HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (236) Citing Articles via Google Scholar Google Scholar Articles by LYNCH, M. A. Search for Related Content PubMed PubMed Citation Articles by LYNCH, M. A.

Long-Term Potentiation and Memory

Trinity College Institute of Neuroscience, Department of Physiology, Trinity College, Dublin, Ireland

ABSTRACT

I. INTRODUCTION

A. The Hippocampus and Spatial Memory

B. Synaptic Modifications and Memory

II. SEVERAL AFFERENT PATHWAYS SUPPORT LONG-TERM POTENTIATION

A. The Amygdala

B. The Visual Cortex and the Somatosensory Cortex

C. The Prefrontal Cortex

D. The Subiculum

III. MECHANISMS UNDERLYING LONG-TERM POTENTIATION

A. NMDA Receptor Activation and LTP

B. NMDA Receptor Activation, Learning, and Memory

C. Metabotropic Glutamate Receptors and LTP

IV. WHAT SIGNALING EVENTS FOLLOW N-METHYL-D-ASPARTATE RECEPTOR ACTIVATION?

A. A Role for CaMKII in LTP

B. A Role for CaMKII in Learning/Memory

C. AMPA Receptors and LTP

D. Silent Synapses, Learning, and Memory

V. INDUCTION OF LONG-TERM POTENTIATION ACTIVATES SEVERAL CELL SIGNALING CASCADES

A. cAMP

B. ERK

C. Phosphatidyinositol 3-Kinase

D. The Consequences of CREB Activation

E. Activation of IEGs and Late-Response Genes in LTP

F. Protein Synthesis and LTP

VI. NEUROTROPHINS, LONG-TERM POTENTIATION, AND MEMORY

VII. SYNAPTIC PLASTICITY AND THE STRESSED BRAIN

A. Behavioral Stress

B. Oxidative Stress

C. Irradiation Stress

D. Age, LTP, Learning, and Memory

E. Cognition and Inflammation

1. IL-1{beta}, LTP, learning, and memory

2. What signaling cascades are activated by IL-1{beta} and ROS in hippocampus?

3. LTP, cognitive function, and {beta}-amyloid

4. Gene expression and age

VIII. LONG-TERM POTENTIATION AND MEMORY: DO THEY SHARE CELLULAR MECHANISMS?

	ABSTRACT

Top Next References

	I. INTRODUCTION

Top Previous Next References

 
Learning may be described as the mechanism by which new information about the world is acquired, and memory as the mechanism by which that knowledge is retained. It is convenient to categorize memory as being explicit, which is defined as that involved in the conscious recall of information about people, places, and things, or implicit, which is characterized by the nonconscious recall of tasks such as motor skills. Explicit memory depends on the integrity of temporal lobe and diencephalic structures such as the hippocampus, subiculum, and entorhinal cortex. Implicit memory includes simple associative forms of memory, such as classical conditioning, and nonassociative forms, such as habituation, and relies on the integrity of the cerebellum and basal ganglia (582).

Although several areas of the brain play a part in consolidation of several forms of learning/memory (Table 1), the hippocampus has been recognized as playing a vital role in formation of declarative memory in particular, which describes the synthesis of episodic and semantic memories. The observations of Scoville and Milner in 1957 (556), showing that bilateral hippocampal removal as a treatment for epilepsy suffered by patient H.M., resulted in anterograde amnesia explicitly identified the importance of the role of the hippocampus and temporal lobe structures in memory. Since then, studies in humans (e.g., Ref. 585) and animals (e.g., Refs. 427, 473) have consolidated the essential finding of that study. More recently, noninvasive methods using direct brain imaging techniques such as magnetic resonance imaging and positron emission tomography (PET) characterized blood flow and oxygen use in the hippocampus and identified fluctuations in these parameters during learning tasks (e.g., Refs. 582, 584). This review focuses principally on a discussion of synaptic plasticity in the hippocampus and only briefly discusses synaptic plasticity in other areas.

A. The Hippocampus and Spatial Memory

One of the most compelling problems in neuroscience is to identify the mechanisms underlying memory, and although a great deal of progress has been made in the past few decades, it remains a significant challenge. Particular emphasis has been placed on analysis of changes that accompany and support spatial memory because of its dependence on hippocampus and because of the well-developed protocols that are available for its analysis. A variety of paradigms are available for investigation of spatial learning, and perhaps the most commonly used is the Morris water maze in which an animal's capacity to remember spatial cues is required to locate a hidden underwater platform (426). Using this paradigm in particular, numerous studies have identified an essential role for the hippocampus in spatial learning; in addition, several studies have built on the original observation of O'Keefe which identified the involvement of specific hippocampal pyramidal cells in encoding information about the location of an animal in a particular space (471). Rats with lesions of the hippocampal and parahippocampal areas perform particularly poorly in spatial learning tasks; in the case of the Morris water maze, although lesioned and nonlesioned rats perform in a comparable manner when the platform is visible, lesioned rats perform very poorly when the platform is not visible. It appears that the key role of the hippocampus in spatial learning is synthesis of the configuration of spatial cues, which is governed, at least to some extent, by temporal events (607). Significantly other forms of learning, like visual discrimination and taste aversion, are not affected by hippocampal lesions.

A careful analysis of performance in different spatial learning tasks has led to the suggestion that the integrity of connections between the hippocampus, subiculum, and cortical areas is necessary for synthesis of all components of spatial learning. Monkeys with large bilateral lesions of the medial temporal lobe, which approximated the damage sustained by H.M., exhibited severe memory impairment on the delayed nonmatching to sample task (420, 663). Impairment was less severe when damage was confined to the hippocampus compared with additional damage to the perirhinal, entorhinal, and parahippocampal cortical regions (33, 661–663, but see Ref. 444). Some (104, 439), but not all (10, 286, 439), authors have reported similar impairments in rats, but results are dependent on the precise nature of the task and the extent of the lesion (104).

Recognition memory can be investigated using the visual paired comparison task, which assesses preference for exploring a new, compared with a familiar, object or picture. It has been reported that performance in this task is not impaired in amnesic patients with hippocampal damage, provided there was no delay between the first and second presentations of the stimuli; a deficit was observed when a delay was introduced (404). This emphasizes the temporal component of hippocampal-dependent memory referred to above. Similar impairments were observed in monkeys with lesions of the temporal lobe (35, 34) or hippocampus (484, 661). Data from the rat are less clearcut (439; but see Ref. 586), and therefore, the role of the hippocampus in recognition memory in mammals requires further elucidation.

Although much emphasis has been placed on assessing the role of hippocampus in memory formation, it is acknowledged that most areas of the cortex are probably capable of supporting various sorts of memory, for example, visual sensory memory, auditory sensory memory, and tactile memory; these are transient or temporary memories, and consolidation is required to enable formation of long-term memory. It has also been shown, using PET in human subjects, that spatial memory is associated with differential activation in area 47 of the prefrontal cortex (573) and that lesions of the posterior parietal associative cortex lead to profound impairments (16).

