AMPA receptor subunits define properties of state-dependent synaptic plasticity

Abstract

Many synapses undergo immediate and persistent activity-dependent changes in strength via processes that fall under the umbrella of synaptic plasticity. It is known that this type of synaptic plasticity exhibits an underlying state dependence; that is, as synapses change in strength they move into distinct ‘states’ that are defined by the mechanism and ability to undergo future plasticity. In this study, we have investigated the molecular mechanisms that underlie state-dependent synaptic plasticity. Using intracellular application of peptides that mimic the C-terminal tail sequences of GluR1 and GluR2 AMPA receptor subtypes, combined with paired recordings of minimal synaptic connections, we have shown that AMPA receptor subtypes present in the membrane at a given time confer some properties of plasticity states. These data show that during synaptic plasticity, AMPA receptor subtypes are differentially stabilized by postsynaptic density proteins in or out of the postsynaptic membrane, and this differential synaptic expression of different AMPA receptor subtypes defines distinct synaptic states.

Introduction

Regulation of the number and type of glutamate receptors present in the postsynaptic membranes of excitatory synapses plays a critical role in the induction and expression of synaptic plasticity. Multiple forms of activity-triggered synapse plasticity, such as long-term potentiation, long-term depression, and depotentiation, can change the strength of synapses by regulating the number of AMPA-subtype glutamate receptors (AMPARs) in the postsynaptic membrane of excitatory synapses (for review see Shepherd & Huganir, 2007). Thus, understanding the mechanisms underlying the specification of receptor type and number present in the postsynaptic membrane is central to understanding synaptic plasticity and higher brain processes, such as memory formation and addiction (Nicoll, 2003; Kauer & Malenka, 2007; Citri & Malenka, 2008; Neves et al. 2008). In general, the factors that determine receptor concentration at the synapse include the rate of receptor exo- and endocytosis (Lüscher et al. 1999; Carroll et al. 1999a,b, 2001; Shi et al. 1999; Daw et al. 2000; Hayashi et al. 2000; Shi et al. 2001; Lee et al. 2002; Beretta et al. 2005; Puthenveedu et al. 2007; Yudowski et al. 2007), receptor diffusion within the surface membrane (Adesnik et al. 2005; Ashby et al. 2006; Groc & Choquet, 2008), the binding of receptors to interacting synaptic ‘scaffold’ proteins (Dong et al. 1997; Garner et al. 2000; Kim et al. 2001; Montgomery & Madison, 2004; Waites et al. 2009) and post-translational modifications of both scaffold and receptor molecules (Roche et al. 1996; Chung et al. 2000; Daw et al. 2000; Lee et al. 2000; Chung et al. 2003; Shepherd & Huganir, 2007).

A number of previous studies have dealt with the differential expression of AMPA receptor subunits in the postsynaptic surface membrane (for review see Shepherd & Huganir, 2007). The AMPA receptor is made up of four subunits arising from a combination of four different genes, GluR1–4 (Hollmann & Heinemann, 1994; Rosenmund, 1998; Palmer et al. 2005). In mature hippocampal excitatory synapses, the majority, if not the entirety, of the population of AMPA receptors are heteromultimers made up of either GluR1 and 2, or GluR2 and 3 subunits (Ozawa & Iino, 1993; Wenthold et al. 1996; Sans et al. 2003). The dominant model of subunit-specific AMPAR trafficking and surface expression holds that AMPARs containing the GluR1 subunit are inserted and removed from the membrane in a manner regulated by synaptic activity. GluR2/3 receptors, on the other hand, are thought to be cycled in and out of the membrane in a constitutive manner and removed at a net rate following synaptic depression (Carroll et al. 1999b; Shi et al. 1999, 2001; Passafaro et al. 2001; Piccini & Malinow, 2002; for review, see Shepherd & Huganir, 2007).

Although the extracellular and transmembrane regions of AMPA receptor subunits are similar, their cytoplasmic regions differ in length, phosphorylation sites, and protein–protein interaction sites (Shepherd & Huganir, 2007). An important component of the regulation of AMPAR subunit-specific trafficking is thought to occur primarily via the interactions between PDZ domain-containing ‘scaffold’ proteins and the PDZ-binding sites found at the ends of the C-terminal tails of the AMPAR subunits (Dong et al. 1997; Kornau et al. 1997; Chung et al. 2000; Daw et al. 2000; Hayashi et al. 2000; Garner et al. 2000; Kim et al. 2001; Piccini & Malinow, 2002; Shepherd & Huganir, 2007). With specific regard to AMPARs, synapse-associated protein 97 (SAP97), protein that interacts with C-kinase 1 (PICK1) and glutamate receptor-interacting protein 1 (GRIP1) have been shown to be important in their trafficking and synaptic localization (Osten et al. 2000; Shepherd & Huganir, 2007). However, the mechanisms of synaptic AMPAR trafficking are still not fully understood. For example, it is not clear where in the synapse the interactions between PDZ domain-containing proteins and particular AMPAR tail sequences occur. A PDZ/AMPAR interaction might well scaffold and stabilize AMPARs inside the cell, on the surface of the extrasynaptic membrane as well as their proposed role of scaffolding within the synaptic membrane (Waites et al. 2009). In addition, there are certainly protein/AMPAR interactions that do not involve PDZ domain-containing proteins. For example, AMPARs containing mutated GluR1 (Shepherd & Hugani, 2007) or GluR2 subunits that lack their C-terminal PDZ ligands are still trafficked to the surface (Panicker et al. 2008). Beyond showing an incomplete understanding of the mechanisms underlying this phenomenon, this probably reflects the participation of other proteins known to be important in AMPAR trafficking, such as transmembrane AMPA receptor regulatory proteins (TARPs), which are required for AMPAR surface expression without subunit specificity (Nicoll et al. 2006).

In our previous work, we have shown that activity-dependent synaptic plasticity exhibits the properties of state dependence. We have found that activity does not merely change the strength of synaptic transmission, but also changes the mechanism by which further plastic changes can occur. For example, activity-dependent synaptic depression occurs by a different mechanism in potentiated and non-potentiated synapses (Montgomery & Madison, 2002). While the final common path to these changes in synaptic strength always appears to be regulation of the surface concentration of AMPARs, the mechanisms that underlie the regulation of AMPAR dynamics change with the state of the synapse (Montgomery & Madison, 2002, 2004). Our previous work defined five distinct mechanistic states in which synapses could dwell. The existence of these states was revealed by the ability to record from unitary synaptic connections, that is, between a single presynaptic and postsynaptic neuron pair. Recording unitary synaptic connections revealed properties of synapses that could not be seen by recording large populations of synaptically coupled neurons simultaneously, much like single channel recordings did for ion channels. In this study, we have examined the activity of postsynaptically injected peptides in unitary synaptic connections between CA3 pyramidal neurons, allowing us to acutely disrupt specific protein–protein interactions of AMPARs. Our data not only show that these peptides change the strength of synaptic transmission, but also provide valuable information about how different AMPAR subunit combinations provide a molecular mechanism for state-dependent synaptic plasticity.

Methods

Ethical approval

The studies and methods of this paper were approved by the Administrative Panel on Laboratory Animal Care (APLAC) of the Stanford University School of Medicine. The studies in this paper were carried out in isolated cultured tissues. These tissues were obtained from the brains of approximately 180 rats. Rats were deeply anaesthetized by the breathing in of enflurane at 2 MAC (minimum alveolar concentration), applied until the animal was completely unresponsive to tail and foot pinch, followed by rapid decapitation using a commercial animal decapitator. We have read the article by Drummond (2008), and our experiments comply with the applicable policies and regulations described therein.

