Mapping Molecular Memory: Navigating the Cellular Pathways of Learning

Abstract

A consolidated map of the signalling pathways that function in the formation of short- and long-term cellular memory could be considered the ultimate means of defining the molecular basis of learning. Research has established that experience-dependent activation of these complex cellular cascades leads to many changes in the composition and functioning of a neuron’s proteome, resulting in the modulation of its synaptic strength and structure. However, although generally accepted that synaptic plasticity is the mechanism whereby memories are stored in the brain, there is much controversy over whether the site of this neuronal memory expression is predominantly pre- or postsynaptic. Much of the early research into the neuromolecular mechanisms of memory performed using the model organism, the marine snail Aplysia, has focused on the associated presynaptic events. Recently however, postsynaptic mechanisms have been shown to contribute definitively to long term memory processes, and are in fact critical for persistent learning-induced synaptic changes. In this review, in which we aimed to integrate many of the early and recent advances concerning coordinated neuronal signaling in both the pre- and postsynaptic neurons, we have provided a detailed account of the diverse cellular events that lead to modifications in synaptic strength. Thus, a comprehensive synaptic model is presented that could explain a few of the shortcomings that arise when the presynaptic and postsynaptic changes are considered separately. Although it is clear that there is still much to be learnt and that the exact nature of many of the signalling cascades and their components are yet to be fully understood, this still incomplete but integrated illustrative map of the cellular pathways involved provides an overview which expands understanding of the neuromolecular mechanisms of learning and memory.

Introduction

Humans, in common with many other organisms, have the remarkable ability to store accumulated information gained through experience in the form of memory. This capacity to learn is an important feature that allows one to modify behaviour in response to subsequent detrimental external stimuli, or to react positively to those stimuli that are beneficial. Memory thus describes any behavioural changes brought about by experience, or the stored knowledge upon which changes in behaviour are dependent. When pursuing a reductionist approach to memory, it is evident that elementary forms of learning may be described at a cellular level in terms of the molecular activities and the cascade of molecular events that occur within specific neurons during the learning process. Cellular and molecular studies (as predominantly performed using a model organism, the marine snail Aplysia californica, in both in vitro and in vivo experiments) have revealed that learning and memory are induced by the modulation of the strength of synaptic connections between these neurons in the brain, a process known as synaptic plasticity (Carew and Sahley 1986; Castellucci et al. 1970; Kandel and Schwartz 1982; Mozzachiodi and Byrne 2009). Synaptic plasticity refers to any lasting upregulation or downregulation of synaptic strength. An extensive range of mechanisms are implicated in changes of synaptic strength, acting at the level of both the presynaptic and postsynaptic neuron. Direct modulation of the process of release of excitatory/inhibitory neurotransmitter and the subsequent changes in the amount of neurotransmitter released, as well as modifications in the function and/or number of neurotransmitter receptors are some examples of the pre- and postsynaptic alterations by which the strength of synaptic connections may be altered (Mozzachiodi and Byrne 2009; Kandel 2001). Subsequent reactivation of the neurons comprising the altered connections constitutes the experience of memory for the external events with which the connection modification was initially associated (Martin and Morris 2002).

The discovery of the neuronal phenomena of short-term and long-term facilitation (STF and LTF, respectively), whereby short- (<30 min) and long-lasting (>24 h) increases in synaptic efficacy are induced by brief yet low (STF) or high (LTF) frequency stimulation (whether electrical or chemical) of a neural pathway, has provided an "experimental analogue of these learning-induced changes in synaptic connectivity" (Bliss and Lømo 1973; and reviewed in Kandel 2001; Martin and Morris 2002). More recent studies have identified a third, intermediate form of synaptic facilitation that is mechanistically and temporally distinct from both STF and LTF. The discovery of this intermediate-term facilitation (ITF), lasting up to 3 h, further emphasized the complex molecular processes that are initiated by different experiences during changes in synaptic plasticity. STF, LTF, and ITF have thus facilitated the discovery of the many cellular signal transduction cascades and the various components involved in the alteration of synaptic strength during stimulation, a process that forms the basis for learning and memory.

An account of the many molecular and cellular events occurring on neuronal stimulation and that lead to the modification of the strength of synaptic connectivity is presented. The primary aim in this description of the cellular mechanisms of learning and memory is to integrate the various findings concerning coordinated neuronal signalling in both the pre- and postsynaptic neurons, in both vertebrates and invertebrates. Figure 1 is a composite graphical representation which illustrates schematically the cellular model of each of the components concerned and forms a useful guide in the following discussion of the various interconnecting pathways important in cellular learning.

Fig. 1.

Molecular memory mapped. A composite cellular model of the signalling pathways involved in short-, intermediate- and long-term synaptic plasticity, based on evidence provided from studies of synaptic plasticity in both vertebrates and invertebrates. Both the presynaptic and postsynaptic neurons are presented, along with the modulatory interneuron (IN) that regulates the release of stimulatory serotonin (5-HT) during behavioural changes such as sensitisation during LTF. The cellular pathways mediated by PKA, PKC and MAPK are indicated as blue, green and red lines, respectively, while those lines that are dashed depict pathways whose involvement in short-, intermediate-, and/or long-term synaptic facilitation is uncertain at present. The orange lines are representative of the pathways proposed for the transsynaptic retrograde signal activated by 5-HT during LTF, one putative signal of which appears to be the binding of neuroligin to neurexin (shown explicitly in the lower part of the figure). While the synaptic interaction of these CAMs may in fact be the only transsynaptic signal that exists, the orange lines have been included to account for the possible pathways that may be followed by one or more additional putative retrograde signals that do not involve CAMs. Positive and negative symbols indicate the enhancement/activation and suppression of cellular processes, respectively. Refer to text for the details of the molecular pathways illustrated (adapted from Fig. 2 in Glanzman 2008)

It is important to note at the outset however, that there are many forms of learning and related behavioural changes. These may be induced upon stimulation by varying the types and frequencies of external stimuli. These behavioural changes include habituation, sensitisation, and classical conditioning, all of which entail the formation of different temporal and spatial cellular combinations of either short-term, intermediate-term, or long-term neuronal facilitation (STF, ITF and LTF, respectively), and/or short-term, intermediate-term, or long-term depression (STD, ITD, LTD) (reviewed in Glanzman 2008). To further compound the apparent complexity, different temporal (application frequency) and spatial (pre- or postsynaptic, cell body or distal cell process) combinations of application of chemical and/or electrical stimuli to neurons may induce activity-dependent changes in the pathway’s described state of facilitation or depression (Glanzman 2008). This variability has caused some controversy in the confidence of certain findings as a result of the relative variability in the experimental methodology employed by different research groups studying the cellular mechanisms of memory. Nevertheless, it is clear that there are common biological features within the signalling cascades involved in the short-term and long-term forms of all behavioural change, no matter what the nature of stimulation. The molecular mechanisms that will be described are those that are therefore accepted as being representative of the general events that lead to STF, ITF and LTF.

The experimental model system used for examining the variety of plastic changes that give rise to associative and non-associative forms of learning and to short-, intermediate-, and long-term memory storage associated with each of these forms of learning, have largely entailed the application of a varying number of 5-min-long pulses of 5-hydroxytryptamine (5-HT; a monoamine neurotransmitter also known as serotonin) to a co-culture system comprising a sensory and motor neuron derived from Aplysia (Kandel 2001; Byrne and Kandel 1996). This model has been used to reproduce the well-known STF, ITF, and LTF that is exhibited during sensitisation of the withdrawal reflex of its gill and siphon by noxious stimuli, such as electrical shocks applied to the animal’s skin since it had been established that 5-HT is released within the central nervous system of Aplysia after sensitising stimulation (Marinesco and Carew 2002). Both the application of 5-HT and the stimulation of serotonergic interneurons facilitate the growth of sensorimotor connections (Brunelli et al. 1976; Mackey et al. 1989), while depletion of 5-HT within the nervous system impairs sensitisation (Glanzman et al. 1989). 5-HT thus plays a central role in the STF, ITF and ITF that produces sensitisation in Aplysia, although other facilitatory transmitters are also known to participate in the process (Ocorr and Byrne 1985). Within the co-culture system described, a single brief application of 5-HT is required for the induction of STF, while five repeated yet spaced applications of 5-HT are sufficient to induce LTF (Mauelshagen et al. 1996; Kandel 2001). Although there are many slight variations in the way in which this is performed, the general protocol involves five 5 min pulses of 5-HT, with 15 min between each pulse (Mauelshagen et al. 1996; Montarolo et al. 1986). The induction of ITF is less straight forward (since it has some of the features of LTF, which has been researched more extensively, ITF will be discussed last; see section ‘Intermediate-Term Synaptic Plasticity’). When studying synaptic plasticity, the growth of new synaptic connections is usually determined by the presence of varicosity formation, while an increase in synaptic strength is measured by monitoring changes in the excitatory postsynaptic potential (EPSP).

Facilitation in Aplysia requires a heterosynaptic form of plasticity where the typical neuronal circuit important for behaviour and learning comprises serotonergic and other interneurons that act on sensory and motor neurons to regulate the strength of their connections (Hawkins et al. 1981). For example, the interneurons may act on the presynaptic terminals of the sensory neurons to enhance the amount of the glutamate (a facilitatory neurotransmitter) that is released, and yet it is increasingly evident that the interneurons also induce changes in the postsynaptic neuron (Glanzman 2008).

Mechanisms of Learning-Related Synaptic Plasticity

Short-Term Synaptic Plasticity

Short-term facilitation (lasting up to 30 min) depends on the covalent modification of pre-existing proteins and the strengthening of pre-existing connections, and seems to be mediated predominantly by changes in the presynaptic sensory neuron (Kandel 2001). There is however some suggestion that some postsynaptic changes may also contribute to STF (Glanzman 2008). Within this process, stimulation (for example by a single tail shock in Aplysia) causes serotonergic interneurons to release a single ‘pulse’ of 5-HT which binds to and activates presynaptic G-protein coupled receptors (GPCRs) that are associated with either adenylate cyclase (AC) or phospholipase C (PLC) (Kandel 2001). The activation of adenylate cyclase by 5-HT and Ca²⁺/calmodulin leads to the synthesis of the second messenger cyclic AMP (cAMP) from ATP, which in turn activates protein kinase A (PKA) (Bernier et al. 1982; Castellucci et al. 1982; Byrne and Kandel 1996). The activation of phospholipase C results in the conversion of phosphatidylinositol bisphosphate (PIP₃) into inositol triphosphate (IP₃) and diacylglycerol (DAG), the latter of which activates protein kinase C (PKC) (Sacktor and Schwartz 1990; Sossin and Schwartz 1992; Sugita et al. 1992; Braha et al. 1993). The resulting IP₃ activates phosphoinositide 3-kinase (PI3K) which plays a significant role in LTF when the neuron is further stimulated by additional pulses of 5-HT.

