Long-term depression in the CNS

Key Points

• Long-term depression (LTD) encompasses a family of synaptic plasticity mechanisms that can be triggered by the synaptic or pharmacological activation of glutamate receptors — in particular NMDARs (N-methyl-D-aspartate receptors) and metabotropic glutamate receptors (mGluRs) — or receptors for other neurotransmitters.

• LTD is expressed by a long-lasting decrease in the efficiency of synaptic transmission, in particular synaptic transmission that is mediated by the synaptic activation of AMPARs (α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid receptors). It may involve presynaptic and postsynaptic mechanisms.

• Complex signalling cascades link the induction of LTD to its expression. The cascades involve Ca²⁺ sensors, protein–protein interactions, protein kinases and phosphatases, proteases and other signalling molecules.

• Historically, a lack of specific inhibitors for LTD has hampered efforts to specify its functional role. The recent development of interference peptide inhibitors targeted at the carboxyl tail of the AMPAR subunit GluA2 subunit has proven important in specifying the functional roles of LTD.

• LTD has diverse roles in cognition, particularly in some forms of learning and memory and in circumstances in which a flexible response is required.

• LTD also seems to be involved in pathological states, including drug addiction, mental retardation and neurodegenerative diseases such as Alzheimer's disease.

Abstract

Long-term depression (LTD) in the CNS has been the subject of intense investigation as a process that may be involved in learning and memory and in various pathological conditions. Several mechanistically distinct forms of this type of synaptic plasticity have been identified and their molecular mechanisms are starting to be unravelled. Most studies have focused on forms of LTD that are triggered by synaptic activation of either NMDARs (N-methyl-D-aspartate receptors) or metabotropic glutamate receptors (mGluRs). Converging evidence supports a crucial role of LTD in some types of learning and memory and in situations in which cognitive demands require a flexible response. In addition, LTD may underlie the cognitive effects of acute stress, the addictive potential of some drugs of abuse and the elimination of synapses in neurodegenerative diseases.

Main

The two major forms of long-lasting synaptic plasticity in the mammalian brain — long-term potentiation (LTP) and long-term depression (LTD) — are characterized by a long-lasting increase or decrease in synaptic strength, respectively. Both processes are thought to be involved in information storage and therefore in learning and memory and other physiological processes. The last few years have seen rapid advances in our understanding of the molecular mechanisms that are involved in the generation of LTD and in the functions of LTD in health and disease.

There are several types of LTD and these can be defined in various ways (Box 1). LTD can be homosynaptic (induced in the conditioned input) or heterosynaptic (induced in a non-conditioned input) and can be induced de novo or following LTP (in which case it is called depotentiation). Although these different forms of LTD may seem similar, they use distinct molecular mechanisms and probably have different functions. A clear definition of, and distinction between, the various forms of LTD is therefore crucial for the understanding of this family of synaptic plasticity mechanisms.

LTD can be induced by prolonged periods of low-frequency stimulation (LFS), by pairing baseline synaptic stimulation with depolarization (known as pairing), by appropriately timed back-propagating action potentials (a form of spike-timing dependent plasticity (STDP)), or by application of an appropriate receptor agonist (known as chemical LTD (chem-LTD)) (Box 1). For a more detailed description of the discovery and characterization of these forms of LTD see Ref. 1.

In this Review, we first discuss the mechanisms that are involved in the generation of LTD, classified according to their induction, expression and signalling mechanisms. We then focus on the role of LTD in physiological and pathological processes in different regions of the CNS.

Mechanisms of induction

Most synapses that undergo LTD use L-glutamate as their neurotransmitter. L-glutamate acts on NMDA (N-methyl-D-aspartate) receptors (NMDARs), AMPA (α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid) receptors (AMPARs), kainate receptors (KARs) and metabotropic glutamate receptors (mGluRs)².

NMDAR-dependent LTD. Depotentiation³ and de novo LTD⁴ require, like LTP⁵, NMDAR activation at many different synapses in the brain (Supplementary information S1 (table)). Thus, NMDARs are widespread triggers for synaptic plasticity, but do not determine the direction of change in synaptic efficiency.

NMDAR-dependent LTD (NMDAR-LTD) is usually induced by LFS but can also be activated as a form of STDP. In addition, a brief application of NMDA can lead to a long-lasting depression of synaptic transmission (a form of chem-LTD)⁶. Mutual occlusion between LFS-induced LTD and NMDA-induced synaptic depression⁷ suggests that they use common mechanisms. However, there are also important differences between LFS-induced and chemically induced LTD^8,9. The following paragraphs refer to LFS-induced NMDAR-LTD except where chem-LTD is specifically mentioned.

NMDARs are tetramers of various subunits (GluN subunits)². They are usually composed of two GluN1 subunits and two GluN2 subunits, although GluN3 subunits sometimes replace GluN2. The two GluN2 subunits can be identical (GluN2A, GluN2B, GluN2C or GluN2D), forming a diheteromer, or they can be different from each other, forming a triheteromer together with two identical GluN1 subunits¹⁰.

Although genetic and pharmacological studies have suggested that NMDAR-LTD of AMPAR-mediated synaptic transmission (LTD(A)) involves the activation of specific NMDAR subtypes¹¹, there is considerable flexibility — for example, GluN2B-containing NMDARs are required for LTD, but only under certain circumstances^{12,13,14,15,16}. It is therefore likely that various NMDAR subtypes can trigger LTD and that the actual subtype(s) that are involved depend on various factors, such as the induction protocol that is employed, expression levels (which vary according to brain region and developmental stage¹⁷) and environmental conditions (for example, access to a running wheel)¹⁸. In addition, in young adult animals proBDNF (the precursor of brain-derived neurotrophic factor (BDNF)), acting via neurotrophin receptor P75 (p75^NTR), increases GluN2B expression and so enables a GluN2B-sensitive form of NMDAR-LTD to occur¹⁹. Furthermore, hippocampal NMDAR-LTD that is induced in adult animals by blocking L-glutamate uptake may also be dependent on GluN2B activation^12,13.

De novo NMDAR-LTD in CA1 is pronounced early in development, but is more difficult to induce in brain slices from adult animals^20,21 or in intact rodent hippocampi in vivo^12,22,23. However, hippocampal LTD can be induced using certain protocols²³ and is facilitated by exposing an animal to novelty²⁴ or mild stress^12,20,21,22. The effects of stress can be mimicked by blocking L-glutamate uptake^12,21. As the CNS develops, glutamate transporter mechanisms may limit activation of the NMDARs that would otherwise trigger LTD. Some studies have also shown strain differences, with LTD being more easily induced in Wistar and Sprague–Dawley strains than in hooded rat strains^25,26. However, there is also evidence that NMDAR-LTD is important for certain forms of learning and memory in the adult animal in the absence of stress (see below).

mGluR-dependent LTD. A second major form of LTD requires the activation of mGluRs. The patterns of activation that are required to induce mGluR-LTD are generally similar to those required to induce NMDAR-LTD, although there are differences depending on the synapse type. For example, NMDAR-LTD is usually induced at CA1 synapses by single-shock LFS, whereas mGluR-LTD is usually induced by paired-pulse LFS²⁷. A form of mGluR-LTD can also be induced by the application of the group I mGluR agonist 3,5-dihydroxyphenylglycine (DHPG)²⁸.

The finding that the weak group I mGluR antagonist L(+)-2-amino-3-phosphonopropionic acid (LAP3) blocks de novo LTD at CA1 synapses provided the first suggestion that mGluRs may be involved in LTD²⁹. The first selective mGluR antagonist that was discovered, α-methyl-4-carboxyphenylglycine (MCPG)³⁰, blocks both depotentiation¹¹ and de novo LTD³¹ in CA1. Subsequent studies revealed that several mGluR subtypes can mediate LTD and that the subtype involved depends on the brain region — for example, mGluR1 and mGluR2 receptors mediate de novo LTD at cerebellar parallel fibre–Purkinje cell synapses³² and hippocampal–mossy fibre synapses³³, respectively.