Whole-cell patch clamp recordings

Interface organotypic hippocampal slices were prepared from postnatal day (P)7–8 Sprague–Dawley rats as described previously (Stoppini et al. 1991; Pavlidis & Madison, 1999). Slices were maintained at 37°C for 2 days, and then transferred into a 34°C incubator for a total of 6–10 days before recording. Individual hippocampal slices were immersed in recording buffer (ACSF; containing in mm: 119 NaCl, 2.5 KCl, 1.3 MgSO₄, 2.5 CaCl₂, 1 Na₂HPO₄, 26.2 NaHCO₃, 11 glucose) and perfused at room temperature at a rate of 2 ml min⁻¹. In some experiments, the μ-opioid antagonist DAMGO ([d-Ala²,NMePhe⁴,Gly-ol]-enkephalin; Tocris Cookson; Ellisville, MO, USA) was added at a final concentration of 2.5–5 mm to reduce polysynaptic inhibition (Hanson et al. 2006). Pyramidal neurons in area CA3 were visualized using infrared differential interference contrast (DIC) optics. Synaptically connected pairs of cells were recorded simultaneously, the presynaptic cell in whole-cell voltage clamp and the postsynaptic cell in whole-cell current clamp. Voltage clamp recordings of excitatory postsynaptic currents were made using an Axopatch 1C amplifier and presynaptic current clamp recordings were made using an Axoclamp 2A (Axon Instruments; Union City, CA, USA); unless otherwise stated, postsynaptic neurons were held at −65 to −70 mV and events were sampled at 10 kHz, and low-pass filtered at 1–2 kHz. NMDAR-mediated EPSCs were recorded at +40 mV in the presence of 10 μm 2,3-dihydroxy-6-nitro-7-sulfamoyl-benzo[f]quinoxaline-2,3-dione (NBQX) and 20 μm bicuculline to block AMPA receptors and GABA_A receptors, respectively. Presynaptic neurons were induced to fire action potentials by brief injection of depolarizing current (typically 10–50 pA for 20 ms). Once a connected pair was established, baseline EPSCs were collected at 0.1–0.03 Hz in response to presynaptic action potentials. Intracellular recording solutions contained (in mm): 120 potassium gluconate (presynaptic cell) or caesium gluconate (postsynaptic cell), 40 Hepes, 5 MgCl₂, 0.3 NaGTP, 2 NaATP, 5 N-(2,6-Dimethylphenylcarbamoylmethyl) triethylammonium (postsynaptic cell only), pH 7.2 with KOH or CsOH.

Synaptic plasticity induction and analysis

Synaptic plasticity, long-term potentiation (LTP) and long-term depression (LTD) were induced as previously described (Pavlidis et al. 2000; Montgomery et al. 2001; Montgomery & Madison, 2002; Montgomery et al. 2005). LTP was induced by pairing presynaptic action potentials at 1 Hz with postsynaptic depolarization between 0 and −10 mV for 1 min. LTD was induced by pairing presynaptic action potentials at 1 Hz with postsynaptic depolarization to −55 mV for 10 min. Baseline stimulation parameters were resumed after the induction of synaptic plasticity. Series resistance (R_s) was continuously monitored throughout the duration of all recordings, and an experiment was discarded if R_s changed more than 30% over the course of the experiment, or if R_s changed suddenly at the time of experimental manipulation (e.g. at the induction of LTP). Online data acquisition and offline analysis of AMPA and NMDA currents were performed using software written in LabVIEW (National Instruments; Austin, TX, USA) by Paul Pavlidis and Eric Schiable. Unless otherwise stated, all results are presented as mean ± standard error of the mean (s.e.m.). Statistical significance was assessed using Student's t test.

Peptide synthesis

Synthetic peptides corresponding to the C-terminus tail of GluR1 (-ATGL: SGMPLGATGL) and a scrambled control peptide (scrambled: LGASGPLGTM) were made by the Stanford Protein and Nucleic Acid Biotechnology Facility and Genscript Corporation (Piscataway, NJ, USA), respectively. The GluR2 C-terminus peptides (-SVKI: KKEGYNVYGIESVKI; -(p)-SVKI: KKEGYNVYGIES-PO₄-VKI; -SGKA: KKEGYNVYGIESGKA) were made by the Johns Hopkins University Synthesis and Sequencing Facility. Peptides were included in the postsynaptic internal solution at 200 μm, along with the protease inhibitors bestatin and leupeptin (100 μm). To inhibit dynamin-dependent endocytosis in some experiments, we used synthesized D-15 peptide (PPPQVPSRPNRAPPG; Genscript Corporation) and included it in the postsynaptic recording solution at a concentration of 2 mm (Lüscher et al. 1999). Membrane-permeant versions of the peptides were synthesized by including eleven arginines and a biotinylated lysine at the N-terminus (for example, the membrane-permeant GluR1 peptide: RRRRRRRRRRR-K_biotin-SGMPLGATGL; Matsushita et al. 2001). Stock solutions were made by dissolving peptide in water at a final concentration of 10–50 mm and stored at −80°C.

Neuronal cultures

Primary hippocampal neurons were cultured in the ‘Banker’ style from Sprague–Dawley rats as described previously (Goslin et al. 1998). Briefly, hippocampi were isolated from E18 rats, dissociated with trypsin, and plated onto coverslips coated with poly-l-lysine. The coverslips were placed in culture dishes containing a monolayer of glial cells and incubated at 37°C in 5% CO₂. The membrane-permeant peptides were added to the neuron culture medium at a final concentration of 10–50 μm and incubated for 30–60 min. Concentrations used for experiments were selected based on efficiency of peptide translocation.

Immunocytochemistry

Immunocytochemistry was performed in cultured neurons to determine surface levels of AMPA receptor subunits. Briefly, cultures were fixed in 4% paraformaldehyde in PBS on ice for 7 min. Cultures were then rinsed several times with PBS over 30 min, blocked in PBS with 3% BSA and 2% normal goat serum for 1 h, and then incubated in primary antibody in block solution overnight at 4°C. Cultures were rinsed in PBS and then incubated in secondary antibody in block solution for 1 h at room temperature. The coverslips were mounted and imaged on a scanning laser confocal microscope designed by Dr Stephen Smith using a Zeiss 40 1.3 NA Fluor objective. The GluR1 N-terminus antibodies used were described previously (Liao et al. 1999) or were obtained from EMD Biosciences (San Diego, CA, USA). Experiments were conducted using coverslips from three different dissections. GluR1 punctum intensities were measured using custom software (written by Noam Ziv, Technion, Israel). Fluorescence intensities were averaged over 4 × 4 pixel squares centred on GluR1 puncta and normalized within a given experiment by taking the ratio of levels from treated cultures to untreated controls. All immunocytochemical data are presented as mean ±s.e.m. and statistical significance was assessed using Student's t test.

GRIP1 over-expression

CA3 pyramidal neurons were transiently transfected using the Helios gene gun system (BioRad, Hercules, CA, USA) according to the manufacturer's instructions. Briefly, Tefzel tubing was lined with gold particles coated with a mixture of cDNA encoding green fluorescent protein (GFP) and GRIP1 (Dong et al. 1997), and cut into short lengths that served as the cartridges for the gun. Organotypic hippocampal slices were ‘shot’ with the gold particles using the gene gun at 100–130 p.s.i., and returned to the incubator for 24 h before recordings were obtained from CA3 neurons, where the postsynaptic cell was fluorescent (GRIP1 transfected) and the presynaptic cell was not (untransfected). Before recording, the location of the GFP-expressing neuron was confirmed by switching back and forth between epifluorescence and DIC (with infrared illumination), and was further confirmed by the loss of soluble GFP fluorescence (by whole-cell dialysis) once the patch clamp recording was obtained. Control recordings, where both pre- and postsynaptic cells were untransfected, were interspersed with recordings from transfected cells. GFP expression alone does not alter synaptic transmission (Waites et al. 2009).

Results

In order to determine the specific roles played by the AMPAR C-terminal tail sequences in both the expression of synaptic plasticity and in defining different plasticity states, we applied short subunit mimicking peptides to the intracellular cytoplasm of CA3 pyramidal neurons.

