The impact of subunit type, alternative splicing, and auxiliary proteins on AMPA receptor trafficking

Abstract

AMPA receptors underlie fast excitatory synaptic transmission in the mammalian nervous system and are critical for the expression of synaptic plasticity. Four genes encode the AMPA receptor subunits, each subject to RNA editing and alternative splicing at multiple positions. In addition, each tetrameric AMPA receptor can harbor up to four auxiliary proteins of which there are multiple types. Subunit type, alternative splicing, and auxiliary proteins are all known to affect AMPA receptor gating and trafficking. However, determining which factors dominate AMPA receptor trafficking requires high-throughput assessment of trafficking across multiple conditions. Here, we deploy two such methods to assess the relative contribution of AMPA receptor subunit type (GluA1 versus GluA2), alternative splicing (flip versus flop), and various transmembrane AMPA receptor regulatory proteins (TARPs) to AMPA receptor trafficking. We find that subunit type is the most important factor, with GluA2 showing a much better surface expression than GluA1, and alternative splicing plays a secondary role, with flip subunits consistently outperforming flop variants in surface expression across all conditions. Type 1 TARPs (γ2-4 and γ8) enhance surface trafficking, while Type 2 TARPs (γ5 and γ7) reduce surface expression, although we could not detect differences within each type. These data will be a helpful resource in comparing surface expression across a variety of AMPA receptor compositions. Our assays will also enable high-throughput assessment of novel disease-associated mutations, chimeras, and auxiliary and chaperone proteins.

AMPA receptors mediate the majority of fast excitatory transmission in the mammalian central nervous system (1). These ionotropic glutamate receptors are tetrameric assemblies of various subunits (GluA1-4) that can form homomeric or heteromeric complexes. Dynamic regulation of synaptic AMPA receptor number and type is a powerful method of controlling synaptic strength in response to activity and during development (1, 2, 3). Regulating synaptic AMPA receptor content involves shifting the balance between endo- and exocytosis as well as lateral diffusion from extra-synaptic spaces (3, 4, 5). In addition, local synthesis of new AMPA receptor subunits in the dendritic spine can contribute additional AMPA receptors as needed (6, 7). Supplying new AMPA receptors to synapses through any of these routes ultimately depends on efficient endoplasmic reticulum synthesis and exit (8). Multiple factors contribute to the efficiency of AMPA receptor synthesis or biogenesis, including the type of subunit (GluA1 versus GluA2, 3, or 4), RNA editing (9, 10), alternative splicing (11, 12), and the presence of chaperone (13) and auxiliary proteins (1, 14, 15, 16).

Native AMPA receptors assemble with a myriad of auxiliary proteins (1, 17). These include the cornichon proteins (18), cys-knot associated proteins (CKAMP) (19), GSG1L (20, 21), and the transmembrane AMPA receptor regulatory proteins (TARPs) (14, 22). TARPs generally act as gain-of-function auxiliary proteins, increasing agonist efficacy and potency (23), attenuating polyamine block (24), slowing desensitization, and accelerating recovery (22, 25) (with some exceptions (26)). Most TARPs also promote the surface expression of AMPA receptors (22, 27, 28, 29) and their differential retention in synaptic versus extra-synaptic spaces (30). Prior work has examined the impact and/or mechanism of specific TARPs, such as γ2 (aka stargazin) (22, 27, 29) or γ8, or compared and measured the impact of several TARPs on AMPA receptor surface (31). However, there has been no systematic assessment of the impact of AMPA receptor subunit type, splice variant, and TARP on surface expression.

Here, we dissect the relative contributions of subunit composition, splice variants, and TARP auxiliary proteins in AMPA receptor trafficking and any potential interactions between these factors through two distinct approaches. First, we implement a flow cytometry method to examine GluA1, GuA2, and GluA1/2 surface expression alone or in the presence of TARPs with single-cell resolution. Second, we develop a split luciferase assay to measure the fraction of AMPA receptors on the cell surface across a population. We systematically measure the plasma membrane levels of GluA1 and GluA2, both flip and flop isoforms. For each subunit and heteromeric assembly, we additionally evaluate the impact of the TARP family of proteins, γ1-8 (excluding γ6). We find that GluA2 traffic is more efficient than GluA1 and flip isoforms traffic better than flop (12). We also find that Type 1 TARPs (γ2-4 and 8) enhance plasma membrane expression, while Type 2 TARPs (γ5 and γ7) impair surface trafficking. In general, all these effects are additive, but some combinations produce synergistic effects.

Results

Flow-based trafficking assay

To assay the plasma membrane levels of AMPA receptors in a quantitative, high throughput manner, we drew upon past flow-based strategies that measure the total receptor number using genetically encoded fluorescent proteins and the surface receptor number using an extracellular epitope and cell-impermeant labels (Fig. 1A) (32, 33, 34, 35). We appended the bright and fast-maturing mNeonGreen (36) (hereafter referred to as GFP) to the cytoplasmic tail of GluA1(Q) and GluA2(Q). An amino-terminal hemagglutinin (HA) tag between the signal peptide and the amino-terminal domain served as the extracellular epitope. These constructs were transfected into the FreeStyle 293-F suspension cell line, and the GFP and HA-APC emission intensities were measured using flow cytometry to assay total and surface receptor expression, respectively (gating scheme in Fig. S1). However, each of these channels contains extraneous components. Since the GFP variant matures quickly (∼10 min (36)), the GFP channel reflects some fraction of mature GFP who’s upstream AMPA receptor subunit is still folding. We routinely added the protein synthesis inhibitor cyclohexamide before flow experiments to reduce this asynchrony. The HA-APC channel may contain signals from anti-HA antibodies trapped in dead or dying cells, leading to false-positive surface signals. AMPA receptors, especially co-transfected with TARPs, can cause excitotoxicity and possibly cause even greater false positive surface signal (Fig. S1B). Therefore, we use DAPI to exclude dead cells, minimizing the effect of antibody trapping (Fig. S1B). Example flow plots of the surface (APC) versus total (GFP) signals from GluA1 and GluA2 (both flip and flop variants) are shown in Figure 1B. GluA2 showed greater surface expression than GluA1 based on histograms of APC fluorescence intensity (Fig. 1B, right side of each plot). To normalize the surface signal to total AMPA receptor content in each cell, we examined the surface-to-total signal ratio for each cell (e.g., single-event APC/GFP intensity). GluA2, both flip and flop, had higher surface/total ratios than GluA1, as seen in the right shifts of cumulative distribution plots toward the higher surface/total ratios (Fig. 1C). Consistent with this, at the population level, GluA2 showed higher percentages of APC + cells that were also GFP+ (Fig. 1D, percentage APC+ of GluA2 flip: 66 ± 3%, n = 8; GluA2 flop: 51 ± 3%, n = 4; GluA1 flip: 40 ± 2%, n = 8, p = 8e-6 versus GluA2 flip; GluA1 flop: 25 ± 3%, n = 4, p = 0.0016 versus GluA2 flop). The flip variants of both GluA1 and GluA2 had stronger surface trafficking than their flop counterparts, also consistent with past work (12, 37).