Working memory, i.e., the ability to maintain and use mental representation for goal-directed behavior, is dependent not only on hippocampus but also on the prefrontal cortex, which has strong connections to the hippocampus. The frontal cortex also plays a significant role in the temporal ordering of spatial and nonspatial events and the planning of responses, and the integrity of other areas of the brain has been identified as being critical for formation of specific memory forms. For instance, the acquisition of motor skills and habits and the memories associated with such skills (procedural memory) relies on the integrity of the striatum and the cerebellum, while the role of the amygdala in emotional memory has also been recognized for many years. The recognition that several areas other than hippocampus, particularly cortical areas, play such an important part in various forms of memory prompted anatomical studies, and therefore, hippocampus-neocortical connections have been studied with great interest. It has been shown that, in addition to the hippocampal-prefrontal cortical connections which are routed through the subiculum, CA1 projects directly to the medial and orbital prefrontal cortices (40). The subiculum also receives inputs from the postsubiculum and entorhinal cortex, and it appears to play a role in processing and integration of the information that it relays to other cortical areas. These connections mean that the subiculum receives positional, directional, sensory, and contextual information. It has been shown that lesions of this area lead to deficits in certain forms of learning (551). In addition to the projection to subiculum, CA1 neurons project to the perirhinal, postrhinal, and entorhinal cortices, and a number of studies have suggested that these pathways play a role in various forms of learning and memory (472). Current evidence suggests that positional information relies on hippocampal-subicular interaction, directional information on the interaction between postsubiculum and subiculum, and sensory information on the interaction between entorhinal cortex and subiculum (472; see below).

While recognizing the primary role of the hippocampus in memory formation, the interaction with cortical structures, particularly in the context of long-term storage of memories, remains an issue of debate, and it has been proposed that sequential activation of the hippocampus and neocortex may be involved in consolidation of memory. One proposal is that although the hippocampus may be largely responsible for recall of recent memories, the neocortex is primarily concerned with recall of more remote memories (582). This idea is linked with the view that the hippocampus allows for rapid learning, permitting the neocortex to undergo synaptic changes required for slow learning. One prediction emanating from the hypothesis that hippocampus and cortex are responsible for maintenance of short- and long-term storage of memory, respectively, is that recall of distant memories would be independent of the hippocampus and, consistent with this, it has been reported that remote childhood memories and general knowledge were not affected in individuals with hippocampal damage. However, data from a recent systematic study on individuals with lesions of the hippocampus, in some cases extending to the temporal cortex, revealed that the extent of the lesion (from that affecting only the CA1, CA3, and dentate gyrus to that involving the entire hippocampal complex and temporal lobe) dictated the degree of impairment in recall and, to some extent, the remoteness of the memory. The development of neuroimaging techniques has allowed further assessment of the role of the hippocampal complex in retrieval of distant memories, and the evidence suggests that activation of hippocampal circuits occurs even when very remote memories are elicited (446).

Analysis of this question in animals has revealed that sectioning the fornix, or damaging the hippocampus or entorhinal cortex, typically impaired very recent memory, but generally spared more remote memory. This suggests that the hippocampus is necessary for memory storage and retrieval for only a limited time after learning and that time-related modification of cortical connections allows for memory retrieval independent of the hippocampus (583). However, it has been pointed out that this might also be explained if representation of older memories was more diffusely distributed in hippocampus. In this case, temporally graded retrograde amnesia could be explained because a partial lesion of the hippocampus will spare a remote memory more than a recent memory, whereas complete hippocampal lesions will affect recent and remote memories equally (445).

It is appropriate to state that while great emphasis has been placed on the role of the hippocampus in spatial memory, a number of studies have identified its importance in nonspatial memory tasks. For instance, in a recent paper, using social transmission of food preference, Clark et al. (104) reported the lesions of the hippocampus and subiculum resulted in anterograde amnesia and temporally graded retrograde amnesia.

B. Synaptic Modifications and Memory

Activity-dependent synaptic plasticity plays a vital role in sculpting synaptic connections during development and has been identified in several synaptic pathways. Although it occurs, in particular, during critical periods of early development, it is also a feature of the adult brain. For example, it is widely accepted that memory formation is dependent on changes in synaptic efficiency that permit strengthening of associations between neurons; indeed, activity-dependent synaptic plasticity at appropriate synapses during memory formation is believed to be both necessary and sufficient for storage of information. Cajal (80) originally hypothesized that information storage relies on changes in strength of synaptic connections between neurons that are active. Hebb (221) supported this hypothesis and proposed that if two neurons are active at the same time, the synaptic efficiency of the appropriate synapse will be strengthened. An enormous effort has been channelled into understanding the mechanism by which strengthening of synaptic connections can be achieved and, in this effort, the importance of one model, above all others, cannot be overestimated; this model is long-term potentiation (LTP).

In 1966, Lomo (340) reported that a single, short test shock, following an initial period of conditioning test shocks to the perforant path, elicited a potentiated response in the dentate gyrus. The first full description of LTP by Bliss and Lomo in 1973 (64) reported that trains of high-frequency stimulation to the rabbit perforant path caused a sustained increase in efficiency of synaptic transmission in the granule cells of the dentate gyrus. This report, and others which followed during the 1970s, confirmed the Hebbian nature of this form of synaptic plasticity, and it was immediately recognized that the synaptic changes that underpin certain forms of learning and memory may be similar to those upon which expression of LTP relied. The three well-described characteristics of LTP, cooperativity, associativity and input specificity (see Ref. 62), and the durability of LTP (8), have been identified as solid arguments that support the hypothesis that LTP may be a biological substrate for at least some forms of memory. Several other pieces of evidence have consolidated this view. 1) LTP is most easily demonstrable in the hippocampus, an area of the brain known to be fundamentally important in memory acquisition. 2) Rhythmic bursts of activity that induce LTP mimic naturally occurring theta rhythm recorded in the hippocampus during exploratory behavior (132, 208, 313, 527). 3) Inhibitors of hippocampal LTP also block hippocampal learning and retention of tasks (425). 4) Several biochemical changes that occur after induction of LTP also occur during memory acquisition (see below). However, a definitive demonstration indicating that memory consolidation requires induction of changes that resemble those necessary for induction of LTP remains elusive. Similarly, it remains to be clearly shown that induction of LTP will result in some form of memory consolidation.

At least two components of memory can be discerned: short-term memory, which endures for a few hours, and long-term memory, which persists for several days and often much longer. At the cellular level, the storage of long-term memory is associated with gene expression, de novo protein synthesis, and formation of new synaptic connections. Consistently, protein synthesis inhibitors can block persistent memory but leave short-term memory unaffected, suggesting that stable, long-lasting memories rely on gene activation that is triggered at, or close to, the time of the experience. Here, there is an interesting parallel between memory and LTP, since it has been revealed that LTP consists of distinct phases involving different molecular mechanisms. The early phase (E-LTP), which lasts 2–3 h, is independent of protein synthesis, while more persistent long-lasting LTP (L-LTP), which lasts several hours in vitro and weeks in vivo, requires synthesis of new proteins.

