6). Minor native populations of homomeric AMPA receptors are formed by GluA1 or GluA4 subunits to produce inwardly rectifying and Ca 2ϩ -permeable channels (7,8).
At present, the molecular logic underlying the formation and cellular processing of specific subunit assemblies in AMPA receptors is still poorly understood. The oligomerization is initiated by formation of dimers between N-terminal domains (NTD) 5 of nascent receptor polypeptides in the endoplasmic reticulum (ER) followed by interactions involving the transmembrane segments and the ligand binding domains (LBD) (9 -11). The relative strength of NTD contacts can partly explain the strong preference of certain subunits like GluA3 for heteromeric assembly, but also the LBD harbors important structural determinants that control the biogenesis of AMPA receptors in a subunit or subunit variant-dependent manner (12)(13)(14). In addition, auxiliary proteins, including transmembrane AMPA receptor-associated proteins (TARP) and cornichon homologs, associate with AMPA receptors already in the ER and, in principle, may influence the assembly process (9,(15)(16)(17).
Recently, we demonstrated that two LBD residues, Tyr-454 and Arg-461, are responsible for the poor delivery of homomeric GluA3 receptors to cell surface both in the presence and absence of NTD (12). These residues are not directly engaged in glutamate binding nor in subunit interactions, and the reason why they inhibit the formation or transport of homomeric GluA3 to the cell surface is unclear. Here, we performed a comparative biochemical analysis of wild-type and mutated GluA2 and GluA3 receptors expressed in HEK293 cells. We found that GluA3 receptors exhibit aberrant, Tyr-454/Arg-461-dependent tendency to disturbed oligomerization and aggregation. Our findings suggest that subunit-dependent balance between transport-competent oligomers and higher-order abnormal assemblies contributes to the preferences of AMPA receptor subunits for homomeric or heteromeric expression.

Experimental Procedures
DNA Constructs-All cDNAs encoding full-length AMPA receptor subunits or mutants thereof represented the rat AMPA receptor subunits; the flip/flop isoform is indicated in "Results" and figure legends; GluA2 is the edited (Q607R) form. All were expressed with an N-terminal signal sequence and tag from pcDNA3.1 (Stratagene) (12,13,18). Plasmids encoding human stargazin and rat GluN1 (NR1) were kind gifts from John L. Black III (Mayo Medical School, Rochester, MN) and Dr. Rolf Sprengel (Max Planck Institute for Medical Research, Heidelberg, Germany), respectively. Mammalian expression constructs of GluA2 and GluA3 LBDs were built on pEGF-C1 (Clontech) by replacing the GFP encoding sequence by a cDNA expression cassette encoding an N-terminal signal peptide, FLAG epitope, requisite GluA sequences, and C-terminal Myc tag as described before for GluA4 (13). Bacterial expression constructs were created by PCR-based cloning and inserted into T7 promoter containing vector and a C-terminal His 6 tag. The constructs were designed to correspond to AMPA receptor ligand binding domains used for crystallization studies (19,20). All constructs were verified by restriction mapping and sequencing of PCR-amplified regions.
Cell Culture and Transfection-HEK293 cells were cultured in DMEM supplemented with 10% (v/v) fetal calf serum and penicillin-streptomycin and transfected with appropriate plasmids by using calcium phosphate coprecipitation as previously described (21). For patch clamp experiments, pEGFP1-C1 (Clontech) was co-transfected for the visualization of transfected cells. For some experiments cells were incubated with tunicamycin (2 g/ml; Sigma T-7765) for 24 h before harvesting.
Electrophysiology-Whole-cell patch clamp recordings and analysis of glutamate-triggered currents in transfected cells were carried out as previously described (22). In brief, recordings from AMPA receptors expressing HEK293 cells were done at a holding potential of Ϫ60 mV. L-Glutamate was diluted in external recording solution containing 150 mM NaCl, 2.5 mM KCl, 2.5 mM CaCl 2 , 1 mM MgCl 2 , 10 mM glucose, and 10 mM HEPES (pH 7.4; 320 mosM) and applied to cells using a piezodriven applicator (Siskiyou Piezo Switcher; Siskiyou Corp.). Electrodes had a resistance of 2-4 megaohms when filled with internal solution containing 140 mM CsCl, 2 mM MgCl 2 , 10 mM EGTA, and 10 mM HEPES (pH 7.2, 315 mosM). Cells were lifted from the bottom of the culture dish by the patch clamp pipette to facilitate the solution exchange. Each drug application was done twice, and traces were averaged for analysis.
Fluorescence Microscopy-Representative micrographs of GluA3-GFP-expressing HEK293 cells were acquired via an EVOS FL digital inverted microscope (AMG, Life Technologies) with a GFP LED light cube (570-nm excitation, 525-nm emission) and 20ϫ objective.
Biochemical Analyses-Transfected cells were extracted in TNE buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 5 mM EDTA) containing 1% (w/v) Triton X-100 followed by ultracentrifugation at 100,000 ϫ g for 1 h. Detergent-soluble and nonsoluble receptor fractions were determined from the immunoreactivity present in the supernatant and in the pellet, respectively. They were quantitated as described below. Protein concentrations were determined by bicinchonic acid assay (Uptima, Interchim, France) using bovine serum albumin as standard. Immunoprecipitation, cell surface biotinylation, and analysis of the secretion of ligand binding domain were performed as described previously (13,21,23).