Specifically, we used two different 10-mer peptides, one having the C-terminal sequence of the GluR1-receptor subunit (-ATGL), and the other having the GluR2 C-terminal sequence (-SVKI; Daw et al. 2000; Kim et al. 2001). It is established that the C-termini of these subunits contain the recognition/binding sequences that allow them to bind to specific intracellular scaffold proteins. In the case of the GluR1 subunit, the last three/four amino acids recognize and bind to Reversion-Induced LIM (RIL) and/or SAP97. For the GluR2 subunit, the last four amino acids recognize and bind to GRIP/AMPAR binding protein (APB) and/or PICK (Shepherd & Huganir, 2007; Fig. 1A). Our experimental strategy, similar to that used by previous studies (e.g. Daw et al. 2000; Kim et al. 2001) was to apply these peptides intracellularly in order to compete a particular subunit from its binding partner and to record the consequences of this on synaptic transmission and plasticity. However, unlike previous studies, we recorded the effects of peptide injection into postsynaptic neurons while stimulating only a single presynaptic neuron, as schematized in Fig. 1A. Thus, we isolated the measured effects down to a unitary synaptic connection. This gave us the advantage of being able to study synapses in the ‘active’ and ‘silent’ state separately. In this manner, we could determine the effects of the peptide specifically on each of these synaptic states.

Figure 1. AMPA receptors are stabilized in the synaptic membrane via GluR1 subunit tail interactions.

A peptide (ATGL) having the sequence of the last 10 amino acids of the C-terminal tail of the GluR1 AMPAR causes EPSC amplitude to rundown when introduced into the postsynaptic intracellular space. A (upper), model of a subunit of the AMPA receptor, showing the amino acid sequence of the C-terminal tails of GluR1 and GluR2. The SAP97/RIL binding site is formed by the last four/three amino acids (A/TGL) of the C-terminus of the GluR1 subunit, while the GRIP/APB/PICK binding site is formed by the last 4 amino acids of the GluR2 subunit (SVKI). (lower). A schematic diagram illustrates the experimental approach used for most experiments in this paper. Simultaneous whole-cell recordings are made between two nearby CA3 pyramidal neurons. The presynaptic neuron remains unclamped and the membrane voltage recorded. The postsynaptic neuron is held in voltage clamp. In this illustration the ‘postsynaptic electrode’ is shown filled in, to indicate that a substance such as a peptide can be introduced into the postsynaptic neuron by including it in the electrode's electrolyte. Example electrophysiological traces show the presynaptic action potential and postsynaptic EPSCs from such a pair of neurons. B, (top) examples of the presynaptic voltage and postsynaptic current recordings made in a pair of CA3 pyramidal neurons. An action potential evoked in the presynaptic cell results in an EPSC in the postsynaptic cell. Graph, plot of a single experiment showing peptide-induced rundown in a single pair of cells. C, averaged data showing that introduction of the GluR1 peptide -ATGL into the postsynaptic neuron causes a large and rapid rundown of the amplitude of the AMPAR-mediated EPSC. A scrambled version of the GluR1 peptide does not cause rundown. (Difference between control and ATGL significant at P < 0.01, no peptide, n= 14; scrambled peptide, n= 7; ATGL peptide, n= 10). D, rundown of the EPSCs in the presence of intracellular GluR1 peptide is caused by the endocytosis of the receptor. The same experimental protocol was used as in Fig. 1C. The GluR1 peptide does not cause rundown when co-applied with peptide D15, a peptide inhibitory to dynamin function. Peptide D15 alone did not change the amplitude of the EPSC at the same time as ATGL caused a rundown, although D15 caused a run-up in the EPSC at later times, as previously reported (Lüscher et al. 1999). ATGL data (filled circles) are replotted from C for the purposes of comparison. Inset: LFS (1 Hz/10 min) does not cause persistent LTD (long-term depression) in the presence of intracellular D15 peptide.

GluR1-containing receptors

Introduction of the GluR1 tail peptide (-ATGL) into the postsynaptic side of a cell pair caused a rundown in the amplitude of AMPAR-mediated EPSCs, shown both in an individual experiment (Fig. 1B) and as an average of all experiments (Fig. 1C). This rundown effect of the peptide is specific for the peptide sequence, as a peptide having the same amino acid composition but a scrambled sequence had no effect on baseline synaptic transmission (Fig. 1C). Application of the -ATGL peptide having the native C-terminal sequence caused a significant rundown of baseline AMPAR EPSC amplitude, after 15 min, of approximately 50%, which was significantly different (P < 0.01) from both the no-peptide and the scrambled peptide conditions (Fig. 1C, mean ±s.e.m.: –ATGL peptide 45.3 ± 2.0% of initial EPSC amplitude after 30 min, n= 10 pairs; scrambled peptide 99.0 ± 3.7% of initial EPSC amplitude after 30 min, n= 7 pairs; no-peptide control 85.3 ± 2.6%, n= 14 pairs). The effect of the -ATGL peptide on synaptic transmission was rapid, as rundown was evident beginning 5 min from the start of the recording (and the beginning of the peptide introduction). However, synaptic transmission was not completely abolished, as the AMPAR EPSC amplitudes levelled out at approximately 50% of baseline amplitude and remained stable for the remainder of the recording (Fig. 1C).

The -ATGL-induced rundown in the amplitude of the AMPAR-mediated EPSCs probably results from the destabilization of the binding between synaptic GluR1-containing receptors and a PDZ scaffold protein. Once destabilized, receptors may be subject to endocytic removal from the membrane, or lateral diffusion of receptors away from the active zone. In order to differentiate between these two possibilities, we co-applied the peptide D15, an inhibitor of dynamin-dependent endocytosis (Lüscher et al. 1999; Xiao et al. 2001). Intracellular application of D15 prevented the ATGL-induced rundown of the EPSP. At 15 min after the initiation of whole-cell recording, a time when ATGL-induced rundown is normally complete, no rundown had occurred. As expected, application of D15 alone induced a run-up of AMPA receptor-mediated synaptic transmission as seen in a previous study (Lüscher et al. 1999), but this run-up did not commence until after this initial period. (At 30 min after initiation of whole-cell recording, mean AMPAR EPSC amplitude was +147 ± 22% of baseline, n= 4 P < 0.01; Lüscher et al. 1999; Xiao et al. 2001.) To further ensure that the D15 peptide worked to block endocytosis in our system, we repeated the experiments of Lüscher et al. (1999) and Xiao et al. (2001) by applying the D15 peptide and then attempting to induce LTD with 10 min of 1 Hz stimulation (low-frequency stimulation, LFS). As they reported, we found that the synaptic depression following LFS was non-persistent in the presence of the D15 peptide (Fig. 1D, inset: 88.1 ± 3.1% of initial EPSC amplitude after 30 min; n= 4 pairs, P= 0.13). When we co-applied D15 along with the GluR1 peptide, the GluR1 peptide-mediated rundown was abolished (Fig. 1D). AMPAR EPSC amplitudes were not significantly different at either 15 or 30 min after introduction of the peptides (106.6 ± 3.9% of initial EPSC amplitude after 15 min, P= 0.10; 89.3 ± 2.6% of initial EPSC amplitude after 30 min, n= 4 pairs, P= 0.19). These results suggest that the entire effect of the GluR1 peptide is mediated by alterations in the endocytosis of GluR1-containing receptors through its C-terminal tail binding proteins.

GluR1 receptors and plasticity

Because postsynaptic intracellular application of -ATGL led to the rundown of synaptic transmission, we asked whether this rundown would occlude LTD. If the receptors that would be endocytosed during LTD have already been removed from the membrane by peptide application, then LTD would be diminished after peptide-induced rundown. The postsynaptic neuron in a recorded pair was injected with either the -ATGL peptide or the scrambled peptide. As before, the GluR1 peptide caused rundown of the AMPAR-mediated EPSC, while the scrambled peptide did not (Fig. 2A). After 30 min, when the rundown was complete, we applied an LTD induction protocol to each cell pair (LFS; presynaptic action potentials at 1 Hz for 1 min, with the postsynaptic neuron depolarized by 5 mV; Mulkey & Malenka, 1992) LTD occurred normally (Fig. 2A) regardless of the presence of the active or scrambled peptide (-ATGL peptide: LTD was 45.9 ± 1.7% of post-ATGL EPSC amplitude following LFS, n= 4 pairs; scrambled peptide LTD: 43.8 ± 2.3% of initial EPSC, n= 5 pairs; P= 0.48). Thus, the ability to induce LTD does not appear to be affected by the pre-removal of a population of AMPARs from the synaptic membrane. That is, a reduction in the number of GluR1-containing AMPARs in the postsynaptic membrane does not impair the expression of LTD.

