|Home | About | Journals | Submit | Contact Us | Français|
A central assumption in neurobiology holds that changes in the strength of individual synapses underlie changes in behavior. This concept is widely accepted in the case of learning and memory where LTP and LTD are the most compelling cellular models. It is therefore of great interest to understand, on a molecular level, how the brain regulates the strength of neuronal connections. We review a large body of evidence in support of the very straightforward regulation of synaptic strength by changing the number of postsynaptic receptors, and discuss the molecular machinery required for insertion and removal of AMPA receptors.
The hypothesis that insertion of AMPA receptors (AMPARs) underlies the increase of synaptic strength associated with LTP was put forward almost 20 years ago1, but was largely ignored until the mid-1990s, resurfacing with quantal analysis of LTP (Ref. 2). However, only recently has it become the subject of direct experimental scrutiny.
The spark for the myriad of research published in the past few years in the field was experiments conducted independently by two groups3,4, in which single connections between CA3 axons and CA1 pyramidal cells in acute hippocampal slices were functionally isolated and in some cases yielded only responses from NMDA receptors (NMDARs) and not AMPARs. Inducing LTP, however, caused the appearance of AMPAR mediated excitatory postsynaptic currents (EPSCs). This led to the proposal that a fraction of ‘silent’ synapses contain only NMDARs, and thus are functionally inactive at normal resting potential, but maintain the capability to undergo LTP, which would be expressed by AMPAR insertion. Although alternative explanations for these results have been proposed5,6, synapses with NMDARs but not AMPARs have now been directly identified anatomically in cultured hippocampal neurons using immunofluorescence7,8 and with immuno-gold preparations visualized by electron microscopy in hippocampal slices9–11.
If LTP expression is caused by postsynaptic AMPAR insertion, it must be the case that AMPARs can be inserted into and removed from the postsynaptic membrane on a relatively rapid time scale. To test this experimentally, CA1 pyramidal cells in acute hippocampal slices have been loaded with toxins that cleave vSNARE proteins or dominant-negative peptides that disrupt dynamin function, thus blocking exo- or endocytosis in the postsynaptic cell12. When exocytosis was blocked, EPSC amplitudes decreased over a time period of 30 min. Conversely, blocking endocytosis caused an increase in the EPSC amplitude, presumably as a result of an accumulation of AMPARs, with unabated exocytosis. Taken together these results suggest that AMPARs are constantly removed from and inserted into the synaptic membrane independent of activity, indicating a constitutive recycling of receptors. The existence of such a mobile pool of AMPARs is now confirmed by biochemical assays of receptor turnover and immunocytochemical methods in cultured hippocampal neurons and rat hippocampal slices13. Constitutive endocytosis of AMPARs has also been demonstrated in cultured hippocampal neurons14 and the finding that this endocytosis is not accompanied by an eventual complete loss of synaptic AMPARs strongly implies constitutive recycling.
The available evidence suggests that the endocytotic arm of this recycling occurs via dynamin-dependent endocytosis of clathrin-coated vesicles12, whereas the exocytotic arm requires some of the players in vesicle fusion that have been described in transmitter release, including SNAP (Ref. 15), synaptobrevin12 and N-ethylmaleimide-sensitive fusion protein (NSF)12,16–18. An intriguing, but poorly understood, aspect of exocytosis is that GluR2 has been shown to specifically interact with NSF (see Table 1), which is enriched in the postsynaptic density (PSD) (for reviews see Refs 19,20). Disruption of this interaction leads to a decrease of synaptic responses (and concomitant loss of surface AMPAR immunostaining) similar in kinetics and magnitude to the decrease caused by the block of exocytosis. To date, the exact role of this interaction between the two proteins is still under investigation. Mice with the targeted disruption of GluR2 have a decreased ratio of AMPA:NMDA responses, corroborating that even on this longer time scale, the absence of GluR2 might also lead to a partial deficit of AMPAergic transmission21.