Figure 1.

Intrinsic AMPA receptor subunit and splice variant trafficking differences revealed by flow cytometry-based assay. A, schematic of flow-based assay where intracellular green fluorescence reports total AMPA receptors and extracellular APC fluorescence indicates the plasma membrane component. Created with Biorender.com. B, dot plots for flip (left) and flop (right) variants of GluA1 (blue) and GluA2 (red) showing the total AMPA receptors on the x axis (GFP channel) and surface AMPA receptors on the y axis (HA-APC) in log scale. The solid line is a piecewise linear fit. The upper histogram represents GFP counts while APC counts are shown on the righthand histogram for either GluA1 (blue) or GluA2 (red). C, Cumulative probability of single event APC over GFP ratios for GluA1 (blue) or GluA2 (red) subunits, both flip (i, darker colors) and flop (o, lighter colors) for the experiment in B. D–F, Summary of the percentage of APC + cells within the GFP + population (D), threshold (E) and (F) slope of piecewise fit of the indicated constructs. Symbols are separate transfections and error bars are SD. p values of less than 0.005 are denoted by ∗∗∗. The p-value for Welch-corrected two-tailed t-tests in panel (D) between GluA1i and GluA2i was 8e-6 and 0.0016 between GluA1o and GluA2o. The p-value for Welch-corrected two-tailed t-tests in panel (E) between GluA1i and GluA2i was 1e-9 and 0.006 between GluA1o and GluA2o. A one-way ANOVA for data in panel (F) was significant (F(3,20) = 85.13, p < 0.0001) with post hoc Holm-Sidak tests having p values less than 0.0001 for all comparisons except GluA2i versus GluA2o, which was non-significant.

A striking feature of the flow plots is a “hockey stick” pattern. Phenotypically, the GFP signal intensity increases without much change to the APC signal until, at some point, both signals increase. Thus, AMPA receptors seem to accumulate intracellularly until they reach some threshold, at which point they emerge on the plasma membrane. However, this point of emergence is different for each subunit, with the GluA2 flip beginning to show detectable plasma membrane expression at much lower levels of total AMPA receptor (e.g. GFP signal) than the GluA1 flip (Fig. 1B, left). A similar effect was observed with GluA1 and GluA2 flop. However, once these receptors began to appear on the surface, the slope of the APC/GFP relation was quite different (Fig. 1B, right). We quantified this hockey stick pattern using a piecewise linear function (see Methods) that separates the data into phases with distinct slopes. In the “initial” phase, AMPA receptors accumulate in cells but with little surface signal; hence, the slope is negligible and constrained to zero. In the “later” phase, AMPA receptors continue accumulating in cells but also appear on the surface; hence, the slope is steeper. We termed the transition between these two phases the “threshold”. The threshold for GluA2, both flip and flop, was lower than that of GluA1 (Fig. 1E, threshold (log(GFP)) GluA2 flip: 2.63 ± 0.01, n = 8; GluA2 flop: 2.72 ± 0.02, n = 4; GluA1 flip: 3.10 ± 0.02, n = 8, p = 1e-9 versus GluA2 flip; GluA1 flop: 3.00 ± 0.05, n = 4, p = 0.006 versus GluA2 flop). Interestingly, the post-threshold slope of the APC/GFP signal also varied between subunits with the following rank order (steepest first): GluA1 flip, GluA2 flip = GluA2 flop, GluA1 flop (Fig. 1F, slope (log(APC)/log(GFP)) GluA1 flip: 1.49 ± 0.03, n = 8; GluA2 flip: 1.27 ± 0.02, n = 8; GluA2 flop: 1.20 ± 0.06, n = 4, GluA1 flop: 0.72 ± 0.03, n = 4). Such differences may arise due to differential rates of folding, dimer and tetramer assembly, ER exit, processing, and the endo/exocytosis balance. The possible processes contributing to the threshold and slope measurements need further clarification (see Discussion). But empirically, GluA2 combined an intermediate slope with a lower threshold, while GluA1, both flip and flop, had a higher threshold (Fig. 1, E and F). Interestingly, GluA1 flip had the steepest slope while GluA1 flop had the shallowest (Fig. 1F). Since AMPA receptor auxiliary or chaperone proteins found in neurons are absent or present at low levels in HEK cells (13, 38), these observed differences arise from the intrinsic folding, multimerization, and trafficking of the AMPA receptor.