A series of fundamentally important findings made in the early 1980s profoundly affected the course of research in LTP and which, for the first time, provided some insight into the mechanisms by which LTP consolidation occurred. The first of these was the observation that LTP in CA1 was inhibited by the N-methyl-D-aspartate (NMDA) antagonist 2-amino 5-phosphonopentanoic acid (AP5) (107), and this, combined with the important discovery that NMDA receptor activation led to influx of calcium through a ligand- and voltage-sensitive calcium channel (27), triggered significant advances in understanding the cellular cascades initiated as a result of tetanic stimulation. It was later established that the majority of synapses which support LTP, in hippocampus and elsewhere, do so in an NMDA receptor-dependent fashion (see below), but that while the resultant increase in postsynaptic calcium concentration was both necessary and sufficient for expression of LTP, NMDA receptor activation, although required in many cases, was not sufficient to result in its induction (62).

Petersen et al. (493) addressed the question of whether LTP at individual synapses was induced in an incremental manner or in an all-or-none manner. In this clever series of experiments, a pairing protocol was used which resulted in a small, nonsaturating amount of potentiation combined with the use of minimal stimulation techniques to activate single fibers. The data indicated that individual synapses had different thresholds, but that once threshold was achieved, the degree of potentiation did not vary, indicating that synapses responded in an all-or-none manner; potentiation was dependent on NMDA receptor activation. It was also shown that, although synapses exhibited a variation in the delay to potentiation, once initiated the response was rapid. These data led the authors to conclude that reaching threshold, in circumstances in which background noise is considerable, requires coincident priming of pre- and postsynaptic elements and suggested that this and the rapid onset of response might be explained by autophosphorylation of calcium/calmodulin kinase II (CaMKII) and the consequent insertion into the membrane of AMPA receptors (see below).

	II. SEVERAL AFFERENT PATHWAYS SUPPORT LONG-TERM POTENTIATION

Top Previous Next References

 
In addition to the principal afferent pathways in the hippocampus, several other afferent pathways have been shown to sustain LTP (see Table 2). One that has been of great significance in promoting the idea that synaptic changes which underlie LTP may also underlie memory is the projection from the thalamus to the amygdala.

A. The Amygdala

A great deal of evidence indicates that fear conditioning, which is a robust form of classical conditioning exhibited by rodents, is amygdala dependent; specifically, neuronal changes mediating the association between the conditioned and unconditioned stimuli occur in the lateral nucleus of the amygdala. Consistently, lesions of the amygdala have also been shown to result in deficits in fear conditioning (114, 200), while phthalic acid lesions of the nucleus basilis magnocellularis, from which there is a dense cholinergic projection to the basolateral amygdala, have been shown to lead to a profound deficit in inhibitory avoidance behavior (499). In humans, as well as animals, activation of the amygdala has been shown to be closely correlated with memory for both aversive and pleasant stimuli (213).

A number of recent findings have led to the suggestion that the amygdala is not a critical long-term information storage site but that its role is to regulate memory consolidation in other brain regions (401, 402). For instance, if the amygdala is lesioned after training, fear-motivated learning is partially retained while certain pharmacological agents when administered following training have been shown to modulate learning (641). It has been concluded that the amygdala is the locus of control for Pavlovian fear conditioning while its role in inhibitory avoidance is to modulate activity of other brain areas (641).

Compelling evidence supporting the hypothesis that LTP represents a valid model for learning/memory has proven to be an elusive goal, but recent analysis in the amygdala has been of major significance. The amygdala is the point of convergence of information from conditioned and unconditioned stimuli and, when the conditioned stimulus is an audible tone, the information is carried to the lateral amygdala via the afferent input from the auditory thalamus; this connection can express LTP (105). Pairing this conditioned stimulus with foot shock (the unconditioned stimulus) increases the response of amygdalar cells to auditory stimulation, and this was shown to be coincident with the animals exhibiting freezing behavior. This enhanced response of the cells was persistent and did not occur when the stimuli were unpaired (522). Predictably neuronal activity in the lateral amygdala was enhanced by the conditioned stimulus, and this preceded behavioral responses (514); more recently, it was shown that drugs that interfere with LTP in these pathways disrupt behavioral fear conditioning (61). Thus, with respect to at least one form of memory, a role for LTP has been identified, and it is important to point out that LTP in the amygdala shares several features with LTP in hippocampus. For instance, it has been shown that it is dependent on NMDA receptor activation, that retrieval of fear memories requires protein synthesis (447), and that activation of the transcription factor cAMP response element binding protein (CREB) is a key element in consolidation of memory, including fear memory (see sect. VD).

In an effort to explain the persistent nature of conditioning, it was proposed that the conditioned stimulus evokes excitatory postsynaptic potentials (EPSP) at sensory input synapses onto pyramidal neurons of the lateral amygdala and that this coincides with depolarization of the same neurons by the unconditioned stimulus. As a consequence of the depolarization, calcium influx through NMDA receptor-associated channels occurs, amygdalar neurons fire action potentials which back-propagate into the dendrites, and this coincident activity, together with the EPSPs generated by the conditioned stimulus, leads to calcium entry through voltage-gated calcium channels. It has been proposed that the increase in intracellular calcium, consequent upon NMDA receptor-associated calcium channel opening, underlies short-term fear memory while the additional calcium entry through voltage-dependent calcium channels is required for long-term memory (61).

Both the hippocampus and entorhinal cortex receive direct projections from basolateral amygdala (497), and therefore, the recent reports indicating a modulatory role of the amygdala on hippocampal LTP have not been surprising. Activation of the basolateral amygdala has been shown to enhance LTP in dentate gyrus (13, 245), but lesions result in impaired LTP (244). Significantly, activation of the basolateral nucleus of the amygdala (within a specific time window) has the capability of transforming short-term potentiation in dentate gyrus of freely moving rats into protein synthesis-dependent persistent LTP. The authors found that this effect was independent of direct activation of glutamatergic inputs and proposed that the convergence of the action of a modulating transmitter as a consequence of amygdalar stimulation and glutamatergic activation following perforant path stimulation was necessary for consolidation of persistent LTP (167). In parallel with its modulatory effect on LTP, amygdalar activation has also been shown to enhance hippocampal-dependent learning (214, 481, 524); however, LTP (and short-term potentiation) in CA1 has been shown to be reduced in slices prepared from rats that were previously exposed to contextual fear conditioning (535).

Emerging evidence has indicated that an intact basolateral amygdala underpins stress-induced modulation of hippocampal LTP (290). Thus lesioning studies have revealed that the inhibitory effect of stress on LTP is suppressed in rats following electrolytic lesions of the amygdala (290), while the poorer performance in spatial learning, which is induced by adrenalectomy (523-525) or stress (290), is dependent on an intact amygdala. Similarly, electrolytic lesions of the amygdala abrogated the behavioral effect of stress induced by restraint and tail shock (290). There is evidence that this modulatory role of the amygdala involves activation of the projection of the stria terminalis to the nucleus accumbens (523).

While the auditory thalamus projects to the lateral amygdala, the mediodorsal thalamic nucleus projects to the medial prefrontal cortex (mPFC); the latter pathway has been shown to support both LTP and long-term depression (LTD), and evidence favors the idea that learned fear may be dependent on plastic changes at these synapses. Thus it appears that extinction of learned fear is associated with LTP in mPFC, while persistence of LTD during extinction is coupled with a return of the learned fear behavior (227).