Although NMDAR-LTD and mGluR-LTD involve distinct receptors and signalling mechanisms, some types of LTD require synergistic interactions between the two receptor subtypes²⁸. In the perirhinal cortex, LTD involves a tripartite interaction between NMDARs, group I mGluRs and group II mGluRs³⁴.

Less common inducers of LTD. Ca²⁺-permeable AMPARs can trigger LTD in a subset of inhibitory neurons in the CA3 area of the hippocampus³⁵ and a form of LTD that was described in the perirhinal cortex is induced by the activation of kainate receptors³⁶. There are also examples of LTD that do not require activation of any class of glutamate receptor³⁷, such as LTD that is induced by activation of muscarinic receptors³⁸. Various forms of chem-LTD do not involve the activation of glutamate receptors, but in most cases it is unknown whether they can also be induced by the action of a synaptically-released neurotransmitter or modulator.

In most studies, LTD involves alterations in AMPAR-mediated synaptic transmission, but can also involve alterations in NMDAR³⁹-, KAR³⁶-, and mGluR⁴⁰-mediated synaptic transmission (known as LTD(N), LTD(K) and LTD(mGluR), respectively). These other forms of LTD have begun to be investigated — for example, it has been shown that LTD(N) is typically triggered by the activation of NMDARs and in some cases³⁹, but not always⁴¹, co-activation of mGluRs is required.

Mechanisms of expression

LTD is mediated by persistent pre- and postsynaptic changes, and the proportion of presynaptic compared with postsynaptic alterations probably depends on various factors, including the type of synapse and the developmental stage of the animal. It is possible that LTD expression involves shrinkage and elimination of both pre- and postsynaptic elements^42,43,44,45, but the mechanisms behind such LTD-associated structural changes are not well understood. Below, we discuss mechanisms of LTD expression about which more data exist.

Alterations in glutamate release. There is evidence that LTD⁴⁶ — including NMDAR-LTD⁴⁷ and mGluR-LTD⁴⁸— can involve a reduction in the probability of glutamate release. This could be triggered by changes in the presynaptic terminal or by postsynaptic changes that are communicated across the synapse via a retrograde messenger. Several retrograde messengers have been proposed to be involved in LTD, including nitric oxide in NMDAR-LTD⁴⁷ and lipoxygenase metabolites in hippocampal mGluR-LTD⁴⁹. Endocannabinoids can also function as retrograde messengers in the striatum⁵⁰, neocortex⁵¹ and cerebellum⁵². In the hippocampus the endocanabinoids act as mediators of a form of heterosynaptic mGluR-induced LTD that is prominent early in development⁵³. Endocannabinoid release during LTP can also lead to LTD of GABA (γ-aminobutyric acid)-mediated synaptic transmission and this affects the subsequent plasticity of the network⁵⁴.

Alterations in receptors. NMDAR-LTD(A) is mainly a postsynaptic phenomenon and this has been most directly demonstrated in hippocampal slices using L-glutamate uncaging in close proximity to synapses at which LTD was induced chemically or with LFS^55,56. Although various mechanisms may underlie the decrease in sensitivity to L-glutamate, most of the evidence points to the removal of AMPARs from the synapse¹¹. However, LTD can also be expressed by an alteration of the conductance properties of the receptors. Specifically, at synapses at which LTP is associated with an increase in single-channel conductance (γ), subsequent depotentiation involves a decrease in this parameter⁵⁷. In the same study, γ was not altered during de novo LTD or during depotentiation when γ had not been increased by LTP. This implies that multiple mechanisms of LTD can coexist at the same population of synapses.

mGluR-LTD can also involve postsynaptic receptor changes. In the cerebellum, mGluR-LTD involves reduced postsynaptic sensitivity to L-glutamate⁵⁸. However, the mechanisms of AMPAR trafficking in NMDAR-LTD and mGluR-LTD may differ — for example, in the CA1, NMDAR-LTD but not DHPG-LTD (a form of mGluR-LTD that is induced by application of DHPG) was associated with a decrease in sensitivity to L-glutamate⁵⁶. It is worth noting that AMPARs are probably not the only type of receptor that are modified in LTD as there is evidence that DHPG-LTD(N) involves the internalisation of NMDARs⁵⁹.

Induction to expression

NMDAR-LTD and mGluR-LTD use different signal transduction mechanisms and these have been most extensively investigated in the context of AMPAR trafficking (Figs 1, 2).

Figure 1: Signalling mechanisms involved in NMDAR-dependent LTD.

a | Calmodulin (CaM) detects Ca²⁺ (shown by graded purple clouds) that enters via NMDARs and this leads, through a Ser/Thr protein phosphatase cascade, to activation of protein phosphatase 1 (PP1) a key enzyme in synaptically-induced LTD. PP1 can dephosphorylate various targets, including ser845 on the AMPAR subunit GluA1 and ser295 of postsynaptic density protein 95 (PSD95). b | GluA2-containing AMPARs are stabilised at synapses by an interaction with N-ethylmaleimide-sensitive factor (NSF). The neuronal calcium sensor protein hippocalcin (HPC) is a high-affinity Ca²⁺ sensor that can target adaptor protein 2 (AP2) to GluA2 and therefore displace NSF and initiate clathrin-mediated endocytosis of AMPARs. Ras-related protein (RalA)-binding protein 1 (RalBP1) may also be involved in the NMDAR-dependent targeting of AP2, where it associates with RalA. c | In some circumstances, protein interacting with C kinase 1 (PICK1) may aid the NMDAR-dependent disassociation of AMPARs from AMPAR-binding protein–glutamate receptor interacting protein (ABP–GRIP), potentially via the targeted phosphorylation of ser880 of GluA2 by protein kinase Cα (PKCα). PICK1, by binding actin-related protein 2/3 (Arp2/3) and F-actin, also acts as a negative regulator of Arp2/3-mediated actin polymerization. d | NMDAR-LTD is associated with phosphorylation (by protein tyrosine kinases (PTKs)) of tyr876 of GluA2 and this may also aid the exchange of PICK1 for ABP–GRIP. e | Glycogen synthase kinase-3β (GSK3β) is required for NMDAR-LTD during which it can be activated by PP1. The upstream regulators of GSK3β (the phosphoinositide 3-kinase (PI3K)–Akt pathway) enable the direct regulation of LTD by LTP. The release of cytochrome c from the mitochondria may activate caspase-9 and caspase-3, which can cleave Akt, possibly resulting in GSK3β activation. AKAP, A-kinase anchor protein; ARAP, AMPAR-associated protein; I-1, inhibitor 1; PP2B, protein phosphatase 2B; RyR, ryanodine receptor.

Figure 2: Signalling mechanisms involved in mGluR-LTD(A).

