PKA-GluA1 coupling via AKAP5 controls AMPA receptor phosphorylation and cell-surface targeting during bidirectional homeostatic plasticity

Summary

Bidirectional synaptic plasticity occurs locally at individual synapses during LTP or LTD, or globally during homeostatic scaling. LTP, LTD, and homeostatic scaling alter synaptic strength through changes in post-synaptic AMPARs, suggesting the existence of overlapping molecular mechanisms. Phosphorylation is critical for controlling AMPAR trafficking during LTP/LTD. Here we addressed the role of AMPAR phosphorylation during homeostatic scaling. We observed bidirectional changes of the levels of phosphorylated GluA1 S845, during scaling, resulting from a loss of PKA from synapses during scaling-down and enhanced activity of PKA in synapses during scaling-up. Altered synaptic PKA signaling, requiring the scaffold AKAP5, alters the effectiveness of neuromodulators and NMDAR activation. Increased phosphorylation of S845 drove scaling-up while a knock-in mutation of S845 blocked scaling-up. Finally we show that AMPARs scale differentially based on their phosphorylation status at S845. These results show that rearrangement in PKA signaling controls AMPAR phosphorylation and surface targeting during homeostatic plasticity.

Introduction

In response to different types of stimuli, synapses of the central nervous system have the ability to change their strength in a bidirectional manner, a phenomenon known as synaptic plasticity. The most well studied forms of synaptic plasticity are long-term potentiation (LTP) and long-term depression (LTD), collectively referred to as Hebbian plasticity. LTP and LTD occur at individual synapses, thus altering the strength of affected synapses relative to nearby unaffected synapses (Malenka and Bear, 2004). It is widely speculated that the changes in relative synaptic strength via Hebbian plasticity form the molecular and cellular basis of learning and memory. Synapses can also undergo a distinct type of plasticity known as homeostatic scaling, during which many or all synapses on a given neuron simultaneously change in synaptic strength in a uniform direction (O'Brien et al., 1998; Turrigiano et al., 1998). Unlike Hebbian plasticity, homeostatic scaling alters the strength of all synapses proportionally (Turrigiano et al., 1998), thus protecting the relative differences in synapse strengths. However, homeostatic plasticity has also been shown to occur at individual synapses (Beique et al., 2011; Lee et al., 2010b). By engaging homeostatic scaling, neurons are able to adjust their own firing rates towards an ideal set point without disrupting differences in synaptic weights that store information. In this way, homeostatic scaling may function to maintain network stability and promote learning and memory by offsetting the destabilizing effects of continued LTP or LTD (Turrigiano, 2008). However, at present the relationship between Hebbian and homeostatic plasticity is not clear (Arendt et al., 2013) and how scaling is able to proceed without disrupting or erasing the information from previous Hebbian plasticity events is unknown. Furthermore, it is unclear how global and local plasticity simultaneously occur to allow network stability and ongoing learning and memory formation. LTP, LTD and homeostatic scaling each alter synaptic strength in large part by altering the abundance of AMPA-type glutamate receptors (AMPARs) in the post-synaptic membrane (Huganir and Nicoll, 2013; Kessels and Malinow, 2009; Malenka and Bear, 2004; O'Brien et al., 1998; Shepherd and Huganir, 2007; Turrigiano, 2008). The shared molecular output of these different plasticity types strongly suggests that local and global plasticity cannot occur independently, but rather that the two plasticity types will necessarily have elaborate cross-talk or feed-back.

Tetrameric AMPARs are made from the subunits GluA1-4, and mediate the majority of fast excitatory synaptic transmission in the central nervous system. The majority of AMPARs in the hippocampus and cortex are composed from GluA1/2 and GluA2/3 subunit combinations (Lu et al., 2009; Wenthold et al., 1996). Phosphorylation of AMPAR cytoplasmic C-terminal tails has been shown to have a prominent role in controlling AMPAR synaptic targeting, as well as channel properties (Shepherd and Huganir, 2007). During the induction and maintenance of LTP and LTD, it has been clearly demonstrated that changes in AMPAR phosphorylation occur, and that these phosphorylation sites regulate LTP and LTD (Lee et al., 2000; Lee et al., 1998; Lee et al., 2003). In particular, phosphorylation sites including the PKA-target GluA1 S845 (Roche et al., 1996), the CaMKII/PKC-target GluA1 S831 (Mammen et al., 1997), and the PKC-target GluA2 S880 (Matsuda et al., 1999; Chung et al., 2000) have been well-characterized. PKA-mediated phosphorylation of GluA1 S845 has been shown to promote GluA1 cell-surface insertion and synaptic retention, increase channel open-probability, and facilitate the induction of LTP (Banke et al., 2000; Ehlers, 2000; Esteban et al., 2003; Lee et al., 2003; Man et al., 2007; Oh et al., 2006), while dephosphorylation of GluA1 S845 is associated with endocytosis and LTD (Ehlers, 2000; Lee et al., 2000; Lee et al., 2003; Man et al., 2007). CaMKII-mediated phosphorylation of GluA1 S831 increases channel conductance and regulates LTP (Derkach et al., 1999; Kristensen et al., 2011; Lee et al., 2000; Lee et al., 2003). Finally, PKC-mediated phosphorylation of GluA2 S880 disrupts the interaction between GluA2 and GRIP, allowing for AMPAR endocytosis and LTD (Chung et al., 2000; Seidenman et al., 2003; Steinberg et al., 2006). Whether changes in AMPAR phosphorylation also occur during homeostatic scaling has not been fully addressed nor is it clear whether AMPAR phosphorylation is necessary for homeostatic scaling to occur. It is possible that AMPARs may scale differentially depending on the patterns of phosphorylation that occurred during earlier LTP or LTD events. In this way AMPAR phosphorylation may form part of the molecular code to coordinate both global and local synaptic plasticity.

