Use-dependent changes in synaptic strength underlie many forms of learning
Arguably the greatest achievement of a brain is its ability to learn and to store the experience and events of the past so that it is better adapted to deal with the future. Memory is the ability to store and to recall learned changes, and nervous systems without memory are extremely handicapped. Although the biological bases for learning and memory are far from understood, certain principles have become clear. First, no single mechanism can explain all forms of memory. Even within a single organism, a variety of types of memory exist—and a variety of mechanisms underlie them. Second, evidence is strong that synapses are the physical site of many if not most forms of memory storage in the brain. As the major points of interaction between neurons, synapses are well placed to alter the processing capabilities of a neural circuit in interesting and useful ways. Third, the synaptic strength (i.e., the mean amplitude of the postsynaptic response) of many synapses may depend on their previous activity. The sensitivity of a synapse to its past activity can lead to a long-term change in its future effectiveness, which is all we need to build memory into a neural circuit.
Some forms of memory last just a few seconds or minutes, only to be lost or replaced by new memories. Working memory is an example. It is the continual series of fleeting memories that we use during the course of a day to remember facts and events, what was just spoken to us, where we put the phone down, whether we are coming or going—things that are useful for the moment but need not be stored longer. Other forms of memory may last for hours to decades and strongly resist disruption and replacement. Such long-term memory allows the accumulation of knowledge over a lifetime. Some memories may be formed after only a single trial (recall a particularly dramatic but unique event in your life), whereas others form only with repeated practice (examples include speaking a language or playing a guitar). Detailed descriptions of the many types of memory are beyond the scope of this chapter, but it seems obvious that no single synaptic mechanism would suffice to generate all of them. Neurophysiologists have identified many types of synaptic plasticity, the term for activity-dependent changes in the effectiveness of synapses, and some of their mechanisms are well understood. However, it has been very difficult to demonstrate that specific forms of memory use particular types of synaptic plasticity, and correlation of memory with synaptic plasticity remains a coveted goal of current research.
Short-term synaptic plasticity usually reflects presynaptic changes
Repetitive stimulation of neuronal synapses often yields brief periods of increased or decreased synaptic strength (Fig. 13-19). The usual nomenclature for the short-term increases in strength is facilitation (which lasts tens to hundreds of milliseconds), augmentation (which lasts several seconds), and post-tetanic potentiation (which lasts tens of seconds to several minutes and outlasts the period of high-frequency stimulation). Not all of them are expressed at every type of synapse. In general, the longer-lasting modifications require longer periods of conditioning stimuli. Short-term decreases in synaptic strength include depression, which can occur during high-frequency stimulation, and habituation, which is a slowly progressing decrease that occurs during relatively low-frequency activation.
FIGURE 13-19 Facilitation, potentiation, depression, and habituation. (Data from Levitan IB, Kaczmarek LK: The Neuron: Cell and Molecular Biology, 2nd ed. New York, Oxford University Press, 1997.)
Three potential explanations may be offered for short-term increases in synaptic strength. First, the presynaptic terminal may release more transmitter in response to each action potential. Second, the postsynaptic receptors may be more responsive to transmitter because of a change in their number or sensitivity. Third, both of these changes may occur simultaneously. Studies involving a variety of synapses suggest that the first explanation is most often correct. In these cases, quantal analysis usually shows that synapses become stronger because more neurotransmitter is released during each presynaptic action potential; postsynaptic mechanisms generally do not play a role. This form of plasticity seems to depend on the influx of presynaptic Ca2+ during the conditioning tetanus. N13-5 Katz and Miledi first proposed that synaptic strength is increased because of residual Ca2+ left in the terminal after a conditioning train of stimuli at the neuromuscular junction. Recent work supports this hypothesis. The idea is that (1) a tetanic stimulus leads to a substantial increase in presynaptic [Ca2+]i that saturates intracellular Ca2+ buffers; (2) the high presynaptic [Ca2+]i takes a relatively long time to decline to baseline, and prolonged stimulation requires prolonged recovery times; and (3) presynaptic action potentials arriving after the conditioning tetanus generate a Ca2+ influx that sums with the residual [Ca2+]i from the preceding tetanus to yield a larger than normal peak [Ca2+]i. Because the dependence of transmitter release on presynaptic [Ca2+]i is highly nonlinear, the increase in release after conditioning stimuli can be large. Several types of Ca2+-sensitive proteins are present in the presynaptic terminal; the Ca2+ binding site on synaptotagmin triggers exocytosis, whereas Ca2+ binding sites on other proteins—perhaps including protein kinase C (see pp. 60–61) and Ca2+-calmodulin–dependent protein kinases (CaMKs; see p. 60) regulate short-term increases in transmitter release.
Short-Term Synaptic Plasticity
Contributed by Emile Boulpaep, Walter Boron
See the following:
1. N8-5 Quantal Nature of Transmitter Release
2. N8-6 Sir Bernard Katz
3. N8-7 Modulation of Quantal Release
Several mechanisms cause short-term decreases in synaptic strength, including the depletion of vesicles and the inactivation of presynaptic Ca2+ channels. Habituation has been studied in the marine invertebrate Aplysia. The animal reflexively withdraws its gill in response to a stimulus to its skin. Vincent Castellucci and Eric Kandel N13-6 found that withdrawal becomes less vigorous—that is, the animal habituates—when the stimulus is presented repeatedly. The basis for this behavioral habituation is, at least in part, a decrease in the strength of synapses made by skin sensory neurons onto gill-withdrawal motor neurons. Using quantal analysis, Castellucci and Kandel showed that synaptic habituation is due to fewer transmitter quanta being released per action potential. Thus, as with the short-term enhancements of synaptic strength, this example of habituation is due to presynaptic modifications.