B. The Visual Cortex and the Somatosensory Cortex

It has been proposed that the mechanisms which underlie LTP and LTD in visual cortex and somatosensory cortex play a contributory role in experience-dependent synaptic plasticity. In the case of the visual cortex, experience-dependent acquisition of visual responsiveness during the critical period requires significant modification of synaptic connections. Synaptic modification depends on neuronal activity, and like LTP in visual cortex and elsewhere (24), threshold stimulation in somatosensory cortex is required to permit synaptic changes. A pivotal role for NMDA receptor activation in LTP induction has been described in both areas, and the profound modifications in synaptic plasticity that occur in early life have been attributed to NMDA receptor subunit expression which alters with maturity (50, 51, 162, 619). Significantly, it has been shown that NMDA-sensitive LTP can be elicited in the adult rat visual cortex in vivo by stimulation of the dorsal lateral geniculate nucleus and that the cortical response to visual stimuli is enhanced after LTP (229). The physiological consequences of the enduring nature of this form of plasticity remain to be established, but if it is the case that experience-dependent plasticity and LTP share common mechanisms, then deficits which occur as a result of visual deprivation during the critical period may be reversible in adulthood.

Development of the barrel cortex is a striking example of synaptic plasticity that is dependent on experience-dependent changes in thalamocortical circuits. Layer IV of the rat somatosensory cortex has a topographic map representing peripheral receptor density, formation of which is exquisitely sensitive to receptor stimulation particularly during development. The thalamocortical synapses of layer IV cortical cells support LTP (156), although it is not clear whether LTP is involved in the plasticity required for formation of topographic maps. However, like use-dependent synaptic modification and LTP in visual cortex, LTP in the developing barrel cortex requires NMDA receptor activation (156), whereas inhibition of the NMDA receptor by AP5 blocks the functional changes associated with mystacial whisker ablation in the neonate (552). The role of the NMDA receptor has been further underlined by the observation that formation of cortical barrels is prevented in the absence of the NR1 subunit in cortical neurons (253), although expression of the NR2 subunit appears to be without effect on plasticity in either visual cortex (495) or barrel cortex (344).

C. The Prefrontal Cortex

Training in an associative learning task was found to be accompanied by enhanced synaptic transmission in hippocampal-prefrontal cortical synapses (141); while early changes in synaptic transmission in hippocampal synapses were recorded, changes in prefrontal cortex were delayed. This is consistent with the idea that the hippocampus plays a special role in rapid learning and acts in concert with the cortex to ensure stabilization of a cortical representation of learned events. Restricted lesions of the prelimbic area of the prefrontal cortex suggest that this area is critically involved in working memory (see Ref. 311).

Consolidation of the hypothesis that the cellular mechanisms underlying LTP are necessary for memory formation would be assisted if it could be shown that pathways that are activated during memory formation sustain LTP. The importance of hippocampal-prefrontal cortex communication in cognition has been recognized for many years (312, 314, 582), and one of the first descriptions of LTP outside the hippocampus was made in the hippocampal input to the prelimbic cortex in vivo (141). LTP in this pathway has been reported several times since the initial report (e.g., Ref. 257), and it has been demonstrated that it requires NMDA receptor activation (257). Like LTP in hippocampus, there is emerging evidence that activation of protein kinase A (PKA) at these synapses leads to CREB activation (see Ref. 311).

In addition to the hippocampal-prefrontal cortical pathway, other cortical pathways have also been shown to support LTP. For instance, LTP can be induced in layer V following stimulation of layer I (231) or layer II (631) and in layer IV following stimulation in layer II/III (215). Although NMDA activation is necessary for expression of LTP in hippocampal-cortical synapses, characterization of the mechanisms underlying LTP in cortico-cortical pathways remain to be clarified. For instance, some authors have reported that NMDA receptor activation is required, whereas others have disputed this (215, 226).

D. The Subiculum

In addition to the direct projections (40), some hippocampal-prefrontal cortical connections are routed through the subiculum (20), an area of the brain which plays a role in processing and integration of information that it relays to other cortical areas. Thus the circuits that connect the subiculum to the presubiculum, the perirhinal cortex, the entorhinal cortex, and the prefrontal cortex have been identified as significant in particular forms of memory and learning, for example, instrumental learning (36), working memory (177), avoidance learning (178), and visual, tactile, and spatial memory (427, 595, 662). Significantly, several of these subicular pathways have been shown to support LTP (110, 296). Perhaps the best characterized is the CA1 to subiculum projection which exhibits paired pulse facilitation and LTP in vivo (110, 472) and in vitro (296). Analysis of subicular unit firing in a pellet-chasing task revealed the existence of place cells in the subiculum, although the firing fields were less discrete than those described in CA1 (471, 560). The expression of synaptic plasticity and the evidence of a representation of space in the subiculum, together with the observations that lesions of the area lead to deficits in spatial learning (427), indicate that, like the hippocampus, the subiculum appears to play a significant role in learning and memory.

	III. MECHANISMS UNDERLYING LONG-TERM POTENTIATION

Top Previous Next References

 
A. NMDA Receptor Activation and LTP

The critical event leading to induction of LTP appears to be the influx of calcium ions into the postsynaptic spine and therefore, predictably, LTP is blocked by injection of EGTA (352b) or BAPTA (433) and induction occurs when the postsynaptic cell is loaded with calcium (370). Therefore, it is agreed that elevation of postsynaptic calcium concentration is both necessary and sufficient for the induction of hippocampal LTP (62).

In the majority of synapses that support LTP (in the hippocampus and elsewhere), the postsynaptic increase in calcium is mediated through activation of the NMDA receptor. Several experimental approaches have been used to consolidate the initial evidence which supported this contention, and some of these are listed in Table 3. Significantly, the characteristics of NMDA receptor activation eloquently explain the properties of LTP: receptor activation leads to opening of the associated calcium channel when occupied by glutamate and when the postsynaptic membrane is depolarized. Therefore, the NMDA receptor complex is dually regulated by ligand and voltage and thereby acts as a coincidence detector. Consistent with its pivotal role in LTP induction are numerous demonstrations that inhibition of NMDA receptor activation blocks LTP. The first of these demonstrations in CA1 in vitro and dentate gyrus in vivo, using the specific competitive NMDA receptor antagonist AP5 and the noncompetitive NMDA-associated channel blocker MK801 (106, 107, 153), was followed by several confirmatory reports. With the exception of mossy fiber-CA3 synapses, induction of LTP in all subfields of the hippocampus is NMDA dependent, although it has been shown that LTP in CA1 can be induced without the participation of NMDA receptors; in this case, the increase in postsynaptic calcium concentration is a consequence of activation of voltage-operated calcium channels, and therefore, calcium channel inhibitors suppress this form of LTP (209).

NMDA activation alone does not induce LTP (280). This observation, together with the demonstration that thapsigargin, which depletes intracellular calcium stores, inhibits LTP (72), suggests that calcium release from intracellular stores augments NMDA receptor-mediated calcium influx. Activation of the NMDA receptor may be critical for induction of many forms of LTP, but it is not necessary for all. In contrast, current evidence is consistent with the hypothesis that a rise in intracellular calcium concentration is a necessary element for the induction of all forms of LTP described to date.

In parallel with the importance of NMDA receptor activation in induction of LTP in hippocampus, it has been repeatedly shown that AP5 markedly attenuates performance in spatial learning tasks (e.g., Ref. 425), although it is now clear that previous exposure to similar tasks alters sensitivity to these inhibitors (see sect. IIIB). Activation of NMDA receptors seems to be necessary not only for acquisition of spatial information, but also for memory retention (240).

