The following antibodies were used: mouse IgG1 anti-GluA1 [Neuromab; Cat# 75-327; RRID: AB_2315840; Immunocytochemistry (ICC) 1:200; Immunoblotting (IB) 1:2,000]; mouse IgG2a anti-SynDIG4 (NeuroMab; Cat# 73-409; RRID: AB_2491106; ICC 1:200; IB 1:2,000); mouse IgG2a anti-SynDIG1 (NeuroMab; Cat# 75-251; RRID: AB_10999753; ICC 1:200); rabbit anti- Interferon-induced transmembrane protein 3 (IFITM3) (ProteinTech; ICC 1:200; IB 1:2,000); rat anti-hemagglutinin (HA) (Roche; ICC 1:50; IB 1:1,000); guinea pig anti-vGlut1 (EMD Millipore; ICC 1:500); Alexa 488-conjugated anti-mouse IgG2a (Molecular Probes; ICC 1:200); Alexa 594-conjugated anti-rat (Jackson ImmunoResearch; ICC 1:200); Alexa 555-cross adsorbed anti-mouse IgG1 (Invitrogen; ICC 1:500); Alexa 649-conjugated anti-guinea pig (Jackson ImmunoResearch: ICC 1:500); mouse anti-beta tubulin (MilliporeSigma; Clone: AA2; IB 1:5,000); goat horseradish peroxidase (HRP) conjugated anti-rat (Invitrogen; IB 1:5,000); and goat HRP-conjugated anti-mouse (Invitrogen; IB 1:10,000).

A full length version of rat SD4 coding sequence was amplified by PCR from pHM6 expression vector and subcloned into pRK5 vector backbone provided by our collaborator Dr. Yael Stern-Bach at the Hebrew University of Jerusalem, Jerusalem, Israel. A full length version of mouse SD1 was expressed using a previously generated pHM6 construct (Kalashnikova et al., 2010). pCMV-HA-mIFITM3 was a gift from Howard Hang & Jacob Yount (Addgene plasmid # 58389; http://n2t.net/addgene:58389; RRID: Addgene_58389). Full length mouse IFITM3 was obtained from AddGene (#58389). SD4/IFITM3 chimeras (Table 1) were generated by sequential PCR amplification using megaprimers. Full length wild-type (WT) GluA1 was provided from the Stern-Bach lab and subcloned from the pGEM vector to the pRK5 expression vector. DNA vectors expressing full length GluA2 and GluK2, as well as the GluK2/A1 (Table 2, chimera #2–4) and GluK2/A2 le 3, chimera #1 and 2) chimeras, were additionally provided by the Stern-Bach lab. Additional GluK2/A1 (Table 2, chimeras #1, 5, and 6) and GluK2/A2 (Table 3, chimera #3) constructs were generated by sequential PCR amplification using the megaprimer method. All constructs contain an in-frame HA tag at the N-terminus for detection. Tables 1–3 identify the amino acid (a.a.) sequences of the indicated protein expressed within each chimeric molecule.

COS cells were seeded in 6-well plates at a density of 300,000 cells per well in COS media. Transfection was performed with 2 μg of DNA using Lipofectamine 2000 (Invitrogen). Cells were lysed for protein extraction 24 h after transfection using a standard lysis buffer (150 mM NaCl, 50 mM Tris(hydroxymethyl)aminomethane (TRIS) pH 7.4, 5 mM ethylenediaminetetraacetic acid (EDTA), 1% Triton x-100, 1 mM phenylmethylsulfonyl fluoride (PMSF), and protease inhibitor cocktail). Cells were lifted using a cell scraper and then passed through a 22.5-gauge needle before transferring lysates to 1.5 ml microfuge tubes. Lysates were then transferred to a rotator at 4^°C for 30 min. After 30 min, lysates were centrifuged at 12,000 rpm for 10 min at 4^°C. Supernatant was removed and flash frozen in liquid nitrogen for long term storage. In preparation for immunoblotting, protein samples were thawed on ice. For all blots, 10 μg protein per sample was denatured at 95^°C and loaded onto freshly poured 8% sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE). Gels were run for 90 min at 120 V and transferred to nitrocellulose membrane for 1 h at 100 V. Membranes were blocked in 5% milk diluted in tris-buffered saline with tween-20 (TBST) for 1 h. For testing the expression of AMPAR chimeras, membranes were incubated with both rat anti-HA antibodies and mouse anti-tubulin antibodies at 4^°C overnight. For testing the expression of SD4/IFITM3 chimeras, membranes were incubated with anti-SD4, anti-IFITM3, and anti-tubulin antibodies at 4^°C overnight. Membranes were washed with TBST and incubated in HRP conjugated goat anti-rat and HRP-conjugated goat anti-mouse secondary antibodies for 1 h at room temperature. Luminata Crescendo reagent was added to membrane for the direct detection of HRP signal (Azure Biosystems).