Figure 2. The GluR1 peptide -ATGL has no effect on LTD, but loses its effectiveness when synaptic transmission is potentiated.

A, LTD is not affected by the GluR1 peptide. Application of the GluR1 peptide causes rundown of the AMPAR-mediated synaptic response. LFS (1 Hz presynaptic action potentials for 1 min, while postsynaptic membrane potential was held 5 mV depolarized from rest), applied after this rundown is essentially complete, causes further depression of the synaptic response. LFS of responses treated with the scrambled peptide, where no rundown occurred, also shows LTD. The absolute post-LFS levels of transmission reached in the peptide/scrambled peptide conditions are not significantly different (scrambled peptide, n= 5; -ATGL, n= 4). B, data from A replotted normalized to a pre-LFS baseline, with the purpose of directly comparing the amplitudes of LTD in ATGL and scrambled conditions. Note that there is no difference in LTD. C, LTP abolishes the GluR1 peptide-induced rundown. Averaged data (scrambled peptide, n= 8; -ATGL, n= 6) showing that when synaptic transmission is potentiated before intracellular GluR1 peptide has exerted its effect, ATGL no longer causes rundown of the EPSC. LTP was induced by pairing (P) depolarization of the postsynaptic membrane potential to ∼0 mV with presynaptic action potentials evoked at 1 Hz for 1 min.

Unlike LTD, we did detect a significant interaction between the presence of the -ATGL and the expression of LTP. In these experiments we induced LTP after 5 min of the initiation of the whole-cell recording to prevent the washout of LTP, before the rundown had commenced (Fig. 2B). The induction of LTP is normally followed by a modest decay of the EPSC amplitude that usually occurs during the first 20 min. We noted, however, that there was no acceleration of EPSC amplitude decay after the induction of LTP when the postsynaptic cell was injected with -ATGL or the scrambled peptide. The level of LTP in the presence and absence of the peptides was not significantly different (-ATGL peptide 215.6 ± 7.7% of initial EPSC amplitude after 30 min, n= 6 pairs; scrambled peptide 190.2 ± 8.3% of initial EPSC amplitude after 30 min, n= 8 pairs; P= 0.055). This suggests that it is a property of LTP to insert and/or alter the nature of the receptors in the membrane, such that they are no longer susceptible to the effects of the -ATGL peptide.

GluR2-containing receptors

We next tested how LTP and LTD were affected by the presence of GluR2 tail peptides which disrupt GluR2 interactions with GRIP and/or PICK1. Unlike in previous studies using these peptides (Daw et al. 2000; Kim et al. 2001), we specifically examined how GluR2 tail peptides altered plasticity in silent synapses and compared these data to the effects seen on synapses in the active state. Surprisingly, the introduction of the GluR2 peptide -SVKI (to disrupt GluR2 interactions with GRIP and PICK), or the phosphorylated GluR2 -(p)-SVKI peptide (to disrupt GluR2 interaction with PICK1), into the postsynaptic cell of a pair displaying a silent connection, resulted in these silent synapses being spontaneously awakened (Fig. 3A). This was a surprising result because in previous no-peptide experiments, silent synapses remained silent unless awakened by a specific protocol of electrical stimulation (Montgomery et al. 2001). -SVKI awakened silent synapses without the need of that protocol and increased the amplitude of the EPSC in active synapses in every case tested. In contrast, the inactive control peptide -SVKA (Fig. 3C) did not cause silent synapses to awaken when injected into the postsynaptic cell of a silent connection. Interestingly, we found that although both the phospho- and unphosphorylated forms of -SVKI awakened silent synapses, the unphosphorylated form of the peptide resulted in unsilenced synapses with AMPAR EPSCs approximately twice the amplitude of those induced by the phosphorylated peptide (Fig. 3A). On average, between 25 and 30 min after -SVKI peptide infusion, AMPAR EPSCs were 7.7 ± 1.4 pA, compared with 4.0 ± 0.6 pA after -(p)-SVKI peptide infusion (difference significant at P < 0.001). We then attempted to induce LTD in these peptide-unsilenced synaptic pairs. LFS was applied 30 min after the beginning of peptide injection in both -SVKI- and -(p)-SVKI-injected pairs. In both cases, synaptic depression was significantly impaired, with a transient depression recovering to a value not significantly below baseline within 30 min (Fig. 3A; mean EPSC amplitudes were 90.0 ± 19.3% and 68.5 ± 26.6% of baseline AMPAR EPSC amplitudes for -SVKI- and -(p)-SVKI-injected pairs, respectively (n= 8), compared with depression induced by LFS in active synapses injected with the control peptide -SVKA (LFS induced an average depression to 28.7 ± 4.9% of baseline EPSC amplitudes; data not shown (n= 6)).

Figure 3. The GluR2 tail peptide increases synaptic transmission in active and silent synapses.

A, effect of the GluR2 tail peptide on silent synapses. Insets: injection of the GluR2 peptide SVKI into a postsynaptic neuron causes the spontaneous awakening of silent synapses. This panel shows examples of two individual silent connections being awakened by postsynaptic intracellular introduction of the unphosphorylated GluR2 tail peptide (SVKI; upper traces) and the phosphorylated peptide ((p)-SVKI; lower traces). Synaptic transmission was tested at 30 s intervals by firing an action potential in the presynaptic cell of the pair, a stimulus that alone is not sufficient to awaken silent synapses. Graph: both phosphorylated and unphosphorylated peptide caused the awakening of silent synapses in the absence of LTP-inducing electrical stimulation, with the unphosphorylated peptide being approximately twice as effective. B, effects of GluR2 tail peptide in active synapses. In active synapses, the phosphorylated peptide appears to be only slightly more effective than the unphosphorylated. C: Injection of the control peptide SVKA (point mutation I/A) has no effect on silent synapses. Inset 1: -SVKA does not awaken silent synapses, Inset 2: NMDA receptor-mediated synaptic currents are seen when postsynaptic cell is depolarized to +40 mV, demonstrating that this is a silent synapse, Inset 3: Even with prolonged application of -SVKA, no awakening of the silent synapse occurs. These results demonstrate the specificity of the injected peptides, since a single point mutation is sufficient to render it inactive.

In neurons connected by active synapses, the introduction of the GluR2 tail peptides resulted in an increase in the AMPAR-mediated EPSC amplitudes (-SVKI injection resulted in AMPAR EPSC amplitudes increasing 254.8 ± 29.6%, measured 30 min after peptide injection; Daw et al. 2000; Kim et al. 2001). However, unlike previous studies in which SVKI increased AMPAR EPSCs in only a subset (36%) of experiments (Daw et al. 2000), we observed that the AMPAR EPSC amplitudes increased in all experiments (11/11 paired recordings). This difference is probably due to our ability to record from exclusively active or silent synapses, and not a population of synapses in varying states. The introduction of the phosphorylated peptide -(p)-SVKI to disrupt the interaction between GluR2 and PICK1 also increased the amplitude of AMPA receptor-mediated EPSCs in active synapses (Fig. 3B; AMPAR EPSC amplitudes increased to 374.3 ± 51.2% of baseline, measured 30 min after peptide injection; Kim et al. 2001). This increase in AMPAR EPSC amplitude at active synapses was significantly higher than the increase observed with -SVKI (P < 0.001; Fig. 3B). Interestingly, the larger effect of -(p)-SVKI over -SVKI on AMPAR EPSC amplitudes at active synapses was the opposite to what was observed at silent synapses where -SVKI induced higher AMPAR EPSC amplitudes than -(p)-SVKI, although the difference was less than half as large. We then attempted to induce LTD at active synapses after -SVKI or -(p)-SVKI by applying LFS (Fig. 3B). As has been reported previously, LTD is significantly impaired in the presence of intracellular -SVKI or -(p)-SVKI (30 min after LFS average AMPA EPSC amplitudes were depressed to only 84.6 ± 9.1% of baseline AMPAR EPSC amplitudes; Daw et al. 2000; Kim et al. 2001). This lack of LTD has been ascribed to the peptide preventing re-tethering of AMPARs that have been endocytosed, so that they return quickly to the synaptic membrane (Daw et al. 2000), or to the peptide preventing internalization of AMPA receptors (Kim et al. 2001).