Cell surface biotinylation was done as described (13). Briefly, transfected HEK293 cells were rinsed with PBS containing 1 mM CaCl 2 , 0.5 mM KCl, 2.5 mM MgCl 2 , and incubated with EZ-Link sulfo-NHS-SS-Biotin (Pierce) at 0.5 mg/ml in the above buffer for 30 min at room temperature. Non-reacted reagent was removed by washing cells with the above buffer. Triton X-100 extracts were made as described above and subject to streptavidin-Sepharose precipitation (GE Healthcare). Immunoprecipitation was done as described (21). Briefly, Triton X-100 extracts of transfected cells were incubated with prewashed Gamma-bind Sepharose (GE Healthcare) and appropriate antibody (at concentration detailed above) and mixed for 2 h at 4°C.
Secretion of GluA2 and GluA3 LBD constructs was analyzed as described (23). Briefly, growth medium was collected from transfected HEK293 cells 40 h after transfection, and 1 mM PMSF was added. Cell debris was pelleted by centrifugation (1500 ϫ g, 5 min, 4°C). The total cell extracts were prepared as above.
Density Gradient Centrifugation-Gradients were prepared by layering 10% (w/v) sucrose in TNE buffer containing 0.5% Triton X-100 with Ͼ40% (w/v) sucrose solution (1:1 ratio) and incubating horizontally for 75 min at room temperature. Thus prepared gradients were stored vertically at 4°C until use. Transfected HEK293 cell extracts (300 l) were layered on top of the gradients that were run in pairs in an ultracentifuge with a Sw-55Ti rotor (Beckman-Coulter) at 175,000 ϫ g for 16 h at ϩ4°C. Fractions (200 l) were collected from the bottom of centrifuge tube by using a peristaltic pump (GE Healthcare P-1).
Gel Electrophoresis Analysis-Samples were run on either Lonza 4 -12% or Bio-Rad 4 -15% TGX Criterion gradient gels for both denaturing (SDS) or blue native (BN)-polyacrylamide gel electrophoresis (PAGE). For SDS-PAGE samples were prepared in Laemmli sample buffer (with final concentration of 60 mM Tris-Cl, pH 6.8, 2% SDS, 10% glycerol, 5% ␤-mercaptoethanol, 12.5 mM EDTA 0.01% bromphenol blue). BN-PAGE was run according to the protocol of Gill et al. (24), and samples were prepared in native loading buffer (with final concentration of 60 mM Tris-Cl, pH 6.8, 10% glycerol, 0.01% bromphenol blue); molecular standards were indicated by Native Mark (Novex, Life Technologies). After gel separation, proteins were transferred onto Hybond membrane (GE Healthcare) and immunoblotted. The samples used for BN-PAGE were obtained from the sucrose density gradients with the exception of the analysis of crude cell extracts. In the latter case Triton X-100 extracts prepared from transfected cells were cleared by centrifugation at 12,000 ϫ g for 20 min, mixed with native loading buffer, and used for electrophoresis.
Immunoblotting and Analysis-For immunoblotting, Clarity TM Western ECL substrate (Bio-Rad, 170-5061) was used. ECL signal was detected and measured either by the ChemiDoc XRS system and Quantity One software (Bio-Rad) or by BioSpectrum 810 and VisionWorksLS software (Ultra-Violet Products Ltd). The pixel density of the immunoreactive bands was obtained from the longest possible exposure before saturation of signal. If required the relative expression levels were normalized to a control, as defined in the respective figure legend. For determination of the percentage of oligomeric state after BN-PAGE, the signal was summed from boxes corresponding to aggregate, tetramer, dimer, and monomer for all fractions run (see Fig. 7A, middle panel). An equivalent box for background only was also measured; if this corresponded to 5% or more of the total signal the experiment was discarded.
Bacterial Protein Expression and Purification-His 6 -tagged proteins were expressed and purified according to standard procedures as described previously (25) with the following modifications; the plasmids containing wild-type or point-mutated GluA2i S1S2 and GluA3i S1S2 segments were transformed into Escherichia coli Origami (DE3)pLysS for protein expression. The first round of purification from E. coli lysates was made via nickel-nitrilotriacetic acid-agarose (Qiagen) against the His 6 tag. The proteins eluted from nickel-nitrilotriacetic acid agarose were dialyzed against 20 mM Tris-HCl, pH 8.0, buffer and then loaded on to a DEAE column (DEAE-Sepharose CL-6B, GE Healthcare). After washing, proteins were eluted with 20 mM Tris-HCl, pH 8.0, buffer containing 0.1 M NaCl. All purified proteins were concentrated by centrifugal devices (microsep 30K, Pall Corp.) to ϳ2 mg/ml and stored in elution buffer solution at 4°C.
Chemical Denaturation-Aliquots of bacterially expressed and purified GluA2 or GluA3 LBD proteins were incubated with increasing concentrations of urea (0 -8 M) in the presence or absence of 10 mM L-glutamate. Intrinsic fluorescence at the 300 -450-nm region produced by excitation at 295 Ϯ 5 nm was measured in quartz cuvettes (Hellma) using a FluoroMax-4 spectrofluorometer (Horiba Scientific) and analyzed by Origin 7.5 software. For calculation of the urea denaturation curves, peak emission intensities at 325 nm and 327 nm were used for GluA2 and GluA3 LBDs, respectively. Fluorescence intensity was plotted against urea concentration and individually best-fit with the sigmoidal dose-response (variable slope and no constraints) model in GraphPad Prism v4 (GraphPad Software Ltd.).