Levels of protein phosphorylation are governed by the opposing activities of protein kinases and phosphatases. Scaffold proteins, such as the A-Kinase Anchor Proteins (AKAPs), can serve as signaling hubs that link second messenger systems, kinases, and phosphatases with particular targets. AKAP5 (also called AKAP79 in humans or AKAP150 in rodents) is a prominent synapse-targeted AKAP that links adenyl-cyclase, protein-kinase A (PKA), protein-kinase C (PKC), and Ca⁺⁺-dependent protein phosphatase calcineurin/PP2B (CaN) with AMPARs (Colledge et al., 2000; Sanderson and Dell'Acqua, 2011; Tavalin et al., 2002). In the current study we show that changes in the coupling of AMPARs with PKA occur via the scaffold AKAP5. Reduction in AMPAR-PKA coupling during scaling-down results in a decrease in the level of phosphorylated AMPARs and a reduced ability to phosphorylate AMPARs upon neuromodulator stimulation. During scaling-up, overall PKA activity is low, but active PKA is concentrated into dendritic spines, resulting in AMPAR S845 phosphorylation and an increased ability to phosphorylate AMPARs upon neuromodulator stimulation. This reorganization of PKA signaling requires the scaffold AKAP5. A GluA1 S845A knock-in mutation that eliminated phosphorylation, abolished scaling-up, while enhancement of S845 phosphorylation can drive scaling-up. Further, we find that phospho-S845 receptors are resistant to scaling-down, showing that AMPARs scale differentially based on their phosphorylation status. Finally, using a glycine- based chemical LTP protocol, we show that rearrangements of the PKA signaling pathway following homeostatic scaling alter the ability of cortical neurons to express LTP. Thus, we demonstrate that, like Hebbian plasticity, homeostatic scaling involves changes in AMPAR phosphorylation, but utilizes distinct mechanisms, and thus interacts with Hebbian synaptic plasticity.

Results

Changes in AMPAR phosphorylation during homeostatic scaling

We first examined AMPAR phosphorylation during homeostatic scaling. Cultured rat cortical neurons were treated with bicuculline or TTX for 24 or 48hrs, treatments previously shown to induce homeostatic scaling-down and scaling-up, respectively (O'Brien et al., 1998; Turrigiano et al., 1998). To monitor the progress of scaling, neurons were surface biotinylated and surface AMPARs were detected by Western blot. Bicuculline treatment resulted in a loss of surface GluA1, 2 and 3 after 48hrs, and to a lesser degree after 24hrs. TTX treatment resulted in increased surface GluA1 and GluA2 only after 48hrs of treatment (Figure 1A, B). Thus, bidirectional homeostatic scaling is clearly induced after 48hrs in our system. Interestingly, TTX treatment did not result in an increase in surface GluA3, even after 48hrs, showing that scaling-up likely involves GluA1/2 and not GluA2/3 receptor subunit combinations (Figure 1A, B). Using phospho-specific antibodies we examined the levels of phospho-S831 and S845 on GluA1, and phospho-S880 on GluA2, well-characterized sites previously shown to be involved in Hebbian plasticity (Shepherd and Huganir, 2007). Bicuculline treatment for 48hrs resulted in a significant decrease in AMPA receptor phosphorylation at all three sites (Figure 1A, C), with the phospho-S845 levels showing the most prominent decrease. TTX treatment for 48hrs resulted in a significant increase in phospho-S845 levels, but not phospho-S831 or phospho-S880 levels (Figure 1A, C). This pattern of S845 phosphorylation was not observed following acute (10min) bicuculline or TTX treatment (Figure S1A, B). Therefore, changes in AMPAR phosphorylation occur during homeostatic scaling.

Figure 1.

Levels of phosphorylated AMPAR change during homeostatic scaling. (A) Cortical neurons (13DIV) were treated 24 or 48hrs with control media (Con), bicuculline (Bic, 20µM) or tetrodotoxin (TTX, 1µM), followed by surface biotinylation and Western blot. (B) Quantification of cell-surface levels of the AMPAR subunits GluA1, GluA2 or GluA3. (C) Quantification of total levels of phosphorylated GluA1-S845 or S831 or GluA2-S880. Values relative to control, indicated by the dashed line. * and ** (p<0.05 and p<0.01 respectively). Error bars indicate ±sem. N=4–6 independent experiments. See also figure S1.

Reorganization of PKA signaling during homeostatic scaling

As the PKA-target GluA1 S845 showed the most prominent changes in phosphorylation during homeostatic scaling we next addressed the molecular basis for changes in AMPAR phosphorylation by PKA. Following a 48hr treatment with bicuculline or TTX, we first examined protein levels of PKA, calcineurin (CaN), AKAP5, and the β2adrenergic receptor, as these proteins are all known to regulate synapse targeting of GluA1 and phosphorylation of S845 (Hu et al., 2007; Joiner et al., 2010; Jurado et al., 2010; Zhang et al., 2013). We observed no change in the levels of the PKA catalytic subunit or the β2adrenergic receptor following scaling. After bicuculline treatment to induce scaling-down, there were small but significant decreases in the levels of the catalytic subunit of CaN and AKAP5 (Figure 2A, B). Decreased CaN phosphatase upon bicuculline treatment would be expected to result in increased GluA1 S845 phosphorylation, contrary to our observations. On the other hand, AKAP5 is known to anchor PKA with the GluA1 AMPAR subunit (Sanderson and Dell'Acqua, 2011). Therefore, the decrease in AKAP5 protein following bicuculline treatment suggested that a loss of coupling between PKA and GluA1 may be responsible for decreased S845 phosphorylation during scaling-down.

Figure 2.

Reorganization of synaptic PKA signaling during scaling. (A–B) Cortical neurons (13-14DIV) were treated with bicuculline/TTX for 48hrs and indicated molecules were quantified by Western blot, N=5. (C) Cortical neurons (13-14DIV) were treated with bicuculline/TTX for 48hrs followed by isoproterenol (Iso) 10nM for 5min. (D) Changes in pS845 were quantified by Western blot. Iso-induced increases in pS845 were reduced by bicuculline and enhanced by TTX treatment, N=6. (E–H) Hippocampal neurons (14DIV) transfected with GFP and treated with bicuculline or TTX. The localization of PKA catalytic subunit (E) or AKAP5 (G) to dendritic spines was observed and quantified (F) and (H) respectively, N=50–75 spines each from 11–17 neurons. Scale bar 5µm. PKA has strong dendritic shaft localization but is observed in dendritic spines. Following bicuculline treatment PKA moves out of spines. (I) AKAP5 was immunoprecipitated from bicuculline/TTX treated cortical neurons (13-14DIV). Co- immunoprecipitation of PKA is reduced following bicuculline treatment, N=3. * and ** (p<0.05 and p<0.01 respectively). Error bars indicate ±sem. See also figures S1 and S3.