TARP effects on AMPA receptor surface expression as measured by flow cytometry

Next, we assessed the impact of TARP auxiliary proteins on the trafficking of GluA1 flip. TARPs are divided into Type 1 and 2, with Type 1 being further subdivided into Type 1a and 1b (1). To systematically assess how each of the TARP types impacts GluA1 trafficking, surface, and total GluA1 flip were measured in cells co-transfected with either empty vector (EV) or each of the TARPs. We included γ1 as an additional negative control as it bears sequence homology with TARPs but has been reported not to modulate AMPA receptor currents (14). Figure 2A shows an example of flow plots of GluA1 flip plus empty vector, type 1a TARP γ2 or Type 2 TARP γ5. As expected, γ2 increased the slope of the APC/GFP ratio. Interestingly, γ5 co-transfection reduced or shallowed the slope. We found a similar pattern across all TARPs where type 1a and 1b TARPs enhanced the surface trafficking of GluA1 flip, producing right shifts in the cumulative probability of surface/total ratios compared to EV or γ1 co-transfection (Fig. 2B). In contrast, the Type 2 TARPs γ5 and γ7 left-shifted the distributions, indicating impaired trafficking (Fig. 2B). To further dissect TARP differences, we fit the single cell flow data (Fig. 2C) to measure threshold and slope. No statistical effect was observed on the threshold of GluA1 flip for any TARP (Fig. 2D, 1-way ANOVA F(7,33) = 0.9254, p = 0.5). However, both type 1a and 1b TARPs did increase the slope of GluA1 flip trafficking, while type 2 TARPs reduced the slope (Fig. 2D, slope (log(APC)/log(GFP)) GluA1 flip with EV: 1.48 ± 0.03, n = 8; with Type 1 TARPs γ2-4, γ8 range from: 1.74 ± 0.03–1.79 ± 0.03, n = 4, p versus EV all less than 0.001; with γ5: 1.04 ± 0.03, n = 4, p < 0.001 versus EV; with γ7: 0.81 ± 0.03, n = 4, p < 0.001 versus EV).

Figure 2.

GluA1 surface trafficking is enhanced by Type 1a and 1b but reduced by Type 2 TARPs. A, dot plots for GluA1 flip either alone (left, black), or co-transfected with Type 1a TARP γ2 (middle, blue) or Type 2 TARP γ5 (right, orange). A solid line is a piecewise linear fit (see Methods). The upper histogram represents GFP counts, while APC counts are shown on the right-hand histogram. B, cumulative probability of single event APC over GFP ratios for GluA1 flip alone (black) or with the indicated gamma subunit. C, dot plots for GluA1 flip alone (black) or with the stated gamma subunit sharing the color scheme of (B). D, summary threshold (left) and slope (right) from linear piecewise fit for each gamma subunit with GluA1 flip. Symbols are separate transfections, and error bars are SD. No statistical effect was observed on the threshold in panel (D) using one-way ANOVA (F(7,33) = 0.9254, p = 0.5). The slopes in panel € were statistically different (one-way ANOVA (F(7,26) = 91.64, p < 0.0001)). Post hoc Dunnett tests identified γ2-8, but not γ1, as being statistically different from the EV control, with all these adjusted p-values < 0.0001. This is denoted by ∗∗∗.

To determine whether the impact of TARPs on AMPA receptor trafficking depended on subunit identity or splice variants, we repeated this experiment with GluA1 or GluA2 (both flip and flop) as well as GluA1/GluA2 co-transfection composed solely of flip or flop subunits. The heatmaps in Figure 3A show the following general trends from these experiments: GluA2 had greater surface expression than GluA1 with heteromers falling in between, flip receptors had greater surface expression than flop, and Type 1 TARPs increased surface trafficking while Type 2 impaired it (Fig. 3A). Representative single-cell surface/total ratios from one such experiment are shown in Figure S2. The percentage of jointly APC+ and GFP + cells, single-cell surface to total ratios as well as fit values of thresholds, and slopes from these experiments are all summarized in Figures S3 and S4 and Table 1, Table 2, Table 3, Table 4. To estimate the relative influence of subunit, splice variant, and TARP on these observable trafficking markers (%APC+ of GFP+ and surface/total ratio), we conducted a three-way ANOVA on the homomeric data with subunit identity (GluA1 versus GluA2), splice variant (flip versus flop), and TARP as factors. All these factors showed significant effects on the observable trafficking markers. Specifically, subunit identity showed the strongest influence (%APC+ of GFP+: 49% of variance, F(1,82) = 461.9, p < 0.0001; APC/GFP ratio: 69% of variance, F(1,73) = 996.6, p < 0.0001) followed by the flip/flop cassette (%APC+ of GFP+: 15% of variance, F(1,82) = 249, p < 0.0001; APC/GFP ratio: 17% of variance, F(1,73) = 251, p < 0.0001) and the type of TARP (%APC+ of GFP+: 15% of variance, F(7,82) = 21, p < 0.0001; APC/GFP ratio: 5% of variance, F(7,73) = 10, p < 0.0001). The individual flow data were fit to distinguish possible effects of threshold and slope, as in Figures 1B and 2A (Fig. 3B) and further analyzed using a similar 3-way ANOVA on the homomeric data. GluA1, both flip and flop, showed greater thresholds than GluA2 while the heteromers were intermediate (Fig. 3B). Indeed, 67% of the variance in the threshold values was accounted for by the subunit identity (F(1,114) = 640.2, p < 0.0001). Interestingly, we found that across TARPs, the flop subunits of GluA1 had lower thresholds than the flip variants, while the converse was true for GluA2 (Fig. 3B, Figs. S3C, S4C, Table 3, Subunit x splice variant interaction effect F(1,114) = 54, p < 0.0001).

Figure 3.

TARP effects on GluA1, GluA2 and heteromeric AMPA receptor surface trafficking. A, left, heatmaps of percentage APC positive cells (APC+) out of GFP positive (GFP+) cells across all flow experiments for the indicated AMPA receptor subunits and TARPs. EV stands for empty vector. A, right, Heatmap of median APC/GFP ratio for all flow runs with the indicated AMPA receptor subunits and co-transfected TARPs. B, heatmaps of threshold (left) and slope (right) fits for flow runs from GluA1, GluA2, or heteromers co-transfected with the indicated construct.

Table 1.

Summary of percent jointly APC+ and GFP + events

TARP	GluA1		GluA2		GluA1/GluA2
	flip	flop	flip	flop	flip	flop
EV	40.5 (6.5)	25.5 (6.3)	66.1 (8.1)	51.0 (6.9)	51.2 (7.1)	38.6 (3.7)
γ1	34.5 (9.0)	20.9 (7.9)	52.2 (4.3)	39.9 (3.3)	48.3 (7.2)	31.4 (3.5)
γ2	30.5 (2.0)	20.2 (2.0)	54.9 (5.5)	44.3 (3.9)	42.9 (4.4)	36.9 (3.1)
γ3	34.6 (2.5)	22.7 (2.4)	55.1 (7.9)	43.0 (5.8)	44.7 (5.5)	35.6 (4.9)
γ4	32.6 (2.4)	18.7 (2.4)	54.1 (4.8)	39.5 (5.8)	40.1 (3.7)	32.3 (4.8)
γ8	22.7 (1.5)	13.9 (1.3)	41.2 (5.4)	25.5 (4.7)	33.6 (4.5)	23.5 (2.0)
γ5	28.4 (3.3)	9.5 (0.9)	55.1 (4.0)	43.8 (6.8)	39.1 (5.1)	29.1 (2.9)
γ7	14.6 (2.5)	2.2 (1.1)	52.9 (4.3)	34.7 (4.0)	27.8 (11.3)	16.3 (8.5)

Data are means (standard deviation). N ranges from three to 8.