A | Signalling mechanisms that are involved in metabotropic glutamate receptor (mGluR)-dependent, AMPAR-mediated long-term depression (LTD(A)) in the hippocampus. Aa | Stimulation of group I mGluRs (here, mGlu5) leads to activation of phosphoinositide-specific phospholipase C (PLC). This can trigger the release of Ca²⁺ from intracellular stores and the activation of protein kinase C (PKC). In some forms of mGluR-LTD, PICK1 may target PKCα to phosphorylate ser880 of GluA2 to displace AMPAR-binding protein and glutamate receptor interacting protein (ABP–GRIP) and permit the removal of AMPARs from synapses. GRIP may then be sequestered by microtubule associated protein 1B (MAP1B). NCS1 has been implicated as a Ca²⁺ sensor and as a targeting molecule that is involved in this cascade. Ab | Several lines of evidence implicate p38 mitogen-activated protein kinase (p38 MAPK) and, to a lesser extent, extracellular signal-regulated kinases (ERKs) in mGluR-LTD, however, the downstream effectors are largely unknown. Ac | Activation of protein tyrosine phosphatases (PTPs) is also required for mGluR-LTD. Although the identity of the PTPs that are involved has not been firmly established, one candidate is striatal-enriched protein phosphatase (STEP). Ad | Arc is also involved in mGluR-LTD and it may help to initiate dynamin-dependent endocytosis of AMPARs. Ae | In some studies mGluR-LTD has been shown to require rapid (in a few minutes) de novo protein synthesis, with Arc, MAP1B and STEP as candidate protein molecules and eukaryotic elongation factor-2 kinase (eEF2K/eEF2) as one of the putative regulators of translation. Af | There is also evidence that the phosphoinositide 3-kinase (PI3K)-Akt-mammalian target of rapamicin (mTOR) pathway may control translation during mGluR-LTD. B | Signalling mechanisms that are involved in mGluR-LTD(A) in the cerebellum. At parallel fibre synapses onto Purkinje cells in the cerebellum, a form of mGluR-LTD(A) that requires activation of mGlu1 receptors has been extensively characterized. There are several similarities between cerebellar mGluR-LTD(A) and hippocampal mGluR-LTD(A) but some striking differences have also been found. Most notably, this form of LTD involves Ca²⁺ entry through voltage-gated Ca²⁺ channels (VGCC) and there is a requirement for 'orphan' glutamate receptor delta 2 (GluD2) subunits. A role for the nitric oxide–cyclic guanosine monophosphate (NO–cGMP) cascade has also been identified. Ser/Thr phosphatases negatively regulate this form of LTD(A) (not shown). CaM, calmodulin; CaMKII, Ca²⁺/calmodulin-dependent protein kinase II; DAG, dyacylglycerol; IP3, inositol trisphosphate; IP3R, inositol trisphosphate receptor; PIKE, PI 3-kinase enhancer; PP1, protein phosphatase 1; PP2A, protein phosphatase 2A; PP2B, protein phosphatase 2B.

NMDAR-LTD(A). In NMDAR-LTD(A), Ca²⁺ that enters through NMDARs binds to calmodulin to activate protein phosphatase 2B (PP2B; also known as calcineurin), which dephosphorylates inhibitor-1 and this leads to the activation of protein phosphatase 1 (PP1)⁶⁰ (Fig. 1). PP1 then dephosphorylates its substrate(s), including ser845 on the AMPAR subunit GluA1 (Ref. 11 and this leads to LTD. In addition, Ca²⁺ entry triggers Ca²⁺ release from intracellular stores⁶¹ and this may serve to activate Ca²⁺ sensitive enzymes that are located further away from the postsynaptic density (PSD), where endocytosis may occur.

The first clue to the molecular mechanism that drives the endocytosis of AMPARs during NMDAR-LTD(A) was the observation that disruption of an interaction between GluA2 and N-ethylmaleimide-sensitive factor (NSF; an ATPase involved in membrane fusion events) causes AMPAR internalisation, mimicking NMDAR-LTD(A)¹¹. Later, it was shown that clathrin-mediated endocytosis was involved in this process and that the clathrin adaptor protein AP2 also binds to the NSF site on the GluA2 subunit. This suggests that AMPARs are stabilised on the membrane by NSF and that AP2 replaces NSF to initiate AMPAR endocytosis during NMDAR-LTD(A)¹¹.

A potential mechanism for the triggering of this exchange involves hippocalcin, a member of the neuronal calcium sensor (NCS) family. On sensing small rises in Ca²⁺ (10⁻⁷ to 10⁻⁵ M) hippocalcin translocates to the plasma membrane, where it forms a complex with AP2 and GluA2 that may initiate clathrin-mediated AMPAR endocytosis⁶². Consistent with this model, inhibition of hippocalcin function blocks NMDAR-LTD(A)⁶². Another AP2 targeting molecule may be Ras-related protein (RalA)-binding protein 1 (RalBP1), which binds to postsynaptic density protein 95 (PSD95) during NMDAR-LTD(A), following activation of the small GTPase RalA⁶³. In addition, the small GTPase Rab5 has been implicated in LTD⁶⁴. Other steps in clathrin-mediated endocytosis that underlies NMDAR-LTD(A) are also beginning to be determined — for example, calcyon, a protein that regulates clathrin assembly is involved in NMDAR-LTD(A) in the hippocampus⁶⁵.

Protein interacting with C kinase 1 (PICK1) is another protein that binds directly to GluA2 and has been implicated in NMDAR-LTD(A). PICK1 is a low-affinity Ca²⁺ sensor⁶⁶ that can also bind protein kinase Cα (PKCα) and that can sense, and perhaps induce, membrane curvature. PICK1 competes with the scaffolding proteins AMPAR-binding protein (ABP) and glutamate receptor interacting protein (GRIP) for binding to the carboxy-terminal (C-terminal) region of GluA2 and it also promotes internalization of GluA2-containing AMPARs¹¹. It was originally proposed that during NMDAR-LTD(A) PICK1 may promote the synaptic removal of AMPARs by inducing the PKCα-mediated phosphorylation of ser880 of GluA2 to dissociate AMPARs from ABP–GRIP. However, most of the experimental data do not support this hypothesis. NMDA-induced internalisation of AMPARs is not dependent on the phosphorylation status of ser880 or on PICK1 (Ref. 67) and PKC inhibitors do not affect NMDAR-LTD(A)^68,69. Experiments that use inhibitors of the GluA2–PICK1 interaction have yielded conflicting results. One study acutely applied an interfering peptide and reported no effect on NMDAR-LTD(A), whereas a subsequent report described a partial inhibition¹¹. A small-molecule inhibitor of the GluA2–PICK1 interaction was also found to partially inhibit NMDAR-LTD(A)⁷⁰. These partial effects are in contrast with the abolition of NMDAR-LTD(A) that results from chronic manipulation of PICK1 (Ref. 71). Thus, the presence of PICK1 may be necessary for NMDAR-LTD(A) in the long-term but it probably plays only a minor part, if any, in the release of AMPARs from their synaptic tethers.

A more important role of ABP–GRIP may be to anchor AMPARs at non-synaptic sites (intracellular or extrasynaptic sites on the plasma membrane)¹¹, as recently confirmed by paired-cell recording experiments⁷². NMDAR-LTD(A) becomes unstable if the ability of AMPARs to bind ABP–GRIP is impaired. This implies that by retaining AMPARs at non-synaptic sites this scaffolding molecule is crucial for the expression of this form of LTD. PICK1 may enable the disassociation of AMPARs from ABP–GRIP at these non-synaptic sites, thereby enabling de-depression (re-potentiation) of synaptic transmission¹¹. Another key function of PICK1 in NMDAR-LTD(A) may be to enable actin depolymerization through an interaction with F-actin and the actin-related protein 2/3 (Arp2/3 complex), and through this process to modify neuronal architecture⁷³.

NMDAR-LTD(A) is classically assumed to require the activation of phosphatases. However, studies using kinase inhibitors have implicated various serine/threonine (Ser/Thr) protein kinases in this process as well. These include protein kinase A (PKA)⁷⁴, cyclin-dependent kinase 5 (Ref. 75), P38 mitogen-activated protein kinase (p38MAPK)⁷⁶ and glycogen synthase kinase-3 (GSK3)^68,77. As with all inhibitor studies, potential off-target effects should be taken into account and this is particularly important for protein kinases because the mammalian genome encodes over 500 protein kinases. A role for GSK3 in NMDAR-LTD(A) is supported by the effects of six different GSK3 inhibitors⁶⁸ including lithium, which may exert some of its therapeutic actions via this mechanism. A direct link between the protein phosphatase cascade and GSK3 was observed during NMDAR-LTD(A); PP1 dephosphorylates GSK3β and its upstream inhibitor Akt and these actions result in activation of GSK3β⁷⁷ (Fig. 3). An additional mechanism of GSK3 activation may occur via caspase-3 (Ref. 78). This protease is activated during NMDAR-LTD(A) through a cascade involving cytochrome c and caspase-9 and is able to cleave Akt, thus removing its tonic inhibition of GSK3 (Ref. 78). The finding that both caspases and GSK3β, an enzyme that is deregulated in patients with Alzheimer's disease, are involved in NMDAR-LTD raises the possibility that the neurodegeneration that underlies Alzheimer's disease and related dementias may be caused at least in part by pathological activation of this form of LTD (Box 2).