Circular Dichroism (CD) Spectroscopy-For CD spectroscopy protein samples were dialyzed overnight against sodium phosphate buffer (50 mM sodium phosphate, pH 7.4) followed by centrifugation (5 min, 13,000 rpm, 4°C). Thermal stability of the GluA2i, GluA3i, and GluA3i (Y454A/R461G) LBDs was monitored by CD spectroscopy using a Jasco J-720 spectropolarimeter. Thermal denaturation scans were recorded using a 0.1-cm cuvette, wavelength 195 nm, a temperature range from 20°C to 70°C with temperature steps of 0.1°C. The protein concentration was 0.2 mg/ml. CD spectra were measured in the presence and absence of L-glutamate at 1 mM for GluA3i and GluA3i (Y454A/R461G) and 0.5 mM for GluA2i. Data were compiled in Excel (Microsoft Office 2007). The spectra were plotted, and T m was calculated with GraphPad Prism v4 (GraphPad Software Ltd).
Statistical Analysis-Data are presented as the mean Ϯ S.E. unless otherwise stated and described, where n is the number of independent experiments. Analysis was by Student's t test or by one or two-way ANOVA followed by Bonferroni's multiple comparison test or by Dunnett's multiple comparison test to control. Statistical significance of p values was indicated by asterisks as follows: *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001. All statistical analyses were done by using GraphPad Prism. Analysis of desensitization kinetics of patch clamp recordings was done by Clampfit 10.2 software (Molecular Devices, Sunnyvale, CA) using single exponential fit paradigm.

Surface Expression of GluA3 Is Rescued by Co-assembly with GluA2 but Not by Association with Stargazin-Previously, we
showed that co-expression with GluA2-flip (GluA2i) promotes the surface expression of the otherwise mostly intracellularly retained GluA3i in transfected HEK293 cells (12). We now analyzed the ability of GluA2 isoforms flip and flop (GluA2o) and of the GluA2i point mutant GluA2i A451Y/G458R (GluA2i AG/YR) to promote GluA3i surface expression as determined by biotinylation assay. Biotinylation with the cell-impermeant biotin derivative interferes with immunoreactivity of extracellular epitopes, so a GFP tag was introduced to the C terminus of the receptor polypeptide. This C-terminal fusion did not alter the basic trafficking of GluA2 and GluA3, as GluA2i-GFP showed significant surface expression, whereas only a minor fraction of Glu3i-GFP was present on the plasma membrane (Fig. 1A). Thus, GluA3i-GFP was expressed in HEK293 cells together with the FLAG-tagged test constructs GluA3i (control), GluA2i, GluA2o, or GluA2i (AG/YR). After surface biotinylation and streptavidin pulldown, GluA3i-GFP was visualized by immunoblotting (Fig. 1, B and C). Consistent with earlier findings, heteromeric assembly improved the surface expression of the GluA3i; in the presence of GluA2i, plasma membrane levels of GluA3i-GFP increased 6-fold. Co-expression with the flop isoform of GluA2 and with the AG/YR double mutants that carry the two residues responsible for the intracellular retention of GluA3 did not the increase surface expression of GluA3i-GFP in a statistically significant manner (Fig.  1C). To determine whether GluA3 displays a significant preference for heteromeric over homomeric assembly, N-terminally tagged GluA2i and GluA3i were expressed with a differently tagged version of the same or the other subunit and subjected to immunoprecipitation. FLAG-tagged GluA3i was co-immunoprecipitated with myc-tagged GluA3i or GluA2i and vice versa to a similar extent. Thus, suggesting that under the experimental conditions of transient expression in HEK293 cells GluA3 had no significant preference for heteromeric assembly (Fig.  1D).
In addition to co-assembly, association with auxiliary proteins is known to promote surface expression of AMPA receptors. Therefore, we studied whether co-expression with stargazin (a prototype member of transmembrane AMPA receptor regulatory proteins) would promote the trafficking of GluA3 to the plasma membrane. Previously, we have shown that stargazin rescues the otherwise poor surface expression of the flop variants of GluA1 and GluA4 homomers (13). However, despite strong expression and surface location of stargazin, no significant improvement was seen for GluA3 (Fig. 2, A and B). Due to this unexpected result, it was important to verify the association of stargazin with GluA3. Co-immunoprecipitation experiments did not show statistically a significant difference in star-FIGURE 1. GluA2 subunit rescues surface expression of GluA3. A, AMPA subunits GluA2-flip (GluA2i) and GluA3-flip (GluA3i) were either N-terminally FLAG-tagged or C-terminally GFP-tagged and expressed in HEK293 cells, as indicated above the panels. Cells were surface-biotinylated, and cell extracts were subjected to streptavidin pulldown (surface). Immunoblots were probed with