The requirement for NMDA receptor activation is not confined to plasticity in the hippocampus, since receptor blockade leads to a deficit in long- and short-term memory of fear conditioning (521, 634). Similarly, NMDA receptor activation is necessary for induction of LTP in amygdala, although LTP in amygdalar interneurons is NMDA independent (376). In addition to the amygdala, experience-dependent synaptic modifications and LTP in visual cortex (50, 51, 162, 229, 619) and frontal cortex (257) rely on activation of NMDA receptors.

Analysis of the subunit composition of the NMDA receptor has revealed differential expression of NR1 and NR2 with brain area, development, and activity (557, 637), and gene targeting has allowed examination of some of the physiological roles of the different subunits. Both hippocampal LTP and spatial learning rely on expression of NR2A, since disruption of this subunit is associated with deficits in both (293, 539), while deletion of the gene encoding NR2B also resulted in impairment of LTP in hippocampus as well as impairment in development of the barrel organ in the trigeminal complex (302, 321). Similarly, mutant mice lacking NR2A exhibit normal responses in tone-dependent fear response (i.e., a hippocampal-independent learned response) but exhibited deficits in contextual fear learning (a hippocampal-dependent response; Ref. 293); this finding discriminates between two forms of fear learning on the basis of their dependence on hippocampal function. In addition to the effects of disruption of NR2 subunits, genetic disruption of the NR1 subunit also leads to impairments in LTP and spatial learning (617, 618). Conversely, overexpression of the NR2B subunit was found to be associated with enhanced LTP and enhanced learning and memory (604). Analysis of the dynamics of different NMDA receptor subunits has revealed that visual experience results in insertion of new receptors with a higher proportion of NR2A subunits, resulting in an increase in the ratio between NR2A and NR2B (505). One consequence of this is that NMDA receptor-associated currents are shortened and, therefore, conditions will favor induction of LTD rather than LTP; this is consistent with the idea that an LTD-LTP modification threshold monitors plasticity and that this threshold alters with maturity (49).

B. NMDA Receptor Activation, Learning, and Memory

A great deal of evidence indicates that NMDA receptor activation plays an essential role in the acquisition of spatial memories. The first data that addressed this question were reported in 1986, when Morris et al. (425) found that blocking the NMDA receptor with AP5 inhibited spatial learning. The specific importance of the finding at that time lay in the fact that this agent also inhibited LTP, and therefore, this suggested an overlap in the mechanisms by which LTP was sustained and by which spatial learning was consolidated. Others, using genetically manipulated mice, arrived at essentially the same conclusion; for example, Tsien et al. (618) generated a mouse in which NMDA receptor was knocked out in CA1 and they reported that these mice exhibited impaired spatial memory, while nonspatial memory was intact, and this was coupled with a deficit in LTP. Similarly, both spatial learning and LTP were impaired in mutant mice that lacked the NR2A (1) subunit (539). Conversely, overexpression of the NR2B subunit yielded mice with enhanced LTP and enhanced learning and memory (604, 605). These genetic correlations between LTP and some forms of learning and memory therefore support the initial findings of Morris et al. (425). However, more recent studies have revealed that inhibition of the NMDA receptor only impairs spatial learning in task-naive animals, whereas pretraining in a spatial task overcomes the inhibition induced by AP5 (39) or another potent and specific NMDA antagonist, NPC17742 (545), even when LTP in dentate gyrus was inhibited. Interestingly, Bannerman et al. (39) reported that pretraining in a nonspatial task induced a similar effect. It was subsequently shown that pretraining in a spatial learning task prevents disruption of a subsequent training session in spatial learning after saturation of LTP (478). It was concluded that all the components of spatial learning (at least in the Morris water maze) do not require NMDA receptor activation. In terms of the question of uncoupling of spatial learning and LTP, it was proposed that spatial learning can take place in the absence of LTP provided episodic aspects of the training context are familiar.

C. Metabotropic Glutamate Receptors and LTP

The first indication of a possible role for metabotropic glutamate receptors in LTP was in 1991 with the observation that the nonselective mGluR agonist 1-amino-1,3-cyclopentanedicarboxylic acid (ACPD) enhanced LTP (403); these findings were subsequently replicated by other groups (375, 512, 513). ACPD was later shown to induce a long-lasting potentiation of the synaptic response in CA1 (71, 72, 100, 374) and in the dentate gyrus (464), and the effect was shown to rely on calcium-dependent changes and on activation of protein kinase C (PKC), since it was prevented by thapsigargin and staurosporine (73). Although it has been reported that mGluR inhibition blocks LTP, this is not a consistent finding (see Ref. 23), and mutant mice have been generated in an effort to identify the precise nature of the dependency of LTP on mGluR activation. One study reported that LTP in CA1 of these mice was unimpaired (111), but another reported that it was blocked (11). Mutant mice lacking mGluR5 have been reported to show attenuated LTP induction in CA1 and dentate gyrus, but LTP in mossy fiber-CA3 synapses was spared, leading the authors to suggest that the modulatory effect of mGluR activation on LTP differed in NMDA-dependent and NMDA-independent pathways (348). It was subsequently shown that potentiation of the NMDA response was absent in mGluR5 mutant mice but that potentiation of the AMPA response was preserved (262); these findings led the authors to conclude that activation of mGluR5 plays a pivotal role in expression of NMDA receptor-dependent LTP. The impaired potentiation of the NMDA receptor-associated response in mGluR5 mutant mice has been identified as being PKC linked, since it could be overcome by activation of PKC (see Ref. 263). One proposed mechanism by which this effect occurs involves PKC-induced activation of src, which increases NMDA receptor-associated channel opening (346), although an alternative substrate for PKC may be homer, which couples mGluR5 to PSD 95 by formation of a cluster with another postsynaptic density protein, Shank (620). LTP has also been assessed in mutant mice lacking mGluR1, and there is certain confusion with respect to mossy fiber-CA3 LTP; one group reported that LTP was absent in mutant mice lacking mGluR1 (111), but this was not supported by the findings of a second group (234). Thus, although mGluR activation appears to contribute to expression of LTP (see Table 3), clarification of the roles of the different receptor groups is necessary (for example, see Refs. 111, 234, and 382).

Several groups have shown that spatial learning is dependent on mGluR activation (46, 70, 111, 348, 517), and inhibitory avoidance and contextual fear learning have also been shown to be dependent on receptor activation (11, 60, 103, 456). Specifically, the mGluR1 antagonist -methyl-4-carboxyphenylglycine (MCPG) reduced spatial learning, whereas a class I agonist trans-azetidine-2,4-dicarboxylic acid (tADA) applied after learning facilitated memory recall (171, 510, 611). Consistently, 4-carboxyphenylglycine (4-CPG), a more selective class 1 antagonist, which blocked LTP, also inhibited learning and memory in some tasks (37). Consistent with these findings is the observation that mGluR5 mutant mice exhibited an impairment in spatial learning, as assessed by performance in the Morris water maze, and also in contextual fear conditioning; both forms of learning are dependent on an intact hippocampus (263). Interestingly, it was also reported that there was a persistent increase in mGluR5 expression after fear conditioning (511). This role for mGluR5 in fear conditioning is consistent with the earlier finding that 1-aminoindan-1,5-dicarboxylic acid (AIDA) resulted in a retention deficit in conditioning to the context but not the cue (103). In contrast to the change in mGluR5 expression, Reidel et al. (511) reported that there was no change in mGluR1 expression after fear conditioning, although an increase in mGluR1 mRNA has been observed after induction of LTP in dentate gyrus (611).