Figure 3: Molecular interactions between long-term potentiation and long-term depression.

a | A schematic of electrode placements for induction of input-specific long-term potentiation (LTP) and long-term depression (LTD) in CA1. The hippocampal slices were obtained from 2-week-old rats. The box shows the dendritic region where enzyme activity was assessed. b | LTP can be induced after 60 shocks at 0 mV as shown by the persistent increase in the amplitude of excitatory postsynaptic currents (EPSCs) after stimulus (stim) application (top left). There is no change in response following an LTP stimulus due to washout of unknown constituents that are required for LTP when baseline recording is extended beyond about 10 minutes (top right). De novo LTD can be induced by 300 shocks at -40 mV and becomes evident by a persistent decrease in EPSC amplitude (bottom left). De novo LTD is inhibited — leaving just a transient depression — if an LTP stimulus (delivered after washout of LTP) is applied first (bottom right). c | Glycogen synthase kinase-3 β (GSK3β) is inhibited after an LTP-inducing stimulus (by phosphorylation at ser9), whereas LTD-inducing stimuli have the opposite effect. d | A diagram showing the cellular mechanism for LTD inhibition by LTP. An LTD-inducing stimulus activates protein phosphatase 1 (PP1), which dephosphorylates GSK3β to activate it and permit the induction of LTD. An LTP stimulus activates the phosphoinositide 3-kinase (PI3K)-Akt pathway, which phosphorylates GSK3β to inhibit it, thus preventing LTD. (Note that the LTP stimulus can inhibit LTD without inducing LTP, as the regulation can occur after washout of LTP.) *, significant difference from control; GluN1, NMDAR (N-methyl-D-aspartate receptor) subunit 1; GluN2, NMDAR subunit 2; GluA1, AMPAR subunit 1; GluA2, AMPAR subunit 2. Parts b and c are modified, with permission, from Ref. 77 © (2007) Cell Press.

NMDAR-LTD(A) is associated with tyrosine phosphorylation of GluA2 and this suggests that protein tyrosine kinases (PTKs) are also involved. PTKs of the sarcoma (Src) family phosphorylate GluA2 at a tyrosine residue in a tyrosine-rich region of the C-terminal tail of GluA2 and this is thought to be required for AMPAR endocytosis¹¹. Consistent with this idea, a peptide that mimics this tyrosine-rich region has been found to block NMDAR-LTD(A)⁷⁹ (Box 3).

What are the targets of enzymes that are activated during NMDAR-LTD(A) and underlie an alteration in the synaptic expression of AMPARs? A major target seems to be PSD95, which positions calcineurin near the mouth of the NMDAR channel through an interaction with A-kinase anchor protein (AKAP)-150 (Ref. 80) and which is dephosphorylated on ser295 during LTD to enable the removal of PSD95 from the synapse and thereby permit AMPAR endocytosis⁸¹.

These mechanisms occur rapidly after NMDAR-LTD(A) is triggered. However, protein synthesis is required for LTD to be sustained as inhibitors of translation cause a recovery of synaptic transmission in a few hours²⁵. How these newly synthesized proteins sustain LTD for longer periods of time is not known but regulators of gene transcription that may be involved in NMDAR-LTD(A) are starting to be investigated⁸².

mGluR-LTD(A). mGluR-LTD(A) involves signal transduction mechanisms that are different from those that underlie NMDAR-LTD(A)⁸³. Any of the seven mGluR subtypes that are expressed in the brain could conceivably trigger LTD, potentially through different cellular mechanisms. mGluR-LTD(A) in CA1 is triggered predominantly through activation of mGlu5 receptors (Fig. 2), although mGlu1 receptors may initiate certain forms of LTD(A) in this region too⁸⁴. A clearer picture of the role of mGlu1 receptors in mGluR-LTD(A) has emerged from studies of the parallel fibre–Purkinje cell synapse in the cerebellum, where their expression levels are high³² (Fig. 2).

The canonical signalling pathway of group I mGluRs involves the hydrolysis of phosphatidyl inositol to generate inositol trisphosphate (IP3) and diacylglycerol (DAG), which in turn can activate PKC. This pathway is involved in mGluR-LTD(A) triggered by both mGlu1 and mGlu5 receptors in the hippocampus⁶⁹, cerebellum⁸⁵ and perirhinal cortex⁸⁶. A common way to induce mGluR-LTD(A) is to apply the group I selective agonist DHPG²⁸. Most strikingly, DHPG-LTD(A) can be induced in the absence of Ca²⁺ (Ref. 87) and is unaffected by inhibition of PKC⁸⁸. This could be because group I mGluRs can signal through both Ca²⁺ dependent and independent pathways, depending on how they are activated.

PICK1 is also required for mGluR-LTD(A) at different synapses^11,86,89. In the perirhinal cortex PICK1 forms a complex with the prototypic member of the NCS family NCS-1, which could act as a high-affinity Ca²⁺ sensor for mGluR-LTD(A)⁸⁶. As PICK1 can also bind PKCα, it is possible that NCS-1 attracts the PICK1–PKC complex to the GluA2 subunit of AMPARs in response to Ca²⁺ signals, resulting in phosphorylation of the subunit and disassociation from ABP–GRIP.

Several other protein kinases have been implicated in various forms of mGluR-LTD(A), including p38MAPK^90,91,92 and possibly extracellular signal-regulated kinase (ERK)⁹³. Phosphoinositide 3-kinase (PI3K) has been associated with DHPG-LTD(A)⁹⁴. In terms of dephosphorylation, mGluR-LTD(A) involves protein tyrosine phosphatases (PTPs)⁹⁰ rather than Ser/Thr protein phosphatases. DHPG-LTD involves the tyrosine dephosphorylation of GluA2 (Ref. 95) and this is associated with the endocytosis of AMPARs⁹⁶. that can be blocked by the GluA23Y peptide¹⁹¹ (Box 3).

In summary, the two most extensively studied forms of LTD — hippocampal NMDAR-LTD(A) and mGluR-LTD(A) — involve different protein kinases and phosphatases as well as different reciprocal changes in tyrosine phosphorylation. In addition, there may be two pools of synaptic AMPARs, one linked to ABP–GRIP and the other bound indirectly to PSD95 via a transmembrane AMPAR regulatory protein (TARP). As both forms of LTD can probably occur at the same synapse, this use of different anchoring proteins and the regulated disassociation via different phosphorylation cascades may preserve the independence of the two processes during the initial phase of the process (Figs 1, 2).

New protein synthesis is required for the expression of mGluR-LTD(A) under certain circumstances⁹⁷. However, mGluR-LTD(A) can also be induced in the presence of protein synthesis inhibitors⁹⁰ with no alterations in its magnitude. This suggests that, at least during its initial phase, mGluR-LTD can be independent of protein synthesis. A number of factors may determine whether protein synthesis is required and these may include the developmental stage of the animal⁹⁸.

In cases of LTD that involve protein synthesis, the protein synthesis-dependent component of LTD is observed within minutes of mGluR-LTD(A) induction, which clearly requires rapid de novo synthesis of one or more proteins⁸⁴. Candidates for these proteins are the immediate early gene Arc/Arg3.1 (Arc)¹⁹², striatal-enriched protein phosphatase (STEP)⁹⁹ and microtubule associated protein 1B (MAP1B)¹⁰⁰. All three proteins seem to be involved in the internalisation of AMPARs following mGluRs stimulation^99,100,101. How can mGluR-LTD(A) be both protein synthesis dependent and independent? One possibility is that crucial proteins, such as Arc, STEP and MAP1B may be in relatively short supply and are only present in sufficient amounts under certain circumstances to enable mGluR-LTD without new protein synthesis. Interestingly, mGluR-LTD(A) is facilitated in mice that lack the gene encoding fragile X mental retardation protein (FMRP)¹⁰² and it is also resistant to protein synthesis inhibitors¹⁰³. Therefore, by suppressing transcription, FMRP may (under normal circumstances) limit the synthesis of proteins and thereby confer susceptibility to protein synthesis inhibition. Collectively, these results suggest that two pathways — potentially one involving ERK and protein synthesis and the other involving p38MAPK — may converge at the level of PTPs and endocytic machinery (Fig. 2).