Activated PKA and PKC have independent but overlapping spheres of action, serving several functions in STF (and LTF as will be discussed later) that involve the phosphorylation and covalent modification of a number of target substrate proteins (Braha et al. 1990, 1993; Sugita et al. 1992; Byrne and Kandel 1996). Both PKA and PKC act by enhancing presynaptic neurotransmitter release when a subsequent action potential is fired in the sensory neuron through two mechanisms. The first of these entails the phosphorylation of two major K⁺ channels, including a voltage-dependent (I _KV) and voltage-independent (I _KS) form (Baxter and Byrne 1989; Byrne and Kandel 1996). Phosphorylation reduces their specific K⁺ ion currents, allowing a greater influx of Ca²⁺ ions into the presynaptic terminal through neuronal spike broadening (longer duration of an action potential) and a related increase in excitability (which describes the ease at which a neuron may trigger a subsequent action potential, and may be manifested through the lowering of the ‘threshold’ potential) (Klein and Kandel 1980; Shuster et al. 1985; Baxter and Byrne 1990; Byrne and Kandel 1996). The resultant increased influx of Ca²⁺ contributes to enhanced transmitter release. The described spike broadening is mediated predominantly by I _KV with a minor contribution from I _KS, while I _KS mediates the majority of the excitability change (Canavier et al. 1991; Belkin et al. 1992; Byrne and Kandel 1996). Furthermore, while both activated PKA and PKC phosphorylate and modulate I _KV, I _KS is only modulated by PKA (Klein et al. 1980; Byrne and Kandel 1996). The second mechanism through which PKA and PKC enhance presynaptic neurotransmitter release involves a spike-broadening independent process acting directly on the mobilisation and exocytosis of presynaptic vesicles containing neurotransmitters (Hochner et al. 1986a, b; Klein et al. 1986; Braha et al. 1990). PKA has been shown to act on some aspect of the release or on the exocytotic mechanism itself, while PKC’s role entails the mobilisation of the vesicles from reserve or storage pools (Braha et al. 1990; Byrne and Kandel 1996).

The relative contributions of the PKA and PKC pathways to short- or long-term presynaptic facilitation depends on both the initial state of the synapse (in terms of being either depressed or non-depressed) and the duration of the exposure to stimulating factors such as 5-HT which facilitates either STF (brief exposure) or LTF (long exposure) (Hochner et al. 1986a, b; Braha et al. 1990; Ghirardi et al. 1992; Byrne and Kandel 1996). In non-depressed synapses and/or synapses which have been exposed briefly to 5-HT to facilitate STF, the cAMP and PKA signalling pathway plays the predominant role through a modulation of vesicle release and its corresponding I _KV and I _KS membrane currents (Klein et al. 1986; Baxter and Byrne 1990; Ghirardi et al. 1992; Sugita et al. 1992; Byrne and Kandel 1996). On the other hand, in depressed synapses which have more prolonged exposure to 5-HT, thereby facilitating LTF, the DAG and PLC pathway plays the dominant role via vesicle mobilisation and the concomitant effect on the I _KV membrane currents which will be discussed (Braha et al. 1990; Ghirardi et al. 1992; Byrne and Kandel 1996).

Long-Term Synaptic Plasticity

Commonalities of Vertebrate and Invertebrate Long-Term Synaptic Plasticity

In this review, in order to present a consolidated map of both the pre- and postsynaptic molecular mechanisms involved in learning and memory, evidence has been derived from studies of both vertebrate and invertebrate long-term synaptic plasticity. This is a consequence of the models that have been presented for invertebrate learning-related synaptic plasticity assuming that plasticity is mediated predominantly by presynaptic changes, while the models that exist for vertebrate plasticity have assumed that it is mediated principally by postsynaptic modifications. However, as stressed in Glanzman (2010) review, there are strong grounds for the idea that a common set of pre- and postsynaptic molecular mechanisms for plasticity are shared between vertebrates and invertebrates, and thus that both the pre- and postsynaptic neurons have critical roles in the expression of long-term memory (see Glanzman 2010). This allows a composite model of the cellular pathways and molecular mechanisms that operate at sensorimotor synapses during changes in synaptic strength to be presented, and a model that may be applicable to the nervous systems of both vertebrates and invertebrates.

Whether the molecular mechanisms of learning and memory in vertebrate synapses, such as those underlying long-term potentiation (LTP) and long-term depression (LTD), are significantly equivalent to those that operate in invertebrate synapses, such as those fundamental to long-term facilitation (LTF) in Aplysia, has been the subject of extensive debate. This has arisen as a result of the described contrasting focus in the proposed synaptic locus (either pre- or postsynaptic) of memory expression in vertebrates and invertebrates. In addition, LTF of the sensorimotor synapse depends on heterosynaptic modulatory input, requiring neurotransmitters such as serotonin (5-HT) originating from modulatory interneurons for full sensitisation-related facilitation. LTP on the other hand is a homosynaptic form of synaptic plasticity that is intrinsic to the sensorimotor synapse. Nonetheless, there is much data to suggest that there is remarkable commonality between the synaptic mechanisms of learning and memory in vertebrates and invertebrates. To illustrate the degree to which the vertebrate and invertebrate synaptic mechanisms of learning and memory (and hence LTP and LTF) have been conserved through evolution, the similarity of their molecular pathways that have been found to be critical for LTP and LTF will be presented. This will provide justification for a general model of the molecular mechanisms governing synaptic plasticity to be presented based on data from both vertebrates and invertebrates.

LTP and LTD are the two neuronal mechanisms that have been proposed to underlie learning and memory in mammals and other vertebrates (Malenka and Bear 2004). LTP was first discovered in the dendate gyrus of the hippocampus, and later identified to occur at many other excitatory glutamatergic synapses in the mammalian brain, such as those that exist in the amygdala, cerebellum and cerebral cortex (Malenka and Bear 2004). Although the induction of LTP has been attributed to several distinct mechanisms and the exact cause of the increase in amplitude of synaptic potentials at potentiated synapses is still to be fully understood, LTP has been described to be induced predominantly by activation of postsynaptic N-methyl-d-aspartate (NMDA)-type glutamate receptors (described below, but see Malenka and Bear 2004). Upon induction, LTP expression is ascribed to both enhanced presynaptic neurotransmitter release, as well as insertion of additional α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA)-type glutamate receptors into the postsynaptic membrane at sites that previously lacked functional AMPA receptors (see section ‘Postsynaptic Mechanisms Contributing to Long-Term Synaptic Plasticity’) (Malenka and Bear 2004). This has become the generally accepted model for synaptic plasticity in the mammalian brain, in which strengthening of excitatory, glutamatergic synapses is accomplished in part by modulating the trafficking of postsynaptic AMPA receptors (Kessels and Malinow 2009).

Despite the resistance to the idea that invertebrate synaptic plasticity exhibits LTP mechanistically equivalent to that seen within mammalian neuronal synaptic plasticity, Lin and Glanzman (1994) presented evidence that the sensorimotor synapse of Aplysia possesses the capacity for Hebbian, NMDA receptor-dependent LTP. Since then, LTP and NMDA receptor dependent plasticity has been reported in the leech (Burrell and Sahley 2004), while NMDA receptor activity has been demonstrated in associative learning in Aplysia (Murphy and Glanzman 1997; Antonov et al. 2003), Drosophila (Xia et al. 2005) and Caenorhabditis elegans (Kano et al. 2008). Hebbian LTP (involving paired pre- and postsynaptic stimulation) has also been reported in the octopus (Hochner et al. 2003; Shomrat et al. 2008) and the honey bee (Menzel and Manz 2005), although these were found to be NMDA-receptor independent. This highlights that LTP and NMDA receptor mediated plasticity is not exclusive to the vertebrate nervous system, and is in fact widespread among many invertebrate species. Correspondingly, synaptic facilitation in Aplysia (Li et al. 2005, 2009) and long-term habituation in C. elegans (Rose et al. 2003) had been initially ascribed entirely to presynaptic processes, but like LTP, it has been shown that modulation of postsynaptic AMPA receptor trafficking also plays a critical role.

Although the postsynaptic AMPA receptor trafficking model of synaptic plasticity in mammals is generally accepted, presynaptic mechanisms have also been shown to contribute to the expression of LTP (Krueger and Fitzsimonds 2006). These presynaptic contributions to LTP remain a subject of controversy however because as induction of this form of synaptic plasticity is postsynaptic, a yet to be identified retrograde signal would be required to mediate the presynaptic expression changes. There are nonetheless many signaling molecules that have been proposed to function as the retrograde signal in LTP (Regehr et al. 2009). In one example, Yeckel et al. (1999) discovered that although NMDA receptor-independent LTP at mossy fibre synapses in the hippocampus is expressed presynaptically, its induction appears to require elevated postsynaptic Ca²⁺. This was disputed by Mellor and Nicoll (2001) who showed that the induction of LTP at these synapses does not require postsynaptic Ca²⁺. However, Contractor et al. (2002) subsequently proceeded to illustrate that a transsynaptic pathway involving EphB-receptor–ephrinB ligand interactions is activated by a rise in intracellular Ca²⁺ within the postsynaptic neuron. This retrograde signal activates PKA, which stimulates the enhancement of neurotransmitter release from the mossy fibres (Contractor et al. 2002). As will be discussed, retrograde signaling is now known to be an important characteristic of plasticity and LTF in invertebrates, particularly Aplysia. The synaptic model of postsynaptic induction and presynaptic expression is thus clearly evident in the Aplysia sensorimotor synapse.