Table 2.

Summary of median APC/GFP ratios

TARP	GluA1		GluA2		GluA1/GluA2
	flip	flop	flip	flop	flip	flop
EV	0.25 (0.04)	0.12 (0.02)	0.54 (0.11)	0.36 (0.03)	0.33 (0.08)	0.25 (0.03)
γ1	0.24 (0.05)	0.13 (0.03)	0.50 (0.04)	0.35 (0.01)	0.36 (0.06)	0.25 (0.03)
γ2	0.27 (0.02)	0.20 (0.03)	0.59 (0.09)	0.40 (0.01)	0.35 (0.07)	0.29 (0.05)
γ3	0.23 (0.01)	0.17 (0.02)	0.58 (0.09)	0.37 (0.03)	0.35 (0.09)	0.26 (0.03)
γ4	0.24 (0.03)	0.15 (0.02)	0.61 (0.04)	0.38 (0.01)	0.36 (0.09)	0.27 (0.04)
γ8	0.23 (0.01)	0.19 (0.02)	0.49 (0.06)	0.33 (0.01)	0.35 (0.08)	0.26 (0.04)
γ5	0.14 (0.01)	0.09 (0.01)	0.48 (0.02)	0.30 (0.02)	0.24 (0.05)	0.20 (0.02)
γ7	0.15 (0.01)	0.10 (0.01)	0.52 (0.03)	0.31 (0.02)	0.27 (0.09)	0.18 (0.03)

Data are means of median surface/total ratios from each flow experiment (standard deviation). N ranges from three to 8.

Table 3.

Summary of GFP thresholds

TARP	GluA1		GluA2		GluA1/GluA2
	flip	flop	flip	flop	flip	flop
EV	3.12 (0.07)	2.97 (0.10)	2.63 (0.03)	2.72 (0.04)	2.82 (0.10)	2.75 (0.08)
γ1	3.11 (0.07)	3.04 (0.11)	2.65 (0.05)	2.78 (0.05)	2.88 (0.11)	2.85 (0.09)
γ2	3.05 (0.10)	2.98 (0.12)	2.58 (0.02)	2.70 (0.05)	2.83 (0.10)	2.79 (0.07)
γ3	3.14 (0.08)	3.03 (0.12)	2.69 (0.03)	2.78 (0.03)	2.90 (0.09)	2.85 (0.07)
γ4	3.06 (0.07)	3.00 (0.13)	2.62 (0.03)	2.71 (0.06)	2.84 (0.10)	2.82 (0.06)
γ8	3.06 (0.14)	3.04 (0.12)	2.64 (0.04)	2.73 (0.03)	2.87 (0.09)	2.82 (0.06)
γ5	3.10 (0.14)	2.94 (0.14)	2.63 (0.04)	2.71 (0.05)	2.83 (0.09)	2.77 (0.08)
γ7	3.01 (0.14)	2.79 (0.06)	2.60 (0.04)	2.71 (0.08)	2.78 (0.08)	2.69 (0.06)

Data are means of log(GFP) channel signal intensity (standard deviation). N ranges from three to 8.

Table 4.

Summary of slopes

TARP	GluA1		GluA2		GluA1/GluA2
	flip	flop	flip	flop	flip	flop
EV	1.48 (0.09)	0.71 (0.07)	1.27 (0.05)	1.20 (0.12)	1.26 (0.10)	0.90 (0.14)
γ1	1.54 (0.08)	0.85 (0.05)	1.35 (0.06)	1.29 (0.13)	1.32 (0.14)	0.99 (0.15)
γ2	1.74 (0.10)	1.07 (0.02)	1.28 (0.10)	1.28 (0.12)	1.37 (0.11)	1.09 (0.17)
γ3	1.75 (0.04)	1.06 (0.06)	1.41 (0.08)	1.38 (0.11)	1.41 (0.08)	1.17 (0.19)
γ4	1.74 (0.06)	1.03 (0.12)	1.36 (0.07)	1.27 (0.08)	1.40 (0.14)	1.14 (0.16)
γ8	1.79 (0.08)	1.16 (0.15)	1.35 (0.12)	1.22 (0.09)	1.39 (0.11)	1.15 (0.17)
γ5	1.04 (0.04)	0.39 (0.06)	1.18 (0.08)	1.09 (0.08)	1.00 (0.19)	0.65 (0.10)
γ7	0.81 (0.08)	0.20 (0.02)	1.28 (0.08)	1.07 (0.07)	1.02 (0.18)	0.66 (0.11)

Data are means of slopes, reflecting log(APC)/log(GFP) channel signal intensity, (standard deviation). N ranges from three to 8.

Next, we examined the slope relating the total AMPA receptors to surface AMPA receptor content (Fig. 3B, right, Figs. S3D, S4D and Table 4). Unlike prior measurements where either subunit identity or splice variant accounted for most of the effects in the data set, the slope was influenced by multiple factors and interactions between factors. The clearest effects were observed in the GluA1 flip, which possesses the largest baseline slope (GluA1 + EV slope 1.48 ± 0.03, Fig. 3B, right, Figs. S3D, S4D, and Table 3). Co-transfection of any Type 1 TARP increased the slope of GluA1 flip to between 1.7 and 1.8 while Type 2 TARPs reduced the slope of GluA1 flip to 1.0 or 0.8 (Fig. 3B, right, Figs. S3D, S4C, and Table 3). A similar TARP effect was observed with GluA1 flop, albeit with the smallest intrinsic slope (Fig. 3B, right, Figs. S3D, S4D and Table 3). Interestingly, the slope of GluA2, both flip and flop, was less impacted by either Type 1 or Type 2 TARPs. For comparison, the slopes of GluA1 flip range from 0.8 with γ7 co-transfection to 1.8 with γ8 while those of GluA1 flop span 0.2 with γ7 up to 1.16 with γ8. In contrast, the slops of GluA2 range for 1.07 for GluA2 flop with γ7 up to 1.4 with GluA2 flip and γ3 (Fig. 3B, right, Figs. S3D, S4D and Table 3). The heteromeric slope values are largely intermediate between the homomers.