To address this possibility, we treated neurons with bicuculline or TTX for 48hrs to induce scaling, and then challenged the neurons for 5 minutes with isoproterenol (10nM) a noradrenaline analogue known to cause cAMP/PKA-mediated phosphorylation of GluA1 S845 through the β2adrenergic receptor (Hu et al., 2007; Joiner et al., 2010). As expected, isoproterenol challenge resulted in a clear increase in phospho-S845 levels in previously untreated neurons. Interestingly phosphorylation of S845 following isoproterenol challenge was significantly reduced in neurons pre-treated with bicuculline and increased in neurons pre-treated with TTX (Figure 2C, D). As the protein levels of the β2adrenergic receptor were not altered, we believe these observations are the result of suppression or enhancement of PKA activity towards GluA1 during scaling down or up, respectively. Indeed, this phosphorylation pattern was not observed at non-synaptic PKA targets such as CREB (S133) or inhibitor protein 1 (I-1, T35) upon isoproterenol treatment (Figure S1C, D) or at the PKC target GluA2 S880 upon PKC activation with phorbol ester PMA (Figure S1E, F). In addition, acute treatment with bicuculline or TTX did not affect isoproterenol-induced phosphorylation of S845 (Figure S1A, B). Therefore, not only are the levels of phosphorylated GluA1 S845 altered during homeostatic scaling but the ability of PKA to phosphorylate GluA1 in response to stimuli is altered. These results suggest the specific rearrangements of PKA signaling at the synapse during scaling.

We also examined the cellular distribution and association of PKA and AKAP5 during homeostatic scaling using immunofluorescence microscopy and co-immunoprecipitation (Co-IP) respectively. The PKA catalytic subunit showed prominent dendritic shaft targeting (Zhong et al., 2009), but nonetheless could also be detected in dendritic spines, shown using a GFP cell-fill or overlap with the synapse marker PSD95. We determined a spine enrichment value by measuring the staining intensity of PKA in the dendritic spines vs. the dendritic shaft. We observed a significant decrease in the localization of PKA to dendritic spines following 48hr bicuculline treatment, but no change following TTX treatment (Figure 2E, F). AKAP5 was more enriched in dendritic spines, and this distribution was not altered during homeostatic scaling (Figure 2G, H). The removal of PKA, but not AKAP5, from dendritic spines during scaling-down suggests that the association between the two proteins may be decreased. Indeed, we detected a decrease in the co-immunoprecipitation of PKA and AKAP5 following bicuculline treatment (Figure 2I).

In parallel, we were also interested in the overall level and distribution of PKA activity during homeostatic scaling. To address this we performed Western blot on control, bicuculline, or TTX treated samples using an antibody raised against a consensus PKA phospho-substrate. We found that the overall Western blot signal was not altered in bicuculline treated samples compared to untreated samples, but was significantly reduced in TTX treated samples (Figure 3A, B). No significant changes were seen under these conditions using a similar PKC phospho-substrate antibody (Figure 3A, B). We confirmed that these antibodies were phospho-specific (Figure S2). Next, we examined the dendritic distribution of PKA activity during homeostatic scaling using a previously developed fluorescence-resonance energy transfer (FRET) sensor, A-Kinase Activity Reporter, AKAR4 (Depry et al., 2011; Zhang and Allen, 2007). This reporter contains a PKA target site, and upon phosphorylation undergoes a conformational shift resulting in an increased FRET signal. Using this FRET reporter we found PKA activity distributed equally in dendritic spines and dendritic shaft in control neurons. Following bicuculline treatment, there was a significant reduction in PKA activity in dendritic spines relative to dendritic shaft. In contrast, following TTX treatment the activity of PKA in dendritic spines was greatly increased over that in dendritic shafts (Figure 3C, D). Finally, we purified post-synaptic densities (PSD) from bicuculline or TTX treated neurons (Figure 3E) and examined PKA activity in the PSD using the PKA substrate antibody. Unlike the pattern observed in whole cell lysate, PKA activity in the PSD was increased following TTX treatment (Figure 3F, G). Furthermore, we observed that the scaffold AKAP5 showed a significant increase in PSD association following TTX treatment (Figure 3F, H). As AKAP5 targeting to dendritic spines is not altered during scaling (Figure 2G, H) it is likely that increased AKAP5 in the PSD is supplied from a peri-synaptic pool of AKAP5 molecules. Overall, these findings show that during scaling-down PKA activity is high, but there is a loss of PKA protein and activity from dendritic spines, and a decrease in PKA phosphorylation of GluA1. During scaling-up, overall PKA activity is low, but active-PKA is enriched in dendritic spines, allowing increased phosphorylation of GluA1 S845 (summarized in Figure S3). These findings further strengthen our hypothesis that PKA signaling, and the access of PKA to its substrate GluA1, is altered during homeostatic scaling and also explain why isoproterenol-induced AMPAR phosphorylation was altered following scaling.

Figure 3.

Levels and cellular distribution of PKA activity during scaling. (A) Cortical neurons (13-14DIV) were treated with bicuculline/TTX for 48hrs. Global PKA or PKC activity was assessed by Western blot of whole cell lysates using pan phospho-substrate antibodies. (B) Quantification of pan phospho-substrate blots indicates a significant reduction in PKA activity in whole cell lysate after TTX treatment, N=6. (C) PKA activity was visualized using a FRET-based A-Kinase Activity Reporter, AKAR4, transfected into cortical neurons (13-15DIV) treated with bicuculline/TTX for 48hrs. The false color scale is indicated. Scale bar (5µm). (D) FRET spine enrichment value, N=50–70 spines each from 9–18 neurons. (E) PSD preparations from cultured cortical neurons. Homogenate (Homo.), cytosol fraction (S2), membrane fraction (P2), PSD fraction (PSD). (F) Western blot for PKA phospho-substrate or AKAP5 in PSD samples from control, bicuculline or TTX treated neurons. (G) PKA substrate blot indicates an increase in PKA activity in the PSD following TTX treatment, N=4. (H) AKAP5 in the PSD was significantly increased following TTX treatment, N=4. * and ** (p<0.05 and p<0.01 respectively). Error bars indicate ±sem. See also figures S2 and S3.