For quantitative analyses, images were taken using either an Olympus FluoView 1000 or Zeiss LSM510 confocal microscope with a 63 × /1.5 NA oil objective with identical settings for laser power, photomultiplier gain, and digital offset. Pinhole (1 AU) and resolution (1,024 × 1,024 pixels) were constant for all images.

Images were analyzed blinded to the experimental condition. Image files were imported into image analysis software (ImageJ) to determine the average size of clusters for each condition. Selected cells were cropped from the original images, saved blinded and subjected to the analysis by an individual not involved in the cell selection and blinding process. The threshold for each independent experiment is determined by averaging the thresholds of at least 25% of images within a dataset. Threshold values were determined by duplicating each image and adjusting the threshold of the duplicated image converted to black and white. Thresholds were determined such that all recognizable puncta were included in the analysis. The average threshold was then applied to all images within a dataset for cluster analysis by inserting values into a pre-written script run through the ImageJ software. The script separates the channels, applies the average threshold values, creates the mask overlay, and analyzes cluster parameters (number and size) defined by the mask. Clusters within the range of 0.1–3.5 μm² were measured. After data collection and the unblinding process, the puncta size of all signals was subjected to statistical analysis. For analysis of puncta size based on co-localization with SD4 (stratification analysis), co-localization was defined as overlap of ≥ 1 pixel. A mask overlay was then created using ImageJ by overlapping the two channels of sGluA1 and SD4. The colocalized puncta in the image representing the receptor coexpressed with SD4 were then used to select unambiguous single puncta manually in the receptor mask overlay. XY coordinates were used to confirm the selected puncta in the receptor mask overlay that corresponded with the colocalized puncta in the image for the receptor coexpressed with SD4. For figure preparation, signals were adjusted for all panels within a figure by using the equal linear adjustments of levels in Photoshop (Adobe Systems).

Data were collected from at least two independent experiments and a minimum n = 10–15 cells per condition per experiment. All graphs and statistical analyses were generated using GraphPad Prism software. Graphs depict the data average and the standard error of the mean (SEM). Statistical significance was assessed by one-way ANOVA with post-hoc Tukey’s test or Student’s t-test. Significance is defined as *p < 0.05; ^**p < 0.01; ^***p < 0.001; and ^****p < 0.0001.

To characterize the relationship between AMPARs and SD4, we used a clustering assay previously established within our lab (Kalashnikova et al., 2010). The full-length AMPAR subunits GluA1 and GluA2, as well as the kainate receptor subunit (GluK2), were expressed in heterologous COS cells either alone or co-expressed with full-length HA-tagged SD4. GluK2 is predicted not to associate with SD4 and served as a negative control. Each receptor subunit contains an N-terminal HA tag for extracellular detection. COS cells were first live-labeled with anti-HA antibodies to examine the distribution of surface expressing GluA1, GluA2, or GluK2. The receptors have extracellular N-termini, while SD4 does not, so only surface GluA1, GluA2, or GluK2 were labeled. After fixation and permeabilization, cells were stained with anti-SD4 antibodies for total SD4. It is important to note that SD4 is a type II transmembrane protein while GluA1 and GluA2 are type I transmembrane proteins; therefore, the HA epitope for the live-labeling of GluA1 or GluA2 is not accessible to HA-SD4 even when the anti-HA antibody is internalized because the HA-tag on SD4 faces the cytoplasm while the anti-HA antibody remains lumenal. Furthermore, the anti-HA antibody used for the surface labeling of HA-GluA1 or HA-GluA2 does not label HA-SD4 when expressed alone (as shown below).