TARP effects on AMPA receptor trafficking assessed by surface luminescence

The flow cytometry-based assay separates trafficking effects into threshold versus slope at the cost of complexity and throughput. In addition, introducing fluorescent proteins at the carboxy terminus of AMPA receptors may compromise trafficking signals associated with the terminal PDZ ligand or intracellular domain. To complement the flow assay, we implemented a split luciferase-based approach. We appended an HA tag followed by an 11-amino acid segment of luciferase called HiBiT sequence immediately after the signal peptide and before the mature protein. HiBiT binds to the complementary large-bit protein (LgBiT) with sub-nanomolar affinity to form a functional nanoluciferase (39). Measuring luminescence before and after cell permeabilization using digitonin yields surface and total measurements of AMPA receptors (40) (Fig. 4A). Using this approach, we measured the surface expression of GluA1(Q) and GluA2(Q), both flip and flop, as well as co-transfections of all possible flip/flop pairings (Fig. 4B, left). We assessed the homomeric data using a 3-way ANOVA as above and, consistent with the flow data, found that subunit identity was the major driver of surface trafficking (47% of variance), with TARP and flip/flop cassette being secondary (22% and 17% of variance, respectively). Figure 4B summarizes these data as a heatmap of raw surface expression. To better visualize the effect of TARPs, we normalized the surface expression obtained with each TARP co-expression to the same-day surface expression of that AMPA receptor subunit and splice variant co-transfected with an empty vector (Fig. 4C). Across all AMPA receptor subunits and splice variants, we found that Type 1 TARPs increase surface expression while type 2 TARPs reduce it. Specifically, the fold change of Type 1 TARPs versus empty vector ranged from 1.4 fold increase for GluA1 flop with γ3 to 2.3 fold greater with GluA1 flip and γ2 while the reduction by γ5 and γ7 ranged from 0.95 with GluA1 flop plus γ7 to 0.3 with GluA1 flip γ7 (Fig. 4C). These data are further summarized in Figures S5 and S6 as well as Table 5.

Figure 4.

Homo and heteromeric GluA1 and GluA2 AMPA receptor surface expression modulated by TARPs. A, schematic of HiBiT/LgBiT surface expression assay where an amino-terminal HiBiT tag binds with recombinant LgBiT to form functional nanoluciferase. Subsequent permeabilization by digitonin allows the separation of the surface AMPA receptors from the total AMPA receptor content. Created with Biorender.com. B, Heatmap showing the percentage of surface-expressed AMPA receptors for GluA1 and GluA2 homomers or GluA1/A2 co-transfection with either empty vector (EV) or the indicated TARP co-transfected. Flip and flop splice variants are denoted by i and o, respectively. C, Heatmap of fold change in surface expression induced by co-transfection versus EV on the same day.

Table 5.

Percentage of homomeric AMPA receptors on cell surface based on luminescence

TARP	GluA1		GluA2
	flip	flop	flip	flop
EV	4.4 (1.0)	1.1 (0.3)	11.4 (1.8)	7.8 (0.9)
γ1	3.7 (0.8)	0.8 (0.1)	17.7 (3.5)	9.7 (1.8)
γ2	10.0 (2.8)	1.8 (0.5)	18.6 (1.9)	14.1 (2.7)
γ3	7.8 (2.0)	1.4 (0.4)	18.2 (2.1)	10.9 (2.2)
γ4	8.4 (2.2)	1.5 (0.6)	17.7 (2.6)	11.1 (2.0)
γ8	9.0 (2.4)	2.2 (0.7)	23.1 (1.5)	13.3 (2.5)
γ5	2.4 (0.6)	1.0 (0.2)	7.8 (0.8)	4.2 (0.7)
γ7	1.16 (0.1)	0.24 (0.2)	8.3 (1.2)	3.7 (0.8)

Data are mean (standard deviation) of the surface percentage. N equals four for all data.

To assess the trafficking of heteromeric AMPA receptors, we repeated co-transfection experiments with the HiBiT tag on either the GluA1 or the GluA2 subunit. As expected, Type 1 TARPs increased surface expression while Type 2 TARPs reduced surface expression (Fig. 4, B and C, Figs. S5 and S6, Type 1 increases ranging from 1.4 fold to 2.4 fold; type 2 decreases ranging from 0.68 to 0.14 versus empty vector). Interestingly, we found that γ2 and γ8 increased surface expression of co-transfected GluA1 and GluA2 more than γ3 or γ4, respectively (Fig. 4, B and C, Figs. S5 and S6). And we found that γ1 systematically elevated the surface expression of GluA2 but not GluA1 (Fig. 4, B and C, Figs. S5 and S6). These data are summarized in Figures S5 and S6 as well as Table 6. Given the effect on surface expression, we hypothesized that γ1 may act as a TARP with GluA2. However, co-transfection of γ1 with GluA2 did not increase the relative efficacy of kainate nor change the desensitization kinetics in excised patches (data not shown).

Table 6.