In addition to group I mGluRs, activation of other Gq-coupled receptors can also induce LTD(A) (known as Gq-LTD(A)) and there are indications that the signalling mechanisms that underlie these other forms of LTD are not necessarily the same. In the hippocampus, pharmacological activation of muscarinic (M) 1 receptors induced LTD(A), with the involvement of ABP–GRIP, liprin-α and the tyrosine phosphatase leukocyte common antigen-related (LAR) family receptor protein tyrosine phosphatase (LAR-RPTP) within the postsynaptic neuron¹⁰⁴ However, DHPG-LTD(A) occurred independently of this cascade, raising the possibility that Gq-LTD(A) may be highly compartmentalised in neurons. M1 receptors and mGluRs may also act synergistically in LTD, with M1 receptor providing a basal activation of PKC, which is further stimulated by mGluR activation resulting in LTD(A) via presynaptic mechanisms¹⁰⁵. Clearly the inter-relationships between different forms of LTD are complex but provide considerable flexibility and scope for modulation. In fact, mGluR-LTD(A) may be the result of interactions with several other neuromodulators, ranging from classical neurotransmitters to ephrins¹⁰⁶ as exemplified by the interactions between mGluRs and endocannabinoids or dopamine in the striatum⁸⁴.

Compared to LTD(A), far less is known about the signalling mechanisms involved in the other forms of LTD. A clearer picture of how different receptors trigger and express LTD and how these processes are modulated and affected is likely to emerge over the next few years.

Physiological functions of LTD

Establishing the functional role of LTD in the CNS has proved challenging due to the difficulty of inducing LTD in vivo and due to a lack of selective inhibitors for LTD that can be used in vivo^24,27. However, the development of novel strategies to target LTD has spurred research in this area^24,27,107 (Box 3). To assess whether LTD has a role in a given learning and memory-related behaviour, we will simplify and adopt three criteria from the 'synaptic plasticity and memory hypothesis' (better known as the SPM hypothesis)¹⁰⁸. These criteria are occurrence (that is, LTD can be induced experimentally in the neural circuits that are relevant for the cognitive function), necessity (that is, blocking LTD disrupts the function) and sufficiency (that is, induction of LTD produces behaviour similar to the function). Although these criteria have not been fulfilled for most of the cognitive functions that have been examined, in a few cases the data support a role for LTD. These cases are described below.

Hippocampus-dependent learning and memory. The hippocampus contributes to learning and memory processes, although its specific role remains a subject of debate^109,110,111. Evidence to support a crucial role for LTD in hippocampus-based learning and memory processes has been missing until recently, because many inhibitors of LTD also affect LTP (Box 3). However, genetic approaches that disrupt specific proteins and enzymes that are involved in hippocampal LTD have been used to investigate the function(s) of LTD. A study using conditional knockout mice showed that the selective ablation of GluN2B subunits in pyramidal neurons in CA1 and cortex specifically impairs CA1 NMDAR-LTD and results in deficits in several hippocampus-dependent learning and memory tasks, providing strong evidence for a key role of this particular form of LTD in memory formation¹⁴. However, the mice also had reduced numbers of dendritic spines and minor alterations in NMDAR-LTP¹⁴ and these may have contributed to the learning and memory deficits.

Genetic disruption of forebrain calcineurin activity impairs LTD at Schaffer collateral–CA1 synapses and these knockout mice also exhibit impairments in working and episodic memory¹¹². When expression of the transcription factor serum response factor (SRF) — which regulates transcription of many immediate-early genes — was disrupted specifically in the forebrain, the mutant mice showed reduced Schaffer collateral–CA1 LTD when LTD was induced by electrical stimulation, NMDA or the cholinergic agonist carbachol, but showed normal LTD when it was induced by DHPG. The mice displayed deficits in habituation to a novel environment and acquisition of hippocampus-dependent memory tasks. A study in transgenic mice in which activity of the Ser/Thr protein phosphatase 2A was inhibited in the forebrain¹¹³ showed reduced NMDAR-LTD and disrupted reversal learning in the water maze as well as non-match to place behaviour in the T-maze¹¹³. This suggested to the authors that hippocampal NMDAR-LTD may have a role in behavioural flexibility. Notably, the mice with disrupted forebrain calcineurin activity¹¹² showed similar electrophysiological and behavioural changes, supporting the assertion that NMDAR-LTD is necessary for behavioural flexibility.

Impaired reversal learning in the water maze was also associated with severely impaired hippocampal LTD in dopamine transporter knockout mice¹¹⁴. Haloperidol reversed both impairments, supporting the role of dopaminergic tone in mediating these effects. Finally, potentiating hippocampal NMDAR-LTD using the NMDAR co-agonist D-serine improved spatial reversal learning in mice, an effect that could be blocked with the GluN2B subunit-specific antagonist Ro25-6981 (Ref. 13). These findings suggest that mechanisms that are consistent with NMDAR-LTD are engaged in hippocampus-dependent tasks that require behavioural flexibility.

Recent studies also suggest the involvement of LTD in other aspects of hippocampus-mediated learning and memory^24,27, such as in novelty detection. In freely moving rats LTD is facilitated during exploration of complex environments that contain novel objects or novel arrangements of objects^115,116, whereas LTP is facilitated and LTD is inhibited during exploration of a novel environment alone^115,117. These effects are modulated by dopamine D1/5 receptors and 5-hydroxytryptamine 4 (5-HT4) receptors^115,117,118. Depotentiation in CA1 and the dentate gyrus has been shown when rats explore a novel environment^119,120 or a familiar environment containing novel objects¹¹⁶. These findings provide evidence that mechanisms that are consistent with LTD in CA1 underlie some aspects of novelty detection^24,27.

Fear conditioning in amygdala. The cellular mechanisms underlying amygdala LTD¹²¹ and fear extinction¹²² have attracted intense attention. The induction of one form of LTD in the lateral amygdala is NMDAR-dependent and its expression depends on GluA2-dependent endocytosis of AMPARs¹²¹. Moreover, the Tat-GluA2_3Y peptide, which prevents hippocampal CA1 LTD by inhibiting AMPAR endocytosis⁷⁹ (Box 3), can also prevent the expression of this amygdala LTD. Blocking AMPAR endocytosis using this peptide disrupts the extinction of Pavlovian fear memory^123,124 (Fig. 4) without affecting the acquisition or expression of fear conditioning¹²³. The effects of the peptide are probably due to the blockade of the depotentiation that is normally induced in the lateral amygdala during extinction training¹²⁴. mGluR and NMDAR antagonists also block this depotentiation during fear extinction, whereas the mGluR agonist DHPG induces it¹²⁴. Furthermore, fear conditioning induces an increase in AMPARs localized to synaptosomes in the lateral amygdala, whereas extinction training reverses this increase¹²⁴. These findings imply that the expression of LTD and depotentiation through regulated endocytosis of GluA2-containing AMPARs in the lateral amygdala plays an important part in the extinction of Pavlovian fear conditioning¹²⁴.

Figure 4: Schematic summary of the use of interference peptides to study the physiological functions of LTD.

The figure shows brain regions that have been targeted for infusion or viral expression of interference peptides in studies that focused on the physiological role of long-term depression (LTD). All peptides are introduced in Box 3 except the mGluR-ct peptide, which prevents binding of mGluR1 with Homer1b/c (see Ref. 172 for details). In the hippocampus GluA2_3Y was found to block the stress-induced disruption of spatial memory retrieval when administered systemically¹² and to prevent the disruptive effects of protein kinase M ζ (PKMζ) inhibition on object location memory¹²⁵. In layer IV of the visual cortex, G2CT was found to block the effects of monocular deprivation on ocular dominance and the depression of deprived-eye responses¹⁴⁸. G2CT was found to disrupt object recognition memory in the perirhinal cortex¹³¹. In the ventral tegmental area, mGluR-ct was found to block the reversal of acute cocaine synaptic potentiation¹⁷². In the amygdala, GluA2_3Y was found to impair conditioned fear extinction^123,124 and to prevent the disruptive effects of PKMζ inhibition on long-term memory for conditioned fear¹²⁵. In the nucleus accumbens, GluA2_3Y was found to block the expression of behavioural sensitization to amphetamine¹⁵⁵ and Pep2-EVKI was found to block drug-induced cocaine reinstatement¹⁸⁰. In the medial prefrontal cortex, GluA2_3Y was found to reduce the cue-induced reinstatement of heroin-seeking behaviour¹⁵⁶.