anti-GFP. Endogenous GAPDH protein was used as a loading control. The presence of C-terminal GFP did not alter the surface trafficking phenotypes of GluA2 and GluA3. B, GluA3i-GFP was co-expressed with GluA3i, GluA2i, GluA2-flop (GluA2o), or with the A451Y/G458R double mutant of GluA2-flip or -flop as indicated and subjected to a biotinylation assay as above. Antibodies used for blotting are indicated on the right. Note that anti-GluA2/3 recognizes both GluA2 and GluA3 C termini but has weaker binding to GluA3. C, quantification of GluA3-GFP cell surface expression after co-expression with the indicated proteins. Anti-GFP IgG immunoblot signals were normalized to the co-expressed GluA3i. Bars show the mean Ϯ S.E., n ϭ 7-14. Statistical analysis of the co-expressed GluA3i-GFP surface expression was determined by one-way ANOVA, p ϭ 0.006; Bonferroni's multiple comparison post-test gave the following p values: F-A3i versus F-A2i, **, p Ͻ 0.01; F-A2i versus F-A2 (AG/YR), *, p Ͻ 0.05; all others, p Ͼ 0.05. Dunnett's multiple comparison post test gave the following p values: F-A3i versus F-A2i, p Ͻ 0.01, and F-A3i versus F-A2o, p Ͻ 0.05, all others p Ͼ 0.05. D, AMPA receptor subunits GluA2i and GluA3i were N-terminally FLAG (F) or Myc (M) -tagged and co-expressed in HEK293 cells, as indicated above the panels. Receptors were immunoprecipitated (IP) with either anti-FLAG or anti-myc antibodies and immunoblotted as indicated below. The presence or absence of stargazin is shown in the middle panel, blotted with anti-stargazin serum. The lower panel shows endogenous GAPDH protein as a loading control. B, quantification of GluA3i Ϯ stargazin surface expression immunoblots; GluA3i ϩ stargazin signal was normalized to GluA3 expression alone. Bars show the mean Ϯ S.E., n ϭ 6. Statistical analysis by paired, one-tailed t test gave p ϭ 0.0762, indicating no significant statistical difference. C, immunoblot showing the association between GluA2i or GluA3i and stargazin. Expressed AMPA receptors were immunoprecipitated by anti-FLAG antibody, and the immunoprecipitates were probed for the associated stargazin (lower panel). The anti-stargazin signal ratio GluA3iϩStg/GluA2iϩStg was 1.23 Ϯ 0.239 (S.E.), n ϭ 7. One sample t test with a control value set to 1 gave p ϭ 0.38, indicating no statistically significant difference. D, summed electrophysiology data from whole-cell patch clamp recordings of HEK293 cells expressing GluA3i either alone (n ϭ 8) or with stargazin (n ϭ 9) and following the application of 10 mM L-glutamate. The bars show the mean Ϯ S.E. for whole-cell peak currents. Statistical analysis by unpaired one-tailed t test gave p ϭ 0.0764, indicating no significant difference. E, comparison of tau of desensitization values from whole-cell patch clamp recordings of HEK293 expressing either GluA3 alone (n ϭ 5) or with stargazin (n ϭ 8). Bars indicate the mean Ϯ S.E. Statistical analysis by unpaired one-tailed t test gave ***, p ϭ 0.0004, indicating stargazin has a significant effect on the desensitization kinetics of GluA3 channels. gazin association between GluA3 and GluA2 (Fig. 2C). Wholecell patch clamp electrophysiology on transfected HEK293 cells did not show a statistically significant increase in the maximal whole-cell currents in the presence of stargazin (Fig. 2D). However, when stargazin was co-expressed, the desensitization of glutamate responses was significantly slower; time constants (des values) for the onset of desensitization were 10.48 ms Ϯ 1.11 for GluA3ϩStg (Stg, stargazin) and 4.50 ms Ϯ 0.38 GluA3 for GluA3 alone, indicating functional association of stargazin with GluA3 (Fig. 2E).
Differences in LBD Stability Did Not Explain the Poor Trafficking of GluA3-Ligand binding domains of GluA2 and GluA3, expressed as soluble secretable fusion proteins, reproduce the trafficking difference between the subunits (12). It is not clear, however, whether the underlying mechanisms are shared. To resolve this question we analyzed whether the two amino acid differences in LBD, which explain the differing surface expression levels of intact GluA2 and GluA3, can similarly affect the secretion of LBDs. Epitope-tagged LBD constructs were expressed in HEK293 cells, and LBD protein levels were determined in the cells and culture medium by Western blotting (Fig. 3, A and B). In agreement with earlier results, wildtype GluA2 LBD was well secreted from the cells, whereas GluA3 LBD was not. Importantly, the double mutants GluA3i Y454A/R461G GluA3 (YR/AG) and GluA2 (AG/YR) reversed the behavior and showed secretion patterns typical for the other subunit; GluA3 (YR/AG) LBD was secreted efficiently to the culture medium, whereas GluA2 (AG/YR) LBD was largely retained in the cells (Fig. 3, B and C). All LBD constructs were equally well expressed within the cells (Fig. 3B, left panels).