AKAP5 is required for homeostatic scaling-up

Our results point towards a prominent role for AKAP5 in shaping signaling events towards AMPARs during homeostatic scaling. To directly address the role of AKAP5 we transfected neurons with a vector control or with an shRNA against rat AKAP5 (shAKAP5) that effectively reduces its expression (Figure S4A). In vector-transfected neurons treated with bicuculline/TTX, we observed the same changes in surface AMPAR subunits and phosphorylation observed in non-transfected cells in Figure 1 (Figure 4A – C). Neurons transfected with shAKAP5 showed reduced levels of AKAP5 protein, but not of the PKA catalytic subunit. In addition, AKAP5 knock-down caused a significant reduction in surface levels of all AMPAR subunits, and reduced phosphorylation at GluA1 S845 and S831 and GluA2 S880. This is consistent with the ability of AKAP5 to anchor both PKA and PKC (Sanderson and Dell'Acqua, 2011). Further decreases in AMPAR phosphorylation and surface levels were still observed in shAKAP5 transfected neurons following bicuculline treatment, suggesting that knock-down of AKAP5 partly mimics scaling-down. However, increased GluA1 phospho-S845 and surface GluA1 and GluA2 normally observed upon TTX treatment were completely blocked in AKAP5 knock-down neurons (Figure 4A–C). Surface GluA1 antibody labeling and fluorescence microscopy confirmed that AKAP5 knock-down blocked scaling-up and partly mimicked scaling-down (Figure 4D, E). Basal levels of surface GluA1 and bidirectional scaling was restored by co-transfection with shRNA-resistant human AKAP5 (Figure 4D, E). Together, these results show that AKAP5 is required for TTX-induced scaling-up and that loss of AKAP5 partially occluded bicuculline induced scaling-down. Further, we found that, like bicuculline-induced scaling-down, AKAP5 knock-down impaired PKA phosphorylation of GluA1 S845 following isoproterenol treatment (Figure S4B, C), consistent with previous results (Zhang et al., 2013).

Figure 4.

AKAP5 knock-down partly occludes scaling-down and blocks scaling-up. (A–C) Cortical neurons (13-14DIV) electroporated with control vector (pSuper) or shRNA against rat AKAP5 (shAKAP5), and treated with bicuculline/TTX for 48hrs followed by surface biotinylation and Western blot. Surface levels of GluA1, GluA2 or GluA3 (B), or total levels of phosphorylated GluA1 S845 or S831, or GluA2 S880 (C) were quantified, N=5. Data are normalized and compared statistically to untreated vector transfected controls. (D) Cortical neurons (13-14DIV) were co-transfected with GFP together with control vector, shAKAP5, or shAKAP5/human myctagged AKAP5 (sh+rescue) and treated with bicuculline/TTX for 48hrs. Surface GluA1 was labeled with an anti-N-terminal antibody followed by immunofluorescence microscopy. GFP (green), surface GluA1 (magenta). Scale bar 20µm, inset 5µm. (E) Surface GluA1 signal from transfected cells was quantified, N=approximately 10 dendrite segments each from 20–35 transfected neurons. Data are normalized and compared statistically to untreated vector transfected controls. * and ** (p<0.05 and p<0.01 respectively). Error bars indicate ±sem. See also figure S4.

GluA1 phospho-S845 is required for scaling-up

Next, we performed scaling treatments followed by cell-surface biotinylation and Western blot, mini-EPSC recording, or immunofluorescence microscopy using wild-type or phospho-deficient transgenic knock-in mice in which key serine residues of GluA1 are mutated to alanine (S831A or S845A)(Lee et al., 2010a). In cultured cortical neurons from wild-type mice treated with bicuculline or TTX, we observed a bidirectional change in surface GluA1 and GluA2, as well as phospho-S845 and decreased surface GluA3 and phospho-S831 and phospho-S880 in response to bicuculline (Figure 5A, B), similar to rat neurons (Figure 1). Bidirectional scaling of surface GluA1 and GluA2 was normal in S831A mutant neurons, but TTX-induced scaling-up was completely eliminated in the S845A mutant neurons (Figure 5A, B). Bicuculline-induced scaling-down of surface GluA3 was not different between genotypes (Figure 5A, B). Bicuculline or TTX treatment resulted in changes in total AMPAR expression, in agreement with previous findings (Anggono et al., 2011), but these alterations in total AMPAR expression were not affected by mutation of S831 or S845 (Figure S5A). Using mEPSC recording, we observed a bidirectional change in mini amplitude in response to bicuculline or TTX in wild-type neurons. TTX treatment resulted in a significant increase in mini amplitude in S831A neurons while in the S845A neurons the TTX increase in mini amplitude was eliminated (Figure 5C, D). Both S831A and S845A neurons showed a clear trend towards decreased mini amplitude in response to bicuculline, but this was not statistically significant (Figure 5C, D). No changes were seen in mEPSC frequency or decay kinetics (Figure S5B, C). Using surface GluA1 immunostaining we observed bidirectional changes in surface GluA1 in WT and S831A neurons in response to bicuculline or TTX treatment, while in S845A neurons scaling-up was completely blocked (Figure 5E, F). Thus, the S845 phosphorylation site of GluA1 is required for homeostatic scaling up. There were no changes in the total or surface levels of GluA1, 2 or 3 between wild-type, S831A and S845A neurons in the absence of scaling induction (Figure S5D, E). Further, mutation of S831 or S845 did not affect the basal phosphorylation levels of the other sites (Figure S5D, E).

Figure 5.

S845 of GluA1 is required for TTX-induced scaling up. (A–F) Mouse cortical neurons (14-15DIV) cultured from wild-type (WT) or S831A or S845A GluA1 knock-in mice were treated for 48hrs with bicuculline/TTX. (A-B) Surface-biotinylation and Western blot. Scaling up is blocked in S845A neurons, N=7. Data from each genotype are normalized to untreated controls. (C) mEPSC recordings from WT, S831A or S845A cortical neurons following 48hrs of drug treatment. (D) Quantification of mEPSC amplitude, N=5–11. Data from each genotype are normalized to untreated controls. (E) Surface GluA1 was labeled using an anti-N terminal antibody under non-permeabilized conditions following 48hrs of drug treatment. Scale bar 20µm, inset 5µm. (F) Surface GluA1 levels were quantified on 10–15 dendritic segments each from 20–24 neurons. Data from each genotype are normalized to untreated controls. * and ** (p<0.05 and p<0.01 respectively). ns, no statistical significance. Error bars indicate ± sem. See also figure S5.