Percentage of heteromeric AMPA receptors on cell surface based on luminescence

TARP	HiBiT-GluA1				HiBiT-GluA2
	A1i, A2i	A1i, A2o	A1o, A2i	A1o, A2o	A1i, A2i	A1i, A2o	A1o, A2i	A1o, A2o
EV	13.0 (1.3)	8.9 (1.0)	8.6 (1.4)	6.8 (1.0)	15.6 (2.2)	12.8 (2.4)	11.6 (2.2)	7.9 (2.1)
γ1	14.0 (1.8)	10.1 (1.6)	11.1 (1.4)	7.3 (1.4)	20.5 (4.3)	19.3 (4.2)	13.9 (2.6)	10.8 (3.5)
γ2	27.1 (1.3)	20.3 (1.4)	17.2 (1.4)	14.2 (0.9)	28.9 (4.8)	26.0 (4.4)	22.5 (4.3)	18.7 (4.7)
γ3	22.5 (1.9)	16.8 (1.7)	15.2 (1.1)	11.0 (0.9)	24.9 (3.9)	24.7 (4.3)	19.8 (4.3)	14.1 (3.8)
γ4	21.1 (2.1)	15.6 (0.7)	12.8 (1.4)	9.6 (0.8)	23.4 (3.3)	22.2 (3.2)	18.1 (1.7)	14.3 (3.2)
γ8	26.5 (2.2)	18.6 (1.1)	18.1 (0.7)	13.9 (1.3)	31.4 (6.1)	27.0 (5.0)	21.7 (5.1)	17.6 (4.3)
γ5	6.3 (0.8)	4.5 (0.5)	3.6 (0.6)	3.2 (0.6)	9.0 (1.8)	7.5 (1.0)	6.0 (1.3)	3.9 (2.2)
γ7	4.2 (0.7)	2.8 (0.6)	2.0 (0.7)	1.0 (1.0)	7.4 (1.3)	7.1 (1.4)	3.5 (2.6)	2.6 (2.9)

Data are mean (standard deviation) of the surface percentage. N equals four for all data.

Finally, cell lines such as suspension HEKs may release transiently transfected proteins into the media through exosomes. Thus, some fraction of our surface expression signal may arise from these release proteins and/or dead cells. To control against differential effects from such release, cells were split into two pools. One pool was washed with PBS prior to luminescence readings to remove released exosome proteins and/or dead cells while the other was not washed. These pools showed indistinguishable surface expression levels, indicating washing is not essential (Fig. S7).

Discussion

Using both flow cytometry and split nanoluciferase-based assays, we report the first systematic assessment of how TARPs, subunit type, and alternative splicing at the flip/flop cassette govern surface trafficking of AMPA receptors. We find that subunit identity is the major driver of surface expression, with GluA2(Q) expressing much better than GluA1(Q). Flow cytometry data analysis reveals this effect is primarily due to a lower threshold for channels to be delivered to the plasma membrane (Figs. 1 and 3). Alternative splicing plays a secondary role, with flip subunits trafficking more effectively across the board than their flop counterparts in identical conditions (Figure 1, Figure 3 and 4). Finally, the presence of TARPs influences surface expression, with Type 1 TARPs (γ2-4 and γ8) enhancing surface expression while Type 2 TARPs (γ5 and γ7) reduce surface expression (Figure 2, Figure 3, Figure 4). These effects are conserved in co-transfection of GluA1 and GluA2, independent of which subunit is being detected (Fig. 4B). Our study focused on the widely expressed GluA1 and GluA2 subunits. In native systems, the GluA2 RNA can be edited at either the Q/R site, placing a positively charged Arginine residue in the channel pore, or the R/G site (41, 42), which alters channel kinetics. Despite GluA2 being nearly 100% edited in the brain, we use GluA2(Q) for all our experiments to permit direct comparison with functional expression studies, the vast majority of which use GluA2(Q). While the extent of R/G editing varies between brain regions (42), adding increases during neuronal maturation (43) and so we opted to use the G edited form of GluA2. We did not include GluA4, given this subunit's limited tissue expression in rodents (44), nor did we include GluA3, as this subunit has greatly reduced surface expression (45) and would likely be difficult to resolve. We anticipate these assays will be useful in assessing the effects of post-translational modifications, ER resident chaperone proteins (8, 46) as well as patient mutations in both AMPA receptor subunits (47, 48) and auxiliary proteins (49). To the best of our knowledge, no other study has systematically assessed trafficking across as many conditions as possible. However, multiple other studies have examined the effects of either GluA1 versus GluA2 (37), flip versus flop (12, 37, 50, 51) or a subset of TARPs on specific AMPA receptors (14, 31, 52, 53). Our work aligns with these past investigations in that type 1 TARPs enhance the surface expression of all AMPA receptor subunits, regardless of flip or flop. This contrasts functional modulation by specific TARPs, where γ2 strongly affects the desensitization kinetics of GluA2 flip but not on GluA2 flop (54). We also observe that γ5 reduces surface expression, as reported previously (53), although this effect was not detected in another study (31). And our observation that γ7 is a negative regulator of surface trafficking aligns with the increase in AMPA receptor currents from granule cells upon γ7 knockdown (55).

Limitations of current work

All our measurements are conducted in HEK suspension cells. These cells differ from neurons in several key respects. HEK cells lack the complement of ER-resident chaperone proteins, auxiliary proteins, and endo/exocytosis machinery present in neurons (although see (13, 38) for examples of HEK cells expressing AMPA receptor chaperone proteins). Beyond differences in protein complement, neurons may utilize both canonical and non-canonical secretory pathways for somatic versus dendritically synthesized proteins (7). Nor can HEK cells accurately reflect the influence of neuron-specific glycosylation on these receptors (56). In addition to these shared limitations, the flow cytometry assay also requires that a fluorescent protein be appended to the channel. We added the fluorescent protein to the C terminal tail, potentially interfering with PDZ or other motifs. However, since our flow and luminescence data strongly agree, adding the C terminal fluorescent protein does not appear to be a major confound. Despite the limitations, HEK suspension cells offer considerable advantages. These include ease of use, reproducibility, and the ability to scale. Also, because suspension cells lack endogenous AMPA receptors and TARP proteins, it is possible to quantify the effect of one specific TARP on one specific AMPA receptor without having to genetically delete or compensate for all the other subunits as one needs to do in neurons. Thus, these assays are well suited to the study of de novo disease-associated variants in either AMPA receptors or auxiliary/chaperone proteins in isolation.

Comparison with previous studies

The elevated surface expression of GluA2 over GluA1, as well as that of flip variants over flop, has been known for some time. The effect of type I TARPs to promote surface trafficking has also been repeatedly demonstrated. Here, we consolidate all these findings in a single report, utilizing two complementary assays to facilitate direct comparison among various conditions and estimate the impact of factors such as subunit identity, splice variants, and TARPs. Our data conflicts with past work in two main respects. First, we find that γ1 expression increases the surface trafficking of GluA2. The effect is small compared to Type 1 TARPs but consistent in both flow and luminescence assays (Fig. S3D, Figs. 4C, S5). In contrast, γ1 transfection into stargazin cerebellar granule neurons did not increase AMPA currents (14) as our data might predict. However, any γ1 effect on GluA2 subunit trafficking may be difficult to detect as granule neurons express all other AMPA subunits as well (57, 58). Another conflict is that we consistently observe γ5 and γ7 reduce surface expression of both GluA1 and GluA2, whereas prior studies have found either no effect (59) or enhancement of surface expression as indicated by rescue experiments in granule neurons or whole-cell currents (60, 61). Inferring surface trafficking levels from functional responses is complicated by TARP effects on open probability and conductance, which may possibly reconcile our datasets (53, 55).