The above results suggest that similar mechanisms contribute to the transport defects of full-length GluA3 and GluA3 LBD. One obvious possibility is that GluA3 LBD is less stable and because of local unfolding GluA3 or GluA3 LBD would not be able to pass the quality control of the secretory pathway. To produce sufficient protein for the measurement of the conformational stability, we engineered the LBD constructs for expression in E. coli. The LBDs contain one essential disulfide bond so E. coli strain Origami, which harbors oxidizing cytosol, was used as the expression host. Initial attempts to express LBDs, which had an overall design as shown in Fig. 3A but lacked the signal peptide, failed to produce any significant levels of protein. Therefore, the LBDs were redesigned to correspond to the GluA2 S1S2J construct (19)which has been used in a similar expression system to produce GluA2 and GluA3 LBDs for crystallization (20) (Fig. 4A). This approach led to successful expression of both LBDs as soluble proteins (although GluA3 yields were consistently lower), and the proteins were purified by using metal-chelation affinity chromatography. Stability was assessed by measuring the intrinsic tryptophan fluorescence in increasing concentrations of urea in the presence and absence of L-glutamate (10 mM). Typical plots of fluorescence intensity against urea concentration are shown in Fig. 4B; they were sigmoidal and showed a marked shift to the right (to higher urea concentrations) in the presence of glutamate, indicative of cooperative unfolding and stabilization by glutamate (Fig. 4B); the EC 50 values for urea-induced unfolding are given (Fig. 4C and Table 1). Interestingly, GluA3 appeared more stable than GluA2, effectively ruling out poor folding of GluA3 LBD as an explanatory factor of the transport defect. To verify the results from chemically induced unfolding, we analyzed the thermostability of GluA2 and GluA3 LBDs by CD spectroscopy. Consistent with the results from chemically induced denaturation, the GluA3 LBD was more stable for thermally triggered unfolding (Fig. 4D). The presence of L-glutamate provided substantial FIGURE 3. Ligand binding domain secretion mimics the cognate fulllength receptor trafficking phenotype. A, schematic structure of the mature GluA3i LBD polypeptide expressed in mammalian cells. The included S1 and S2 residues are indicated, as are the point mutation sites; the S1 and S2 domains are joined by a flexible 13-residue linker with the sequence STE-GEVNAEEEGF (13,28). For efficient secretion, the expression plasmid encoded a viral signal peptide (MTILCWLALLSTLTAVNA) placed N-terminally from the FLAG epitope (28). The numbering is from rat GluA3, Uniprot sequence P19492 (12). All LBD constructs for mammalian expression encode an N-terminal FLAG tag and C-terminal Myc tag. B, representative immunoblots showing expression of GluA2i and GluA3i LBD proteins from transfected HEK293 cells, both "wild type" and point-mutated versions (upper panels, anti-Myc IgG). The left panels show protein expression in cell extracts; right panels show expression of constructs in cell media after secretion. Endogenous GAPDH protein is shown as a loading control. C, quantitation of LBD constructs expression from cell extract (solid shading) and media (hatched shading). Immunoblot signals were normalized to GluA2 LBD expression. Bars show the mean Ϯ S.E., n ϭ 4 -11. For cell extract samples one-way ANOVA gave p ϭ 0.0006, and for media samples one-way ANOVA p Ͻ 0.0001. Significance was determined by Dunnett's multiple comparison to GluA2i LBD as control is shown; ** indicates p Ͻ 0.01. stabilization for both LBDs, the T m values increased by Ͼ10°C for both GluA2 and GluA3 LBDs (Fig. 4D and Table 1). Interestingly, the thermostability of the LBD mutant GluA3 (YR/ AG), which reverses the trafficking phenotype to GluA2-like, showed similar stability as the GluA3 wild-type both in the presence and absence of L-glutamate (Fig. 4D and Table 1). The results indicate that the differences between GluA2 and GluA3 in LBD stability and in the transport can be separated. Thus, stability and transport competency are likely independent from each other.
Homomeric GluA3 Receptors Have a Tendency for Abnormal Oligomerization and Aggregation-To identify possible gross differences in the biochemical properties of GluA2 and GluA3, we determined the detergent solubility of homomeric GluA2 and GluA3 receptors expressed in HEK293 cells. For this pur-pose, Triton X-100-soluble and -insoluble fractions were separated by ultracentrifugation and analyzed by SDS-PAGE and immunoblotting (Fig. 5A). There was a substantial difference between the subunits; 70% of GluA2 protein (determined by using anti-FLAG antibody) was solubilized by Triton X-100, whereas the fraction of detergent-soluble GluA3 was only half that (Fig. 5, A and B). If the poor solubility of GluA3i is due to aggregation, it is possible that it could increase aggregation of other proteins coexpressed with it. To test this we studied the detergent solubility of the GluN1 subunit of the NMDA receptor when coexpressed with either GluA3i or with GFP. Expression with GluA3i led to a significant increase in the amount of insoluble GluN1 (Fig. 5, C and D), supporting aggregation as an explanation for the poor detergent solubility of GluA3i homomers. To find out whether the detergent solubility is related to the trafficking phenotypes, the detergent-soluble fractions of wild type and the reciprocal double-mutated GluA2 and GluA3 flip and flop receptors were analyzed in parallel and normalized to GAPDH (serving as a loading standard for cellular protein). For both flip (Fig. 5, E and F) and flop (Fig. 5, G and H) isoforms, the point mutations completely reversed the solubility properties of the parental receptors; GluA3 YR/AG behaved like wild-type GluA2, and GluA2 AG/YR showed GluA3-like low detergent solubility. Next, we examined the effects of expression of GluA2i and GluA3i on GRP78, an indicator of stress in the endoplasmic reticulum. In comparison to mock-transfected cells, the amount of GRP78 protein was slightly elevated in cells expressing GluA2i and GluA3i, but there was no difference between the subunits (Fig. 5I). In contrast, the GRP78 level was strongly increased in tunicamycintreated cells (2 g/ml; 24 h), used as a positive control. Anti-FLAG immunoprecipitations showed that GRP78 is associated with both subunits to the same extent (Fig. 5J).