We observed diffuse and even distribution of GluA1, GluA2, and GluK2 when expressed alone. When co-expressed with SD4, a change in the overall distribution of both GluA1 and GluA2 was observed (Figure 1). No difference was observed when GluK2 was co-expressed with SD4, indicating the specificity of SD4 for AMPARs (Figure 1A). Quantification indicates a significant increase in the mean cluster size of GluA1 and GluA2 puncta when co-expressed with SD4 compared with receptor alone (Figure 1B). Although GluK2 puncta are larger at baseline, there is no significant change in puncta size when co-expressed with SD4 (Figure 1B).

We noted a distribution of GluA1 or GluA2 cluster sizes in SD4 co-expressing cells. To determine whether cluster size was related to overlap with SD4, which was not captured in the previous analysis, we stratified populations in co-expressing cells into two groups representing glutamate receptor puncta co-localized with SD4 (w/SD4) or not co-localized with SD4 (w/o SD4). The stratification of GluA1 puncta co-localized with SD4 showed that the mean cluster size is significantly greater when SD4 and GluA1 are co-localized compared to GluA1 expressed alone (Figure 1C). In contrast, the size of puncta not co-localized with SD4 were not significantly different compared with GluA1 alone (Figure 1C). In addition, the mean size of GluA2 clusters is significantly greater when co-localized with SD4, while non-colocalized clusters are not significantly different compared with GluA2 alone (Figure 1D). No significant differences were observed in the size of GluK2 clusters co-localized with SD4 or not co-localized compared with GluK2 alone (Figure 1E). These results provide evidence that the co-expression of SD4 is sufficient to re-distribute and cluster GluA1 and GluA2-containing AMPARs in heterologous cells, and the increased cluster size is dependent on co-localization with SD4.

To identify the region of SD4 sufficient for clustering with GluA1 or GluA2, we generated chimeric proteins by swapping domains between SD4 and the distantly related Dispanin family (Sällman Almén et al., 2012) member IFITM3 with a similar topology (Yount et al., 2010; Ling et al., 2016). One chimera was generated using the intracellular N-terminal region of SD4 and the C-terminal domains of IFITM3, such as the hydrophobic segment that does not span the lipid bilayer, the small intracellular loop, the transmembrane domain, and the short extracellular portion (SD4-NTD/IF-M), and a second chimera was generated using the intracellular N-terminal region of IFITM3 and the corresponding C-terminal domains of SD4 (IF-NTD/SD4-M) (Figure 2A and Table 1). For brevity, we refer to the C-terminal domains as the “membrane associated region.” Constructs were first verified by immunoblot with antibodies that only recognize the N-terminus of their respective proteins (Figure 2B). Therefore, signal is only present when the N-terminus is expressed.

Furthermore, GluA1 was expressed in COS cells either alone, or co-expressed with full-length SD4, SD4-NTD/IF-M, or IF-NTD/SD4-M (Figure 2C). We observed no difference in the mean cluster size of GluA1 populations either co-localized or not co-localized with SD4-NTD/IF-M compared with GluA1 alone (Figure 2D). However, the stratification of GluA1 populations indicated a significant increase in the mean size of GluA1 clusters when co-localized with IF-NTD/SD4-M compared with GluA1 alone (Figure 2E). Next, GluA2 was expressed in COS cells either alone, or co-expressed with either IF/SD4 chimeras (Figure 2F). The stratification of GluA2 populations co-localized or not co-localized with SD4-NTD/IF-M resulted in no change in the mean size of GluA2 clusters (Figure 2G). However, we observed a significant increase in the mean size of GluA2 clusters when co-localized with IF-NTD/SD4-TM compared with GluA2 alone (Figure 2H).