Despite the current controversies and the gaps within the overlapping spheres of pre- and postsynaptic processes within vertebrates and invertebrates during LTP and LTF, based on the evidence described, a consolidated model of persistent synaptic enhancement may become appropriate for describing the molecular mechanisms of learning and memory in both vertebrates and invertebrates. In general, long-term synaptic plasticity can be distinguished from short-term plasticity by its requirement for gene transcription and protein synthesis, and the consequential longer duration (of one or more days) of the increased synaptic efficiency produced. Unlike short-term synaptic plasticity, long-term plasticity also requires the growth of new synaptic connections between the sensory/presynaptic and motor/postsynaptic neurons, implying that the protein synthesis and gene transcription accompanying long-term plasticity is required for this new growth (Frost et al. 1985; Montarolo et al. 1986; Bailey and Chen 1988a, b). LTP and LTF may also involve the insertion of new active zones, through additional protein synthesis, into pre-existing but empty presynaptic terminals without the need for new growth (Bailey and Chen 1991; Kim et al. 2003). These less obvious morphological changes, whereby previously 'silent' presynaptic terminals lacking vesicles are activated, may involve the filling of the empty terminals with synaptic vesicles so that they become functional (Kim et al. 2003). LTP and LTF thus strongly resemble their short-term forms with respect to their stimulation by either tetanus (LTP) or 5-HT (LTF), and in their induction of enhanced transmitter release and enhanced presynaptic excitability, but may also be considered to be extensions of certain aspects of STP and STF in a manner that depends on RNA and protein synthesis.

In addition to the roles that PKA and PKC have in enhancing presynaptic neurotransmitter release specifically during LTP/LTF, these kinases act in other pre- and postsynaptic signalling pathways of facilitation that lead to gene transcription, protein synthesis and the subsequent construction of new sensorimotor synapses.

Presynaptic Mechanisms Contributing to Long-Term Synaptic Plasticity

PKA is known to possess both catalytic and regulatory subunits, the latter of which normally inhibits the catalytic subunit (Gill and Garren 1971). When the level of cAMP rises as a result of the activation of the receptor linked adenylate cyclase by 5-HT stimulation, the binding of cAMP to the regulatory subunit of the PKA induces a conformational change that allows the regulatory subunit to dissociate from the catalytic subunit (Gill and Garren 1971). The free catalytic subunit is then able to catalyse the phosphorylation of its specific substrate proteins in the presynaptic terminal. Repeated stimulation by 5-HT, such as that occurring in the initial stages of LTF, causes the levels of cAMP to rise extensively and to persist for a period long enough to free the catalytic subunit of PKA and to allow it to translocate to the nucleus (Brunelli et al. 1976; Bacskai et al. 1993; Byrne and Kandel 1996). In its nuclear translocation, PKA also recruits and activates mitogen-activated protein kinase (MAPK) (Martin et al. 1997b; Impey et al. 1998). Within the nucleus, MAPK interacts with and phosphorylates the cyclic AMP response element-binding protein-2 (CREB-2), which is normally responsible for the regulated inhibition of CREB-1 (Bartsch et al. 1995; Michael et al. 1998; Impey et al. 1998). As the repressive action of CREB-2 is being removed by MAPK, PKA is simultaneously able to activate CREB-1 (Kaang et al. 1993; Bartsch et al. 1998). CREB-1 is a transcription factor whose phosphorylative activation in turn stimulates (along with other co-activators) transcription and mRNA production from a variety of genes containing the cyclic AMP response element (CRE) promoter to which CREB-1 binds (Dash et al. 1990; Bartsch et al. 1998). The immediate-response genes which are activated include one which encodes ubiquitin hydrolase. This is a component of a specific ubiquitin protease that facilitates the controlled proteolytic cleavage of the regulatory subunit of PKA to relieve its inhibitory effect on the catalytic subunit (Hegde et al. 1997; Chain et al. 1999). This cleavage therefore results in the PKA acquiring persistent activity, which leads to the persistent phosphorylation of its substrate proteins (Kandel and Schwartz 1982). These include CREB-1 in a now long-lasting positive feed-back system, as well as the proteins involved in the STF process such as the specific K⁺ ion channels and those involved in synaptic vesicle release. The CAAT box enhancer binding protein (C/EBP) is another product of the immediate response genes activated by CREB-1 which is an additional transcription factor that in turn stimulates the transcription of downstream genes that trigger long-term structural changes in the sensory neurons (Alberini et al. 1994). C/EBP may either act by itself as a homodimer to recognise and activate transcription of late CAAT box promoter genes, or with activating factor (AF) as a heterodimer to activate TAAC genes (Alberini et al. 1994; Bartsch et al. 2000). These late genes express proteins such as elongation factor 1α (EF1α) that give rise to the formation of new synaptic connections between the sensory and motor neurons (Glanzman 2008).

Synapse-Specificity of Long-Term Synaptic Plasticity

Long-term synaptic changes occur only at those synapses that have been stimulated by previous activity which suggest that the changes are highly synapse-specific. In view of the "extreme functional and morphological polarity of each neuron (with each of the potentially thousands of presynaptic terminal boutons and postsynaptic spines made by a single neuron capable of operating as an autonomous compartment)" (Martin et al. 2000), synapse-specificity facilitates the formation of the highly specific neuronal interconnections important in memory generation in the hippocampus of the brain (Kandel 2001).

As long lasting forms of memory and synaptic plasticity require the synthesis of new mRNA and protein, the synaptic specificity is interesting when one considers the requirement for transcription within the central nucleus. The means by which these transcription-dependent forms of plasticity occur in a synapse-specific manner may be described by three models. According to the first model, the mRNA or protein products of signal activated gene expression are delivered unspecifically throughout the cell but only increase synaptic strength at a synapse that has been ‘tagged’ by the signal (Martin et al. 2000). According to a second model, the mRNA or protein produced upon synaptic signalling is targeted specifically to the stimulated synapse by the synaptic signalling. The third model has similarities to the second as the mRNA is specifically localised to the stimulated synapse where it undergoes local translation (reviewed in Martin et al. 2000).

There is much evidence that demonstrates that the mRNA transcription products of many of the early and late genes described above undergo translocation and localisation to distinct neuronal compartments before translation (Bassell et al. 1999). mRNA localisation and their local regulated translation decentralises part of the control of gene expression from the nucleus to discrete subcellular compartments (Bassell et al. 1999). This allows the rapid modification of the macromolecular composition of distinct locations in neurons such as specific synapses, thereby facilitating new synaptic growth and the formation of new sensorimotor contacts at these synapses (Martin et al. 2000). In addition to the stabilisation of new synaptic growth, the locally produced proteins are also important in retrograde signalling to the nucleus to initiate further transcription and additional local protein production, as well as acting as a local synaptic mark that stabilises subsequent functional and structural changes at the activated synapse (Martin et al. 1997a; Casadio et al. 1999; Sherff and Carew 1999). The occurrence of this ‘synaptic mark’ in Aplysia is evident from the finding that a synapse specific LTF induced in one branch can be captured at another by the application of a single pulse of 5-HT, which would normally only induce STF when employed by itself (Martin et al. 1997a). Thus, repeated pulses of 5-HT serve to mark the activated synapse so that newly synthesised mRNAs or proteins required for LTF, whether transported to all the synapses of the neuron or, more probably, to the specific synapse activated, are only actively employed by the marked synapse (Martin et al. 1997a).

When the molecular nature of the synaptic mark was investigated, it was discovered that it has a PKA-dependent component needed for the initial capture of the synapse-specific LTF (Casadio et al. 1999), which is what might be expected if one considers that most pathways leading towards LTF-related synaptic changes require the early activation of PKA. The tag should thus be made or activated at the synapse by a signal for STF that stimulates the PKA, since this is all that is required for synaptic capturing. Another component of the mark was a rapamycin-sensitive, local protein synthesis-dependent factor required for the long-term maintenance of LTF (Casadio et al. 1999), a process that will be elaborated on. The fact that mRNAs are produced within the nucleus means that there is a requirement for the local translation of some mRNAs at specific synapses, suggesting that they may remain 'dormant' before they reach these sites of translation (Si et al. 2003). If this model proves to be correct, then a regulator of translation capable of activating translationally dormant mRNAs may act as the synaptic mark.

The localisation of the mRNA may depend on one or more cis-acting sequences in the 3′ untranslated region (UTR), which interacts with trans-acting proteins such as Staufen and/or Vera-like homologs (that mediate similar processes in non-neuronal cells, as discovered in Drosophila and Xenopus oocytes, respectively) that transport complexes along the microtubule/microfilament network (Bassell et al. 1999; Martin et al. 2000). It has also been shown that transport of the mRNA occurs in RNA granules as ribonucleoprotein particles, while their induced translocation is a highly specific and regulated process mediated by the pre- and postsynaptic neuron (Knowles et al. 1996; Kiebler et al. 1999; Martin et al. 2000).

The localisation of mRNA is a means of spatially regulating gene expression that must be accompanied by locally regulated translation for new protein synthesis. Translation of mRNA encoding Ca²⁺/calmodulin-dependent protein kinase IIα (CaMKIIα; the most abundant postsynaptic dendritically localised mRNA) increases specifically in dendrites of the superior colliculus of young rats following binding of the synthetic neurotransmitter, N-methyl-d-aspartic acid (NMDA) to NMDA receptors (Scheetz et al. 2000). The activation of NMDA receptors on the postsynaptic neuron causes a decrease in overall protein synthesis by activating eukaryotic elongation factor 2 (eEF2) (Scheetz et al. 2000) resulting in a decrease in the overall translation of mRNA which might thus promote the translation of the specific populations of mRNA present in dendrites (Scheetz et al. 2000; Martin et al. 2000).

The translation of the specific RNA populations in mammals has also been shown to be regulated by cytoplasmic polyadenylation, a process generally mediated by two cis-acting regions in the 3′UTR of many transcripts. The first is termed the cytoplasmic polyadenylation element (CPE) of the general structure UUUUUAU that, upon stimulation, binds the CPE binding protein (CPEB) to recruit poly (A) polymerase (PAP) (Wu et al. 1998). PAP interacts with the second element of a polyadenylation sequence, AAUAAA, causing an increase in the length of the poly (A) tail and promotion of the translation of the translationally dormant mRNA (Wu et al. 1998). It has been shown in the hippocampus that CPEB is present in dendrites and that CaMKIIα transcripts contain a CPE in their 3′UTR, indicating that synaptic activation may regulate local translation of CaMKIIα through cytoplasmic polyadenylation (Wu et al. 1998). Although the role of CaMKIIα in LTF (which will be expanded on) is exhibited mostly in the postsynaptic neuron, there is evidence that the described mechanism of regulated mRNA localisation and translation may extend to similar mRNA transcripts involved in LTF, acting in both the pre- and postsynaptic neurons. Synaptic CPE-containing mRNAs discovered in Aplysia include those encoding regulatory proteins such as ephrin A2 (EphA2) that determine where and to what extent synapses should grow (Brittis et al. 2002), as well as structural proteins such as N-actin and Tα1-tubulin that stimulate and maintain long-term synaptic growth and plasticity (Kim and Lisman 1999; Moccia et al. 2003). The dynamic reorganisation of the cytoskeleton controls part of the structural aspect of synapse formation in LTF, achieved either through the local synthesis of cytoskeletal components, or through the redistribution of those that were pre-existing (Bonhoeffer and Yuste 2002). As N-actin and Tα1-tubulin mRNAs are dendritically localised, contain a CPE, and become polyadenylated in response to repeated pulses of 5-HT, it has been rationalized that CREB-mediated local protein synthesis must control at least some of the structural components required for new synaptic growth during LTF (Martin et al. 2000; Si et al. 2003).