Increasing phospho-S845 promotes GluA1 surface targeting and scaling-up

In order to test how manipulations of phospho-S845 would affect scaling neurons were treated with bicuculline/TTX in combination with forskolin/rolipram (FR: 2.5µM/100nM, 48hrs) to activate PKA or FK506 (2.5µM, 48hrs) to inhibit CaN, manipulations expected to increase the levels of phospho-S845. FR treatment increased the phosphorylation of GluA1 S845 as expected for this PKA target (Figure 6A, B). Moreover, FK506 treatment significantly increased the levels of GluA1 phospho-S845, confirming that CaN is a phosphatase that controls basal levels of phospho-S845 (Colledge et al., 2000; Tavalin et al., 2002) (Figure 6A, B). Interestingly, under these conditions we found that FR or FK506 treatment alone resulted in increased surface targeting of GluA1 that completely occluded the effect of TTX (Figure 6A, B). Bicuculline-induced scaling-down in the presence of FR or FK506 seemed to be reduced compared to bicuculline-only treatment but was still clearly occurring (Figure 6A, B). Again we observed that bicuculline treatment alone caused a loss of phosphorylation at S845. However, when combined with bicuculline, FR treatment prevented the loss of phospho-S845 and maintained this phosphorylation at a high level. Interestingly, during bicuculline treatment, FK506 did not prevent the loss of phosphorylation at GluA1 S845 suggesting that the loss of phospho-S845 is not due to activation of CaN but likely from a reduction in PKA activity towards GluA1 (Figure 6A, B). Treatment of wild-type mouse neurons with FR resulted in increased GluA1 surface targeting but this was not observed in phospho-deficient GluA1 S845A mouse neurons indicating FR’s effect on GluA1 surface expression was mediated by S845 phosphorylation (Figure S5F, G). Thus, treatments leading to phosphorylation of GluA1 S845, either kinase activation or phosphatase inhibition, promote GluA1 surface targeting.

Figure 6.

PKA activity drives scaling-up. (A–F) Cultured cortical neurons (13-14DIV) were treated for 48 hours with control media or combinations of bicuculline (20µM), TTX (1µM), forskolin + rolipram (FR, 2.5µM/100nM), or FK506 (2.5µM). (A) Drug treatment was followed by surface biotinylation and Western blot. (B) Quantification of surface GluA1 or total phospho-S845. Treatment with FR or FK506 increased GluA1 surface levels occluding the effects of TTX. FR causes a high level of S845 phosphorylation that is maintained throughout scaling. N=6. (C) Following drug treatments, surface GluA1 (magenta) was labeled followed by total VGlut1 (green). Scale bar 5µm. (D) Surface GluA1 fluorescence intensity was quantified only at areas that overlapped with the synapse marker VGlut1. N=24 fields. FR treatment increased synaptic surface GluA1 to a similar degree as TTX. FR and TTX showed an additive effect. (E) mEPSC recording from cortical neurons following 48hrs of drug treatment. (F) Quantification of mEPSC amplitude. FR treatment increased mEPSC amplitude to a similar degree as TTX. FR and TTX showed an additive effect. N=7–15. * and ** (p<0.05 and p<0.01 respectively). ns, no statistical significance. Error bars indicate ±sem. See also figure S6.

In order to further characterize the effects of PKA activation during scaling surface GluA1 was labeled in neurons treated with bicuculline, TTX, or FR. We again observed that the increase in surface GluA1 induced by TTX was occluded by FR treatment, but that bicuculline-induced scaling-down could still occur (Figure S6A, B). However, when we co-stained surface GluA1 with the excitatory synapse marker VGlut1, we observed that FR and TTX treatment had an additive effect in recruitment of surface GluA1 to synaptic sites (Figure 6C, D). Using electrophysiological recordings, we found that treating neurons with FR or TTX increased mEPSC amplitude similarly, compared to untreated neurons (Figure 6E, F). When FR was combined with TTX we observed a further enhancement of mEPSC amplitude (Figure 6E, F), consistent with enhanced synaptic recruitment of surface GluA1. This finding is in agreement with earlier studies suggesting that PKA activity primarily acts to increase AMPAR surface targeting, and that additional signaling can recruit these receptors into the post-synaptic density (Esteban et al., 2003; Oh et al., 2006). Treatment of FR in combination with bicuculline reduced mEPSC amplitude compared to FR treatment alone (Figure 6E, F). No changes in mEPSC frequency or decay kinetics were observed (Figure S6C, D). Together, these results indicate that elevated PKA activity can promote synaptic strengthening and scaling-up, but that scaling-down can still occur.

Phospho-S845 GluA1 is resistant to scaling-down

In the previous experiments, we found that under conditions of high PKA activity, the loss of phospho-S845 during scaling-down was prevented, but a reduction in surface GluA1 still occurred. This suggests that under certain signaling conditions phospho-S845 AMPARs may be resistant to scaling-down and that non-phosphorylated receptors may be preferentially removed. To further test this possibility, we examined the surface and cellular distribution of GluA1 phospho-S845 during homeostatic scaling. Bicuculline, TTX and FR treated neurons were biotinylated, and surface GluA1 was visualized by Western blot. Blots were then stripped and re-probed with the anti-phospho-S845 antibody. Following FR and bicuculline treatment there was no decrease in phospho-S845 receptors on the cell-surface even while a reduction in overall surface GluA1 was observed (Figure 7A, B), suggesting that phospho-S845 receptors were preferentially retained on the cell-surface while non-phosphorylated receptors were removed. As an alternate strategy we prepared PSD samples from bicuculline, TTX, and FR treated neurons and examined the PSD levels of phospho-S845 receptors. Again we observed that FR treatment greatly increased the level of phospho-S845 in the PSD and that this level was maintained in the presence of bicuculline even though GluA1 levels in the PSD were reduced (Figure 7C, D).

Figure 7.

Phosphorylated S845 AMPARs are resistant to scaling-down. (A–E) Cortical neurons (13-14DIV) were treated with bicuculline (20µM), TTX (1µM), or forskolin/rolipram (FR, 2.5µM/100nM) for 48hrs. (A) Surface biotinylation and Western blot. Blots were probed for surface GluA1 and then re-probed for surface phospho-S845. (B) Quantification of surface phospho-S845. Surface phospho-S845 receptors are maintained at a high level in the presence of FR throughout the scaling process despite the observation that total surface GluA1 levels are reduced in the presence of bicuculline. N=4. (C) PSD preparation from drug treated neurons. PSD samples were probed for GluA1 and phospho-S845 by Western blot. (D) Phospho-S845 receptors are retained in the PSD during scaling under conditions of high PKA activity (FR), N=3. (E) phospho-S845 receptors were localized by comparing their distribution to total GluA1 and VGlut1. Scale bar 5µm. Phospho-S845 receptors are localized to synapses as seen from the triple co-localization and are maintained at the synapse in the presence of bicuculline. Note that phospho-S845 receptors are only visible by immunofluorescence in the presence of FR. * and ** (p<0.05 and p<0.01 respectively). ns, no statistical significance. Error bars indicate ±sem. See also figures S7.

Next we examined the cellular distribution of phospho-S845 receptors. Phospho-S845 receptors could be visualized as punctate structures only when phosphorylation was increased by FR treatment in wild-type, but not in S845A mutant neurons (Figure S7). When we compared the distribution of phospho-S845 punctae in FR treated neurons to the excitatory synapse marker VGlut1 we saw a very clear overlap, suggesting that phospho-S845 receptors are localized at synapses (Figure 7E). When neurons were also treated with bicuculline or TTX, we again observed that the phospho-S845 punctae were synaptic (Figure 7E). Together, these results show that, under conditions of high-PKA activity, scaling-down of GluA1 containing AMPARs can still occur, but that phosphorylated receptors remain on the cell surface and at synapses. Thus, AMPARs scale differentially based on their phosphorylation status.

Homeostatic scaling and LTP

Finally, we tested how the rearrangements in PKA signaling during scaling influenced LTP. Cortical neurons were first treated with bicuculline or TTX to induce homeostatic scaling and then treated with a glycine-based media, well-established to induce a form of chemical LTP (cLTP) (Liao et al., 2001; Lu et al., 2001). Using cell-surface biotinylation and Western blot we observed that glycine treatment in control cells resulted in a significant increase in surface GluA1 as well as phosphorylation of S845. Increases in surface GluA1 and phospho-S845 were completely blocked in neurons pre-treated with bicuculline or occluded in neurons pre-treated with TTX (Figure 8A, B). This result was confirmed using cell-surface immunolabeling and fluorescence microscopy (Figure 8C, D). However, we considered that while scaling blocked/occluded the increase in surface GluA1 by cLTP, pre-existing surface receptors could still become enriched at synapses. To test this possibility we isolated PSD samples from scaled/cLTP treated neurons and examined the synaptic levels of AMPAR subunits by Western blot. In control neurons cLTP resulted in a clear increase in PSD-associated GluA1 and GluA2, but not NMDAR subunit GluN2B (Figure 8E, F). Synaptic GluA3 showed a trend to increase upon cLTP that was not statistically significant. Interestingly, we saw no change in synaptic AMPAR subunits upon cLTP in neurons pre-treated with bicuculline or TTX (Figure 8E, F). As an alternate strategy we immunolabeled surface GluA1 followed by staining of the excitatory synapse marker VGlut1. We then examined surface GluA1 levels only at regions positive for VGlut1 and we again conclude that there was an increase in synaptic GluA1 in control but not in scaled neurons (Figure 8G, H).

Figure 8.

Scaling and chemical LTP. (A–H) Cortical neurons (13-14DIV) were treated with bicuculline (20µM)/TTX (1µM) for 48hrs, followed by chemical LTP (cLTP) using glycine (Gly, 200µM, 5min treatment, 20min chase). (A) Neurons were then surface-biotinylated and lysed for Western blot. (B) Quantification of surface GluA1 and phospho-S845 Western blots. Glycine treatment in control neurons results in increased surface GluA1 and phospho-S845 levels. N=5. Increases are blocked by treatment with bicuculline or occluded by treatment with TTX. (C) Neurons were treated as in A, followed by fixation and surface GluA1 labeling with an anti-N terminal antibody. Scale bar 20µm, inset 5µm. (D) Surface GluA1 levels were quantified on 10–15 dendritic segments each from 28–32 neurons per condition. (E) Following scaling and cLTP treatments cortical neurons were fractionated to obtain PSD samples. PSD material was analyzed by Western blot. (F) cLTP results in increased levels of synaptic GluA1 and GluA2 in control neurons but not in neurons pre-treated with bicuculline or TTX. For each scaling condition, data are presented as glycine treated relative to untreated, indicated by the dashed line. N=3. (G) Surface GluA1 (magenta) and total VGlut1 (green) were labeled following scaling and cLTP. Scale bar 5µm. (H) Surface GluA1 fluorescence intensity was quantified only at areas that overlapped with the synapse marker VGlut1. For each scaling condition data are presented as glycine treated relative to untreated. N=35 fields. * and ** (p<0.05 and p<0.01 respectively). ns, no statistical significance. Error bars indicate ±sem.

Similarly to bicuculline-induced scaling-down, knock-down of AKAP5 also blocked the effect of glycine on surface GluA1 and phospho-S845 (Figure S8A–D). Importantly, expression of human AKAP5 could rescue glycine induced increases in surface GluA1 in AKAP5 knockdown neurons (Figure S8C, D). These findings again demonstrate the importance of GluA1-PKA coupling via AKAP5 in controlling AMPA receptor phosphorylation and surface targeting. Further, these results show that rearrangements of the PKA signaling pathway during homeostatic scaling have important impacts on the ability of cortical neurons to express LTP.

Discussion

Changes in PKA signaling during homeostatic scaling

Previous studies have shown that changes in phosphorylation play a key role in controlling AMPAR synaptic targeting during Hebbian forms of synaptic plasticity (Hu et al., 2007; Lee et al., 2000; Lee et al., 2003; Seol et al., 2007; Steinberg et al., 2006). In the current study, we investigated the role of AMPAR phosphorylation in homeostatic scaling, a form of global synaptic plasticity. We were also interested in the possibility that AMPAR phosphorylation may mediate interactions between local (Hebbian) and global (homeostatic) plasticity. AMPAR phosphorylation is governed by the opposing activities of kinases and phosphatases. The activities of these enzymes towards specific substrates are coordinated by scaffold proteins, such as AKAP5 in the case of AMPARs (Sanderson and Dell'Acqua, 2011). Focusing on three well-characterized phosphorylation sites, we have shown that indeed the level of phosphorylated AMPARs does change during homeostatic scaling. In particular, the level of GluA1 phosphorylated at S845 showed bidirectional changes during homeostatic scaling, and these changes are brought about through changes in the coupling of PKA and GluA1 via the signaling scaffold AKAP5 (Figure 9).

Figure 9.