These results indicate that the intracellular proline-rich N-terminal portion of SD4 is dispensable for clustering with GluA1 and GluA2. Furthermore, the C-terminal portion of SD4, which consists primarily of membrane associated regions, is sufficient for clustering with GluA1- and GluA2-containing AMPARs, and that clustering of AMPARs is dependent on co-localization with SD4.

To identify the region of GluA1 that is sufficient for clustering with SD4, we generated chimeric GluA1 proteins using the homologous domains of GluK2 (Table 2). The expressions of all GluK2/GluA1 chimeras were verified by immunoblot (Figure 3A). The chimeras were then transfected and expressed in COS cells either alone or with full-length SD4 (Figures 3B–F). The quantification of mean area of clusters shows an increase in puncta size when SD4 is co-expressed with GluA1, but no significant difference when co-expressed with the chimera expressing the M1-3, S2, M4, and CT domains of GluA1 (Figure 3G). However, the stratification of GluK2/A1/M1-3/S2/M4/CT chimeric puncta (Figure 3B) depicts a significant increase in puncta size when co-localized with SD4 (Figure 3H). Therefore, we conclude that the N-terminus of GluA1 is dispensable for clustering with SD4, and cluster size is dependent on co-localization with SD4.

We next generated GluA2 chimeras using the homologous domains of GluK2 (Table 3). An expression of GluA2 chimeras was verified by immunoblot (Figure 4A). All chimeras were transfected in COS cells either alone or with full-length SD4 (Figures 4B–D). We found that only the chimera expressing the M1-3, S2, M4, and CT domains of GluA2 resulted in an altered distribution when co-expressed with SD4 (Figure 4B). Additionally, the GluK2/A2/M1-3/S2/M4 chimera, where the GluA2-CT domain was not present, resulted in a loss of the clustering phenotype (Figure 4C). Therefore, these experiments show the importance of GluA2 C-terminal domain for clustering with SD4. The quantification of mean area of clusters shows a significant increase in cluster size when SD4 is co-expressed with either full-length GluA2 or the GluK2/A2/M1-3/S2/M4/CT chimera (Figure 4E). Furthermore, the stratification of puncta from Figure 4B shows a significant increase in puncta area only when co-localized with SD4 (Figure 4F). Therefore, we conclude that the N-terminus of GluA2 is dispensable for clustering with SD4, and cluster size is dependent on co-localization with SD4.

For comparison, we sought to identify a region of AMPAR necessary for clustering with the SD4-related family member SD1. Full-length GluA2 had previously been observed to cluster with SD1 (Kalashnikova et al., 2010); therefore, GluA2 and GluK2 were used as positive and negative controls, respectively. We observed altered distribution of GluA2, but not GluK2, when co-expressed with SD1 (Figure 5A). Next, we co-expressed two of the key GluA2 chimeras from Figure 4. Similar to SD4, we observed clustering when the GluK2/A2/M1-3/S2/M4/CT chimera was co-expressed with SD1. However, in contrast to SD4, we observed clustering when the GluK2/A2/M4/CT chimera was co-expressed with SD1 (Figure 5B). Quantification depicts a significant increase in the mean area of clusters when SD1 is co-expressed with GluK2/A2/M1-3/S2/M4/CT and GluK2/A2/M4/CT, but not with GluK2 (Figure 5C). Last, the stratification of M4 and CT chimera puncta co-localized with SD1 results in a significant increase in the mean area of clusters (Figure 5D). Therefore, the M4 and CT of GluA2 is sufficient for clustering with SD1.

Next, we were interested in whether the co-expression of AMPARs with SD4 results in a reciprocal increase in the cluster size of SD4. For these experiments, SD4 was expressed in COS cells either alone, or co-expressed with either GluA1, GluA2, or GluK2 (Figure 6A). The stratification of SD4 puncta co-localized with GluA1 results in a significant increase in the mean cluster size of SD4 puncta co-localized with GluA1 compared with SD4 alone (Figure 6B). Additionally, the mean cluster size of SD4 puncta co-localized with GluA2 is also significantly increased (Figure 6C). We observed no difference in the mean cluster size of SD4 puncta whether co-localized or not-colocalized with GluK2 compared with SD4 alone (Figure 6D). We conclude that not only does the co-localization of SD4 with AMPARs increase the mean cluster size of the receptor, but colocalization with AMPARs also significantly increase the cluster size of SD4.

A clustering of GluA2 by SD1 requires a 37^°C incubation after the surface labeling at 4^°C (Kalashnikova et al., 2010). This observation suggests that a biological process, such as endocytosis is necessary for clustering by SD1. Then, we were interested to determine whether SD4 clustering of AMPARs is temperature dependent. For these experiments, duplicate plates of COS cells expressing GluK2, GluA1, or GluA1 either alone or co-expressed with SD4 were prepared. For the surface labeling of receptors, all plates were incubated at 4^°C. Next, one plate was transferred to a 37^°C incubator, while the other plate remained at 4^°C. After incubation, coverslips were fixed and imaged. We observed that incubation at 4^°C does not result in a change of distribution for any of the receptors, either alone or co-expressed with SD4 (Figure 7A). Quantification indicates no significant increase in the mean area of clusters after the 4^°C incubation (Figure 7B). However, incubation at 37^°C resulted in the clustering of receptors as expected (Figure 7C). Quantification shows a significant increase in the mean area of clusters when either GluA1 or GluA2 are co-expressed with SD4, but not GluK2 (Figure 7D). Therefore, we conclude that the endocytosis of surface labeled AMPARs is most likely to be necessary for clustering with SD4.

To test the role of SD4-dependent AMPAR clustering in synaptic plasticity, we utilized the primary culture of dissociated rat hippocampal neurons. Neurons were treated with 200 μM glycine or vehicle in aCSF without magnesium at 37^°C for 5 min to induce chemical-LTP. Neurons were then transferred to aCSF recovery buffer for 20 min and then stained with anti-GluA1 antibody at 37^°C for 1 h. Neurons were then fixed, permeabilized, and stained for total SD4 and vGlut1 (Figure 8A; three representative stretches from three individual neurons are shown). The percentage of synaptic GluA1 puncta (defined as overlap with the pre-synaptic marker vGlut1) increased by 2-fold after glycine treatment compared with vehicle (Figure 8B), indicating successful chemical-LTP induction. Additionally, the percentage of synaptic GluA1 puncta co-localized with SD4 significantly increased after chemical-LTP (Figure 8B), suggesting that at least a portion of SD4 redistributes to the synapse during glycine induced chemical-LTP. The average number of vGlut1 puncta per stretch was equivalent (vehicle: 27.02 ± 2.39, n = 42; glycine: 25.27 ± 1.73, n = 45; p = 0.553), indicating that the increased percentage of synaptic GluA1 which colocalizes with SD4 is not a result of changes in vGlut1.

Next, we looked at changes in the mean area of GluA1 clusters as a result of chemical-LTP. We stratified these data to determine whether changes in the size of these clusters are dependent on co-localization with SD4. We observed that there was no change in the mean area of GluA1 clusters not colocalized with SD4 after treatment with glycine compared with vehicle (Figure 8C). Interestingly, there was a significant increase in the size of GluA1 clusters during vehicle treatment only when GluA1 was co-localized with SD4. Furthermore, we observed a significant increase in the mean area of GluA1 clusters co-localized with SD4 after glycine induced chemical-LTP (Figure 8C). As a result of these experiments, we conclude that the co-localization of SD4 with synaptic GluA1 increases after chemical-LTP, and this co-localization results in an increase in the mean area of GluA1 clusters.