	IV. WHAT SIGNALING EVENTS FOLLOW N-METHYL-D-ASPARTATE RECEPTOR ACTIVATION?

Top Previous Next References

 
A. A Role for CaMKII in LTP

When it was established that increased calcium concentration in the postsynaptic cell, as a consequence of NMDA receptor activation, was a critical factor in the induction of LTP, attention turned to analysis of the downstream cellular consequences of this increase. Among the early findings was that postsynaptic entry of calcium led to activation of CaMKII; this observation turned out to be a finding of major importance. CaMKII is one of the most abundant proteins in neurons comprising 1–2% of the total. Although it is expressed presynaptically and postsynaptically, its expression is particularly high in the postsynaptic density, where it is ideally located to respond to changes in calcium concentration. There are more than 30 isoforms of CaMKII and numerous substrates, many of which are located in the postsynaptic density (see Ref. 159). CaMKII appears likely to be a mediator of primary importance in linking transient calcium signals to neuronal plasticity.

In 1989, two groups reported the important finding that inhibitors of CaMKII blocked LTP in CA1 (369, 373). Since that time, the requirement for CaMKII activation in expression of LTP has been confirmed many times using many different approaches. Thus it has been shown that CaMKII activation is triggered by the LTP-induced NMDA receptor-driven increase in intracellular calcium and that activation of the kinase persists after induction of LTP (176, 479). This persistent activation of CaMKII occurs as a result of autophosphorylation at Thr-286, and it has recently been shown that if the kinase is mutated at this residue (by replacement of threonine with alanine), then autophosphorylation is prevented; in these circumstances, both LTP and spatial learning are impaired (194). These findings and a range of related findings, specifically those obtained from analysis of changes in transgenic mice, provide convincing evidence that CaMKII activation is necessary for expression of LTP. Significant among these reports are the observations by Silva and colleagues (570, 572), who demonstrated that deletion of the CaMKII gene in mice resulted in impairment in LTP and also impairment in spatial memory. Similarly, introduction of an activated calcium-independent form of CaMKII into CA1 neurons potentiated synaptic transmission and occluded LTP (219, 337, 494). However, LTP in CA1 in transgenic mice expressing a constitutively active calcium-independent mutant form of CaMKII (-CaMKII^T286A mice) was similar to that in wild-type mice (388, 389), a surprising result if it is argued that CaMKII alone is sufficient to induce LTP. It was established that, in these mice, the threshold to induce LTP was increased, while the threshold to induce LTD was reduced, suggesting that the extent of activation of CaMKII was important in modulating the response to stimulation. These findings were elaborated upon and clarified in a recent study. Using transgenic mice in which the level of transgene expression was regulated, Bejar et al. (53) reported that expression of the transgene was associated with significant impairments in contextual fear conditioning and in spatial memory. The authors also reported that the level of expression of the transgene was important; significantly, LTP, induced by 5-Hz stimulation, was enhanced in mice expressing low levels of the transgene but was markedly depressed in mice expressing high levels of the transgene (53). The idea that the degree of stimulation of CaMKII may have a modulatory effect on plasticity has also been addressed in another study. In this case it was proposed that CaMKII may act as a memory molecule. Thus a strong stimulus can prime CaMKII so that subsequent stimuli can lead to greater association with the postsynaptic density (562).

The requirement for CaMKII activation in expression of LTP is therefore generally accepted, and there is strong evidence suggesting that activation of CaMKII is sufficient to induce LTP. Evidence favoring the view includes the demonstration that injection of a constituitively active form of CaMKII induces LTP (337, 494). A similar conclusion was drawn in experiments that used occlusion to address the question; thus cells that exhibited synaptic potentiation induced by CaMKII were insensitive to tetanic stimulation, and vice versa (335). Perhaps the most powerful arguments used to support the view that CaMKII is sufficient to induce LTP have been developed on the back of the "silent synapse" theory of LTP (see below). The increase in responsiveness to applied glutamate following LTP, which is due largely to increased AMPA conductance, is a consequence of CaMKII-induced phosphorylation of GluR1 on Ser-831 (125), and it has been proposed that this contributes to the LTP-associated increase in conductance (120). Of importance is the observation that the increase in GluR1 phosphorylation that accompanies LTP is inhibited by CaMKII antagonists. In addition, there is now compelling evidence to suggest that delivery of AMPA receptors to the spine after induction of LTP, allowing the transition from silent to nonsilent receptor, is closely linked with, and may even be dependent on CaMKII activation (324, 564, 565; see below and see Fig. 1).

View larger version (20K): [in this window] [in a new window]   FIG. 1. Among the consequence of the increase in intracellular Ca2+ concentration ([Ca2+]i) which accompanies N-methyl-D-aspartate receptor (NMDA-R) activation is increased calmodulin kinase II (CaMKII) activity which exerts multiple actions. One significant effect is increased AMPA conductance as a result of AMPA receptor (AMPA-R) phosphorylation and increased recycling of AMPA-R, which is due to CaMKII-induced changes in cytoskeletal proteins (see text for details). NSF, N-ethylmaleimide-sensitive factor.

In addition to the pivotal role for CaMKII in LTP, evidence has been emerging which suggests that CaMKIV may also play a role. CaMKIV, which is localized predominantly in neuronal nuclei, modulates CREB-regulated gene expression during LTP in CA1. Its activity is transiently increased after induction of LTP and is accompanied by increased phosphorylation of the transcription factor, CREB, and increased expression of the immediate early gene c-fos (275). Transgenic mice, in which the expression of a dominant-negative form of CaMKIV was restricted to the postnatal forebrain, exhibited a deficit in L-LTP but not E-LTP. In parallel with the impairment in LTP, CREB phosphorylation and c-fos expression were significantly attenuated in these mice (274).

Identification of the cellular events leading to activation of CaMKII after NMDA receptor activation has been a subject of intense interest. CaMKII binds to several postsynaptic density proteins including -actinin and PSD95 and the synaptic adhesion molecule, densin-180. CaMKII activation has also been shown to lead to phosphorylation of microtubule-associated protein 2 (MAP2) and neurofilament L, both of which play a role in cytoskeletal regulation; on the basis of these findings, it has been proposed that CaMKII activation may contribute to the morphological changes that accompany the more persistent components of LTP (see below). However, it is important to recognize that there are several CaMKII substrates on the presynaptic side of the synapse; these include synapsin, synaptotagmin, and synaptophysin, which play a role in neurotransmitter release. The significance of this is that LTP, at least in dentate gyrus, has been coupled with enhanced transmitter release (see Refs. 62, 353). On the postsynaptic side of the synapse, current evidence suggests that NMDA receptor activation leads to translocation of CaMKII from dendrites, where it is associated with actin, to the postsynaptic density; it has been proposed that this requires calcium but not autophosphorylation (159). It seems that binding to the NR2B subunit occurs and that this leads to increased association of calmodulin with CaMKII and, subsequently, an increased association with the postsynaptic density (48).