The above-described results suggest that although the crucial difference between GluA2 and GluA3 homomeric receptors is strictly related to specific LBD residues, the mechanistic explanation for the different transport competences of GluA2 and GluA3 receptors does not lie in the primary folding of the LBD but more likely in the solution behavior of the cognate receptors. This prompted us to analyze the oligomeric state of GluA2, GluA3, and GluA3 (YR/AG) receptors by using fractionation of detergent-soluble receptors in sucrose density gradient centrifugation. Anti-FLAG immunoblots of sucrose gradient fractions showed remarkable differences between the receptor variants (Fig. 6A). The majority of GluA3 the 110-kDa signal was associated with the higher density sucrose fractions with a broad peak (P1) at fractions 5-7 with substantial immunoreactivity also present at the bottom of the gradient, likely representing aggregation (Fig. 6B). A minor population of GluA3 sedimented at a lower speed, forming a distinct peak at fractions 13-14 (P2). In contrast, most of GluA2 sedimented in a uniform manner producing a peak coinciding with P2. A minor fraction of GluA2 was present in the higher density fractions. The mutant GluA3 YR/AG showed intermediate behavior distributing equally between the two peaks P1 and P2 but was also present in the fractions between these peaks (Fig. 6B). In addition to the 110-kDa receptor subunit monomer band a diffuse ϳ250-kDa band was prominently present in the peak FIGURE 4. Instability of the LBD does not explain the poor trafficking of GluA3. A, schematic structure of the GluA3i LBD polypeptide expressed intracellularly in bacterial cells. The included S1 and S2 residues are indicated, as are point mutation sites; the S1 and S2 domains are joined by a two-amino acid linker. All LBD constructs for bacterial expression have a C-terminal His 6 tag for purification. B, typical graphs for urea denaturation of purified GluA2i and GluA3i LBD proteins after expression and purification from bacteria. Measurement was by intrinsic fluorescence spectroscopy, and LBDs were denatured in the absence or presence of 10 mM L-glutamate. C, a scatter plot of measured EC 50 values for urea denaturation of GluA2i and GluA3i LBDs with or without ligand. Points indicate individual values, and the line is the mean. Statistical analysis by one-way ANOVA gave p Ͻ 0.0001. Bonferroni's multiple comparison test indicated a significance effect (p Ͻ 0.001) for the presence of L-glutamate for both LBDs; it also indicated GluA3i LBD was significantly more stable than A2i LBD both without L-Glu (p Ͻ 0.05) and with L-Glu (p Ͻ 0.01). D, best fit non-linear plots of thermal denaturation of purified GluA2i, A3i, and A3i (YR/AG) LBD proteins. Measurement was by CD spectroscopy. LBDs are with or without L-glutamate. GluA2i is shown in blue, GluA3i in red, and A3i (YR/AG) in gray. Statistical analysis by paired t test of measured T m in the presence or absence of L-Glu indicates a significantly stabilizing effect of glutamate (p ϭ 0.0038). In the absence of L-glutamate, symbols and lines are solid; with L-glutamate, symbols and lines are open or dashed, respectively.  fractions of GluA2 and GluA3 in amounts that correlated roughly with the monomer band. This had been noticed previously and assigned to SDS-resistant assembly intermediates (26). Additionally, with GluA3 samples, the densest fractions at the bottom of tube produced further intense diffuse smear-like immunoreactivity at very high molecular weight region (Fig.  6A). Due to this complex distribution of immunoreactivity we quantitated the immunoreactivity separately for the 110-kDa band (Fig. 6B) and for the entire sample lane (Fig. 6C). Inclusion of the diffuse immunostaining at a Ն250-kDa region did not significantly affect the overall result but blurred the P1 peak of GluA3 protein because of the large amount of aggregates in the bottom fractions (Fig. 6C).
To determine the major oligomeric state(s) of the receptor in P1 and P2, we subjected a sample of every other sucrose gradient fraction to BN-PAGE followed by immunoblotting. The samples were separated to four populations by the rate of migration in the gel, which were assigned as aggregates (Ag), tetramers (T), dimers (D), and monomers (M) (Fig. 7A) as based on previous studies (24,27). The highest density sucrose fractions resolved into diffuse large-sized aggregates, whereas P1 and P2 peaks best coincided with the tetramer and dimer populations. The monomeric form was only (occasionally) seen in the post-P2 fractions of GluA2 protein (Fig. 7A). Quantitative analysis of the respective band intensities from five indepen-dent experiments revealed the strong presence of aggregates in GluA3 and dimers in GluA2 as the most striking differences between the subunits (Fig. 7B). Again, the double mutant GluA3 YR/AG showed intermediate behavior. Tetramers accounted for 25-28% of the total immunoreactivity in all three receptor variants. The ratio of dimers-to-tetramers was ϳ0.40 for GluA3 and 2.0 for GluA2 (Fig. 7C), the predominance of dimers in GluA2 being in agreement with previous studies of GluA2 biogenesis (26,27). We also analyzed the Triton X-100 extracts directly by BN-PAGE without prior fractionation in a sucrose gradient. The results confirm the strong presence of the dimeric species in wild-type GluA2i and its virtual absence in GluA3i. Moreover, the double mutants GluA2i AG/YR and GluA3i YR/AG showed banding patterns that were intermediate of their parental wild-type subunits (Fig. 7D).
The sucrose density gradient fractionation and BN-PAGE analysis indicate that GluA3 receptors have a much stronger tendency to form tetramers and larger oligomers/aggregates than GluA2, and this difference was largely but not totally eliminated by the YR/AG double mutation.