Model of signaling mechanisms affecting AMPA receptor phosphorylation during scaling, LTP, and LTD. Scaling involves a change in the coupling of basal PKA with AMPA receptors via AKAP5, resulting in increased or decreased AMPA receptor phosphorylation. LTP and LTD involve the activation and recruitment of kinases or phosphatases respectively to increase or decrease AMPA receptor phosphorylation respectively. Note that AKAP5 also participates in LTP and LTD.

During scaling down, PKA becomes uncoupled from the scaffold protein AKAP5 and exits dendritic spines. Even though general PKA activity remains high, this reorganization results in a reduced ability of PKA to access its substrate GluA1, leading to a loss of GluA1 S845 phosphorylation, reduced effectiveness of noradrenaline and a block of cLTP. During scaling-up, general PKA activity is low, but remaining active PKA is enriched in dendritic spines, where it can promote the phosphorylation of GluA1 S845. This rearrangement results in enhanced effectiveness of noradrenaline and an occlusion of cLTP. Both AKAP5 and GluA1 S845 are required for scaling-up, and scaling-down can be partially occluded by knocking-down AKAP5. A recent study found that dark-induced scaling-up in the visual cortex also required GluA1 S845 (Goel et al., 2011). Interestingly, both WT and S845A knock-in neurons showed increased expression of GluA1 after TTX treatment, and yet S845A neurons completely lacked scaling-up of surface or synaptic AMPARs. This finding suggests that an increase in GluA1 expression levels are not sufficient to promote scaling without phosphorylation of S845. The observed changes in GluA1 phospho-S845 during scaling are consistent with earlier literature showing that phosphorylation of GluA1 S845 promotes surface targeting and LTP, and that dephosphorylation promotes GluA1 endocytosis and LTD (Ehlers, 2000; Esteban et al., 2003; Lee et al., 2003; Man et al., 2007; Oh et al., 2006).

AMPAR phosphorylation and global synaptic plasticity

One of the defining features of homeostatic scaling is its multiplicative nature, the observation that all synapses on a given neuron seem to be affected to an equal degree. However, the mechanism for multiplicative scaling remains largely unknown. A prevailing model of the excitatory synapse postulates that the post-synaptic density (PSD) contains “slots” which can be occupied by trafficking AMPARs (Huganir and Nicoll, 2013; Kessels and Malinow, 2009). Synaptic AMPARs undergo continuous turnover as receptors exit the PSD, leaving empty slots behind, and are replaced from pools of surface extra-synaptic receptors. A recent study demonstrated that an increased off-rate of PSD-anchored AMPARs was one key mechanism contributing to synaptic weakening during homeostatic scaling-down (Tatavarty et al., 2013). This report demonstrated that increased AMPAR off-rates would allow for uniform adjustments to synaptic strength enabling multiplicative scaling-down (Tatavarty et al., 2013; Turrigiano, 2008). However, scaling-up appeared to utilize a different mechanism (Tatavarty et al., 2013). Dephosphorylation of GluA1 S845 is believed to destabilize AMPARs on the cell-surface and at the synapse (Ehlers, 2000; Lee et al., 2000; Lee et al., 2003; Man et al., 2007). Uncoupling of AMPARs from PKA favors dephosphorylation and may form the molecular basis of this observed destabilization of synaptic AMPARs during scaling-down. Conversely, enhanced coupling between PKA and GluA1, via AKAP5 during scaling-up may increase the surface pool of receptors that could occupy slots in the PSD, and perhaps increase the slot occupancy in a global fashion, enabling multiplicative scaling-up.

LTP has a strong destabilizing effect on neuronal networks due to its feed-forward nature, and this is speculated to be offset by scaling-down (Turrigiano, 2008). Another predicted function of scaling-down, yet untested, is to increase synaptic signal to noise ratios (Turrigiano, 2008). In this scenario, naïve synapses would be more affected than previously potentiated synapses. Either, smaller naïve synapse could be weakened below a functional threshold, or previously potentiated synapses could be resistant, or even escape scaling-down altogether. We have shown that when PKA is highly active, there is an increase in phosphorylated S845 receptors, and during scaling-down, these remain on the cell surface, at the synapse and within the PSD, despite the overall loss of AMPARs from these locations. These results indicate that non-phosphorylated receptors are preferentially removed and that PKA-phosphorylated receptors are resistant to scaling-down. It is possible that potentiated synapses may maintain a pool of active kinases and a larger pool of scaling-resistant phosphorylated AMPARs, consistent with synaptic tagging (Huang et al., 2006; Sajikumar et al., 2007) Neuromodulators such as noradrenaline that regulate the cAMP/PKA pathway and modulate AMPAR trafficking may also influence scaling (Hu et al., 2007; Joiner et al., 2010; Seol et al., 2007). During global scaling-down, localized innervation of neuromodulators could protect certain synapses from scaling. Therefore, global decreases in kinase-AMPAR coupling, combined with localized PKA signaling during/following LTP or neuromodulator release offer a mechanistic insight into how scaling- down may proceed in a manner sensitive to the previous history of the synapse and achieve an enhanced signal to noise ratio in neural networks.

Homeostatic scaling vs. LTP/LTD

During LTP, kinases such as PKA and CaMKII are activated/recruited to phosphorylate AMPARs, promoting receptor insertion and synaptic potentiation (Huganir and Nicoll, 2013; Lee et al., 2000; Malenka and Bear, 2004). Blocking protein kinases or removing the AMPAR phosphorylation sites can impair LTP (Lee et al., 2000; Lee et al., 2003; Malenka and Bear, 2004). LTD involves the activation or recruitment of phosphatases such as PP1, PP2A and CaN, which can dephosphorylate AMPARs, promoting their removal from the synapse (Figure 9) (Huganir and Nicoll, 2013; Lee et al., 2000; Lee et al., 1998; Malenka and Bear, 2004). In our model, there is no overt activation or deactivation of kinases or phosphatases during scaling, but rather changes in the ability of basal activities of these enzymes to access AMPARs through changes in coupling via AKAP5 (Figure 9). Indeed, overall PKA activity was even supressed during scaling-up. Furthermore, inhibition of CaN with FK506 resulted in increased GluA1 S845 phosphorylation showing that there is basal PKA activity acting on this site and that this basal PKA activity is balanced by CaN phosphatase activity. During LTD AMPAR dephosphorylation and synaptic depression can be blocked by FK506 (Lee et al., 2000). Loss of AMPAR phosphorylation during scaling-down, however, was not prevented by FK506, suggesting that CaN phosphatase is not activated during scaling-down. Therefore, our results show that changes in AMPAR phosphorylation occur during scaling, and Hebbian plasticity. However, homeostatic scaling employs a fundamentally distinct mechanism (Figure 9).

While homeostatic scaling and Hebbian plasticity have important differences in their molecular mechanisms, these plasticity types share substantial overlap as well. An important question in the field is to understand the relationship between local and global plasticity types. We found that cLTP was occluded by scaling-up or blocked by scaling-down, suggesting that adjustments to synaptic weights during homeostatic scaling may come at the cost of the neuron’s capacity for LTP. Loss of PKA from the synapse during scaling-down may explain the block of LTP. Indeed, knock-down or knock-out of AKAP5, which anchors PKA in the synapse (Tunquist et al., 2008; Weisenhaus et al., 2010; Zhang et al., 2013), also blocks LTP, mimicking the effects of scaling-down. It has been speculated that scaling-down may prevent the feedforward effects inherent in LTP (Turrigiano, 2008). Therefore, it makes physiological sense that LTP induction (but not maintenance) should be inhibited during scaling-down. The occlusion of LTP by scaling-up may be explained by a ceiling effect. Increased GluA1 surface targeting and S845 phosphorylation during scaling-up cannot be further enhanced by cLTP.

A recent study has shown that LTP is enhanced in relatively young hippocampal slices previously treated with TTX to induce scaling-up (Arendt et al., 2013). The authors observed that scaling-up with TTX treatment induced the formation of silent synapses and that LTP induction unsilenced these synapses resulting in substantial potentiation (Arendt et al., 2013). Whether, active (AMPAR containing) synapses also underwent potentiation or were already at a ceiling of synaptic strength was not clear. Whether LTP is also enhanced, or in fact occluded in older hippocampal slices following scaling-up will need to be tested (Soares et al., 2013). Another possibility is that the relationship between local and global plasticity may be unique in different brain regions or developmental stages, dictated by the physiological function of specific neuron types. Homeostatic plasticity has also been demonstrated at single synapses (Beique et al., 2011; Lee et al., 2010b). The interaction between single synapse homeostatic plasticity and Hebbian plasticity also requires further investigation.

In conclusion, we have shown that homeostatic scaling involves significant alterations in the PKA signaling pathway with important consequences on AMPAR phosphorylation and synaptic strength. Further, AMPARs scale differentially based on their phosphorylation status which may form the basis of interactions between local and global plasticity types.

Experimental procedures

Chemical LTP and drug treatment

To induce homeostatic scaling neurons were treated (unless otherwise stated) for 48 hours with bicuculline methobromide (20µM) or tetrodotoxin (TTX, 1µM). For some experiments these treatments were used in combination with forskolin/rolipram (2.5µM/100nM), or FK506 (2.5µM). For some experiments, neurons were treated with isoproterenol (10nM) or phorbol-myristoyl-acetate (PMA, 10nM) for 5minutes at the end of the 48hrs scaling treatments or following a 5min pre-treatment with bicuculline or TTX. For chemical LTP experiments, neurons were first pre-incubated for 15min at 37°C in chemLTP buffer (125mM NaCl, 2.5mM KCl, 2mM CaCl₂, 1mM MgCl₂, 33mM glucose, 25mM HEPES pH 7.4, 20µM bicuculline, 500nM TTX, 1 µM strychnine), followed by glycine treatment for 5 min (chemLTP buffer, 200µM glycine, 0 magnesium), and then returned to the original buffer (without glycine) for 20min prior to lysis/fixation. Neurons were used at 13–15DIV for all experiments.

Surface biotinylation

Neurons were rinsed with ice-cold phosphate-buffered saline containing 0.1mM CaCl₂ and 1mM MgCl₂, pH8.0 (PBSCM), incubated in PBSCM containing 0.5mg/mL Sulfo-NHS-SS-biotin (Thermo Scientific, 30 minutes, 4°C) then rinsed in PBSCM and unreacted biotinylation reagent was quenched in PBSCM containing 20mM glycine (2 × 7minutes, 4°C). Cells were lysed in lysis buffer (PBS containing 50mM NaF, 5mM sodium pyrophosphate, 1% NP-40, 0.5% sodium deoxycholate, 0.02% SDS, 1 µM okadaic acid, 1mM Na₃VO₄, and protease inhibitor cocktail (Roche). Protein concentration of each lysate was quantified using Bradford reagent (BioRad) and equal amounts of protein were incubated overnight with NeutrAvidin coupled-agarose beads (Thermo Scientific). Beads were washed three times with ice-cold lysis buffer and biotinylated proteins were eluted with 2×SDS sample buffer. Cell-surface or total proteins were then subjected to SDS-PAGE and analysed by Western blot.

Immunocytochemistry

For surface AMPA receptor labeling, neurons were fixed for 5 minutes at room temperature in PBS containing 4% paraformaldehyde (PFA)/4% sucrose, rinsed with PBS, blocked with 10% goat serum in PBS for 30 minutes, and incubated with mouse anti-GluA1 (anti-N-terminus) antibodies in PBS with 1% goat serum (2 hours). Coverslips were washed with PBS. In some experiments, neurons were permeabilized with 0.1%TX-100 followed by labeling with guinea pig anti-VGlut1 (2 hours). Coverslips were incubated with Alexafluor 568-conjugated goat anti-mouse and Alexafluor 647 goat anti guinea pig secondary antibodies in PBS containing 1% goat serum for 45 minutes. After final washes with PBS coverslips were mounted onto glass slides using Fluoromount-G (Southern Biotech). All steps were performed at room temperature. For all other immunocytochemistry experiments neurons were fixed as above for 10 minutes, followed by PBS rinse and then permeabilized with 0.1%TX-100 in PBS for 10 minutes. Fixed neurons were then blocked as above and incubated with primary antibodies in PBS containing 1% goat serum overnight at 4°C. Neurons were rinsed, incubated with fluorescently labeled secondary antibodies, and mounted onto glass slides as above. Images were obtained using a 510-laser scanning confocal microscope (Zeiss).

Highlights.

• Synaptic PKA activity and phosphorylation of AMPARs is altered during scaling

• AKAP5 and GluA1 S845 are required for homeostatic scaling-up

• AMPA receptors scale differentially based on their phosphorylation state

• The effect of neuromodulators and NMDAR activation is altered during scaling