Long-term Potentiation

Transcription

1 John Lisman, Brandeis University, Massachusetts, USA Long-term potentiation is an activity-dependent strengthening of synapses that is thought to underlie memory. Introduction One of the major unsolved problems in neuroscience is the mechanism by which memory is stored in the brain. Some property of our neurons must change when we learn, but what parts of the neuron change and the molecular basis of these changes is not known. Long-term potentiation (LTP) has been a focus of efforts to understand memory. The term refers to a long-lasting strengthening of synapses that can be triggered by particular brief patterns of synaptic stimulation. Other patterns can produce a long-term weakening of synapses; this result is termed as long-term depression (LTD) or depotentiation. These findings show that synapses have a defined strength and that this strength can be bidirectionally modified. The evidence that such changes in synaptic strength are actually involved in memory has strengthened in recent years. One class of experiments involved the selective genetic removal of N-methyl-D-aspartate (NMDA) receptor channels from CA1 cells (Tsien et al., 1996). These mice lack the ability to induce LTP. Behavioural studies showed that the animals are defective in learning and memory tasks. In a complementary approach, two groups of experimenters (Gruart et al., 2006; Whitlock et al., 2006) have demonstrated the converse correlation, namely that learning tasks induce LTP in CA1 cells. Several fundamentally different forms of LTP have been discovered. This article will focus on the most studied form that found at the glutamatergic synapses of the CA1 region of the hippocampus. A typical experiment starts by measuring the strength of a group of synapses. This is done by firing a single action potential in some of the axons that enter this region. These axons make synapses with pyramidal cells and generate a graded excitatory postsynaptic potential (EPSP). The strength of the synapses is defined by the magnitude of the EPSP. LTP is then induced by stimulating the axons to fire at high frequency (typically 100 Hz for 1 s), a stimulus referred to as a tetanus. The remarkable finding is that this brief tetanus causes a long-lasting potentiation of the strength of the synapses. The size of the EPSP typically increases by % but can increase by as much as 800%. In the brain-slice preparation used for most studies, potentiation persists until the slice is no longer viable (5 12 h). LTP can also be induced in living animals, and there it can persist for at least a year.. Introduction Advanced article Article Contents. Synapse Specificity and Associative Nature of Long-term Potentiation. Long-term Potentiation as a Mechanism for Memory Storage. Role of Calcium Entry through NMDA Receptor Channels. Presynaptic versus Postsynaptic Sites for Expression of Long-term Potentiation. Molecular Mechanisms of Long-term Potentiation. Structural ModificationofSynapsesandChangesin Gene Expression. Prospects doi: / a pub2 Synapse Specificity and Associative Nature of Long-term Potentiation There are two properties of LTP that have made it a particularly attractive candidate for memory storage. The first is the finding that synapses can be independently modified. The evidence for this synapse specificity comes from experiments in which two sets of synapses on to the same neuron are stimulated. If a tetanus is given to one set, these synapses will be strengthened, but the synapses in the other set will not. A more recent experiment used two-photon uncaging of glutamate in the direct vicinity of an individual synapse (Matsuzaki et al., 2004). Such stimulation could be used to induce LTP at that synapse in much the way LTP is induced by presynaptically released glutamate. The remarkable finding was that other nonstimulated synapses only microns away did not undergo LTP. These results indicate that each synapse can be used to store information. Because most neurons have over excitatory synapses, the potential for information storage is vast. See also: Neurons The second attractive property of LTP is that it is associative. According to nerve network theory (see next section), it is this property that enables small changes in many synapses to produce distributed storage of a complex memory in a nerve network. What associativity means is that LTP cannot be triggered by activity in a single input axon, but can be triggered if that activity is associated with other active excitatory inputs. The basis of this associativity requirement is that LTP induction requires the membrane voltage of the pyramidal cell to become very depolarized (a positive deflection from resting potential) by an amount much larger than that produced by a single excitatory synaptic input (500 mv). This requirement has been demonstrated by artificially depolarizing the cell with positive current. Under these conditions, even stimulation ENCYCLOPEDIA OF LIFE SCIENCES & 2007, John Wiley & Sons, Ltd. 1

2 of a single input can produce LTP. Conversely, if many inputs are stimulated, but the cell is hyperpolarized (made more negative) by current injection, LTP does not occur. Thus, a synapse will be strengthened if two conditions are met: the synapse is activated by presynaptic activity and there is substantial postsynaptic depolarization due to other excitatory inputs (and not too much inhibitory input). This dual requirement for strengthening is often referred to as the Hebb rule. See also: Membrane Potential One of the elegant aspects of the Hebb rule is that it can be executed at each synapse using information that is locally available. A synapse can know that its presynaptic input is active because its receptors will be activated by the neurotransmitter released. The synapse can also know about the overall depolarization of the neuron because this spreads fairly uniformly over the cell membrane; to a first approximation, the voltage at any point reflects the combined influence of all the excitatory and inhibitory inputs on to a cell. Membrane ion channels in the postsynaptic membrane have the capability of sensing this voltage. As we will see later, an ion channel called the (NMDA) receptor can in fact execute the Hebb rule by sensing both presynaptic glutamate release and membrane depolarization. Long-term Potentiation as a Mechanism for Memory Storage The theory of nerve networks provides us with a rough outline of how synapses in a network of neurons might store a memory in a distributed way using a Hebbian form of LTP. Let us consider how a visual scene might be stored. The scene would first be processed by low-level networks in which neurons responded to elementary visual features. These features would then be synaptically associated in a higher-level network specialized for memory storage. In an idealized version of such a memory network, each neuron makes a synaptic connection with every other neuron and it is at these recurrent connections that memory can be stored by the Hebbian modification. Let us say that when the visual scene is present, a subset of cells in the memory network will fire, specifically cell x and a group of other cells, G. Focusing on the connection of the G cells to the x cell, we can see that the Hebb rule will be satisfied at all these synapses because there is both presynaptic input and postsynaptic depolarization. These connections will therefore undergo LTP. This strengthening makes it possible for G cells to collectively fire the x cell. To see the utility of this change, consider a recall process in which the same scene is presented, but missing certain features, in particular the ones that would normally cause the x cell to fire. Cell x will nevertheless fire because it receives strong inputs from the G cells. In this way, a complete memory can be recalled when the network is cued with only a part, a fundamental property of associative memory. See also: Learning and Memory Role of Calcium Entry through NMDA Receptor Channels A major discovery is that the properties of a single receptor, the NMDA receptor, can account for the Hebbian property of LTP. The NMDA receptor channel is one of the two major subtypes of ionotropic receptors for the neurotransmitter glutamate, the neurotransmitter at most excitatory synapses in the central nervous system. The other major subtype of ionotropic receptor is the a-amino-3-hydroxy- 5-methyl-4-isoxazolepropionate (AMPA) receptor. These receptors are also ion channels and they produce the bulk of the excitatory postsynaptic potential. The function of NMDA receptors was unclear until it was shown that they have a special role in synaptic plasticity. This was demonstrated by showing that if NMDA receptor channels are pharmacologically blocked, LTP cannot be induced (Bliss and Collingridge, 1993). Subsequent work used genetic methods to eliminate the NMDA receptor channel came to the same conclusion. See also: Glutamate as a Neurotransmitter; NMDA Receptors An unusual property of NMDA receptor channels explains how these channels trigger LTP in accord with the Hebb rule. Most transmitter-activated channels are opened by transmitter alone; their opening does not depend on membrane voltage. In contrast, NMDA receptor channels open only if there is also substantial postsynaptic depolarization. This voltage dependence has a simple mechanism. If the postsynaptic neuron is near resting potential, glutamate cannot activate the channel because the pore of the channel is blocked by external magnesium. However, if the postsynaptic neuron is depolarized, Mg 2+ is electrically driven out of the pore, and the channel can open. These dual requirements for opening the NMDA receptor channel are exactly the requirements for synaptic modification according to the Hebb rule. See also: Neural Activity and the Development of Brain Circuits But what is it about the opening of the NMDA receptor channels that leads to LTP induction? It is now clear that NMDA receptor channels are much more permeable to Ca 2+ than are most AMPA receptor channels and that it is the influx of Ca 2+ through the NMDA receptors that triggers LTP. A key finding was that LTP could be blocked by intracellular Ca 2+ buffers (Lynch et al., 1983), molecules that bind Ca 2+ and prevent elevations in Ca 2+ concentration. Not only is Ca 2+ elevation necessary for inducing LTP, it also appears to be sufficient: LTP occurs when light releases Ca 2+ from special light-sensitive buffers that have been injected into the cell. See also: Calcium Signalling and Regulation of Cell Function Fluorescent Ca 2+ indicators and sensitive light detectors have been used to directly measure the intracellular Ca 2+ elevation that occurs during LTP. The actual sites of synaptic input into pyramidal cells are primarily on small (1 mm) protuberances from dendrites called spines. During LTP, spines at which the Hebbian condition is met undergo Ca 2+ elevation to levels higher than 10 mmol L ENCYCLOPEDIA OF LIFE SCIENCES & 2007, John Wiley & Sons, Ltd.

3 Remarkably, the Ca 2+ level in the spine itself can be higher than in the parent dendrite and nearby spines, even though the distances involved are very small (Connor et al., 1994). The localization of Ca 2+ may be responsible for the synapse specificity of LTP. See also: Dendritic Spines; Fluorescent Probes Used for Measuring Intracellular Calcium An important unanswered question about LTP is the identity and properties of the postsynaptic depolarizing event or events that allow the NMDA receptor channels to open. One candidate is Na + action potentials. It is known that synaptic stimuli strong enough to induce LTP produce action potentials. Many action potentials are initiated in the axon hillock and actively backpropagate into the dendrites. If subthreshold synaptic stimulation too weak to evoke LTP is repeatedly coupled with a spike produced by somatic current injection, LTP occurs. However, the role of Na + action potentials during synaptically induced LTP has not been demonstrated. Indeed to the contrary, experiments showed that if the back propagating action potential is blocked, the LTP induced by strong synaptic inputs still occurs normally (Golding et al., 2002). Other depolarizing events that could activate NMDA receptors include Ca 2+ action potentials or the dendritic EPSP itself, some of which is caused by current through the NMDA receptor channel itself. We will return later to a discussion of the biochemical processes triggered by Ca 2+ elevation. See also: Action Potential: Ionic Mechanisms Presynaptic versus Postsynaptic Sites for Expression of Long-term Potentiation The expression of potentiated transmission during LTP could be due to changes that enhance the responsiveness of the postsynaptic cell to neurotransmitter. Alternatively, potentiation could also be due to presynaptic changes that result in more transmitter being released. Since LTP is induced by postsynaptic events, if expression is presynaptic there would have to be a retrograde message that carried information from the postsynaptic cell back to the presynaptic cell. Determining the site of expression of LTP has been surprisingly difficult, but there is now strong evidence that postsynaptic modification occurs and specific mechanisms for producing it have been established. There is also increasing evidence for presynaptic modifications. The strong evidence (Nicoll, 2003) for postsynaptic modification comes from several lines of experiments: (1) AMPA receptor channels have enhanced conductance after LTP; (2) Miniature synaptic current responses, the amplitude of which is thought to be controlled postsynaptically, are increased after LTP; (3) By marking the GluR1 form of AMPA receptor channels with an electrophysiologically observable tag, it has been shown that GluR1 receptors become incorporated into the synaptic membrane after LTP; (4) The two-photon uncaging method mentioned earlier has been used to induce LTP under conditions where the presynaptic axon is not functioning. See also: AMPA Receptors The evidence that LTP produces an increase in transmitter release from the presynaptic cell comes from several lines of experiments. An indirect class of evidence is the observation that synapses fail less often after LTP induction. According to the classical interpretation of failures derived from study of the neuromuscular junction, failures occur because a presynaptic action potential sometimes fails to produce the release of even a single synaptic vesicle. Thus, a change in failures after LTP induction is consistent with a presynaptic change in vesicle release. However, there is now substantial evidence that transmission at central synapses is different from that at the neuromuscular junction and that changes in failures can sometimes arise through a postsynaptic change in the following way. A synapse might initially have NMDA receptor channels, but no AMPA receptor channels (Liao et al., 1995). Because NMDA receptor channels do not produce significant postsynaptic current near resting potential, the synapse would appear to be silent when synaptic transmission was assayed under normal conditions (i.e. near resting potential). After LTP, postsynaptic modifications would lead to the introduction of functional AMPA receptor channels at the synapse and these would generate synaptic responses. It follows that a decrease in failures can be due to a postsynaptic modification. See also: Heterosynaptic Modulation of Synaptic Efficacy; Neurotransmitter Receptors in the Postsynaptic Neuron; Neurotransmitter Release from Presynaptic Terminals Because of this ambiguity, there has been the need for more direct methods to determine whether there are presynaptic changes during LTP. Direct evidence (Zakharenko et al., 2003) for such changes has been provided by experiments that use the dye FM1-43. The dye can be loaded into synaptic vesicles. When the dye is released from the vesicle along with glutamate as a result of synaptic stimulation, the release can be monitored by a change in fluorescence of the dye (the dye is responding to ph changes when the dye is released from the very acid interior of the vesicle). This method shows that release is enhanced after LTP. Importantly this enhancement develops slowly (30 min) after LTP induction. By comparison, postsynaptic effects develop within seconds. In addition to changes in the number of vesicles released, there are indications that a change in the mode of vesicle release may occur (Choi et al., 2003). In one mode, glutamate is released from the vesicle so slowly that AMPA channels are not activated. Such synapses are termed whispering synapses. After LTP induction, release becomes fast enough to activate AMPA channels, thereby contributing to the potentiation of transmission. See also: Nitric Oxide as a Neuronal Messenger As LTP in the CA1 region is induced postsynaptically as a result of activation of NMDA receptor channels, retrograde messengers would seem to be required to affect presynaptic release. The molecules nitric oxide (NO) and various metabolites of arachidonic acid ENCYCLOPEDIA OF LIFE SCIENCES & 2007, John Wiley & Sons, Ltd. 3

4 (endocanabinoids, prostaglandins, eicosinoids) are candidates. Early pharmacological studies suggested that LTP could be blocked by interfering with NO, but subsequent work showed that LTP in the CA1 hippocampal region could be induced even in the absence of neuronal nitric oxide synthase, the isoform activated by synaptic activity. However, LTP under these conditions may have been primarily postsynaptic. A clear demonstration of a presynaptic component of LTP dependent on NO activity has been found in the barrel cortex of mice (Hardingham and Fox, 2006). In these mice the postsynaptic component of LTP dependent on the GluR1 receptor was eliminated genetically. The remaining LTP was dependent on nitric oxide synthetase (NOS) activity and this LTP had the properties of presynaptic changes in transmitter release. Molecular Mechanisms of Long-term Potentiation What kinds of biochemical processes could underlie bidirectional synaptic modification and provide the basis for the stable storage of information (Lisman, 1989). No definitive answer is available yet, but a general framework (Figure 1) is as follows: 1. The high Ca 2+ elevation that occurs during LTP induction triggers enzymatic processes that strengthen the synapse, whereas the more moderate Ca 2+ elevation that occurs during induction of LTD or depotentiation triggers processes that weaken the synapse. 2. The high Ca 2+ elevation activates a kinase, probably CaMKII, whereas moderate Ca 2+ elevation triggers a phosphatase cascade. These produce reversible phosphorylation and dephosphorylation, respectively, of target proteins that control the strength of the synapse. See also: Protein Phosphorylation and Long-term Synaptic Plasticity 3. The memory of synaptic modification may arise from the persistent nature of the kinase activity. In the case of CaM-kinase, persistence arises from the maintained autophosphorylation. According to this view, the kinase is a molecular switch that is turned on by Ca 2+ and remains on until it is turned off by phosphatases activated during depotentiation. In the on-state, the switch enhances synaptic transmission by phosphorylating AMPA receptor channels or stimulating their insertion into the synapse. See also: Cellular Neuromodulation The role of Ca 2+ in triggering LTP is well accepted, as is the role of more moderate Ca 2+ elevation in triggering synaptic weakening and depotentiation. Proposition (1) above is thus on reasonably firm ground. The evidence for the role of CaMKII and perhaps other kinases in LTP is now strong, but their role in maintaining LTP (proposition (3)) is uncertain. These issues will be discussed in the next section. The evidence for the role of phosphatases in some forms of long-term depression and depotentiation (proposition 2) is reasonably good; this is reviewed in the related article on long-term depression. See also: Long-term Depression and Depotentiation CaMKII is a major component of the postsynaptic density, a submembrane structure attached to the intracellular side of the postsynaptic membrane. The kinase is thus strategically located to sense Ca 2+ entry through NMDA receptor channels, the signal that we know triggers LTP. Furthermore, by being localized at the synapse, CaM-kinase can regulate the proteins that control synaptic strength (e.g. the AMPA receptor channels) in a synapse-specific manner. Several strategies (Lisman et al., 2002) have been used to test the role of CaMKII in LTP. In early experiments, inhibitors of this kinase were injected into the postsynaptic cell and it was found that this blocked LTP induction. Subsequently, genetic methods were used to knockout the kinase and this also blocked LTP induction. Of particular note is a set of experiments in which a mutated form of CaMKII was knocked in, replacing the normal form (Giese et al., 1998). This mutated form was altered in only one amino acid Thr286, a phosphorylation site crucial for the persistent activation of the kinase (see later). In this mutant, LTP was blocked. Notably, behavioural experiments showed that memory was strongly reduced by this mutation. It has also been shown that introduction of active CaM-kinase into the postsynaptic cell can itself potentiate synaptic transmission. This not only mimics LTP, but prevents subsequent LTP induction by a tetanus, as if the two methods for enhancing transmission share a common mechanism. Progress has been made in understanding how CaMkinase can make synapses stronger. One mechanism is by the direct phosphorylation of AMPA receptor channel. A specific site (Ser831) on the GluR1 form of AMPA receptor channels can be phosphorylated by both CaM-kinase and C-kinase and enhances the function of the channels (Lee et al., 2000). In addition to increasing the conductance of AMPA channels by such modulation, CaMKII is likely to control the anchoring of these channels at the synapse. CaMKII is the most abundant protein in the postsynaptic density (PSD), suggestive of a structural role. One way CaMKII can affect anchoring is by binding to SAP97, a protein that can bind AMPA channels. CaMKII is also bound indirectly to other AMPA binding proteins. During LTP, CaMKII translocates to the synapse and remains there for a long period. This additional CaMKII may serve as structural seed for complexes that anchor additional AMPA channels at the synapse. There is also evidence to suggest that CaM-kinase has a role in L-LTP protein synthesis through phosphorylation of CPEB. Although most research has focused on the role of postsynaptic CaMKII in LTP, there is also evidence supporting a role for presynaptic CaMKII (Lu and Hawkins, 2006); introduction of CaM-kinase inhibitory peptide into the presynaptic cell decreases a component of LTP. See also: Glutamatergic Synapses: Molecular Organization 4 ENCYCLOPEDIA OF LIFE SCIENCES & 2007, John Wiley & Sons, Ltd.

5 Summed EPSPs and IPSPs L-VDCC KV RyR Calpain Negative feedback α-actinin calmodulin V post IP3R IP3 Calmodulin release High Ca 2+ Moderate Ca 2+ Positive feedback Neurogranin PKC NDMA SRC mgur5 PLC AA PLA MAPK Presynaptic glutamate release Back propagating Na + spikes Ca 2+ spikes ACI NOS camp PKA RAS Syngap AMPA CaMK mrna PKM-ζ CPEB Synthesis I1p PP2b CaMK I1 PP1 CaMK p Binding NMDA CREB, MAPK Activity-dependent redistribution Actin Snap NSF AMPA channel insertion Integrin NCAM N-CADHERIN AMPA AMPA Figure 1 Postsynaptic mechanisms underlying long-term potentiation (LTP), long-term depression (LTD) and depotentiation (adapted from (Lisman, 1994)). A dendritic spine is shown protruding from a small region of dendrite. The synaptic transmission (far right) is mediated by the neurotransmitter glutamate. The postsynaptic membrane contains AMPA and NMDA ionotropic glutamate channels and the metabotropic glutamate receptor, mglur5. If the postsynaptic cell is strongly depolarized and if glutamate is being released presynaptically, then the NMDA receptor channel will open and LTP will be induced. The NMDA receptor channel itself is under complex control through positive and negative feedback loops. The final consequence of NMDA receptor channel opening is a high elevation of intracellular Ca +, which then triggers processes that lead to the upregulation of AMPA receptor channels. The upregulation occurs either by phosphorylation of existing AMPA receptor channels orbyaddition of new channels (seebottom of figure). During LTP induction theactivityof CaM-kinase is enhanced and this produces the phosphorylation of AMPA receptor channels. CaM-kinase itself becomes phosphorylated and in this state, its phosphorylation is self-sustaining. Other protein kinases, PKA and PKC may also be involved in the phosphorylation of AMPA receptor channels. The controls on PKC appear to be complex and involve the synthesis of new forms (PKM-z) and the control by a positive feedback pathway involving arachidonic acid (AA), phospholipase A 2 and RAS. The addition of new AMPA receptor channels depends on the movement of vesicles containing AMPA receptor channels into the spine during LTP induction and the fusion of vesicles containing AMPA receptor channels into the plasma membrane. Fusion involves two proteins, SNAP and NSF. The Ca 2+ elevation that occurs during synaptic signalling may also depend on Ca 2+ released from intracellular stores by IP3 receptors and ryanodine receptors and by Ca 2+ entry through L-type voltage-dependent Ca 2+ channels located in the spines. If postsynaptic depolarization is not strong, NMDA receptor channels will be only moderately activated and this will lead to a moderate elevation of Ca 2+ that induces synaptic weakening (LTD or depotentiation). One form of this weakening is controlled by activation of a phosphatase pathway involving phosphatase 2B (pp2b), which dephosphorylates Inhibitor 1 (I1), and leads to activation of phosphatase 1 (PP1). One role of PP1 is to dephosphorylate CaMK and this may lead to synaptic weakening. During LTP induction, when it is important for CaMK to become phosphorylated, it is undesirable to activate PP1. This is prevented by a pathway involving adenylate cyclase 1 (AC1), camp and PKA. This pathway acts to counteract theeffect ofpp2boni1. OthermoleculesofpotentialsignificanceforLTP includethe celladhesionfactorsshown onthe bottom rightandthe activitydependent mechanism that controls the translation of CaMK mrna. CaM-kinase is an interesting candidate as a memory molecule because it has switch-like properties that make its activity persist after calcium elevation returns to baseline. This memory occurs because Thr286 sites become autophosphorylated during the calcium elevation. Experiments show that this phosphorylation can persist for many hours after LTP induction. What potentially limits long-term information storage by such a process is spontaneous phosphatase activity, which could dephosphorylate Thr286. Furthermore, all proteins undergo protein turnover, so information stored in any one molecule might be slowly eliminated by this process. Theoretical work, however, points to the possibility that on CaMKII holoenzyme (there are 12 similar subunits in a holoenzyme) could ENCYCLOPEDIA OF LIFE SCIENCES & 2007, John Wiley & Sons, Ltd. 5

6 retain the on-state indefinitely. Dephosphorylation of Thr286 sites could be counteracted by rapid rephosphorylation of these sites by neighbouring subunits. Cooperative interactions between holoenzymes could also lead to phosphorylation of holoenyzmes newly inserted in the process of protein turnover. In this way, information could be stably stored by groups of CaMKII holoenzymes. Although there are indications that CaMKII is responsible for the maintenance of LTP, the evidence is not yet conclusive. Evidence is accumulating that another kinase, the PKM-z isoform of C-kinase, has an important role in LTP maintenance. The concentration of this enzyme is increased during LTP and decreased during long-term depression. The increase of PKM-z after LTP induction is due to protein synthesis. Importantly, the gene produces a protein that lacks a regulatory domain and is therefore constitutively on. Inhibition of PKM-z does not affect the early phase of LTP. Rather late phase LTP will not develop or will be reversed if inhibitor is applied after LTP is already established. Basal synaptic transmission is not affected. Interestingly, a membrane-permeant peptide inhibitor of PKM-z, ZIP, not only reverses LTP when applied in the slice preparation, but can irreversibly destroy a memory (place avoidance), when applied in vivo after learning. (Pastalkova et al., 2006). See also: Arachidonic Acid Signalling in the Nervous System; Receptor Transduction Mechanisms An important area of progress has been in understanding the role of different AMPA receptor subunits in LTP and basal synaptic transmission (Malinow and Malenka, 2002). Different subunits appear to have different roles. Notably GluR1 is of particular importance in the earliest phases of LTP. Knockout of GluR1 preferentially reduces the early phase. Interestingly, this knockout totally eliminates extrasynaptic AMPA receptors and a phenomenon called distance-dependent scaling. There is increasing evidence that extrasynaptic GluR1 is in a diffusional equilibrium with binding sites at the synapse. GluR1 appears to be in a complex with a protein termed stargazin and it is the binding of stargazin to a synaptic anchoring protein, PSD- 95, that captures GluR1 at the synapse. The importance of PSD-95 is underscored by the finding that overexpression of this protein can enhance synaptic strength, while eliminating it and other proteins of the same function reduces synaptic strength (El-Husseini Ael et al., 2002). One of the difficulties in understanding transmission under basal conditions is that different subunits and subunit combinations appear to have different roles, but the exact nature of these roles remains to be elucidated. Scores of substances affect the induction of LTP, giving the impression that synaptic modification may be an extremely complex process. However, many of these may not interfere with the biochemistry of synaptic modification itself, but rather act indirectly to affect postsynaptic depolarization. Since this will affect the opening of NMDA receptor channels, LTP will be secondarily affected. For instance, in some regions the inhibition that is also evoked by the stimulating electrode is so strong the LTP cannot be induced. Substances that reduce this inhibition can enhance LTP induction. Perhaps less obvious is that the tetanic stimulation usually used to induce LTP places high demands on the mechanisms that keep vesicles available for release in the presynaptic terminal. Results suggest that brain-derived neurotrophic factor (BDNF) affects LTP induction in part by modulating such mechanisms, and the same may be true for agents that work through cgmp. Other work shows that activation of MAP-kinase through the insulin receptor can enhance the BK K + conductance in hippocampal neurons thereby reducing excitability. The resulting interference with synaptically induced depolarization may account for the reduction of LTP by these inhibitors. Also of clear importance in LTP are tyrosine kinases, particularly Src. Results indicate that a major role of Src is in the activity-dependent upregulation of the NMDA receptor channel itself. Structural Modification of Synapses and Changes in Gene Expression The structural basis of changes in synaptic strength is of increasing interest. When LTP is induced, spines get larger. Conversely, LTD makes spines become smaller. These changes in spine size involve changes in the cytoskeleton, notably changes in the amount of F-actin within the spine. Only electron-microscopic (EM) methods have sufficient resolution to image the synapse itself. This form of microscopy requires fixation and sectioning, and it is therefore not possible to examine the same synapses before and after LTP induction. So far the only available approach has been to compare the average synapse structure in different slices, before and after LTP induction. This type of work strongly suggests that the synapse itself enlarges after LTP induction (Harris et al., 2003). Even nearby synapses on the same dendrite differ greatly in structure. Interestingly, specializations on both sides of the synapse, the presynaptic grid and postsynaptic density, are both highly variable in size, but have exactly the same size at any given synapse. Indeed their edges are exactly in register. This suggests that the synapse grows through a structurally coordinated transsynaptic process. If such growth occurred during LTP, it would probably lead to both a change in the probability of release and an enhanced postsynaptic response. The biochemical factors involved in structural control are beginning to be elucidated. For instance, LTP is inhibited by agents that interfere with adhesion molecules. These adhesion molecules, including L1, NCaM, integrins and cadherin, are found in synaptic junctions and may be particularly concentrated at the punctaadherin, a specialized subregion of the synaptic region. The control mechanisms that regulate actin changes are also being elucidated. One such mechanism involves cofilin, a powerful agent for actin reorganization through depolymerization and cutting of actin filaments. Genetic removal of LIM-kinase, which inactivates cofilin, leads to reduction of 6 ENCYCLOPEDIA OF LIFE SCIENCES & 2007, John Wiley & Sons, Ltd.

7 spine size and deficits in long-term potentiation. Complementary experiments involving inhibition of the phosphatase that activates phospho-cofilin, showed deficits in longterm depression (Morishita et al., 2005). The early phases of LTP develop with seconds after induction and thus must involve purely modulatory processes. However, on a longer time scale changes in gene transcription and translation may be involved. There is now clear evidence that later phases of LTP are inhibited by protein synthesis inhibitors. Moreover, the induction of LTP triggers an increase in protein synthesis. For instance, it is known that a few forms of messenger ribonucleic acid (mrna) are translated locally in the dendrites and that local regulators of translation can themselves be controlled by NMDA receptor channel activation. The utilization of these proteins is affected by a process called synaptic tagging. This was shown by inducing LTP in one pathway and thereby activating protein synthesis. Then a protein synthesis inhibitor was applied and LTP induced in a second pathway. The important observation is that the LTP in this second pathway had a normal late phase, despite the presence of protein synthesis inhibitor, presumably because it was able to use the proteins whose synthesis was triggered by the first pathway. Interestingly, this ability of one pathway to enhance another has a limited lifetime; if the pathways were stimulated more than a few hours apart, there was no interaction. To explain these results, it was proposed (Frey and Morris, 1997) that LTP induction produces a transient synapse-specific biochemical modification that functions as a tag to incorporate proteins whose synthesis is triggered by a process that is not synapse-specific. See also: Integrin Superfamily LTP induction causes changes in the transcription of many genes and triggers new protein synthesis. The early, transient burst of protein synthesis in the postsynaptic dendrite utilizes pre-existing mrna transcribed earlier from what are known as intermediate-early genes (IEGs). These genes encode proteins such as the transcription factors zif/268, c-fos and jun; growth factors such as BDNF; and enzymes such as CaMKII. The immediate early gene product, Arc/Arg3.1 interacts with dynamin and endophilin to regulate AMPA channel receptor endocytosis (Tzingounis and Nicoll, 2006). These proteins form the bridge between the temporary mechanistic changes comprising early LTP and the sustained late LTP structural and functional changes. Experiments in which inhibitors of transcription were found to block synaptic plasticity and the findings that some IEGs are transcription factors have suggested a role for the nucleus in LTP. In support of this hypothesis, in mutant mice lacking the transcription factor CREB, LTP is nearly normal soon after induction but is reduced at late times. However, the accumulating evidence about specific gene products and LTP mechanisms suggests that local dendritic translation of pre-existing mrna is sufficient for synaptic plasticity. See also: Protein Synthesis and Long-term Synaptic Plasticity BDNF has a major role in late LTP (Bramham and Messaoudi, 2005). The peptide is secreted from the presynaptic cell during LTP induction, primarily as proto-bdnf, and is converted to BDNF through the action of plasmin, a protease that itself undergoes activitydependent secretion and activation by tpa and possibly MMP-9. BDNF acts on presynaptic TrkB receptors to enhance the vesicle release process during high frequency stimulation. It also acts postsynaptically to enhance depolarization directly and, through multiple kinase cascades, enhances protein synthesis. Among the proteins synthesized is TrkB itself. The phase of LTP that occurs during the first 30 min does not appear to require BDNF action, but later phases do. Notably, interfering with TrkB function at 30 min after LTP induction produces a rapid decrease in potentiation, suggesting that BDNF action is required for a considerable time after induction to maintain LTP. Experiments with dopamine antagonists or knockout of D1 receptors indicate that dopamine is required for late LTP. This raises the question of what causes the firing of the dopamine cells in the ventral tegmental area (VTA) that provides dopamine to the hippocampus. It now seems likely that the hippocampus itself computes the novelty of incoming information and that the resulting novelty signal is sent to the VTA. In this way, the transition to long-term information storage (late LTP) may be conditional upon the novelty of the incoming information (Lisman and Grace, 2005). Prospects There are many developing areas of research that will yield new insight into the mechanisms of LTP and memory. In the past, physiological studies have dealt almost exclusively with the study of populations of synapses, but techniques are now being developed to study individual synapses. Combined with imaging methods, it should be possible to watch synapses as they undergo LTP and to obtain insight into structure function issues. Rapid progress is also being made in understanding the molecular composition of the synapse. Once a molecule is identified, it is possible easily to screen for other molecules that bind to that molecule. In this way, a picture is beginning to emerge of the molecular complexes on both sides of the synapse. However, this effort is hindered by lack of direct structural information about the PSD. Refinements of EM methods using tomography offers hope in this regard. See also: G Proteins; Synapses; Visual System Development in Vertebrates Ultimately, the testing of hypotheses regarding the molecular basis of memory requires the ability to assay memory itself. There is increased understanding that there are many forms of memory and that they may be mediated by different mechanisms. New behavioural tests are being developed that allow particular forms of memory to be assessed. The ability of new genetic methods to produce defined molecular changes at precise times and precise locations in living animals will make it possible to test whether particular forms of memory rely on particular molecules. See also: Memory: Clinical Disorders; Alzheimer Disease ENCYCLOPEDIA OF LIFE SCIENCES & 2007, John Wiley & Sons, Ltd. 7

8 We are who we are largely because of our accumulated experience stored in memory. As more is learned about LTP, it will make possible the manipulation of memory and provide solutions to memory problems. It would clearly be undesirable to wipe our memory clear. But as we come to understand the biochemical basis of memory, it may become possible to affect specific memories. This may help in the treatment of unwanted memories, such as those involved in posttraumatic stress disorders. The implications for understanding the molecular basis of memory are vast and probably largely unanticipated. The obvious hope is that this understanding will lead to treatments of memory disorders, such as Alzheimer disease. Recently the science of LTP and the science of Alzheimer disease have begun to converge. Of particular import may be the understanding of the gating process that enhance LTP and allow the transition from early LTP to late LTP. Stimulation of such processes offers the hope for enhancing memory processes. See also: Posttraumatic Stress Disorder References Bliss TV and Collingridge GL (1993) A synaptic model of memory: long-term potentiation in the hippocampus. Nature 361(6407): Bramham CR and Messaoudi E (2005) BDNF function in adult synaptic plasticity: the synaptic consolidation hypothesis. Progress in Neurobiology 76(2): Choi S, Klingauf J and Tsien RW (2003) Fusion pore modulation as a presynaptic mechanism contributing to expression of longterm potentiation. Philosophical Transactions of the Royal Society of London, Series B: Biological Sciences 358(1432): Connor JA, Miller LD, Petrozzino J and Muller W (1994) Calcium signaling in dendritic spines of hippocampal neurons. Journal of Neurobiology 25(3): El-Husseini Ael D, Schnell E, Dakoji S et al. (2002) Synaptic strength regulated by palmitate cycling on PSD-95. Cell 108(6): Frey U and Morris RG (1997) Synaptic tagging and long-term potentiation. Nature 385(6616): Giese KP, Fedorov NB, Filipkowski RK and Silva AJ (1998) Autophosphorylation at Thr286 of the a calcium-calmodulin kinase II in LTP and learning. Science 279(5352): Golding NL, Staff NP and Spruston N (2002) Dendritic spikes as a mechanism for cooperative long-term potentiation. Nature 418(6895): Gruart A, Munoz MD and Delgado-Garcia JM (2006) Involvement of the CA3-CA1 synapse in the acquisition of associative learning in behaving mice. Journal of Neuroscience 26(4): Hardingham N and Fox K (2006) The role of nitric oxide and GluR1 in presynaptic and postsynaptic components of neocortical potentiation. Journal of Neuroscience 26(28): Harris KM, Fiala JC and Ostroff L (2003) Structural changes at dendritic spine synapses during long-term potentiation. Philosophical Transactions of the Royal Society of London, Series B: Biological Sciences 358(1432): Lee HK, Barbarosie M, Kameyama K, Bear MF and Huganir RL (2000) Regulation of distinct AMPA receptor phosphorylation sites during bidirectional synaptic plasticity. Nature 405(6789): Liao D, Hessler NA and Malinow R (1995) Activation of postsynaptically silent synapses during pairing-induced LTP in CA1 region of hippocampal slice. Nature 375(6530): Lisman J (1989) A mechanism for the Hebb and the anti-hebb processes underlying learning and memory. Proceedings of the National Academy of Sciences of the USA 86(23): Lisman J (1994) The CaM kinase II hypothesis for the storage of synaptic memory. Trends Neuroscience 17(10): Lisman JE and Grace AA (2005) The hippocampal-vta loop: controlling the entry of information into long-term memory. Neuron 46(5): Lisman J, Schulman H and Cline H (2002) The molecular basis of CaMKII function in synaptic and behavioural memory. Nature Reviews Neuroscience 3(3): Lu FM and Hawkins RD (2006) Presynaptic and postsynaptic Ca(2+) and CamKII contribute to long-term potentiation at synapses between individual CA3 neurons. Proceedings of the National Academy of Sciences of the USA 103(11): Lynch G, Larson J, Kelso S, Barrionuevo G and Schottler F (1983) Intracellular injections of EGTA block induction of hippocampal long-term potentiation. Nature 305(5936): Malinow R and Malenka RC (2002) AMPA receptor trafficking and synaptic plasticity. Annual Review of Neuroscience 25: Matsuzaki M, Honkura N, Ellis-Davies GC and Kasai H (2004) Structural basis of long-term potentiation in single dendritic spines. Nature 429(6993): Morishita W, Marie H and Malenka RC (2005) Distinct triggering and expression mechanisms underlie LTD of AMPA and NMDA synaptic responses. Nature Neuroscience 8(8): Nicoll RA (2003) Expression mechanisms underlying long-term potentiation: a postsynaptic view. Philosophical Transactions of the Royal Society of London, Series B: Biological Sciences 358(1432): Pastalkova E, Serrano P, Pinkhasova D et al. (2006) Storage of spatial information by the maintenance mechanism of LTP. Science 313(5790): Tsien JZ, Huerta PT and Tonegawa S (1996) The essential role of hippocampal CA1 NMDA receptor-dependent synaptic plasticity in spatial memory. Cell 87(7): Tzingounis AV and Nicoll RA (2006) Arc/Arg3.1: linking gene expression to synaptic plasticity and memory. Neuron 52(3): Whitlock JR, Heynen AJ, Shuler MG and Bear MF (2006) Learning induces long-term potentiation in the hippocampus. Science 313(5790): Zakharenko SS, Patterson SL, Dragatsis I et al. (2003) Presynaptic BDNF required for a presynaptic but not postsynaptic component of LTP at hippocampal CA1-CA3 synapses. Neuron 39(6): ENCYCLOPEDIA OF LIFE SCIENCES & 2007, John Wiley & Sons, Ltd.

9 Further Reading Kelleher RJ 3rd, Govindarajan A and Tonegawa S (2004) Translational regulatory mechanisms in persistent forms of synaptic plasticity. Neuron 44(1): Malenka RC and Bear MF (2004) LTP and LTD: an embarrassment of riches. Neuron 44(1): Milner B, Squire LR and Kandel ER (1998) Cognitive neuroscience and the study of memory. Neuron 20: Roberson ED, English JD and Sweatt JD (1996) A biochemist s view of long-term potentiation. Learning and Memory 3(1): Sheng M and Hoogenraad CC (2006) The postsynaptic architecture of excitatory synapses: a more quantitative view. Annual Review of Biochemistry 76 EPUB: abs/ /annurev.biochem ENCYCLOPEDIA OF LIFE SCIENCES & 2007, John Wiley & Sons, Ltd. 9