Source of intrinsic differences between subunits

GluA2 shows more abundant surface expression than GluA1 (Figs. 1 and 4B), consistent with past work (37). Flow cytometry analysis reveals GluA1 has a higher “threshold” than GluA2 (Fig. 1E), likely accounting for the difference in surface expression. We suggest this greater “threshold” arises due to a longer maturation time of GluA1. The molecular basis for this longer maturation time is unclear but may reflect differences in glycosylation requirements. GluA1 possesses 6 N-glycosylation sites, two of which (N63 and N363) are critical for HEK cell surface expression (62). In contrast, GluA2 has four N-glycosylation sites, with only one of these (N370) being essential for surface expression in HEK cells (63). The more stringent glycosylation requirement of GluA1 may lead to intracellular accumulation and account for the greater threshold prior to surface expression (Fig. 1E). Another possibility is that GluA1 homomers are more difficult to fold properly in HEK cells, leading to increased misfolded proteins accumulating in the ER. In this view, the “threshold” measurement we observe does not reflect GluA1 homomers awaiting surface delivery but an accumulation of misfolded protein. Future work using pulse chase assays, domain swapping and mutations will be able to determine the molecular basis for this subunit difference.

Within the last decade, many disease-associated variants in GRIA and other iGluR genes have been reported (1, 48, 64). These variants may possess alterations in channel gating, trafficking, or both (1, 48). In addition, new chaperone proteins have been uncovered, shedding light on the intricate biogenesis of AMPA receptors (8, 13, 20, 46). The throughput of methods developed here are ideally suited to assay trafficking across variants and with specific chaperone proteins. Thus, despite decades of research on AMPA receptors, new technologies and approaches continue to offer up new questions and avenues of inquiry.

Experimental procedures

Plasmids and cloning

The cDNA encoding mouse γ1 was a kind gift from Dr Robert Dirksen. γ2 (aka stargazin), γ3, γ4, γ8, GluA1, and GluA2 (short form, UniProt [P19491]) were kind gifts from Drs. Bowie or Jayaraman. γ5 and γ7 were synthesized as gBlocks (Integrated DNA Technologies). Both GluA1 and GluA2 contained Q at the Q/R site (10) and GluA2 contained a G at the R/G site (41). All TARP and AMPA receptor subunit cDNAs were subcloned into pcDNA3.1(+) by PCR amplification with high-fidelity Q5 polymerase (New England Biolabs) using primers with restriction sites in the overhangs. A consensus Kozak sequence (GCCACC) was also added to precede the intended start codon. The mNeonGreen coding sequence was inserted at the carboxy terminus of GluA1 and GluA2 using NEBuilder HiFi DNA Assembly Master Mix from PCR-derived fragments (New England Biolabs). For flow cytometry, HA epitope tags were inserted following the signal sequences for GluA1 and GluA2 using Q5 mutagenesis/KLD enzyme mix (New England Biolabs). The sequence corresponding to the flip isoform was converted into the corresponding flop sequence using DNA fragments generated by polymerase chain assembly (GluA1) or a gBlock (GluA2) and NEBuilder HiFi DNA Assembly Master Mix. For luciferase-based surface trafficking, the signal peptides were replaced with that of the hemagglutinin signal peptide (HAsp) (65), followed by an HA epitope, a GS linker, the 11-amino acid HiBiT tag (66) and a GSTG linker before the start of the mature AMPA receptor sequence. All cDNA constructs were verified by Sanger sequencing (Eurofins Genomics) or Oxford Nanopore sequencing (Plasmidsaurus).

Cell culture and transfection

FreeStyle 293-F (HEK293 F) cells were purchased from ThermoFisher Scientific and maintained in FreeStyle 293 Expression Medium in a 125 ml tissue culture flask under constant shaking at 135 rpm according to the manufacturer’s instructions. For transfection, 5 x 10⁵ cells were seeded into each well of a 12-well plate and transfected with 1.5 μg of cDNA (0.75 μg TARP, 0.75 μg AMPA receptor subunit) using a 3:1 mass ratio of PEI:cDNA. NBQX was added to a final concentration of 20 μM at the time of transfection to promote viability. The transfection plates were then returned to the shaking platform to keep cells in suspension.

Flow cytometry

Two hours before flow cytometry analysis, cells were treated with 100 μM cycloheximide to inhibit protein synthesis and synchronize fluorescent protein maturation (67). Cells were pipetted from their 12-well plate into 2 ml of FACS buffer (divalent-free DPBS containing 0.3% BSA and 1 mM Na₄EDTA) and pelleted by centrifugation (200 x g, 5 min, 4 °C). The cell pellets were resuspended in 100 μl of staining mix (10% FBS in FACS buffer) and 2 μl of allophycocyanin (APC) anti-HA.11 [clone 16B12] (BioLegend). This amount of antibody was empirically tested to yield the highest stain index. Cells were stained on ice, protected from light, for 20 min before being washed with 4 ml of FACS buffer and recollected by centrifugation as before. Cells were resuspended in 100 to 200 μl of FACS buffer containing 0.2 μg/ml of DAPI (Cell Signaling Technologies) to assess viability.

Cells were run on an LSRII following instrument compensation, collecting 50,000 events per condition. DAPI was excited with a 405 nm laser and detected between 425 to 475 nm, GFP with a 488 nm laser and detected between 505 to 525 nm, and APC with a 633 nm laser and detected between 650 to 670 nm. Single colored GFP samples were run after compensation to confirm that fluorescence spillover was being compensated.

Data analysis was performed using FCS Express 7 Flow (De Novo Software). Single cells were gated for using forward scatter and side scatter, and only viable cells based on DAPI-exclusion were used for analysis (typically 25,000–30,000 events of the 50,000 collected). The Surface/Total GluA parameter was created using the parameter math function in FCS Express, where the fluorescence intensity of the HA-APC channel was divided by the fluorescence intensity of the GFP channel for each cell.