Because the -SVKI peptide could awaken silent synapses in the absence of the usually required electrical stimulation, we wondered if the properties of those peptide-awakened synapses were the same as the properties of synaptic-activity-awakened silent synapses. One unique characteristic of an electrical-stimulus-awakened silent synapse is that it cannot undergo LTD of either the AMPAR- or NMDAR-mediated EPSC for 30 min after awakening, a property that defines the ‘recently silent’ state of the synapse (Montgomery & Madison, 2002, 2004). We wanted to test if peptide-awakened synapses similarly spend their first post-awakening half-hour in the recently silent state in which no plasticity occurs. However, this is technically more difficult than with an electrically awakened synapse because the presence of the -SVKI peptide itself prevents LTD of AMPAR-mediated synaptic responses (Fig. 3B; Daw et al. 2000; Kim et al. 2001), thus eliminating the ability to test if LTD of AMPA responses is influenced by synapse awakening. We were, however, able to examine LTD of NMDAR-mediated EPSCs. As we have shown previously (Montgomery et al. 2005), in electrically awakened silent synapses, LTD of the NMDAR-mediated responses is frozen along with the same time course as that of the AMPAR responses. Thus, we could compare the plasticity of the NMDAR-mediated EPSCs between peptide- and stimulus-awakened silent synapses to see if they differ.

As shown in Fig. 4, pairs of cells having silent connections were obtained with -SVKI peptide present in the postsynaptic recording electrode. As in our previous experiments (Fig. 3), 3–4 min after the beginning of the recording the silent synapses would spontaneously awaken. Ten minutes after the beginning of the recording, we introduced 10 μm of the AMPAR antagonist NBQX, allowing us to record isolated NMDAR-mediated responses. Unlike in activity-unsilenced synapses, the NMDAR response in peptide-unsilenced synapses could be significantly depressed by the LFS (average depression to 39.3 ± 8.7% of baseline amplitudes, P < 0.0001; n= 6 silent pairs). Note that the LFS was complete before 30 min had elapsed since synapse awakening, a time when the synaptic responses, including those mediated by NMDARs, would still be ‘frozen’ in stimulus-awakened silent synapses. Thus, peptide-awakened silent synapses lack the ‘recently silent’ state seen in stimulus-awakened silent synapses, suggesting that they differ markedly in some underlying property.

Figure 4. Synapses unsilenced by GluR2 C-terminal peptides are able to express LFS-induced LTD of the NMDAR EPSC even within 30 min of unsilencing.

AMPAR-mediated responses induced by GluR2 tail peptide injection into the postsynaptic neuron of cell l pairs were subsequently blocked by bath application of 10 μm NBQX. Once these nascent AMPA responses were blocked, the postsynaptic cell was depolarized to +40 mV to reveal the NMDA-mediated synaptic current. LFS was applied for 10 min, followed by a return of the postsynaptic cell to +40 mV. The NMDAR-mediated synaptic responses were significantly depressed by LFS when compared to baseline NMDAR EPSC amplitudes prior to LFS (P < 0.0001; n= 6 silent pairs). In these experiments, LFS was always completed within 30 min after the start of recording.

GluR2-lacking AMPARs

AMPA receptors that lack a GluR2 subunit are uniquely sensitive to the channel-blocking toxin philanthotoxin (PhTx; Plant et al. 2006). Thus, we used this toxin to determine if there was a difference in the GluR2 content of receptors present in the postsynaptic membrane during potentiation of active synapses vs. silent synapses. PhTx (1 μm) was applied in the bath immediately after the induction of LTP. In active synapses (Fig. 5A), application of PhTx had no effect on the magnitude of LTP maintained. At 35–40 min after the induction of LTP, the average AMPAR EPSC amplitude was 289.5 ± 78.9% of baseline in controls and 291.8 ± 76.9% of baseline in the presence of PhTx (n= 5 pairs, difference between LTP in control and PhTx not significant (P > 0.5)). However, in silent synapses, PhTx clearly reduced LTP (Fig. 5B). Because the baseline of silent synapse transmission is 0 pA, these data could not be normalized to baseline, so absolute EPSP amplitude is plotted in Fig. 5B. LTP in the absence of PhTx reached an absolute level of 18.1 ± 4.6 pA after the induction of LTP (mean of all responses between 35 and 40 min post tetanus), compared with 4.9 ± 1.9 pA in the presence of the toxin (n= 6 pairs; P < 0.001). These data strongly suggest that during potentiation of active synapses, the AMPA receptors inserted into the postsynaptic membrane all contain a GluR2 subunit, while in silent synapses, the majority of inserted AMPARs lack GluR2. However, it is also noted that PhTx did not completely abolish synaptic transmission in ‘potentiated’ silent synapses. We observed that PhTx reduced transmission in these synapses to within about 1/3 over the control level, suggesting that up to 1/3 of AMPA receptors inserted into silent synapses do contain GluR2 subunits.

Figure 5. Awakening of silent synapses is accompanied by the postsynaptic insertion of GluR2-lacking AMPA receptors.

Active (A) or silent (B) synapses were potentiated, followed by the application of the GluR2-lacking AMPA-receptor channel blocker PhTx. In active synapses (A) there was no difference between toxin-treated and -untreated synapses, suggesting that AMPA receptors inserted postsynaptically during the potentiation of active synapses all contain GluR2 subunits (control LTP: 289.5 ± 78.9% of baseline; PhTx: 291.8 ± 76.9% of baseline, no significant difference: n= 5, P= 0.9). In contrast, in recently silent synapses (B), PhTx clearly reduced the amplitude of potentiated synaptic transmission, suggesting that AMPA receptors without GluR2 subunits are inserted during silent synapse awakening (control: 18.1 ± 4.6 pA after the induction of LTP, compared with 4.9 ± 1.9 pA in the presence of the PhTx (n= 6; P < 0.001). The magnitude of the reduction compared to control recordings suggests that at least 2/3 of the AMPA receptors inserted during the awakening of silent synapses lack GluR2 subunits.

Immunocytochemical localization of AMPARs

In an attempt to differentiate between the possibility that non-synaptic AMPARs are scaffolded inside the cell as opposed to being on the extrasynaptic surface membrane, and that the -ATGL and -SVKI affect receptors at one or the other of those sites, we modified the GluR1 peptide to make it membrane permeant (11R-GluR1). In this peptide we inserted a biotinylated lysine between the poly-R and GluR1 tail sequence (full sequence: 11R-(K_biotin)-SGMPLGATGL). The biotinylated lysine residue allowed us to use an anti-biotin antibody to track the location of the peptide, in order to assay whether it does in fact permeate the cell. We found that in organotypic slices, 11R-GluR1 peptide uptake occurred in only a few neurons. In hippocampal neurons in primary culture, however, virtually every neuron took up the peptide at high levels, showing that the peptide was permeating cells (see Supplemental Fig. 1, available online only). This allowed us to test the effect of the 11R-GluR1 peptide on surface AMPA receptors by immunostaining methods. Immunostaining unpermeabilized cells with an antibody directed against an extracellular epitope of the GluR1 antibody resulted in a distinct punctuate staining pattern (Fig. 6A). As seen in Fig. 6A, application of the 11R-GluR1 peptide caused an approximately 25% decrease in the average punctum intensity compared to scrambled peptide controls (11R-Scram). The reduction in surface AMPAR intensities following 11R-GluR1 application is statistically significant (75.35 ± 0.01% of no-peptide control intensities), compared to scrambled peptide controls (91.76 ± 0.02%; P < 0.05; see also Supplemental Fig. 2, available online only). These results further reinforce the conclusion that destabilization of GluR1-containing AMPA receptors through disruption of GluR1 C-terminal interactions leads to a reduction in surface levels of AMPA receptors. In contrast, application of a membrane-permeant version of -SVKI (11R-GluR2; Fig. 6B), caused an increase in surface expression of GluR1-containing AMPARs (187.20 ± 10.04% of no-peptide controls) compared to scrambled peptide controls (96.80 ± 2.87%; P < 0.05). Taken together, these results suggest that these two peptides cause the endocytosis and exocytosis of GluR1-containing receptors, rather than simply moving them in or out of the extrasynaptic surface membrane.