A common 5′ translational control mechanism found in both vertebrates and invertebrates referred to as rapamycin-sensitive translation, involves the presence of terminal oligopyrimidine tracts in the 5′UTR of transcripts whose translation requires the regulation of eukaryotic initiation factor 4e (eIF4e) and ribosomal protein S6 kinase (rS6K) (Brown and Schreiber 1996). It has been shown that signal induced translation and subsequent synapse facilitation in Aplysia in isolated neuron processes is rapamycin-sensitive, which makes it dependant on this form of protein synthesis (Yanow et al. 1998; Casadio et al. 1999). It has also been illustrated that certain components of the rapamycin-sensitive pathway, including eIF4e and rS6K, which were described above, are present in dendrites of the hippocampus (Tang et al. 1998). These findings suggest that 5′-mediated mechanisms are also required to control synaptically regulated translation.

PI3K is activated by IP₃ which has been generated through the same GPCR that activates PKC (see Fig. 1), and it has been shown that one of the downstream effects of PI3K activity in Aplysia is the activation of rapamycin-sensitive protein synthesis from specific mRNA transcripts (Yanow et al. 1998; Dufner and Thomas 1999). These include CPEB mRNA, whose translation and consequent production of CPEB protein is necessary for the further production of protein that is required for LTF (Si et al. 2003).

The Role of Cell-Adhesion Molecules in Long-Term Synaptic Plasticity

Cell-adhesion molecules (CAMs) are glycoproteins present on the surface of the cell that mediate both cell-to-cell and cell-to-extracellular-matrix interactions. Their cytoplasmic domains are able to modulate their function by interacting with various intracellular signalling proteins that include kinases/phosphatases, second messengers, and adaptor molecules (Juliano 2002). Apart from the role they play in cell-migration, neurite outgrowth and de novo synapse formation in the central nervous system, neuronal CAMs have been found to be involved in the formation and functional expression of LTF, where their downstream signalling pathways are important for inducible synaptic plasticity, growth and development (Murase and Schuman 1999).

It is known that TM-apCAM (the transmembrane form of Aplysia cell-adhesion molecule) is specifically down-regulated through endocytosis at the sensory neuron membrane surface, rather than at the postsynaptic cell membrane, when the cell is stimulated by 5-HT (Mayford et al. 1992; Bailey et al. 1992). The two other isoforms of apCAM that exist, a large and small isoform of a glycosyl phosphatidylinositol-linked apCAM (GPI-apCAM), do not possess a cytoplasmic domain and are not internalised upon 5-HT stimulation (Bailey et al. 1997). apCAM has also been implicated in the neuritic fasciculation (bundling) of growth cones with presynaptic neurites through its adhesive extracellular domains, in addition to influencing synaptic strength enhancement and the formation of new synaptic connections in response to 5-HT (Bailey et al. 1992; Han et al. 2004). As will be discussed, this occurs via its intrinsic interaction with cytosolic proteins such as cytoskeletal elements. It also influences the downstream signalling pathway involving CREB-1 transcription (Lee et al. 2007). The clathrin-mediated internalisation and consequent downregulation of the TM-apCAM is thus an initial step in initiating the defasciculation (debundling) and the synaptic growth associated with LTF, as well as initiating a synaptic growth-independent increase in synaptic strength (Mayford et al. 1992; Bailey et al. 1992, 1997; Han et al. 2004).

The question arises as to how the internalisation of TM-apCAM contributes to the initiation of the transcription within the nucleus during LTF, particularly since it is known that it interacts with intracellular signalling molecules via its cytoplasmic tail, and it has been established that modification of this tail is required for internalisation and downregulation of TM-apCAM during LTF (Bailey et al. 1992). However, an apCAM-binding protein (CAMAP; CAM-associated protein) that has approximately 100 residues in contact with the cytoplasmic tail of apCAM has been found (Lee et al. 2007). Lee et al. (2007) "identified a cell-adhesion-associated molecule (CAMAP) that translocates to the nucleus of the sensory neuron following repeated application of 5-HT at the synapse and coactivates CREB-mediated gene transcription required for LTF". It had been discovered previously that downregulation and internalisation of TM-apCAM occurs through the ubiquitination or phosphorylation of its PEST sequence by activated MAPK (Mayford et al. 1992; Bailey et al. 1997). Once again, this process is known to be important in the formation of LTF, indicating that the CAM may regulate critical downstream signalling molecules through its cytoplasmic domain (Lee et al. 2007). This has led to the conclusion that the apCAM-CAMAP complex is important for synaptic signalling and synaptic growth during LTF.

CAMAP possesses both an N-terminal (N) and C-terminal (C) domain, the latter of which has an autoinhibitory effect upon the functional activity of the N-terminal domain that requires activation for transcriptional signalling (Lee et al. 2007). It has also been shown that the phosphorylation on Ser148 of CAMAP by PKA, (which, as described above, is stimulated by repeated application of 5-HT), causes its dissociation from TM-apCAM. This induces a conformational change that alleviates the autoinhibition of CAMAP’s functional N-terminal by its C-terminal domain (Lee et al. 2007). The dissociation allows the liberated TM-apCAM to be modified through phosphorylation (P) and/or ubiquitination (Ub) by MAPK and its subsequent internalisation through endocytosis (Lee et al. 2007). CAMAP may thus act as a presynaptic scaffolding protein that stabilises TM-apCAM, while 5-HT signalling induces the complex’s dissociation and the described internalisation/downregulation, ultimately allowing for new synaptic growth (Lee et al. 2007) via the various mechanisms which have just been described. The phosphorylated CAMAP then translocates into the nucleus possibly via importin (Imp) pathways which commonly mediate nuclear transport of macromolecules (Goldfarb et al. 2004). Once in the nucleus, it coactivates CREB-1-mediated transcription of C/EBP in association with PKA and CBP (Lee et al. 2007). C/EBP then triggers the transcription of downstream target genes, either as a homodimer or in association with activated AF as a heterodimer (Alberini et al. 1994; Bartsch et al. 2000), to induce LTF and related synaptic growth.

CAMAP nuclear transport is not only dependent on PKA signalling, which is the critical factor for retrograde signalling for transcription in the nucleus and for subsequent initiation of LTF, but it has been shown to be independent of local protein synthesis (Lee et al. 2007). This reinforces the idea that synapse-specific LTF has two time-dependent phases; initiation and maintenance. Initiation is dependent on PKA-signalling, where retrograde signals (such as activated CAMAP and MAPK) are generated to stimulate nuclear CREB-dependent gene expression (Martin et al. 1997b; Casadio et al. 1999; Kandel 2001). In contrast, the long-term maintenance phase of LTF depends on a rapamycin-sensitive component of local protein synthesis from the mRNA transcripts induced by the retrograde signals (Casadio et al. 1999).

TM-apCAM also has a CAMAP-independent function that acts both when complexed with CAMAP and when the two proteins have disassociated as a result of phosphorylation of each component. Although the nature of this function is not completely clear, TM-apCAM is thought to act, via distinct mechanisms, as a suppressor of both synaptic strength enhancement and of new synaptic varicosity formation in pre-existing synapses and in the non-synaptic region, respectively (Han et al. 2004). Since the presence of TM-apCAM does not affect the 5-HT-activated intracellular signalling cascade involving C/EBP-mediated transcription when suppressing LTF (the influence on CREB and C/EBP is presumably left to CAMAP as detailed above), the TM-apCAM down-regulation may act as a synaptic tag or facilitate the formation of a synaptic tag (Han et al. 2004). In this regard, TM-apCAM might serve to stabilise the synaptic structure and may need to be removed after the initiation phase and dissociate from CAMAP to allow synaptic vesicles and newly synthesised proteins, which may have been produced locally or via the nucleus, to be incorporated at pre-existing synapses, thereby facilitating EPSP enhancement during LTF (Lee et al. 2007). Furthermore, the extracellular domain of TM-apCAM inhibits defasciculation but does not inhibit new varicosity formation. This indicates that the internalisation of TM-apCAM may thus not be required to achieve defasciculation, but rather to remove the functional suppression on new-growth and synaptic enhancement that has been attributed to TM-apCAM’s cytoplasmic domain (Han et al. 2004). While the mechanism of this suppression is unknown, it has been suggested that the cytoplasmic domain may inhibit new varicosity formation by the stabilisation of cytoskeletal proteins (Han et al. 2004). Therefore, although the role of TM-apCAM down-regulation in expression of LTF has not been fully elucidated, the downregulation of TM-apCAM does seem to be a requirement for enhancement of synaptic strength in pre-existing synapses and new synaptic growth in non-synaptic regions, contributed in part by the suppressive function of its cytoplasmic domain which must be removed to allow full LTF formation (Han et al. 2004).

It has been found recently that the CAMs neurexin and neuroligin, that interact heterophillically with each other at the synapse, are situated predominantly in the pre- and postsynaptic neuron, respectively and are critically involved in the formation of long-term synaptic plasticity (Choi et al. 2011). It is generally accepted that the transsynaptic neurexin and neuroligin interaction affects the remodeling, differentiation and maturation of synapses during development, rather than de novo synaptogenesis (Missler et al. 2003; Varoqueaux et al. 2006; reviewed in Sudhof 2008). This is achieved through their cytoplasmic domains, whose PDZ-binding motifs stimulate signalling cascades suspected to induce the differentiation of the respective synaptic compartments from which the CAMs are derived.

However, there is an increasing body of evidence that suggests that neurexin and neuroligin are also critical components of the signalling cascades that function in the cellular expression of learning and memory (Kim et al. 2008; Etherton et al. 2009; Blundell et al. 2010). While investigating the role played by the neurexin-neuroligin transsynaptic interaction in molecular memory, Choi et al. (2011) discovered that the interaction contributes significantly to the initiation, stabilization and persistence of activity-dependent long-term synaptic plasticity and the associated 5-HT-induced strengthening and growth of sensorimotor connections. They showed by eradicating expression of either presynaptic neurexin or postsynaptic neuroligin, or by inducing over-expression of either neurexin or neuroligin alone, that the functional and structural expression of LTF is blocked (Choi et al. 2011). This suggested how important the coordinated functioning and transsynaptic interaction of both neurexin and neuroligin is in stabilizing the new synaptic growth and long-lasting forms of synaptic facilitation induced by 5-HT.