NanoBiT surface trafficking assay

HEK293 F cells were added to a 12-well plate, 1 ml of 1 x 10⁶ cells/ml in FreeStyle media supplemented with 20 μM NBQX and immediately transfected with 500 ng of AMPA receptor and 1000 ng of auxiliary protein or empty vector using PEI at a 3:1 PEI to DNA ratio. 20 to 24 h post-transfection, 50 μL of cells from each transfection were transferred to a well of an opaque 96-well plate, supplemented with 50 μL of fresh detection reagent (see below), and incubated for at least 5 min. Luminescence measurements were conducted using FLUOstar plate reader (BMG) without a filter, with readings once per minute and a 480 ms integration time. Following 10 min of acquisition, 11 μL of 4 mg/ml digitonin (in 90% PBS, 10% DMSO) was manually mixed into each well. Luminescence was continuously measured for 30 min to monitor cell permeabilization by digitonin. The luminescence at the 10-min time point (immediately before digitonin addition) and the end of the 30-min digitonin incubation were taken as the surface and total measurements, respectively. In all experiments, a non-transfected condition was included and used for background subtraction. For “wash” experiments, cells were first collected, washed three times with PBS (supplemented with 20 μM NBQX), and re-suspended in 1 ml of PBS containing NBQX before 50 μL was taken for luminescence readings.

The detection reagent was a mix of in-house purified LgBiT protein and hydrolyzed Hikarazine 103, each diluted 1:200 in PBS. To purify LgBiT, a g-Block (IDT) encoding an E.coli optimized LgBiT coding sequence with N-terminal hexa-histidine and TEV cleavage sequence was inserted into the pET28a(+) expression vector. A single colony of BL21(DE3) E.coli (NEB), transformed with this vector, was expanded to 2L with an OD₆₀₀ of 0.6 using a constant temperature of 37°C. Expression was induced with 1 mM IPTG. Following overnight incubation at 18°C, bacteria were pelleted and stored at −80°C. Thawed bacteria were resuspended in purification buffer (20 mM sodium phosphate, 500 mM NaCl, pH 7.5) with all subsequent steps performed at 4°C, lysed by sonication, and incubated with 1 mg/ml Lysozyme for 1 h. Lysates were cleared by centrifugation first at 6000g for 30 min, then 50,000g for 30 min, and finally filtered using 0.45 μM PVDF syringe filter. The cleared lysate was supplemented with 10 mM imidazole and flowed over a pre-equilibrated HisPur Ni-NTA Spin Column (Thermo Fisher). After 20 column volume washes in wash buffer (20 mM sodium phosphate, 500 mM NaCl, 25 mM imidazole pH 7.5), protein was eluted in 1 ml fractions using elution buffer (20 mM sodium phosphate, 500 mM NaCl, 250 uM imidazole, pH 7.5) and fractions analyzed using SDS-PAGE and luminescence. LgBiT-containing fractions were pooled, and buffer exchanged into storage buffer (20 mM sodium phosphate, 150 mM NaCl, 50% glycerol, pH 7.5) using a PD-10 desalting column (Cytiva). LgBiT protein was stored at −80°C for long-term or −20°C for short-term at concentrations comparable to commercial LgBiT protein, as estimated by SDS-PAGE and luminescence.

The o-acylated pro-luciferin Hikarazine 103 was purchased from Yves L. Janin (Museum National d’Histoire Naturelle, Paris) (68, 69) and converted to an active form by solubilizing 1 mg of Hikarazine 103 in 0.2 ml of DMSO, mixing 0.3 ml of acidic ethanol (0.1 mM HCl in 200 proof ethanol). The hydrolysis reaction was placed in a 50°C water bath for 2 h, then transferred to −20°C for storage. Hydrolysis was performed in 1 mg batches as needed.

Electrophysiology

Culture dishes were visualized with phase contrast on a Nikon Ti2 microscope using a 20x objective. Outside-out patches were excised using heat-polished, thick-walled borosilicate glass pipettes of three to 5 MΩ resistance. The pipette internal solution contained (in mM) 135 CsF, 33 CsOH, 11 EGTA, 10 HEPES, 2 MgCl_2, and 1 CaCl₂ (pH 7.4), and the external solution contained 150 NaCl, 10 HEPES, 1 MgCl₂ and 1 CaCl₂ and was supplemented with 10 mM monosodium glutamate as an agonist. All recordings were performed at room temperature with a holding potential of −60 mV using an Axopatch 200B amplifier (Molecular Devices) and Clampex 10 or 11. Data were acquired 50 kHz, filtered at 10 kHz with series resistance was routinely compensated by 90 to 95% where the peak amplitude exceeded 100 pA. Rapid perfusion was performed using home-built, double or triple-barrel application pipettes (Vitrocom), manufactured as described previously (70). Translation of application pipettes was achieved using piezo actuators driven by voltage power supplies. The command voltages were generally low-pass filtered (50–100 Hz, eight-pole Bessel).

Statistics and data analysis

Log-log flow plots of the GFP versus APC signal were fit with a piecewise linear or segmented function with two phases or segments. Data where x < k, were fit with:

$$y=m\ast x+b$$

where x and y are the GFP and APC signal intensities, respectively, m is the slope, and b is the y-intercept signal intensity. In this first phase, the slope, m, was constrained to 0. The free parameter k is the “threshold” and data where x > k were fit with:

$$y=m_2\ast \left(x-k\right)+m\ast x+b$$

where m₂ is a second slope, reflecting the surface trafficking of receptors over the threshold k.

Statistical comparisons between two samples were made using two-tailed heteroscedastic t-tests. Comparisons between many samples across conditions were done using two- and three-way ANOVAs, implemented in Prism (GraphPad). The Benjamini, Krieger and Yekutieli method of multiple comparison testing was used. P-values less than 0.05 were statistically significant. A single transfection for any given condition, run on either a flow cytometer or plate reader, was taken to be a biological replicate or n.

Data availability

All data are available in the main text and/or supporting information.

Supporting information

This article contains supporting information.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.

Reviewed by members of the JBC Editorial Board. Edited by Mike Shipston