Figure 6. Surface immunostaining of the GluR1 AMPA receptor subunit is decreased by application of the GluR1 peptide.

In these experiments a membrane-permeant analogue of the GluR1 peptide was used. Aa, immunostaining of an extracellular epitope of the GluR1 subunit in the absence of the GluR1 peptide. Ab, immunostaining of an extracellular epitope of the GluR1 subunit in the presence of the scrambled version GluR1 peptide. Ac, immunostaining of an extracellular epitope of the GluR1 subunit in the presence of the GluR1 peptide (n= 35 fields). Note that surface staining of the GluR1 receptor subunit is significantly decreased by application of the GluR1 peptide both with respect to the no-peptide and scrambled peptide conditions (GluR1 peptide, n= 35 fields; scrambled peptide, n= 34 fields). B, same conditions as in A (immunostaining for an extracellular epitope of GluR1), except with application of the membrane-permeant GluR2 tail peptide. Ba, no GluR2 peptide; Bb, scrambled GluR2 peptide; and Bc, the active form of the GluR2 peptide. Note that application of the active GluR2 tail peptide results in the appearance of AMPA receptors containing the GluR1 peptide on the cell surface (GluR2 peptide, n= 30 fields; scrambled peptide, n= 32 fields). C and D, quantification of the effects of the GluR1 and GluR2 peptide, respectively, on the surface expression of GluR1-containing AMPA receptors.

GRIP1 over-expression

The results using the -SVKI peptide suggest that the role of the GluR2 C-terminal tail is to stabilize receptors at a site away from the postsynaptic membrane, and that destabilizing them results in their spontaneous insertion into this synaptic membrane. One of the proteins that is a binding partner and scaffold for AMPARs is GRIP1 (Dong et al. 1997). To gain evidence on the site at which GRIP1 tethers AMPARs, we biolystically transfected pyramidal neurons in organotypic hippocampal slices using gold beads coated with cDNA for both GRIP1 and GFP. We found that over-expressing GRIP1 in a postsynaptic pyramidal neuron caused a dramatic reduction in the amplitude of action-potential-evoked AMPAR EPSCs in cell pairs compared to control recordings (Fig. 7A). The mean AMPAR EPSC amplitude was 29.94 ± 3.87 pA in control pairs (n= 9), vs. 11.06 ± 0.09 pA in GRIP1 over-expressing pairs (n= 10). Furthermore, the occurrence of apparently unconnected cell pairs, those that had no visible synaptic AMPAR- or NMDAR-mediated currents, was increased approximately 2-fold in GRIP1-overexpressing pairs (11/20 pairs connected, control; 10/36 connected in GRIP1-expressing cells). In addition, we found that LTP was impaired when the postsynaptic cell was over-expressing GRIP1 (Fig. 7B). In control pairs, LTP was induced in 5/6 pairs (average AMPAR EPSC amplitude potentiation was 42 ± 5.2% above baseline), while in pairs over-expressing GRIP1 postsynaptically, LTP was absent in 4/5 pairs (average AMPAR EPSC amplitude after attempting LTP induction was –52.8 ± 1.5% below baseline). Thus, postsynaptic GRIP1 over-expression both reduces the amplitude of synaptic currents and prevents insertion of AMPARs into the postsynaptic membrane. This suggests that an over-abundance of GluR2 binding sites over-stabilizes AMPARs inside the cell, both removing them from the surface and preventing them from being inserted during LTP. Thus, the role of GRIP1 appears to be to stabilize AMPARs at a site outside of the postsynaptic membrane, either inside the cell, or on the surface, but in extrasynaptic membrane. The additional finding that GRIP1 overexpression increases the number of unconnected pairs of cells suggests that long-term removal of AMPARs causes synapse elimination.

Figure 7. Over-expression of the protein GRIP1 reduces synaptic transmission and prevents the induction of LTP.

Pyramidal neurons in organotypic slices were made to express cDNA for GRIP1 and GFP by biolystic transfection. Twenty-four hours after transfection, whole-cell recordings were made from pairs of cells where the postsynaptic cell was expressing GFP (and over-expressing GRIP1). A, GRIP1 over-expression reduces the amplitude of action potential-evoked EPSCs in cell pairs (stimulated by the action potential of a single presynaptic neuron, recorded in a single postsynaptic neuron), and increases the proportion of cell pairs that are apparently unconnected (i.e. have no AMPAR- or NMDAR-mediated EPSC; control, n= 11/20 pairs connected; GRIP1-expressing cells, n= 10/36 pairs connected). B, scatter plot showing the amplitude of EPSCs before and after inducing LTP in control non-transfected cell pairs (filled triangles) and pairs with postsynaptic GRIP1 transfection (open triangles). LTP was absent in pairs where the postsynaptic cell was transfected with GRIP1 cDNA. Small symbols connected by lines show the results for each individual pair of cells tested. Offset large symbols show the mean and s.e.m. for each group.

Discussion

Determining how AMPA receptors are localized to the synapse, inserted and removed from the synaptic membrane upon stimulation is imperative to understanding the molecular mechanisms of synapse plasticity in the central nervous system. While synapse strength can be altered by the common path of changing the number of synaptic AMPA receptors, with AMPARs inserted during the induction of LTP and removed during induction of LTD, it is now known that the mechanisms underlying these movements of AMPARs change, depending on the synapse's plasticity history. Put another way, the mechanistic state of the synapse changes as the synapse undergoes activity-dependent changes in potency. These states not only reflect the history of the synapse, but also dictate and limit the future plastic changes available to that synapse (Montgomery & Madison, 2002, 2004). The molecular mechanisms that underlie and define each state have previously been unclear. Here our work now shows that differential synaptic trafficking of GluR1- versus GluR2-containing AMPA receptors not only occurs during synaptic plasticity, but serves as the mechanism underlying at least some of these synaptic plasticity states.

In our experiments, we have determined specific roles played by GluR1- versus GluR2-containing AMPA receptors in state-dependent synaptic plasticity by examining the effects of GluR1 and GluR2 C-terminal tail-mimicking peptides on synaptic transmission and plasticity. Initial experiments showed that the injection of the GluR1 tail peptide weakened synaptic transmission via an endocytosis-dependent process. This suggests that when the peptide replaces the tail of the GluR1 receptor on its scaffold protein, it destabilizes the receptor in the synaptic surface membrane, allowing for its removal. The rundown of synaptic transmission that is caused by the GluR1 tail peptide is prevented by the co-application of D15, a peptide known to block endocytosis (Fig. 1D; Lüscher et al. 1999; Xiao et al. 2001), and is accompanied by a decrease in surface GluR1 expression (Fig. 6). Thus, the GluR1 tail peptide-induced removal of GluR1-containing receptors from the synaptic membrane occurs via their untethering in the postsynaptic membrane and subsequent endocytosis, rather than a dispersal of the receptors to the extrasynaptic surface membrane.

In contrast, the injection of the GluR2 tail peptide -SVKI had the opposite effect to the GluR1 tail peptide -ATGL. As was previously shown (Daw et al. 2000; Kim et al. 2001), synaptic strength increases within a few minutes of postsynaptic introduction of the GluR2 tail peptide -SVKI. These results suggest that the tethering of the GluR2 tail to its scaffold protein is responsible for stabilizing GluR2-containing AMPARs at a site outside of the synaptic membrane (Daw et al. 2000; Kim et al. 2001). The finding that synapses of neurons postsynaptically overexpressing GRIP1 have reduced amplitude and that LTP cannot be induced at these synapses further supports this conclusion. The fact that LTD is present but transient in the presence of intracellular GluR2 tail peptide suggests that the initial processes that underlie synaptic depression, such as endocytosis, are intact in the presence of the peptide, but that the receptors are not retained as they are normally at a site outside the synapse. Rather, unable to re-tether to a GluR2 scaffold, they return to the synaptic membrane. Thus, synapses exposed to the GluR2 tail peptide are essentially re-potentiated because of the inability of the GluR2 tail of the AMPA receptor to bind to its normal scaffold outside of the synapse.