In addition to the above described constitutive recycling, several forms of receptor activation strongly accelerate dynamin-dependent endocytosis of AMPARs (Ref. 22). NMDAR activation induces AMPAR internalization and recycling. This internalization requires the activation of protein phosphatases13, one of which is the Ca2+-dependent protein phosphatase calcineurin23. The crucial dephosphorylated target is still unknown. Internalization is accompanied by dephosphorylation of GluR1 at a PKA phosphorylation site13; however, dephosphorylation of a component of the endocytotic machinery might alternatively be required. Indeed, calcineurin is known to dephosphorylate several proteins involved in endocytosis, including dynamin24,25.
AMPAR activation can also induce AMPAR internalization26, but the mechanisms involved are controversial. Some evidence suggests that ligand binding by itself is a trigger for endocytosis13,14, but it has also been reported that AMPAR-dependent internalization is caused by an AMPAR-dependent depolarization, which then induces Ca2+ influx through activation of voltage-gated Ca2+ channels14,23,27, and would presumably act through the same mechanisms as NMDA-induced internalization. The ligand-dependent AMPAR internalization is not dependent on intracellular Ca2+ and is not accompanied by dephosphorylation of GluR1; notably, AMPARs internalized by this ligand-dependent mechanism are targeted to lysosomes rather than recycled14. Based on this, it has been suggested that the phosphorylation state of GluR1 at the PKA site might regulate the sorting of internalized receptors13 (see Fig. 1).
In addition to glutamate, insulin, which is known to promote endocytosis in many cells, has also been found to enhance AMPAR internalization28. The mechanisms underlying this internalization remain largely unknown. Calcineurin could possibly be involved14, but this has been contested23. Curiously, AMPARs internalized by insulin are initially targeted to early endosomes, but are then sorted into a separate, non-lysosomal internal compartment14. The insulin-induced internalization also requires the C-terminal tail of GluR2, but not GluR1, although the PDZ (PSD-95, Discs-large, ZO-1) recognition sequence at the C-terminal (see Table 1 and Box 1) is apparently not required14. Considerable work remains to elucidate how receptors are sent into this distinct sorting path, and to determine the physiological significance of this signaling cascade.
Recent attention has been focused on AMPA receptor movement and localization, but much research has also been carried out on other receptors. We briefly outline the ins-and-outs of two other major ionotropic receptors in the CNS: NMDA receptors and GABAA receptors.
It is clear that few of the mechanisms involved in AMPA receptor (AMPAR) trafficking apply to NMDA receptors (NMDARs); indeed, stable NMDAR EPSCs are frequently used as a control when manipulations that disrupt trafficking cause changes in AMPAR EPSCs. However, a recent study found that the subunit composition of synaptic NMDARs in visual cortex can change in an activity-dependent fashion over a few hours, suggesting that regulated receptor insertion can occur for NMDARs (Ref. a), and recent evidence suggests that protein kinase C (PKC) activity can trigger NMDAR insertionb. The mechanisms by which the NMDAR number at the synapse is regulated are unknown. Surprisingly, however, the widely examined interaction between PSD-95 and NMDARs seems not to be involved, because mice with a truncated PSD-95 that does not localize at synapses have normal synaptic NMDAR responsesc and disruption of PSD-95 localization to the postsynaptic density does not affect synaptic clustering of NMDARs (Ref. c). Presumably other proteins regulate NMDAR number at the synapse; α-actinin is known to bind to NMDARs (Ref. d) and might stabilize NMDARs at the synapse, and PDZ-domain-interacting proteins other than PSD-95 might regulate synaptic expression of NMDARs.