Discussion
Native GluA3 subunits are predominantly associated with GluA2 (5, 6), and coexpression with GluA2 (Ref. 12; this study) rescues the otherwise poor surface expression of recombinant FIGURE 6. Sucrose gradient analysis reveals assembly differences between the receptors. A, typical SDS-PAGE immunoblots of fractions following sucrose gradient centrifugation of Triton X-100 extracts from HEK293 cells expressing N-terminally FLAG-tagged GluA3, GluA3 (YR/AG), and GluA2. Fraction 1 is the densest, and fraction 23 is the lightest, as indicated by the bar underneath. The subunit monomer signals are indicated by an arrowhead. A band at 250 kDa and a higher aggregate smear is seen for all receptors; however, the fraction location and intensity varies. All blots were probed with anti-FLAG antibody, and the molecular weight markers are indicated. B and C, signal intensity profiles derived from the immunoblots shown. B shows the signal profile from the monomer band in each fraction, where (Fn/⌺F1 Ϫ F24). C shows the signal profile from entire detected signal (monomer, oligomer, and aggregate) of each fraction. Peak 1 (P1) corresponds to fractions F5-8, and peak 2 (P2) corresponds to fractions F12-15.
GluA3 subunits (12,29), suggesting that GluA3 is an "obligatory heteromeric" subunit. Two amino acid residues, Tyr-454 and Arg-461, were previously identified as critical determinants of the poor surface expression of homomeric GluA3 receptors (12). Substituting the residues with their GluA2 counterparts, alanine and glycine, respectively, rescues the surface expression of GluA3 homomers to GluA2 levels, and conversely, their incorporation to GluA2 leads to dramatic decrease in the transport to cell surface (12). The location of these residues in the ligand binding domain but outside of known subunit interfaces and the ligand binding cleft makes their striking effect on receptor transport enigmatic. Here, we wished to elucidate the biochemical mechanisms responsible for the inefficient expression of homomeric GluA3 receptors on the plasma membrane by examining the differences in molecular properties between GluA2 and GluA3 receptors and their dependence on the YR/AG residues.
Our analysis of full-length receptors identified detergent solubility as a biochemical property that clearly distinguished between GluA2 and GluA3 and was strongly dependent on the YR/AG residues. The majority of GluA2 expressed in HEK293 cells was soluble in Triton X-100, whereas most of immunoreactive GluA3 appeared as insoluble aggregates. Double mutations at YR/AG positions shifted the solubility of GluA2 or GluA3 toward the other subunit, indicating that Tyr-454 and Arg-461 promote aggregation. To find out if this is caused by poor folding or instability of GluA3 LBD, we expressed wildtype and mutated GluA2 and GluA3 LBDs as soluble fusion proteins. The secretion efficiencies of GluA2 and GluA3 LBDs in HEK293 cells closely mimicked the transport phenotypes of the cognate membrane-bound receptors and were completely reversed by mutations at the YR/AG residues. Spectroscopic analysis of the unfolding of purified LBDs, purified from E. coli expression, yielded classical cooperative unfolding curves for both GluA2 and GluA3. Unexpectedly, GluA3 LBD appeared more stable than GluA2 against both urea-induced and thermal denaturation, but this effect was unrelated to the trafficking difference as the GluA3 YR/AG double mutant showed similar stability to wild-type GluA3. These findings suggest that the aggregation-promoting effect of Tyr-454/Arg-461 residues is not caused by an unstable fold of GluA3 LBD. Of note, the occurrence of thermal denaturation of ligand-free LBDs at or near physiological temperature together with the strong stabilizing effect by glutamate is consistent with the emerging concept that glutamate in the ER acts as a physiological chaperone in AMPA receptor folding (23, 24, 30 -32).
Analysis of the detergent-soluble receptors, assumed to represent the properly folded protein by sucrose density gradient centrifugation and blue native gel electrophoresis, revealed differences in the oligomeric status of GluA2 and GluA3 proteins. Major immunoreactive peaks from sedimentation in density gradients resolved in higher oligomers/microaggregates, tetramers, and dimers. Consistent with aggregation in the macroscopic scale, more than half of immunoreactive GluA3 existed as microaggregates or complexes larger than tetramers. GluA2 and GluA3 preparations contained tetramers at similar relative amounts, but the proportion of dimers was much higher for GluA2, in agreement with an earlier study (27). The double mutation YR-to-AG shifted the hydrodynamic behavior and dimer-to-tetramer ratio of GluA3 toward GluA2, albeit not completely. The lower dimer-to-tetramer ratio (0.4:1) suggests that GluA3 tetramers are more stable/less dynamic than those of GluA2 (2:1). Considering this fact with the failure of stargazin, which associates selectively with tetrameric AMPA receptors in the ER (15,33) to significantly promote GluA3 surface expression, we speculate that the (majority of) GluA3 tetramers do not represent the native conformation but some incompletely matured intermediate. . BN-PAGE analysis reveals oligomeric steady-state differences between the receptors. A, typical BN-PAGE immunoblots of every other fraction after sucrose gradient centrifugation of Triton X-100 extracts from HEK293 cells expressing N-terminally FLAG-tagged GluA3, GluA3 (YR/AG), and GluA2 (same samples as shown in SDS-PAGE immunoblots). Signals corresponding to aggregate (Ag), tetramer (T), dimer (D), and monomer (M) species are indicated. Boxes in the middle panel highlight the oligomeric state and regions used for the percentage analysis. Fraction numbers are indicated above and below, and P1 and P2 indicate the analyzed fractions from the peaks seen in the Fig. 6 SDS-PAGE analysis. B, summary graph of the oligomeric state for each receptor. Bars correspond to mean Ϯ S.E., n ϭ 5. Statistical analysis by two-way ANOVA gave an overall p Ͻ 0.0001. Bonferroni's post-test comparison of oligomeric states between receptors indicated significant difference in aggregate and dimer states between GluA2 versus GluA3 (p Ͻ 0.001 for both) and GluA2 versus A3 (YR/AG) (p Ͻ 0.01 for both). However, the GluA3 wild-type versus mutant did not show any significant difference for any oligomeric state. Notably, there was no significant difference in tetramer or monomer percentage between any of the receptors. C, graph highlighting the variation in dimer:tetramer ratio between the receptors. Bars correspond to mean Ϯ S.E., n ϭ 5. Statistical analysis by one-way ANOVA gave an overall p ϭ 0.0342. Bonferroni's post-test comparison indicated significant difference in between GluA2 versus GluA3 (*, p Ͻ 0.05) but not for any other comparison. D, BN-PAGE immunoblots (anti-FLAG) of crude Triton X-100 extracts prepared from HEK293 cells expressing FLAG-tagged GluA2i, GluA3i, GluA2i (AG/ YR), or GluA3i (YR/AG). Blots from two independent transfections are shown.