The comparison between the effects of the GluR1 and GluR2 tail peptides suggests an interesting principle of AMPA receptor trafficking during plasticity. While the details are more complex, at its core the interplay between LTP and LTD is a hand-off between the stabilization of the GluR2 tail outside the synaptic membrane and the GluR1 tail inside that membrane. One important issue is whether the receptors added or subtracted from the membrane come to and from an intracellular pool, or instead dispurse to and return from the extrasynaptic membrane. Our findings tracking receptors with immunocytochemistry showing that receptors appear and disappear from the membrane with -SVKI and -ATGL, respectively, suggest that receptors are trafficked between intracellular and membrane sites. Whether receptors trafficking to and from an intracellular pool transition through the extrasynaptic membrane is not addressed by our experiments.

Our findings extend previous studies in several additional important ways. Because our recordings were made in pairs of cells where the state of the connecting synapses were known, we could separately test the effects of the peptide on active and silent synapses. Furthermore, we could then test the properties of other individual states that began from the active or silent state. This has provided several important pieces of information about the mechanisms of LTP. Among these are:

1. Because AMPA receptors are known to constitutively cycle in and out of the membrane in active synapses, previous experiments did not differentiate between an effect of the GluR2 peptide on endocytic, exocytic or other processes. Silent synapses, on the other hand, lack AMPA receptors in their synaptic membrane, and therefore have no constitutive cycling. Thus, the increase in transmission seen when applying the GluR2 tail peptide must arise from an increase in receptor delivery to the synaptic membrane, rather than an increase in receptor retention in that membrane. As an aside, while the dominant model of AMPAR trafficking holds the GluR1-containing receptors to be regulated in their trafficking, and GluR2/3 receptors to cycle constitutively, it is clear from the difference in silent and active synapses that there is some regulation of the cycling of GluR2/3 receptors.

2. The -ATGL-induced loss of GluR1-containing receptors from the synaptic membrane does not reduce the percentage of LTD that is induced by low-frequency stimulation. This suggests several things about the properties of LTD. First, that the presence of the GluR1 subunit is not necessary for the induction of LTD. Furthermore, the finding that LTD reduces synaptic transmission by the same percentage with or without ATGL injection suggests that once induced, LTP either does not differentiate between AMPAR receptors containing different subtypes, or it prefers GluR2-containing receptors. If, for example, LTP removed primarily GluR1-containing receptors, then the percentage of LTD achieved would be reduced after removal of these receptors. Alternatively, if GluR3-containing receptors were preferentially removed during LTD, then the percentage of LTD would be increased after removal of GluR1-containing receptors.

3. The finding that the induction of LTP abolishes the rundown of synaptic potentials caused by injection of ATGL, suggests that AMPA receptors are stabilized in the membrane by a different mechanism in potentiated synapses to that in active synapses. This could reflect a modification of the receptors themselves, such as phosphorylation or a change in the properties of the scaffold. Given the well-documented role of receptor phosphorylation in LTP (see Shepherd & Huganir, 2007), we favour the former explanation.

4. The finding that the GluR2 tail peptide can awaken silent synapses has several interesting implications for understanding the mechanisms underlying LTP. Because the GluR2 tail peptide causes AMPAR insertion in silent synapses in the absence of an electrical potentiating stimulus, we conclude that the sole factor required to potentiate these synapses is that GluR2-containing AMPARs are made available for insertion. Any ‘slots’ in the postsynaptic membrane needed to receive these receptors are apparently already present and any exocytic processes needed to move them to the synaptic membrane are either already active or have already occurred.

5. By comparing the efficacy of the phosphorylated and unphosphorylated GluR2 tail peptides (Fig. 4), we are able to conclude that AMPA receptors inserted during the induction of LTP and those inserted during the potentiation of silent synapses come from different sources. We found that the relative efficacy for the phosphorylated and unphosphorylated GluR2 tail peptides was different in active and silent synapses. In silent synapses, -SVKI was approximately twice as potent as -(p)-SVKI, while in active synapses, -(p)-SVKI and -SVKI were equally potent. Since -SVKI will compete to remove the GluR2 receptor from both GRIP and PICK, while the phosphorylated peptide competes primarily with the receptor for the binding site on PICK, we interpret this result to suggest that in silent synapses, AMPA receptors are released to the membrane from both PICK and GRIP scaffolds, while in active synapses most AMPARs source from a PICK scaffold. This might appear to raise an issue with the efficacy of over-expression of GRIP in preventing LTP if some or all receptors are sourcing from PICK, but it does not. An over-abundance of GRIP binding sites would be expected to bind any free GluR2-containing AMPARs, preventing them from reaching the membrane, regardless of what scaffold they were released from.

6. The finding that PhTx reduces transmission in recently potentiated silent synapses, but not in recently potentiated active synapses shows that when silent synapses are potentiated, AMPA receptors lacking GluR2 are inserted into the postsynaptic membrane, whereas when active synapses are potentiated, only GluR2-containing receptors are inserted. However, not all of the receptors inserted during the awakening are GluR2-lacking, since PhTx is only partially effective. Because we know that GluR2 and GluR1 subunits are present in the membrane of recently silent synapses, and also that GluR1 subunits are not necessary for LTD induction, we reason that it is the presence of the GluR3 subunit that confers the ability of a synapse to undergo LTD. Further support for this idea is provided by our finding that silent synapses that are activated by intracellular injection of GluR2 tail peptides have LTD, unlike recently silent synapses that have been awakened by electrical stimulation. The -SVKI peptide releases all GluR2-containing receptors from their intracellular scaffolds, and thus would put GluR2/3 receptors in the membrane. Since GluR1- and 2-containing receptors are clearly in the membrane of stimulus-awakened silent synapses, we propose that the characteristic that defines the recently silent state is the lack of GluR3-containing receptors in the membrane.

Together, our data suggest that the presence, absence or differential stabilization of particular subunits of AMPA receptors changes when synapses undergo plasticity, and furthermore, that these changes limit the plastic potential of synapses and thus form at least part of the mechanisms that cause the state-dependent behaviour of synapses. In Fig. 8, we propose a model that takes into account our current and previous data, along with the general model of AMPAR trafficking in plasticity. The basic framework that informs our current model is that GluR1-containing receptors (generally thought to be GluR1/2 heteromers) are inserted and removed in a regulated fashion, while the default activity of GluR2/3 AMPARs is to cycle constitutively into and out of the synaptic membrane (see Shi et al. 2001). In our previous work (Montgomery & Madison, 2002, 2004), we described five distinct mechanistic plasticity states in which synapses could reside. We have proposed hypothetical roles of AMPA receptor subunits as the underlying mechanism that define or accompany these states. Our data now provide direct evidence that the differential presence of particular subunits of AMPARs underlies the properties of at least three of the mechanistic plasticity states: Silent, Recently Silent and Active (Fig. 8).

Figure 8. A hypothetical model for the role of receptor subunit types in defining the properties of synaptic plasticity states.