It has been known for some time that GABAA receptor agonists can cause GABAA receptor internalization in neurons, and an agonist-independent receptor internalization that is activated by phorbol esters has been shown in heterologous cells (reviewed in Ref. e). Rapid receptor insertion at GABAergic synapses in response to insulinf and kindling-induced epilepsyg has also been reported. Moreover, blockade of postsynaptic endocytosis increases synaptic GABAA responses, indicating that constitutive endocytosis of these receptors regulates GABAergic synaptic strengthh. The machinery for regulated GABAA receptor insertion and removal from the synapses therefore seems to be present, but little is known about the nature of this machinery. A probable player in these phenomena is gephyrin, a microtubule-binding protein that was first shown to cluster synaptic glycine receptors and subsequently also GABAA receptors i. However, a strong direct interaction between GABAA receptors and gephyrin has not been foundj, suggesting that other linker proteins might regulate GABAA receptor number at the synapse.
Exocytosis of AMPARs is more difficult to investigate, partly because most stimulation protocols applied to cultured neurons lead to AMPAR endocytosis. Recently, using brief application of glycine to stimulate NMDARs of cultured hippocampal neurons29 it was possible to reliably increase surface expression of AMPARs and synaptic efficacy. Both results were blocked by NMDAR antagonists and postsynaptically applied tetanus toxin. In organotypic slice cultures overexpressing GluR1, activation of NMDARs was sufficient to translocate this subunit into spines30 and drive the synaptic insertion of AMPARs made exclusively of GluR1 subunits. It was found that perfusion of neurons with activated Ca2+–calmodulin kinase type II (CaMKII) was sufficient to drive receptor insertion. Subcellular fractionation assays of receptor movement similarly suggest that NMDAR activation can result in AMPAR insertion31. Overexpression of different AMPAR subunits in CA1 neurons of organotypic hippocampal slice cultures indicates that receptors made of different subunits participate preferentially in various forms of AMPAR insertion. Although GluR2- and GluR3-containing receptors make up the bulk of constitutively recycling AMPARs, GluR1 dominates if receptors are inserted after appropriate NMDAR activation32,33. It is interesting to note that subunit-specificity of these interactions might vary from synapse to synapse. At the cerebellar parallel fiber-stellate cell synapses, high-frequency stimulation leads to insertion of GluR2-containing AMPARs and removal of GluR2-lacking AMPARs (Ref. 34).
Variation in subunit composition could also explain differences between synaptic and extrasynaptic AMPARs. Knockout mice lacking GluR1 have a normal AMPAR complement as assessed by synaptic currents, but massive deficits in AMPAR responses to an exogenous agonist35. This indicates that under normal conditions, the bulk of surface GluR1 is not located at synapses, but is presumably at extrasynaptic sites. This raises the possibility that distinct mechanisms would be responsible for surface insertion and synaptic targeting. Additional support for this idea comes from the analysis of the stargazer mutant mouse. Stargazin mutants have a complete lack of AMPAR surface expression in cerebellar granule cells36,37, and it appears that stargazin, through a direct interaction with AMPARs and PDZ-containing tethers, is directly involved in receptor targeting38. The stargazin protein has a PDZ-binding domain, and expression of a truncated form of stargazin lacking the C-terminal PDZ interaction site rescues responses to extrasynaptic AMPAR activation, but not spontaneous EPSCs (and therefore synaptic AMPARs). This dissociation suggests distinct pathways for surface insertion and synaptic targeting, with an, as yet, undetermined region of stargazin required for trafficking of AMPARs to the surface, and stargazin-PDZ interactions required for synaptic expression of AMPARs.
The numerous results demonstrating AMPAR movement raise the question of when, or whether, AMPARs ever stop moving. Is there a population of AMPARs that under normal conditions are stably held at the synapse? One suggestion that such a stable population exists is that the rundown of AMPAR-mediated EPSC amplitude induced by inhibitors of exocytosis is incomplete12. Another suggestion of such a stable population is that the glutamate-induced removal of AMPARs from the surface is not accompanied by an increase in the rate of endocytosis27, suggesting that internalization is rate-limited by release of AMPARs from anchors at the synapse.