B. A Role for CaMKII in Learning/Memory

In parallel with the finding that CaMKII activation is necessary for induction of LTP, several studies have indicated that activation of CaMKII is required for consolidation of various forms of memory. For instance, heterozygote mice in which CaMKII expression was reduced by 50% exhibited normal hippocampal-dependent memory initially after training, but it was reported that memory 10–50 days after training was significantly impaired (164). Similarly, spatial learning has been shown to be impaired in -CaMKII^T286A mice (194). Interestingly, hippocampal place cells are unstable in -CaMKII^T286A mice, and it has been proposed that this may significantly impact on spatial learning (101, 335). The situation with respect to a role for CaMKIV in memory requires clarification. One report has indicated that the deficit in LTP in CaMKIV transgenic mice was correlated with an impairment in long-term memory (i.e., consolidation/retention rather than acquisition; Ref. 274), but it has also been reported that targeted gene disruption of CaMKIV, which resulted in impaired LTP in CA1 coupled with impaired CREB phosphorylation, was not associated with any evidence of deficits in spatial learning or memory (232).

The dependence of plasticity in the somatosensory cortex on CaMKII activation has been assessed by deletion of CaMKII and by assessing changes in -CaMKII^T286A mice. Both sets of animals failed to exhibit plasticity, and both were incapable of sustaining LTP (164, 196). These coupled findings suggest that this form of plasticity may be reliant on molecular changes that contribute to maintenance of LTP. A similar argument has been advanced with respect to plasticity in the visual cortex. In this case also, experience-dependent synaptic plasticity is markedly attenuated in -CaMKII^T286A mice (597). Consistent with the idea that CaMKII activation plays a role in fear conditioning, as it does in spatial learning, are the observations that transgenic animals with regulated expression or deficits of CaMKII demonstrate impairments in fear conditioning (571, 638), while fear conditioning is blocked by the CaMKII inhibitor KN62 (648). CaMKII activation also appears to play a role in other forms of learning. Thus intrahippocampal infusion of the CaMKII inhibitor KN62 caused full retrograde amnesia of inhibitory avoidance learning in rats, when given immediately, but not three or more hours after training (600), while activity of CaMKII was increased after training rats in a one-trial inhibitory avoidance task (see Ref. 254). Interestingly, an increase in the phosphorylation state of CREB in the hippocampus has been described after inhibitory avoidance training (57), and this finding, together with the observation that inhibition of CaMKII by KN62 prevents induction of the immediate early genes zif268 and c-fos (577), provides an insight into mechanisms underlying the enduring nature of CaMKII-dependent learning.

The data obtained from the study of synaptic plasticity in CaMKII knockout mice has, to a large extent, been paralleled by data in other organisms; the clear message is that several models expressing various forms of synaptic plasticity exhibit a requirement for CaMKII activation. For instance, CaMKII knockout in Drosophila exhibits impaired associative learning, but motor and sensory systems remain unaffected (265). Similarly, knockout of unc43 (the CaMKII analog in Caenorhabditis elegans) affects the stability of synapses and general neuronal behavior, ultimately affecting function of olfactory neurons (536).

C. AMPA Receptors and LTP

The importance of AMPA receptors in fast excitatory synaptic transmission has been acknowledged for decades, and because of this, it has been recognized that modulation of AMPA receptor activity could significantly contribute to expression of LTP. The production of mutant mice expressing different receptor subunits provided some insight into the role of AMPA receptors, particularly in relation to control of calcium fluxes. Calcium entry is modulated by the GluR2 subunit of the AMPA receptor; specifically, high expression of GluR2 mRNA has been correlated with low calcium entry (186). Predictably, AMPA receptors assembled from GluR2 subunits, in contrast to those assembled from GluR1, GluR3, or GluR4 subunits, are impermeable to calcium ions (186). Thus AMPA receptor-associated calcium permeability is low in pyramidal and granule cells of the hippocampus where there is a relatively high expression of GluR2-containing AMPA receptors. LTP was found to be enhanced in GluR2 mutant mice (261), whereas LTP was markedly attenuated in mice lacking the GluR1 subunit (423); specifically, LTP in dentate gyrus was recorded, albeit attenuated, in GluR1 knockout mice, but it could not be induced in area CA1 (659).

A great deal of evidence has suggested that increased expression of AMPA receptors on the postsynaptic membrane is likely to be the primary requirement leading to expression of LTP. The initial finding suggesting that postsynaptic glutamate receptor expression might be modulated after induction of LTP came from analysis of LTP-associated changes in sensitivity of CA1 neurons to ionophoretically applied glutamate receptor ligands. The data indicated a slow increase with time after LTP induction (120), suggesting that LTP increased the sensitivity, or the number, of receptors. Subsequent evidence revealed that increased receptor number was responsible for this finding. The primary work leading to the development of the so-called silent synapse theory of LTP was initiated with the recognition that certain synapses were functionally silent because of a lack of AMPA receptors, although NMDA receptors were present (252, 323). Thus when single connections between CA3 axons and CA1 pyramidal cells were assessed, only NMDA receptor-generated excitatory postsynaptic currents (EPSCs) could be elicited in a proportion of CA1 pyramidal cells; however, stimulus paradigms that induced LTP resulted in the recruitment of AMPA receptor-generated responses (252, 323). This was interpreted as evidence that AMPA receptors were inserted into the postsynaptic membrane after induction of LTP. Since then, a great deal of evidence has been accumulated indicating that AMPA receptor expression on cells is a dynamic process and is controlled by a cycle of exocytosis and endocytosis (351, 372). It has also been repeatedly shown in cultured cells that this cycle is modulated by NMDA receptor activation which leads to increased recruitment of AMPA receptors and increased AMPA-mediated miniature EPSPs (324, 345, 565). Lu et al. (345) reported that the punctate expression of GluR1 which colocalized with synaptophysin was consistent with synaptic localization of the AMPA receptors. They further reported that activity-dependent expression of these receptors was blocked by NMDA receptor inhibition, by sequestering calcium in the cells using BAPTA or by application of tetanus toxin which inhibits exocytosis by cleaving vesicle-associated membrane protein (VAMP), the SNARE protein which is necessary for exocytosis.

It appears that there is a fairly constant turnover of AMPA receptors containing GluR2/3 subunits at the synapse and that delivery is dependent on their interaction with a number of cytoskeletal proteins including N-ethylmaleimide-sensitive factor (NSF; Refs. 458, 579). In contrast, GluR1-containing AMPA receptors are inserted into dendritic spines in an activity-dependent manner, and this is associated with CaMKII activation and requires an interaction with a different family of postsynaptic density proteins (type II PDZ domain proteins; Refs. 496, 564). Indeed, it has been reported that activation of CaMKII drives synaptic incorporation of GluR1 subunits (see Ref. 578); however, it appears that the substrate protein is the PDZ domain protein that complexes with GluR1 to allow membrane insertion (220; see Fig. 1). In a recent study, it was shown that brain-derived neurotrophic factor (BDNF) may induce the accumulation of AMPA receptors at synapses previously devoid of these receptors, as has been proposed for silent synapses (406). Interestingly, while LTP has been linked to an increase in redistribution of AMPA receptor leading to increased membrane expression, there is also some evidence that NMDA receptor-dependent LTD is associated with a decrease in the proportion of surface-expressed synaptically localized GluR1 subunits (88).