In addition, since some 'empty' presynaptic neuronal processes are enriched with neurexin and neuroligin upon treatment with 5-HT (which consequentially induces LTF) (Choi et al. 2011), and because this treatment leads to increased kinesin-mediated transport of both neurexin and neuroligin from the cell body to the synapse (Puthanveettil et al. 2008), it has been proposed that their coordinated increase in concentration at both the pre- and postsynaptic neurons of stimulated synapses probably involves 5-HT-induced, kinesin-mediated axonal mobilisation. Choi et al. (2011) also discovered that the persistence of LTF requires continuous synthesis of new neurexin and neuroligin beyond 24 h. Thus if one considers that neurexin mRNA contains a CPEB binding element in its 3′UTR (that regulates the local protein synthesis required for the long-term maintenance of LTF and synaptic growth; see section ‘Synapse-Specificity of Long-Term Synaptic Plasticity’) and that neuroligin mRNA is a target of CPEB in Drosophila (Mastushita-Sakai et al. 2010), it could be hypothesized that the presence of transsynaptic signalling molecules at stimulated synapses is also regulated by CPEB mediated local protein synthesis.

After consideration of both of the mechanisms that participate in neurexin and neuroligin synaptic enrichment after 5-HT induction, it has been suggested that the local increase in the concentration of these CAMs during LTF could be regulated in one of two possible ways. Either both processes of enrichment occur in a single varicosity, or the increase in kinesin-mediated transport and CPEB-mediated local protein synthesis occur separately and independently in distinct populations of varicosities (Choi et al. 2011).

It was also found that the Aplysia neurexin and neuroligin proteins possess a domain structure and subcellular location that exhibits many similarities to their corresponding vertebrate homologs (Choi et al. 2011). This, together with their existence in the genomes of Drosophila and C. elegans, provides further evidence for the large degree of evolutionary conservation of the neurexin-neuroligin transsynaptic interaction, as well as the mechanisms of synaptic plasticity as a whole (Tabuchi and Sudhof 2002). Ultimately, Choi et al. (2011) show that the activity-dependent regulation of the neurexin-neuroligin interaction (and their downstream signalling cascades) might subserve the transsynaptic signalling at the Aplysia sensorimotor synapse required for long-term memory storage.

The Autocrine Action of Sensorin in Long-Term Synaptic Plasticity

As illustrated (Fig. 1), 5-HT induces LTF by activating a number of complex signalling pathways that produce structural plasticity through gene and protein expression that is mediated by the timely activation of PKA, PKC, PI3K and MAPK (Kandel 2001). Both the local cytoplasmic substrates and transcription factors that are phosphorylated by these kinases, as well as the mechanisms through which the three former kinases are activated have been discussed in detail. With regards to MAPK however, it is known that its neuronal activation is regulated by activity-dependant increases in the production, secretion, and autocrine action of factors that activate receptor tyrosine kinases (TrkB) (Purcell et al. 2003), and, in particular, the sensory neuron neuropeptide called sensorin (Hu et al. 2004, 2006). In brief, the rapid increase in sensorin expression induced upon repeated stimulation by 5-HT is mediated predominantly by PI3K and rapamycin sensitive translation, while the subsequent release of the newly synthesised sensorin and the regulation of its signalling is mediated by PKA and PKC (Hu et al. 2006). It should be noted again that the activation of rapamycin-sensitive protein synthesis required for LTF is one of the known downstream effects of PI3K activity. This pathway can be activated through the same GPCRs that activate PKC or PKA (Yanow et al. 1998; Casadio et al. 1999).

When considered more carefully, it has been found that the synthesis and secretion of sensorin is regulated by a number of converging signalling pathways that respond to different stimuli (Hu et al. 2007). The use of five applications of 5-HT (referred to as nonassociative LTF, where a behavioural change occurs in response to a single type of stimulus such as multiple, spaced tail shocks) induces the rapid and local increase in sensorin synthesis through PI3K activity (Hu et al. 2006). However, with a combination of tetanus (an action potential) and 5-HT (associative LTF, where a behavioural association between two different types of stimuli are formed), PKC activity is required for the same production of sensorin but without participation by PI3K (Hu et al. 2007). As previously described, the synthesis of sensorin is also rapamycin sensitive, and thus, like many other proteins, its local synthesis contributes to long-term synaptic plasticity (Hu et al. 2006). The same signalling pathway for both nonassociative and associative forms of LTF follows the earlier increase in sensorin synthesis, and is mediated by both PKA and PKC (Hu et al. 2007). In light of this, both PKA and PKC activities contribute to the sensorin secretion regardless of whether they had been stimulated by five applications of 5-HT or by tetanus plus 5-HT.

Active synaptic type II PKA has been found retained close to the plasma membrane through interactions between its regulatory subunit and membrane anchoring proteins (Liu et al. 2004). This interaction is required for the release of sensorin (Hu et al. 2006). Increased phosphorylation of specific proteins, such as synapsin, that upregulate the secretion of sensorin may result from the persistent activation of this PKA by cAMP (Jovanovic et al. 2000; Angers et al. 2002), in a manner similar to the way in which serotonergic stimulation of PKA during LTF causes increased release of synaptic vesicles containing neurotransmitters (Byrne and Kandel 1996; Hu et al. 2007; Purcell et al. 2003), as was described earlier. In the same way, persistent activation of synaptic PKC may also lead to phosphorylation of synaptic proteins necessary for secretion of neuropeptides (Sossin et al. 1994; Sieburth et al. 2007), or more predictably, the phosphorylation of cytoskeletal proteins that regulate the mobilisation of synaptic vesicles containing sensorin for release at the required locations at the cell membrane (Knox et al. 1992; Nagy et al. 2002; Nakhost et al. 2002). This is supported once again by the role PKC plays in mobilisation of neurotransmitter containing synaptic vesicles encountered when a Ca²⁺ ion influx leads to increases in synaptic strength during LTF (Byrne and Kandel 1996). Thus, the two kinases, PKA and PKC, are thought to act together during persistent activation during the different forms of LTF by phosphorylating either distinct combinations of synaptic proteins that have complementary functions, and/or different sites on the same proteins, all of which regulate the release of newly yet locally produced neuropeptides such as sensorin (Hu et al. 2006, 2007). These proteins may thus include sensorin receptors, those that mediate receptor trafficking, relevant cascade proteins that activate downstream signalling, as well those involved in exocytosis (Meyer-Franke et al. 1998; Du et al. 2000; Patterson et al. 2001). These pathways parallel the activity-dependant regulation of local synthesis and secretion of brain-derived neurotrophic factor (BDNF), a neuropeptide that also plays an important role in long-lasting forms of LTF in the brain (Hall et al. 2000; Pang et al. 2004).

As mentioned, the MAPK pathway is one such pathway that is activated by sensorin and BDNF upon release and subsequent binding to TrkB-like receptors. The translocation of the activated phosphorylated form of MAPK into the nucleus of the sensory neuron is facilitated by PKA, whose 5-HT induced activation is required for the activation and translocation of MAPK by sensorin (Hu et al. 2004, 2006). Therefore, activated PKA not only stimulates the translocation of MAPK into the nucleus to phosphorylate transcription factors and other components that regulate the gene expression required for LTF, but is also responsible for the release of some of the neuropeptides necessary for the auto-activation of the MAPK requiring translocation (Hu et al. 2007; Purcell et al. 2003). Moreover, PKC is required not only for the synthesis of the neuropeptides required for MAPK activation, but is also essential for their synaptic release to facilitate the subsequent autocrinal activation of MAPK. Therefore, the long-lasting synaptic changes that are required for LTF are "orchestrated by the timely and persistent activation of signalling pathways, first by the initial stimuli and then followed by the actions of a secreted neuropeptide" (Hu et al. 2007).

It has been suggested that the changes in the synthesis and secretion of sensorin that may occur locally in specific synapses (and the subsequent activation of the related downstream signalling cascades) may also contribute to the ‘tagging’ phenomenon experienced by activated synapses, marking them for expression of long-term synapse-specific plasticity (Casadio et al. 1999; Hu et al. 2006). Furthermore, the increased activity of transcription factors activated by MAPK translocation and persistent PKA activity in the sensory neuron nuclei induced upon sensorin signalling, may also be responsible for the following upsurge of sensorin synthesis found to be necessary for the long-lasting form (>72 h) of LTF (Casadio et al. 1999; Sherff and Carew 1999). As the long-term maintenance of synaptic contacts require the continuous secretion of sensorin (Hu et al. 2004), the second increase in sensorin protein production described may induce a relative increase in sensorin secretion, thereby facilitating the required long-term maintenance (Hu et al. 2006). The initiation and maintenance of long-term changes in synapse function and structure are therefore contributed to by the early changes induced by the activation of the complex signalling pathways by 5-HT and sensorin, as well as by the consequential synthesis and secretion of further sensorin neuropeptides (Hu et al. 2006).

Presenilins and Their Role in Synaptic Plasticity

Presenilins are essential components of γ-secretase, a multiprotein protease complex that is responsible for the cleavage of the Notch receptors and the amyloid precursor protein in mammals (De Strooper et al. 1998, 1999). Mutations in presenilin 1 and presenilin 2 are the primary cause of familial Alzheimer’s disease (FAD) (Hutton and Hardy 1997), a neurodegenerative illness characterised by progressive cognitive deterioration by progressive loss of neurons and synapses, as well as by formation of amyloid plaques and intracellular neurofibrillary tangles. Mutations in human presenilins that have been linked to FAD are known to enhance production of the amyloidogenic 42-residue β-amyloid (Aβ) peptide specifically over that of the less amyloidogenic Aβ40 generation (Moehlmann et al. 2002; Schroeter et al. 2003). In light of this it has been suggested that the pathological mechanism involves a toxic gain-of-function, yet there are possibilities that a partial loss of presenilin function may play a role in the disease process (Saura et al. 2004). However, the strategy whereby presenilin mutations cause memory loss and neurodegeneration in FAD still remains unclear.

A number of the molecules that play a significant role in the stimulation of LTP, including BDNF and CREB, are also implicated in neuronal survival. The discovery that inactivation of CREB in the postnatal forebrain causes significant neurodegeneration (Mantamadiotis et al. 2002), and that a reduction in CRE-dependent gene expression by reduced CBP function has been implicated in the pathogenesis of neurodegeneration in certain disorders (Nucifora et al. 2001), implies the existence of a link between synaptic plasticity (LTP) and neuronal survival. Unfortunately, the impact of genetic defects in FAD on the pathways fundamental to LTP and memory remain to be understood.

Nevertheless, it has been shown that normal presenilin 1 plays a critical role in neurogenesis and on the Notch signalling cascade (Handler et al. 2000; Shen et al. 1997), which is a pathway that contributes to CREB-1 transcription and the subsequent synaptic changes necessary for LTP. It has been documented that presenilin participates in the proteolytic cleavage of the Notch receptor to produce the constitutively active Notch intracellular domain (NICD) (Saura et al. 2004). NICD then translocates to the nucleus where it interacts with the sequence-specific DNA binding factor CBF-1 (C-promoter binding factor 1) (Saura et al. 2004). Following this, the co-activated NICD-CBF-1 complex binds to the CBP proximal promoter of the CBP gene, thereby activating the transcription of the CREB binding protein (CBP) (Saura et al. 2004). CBP is an important cofactor (acting together with CAMAP) for transcriptional activation by CREB-1, indicating that CRE-dependent gene expression is indirectly regulated by presenilin through its regulation of Notch-dependent CBP expression (Saura et al. 2004).

Presenilin is also essential for neuronal survival as it has been established that the loss of presenilin function causes progressive neurodegeneration (Saura et al. 2004). Synaptic and neuronal loss is also preceded by defects in LTP when presenilin is inactivated. This illustrates presenilin’s importance in long-term memory as well as in the neuronal survival that permits the maintenance of LTP (Saura et al. 2004).

Long-Term Memory Endurance Through Prion Formation

Miniaci showed that it is necessary for CPEB to mediate local synaptic protein synthesis continuously for at least 72 h to induce persistent synaptic facilitation in Aplysia (Miniaci et al. 2008). However, although synaptic stimulation increases the activity of CPEB, it is logical to suppose that this change should be effective for a limited duration. How then can functional CPEB be maintained for extended periods to allow its required persistence? Even with the continuous turnover of active proteins (occurring every few hours) that might participate in the expression of LTF, the molecular basis of long-term memory must account for its lasting endurance of up to many years (Bailey et al. 2004). Self-perpetuating protein-based phenomena such auto-phosphorylation loops, complex signalling networks comprising self-sustaining feedback loops, and transcription factors that stimulate their own transcription, are a few possible mechanisms that explain information storage despite molecular turnover and how a transiently active molecule can produce enduring changes (Reviewed in Shorter and Lindquist 2005). Another plausible explanation for this however is provided by the conformational replication of prions. Prions comprise a distinctive type of protein that can assume two functionally and morphologically distinct conformational states, one of which is multimeric and self-perpetuating (Prusiner 1998). These proteins may thus act as a template for proteins possessing similar amino acid sequences, causing them to adopt the prion conformation in a protein-folding like chain reaction (Prusiner 1998). Prions are therefore capable of acting as genetic elements that facilitate the self-replication of conformational information (Prusiner 1998; Shorter and Lindquist 2005). Lending confidence to the involvement of prions in LTF, Si et al. (2003) discovered that ApCPEB has prion-like properties and could (much like other prion proteins) exist in at least two distinct conformational states; a monomeric and multimeric, prion-like form.

In an unstimulated synapse, it was found that CPEB is monomeric and is either inactive or acts as a repressor of LTF (Si et al. 2010). As was described above, following synaptic stimulation by repeated pulses of 5-HT, PI3K activation induces rapid localised bursts of translation of neuronal CPEB mRNA at the specific synapses stimulated by the neurotransmitter (Shorter and Lindquist 2005). The local overexpression and increased concentration of CPEB induces the formation of non-toxic amyloidogenic CPEB prion conformers that exist as active, dominant and self-sustaining multimeric states resulting from homotypic CPEB interactions (Shorter and Lindquist 2005; Si et al. 2010). The CPEB prions formed thereafter allow a sustained period of translation of dormant CPE-containing mRNAs required for LTF only at those specific synapses initially stimulated (Shorter and Lindquist 2005). This active CPEB prion state thus self-propagates and induces the formation of further CPEB multimers. They are therefore able to persist without any further stimulation or activation, facilitating the maintenance of the synaptic changes generated during LTF and may represent the self-sustaining synaptic ‘tag’ that is limited to activated synapses (Shorter and Lindquist 2005; Si et al. 2010). Not only does the stability of the prion fibres mean that they are less likely to be degraded than soluble proteins (Venkatraman et al. 2004), but the CPEB prion conformer’s state of aggregation limits them to the activated synapses, leaving the unstimulated synapses of the same neuron inactivated (Shorter and Lindquist 2005). CPEB prions may thus, in part, represent the physical manifestation of a memory in neuronal tissues. Another advantage of having the prion-like multimerisation is that if the monomeric form of the CPEB protein exhibits little activity, having multiple entities would essentially increase the apparent activity (Si et al. 2010). This hypothesis has much support, particularly as it has been found that the non-neuronal forms of CPEB (such as that found in Xenopus oocytes) are activated through phosphorylation by Eg2 of a canonical LDS/TR site, a sequence that is not present within the neuronal CPEB (Mendez et al. 2000; Si et al. 2003). As the phosphorylation of CPEB by Eg2 increases the oocyte CPEB activity by four to fivefold, the inactive unphosphorylated neuronal CPEB protein may be able to achieve the same degree of activation through a four to fivefold increase in its concentration, such as that provided by the formation of the prion multimer state induced upon neuronal CPEB mRNA overexpression (Si et al. 2003). ApCPEB homologues have also been discovered in Drosophila (Keleman et al. 2007) and in mammals (Alarcon et al. 2004), and may possibly contribute to memory maintenance in a similar manner.

With regard to synapse-specific tagging, the fact that CPEB is activated through extracellular signals to activate translationally dormant mRNAs that are implicated in synaptic growth (Wu et al. 1998), as well as being spatially restricted and self-perpetuating (Shorter and Lindquist 2005), makes CPEB an attractive candidate for a synapse-specific mark for stabilisation.

In light of this, one might question how long-term memory may be forgotten in the context of the self-perpetuating nature of prion-based memory. There is suggestion however that neurons may express a different CPEB isoform lacking a prion domain that causes reversal of prion formation as is the case in many other yeast prions (Komar et al. 2003; Shorter and Lindquist 2005). In addition, functionally diverse ATPases or molecular chaperones similar to Hsp104 that function in disassembly of the CPEB fibres at synapses, thereby reversing the LTF may exist (Si et al. 2003).

Based on the evidence derived from studies of a relatively novel signalling molecule, protein kinase M (PKM), a distinctive and contrasting molecular mechanism for the long-term maintenance of long-term memory has recently been proposed. Both PKMζ, the constitutively active fragment of the mammalian atypical protein kinase C (PKCζ), and PKM Apl III, the constitutively active fragment of the Aplysia atypical protein kinase C (PKC Apl III), have been shown to play critical roles in the persistence of LTP and LTF, respectively (Ling et al. 2002; Pastalkova et al. 2006; Cai et al. 2011). The only notable difference between PKMζ and PKM Apl III is the means by which they are generated. While PKMζ is produced by transcription from an alternate start site within the atypical PKCζ gene (Hernandez et al. 2003), PKM Apl III is produced when PKC Apl III undergoes site-specific proteolytic cleavage by calpain to generate a PKM fragment (Bougie et al. 2009).

While challenging the traditional model of memory persistence and consolidation (which includes the prion-like activity of CPEB described above), and expanding upon the work of Sacktor (see Sacktor 2011 for a review of this work), Cai et al. (2011) showed that LTF is erased in Aplysia when PKM Apl III is inhibited by either ζ inhibitory peptide (a pseudosubstrate inhibitor) or chelerythrine (a PKC inhibitor). Interestingly, despite the fact that 5-HT appears to activate PKM Apl III in motor neurons of Aplysia (Villareal et al. 2009), there is neither spontaneous recovery of the long-term synaptic changes even long after treatment with the described inhibitors, nor is there reinstatement of the long-term synaptic changes when the cells are stimulated again in a manner that normally induces STF (Cai et al. 2011). More specifically, when PKM Apl III was inhibited at 24 h after 5-HT treatment, continued expression of LTF was disrupted 24 h later. Inhibition of ApCPEB at 24 h produces the same effect (see above, and Si et al. 2003), indicating that there is some degree of temporal overlap of these two mechanisms of memory-maintenance. The exact nature of their interaction is yet to be understood, but in an attempt to delineate the relative roles of ApCPEB and PKM Apl III in the persistence of memory, Cai et al. (2011) proceeded to temporarily inhibit protein synthesis to test whether it disrupts long-term memory. They discovered that at 1 week post stimulation, temporarily preventing the synthesis of protein does not affect the expression of long-term memory. This is in contrast to the reversal of LTF upon protein synthesis inhibition at 24–48 h after stimulation, as seen by Miniaci et al. (2008) while investigating the role of ApCPEB in LTF as described above.

How then does PKM Apl III allow for the persistence of LTF once the requirement for protein synthesis has diminished? The mechanisms by which PKMζ and PKM Apl III achieve this are unfortunately still poorly understood. The finding that temporarily inhibiting either form of PKM completely eradicates further LTF or LTP is difficult to explain. It could be expected that after the inhibitor concentration has become low enough through either diffusion or degradation to have no effect, that the previously established long-term memory would be restored. According to one model suggested by Sacktor (2011), PKMζ remains active at potentiated synapses by a positive-feedback loop that employs GluR2 subunit-containing AMPA receptors and their trafficking to these potentiated synapses. (for a description of the role that AMPA receptors play in vertebrate and invertebrate long-term memory, see section ‘Postsynaptic Mechanisms Contributing to Long-Term Synaptic Plasticity’). This model proposes that AMPA receptors are transported to the postsynaptic membrane, where it may act as a synaptic tag, when the GluR2 subunit on the receptor is phosphorylated by PKMζ (Migues et al. 2010). Inhibitors of PKMζ interrupt the positive-feedback loop by preventing the continuous phosphorylation of the GluR2 subunit, causing the endocytosis of the AMPA receptor. With this synaptic tag now fully removed, the associated PKMζ is dislocated from the synaptic region and cannot be stimulated through its pathways of activation to induce trafficking of AMPA receptors to the formerly potentiated synapses, even when PKMζ inhibitors are removed. This results in the synapses reverting to their naive form. Although these mechanisms have been proposed for mammalian LTP based on studies of memory formation in the hippocampus of rodents (see Sacktor 2011 for a review of this work), support for its existence in Aplysia LTF arises from the finding of Cai et al. (2011) that temporarily inhibiting protein synthesis has no effect on the expression of established long-term memory, yet PKM Apl III inhibition does.

There is thus a possibility that Aplysian long-term synaptic plasticity is preserved through a positive-feedback loop that involves on-going protein phosphorylation by PKM Apl III and the consequential enhanced trafficking of AMPA receptors to postsynaptic sites, as suggested by Sacktor’s model for mammalian memory. Cai et al. (2011) have also proposed that continued PKM Apl III activity may play a role in the maintenance of stable synaptic facilitation through the preservation of learning-induced changes in synaptic structure (Liu et al. 2009).

At around the same time, Hu et al. (2011) reported findings remarkably consistent with those of Cai et al. (2011), yet their different focus on questioning how long-term memory is maintained, brought about the discovery that an additional rapid increase in the synthesis and secretion of sensorin is also induced by a second batch of 5-HT stimulation at 24 h and is required for the described persistence of LTF in Aplysia (see section 'The Autocrine Action of Sensorin in Long-Term Synaptic Plasticity'). Hu et al. (2011) confirmed that the additional stimulus extends the duration of LTF by newly inducing PKC (potentially the atypical form, PKM, as described above by Cai) and its downstream signalling pathways to control the persistent expression of sensorin, the same neuropeptide that participates in initiating LTF during the first stimulus.

This was realized after the inhibition of the downstream actions of the newly synthesised sensorin produced by the second set of 5-HT treatments disrupted persistent LTF, which indicated that it is indeed required for the long-term maintenance of memory after 24 h (Hu et al. 2011). Corresponding to their previous findings regarding sensorin, they showed that while the first wave of the rapid 5-HT-induced increase in sensorin synthesis required to initiate LTF is mediated by PI3K (Hu et al. 2006, and see above), the new persistent increase in sensorin synthesis and signalling, and the consequent persistence in LTF, is induced by activation of PKC by the additional stimuli. The use of rapamycin or chelerythrine [the PKC inhibitor also employed by Cai et al. (2011) in their related experiments] during the second series of stimulation by 5-HT presented a rapid decrease in the levels of sensorin, indicating that the inability to maintain high levels of sensorin expression (and rapamycin-sensitive protein synthesis) by PKC may lead to the abolishment of LTF (Hu et al. 2011). Also, inhibiting the activity of PKC during the first and PI3K during the second set of 5-HT treatments had no effect on persistent LTF or the rapid increases in sensorin expression, indicating that PI3K and PKC function sequentially and alone in the first and second stages of stimulation, respectively (Hu et al. 2006, 2011).

Hu et al. (2011) do not exclude the possibility that the activities of the same or different PKC isoforms, including that of the atypical PKC isoform PKMζ/PKM Apl III, converge to regulate the described synthesis of sensorin. Moreover, it is suggested that the PKC isoforms may act either in the presynaptic neuron to regulate the translation of sensorin mRNA (Hu et al. 2006) by a CPEB- and or CREB-dependent mechanism, or may act in the postsynaptic neuron to produce a retrograde signal that controls the synthesis of sensorin in the presynaptic neuron (see section 'Postsynaptic Mechanisms Contributing to Long-Term Synaptic Plasticity') (Hu et al. 2011).

Considering the recent findings of both Cai et al. (2011) and Hu et al. (2011), it is evident that the PKC signalling cascade (that of the atypical protein, PKM, or both PKC and PKM) is newly recruited and sequentially activated after PI3K within the later stages of LTF (seen to be enhanced by additional stimuli 24 h after LTF induction) to regulate the constitutive synthesis and secretion of sensorin to produce a persistent form of synaptic plasticity (Hu et al. 2011). The requirement for CPEB-dependent protein—and thus sensorin—synthesis then diminishes by about 1 week after stimulus. Thereafter potential positive-feedback loops involving the trafficking of AMPA receptors by PKM allow for the further persistence of long-term memory (Cai et al. 2011). This is in contrast to the prion-like CPEB model for memory maintenance, since it states that persistent macromolecular synthesis is required for extended periods of long-term forms of synaptic plasticity (see above, and Si et al. 2010).

Postsynaptic Mechanisms Contributing to Long-Term Synaptic Plasticity

Investigations into the neuromolecular mechanisms underlying memory and LTF in the marine snail Aplysia have primarily focused on the changes in plasticity that occur within the presynaptic sensory neuron. Much controversy and debate has therefore grown over whether the site of LTF expression is predominantly presynaptic or postsynaptic. Recently however, much experimental evidence has demonstrated that postsynaptic mechanisms make a critical contribution to activity-dependent LTF, particularly through the modulation of the trafficking of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA)-type glutamate receptors (Roberts and Glanzman 2003; Nicoll 2003; Cai et al. 2008).

As described initially, the typical neuronal circuit important in LTF comprises serotonergic interneurons that act on both sensory and motor neurons to regulate the strength of their connections (an activity-independent, heterosynaptic form of plasticity), a process that has only been discovered in invertebrates. Therefore, within the sensitisation of the withdrawal reflex by noxious stimuli of Aplysia’s gill and siphon for example, the modulatory interneurons secrete serotonin that not only functions to stimulate the presynaptic sensory neurons, but also acts upon the postsynaptic motor neuron. This stimulation causes the activation of signalling pathways within the postsynaptic neuron that mediate persistent presynaptic and postsynaptic plastic changes for LTF. Conversely, NMDA receptor-dependent LTP is a homosynaptic form of plasticity that requires no external serotonergic modulatory input, but rather occurs intrinsically by the sensorimotor synapse upon tetanic stimulation (spike activity). This form of synaptic plasticity has been described in both vertebrates and invertebrates (see section 'Commonalities of Vertebrate and Invertebrate Long-Term Synaptic Plasticity'). During LTP, tetanic stimulation of the sensory neuron leads to the presynaptic release of the neurotransmitter glutamate through a variety of interconnected pathways not unlike the pathways described in the section 'Presynaptic Mechanisms Contributing to Long-Term Synaptic Plasticity'. Glutamate acts on several postsynaptic receptor types that include NMDA and AMPA receptors which are permeable to Ca²⁺ ions and monovalent cations (Na⁺ and K⁺ ions), respectively (Collingridge 1985). Under normal conditions, glutamate released presynaptically during short-term potentiation binds to both AMPA and NMDA receptors, yet blockage of the NMDA receptor systems by Mg²⁺ ions present within the extracellular fluid prevents the activation of this ion channel (Nowak et al. 1984; Herron et al. 1985). Mg²⁺ inhibition is only removed by sufficient depolarisation of the postsynaptic cell, to the extent that a single EPSP does not provide the magnitude and duration of depolarisation required (Nowak et al. 1984). This is significant if one considers the ‘slow build-up’ of the NMDA receptor mediated response during postsynaptic expression of LTP which is known to occur (Collingridge 1985). The initial excitatory synaptic response required for depolarisation is therefore mediated predominantly by monovalent cations such as Na⁺ and K⁺ ions permeating through the AMPA receptors that are activated after binding glutamate (Collingridge and Bliss 1987). However, the stronger and longer lasting depolarisation facilitated by increased permeation of AMPA receptors during high frequency stimulation, which would result in the progress of LTP, alleviates the Mg²⁺ block of the NMDA channels (Nowak et al. 1984). Ca²⁺ is then able to enter the cell via the unblocked NMDA ion channels. This triggers a signal cascade that leads to the enhanced synaptic activity and plastic change required for LTP. Other factors contributing to sufficient depolarisation may include the temporal summation of EPSPs originating from AMPA receptors or from the NMDA receptor itself, and/or accumulation of extracellular potassium (Collingridge and Bliss 1987).

The opening of NMDA channels therefore has two requirements; the association of the NMDA receptor with an agonist, as well as strong local depolarisation (Collingridge and Bliss 1987). These requirements explain a few important characteristics of LTP, among them, co-cooperativity, which has an intensity threshold for LTP initiation, and associativity, in which a weak stimulus is able to induce LTP if applied in conjunction with a strong stimulus impinging upon a converging pathway (as observed in synaptic tagging and capturing) (Collingridge and Bliss 1987). Both properties are explained by the requirement for strong-local depolarisation to reduce the blockage of NMDA channels by Mg²⁺ ions.

In addition to the usual mechanisms contributing to the maintenance of LTP, (including the sustained increase in presynaptic transmitter release and the growth of new synaptic connections), it has also been found that application of a train of action potentials and the associated postsynaptic entry of calcium causes the activation of calpain, a calcium dependent protease that relieves the inhibition of a covert sub-population of NMDA glutamate receptors (Lynch and Baudry 1984). Although these do not include the AMPA subset of receptors, calpain assists in the maintenance of LTP by increasing the number of functional postsynaptic receptors available for stimulation (Collingridge and Bliss 1987).

An examination of the related postsynaptic pathways of invertebrate LTF reveals that when the interneuron modulating LTF is stimulated during sensitisation, it secretes 5-HT that also acts on the postsynaptic neuron. The 5-HT released activates PLC in the motor neuron via a GPCR, whose activation leads, as before, to the conversion of PIP₃ into IP₃ and DAG. IP₃ then causes a further increase in postsynaptic Ca²⁺ (along with the Ca²⁺ influx resulting from activation of NMDARs) by inducing its release from IP₃-receptor mediated intracellular stores (Bezprozvanny et al. 1991; Hagar et al. 1998). The Ca²⁺ ions released induce an additional increase in their own concentration by activating ryanodine receptors (RyRs), another class of intracellular calcium channels that mediate calcium-induced calcium release (CICR) in animal cells (Li et al. 2005). These Ca²⁺ sources may thus act synergistically, as an increase in intracellular Ca²⁺ ions also augments IP₃-receptor mediated release from the described intracellular stores (Hagar et al. 1998).

The resulting prolonged increase in the concentration of postsynaptic intracellular Ca²⁺ ions leads to many downstream consequences. Predominantly these include the activation of PKC (Villareal et al. 2009). PKC and CaMKIIα are also activated by the DAG produced on PLC stimulation and the consequent catalytic cleavage of PIP₃. The increased Ca²⁺ ions, as well as both of the activated PKC and CaMKIIα kinases, then drive local postsynaptic protein synthesis and the upregulation of AMPA receptor function (Trudeau and Castellucci 1995; Zhu et al. 1997). Enhancement of AMPA receptor function may occur through the synthesis of new AMPA receptors concomitantly with the increased synthesis of postsynaptic protein, the exocytotic delivery and insertion of further AMPA receptors into the postsynaptic membrane, and/or a combination of both (Li et al. 2005; Villareal et al. 2007).

Most intriguingly, the rise of Ca²⁺ concentration in the motor neuron during both LTP and LTF also stimulates one or more retrograde signals that are released postsynaptically and that influence certain presynaptic signalling cascades (Roberts and Glanzman 2003). Although the exact nature of the retrograde signal is not fully known, it has been found to contribute to presynaptic expression and secretion of sensorin through the activation of PI3K and PKA, respectively (Hu et al. 2004, 2006). As outlined above, after binding to its auto-receptors, sensorin induces the phosphorylative activation of MAPK, its subsequent translocation into the presynaptic nucleus together with PKA, and the regulation of the gene expression required for LTF (Hu et al. 2006, 2007). Changes in postsynaptic transcription during LTF have been proposed to accompany these necessary modifications to presynaptic transcription (Cai et al. 2008).

According to this postsynaptic model, only these retrograde signals are able to induce the persistent presynaptic cellular changes that are required for LTP or LTF, including the persistent increase its excitability (Antonov et al. 2003). Roberts and Glanzman (2003) state it well when commenting that the “persistent associative changes are induced entirely postsynaptically but are expressed, in part, via transsynaptic activation of PKA within the sensory neuron”. The retrograde signal may also be the stimulating factor that is required for the persistent activation of the presynaptic PKA which is produced through the liberation of its catalytic subunit from its inhibitory subunit, thereby facilitating its translocation into the nucleus. The uncertainty of a 'sufficient rise and persistence in cAMP levels' would thus not be required for this purpose as was described in the presynaptic model. While cAMP may induce the initial activation of PKA required in short-term plasticity, once the dendritically localised PKA has produced a significant increase in synaptic strength during the early stages of LTP/LTF, the resulting increase in glutamate release and the eventual postsynaptic release of the retrograde signal may be the factor that stimulates the nuclear translocation of PKA required in the later stages of LTP/LTF. This has been hypothesised because the initial postsynaptic activation of PKA causes plastic changes. These include short-term increases in transmitter release and neuronal excitability that does not persist in the absence of a signal from the postsynaptic neuron (Roberts and Glanzman 2003; Cai et al. 2008). It is more likely however when considering the evidence obtained from both presynaptic and postsynaptic based studies, that the retrograde signal acts cooperatively with the signalling events induced within the presynaptic neuron that are required for activation of PKA, PI3K, and MAPK and the consequent synaptic changes accompanying LTP or LTF.

Intermediate-Term Synaptic Plasticity

Intermediate-term facilitation (ITF) is a memory phase which has only been fully recognized within the last decade. ITF can be distinguished from STF and LTF through both temporal and mechanistic differences in its expression (see Sutton and Carew 2000 for an extensive review). Two distinct forms of ITF have been identified. These are based on the means by which they are induced and the molecular events required for their expression. The first form can be induced by multiple, spaced applications of 5-HT in the absence of sensory neuron activity, in a manner identical to LTF. Here, 5 individual 5-HT applications stimulates synaptic facilitation that is eliminated within 3 h after stimulus (ITF), but fully reappears by 24 h after stimulus (LTF) (Ghirardi et al. 1995; Mauelshagen et al. 1996). This form of ITF (activity-independent ITF) thus comprises part of the first phase of a biphasic pattern of facilitation produced by multiple 5-HT applications, concluding with LTF as the second phase. This activity-independent ITF can be differentiated from STF in its dependence on protein synthesis, and differentiated from LTF as it is not dependent on gene transcription (Sutton and Carew 2000). However, the maintenance of ITF induced by multiple applications of 5-HT, like LTF, also requires the persistent phosphorylative activation of presynaptic PKA (Sutton and Carew 2000).

The second form of ITF that has been identified can be induced by coupling a single pulse of 5-HT with coincident sensory neuron activity, and for this reason is referred to as activity-dependent ITF (Bao et al. 1998; Sutton and Carew 2000). This form of ITF is distinct from activity-independent ITF in that it is independent of protein synthesis for its induction, and it is dependent on persistent activation of presynaptic PKC, rather than PKA (Sutton and Carew 2000; Zhao et al. 2006). Interestingly, since both persistent PKA and PKC contribute to the expression of LTF (see above), their initial constitutive activation in each of their respective forms of ITF may contribute to the later induction of LTF during the second phase of the described biphasic pattern of synaptic facilitation.

It was initially unclear whether postsynaptic mechanisms contribute to the expression of ITF, despite evidence that they play a significant role in activity-independent ITF (Chitwood et al. 2001; Li et al. 2005; Villareal et al. 2009). As described previously, many of these features are also critical for LTF expression, suggesting the existence of temporal and mechanistic overlap of ITF and LTF mechanisms. This also explains their manifestation as sequential biphasic patterns of facilitation. Initial results showed that brief pulses of glutamate, coincident with a 10 min application of 5-HT, to an isolated motor neuron, produced an enhancement of glutamate action potentials that persisted for ≥2 h. This enhancement was found to be dependent on elevated intracellular Ca²⁺ levels (Chitwood et al. 2001). Using sensorimotor co-cultures, Li et al. (2005) subsequently demonstrated that synaptic facilitation, specifically ITF, requires elevated intracellular Ca²⁺ levels, which is released from the receptor mediated-intracellular stores in response to both inositol 1,4,5-trisphosphate (IP₃) and ryanodine. Concurrently it was discovered that, during ITF, 5-HT modulates the functional expression of postsynaptic AMPA-type receptors by inducing their exocytotic insertion into the plasma membrane (Chitwood et al. 2001; Li et al. 2005).

Questioning whether the protein synthesis required for activity-independent ITF was located postsynaptically, Villareal et al. (2004, 2009) showed that the enhancement of the glutamate response produced by a 10 min application of 5-HT in an isolated motor neuron required local and rapid protein synthesis in the neurites of the motor neuron, rather than in the cell body. The exact identity of the postsynaptic proteins whose synthesis is induced by the 5-HT in ITF is unknown. However, it has been shown that dendritic synthesis of AMPA receptors in hippocampal neurons is stimulated by application of dopamine (Smith et al. 2005). When considered with respect to the enhanced exocytotic insertion of AMPA receptors into the plasma membrane after a 10 min application of 5-HT, the upregulation of local AMPA receptor synthesis would be an attractive mechanistic feature of ITF. The signalling pathway that regulates this postsynaptic protein synthesis also remains to be elucidated, but there is evidence that PKC may play a role in the process (Villareal et al. 2003).

Although it evident that there are still large gaps in the understanding of the molecular mechanisms of intermediate-term facilitation, it is apparent that distinct sets of synaptic processes are initiated depending on the features of various specific learning experiences, even when the resulting effects of those processes are exhibited in the same time domain.

Concluding Remarks

The timely and sequential activation of many different interrelated signalling pathways within both the presynaptic/sensory and postsynaptic/motor neurons, mediated in part by the secretion of glutamate, newly synthesised sensorin and retrograde signals, is required for the expression of LTP and LTF and its corresponding structural plasticity. The short- and intermediate-term processes of STF and ITF, and the more persistent long-term processes of LTP and LTF overlap both spatially and temporally, while the phosphorylating activities of specific kinases located at synaptic terminals must be coordinated with the events induced in the nucleus to allow long-lasting functional and structural changes to be initiated and maintained at activated or ‘tagged’ synapses (Roberts and Glanzman 2003). Thus, the many sequential, temporal, and interconnected activities of the kinases PKA, PKC, PKM, PI3K, MAPK and CaMKIIα, are regulated by serotonin, sensorin and unknown retrograde signals to produce the synaptic changes implicated in the many forms of learning and memory.

Nevertheless, although the synaptic model presented for the pathways involved in cellular learning is comprehensive, it is not wholly embracing. Much is still to be discovered, and the true molecular nature of many of the components described (such as the properties of the synapse specific ‘tag’ and that of the postsynaptic retrograde signal to name just a few) is yet to be fully elucidated. Furthermore, the fact that other MAP kinases (of which extracellular-signal-regulated kinases is one example), hormones and growth factors, as well as the involvement of other signalling molecules such as insulin, have recently been revealed to play additional roles in long-term synaptic plasticity indicates that although the ‘map to molecular memory’ presented here has integrated much of the published research data, it is still far from complete.

Abbreviations

5-HT

5-Hydroxytryptamine

AC

Adenylate cyclase

AF

Activating factor

AMPAR

α-Amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid-type glutamate receptor

BDNF

Brain-derived neurotrophic factor

CAM

Cell-adhesion molecule

CAMAP

CAM-associated protein

CaMKIIα

Ca²⁺/calmodulin-dependent protein kinase IIα

CBF-1

C-promoter binding factor 1

CBP

CREB binding protein

C/EBP

CAAT box enhancer binding protein

CICR

Calcium-induced calcium release

CPE

Cytoplasmic polyadenylation element

CPEB

CPE-binding protein

CNS

Central nervous system

CRE

Cyclic AMP response element

CREB

CRE-binding protein

DAG

Diacylglycerol

eEF2

Eukaryotic elongation factor 2

EF1α

Elongation factor 1α

eIF4e

Eukaryotic initiation factor 4e

EphA2

Ephrin A2

EPSP

Excitatory postsynaptic potential

ERK

Extracellular-signal-regulated kinase

FAD

Familial Alzheimer’s disease

GPCR

G-protein coupled receptor

GPI-apCAM

Glycosyl phosphatidylinositol-linked isoform of Aplysia CAM

I_KS

Voltage-independent K⁺ channel

I_KV

Voltage-dependent K⁺ channel

Imp

Importin

IN

Interneuron

IP₃

Inositol triphosphate

IP₃R

Inositol triphosphate-mediated receptor

ITD

Intermediate-term depression

ITF

Intermediate-term facilitation

LTD

Long-term depression

LTF

Long-term facilitation

LTP

Long-term potentiation

MAPK

Mitogen-activated protein kinase

NICD

Notch intracellular domain

NMDAR

N-methyl-d-aspartic acid receptor

P

Phosphate group

PAP

Poly (A) polymerase

PI₃K

Phosphoinositide 3-kinase

PIP₃

Phosphatidylinositol bisphosphate

PKA

Protein kinase A

PKC

Protein kinase C

PKM

Protein kinase M

PLC

Phospholipase C

PS

Presenilin

rS6K

Ribosomal protein S6 kinase

RyR

Ryanodine receptor

STD

Short-term depression

STF

Short-term facilitation

STP

Short-term potentiation

TM-apCAM

Transmembrane isoform of Aplysia CAM

TrkB

Receptor tyrosine kinase

Ub

Ubiquitin