Tethering of AMPARs at synapses might ultimately be accomplished by interactions with the cytoskeleton; indeed, GluR1 is reported to bind directly to the membrane cytoskeletal protein 4.1 (Ref. 39). Depolymerization of actin leads to a loss of AMPARs from the surface27,40 and a decrease in synaptic efficacy41. Conversely, stabilization of actin has been reported to prevent glutamate-induced AMPAR internalization27; however, this result is controversial23, and interpretation is further complicated by the observation that actin stabilization also interferes to some degree with the endocytotic process27.
PDZ domain-containing proteins have also been described – glutamate receptor-interacting protein (GRIP), AMPAR binding protein (ABP), and protein interacting with C kinase (PICK1) – that bind to GluR2 and GluR3 through their class II PDZ domain recognition site, and could be involved in regulating the number of AMPARs at the synapse. Removal of the PDZ binding site from GluR2 reduces the accumulation of AMPAR levels at the synapse42. Moreover, disruption of the interaction between GluR2 and GluR3 and PDZ domains has been reported to block LTD in the hippocampus43 and cerebellum44. These effects appear to be regulated by protein kinase C (PKC), which reduces binding of GluR2 to GRIP and ABP, but not PICK1, by phosphorylating AMPARs at a serine residue (Ser-880) near the C-terminal45. However, disruptions of the interaction between AMPARs and PDZ domains have little consistent effect on baseline synaptic transmission43. Moreover, the GluR2–PICK1 interaction appears to be required for cerebellar LTD (Ref. 44), but hippocampal LTD requires the GluR2–GRIP/ABP interaction43. Given these complexities, a comprehensive model seems far off; nevertheless, the evidence is strong for regulation of AMPAR localization by specific protein–protein interactions at the PDZ binding site.
Triggered by the silent synapse hypothesis, the main drive for investigating AMPAR recycling derives from the desire to understand the cellular mechanism of synaptic plasticity, and, in particular, LTP and LTD. If the enhanced synaptic efficacy associated with LTP is the result of an increase in the number of synaptic AMPARs, one has to postulate a mechanism, which delivers more receptors to the synapse. This could be achieved by a transient increase of the rate of exocytosis. Alternatively, in the presence of constant vesicle turnover, vesicles could be filled with more AMPARs, causing a larger number to be delivered to the surface with each fusion event. Both of these models predict exocytosis would be essential for LTP. Indeed cleavage of postsynaptic v-SNAREs by introducing either botulinum toxin or tetanus toxin into CA1 neurons prevented LTP expression15. In these experiments, the effects of these toxins on constitutive cycling were not observed. However, unlike studies directly examining constitutive cycling, these experiments were carried out with sharp electrodes. This has the technical advantage that wash-out of LTP seen with patch-clamp recording is prevented, but also has the limitation that considerable time (10–15 min) is required for the input resistance of the cell to recover after impalement by the electrode. The synaptic rundown caused by toxins that block exocytosis occurs on a 10–20 min time-scale, and is expected to have largely occurred before stabilization of sharp electrode recordings. Because of these technical considerations, the effects of these agents on constitutive recycling and LTP induction have not yet been examined by the same methods in the same cells. Nevertheless, by monitoring fluorescence of the styryl dye FM1-43 (Ref. 46) incorporated into postsynaptic membrane organelles, it was shown that high-frequency stimulation similar to the kind used to induce LTP led to Ca2+-dependent exocytosis. Evidence that this postsynaptic exocytosis delivers AMPARs comes from experiments in which transfection of CA1 neurons in organotypic slice cultures with a green fluorescent protein (GFP)–GluR1 fusion protein and application of an LTP induction protocol triggers the translocation of this protein from dendritic shafts to the spines30. Using rectification as an electrophysiological tag, the authors could show that GluR1 homomeric receptors were indeed synaptically inserted47. LTP-induced AMPAR insertion appears to require the GluR1 subunit, as mice deficient in GluR1 lack LTP (Ref. 35), but can be rescued by transfection of the GluR1 subunit into CA1 pyramidal cells of organotypic cultures made from GluR1−/−mice48. Experiments with mutant GluR1 subunits determined that a PDZ interaction site located at the C terminal was needed for synaptic insertion47.
If LTD is a functional reversal of LTP, and if LTP is the result of receptor insertion, then LTD might be expected to be caused by receptor internalization. Indeed, blockade of postsynaptic endocytosis using peptides that interfere with dynamin function prevents the induction of LTD (Ref. 12). In cultured hippocampal neurons, where changes in the synaptic AMPARs can be directly visualized, LTD induction led to decreases in synaptic efficacy that were associated with a loss of synaptic GluR1 puncta but not NMDAR1 puncta49. The pool of AMPARs internalized by LTD appears to overlap to a large extent with the constitutively recycling pool, as the rundown of EPSCs in the presence of botulinum toxin occludes subsequent LTD (Ref. 12). Furthermore, disrupting the NSF–GluR2 interaction also occluded LTD (Refs 12,50). However, LTD is also blocked by peptides that prevent the interaction between GluR2 and PDZ domains (see above), even though these peptides only have sporadic effects on baseline transmission. This result could indicate that some of the receptors internalized by LTD come from a stabilized pool, or alternatively that AMPAR–PDZ interactions do not exclude AMPARs from the recycling pool. It is interesting to note that in cerebellar LTD (an NMDAR-independent process), interference with the clathrin endocytotic complex or PDZ-domain interactions also block changes in synaptic efficacy44,51. This suggests that the dynamic modulation of the number of postsynaptic receptors is a mechanism to regulate synaptic strength shared by several forms of long-term synaptic plasticity.
In conclusion, a substantial body of evidence supports the idea that AMPA receptors are dynamic components of the postsynaptic density. It is clear that at least some AMPARs are mobile, even under normal conditions, and that changes in AMPAR traffic can be induced by exogenous manipulations. Long-lasting forms of synaptic plasticity also appear to cause changes in the synaptic complement of AMPARs, although the available evidence does not preclude additional changes through other mechanisms (e.g. conductance change52). Preliminary insights have also been made in several areas, including how receptors are sorted intracellularly, what signals determine synaptic localization on the surface, and what proteins interact with AMPARs to allow trafficking to occur.
However, several questions remain. If surface AMPARs are constantly being exchanged with an intracellular pool, where are the intracellular AMPAR-containing vesicles? A pool of AMPARs in dendritic spines has been reported, but only at synapses that already have a pronounced synaptic AMPAR population11. One possibility is that these vesicles are located on dendrites near the spine for weaker synapses, but the relative paucity of these vesicles is troubling and a comprehensive, high-resolution search is warranted. Another issue to be resolved is whether there are distinct pools of AMPARs that recycle, or that are moved during synaptic plasticity, or that interact with proposed tethers. If such a distinction can be made, the features of AMPARs that restrict them to a given pool will also be of interest. Current evidence suggests that subunit composition might be such a feature, but the rules that regulate endogenous receptors, most of which contain multiple subunit types, are not yet elucidated. Another obvious possibility with some support is that phosphorylation and dephosphorylation events represent the cue that allows switching between pools. Although the answers to these questions remain obscure, one thing is clear – AMPA receptors don’t get much rest, and neither will the people who study them.
We thank R. Carroll and R. Nicoll for helpful discussions and G. von Kaenel for technical assistance with the figure. C.L. is supported by grants of the Swiss National Science Foundation and the Human Frontier Science Program.
Christian Lüscher, Depts of Pharmacology (APSIC), Physiology, and Neurology (NEUCLID), University of Geneva, 1211 Geneva 4, Switzerland.
Matthew Frerking, Neurological Sciences Institute, Oregon Health Sciences University, 505 NW 185th Ave, Beaverton, OR 97006, USA.