Several groups have reported that induction of LTP is dependent on development; neonatal rats, up to postnatal day 8, are incapable of sustaining LTP, but this inability disappears shortly afterwards (47, 78, 147). In an interesting parallel it has been established that silent synapses demonstrate a developmental profile; thus the number of synapses expressing NMDA, but not AMPA, receptors in the hippocampus decreases with age (325).

Evidence for conversion of silent to functional synapses in plasticity associated with development has been accumulating, and the presence of silent synapses in developing cerebellar granule cells, which disappear as development progresses, has been documented recently (343). Similarly, during the so-called critical period, many thalamocortical synapses exhibit NMDA-, but no detectable AMPA-generated, receptor currents, suggesting the presence of functionally silent receptors. It has been proposed that LTP may lead to conversion of thalamocortical synapses from silent to functional and that this is fundamental in modulation of experience-dependent changes in thalamocortical circuits (156, 251). Silent synapses have also been identified in immature pyramidal neurons in the neonatal rat visual cortex, and these decrease during early postnatal development; the evidence suggests that, in this area also, the conversion from silent to functional synapses is dependent on NMDA receptor activation and occurs as a consequence of LTP-like activity (533).

D. Silent Synapses, Learning, and Memory

The initial findings that indicated an increase in AMPA receptors sensitivity after LTP induction sparked some parallel investigations in tissue prepared from animals that underwent training of various sorts. Thus Tocco et al. (613) reported that AMPA receptor binding was increased after training in a classical conditioning paradigm, but that no parallel change was observed in NMDA receptor binding. While facilitation of AMPA-mediated responses was shown to improve memory (204, 588), inhibition resulted in retrograde amnesia of inhibitory avoidance training in rats (82, 84, 260). A rapid and selective increase in the density of AMPA receptors in the hippocampus was also reported after training (82, 84), and recent evidence has indicated that this increase in binding reflected enhanced GluR1 expression (83). It was also shown that training led to a time-dependent increase in expression and activity of CaMKII and that this was accompanied by increased AMPA binding in hippocampal synaptosomal membranes and by increased GluR1 subunit phosphorylation (83).

	V. INDUCTION OF LONG-TERM POTENTIATION ACTIVATES SEVERAL CELL SIGNALING CASCADES

Top Previous Next References

 
Since the 1980s, many groups have concentrated their efforts with a view to identifying the signaling pathways that enable LTP to be sustained, and for several years a great debate continued about whether changes occurred presynaptically, postsynaptically, or both. It appears to this author that the predicted outcome, i.e., that changes occur on both sides of the synapse, has long been realized, and the following sections and Table 4 attempt to highlight at least some of the changes reported.

A. cAMP

Several studies have indicated that LTP is dependent on a cascade of cellular signaling events that are stimulated by an increase in intracellular cAMP concentration; these events include activation of PKA, which leads ultimately to activation of transcription factors such as CREB and translation. It has been shown that cAMP concentration and PKA activation are enhanced after induction of LTP and that LTP is inhibited by activators of PKA.

Earlier experiments tended to focus on the role of cAMP/PKA in stimulating changes responsible for sustaining the more persistent components of LTP, and this was mainly as a consequence of studies conducted by the Kandel group (236). For instance, it was shown that delivery of one high-frequency train (100 Hz) induced LTP that persisted for 1–3 h; this was inhibited by inhibitors of CaMKII but was not affected by PKA inhibitors or protein synthesis inhibitors. In contrast, delivery of three high-frequency trains of stimuli was shown to induce LTP that persisted for between 6 and 10 h, and this was blocked by a PKA inhibitor (236). Similarly, late-phase LTP in perforant path-granule cell synapses was shown to be inhibited by the PKA inhibitor Rp-cAMPS and mimicked by the adenylate cyclase activator forskolin (453), while the protein synthesis inhibitor emetine exerted the same effect as Rp-cAMPS (453). This latter observation, as well as earlier observations which pinpointed the coincident timing of the effects of protein synthesis inhibitors and PKA inhibitors, suggested that the effects might be coupled and therefore led to the proposal that the primary effect of PKA was to stimulate protein synthesis. These findings were backed up by observations made in transgenic mice that express R(AB), an inhibitory form of the regulatory subunit of PKA. In these mice, in which PKA activity is markedly reduced, L-LTP was decreased in area CA1, but no effect on the early phase of LTP was observed (3). However, some evidence has suggested that PKA may also play a role in early LTP. For example, PKA was transiently activated 2 and 10 min after induction of LTP in CA1 of the hippocampus, but there was no evidence of a persistent change, suggesting that its role was confined to earlier, rather than later, events (520); it was proposed that this was a consequence of NMDA receptor activation and subsequent activation of calmodulin-dependent adenylyl cyclase (649). A role for PKA in early LTP was also proposed by Otmakhova et al. (477) on the strength of their observation that application of the PKA inhibitor H-89 suppressed the early LTP induced by a single tetanus and that LTP induced by a pairing protocol was decreased by intracellular perfusion of the peptide PKA inhibitor PKI(6—22) amide.

It had been assumed that the primary, and possibly sole, action of cAMP is to activate PKA, but a recent report requires this idea to be revisited. Perfusion of the cAMP analog Rp-cAMPS into CA1 pyramidal cells after induction of LTP decreased the amplitude of the synaptic response in a dose-dependent manner, and the expectation was that this effect was due to its reported inhibitory action on PKA. However, the effect was not mimicked by a specific PKA inhibitor, which suggested a novel action of cAMP (476); these authors proposed that the rapid cAMP-dependent stimulation of the BDNF receptor TrkB may lead to BDNF-dependent synaptic potentiation in CA1 which has been reported by several groups (see below; Ref. 486).

Transgenic mice in which PKA activity is decreased [because they express R(AB); see above], and which exhibited an impairment in L-LTP, also exhibited deficits in spatial memory, indicating that PKA plays a critical role in the consolidation of at least this form of memory (3). In support of a requirement for PKA activation in other forms of memory, it was demonstrated that long-term, but not short-term, contextual fear memory was also impaired in these mice. Consistently, activation of PKA was shown to accompany contextual fear conditioning while Rp-cAMPS inhibits both LTP in amygdala and long-term contextual fear memory (547–549). Similarly, investigation of L-LTP in the cortico-amygdala and the thalamoamygdala pathways revealed the predicted dependence on gene expression and on new protein synthesis; these changes were mediated by activation of PKA as demonstrated by the ability of forskolin to induce LTP in both pathways (238). Interestingly, mice that were maintained in an enriched environment and exhibited enhanced LTP also showed improved memory for contextual, but not cued, fear conditioning. These data suggest that exposure of mice to an enriched environment may alter signaling through PKA and consequently may modulate synaptic plasticity (144).

The dependence of different forms of plasticity on cAMP/PKA is a recurring and unifying theme. For instance, the plasticity induced in the visual cortex as a consequence of monocular deprivation h