A recent study suggests that these processes may also have a role in long-term memory stabilization¹²⁵. Inactivation of the protein kinase C isoform M ζ (PKMζ) with the ZIP peptide impairs the maintenance of amygdala LTP and erases fear memory in rats¹²⁵. Application of the Tat-GluA2_3Y peptide before ZIP prevented both ZIP-induced LTP reversal (depotentiation) and behavioural impairment¹²⁵ (Fig. 4). Similar results were obtained in a hippocampus-dependent object location memory task¹²⁵, suggesting that the inhibition of GluA2-containing AMPAR endocytosis by PKMζ may be a general mechanism for memory maintenance. The delayed extinction and prevention of ZIP-induced memory erasure by the Tat-GluA2_3Y peptide¹²⁵ suggests that the expression of LTD and depotentiation may have a common mechanism that involves GluA2-dependent AMPAR endocytosis. Furthermore, it is also in agreement with the involvement of LTD and depotentiation in reversal learning mentioned above, further supporting a crucial role of LTD in response flexibility.

Recognition memory in perirhinal cortex. The perirhinal cortex (PRH) is involved in several types of learning, including recognition memory for objects^126,127. Consistent response decrements in the activity of PRH neurons during familiarity discrimination¹²⁸ suggest that mechanisms that are consistent with LTD in the PRH may be involved in recognition memory. The muscarinic receptor antagonist scopolamine has been shown to block LTD in PRH slices, disrupt the normally observed increase in PRH Fos expression in response to the animal seeing novel pictures and to impair performance in an object-based recognition memory task¹²⁹. The G2CT interference peptide inhibits GluA2-dependent AMPAR endocytosis by preventing the interaction between AP2 and the GluA2 subunit of AMPARs¹³⁰ (Box 3; Fig. 4), and blocks an NMDAR-dependent form of LTD in both hippocampal¹³⁰ and PRH¹³¹ slices. Lentiviral expression of G2CT in the PRH (used to examine the effects of blocking LTD expression on visual recognition memory¹³¹) disrupted object recognition (Fig. 4). Interestingly, NMDAR-LTD could be readily induced in PRH slices from control rats after performance of the object recognition task, but not in PRH slices from animals in which the peptide had been expressed¹³¹. Together, these data provide evidence that LTD is necessary for object recognition.

Cerebellar learning. Given the propensity of parallel fibre–Purkinje cell synapses for LTD^132,133, the role of LTD in cerebellar function has been the subject of considerable interest²⁷ but findings have been inconsistent. For example, (1R)-1-benzo theophen-5-yl-2[2-(diethylamino) ethoxy] ethanol hydrochloride, which blocks cerebellar LTD in vivo, fails to affect rotarod learning or classical eyeblink conditioning in rats¹³⁴. By contrast, genetic rescue techniques have implicated the C-terminus of the 'orphan' glutamate receptor delta 2 (GluD2) in LTD induction at parallel fibre–Purkinje cell synapses and in eyeblink conditioning¹³⁵. Mice deficient in the class I major histocompatability complex (MHCI) molecules H2-K^b and H2-D^b have a reduced threshold for the induction of LTD in the cerebellum and show enhanced motor learning and memory in the rotarod test¹³⁶. Studies using L7-PKCI transgenic mice — which have chronic PKC inhibition restricted to cerebellar Purkinje cells and compromised LTD at parallel fibre–Purkinje cell synapses — have implicated cerebellar LTD in the adaptation of the vestibulo-ocular reflex¹³⁷ and in the learning of procedural components of spatial tasks in which the solution requires the development of behavioural strategies¹³⁸. Mice lacking Ca²⁺/calmodulin-dependent protein kinase IV are deficient in the maintenance, but not the induction, of cerebellar LTD¹³⁹ and in memories that relate specifically to training-induced increases in the gain of the vestibulo-ocular reflex¹⁴⁰. These studies support an association between LTD and learning and memory in the cerebellum but a causal relationship remains to be demonstrated. As expression of LTD in the cerebellum seems to be dependent on GluA2-dependent AMPAR endocytosis^141,142,143, the interference peptides that disrupt AMPAR endocytosis (Box 3) may become a useful tool to clarify this role.

Development of the visual cortex. Synaptic plasticity is also important for experience-dependent maturation of neural circuits, such as the sensory experience-dependent alterations in ocular dominance columns of the visual cortex following monocular deprivation¹⁴⁴. It has been proposed that LTD mechanisms underlie the reduced responsiveness of cortical neurons to inputs from the deprived eye¹⁴⁵. Monocular deprivation induces synaptic depression via distinct mechanisms in layers II/III and layer IV of the visual cortex^146,147. A recent study showed that blocking GluA2-dependent AMPAR endocytosis using the G2CT peptide blocks the effects of monocular deprivation on ocular dominance and the depression of deprived-eye responses in layer IV of the visual cortex¹⁴⁸ (Fig. 4).

Acute-stress-induced impairments in learning and memory. Stress has diverse effects on the brain^107,149 and limbic structures are strongly influenced by stress^150,151. In the CA1 region of the dorsal hippocampus, acute stress enables NMDAR- and mGluR-dependent forms of LTD^12,20,22,152 but impairs the induction of LTP⁴⁸. The effects of stress on synaptic plasticity are the subject of several previous reviews^107,149.

Acute stress disrupts hippocampus-dependent memory retrieval^12,153 and hippocampal LTD seems to have a key role in this effect. First, acute stress facilitates the occurrence of hippocampal LTD^12,21,22, an effect that is mediated by corticosterone release and increased extracellular glutamate concentrations in the hippocampus^21,154. The increased glutamate concentrations enable the induction of LTD through spill-over activation of extrasynaptically localized, GluN2B-containing NMDARs^12,21. Second, hippocampal LTD is necessary for stress-induced memory impairment. This was demonstrated in a study that showed that the GluN2B subtype-specific NMDAR antagonist Ro25-6981 (which prevents LTD induction^12,13,21) or the Tat-GluA2_3Y peptide (which blocks regulated AMPAR endocytosis and therefore prevents LTD expression (Box 3)^79,155,156) prevents the deficits in spatial memory retrieval that are caused by acute stress in a water maze task¹² (Fig. 4). Ro25-6981 administration also blocks stress-induced deficits in spatial and object recognition memory and this suggests a common role for GluN2B NMDAR in the disruptive effects of stress on memory retrieval¹⁵⁷. Finally, hippocampal LTD is sufficient to induce memory impairments. Intra-cranial injection of the glutamate transporter inhibitor threo-β-benzyloxyaspartate (TBOA) mimics the effects of stress exposure by facilitating the induction of LTD (via a spill-over activation of GluN2B receptors) and disrupts spatial memory retrieval¹². These effects were reversed by systemic administration of the GluN2B receptor antagonist Ro25-6981 (Ref. 12). Together, these studies suggest that mechanisms that are consistent with hippocampal LTD mediate the effects of acute stress on spatial memory retrieval.

Fragile X syndrome. There is strong evidence that mGluR-LTD is involved in the pathogenesis of fragile X syndrome¹⁵⁸. Electrophysiological hippocampal recordings have demonstrated that mice lacking fragile X mental retardation protein (FMRP) exhibit enhanced protein synthesis-independent LTD mediated by activation of group I mGluRs¹⁰². This enhancement may account for the abnormally long thin dendritic spines and the cognitive deficits observed in patients with fragile X syndrome¹⁵⁹.

FMRP knockout mice show normal learning in the inhibitory avoidance paradigm. However, their extinction rates are increased — an effect that can be reversed by genetically reducing mGluR5 signalling¹⁵⁸. Although this facilitated extinction has not been directly linked to the exaggerated hippocampal mGluR-LTD, there is an interesting parallel between this form of extinction learning and the extinction of Pavlovian fear conditioning that seems to depend on the expression of LTD and GluA2-containing AMPAR endocytosis in the amygdala^123,124. Therefore, both NMDAR and mGluR-dependent forms of LTD may be important for the extinction of previously learned memories.

Psychiatric disorders. Alterations in neuroplasticity may underlie some of the symptoms of psychiatric disorders such as depression, schizophrenia and bipolar disorder^160,161,162. Enhanced CA1 LTD was demonstrated in a rat model of depression and could be blocked by the antidepressant fluvoxamine¹⁶³. Altered synaptic plasticity has also been noted in several animal models of schizophrenia. Calcineurin knockout mice specifically lacks LTD in CA1 and shows deficits in working and reversal memory¹¹², and behavioural alterations similar to those seen in patients with schizophrenia¹⁶⁴. Mice with a knock-in mutation that results in the blockade of inhibitory phosphorylation of GSK3α and GSK3β display impaired hippocampal CA1 LTD (but intact LTP) and behavioural abnormalities, such as increased locomotor activity in response to amphetamine and stress-induced depression-like behaviours which may be relevant to mood disorders¹⁶⁵. However, whether these behavioural alterations are caused by a disruption in LTD remains to be tested experimentally.

Drug addiction and cortico-limbic-striatal circuits. Strong evidence supports a role for synaptic plasticity in drug addiction^{166,167,168,169,170} and in some circumstances, a specific role for LTD (Box 4). In particular, studies using the GluA2_3Y peptide have illuminated roles of nucleus accumbens (NAc) LTD(A) in the expression of behavioural sensitization to amphetamine¹⁵⁵ and of medial prefrontal cortex LTD(A) in cue-induced relapse to heroin-seeking behaviour (Box 4)¹⁵⁶. However, recent studies also suggest an important role of GluA2-lacking AMPARs in various models of addiction. For example, the insertion of GluA2-lacking AMPARs in the NAc underlies the enhanced response to cocaine-associated cues following extended drug withdrawal¹⁷¹ and a single injection of cocaine increases the proportion of GluA2-lacking AMPARs in the ventral tegmental area. This effect was reversed by mGluR1-dependent LTD (Fig. 4)^84,89,172. Whether the increased insertion of GluA2-lacking AMPARs that was observed in these studies is a consequence of endocytic removal of GluA2-containing AMPARs (due to enhanced GluA2-dependent LTD) remains to be determined. Thus, a complex interplay among various forms of LTD probably mediates different aspects of drug addiction.

Conclusions

Considerable progress has been made in the understanding of the molecular cascades responsible for LTD in the CNS. Most of the data focuses on the role of LTD in the hippocampus, in particular the forms that are triggered by activation of NMDARs or mGluRs and are expressed by AMPAR-mediated synaptic transmission. However, LTD can also be triggered by the activation of other types of glutamate receptors and other neurotransmitters, and can be expressed by alterations in other forms of synaptic transmission. Therefore, LTD — like its counterpart, LTP — probably has many roles in health and disease. Future research is likely to uncover the mechanisms by which LTD is induced by neurotransmitters other than L-glutamate (for example, by acetylcholine) and to explore the relationships among the various forms of LTD, as well as the relationships between forms of LTD and other forms of synaptic plasticity — LTP in particular.

LTD occurs in numerous circuits in the mammalian brain. Although the function of hippocampal LTD has received the most attention, compelling evidence exists for a role of LTD in other areas (Fig. 4). Novel pharmacological and genetic approaches that specifically affect LTD have provided direct evidence for its involvement in several physiological processes that include learning and memory and the development of the visual system. In addition, the involvement of LTD in pathological states has been demonstrated in models of drug addiction, acute stress, developmental conditions and neurodegenerative disorders, such as Alzheimer's disease (Box 2). These findings underscore the need for further research into the mechanisms and functions of LTD that may prove useful in the development of novel therapeutic agents and treatments for a diverse range of disorders.

Box 1 | The taxonomy of long-term depression

The term long-term depression (LTD) encompasses several forms of synaptic plasticity. LTD is often input-specific, in which case it is called homosynaptic (see the figure, part a, left panel) because LTD occurs at the synapses of the conditioned input (S1 in the figure). Under some circumstances it can be observed at a non-conditioned input, in association with LTD (see the figure, part a, central panel) or long-term potentiation (LTP) (see the figure, part a, right panel) that occur in synapses that receive the conditioning input and in this case it is called heterosynaptic LTD. Examples of homo- and heterosynaptic LTD can be seen in the figure (part a) — the green dots indicate the response in the synapses that are activated by low-frequency or high-frequency stimuli (LFS and HFS, respectively) and the yellow dots show the response of a non-condition input. An LTD experiment is typically run for 30–60 minutes.

Homosynaptic LTD (and potentially heterosynaptic LTD) can be further divided into two main types (see the figure, part b). LTD is often observed only after the induction of LTP, in which case it is referred to as depotentiation (see the figure, part b, left panel). In some cases, however, LTD can be observed from baseline conditions and this has been termed de novo LTD (see the figure, part b, right panel). Often the type of LTD that is being observed is unknown (for example, LTD after LTP in an input that can exhibit de novo LTD may be a mixture of depotentiation and de novo LTD).

All of these forms of LTD can be recorded using either extracellular recording (see the figure, part c, left panel) or whole-cell (patch-clamp) recording (see the figure, part c, right panel). Input-specificity studies of LTD typically involve a set-up that includes two input electrodes (labelled S1 and S2) and a recording electrode (R) — in the case of extracellular recording, this electrode is extracellular — and LTD is normally induced using LFS (typically 900 stimuli at 1–3 Hz). Whole-cell recording allows the membrane potential to be altered (see the figure, part d) so that LTD can also be induced by pairing baseline stimulation with depolarisation (typically to −40 mV; see the figure, part d, left panel) or by appropriately timed back-propagating action potentials (in current clamp mode). In this case the resulting LTD is a form of spike-timing dependent plasticity (STDP) (see the figure, part d, right panel). Finally, a form of LTD termed chemical LTD can be induced by applying a receptor agonist, such as 3,5-dihydroxyphenylglycine (DHPG) to the perfusing medium (see the figure, part e).

LTD can be further categorised based on which receptors trigger the process. NMDARs and metabotropic glutamate receptors (mGluRs) are the most commonly occurring among these receptors (NMDAR-LTD and mGluR-LTD, respectively). Finally, LTD can also be categorised based on the type of synaptic transmission that is being studied. This is typically AMPAR-mediated synaptic transmission (for example, NMDAR-LTD(A)), but other forms of synaptic transmission can also support LTD, for example, NMDAR-mediated synaptic transmission (NMDAR-LTD(N)).

Box 2 | Long-term depression and neurodegeneration

There is growing evidence that long-term depression (LTD) and the apoptotic pathway have common molecular mechanisms including activation of glycogen synthase kinase-3β (GSK3β)⁷⁷ and caspase-3 (Ref. 78). We propose that NMDAR-dependent LTD of AMPAR-LTD (known as NMDAR-LTD(A)) contributes to the elimination of superfluous synapses during development and that this involves a classic apoptotic pathway. However, once synapses are stabilized this process has to be downregulated to preserve synapse integrity. One regulator of this process may be GSK3β, as its activity is required for NMDAR-LTD(A) and it is inhibited at synapses that have recently experienced LTP(A)⁷⁷ (Fig. 3). In this model LTD requires the activation of GSK3β and this then leads to the removal of some AMPARs from the synaptic membrane. Additional LTD may eventually result in 'silent' synapses as part of CNS development (see the figure, top part and bottom line) but not in the context of cognitive plasticity in adulthood (see the figure, middle part). Synapses that are devoid of AMPARs may be eliminated as part of CNS development (see the figure, top part). Clearly, the activity of GSK3β has to be tightly regulated (by phosphorylation (P in figure) and dephosphorylation) to prevent complete elimination of AMPARs from the membrane and spurious synaptic pruning. However, during various neurodegenerative conditions such as Alzheimer's disease and schizophrenia GSK3β activity may become deregulated due to increased expression of GSK3β or as a result of alterations in the upstream regulators of GSK3β¹⁷³. This could result in a 're-awakening' of NMDAR-LTD(A), leading to synaptic pruning and eventually to neurodegeneration (see the figure, bottom part). Consistent with this hypothesis, neurotoxic species of amyloid β_1-42 (Aβ_1-42) have been shown to facilitate NMDAR-LTD via a GSK3β-dependent pathway¹⁷⁴.

Box 3 | The development and use of GluA2-derived inhibitors of LTD

Early research regarding the functional roles of synaptic plasticity was focused on studying NMDARs. However, because activation of these receptors results in both long-term potentiation (LTP) and long-term depression (LTD), this strategy was not useful for specifying the distinct functional roles of these forms of plasticity (see the figure, part a). In addition, NMDARs are involved in synaptic transmission, particularly during high-frequency discharges¹⁷⁵. Therefore, the behavioural consequence of their inhibition could be due to an impairment of LTP, impairment of LTD or inhibition of synaptic transmission. More recently, the realization that distinct molecular signalling cascades regulate AMPAR exocytosis and endocytosis has led to strategies to disrupt these intricate molecular events (see the figure, part a)^{12,131,148,155,176}.

Considerable evidence supports the role of AMPAR endocytosis that is dependent on the AMPAR subunit GluA2 in the expression of various forms of LTD in a number of brain regions¹¹. Several putative endocytic motifs have been identified in the carboxyl terminal (C-terminal) region of the GluA2 subunit, including the adaptor protein 2 (AP2)-binding domain¹³⁰ (which overlaps with the N-ethylmaleimide-sensitive factor (NSF)-binding region), the tyrosine-containing cluster and the PDZ domain that binds protein interacting with C kinase 1 (PICK1) and AMPAR-binding protein/glutamate receptor interacting protein (ABP–GRIP). These findings have lead to the development of several interference peptides (see the figure, part b) that can disrupt the interactions between the GluA2 subunit and endocytic machineries¹⁷⁷. These peptides include the GluA2_3Y peptide (a short section of the GluA2 carboxy-terminus (C-terminus) (YKEGYNVYG) that is required for regulated GluA2-containing AMPAR endocytosis^{12,79,156,178,179} (see the figure, part a) as well as the G2CT peptide (KRMKLNINPS). The G2CT peptide is derived from the NSF/AP2 binding region¹³⁰ in the GluA2 C terminus and disrupts AP2 but not NSF binding and therefore prevents AMPAR endocytosis and LTD without affecting basal synaptic transmission^130,131,148 (see the figure, part a). A third class of peptide (Pep2-EVKI and Pep2-SVKI) disrupts the interaction of PICK1 and ABP–GRIP with the C-terminus of GluA2 (Refs 143, 180, 181). However, these last peptides may have the drawback of blocking both LTD and LTP⁷¹ (see the figure, part a).

As interference peptides specifically prevent AMPAR endocytosis they have become useful tools for examining the role of LTD in specific brain regions during various cognitive tasks. One potential shortcoming of interference peptide inhibitors is the presence of unknown, off-target effects. As clathrin-mediated AMPAR endocytosis seems to be involved in the expression of various forms of LTD, these peptides can potentially block multiple forms of LTD.

A number of strategies have been developed to deliver the peptides to brain areas of interest. These include expression of peptides with lentiviral vectors^131,148,176, conjugation of peptides to a modified tetrapeptide¹⁸⁰ or fusion of peptides with the cell membrane transduction domain of the human immunodeficiency virus 1 (HIV1) Tat protein^{12,155,156,182}. Studies that use these peptides have shown important roles for AMPAR endocytosis — and hence also for LTD — in memory, drug addiction, fear extinction and the development of the visual system (Fig. 4).

Box 4 | Long-term depression and drug addiction

Addiction to psychoactive drugs is characterized by compulsive drug use and relapse to drug-taking behaviour after a period of abstinence¹⁶⁶. Long-lasting drug-induced changes in neuroplasticity have been studied using a number of paradigms in rodents. These paradigms include behavioural sensitization of motor activity by repeated, intermittent administration of a drug¹⁸³ and inducing relapse to drug-seeking behaviour by presenting cues that were previously paired with drug administration¹⁸⁴. Both paradigms result in complex changes in regions of the mesocorticolimbic circuitry such as the ventral tegmental area (VTA), the nucleus accumbens (NAc) and the medial prefrontal cortex (mPFC). For example, acute NAc slices from mice that are sensitized to cocaine exhibit a decreased postsynaptic AMPA:NMDA current ratio and this suggests a role for LTD in behavioural sensitization¹⁸⁵. Furthermore, LTD that is induced in NAc slices involves AMPAR subunit GluA2-dependent endocytosis and is prevented by the Tat-GluA2_3Y peptide¹⁵⁵ (Box 3). When administered systemically before an amphetamine challenge, this peptide blocks the expression of behavioural sensitization in rats, an effect that also occurs following infusion of Tat-GluA2_3Y locally into the NAc but not the VTA. These results provide evidence that GluA2-mediated AMPAR endocytosis in the NAc is required for the expression of behavioural sensitization to amphetamine¹⁵⁵ (Fig. 4).

Recent studies suggest that stimulant drugs produce complicated patterns of changes in NAc synaptic plasticity that are drug- and treatment protocol-specific. Sensitization to amphetamine or cocaine increases dendritic complexity in the NAc¹⁸⁶. By contrast, the cell surface expression of AMPARs in the NAc is increased in rats that are sensitized to cocaine and that are undergoing withdrawal, suggesting that the NAc is in a potentiated state¹⁸⁷. However, no changes in AMPAR expression have been found following amphetamine sensitization¹⁸⁸. Whole-cell recordings from NAc neurons of mice that are sensitized to cocaine and that are undergoing withdrawal confirm the development of robust potentiation in AMPAR-mediated currents¹⁸⁹. However, more recent evidence suggests that administration of a challenge dose of a drug to sensitized rats undergoing withdrawal produces a robust depression in NAc AMPAR currents¹⁸⁹ and AMPAR cell surface number¹⁹⁰. Thus, there may be a gradual increase in the membrane surface expression of AMPARs in these conditions. This would ensure an enhanced LTD in response to a challenging dose of the drug and explain the efficacy of the Tat-GluA2_3Y peptide in blocking the expression of behavioural sensitization¹⁵⁵.

LTD in mPFC has also been linked to cue-induced relapse to drug-seeking behaviour¹⁵⁶. In one study, rats were trained to nose-poke for heroin infusions in the presence of audio–visual cues¹⁵⁶. Following abstinence or extinction training, re-exposure to the cues induced heroin-seeking behaviour, reduced levels of several proteins including the AMPAR subunits GluA2 and GluA3, and increased levels of adaptor protein 2 (AP2) subunit mu1 (Ap2m1) in synaptic membranes from mPFC. Moreover, the AMPA:NMDA current ratios and the rectification index measured in slices that were taken from the mPFC were reduced. After extinction training, treatment of rats with the Tat-GluA2_3Y peptide before cue re-exposure reduced the alteration of AMPAR expression and the AMPA:NMDA ratio, and also reduced cue-induced reinstatement of heroin-seeking behaviour. These results implicate mechanisms that are consistent with AMPAR endocytosis and LTD in cue-induced relapse¹⁵⁶ (Fig. 4). In another study, reinstatement of cocaine-seeking behaviour to a priming injection of cocaine was abolished by intra-NAc administration of a peptide (Pep2-EVKI; Box 3) that blocks the interaction of GluA2-subunit containing AMPARs with PICK1, thereby blocking AMPAR endocytosis. However, LTD was not directly measured in this study¹⁸⁰. Thus, mechanisms consistent with AMPAR-dependent LTD are involved in various aspects of addiction, raising the possibility that blocking LTD may prove a valuable therapeutic strategy for treating addiction.