Experimentally, synapses are encountered in one of two states: either Active or Silent. Silent synapses are not merely active synapses without surface AMPARs, but represent a distinct mechanistic state based on their different behaviours following potentiation. We illustrate the following states and the behaviour of AMPARs in each of those states as follows. Silent synapses besides simply lacking AMPA receptors in their postsynaptic membranes have three other interesting properties. First, they cannot be made to undergo LTD of their NMDAR-mediated EPSCs (Montgomery et al. 2004). Second, when unsilenced, these synapses do not transition directly to the Active state, but rather enter a unique state that we have termed Recently Silent (Montgomery & Madison, 2002, 2004). Finally, when unsilenced, the AMPARs that are trafficked to the synaptic membrane come from GRIP and PICK scaffolds. This latter conclusion is based upon the current finding that injection of the SVKI peptide, which competes with the GluR2 tail from both GRIP and PICK, causes the appearance of an AMPAR-mediated EPSC that is approximately twice the amplitude of that produced by injection of -(p)-SVKI, which replaces with the GluR2 tail only from PICK. These data suggest that one mechanism underlying the difference between the Silent and Active state is the scaffolding proteins employed to tether AMPA receptors outside of the postsynaptic membrane. Recently Silent state: when silent synapses are awakened, they do not transition directly to the Active state, but instead enter a unique state that is defined by the fact that these synapses cannot be made to undergo synaptic depression of either the AMPAR- or NMDAR-mediated EPSC (Montgomery & Madison, 2002, 2004). Our current finding that silent synapses awakened by peptide injection bypass this Recently Silent state, instead going directly to the LTD-capable Active states, suggests that the ability to undergo synaptic depression is conferred by the presence of a subset of AMPAR in the postsynaptic membrane. The explanation most parsimonious with the prevailing model and consistent with our observation that PhTx significantly decreases the potentiation of silent synapses, is that GluR1-containing receptors are inserted alone during silent synapse unsilencing by electrical stimulation. The eventual return to the Active state would then correspond to the delayed activation of cycling of non-GluR1-containing receptors. In contrast to the GluR1 peptide, the GluR2 tail peptide should not differentiate between any of the GluR2-containing AMPARs. Thus, a likely explanation for the difference in silent synapses awakened with the GluR2 peptide rather than electrical stimulation is that both GluR1/2 and GluR2/3 receptors are immediately put into the membrane. Taking this reasoning one step further, it would follow that the non-plastic nature of the ‘Recently Silent’ state awakened with electrical stimulation occurs because only GluR1/2 receptors are present in the synaptic membrane, and the expiration of this state occurs when the constitutive cycling of GluR2/3 receptors begins with a delay after stimulation. This result then suggests that not only does synaptic depression require the presence of the GluR2/3 receptor in the membrane, but that constitutive cycling of these receptors is regulated by synaptic activity, albeit, not in the immediate way of GluR1/2 receptors. The central state, the Active state, is the most pluripotent of the states, able to undergo either potentiation or depression, as well as silencing. Plasticity in either direction from the Active state is mediated by activation of the NMDA receptor. The fact that in active synapses the presence of the GluR1 peptide results in an average rundown of 50% of baseline amplitudes suggests that half of the receptors present in the membrane in this state contain GluR1 subunits, probably GluR1/2. Our findings would suggest that it is the GluR2 tail that is limiting for activity-induced insertion of AMPARs in the membrane, and that once there, they are stabilized by a mechanism that differs from the GluR1 tail/PDZ interaction that exists in the basal state. The finding that LTD in active synapses treated with the GluR2 tail peptide is non-persistent (Fig. 4), further suggests that when AMPARs are endocytosed, they are quickly returned to the membrane in the presence of the peptide, due to the fact that they cannot be stabilized inside the cell (see also Daw et al. 2000). GluR2-containing AMPA receptors destined for the synaptic membrane could be tethered at an intracellular site, such as a recycling endosome (Ehlers, 2000; Lee et al. 2000; Park et al. 2004), or they could already be at the membrane surface at an extrasynaptic site. Our immunocytochemical staining experiments (Fig. 6) show that the GluR2 peptide causes an increase in surface AMPA receptors containing GluR1 subunits, suggesting that the receptors were recently inserted in the membrane. In other words, these results suggest that the GluR2-containing receptors were at an intracellular site before being liberated by the GluR2 peptide. However, it is also possible that the receptors were already at an extrasynaptic site, but too dispersed to show up clearly in our fluorescence measurements. Potentiated synapses are, of course, stronger than active synapses, but these synapses comprise a distinct plasticity state in that activation of mGluRs and not NMDARs changes synaptic strength (Montgomery & Madison, 2002, 2004). Our data do not relate directly to the issue of whether differential AMPA receptor subunit surface expression participates in the mechanisms that define the Potentiated state (Fig. 7). What we do know from our data is that once synapses are in the Potentiated state, they contain the GluR2 subunit (Fig. 6) and that the GluR1 tail peptide loses its ability to cause rundown of the potentiated synaptic currents. This could arise from multiple possible mechanisms. The first possibility is that GluR1-containing receptors are not inserted during LTP of active synapses, but that GluR2/3 receptors are. However, previous studies (Hayashi et al. 2000; Shi et al. 2001) have strongly suggested that at least the receptors inserted initially during LTP are GluR1-containing. As our PhTx data have shown that potentiated synapses contain the GluR2 subunit, the GluR1-containing receptors would be GluR1/2 heteromers. A second possibility is that GluR1-containing AMPARs do not require the GluR1 C-terminal tail to be stabilized in the postsynaptic membrane. This idea is supported by the finding that LTP seems to occur normally in mice where GluR1 receptors lack their C-terminal PDZ ligand (Kim et al. 2005). Further support for this idea comes from the established findings that the AMPAR accessory proteins, TARPs, participate in the stabilization of GluR1 receptors, but do not have specificity for the tail region of the receptor (Tomita et al. 2004; Ziff, 2007). A third possibility is that GluR1-containing receptors are inserted during LTP, but are replaced by GluR2/3 receptors during the half-hour following LTP induction. In this regard, it is interesting to note that LTP does run down for approximately 30 min following LTP induction, and with a time course that is similar to that caused by the GluR1 tail peptide in control conditions. The fact that no additional rundown occurs in the Potentiated state in the presence of the GluR1 tail peptide might simply represent the fact that GluR1-containing receptors are leaving the postsynaptic membrane anyway. Finally, it is possible that GluR1-containing receptors are inserted during LTP, but that phosphorylation of their tails leads to a stabilization mechanism that is not sensitive to the GluR1 tail peptide. Certainly there is abundant evidence that phosphorylation is necessary for LTP, and that the tails of the GluR1 receptor are phosphorylated during its induction (see Shepherd & Huganir, 2007). Because of the overwhelming evidence for a role of phosphorylation in inducing LTP, we have chosen to illustrate only this fourth possibility in our model, illustrated in this figure. Key: As labelled; * indicates the intracellular scaffold and/or cycling waypoint for GluR1-only AMPARs, the exact location of which is not suggested by the data.

In summary, our current experiments provide new information about the role of specific subunits of AMPA receptor trafficking during synaptic plasticity. Our data suggest that the central feature of AMPAR trafficking during synaptic plasticity is a hand-off between PDZ/GluR2 tail interactions, which stabilize AMPARs outside of the postsynaptic membrane, and GluR1-associated processes that stabilize AMPARs within the postsynaptic membrane. The postsynaptic density scaffold proteins PICK and GRIP play a pivotal role in subunit-dependent AMPAR trafficking in the Active and Silent state, respectively, by stabilizing GluR2-containing receptors outside the postsynaptic membrane, most likely at an intracellular site. While regulated insertion/deletion of GluR1/2 receptors and constitutive cycling of GluR2/3 receptors probably remain the dominant trafficking characteristics, it is clear from our results that this previous model is too rigid and that receptor cycling can be regulated in a state-dependent manner. Indeed, it appears that the regulation of this cycling has a major impact on the plastic potential of synapses and the molecular mechanisms that define synaptic states.

Glossary

Abbreviations

AMPAR

AMPA receptor

GFP

green fluorescent protein

GluR

glutamate receptor

GRIP1

glutamate receptor-interacting protein 1

LFS

low-frequency stimulation

LTD

long-term depression

LTP

long-term potentiation

PhTx

philanthotoxin

PICK1

protein that interacts with C-kinase 1

SAP97

synapse-associated protein 97

Author contributions

This project was conceived by D.V.M., J.M.M. and M.R.E. The experiments in this paper were designed by J.M.M., M.R.E. and D.V.M. and were performed by M.R.E., J.M.M., J.E.H. and M.L.H. Data were analysed and interpreted by D.V.M., J.M.M., M.R.E., R.L.H., M.L.H. and L.M. cDNA for GRIP1 was made and provided by L.M. and R.L.H., who also assisted in the design of the experiment where it was used. All authors participated in the preparation of the manuscript and all have approved the final copy. All experiments were performed at Stanford University School of Medicine, CA, USA. J.M.M. and M.R.E. contributed equally to this study and are co-first authors.

Supplemental material

Supplemental Figure 1

Supplemental Figure 2

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer-reviewed and may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors