An increase in synaptic efficacy that persists for hours to more than days after the delivery of a brief induction stimulus

In the field of memory research, interest in activity dependent lasting synaptic "plasticity is a natural sequel to the tenet that learning involves synaptic modifications. Over the years, cellular physiologists have identified a number of stimulation protocols that unveil synaptic plasticity (Johnston and Wu 1995). For example, in many types of synapses, when a pair of stimuli are delivered sequentially within a fraction of a second, the response to the second stimulus is larger than that to the first (del Castillo and Katz 1954). This is 'paired-pulse facilitation' (PPF). Stimulation by a train of stimuli ('tetanic stimulation') could result in augmentation of synaptic response that lasts seconds to minutes (Larrabee and Bronk 1947; Lloyd 1949). This is 'post-tetanic potentiation' (PTP). The short life of PPF and PTP, and the fact that they were initially investigated in synapses that do not connote learning such as the neuromuscular junction, did not stir much excitation in the memory community. This situation changed when investigators in the laboratory of Per Andersen in Oslo noted that in the "hippocampus, certain tetanic stimulation protocols resulted in enhanced synaptic efficacy (Lomo 1966; Andersen and Lomo 1967), which could "persist for hours to days (Bliss and Lomo 1973; Bliss and Gardner-Medwin 1973).1 Enter long-term potentiation (LTP).

LTP is a generic term. It is now used to refer to a heterogeneous form of neuronal plasticity observed in many types of synapses. Many neuroscientists consider LTP a mechanism that implements learning, short-term and intermediate-term memory at the cellular "level. But by no means is this accepted by all. To gain insight into the phenomenon and the controversy that encircles it, it helps to recall the original observation. Bliss and Lomo (1973) stimulated the perforant path leading from the entorhinal "cortex to the dentate gyrus in the hippocampal formation of the anaesthetized rabbit, and found that after a conditioning train at 100 Hz for 3-4 s or 10-20 Hz for 10-15, the dentate response to a single afferent volley was potentiated for hours, and in the unanaesthetized rabbit, even for weeks (Bliss and Gardner-Medwin 1973). This change was noted in the amplitude of the excitatory postsynaptic potential (EPSP) and in the amplitude and latency of the population spike (EPSP-to-spike potentiation, abbreviated E-S potentiation). This implied that both the synaptic strength and postsynaptic excitability have changed. LTP was hence discovered by applying non-physiological stimuli in a paradigm that did not involve learning. Bliss and Lomo (1973) were well aware of that, but also realized the potential of their finding: 'Our experiments show that there exists at least one group of synapses in the hippocampus whose efficiency is influenced by activity that may have occurred several hours previously—a time-scale long enough to be potentially useful for information storage. Whether or not the intact animal makes use in real life of a property which has been revealed by synchronous, repetitive volleys to a population of fibers... is another matter'. With time, variants of LTP have been demonstrated in other hip-pocampal synapses, the Schaffer collateral/commissural synapses in area CA1 and the mossy fibres synapses in area CA3. It was also demonstrated in other brain regions, including the cerebral cortex, but hippocampal LTP remained the dominant "model of activity dependent synaptic plasticity in the mammalian brain (Bliss and Collingridge 1993; Nicoll and Malenka 1995). The Yin of the LTP-Yang is long-term depression (LTD), elicited after specific stimulation protocols in synapses that can or cannot sustain LTP (Bear and Malenka 1994; Lisberger 1998). A dominant model system for investigating LTD and its potential role in learning is the "cerebellum.

Breakthroughs in understanding LTP started to emerge when the analysis shifted from the macroscopic to the microscopic level. This benefited from the use of brain slices, and from a variety of new techniques in cellular physiology, neuropharmacology, molecular biology, and "neurogenetics. In the hippocampus, LTP appears in two major forms, one in which induction is dependent on N-methyl-D-aspartate (NMDA) "receptors (NMDAR, "glutamate), the other on which it does not. The original dentate LTP as well as CA1 LTP are NMDAR dependent, whereas CA3 LTP is NMDAR independent. Here is a partial sketch of the current picture of NMDAR-dependent LTP, with a few comparisons with NMDAR-independent LTP.

1. Induction. LTP is induced by glutamate activation of NMDAR under conditions that remove the magnesium block from the NMDAR-"ion channel ("coincidence detector). This leads to calcium (Ca2+) influx, resulting in activation of Ca2+-/calmodulin-activated "protein kinase (CaMKII) and other kinase systems, and in the modulation of activity of a variety of "intracellular signal transduction cascades. Other types of glutamate receptors, such as the metabotropic receptors, might also be required in the induction "phase. The site of induction is postsynaptic. In contrast, NMDAR-independent LTP is probably induced by presynaptic Ca2+ influx.

2. Maintenance and expression. Short-term changes ('early LTP', less than an hour) involve modification in existing proteins. Longer-lasting LTP ('late LTP') involves modulation in gene expression (Frey et al. 1988; Nguyen et al. 1994; "immediate early genes, "late response genes, "protein synthesis). In both phases, a major part is played by glutamatergic receptors of the a-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid (AMPA) type (APMAR). They are phosphorylated (Barria et al. 1997), and their synthesis (Nayak et al. 1998) and density in the synapse (Hayashi et al. 2000) increase. This results in enhanced AMPAR-mediated transmission. In addition, growth processes possibly contribute to enhanced synaptic efficacy (e.g. Andersen and Soleng 1998). All in all, the long-term processes involve an intricate step-wise dialogue among synaptic and cell-wide mechanisms (Frey and Morris 1997; Dudai and Morris 2000). Although the focus of change in NMDAR-dependent LTP is postsynaptic, presynaptic mechanisms, possibly regulated by a message that travels from the postsynaptic to the presynaptic terminal ('retrograde message', e.g. nitric oxide), also contribute to potentiation. In contrast, in NMDAR-independent LTP, the major site of expression is assumed to be presyn-aptic, involving enhanced transmitter release, which is induced by activation of intracellular signal transduc-tion cascades such as the cAMP cascade (Nicoll and Malenka 1995).

The aforementioned sketch does no justice to the potential complexity of LTP (to tell the truth, discovering that a molecule is involved in LTP is now the rule rather than the exception; Sanes and Lichtman 1999).

Still, it conveys the flavour of the core mechanisms involved. The $64000 question (inflation notwithstanding, Stevens 1998) is whether LTP implements memory. The phenomenology of LTP does look attractive. In addition to the initiation by a brief stimulus and to poststimulus persistence, other properties of LTP are frequently taken as evidence that it is indeed related to learning. These properties include: (a) input specificity (LTP is restricted to the conditioning path; Andersen et al. 1977; Lynch et al. 1977); (b) cooperativity (LTP has a stimulus intensity threshold, below which only PTP may develop, and above which LTP is a function of the number of activated fibres; McNaughton et al. 1978); and (c) associativity (activation of adjacent, convergent afferents can yield greater LTP in one of these afferents, and, furthermore, a weak input, incapable of sustaining LTP, can sustain it if activated concurrently with a strong stimulus to another, convergent path; Levy and Steward 1979; Gustafsson and Wigstrom 1986). Note, however, that the similarity argument suffers from the "homunculus fallacy: it assumes that within the brain resides a miniature creature, the long-term potentiated synapse, which displays the properties of the behaving organism, whereas "reduction does not entail that the part displays the properties of the whole. We must look for better arguments. Those were supplied by a combination of "methods. Here is a sample:

1. Correlation. Auditory "fear conditioning, which is subserved by the "amygdala, alters the auditory evoked responses in the amygdala in the same way as LTP induction. The change parallels the acquisition of the fear behaviour, and does not occur if the tone and the shock remain unpaired (Rogan et al. 1997).

2. Perturbation. This is a popular approach. Drugs that block NMDAR also block certain types of learning, including "maze tasks that depend on hippocampal function (Morris et al. 1986). Some versions of these tasks are, however, unaffected (Bannerman et al. 1995). A few mutations that impair LTP impair learning (Silva et al. 1992), others do not (Zamanillo et al. 1999). The pharmacological and genetic data can be used to show that certain cellular components and mechanisms are shared by LTP and learning, but not that both are the same. A different type of interventional approach is based on the prediction that if learning is subserved by LTP, driving all the synapses to their maximal LTP might saturate the "capacity of the system and block the ability to acquire new information. Saturation of hippocam-pal LTP in the rat was indeed found to impair maze learning (E.I. Moser et al. 1998). The inverse is also true: training rats on a reach-and-grasp motor task results in reduced ability of the neurons in the motor cortex to sustain LTP, as if learning has exploited a part of the LTP capacity in that region (Rioult-Pedotti et al. 2000). 3. Modelling. The concept of LTP is useful in some models that mimic the performance of brain circuits (e.g. Mehta et al. 2000).

So what are the conclusions? It is heated debate. The participants are divided into three congregations: those that adhere to St Anselm's motto, credo ut intelligam (Anselm ~ 1100), 'unless I believe I shall not understand'; the opposing atheists, admittedly a minority; and in between, those that do not feel that questioning the role of LTP in learning is blasphemy. All in all, the

10ms

20 ms

20 ms

Fig. 42 A highly simplified scheme of the cellular mechanisms of LTP in 'hippocampal area CA1. In the absence of an LTP induction stimulus (a), excitatory 'neurotransmission is mediated via two major types of 'glutamate 'receptors: AMPA (AMPAR) and metabotropic (mGluR) receptors. The first is an 'ion channel preferentially permeable to sodium (small open circles), the latter is linked to 'intracellular

20 ms

Fig. 42 A highly simplified scheme of the cellular mechanisms of LTP in 'hippocampal area CA1. In the absence of an LTP induction stimulus (a), excitatory 'neurotransmission is mediated via two major types of 'glutamate 'receptors: AMPA (AMPAR) and metabotropic (mGluR) receptors. The first is an 'ion channel preferentially permeable to sodium (small open circles), the latter is linked to 'intracellular

Fig. 43 Long-term potentiation in the popularity of LTP. The graph depicts the number of papers that mentioned either 'long-term-potentiation' or 'LTP' in their title, abstract or keywords, per year, in the period 1989-99 (Full circles). The percentage of these papers of all the papers listed in the Science Citation Index throughout this period is also presented (open circles). Both plots show a more than an order of magnitude increase in the popularity of LTP over that decade, with an almost step function increase in popularity between 1990 and 1991. Note, however, the plateau in the last years in the graph, which appears also to be retained in the year 2000 (data not shown). This indicates that the interest in LTP, or the capacity of the neuroscience community to deal with it, may have reached at least a temporary saturation. About a third of all the papers that have mentioned LTP throughout the above decade also referred specifically to learning and memory. See also *zeitgeist. (Compiled from the Science Citation Index Expanded, Web of Science V. 4.1, ©ISI, Institute for Scientific Information.)

signal transduction cascades. A third type of glutamatergic receptor, the NMDA receptor channel (NMDAR), is a 'calcium (Ca2+) channel, blocked under resting conditions by magnesium (closed circle). Induction of LTP (b) involves removal of the magnesium block, resulting in an NMDAR-mediated Ca2+ influx (large open circles). Ca2+ activates, either directly or indirectly, a number of signal transduction cascades, involving a number of 'protein kinases. A key role is played by the Ca2+-dependent kinase CaMKII. This leads to phosphorylation and activation of AMPAR, and, furthermore, to translocation of new functional AMPAR molecules into the synapse. A retrograde message (nitric oxide, NO?) could modify presynaptic activity. The potentiation involves additional processes, including, in the case of long-lasting ('late') LTP, modulation of gene expression (see 'immediate early genes, 'late response genes, 'protein synthesis). Proliferation of synaptic contacts may also ensue ('development, not shown for simplicity)^ stimulus arriving at the presynaptic terminal of the potentiated synapse (c), releases glutamate again, but now the transmitter (whose release may be augmented because of the aforementioned presynaptic modifications) encounters additional AMPAR, resulting in marked synaptic facilitation (compare the synaptic response, inset in a and c, respectively). (Based on Nicoll and Malenka 1995; Hayashi et al. 2000.)

jury is yet out (e.g. Shors and Matzel 1997). As it is an umbrella term, the simplistic question 'is LTP = memory' is meaningless; one must specify exactly which LTP. In each system, the relevance of the phenomenon to learning should be assessed on the basis of physiological and behavioural data that are relevant to this same system. Next, one should consider the possible role of LTP in learning at the computational, algorithmic, and implementational levels. The computational level is a tricky issue, because in spite of the intuitive idea that a stronger synapse means stronger memory, the ultimate contribution of enhanced synaptic efficacy to the representational properties of circuits is far from simple (e.g. Markram and Tsodyks 1996). An example of a role at the algorithmic level is the AND gate function provided by NMDAR-dependent LTP. And as to implementation, similar algorithms may be implemented in different neurons by different receptors and signalling cascades. In most cases we are still ignorant as to what the crucial parameters of intracellular molecular networks are, and what they actually represent (e.g. Barkai and Leibler 1997). Nevertheless, it is safe to conclude that multiple receptors and signalling cascades are shared by LTP and other processes of use-dependent synaptic plasticity that are not LTP (e.g. "Aplysia, "development; Constantine-Paton and Cline 1998). In other words, LTP itself may not be learning, but it surely unveils cellular mechanisms that are used in learning.

Selected associations: Associative learning, Model, Plasticity, Synapse

1At the time of LTP discovery, the probable role of the hippocampus in learning was not on the mind of most cellular physiologists. The hippocampus was initially chosen because it is a convenient preparation for investigating cellular physiology (Andersen, personal communication).

0 0

Post a comment