Our study shows that homomeric GluA3 receptors are prone to aggregation at both molecular and macroscopic levels and that this tendency is strongly linked to the two residues that determine the intracellular retention of GluA3 homomers. In an attempt to provide a mechanistic explanation for these findings, we present a tentative model (Fig. 8) that accommodates our experimental results with the body of current research literature on AMPA receptor assembly (for reviews, see Refs. [35][36][37][38]. The crystal structure of GluA2 tetramers shows that the LBDs and NTDs are organized as pairs of dimers, one subunit in each pair occupying a proximal and the other a distal position around the central axis, whereas the membrane-associated regions are arranged around the ion channel in a 4-fold symmetrical fashion (10). Based on the GluA2 tetramer struc-ture, the critical GluA3 residues Tyr-454 and Arg-461 would point outward from the tetramer in the proximal LBDs and inward in the distal LBDs (12). Heteromeric AMPA receptors are believed to have a 2:2 stoichiometry with the subunits, adopting an alternating arrangement (1-2-1-2; Refs. 3 and 39), which in the case of GluA2/3 is ensured by preferred formation of NTD heterodimers (11,34,35). Thus, depending on whether the GluA3 subunits would favor the proximal or the distal positions in the LBD layer, the critical residues would be either exposed on the tetramer's outer surface or masked inside. The rescue of plasma membrane expression and decrease of GluA3 aggregation would then arise from preferential adaptation of GluA3 LBDs of one of the two alternative orientations upon coassembly with GluA2. Whether the outer or inner orientation is detrimental to proper maturation and surface expression is presently unclear. However, we favor a scenario in which the outward orientation of Tyr-454/Arg-461 residues would lead to aggregation through lateral interactions, as we suggested earlier (12) based on the location of homologous GluA2 residues at an intermolecular interface present in an LBD crystal structure (19,20). Therefore, we propose that upon heteromeric assembly with GluA2, GluA3 LBDs would strongly favor the distal positions (as in heteromer 2; Fig. 8B). In contrast, in GluA3 homomers (and in the unfavored heteromer 1-like GluA2/3 receptors) the exposed Tyr-454/Arg-461 residues would drive the formation of higher oligomers, impeding maturation and transport to cell surface (Fig. 8C). Similarly, the poor secretion of GluA3 LBD would be caused by a tendency to form head-totail intermolecular contacts, which may interfere with efficient secretion. The other scenario, which gives the critical role to the inwardly oriented residues, is more difficult to reconcile with the LBD data.
To summarize, one of the two orientations of GluA3 residues Tyr-454 and Arg-461 promotes aggregation and formation of transport-incompetent tetramers, resulting in intracellular retention of homomeric GluA3 receptors. Coassembly with GluA2 leads to elimination of the deleterious orientation allowing the forward trafficking of GluA2/3 heteromers to the cell surface. In conclusion, our results demonstrate that the propensity to aggregate can explain subunit-dependent differences in surface expression of AMPA receptors. Interestingly, the critical determinant for the poor surface transport of homomeric GluA1 and GluA4 flop receptors, a single Leu/Val residue, is positioned quite close to the Tyr-454 and Arg-461 studied in GluR3 (13,20). Future studies will reveal whether the mechanisms identified here for GluA3 will also apply to the poor trafficking of flop variants.  The proximal and distal LBDs chains (relative to the 2-fold axis of symmetry) are colored blue and green, respectively, and the Ala-451 and Gly-458 residues are shown as red van der Waals spheres. The upper model shows the LBD layer from above, whereas the lower model shows the tetramer from the side with NTD layer in light gray. Note the peripheral position of the residues in the proximal and central location in the distal LBDs. B, simplified schematic of LBD organization in GluA3 homomers and in GluA2/3 heteromers. The proximal LBDs are shown as circles, and the distal LBDs are shown as squares. GluA3 residues Tyr-454 and Arg-461 are shown as black circles. GluA3 and GluA2 subunits are colored red and blue, respectively. Note that in GluA2/3 heteromers, Tyr-454/Arg-461 residues can face either outward (heteromer 1) or inward (heteromer 2). C, schematic of major assembly and transport pathways of homomeric GluA3 and heteromeric GluA2/3 receptors. After tetramer assembly, the transport-competent heteromers (left) undergo maturation, leave the ER, and traffic to the plasma membrane. GluA3 homomers predominantly give rise to aggregation via molecular interactions involving the proximal LBDs and remain stuck within the ER. D, the primary sequence surrounding the identified residues of interest (shown in blue and red) within the S1 domain in the four AMPA receptor subunits. Asterisks indicate identical residues; double dots indicate strong similarity; single dots indicate weak similarity.