Eric R. Kandel
For more information about Eric R. Kandel and the work that led to his Nobel Prize, visit http://www.nobel.se/medicine/laureates/2000/index.html (accessed October 2014).
Long-term potentiation in the hippocampus may last for days or weeks
In 1973, Timothy Bliss and Terje Lømo described a form of synaptic enhancement that lasted for days or even weeks. This phenomenon, now called long-term potentiation (LTP), occurred in excitatory synapses of the mammalian cerebral cortex. LTP was generated by trains of high-frequency stimulation applied to the presynaptic axons, and it was expressed as an increase in the size of EPSPs. Several properties of LTP, including its longevity and its location in cortical synapses, made it immediately attractive as a candidate for the cellular basis of certain forms of vertebrate learning. Years of intensive research have revealed many details of the molecular mechanisms of LTP. They have also provided some evidence, albeit still indirect, that LTP is involved in some forms of learning. Numerous mechanistically distinct types of LTP exist; here we discuss only the best-studied example.
LTP is easily demonstrated in several synaptic relays within the hippocampus, a part of the cerebral cortex that has often been considered essential for the formation of certain long-term memories. The best studied of these synapses is between the Schaffer collateral axons of CA3 pyramidal neurons (forming the presynaptic terminals) and CA1 pyramidal neurons (the postsynaptic neurons). In a typical experiment, the strength of synapses to the CA1 neuron is tested by giving a single shock about once every 10 seconds. Stimuli are applied separately to two sets of Schaffer collateral axons that form two different sets of synapses (Fig. 13-20A). If we stimulate a “control” Schaffer collateral once every 10 seconds, the amplitude of the EPSPs recorded in the postsynaptic CA1 neuron remains rather constant during many tens of minutes (see Fig. 13-20B). However, if we pair the presynaptic test shocks occurring once every 10 seconds to the “test” pathway with simultaneous postsynaptic depolarization of the CA1 neuron, the amplitude of the EPSP gradually increases several-fold (see Fig. 13-20C), which is indicative of LTP. In this case, the trigger for LTP was the pairing of a low-frequency presynaptic input and a strong postsynaptic depolarization. We already saw that Bliss and Lømo originally induced LTP by activating the presynaptic axons with brief bursts of tetanic stimulation (50 to 100 stimuli at a frequency of ~100 Hz). Both strategies—presynaptic-postsynaptic pairing protocol and tetanic presynaptic stimulation—are effective ways to induce LTP.
FIGURE 13-20 Causing LTP by pairing presynaptic and postsynaptic stimuli. A, Pyramidal CA3 neurons in the hippocampus send axons (Schaffer collaterals) to synapse on pyramidal CA1 neurons. In the case of the “control” stimulus, the stimulating electrode stimulates collaterals that activate one set of synapses on the postsynaptic CA1 neuron. In this case, the CA1 neuron receives only presynaptic stimuli. In the case of the “test” stimulus, a second electrode stimulates a different set of collaterals that activate a different set of synapses on that same CA1 neuron. However, in this case, the presynaptic stimuli will be paired with a postsynaptic depolarization that is delivered by a third microelectrode. Aside from pairing or not pairing the presynaptic stimuli with a postsynaptic stimulus, the test and control pathways are equivalent. The third microelectrode records the EPSPs from the CA1 neuron in both the test and control experiments. B, In this case, the control Schaffer collaterals are stimulated. Each test pulse is represented by a point on the graph. However, because the CA1 neuron is not depolarized, the amplitude of the EPSPs remains constant (i.e., there is no LTP). C, In this case, the test Schaffer collaterals are stimulated. When the CA1 neuron is also depolarized, the amplitude of the EPSPs greatly increases (i.e., LTP has been induced). (Data from Gustafsson B, Wigstrom H, Abraham WC, Huang YY: Long-term potentiation in the hippocampus using depolarizing current pulses as the conditioning stimulus to single-volley synaptic potentials. J Neurosci 7:774–780, 1987.)
The induction of LTP has several interesting features that enhance its candidacy as a memory mechanism. First, it is input specific, which means that only the activated set of synapses onto a particular cell will be potentiated, whereas unactivated synapses to that same neuron remain unpotentiated. Second, induction of LTP requires coincident activity of the presynaptic terminals plus significant depolarization of the postsynaptic membrane. We saw this effect in Figure 13-20C, which showed that inducing LTP required coincident synaptic input and depolarization. Because single hippocampal synapses are quite weak, the requirement for substantial postsynaptic activation means that LTP is best induced in an in vivo situation by cooperativity—enough presynaptic axons must cooperate, or fire coincidentally, to strongly activate the postsynaptic cell. The cooperative property of LTP can be used to form associations between synaptic inputs. Imagine that two sets of weak inputs onto one cell are, by themselves, too weak to induce LTP. Perhaps each encodes some sensory feature of an object: the sight (input 1) and sound (input 2) of your pet cat. If the two firing together are strong enough to induce LTP, both sets of synaptic inputs will tend to strengthen, and the features that they encode (the sight and sound of the cat) will become associated in their enhanced ability to fire the postsynaptic cell. In contrast, for example, the sight of your cat and the sound of your alarm clock will rarely occur together, and their neural equivalents will not become associated.
The molecular mechanisms of one form of LTP in the CA1 region of hippocampus have been partially elucidated. The synapse uses glutamate as its transmitter, and both AMPA and NMDA receptors are activated to generate an EPSP. Induction of this type of LTP depends on an increase in postsynaptic [Ca2+]i levels beyond a critical level and lasting for about 1 to 2 seconds. (Recall that the short-term forms of enhancement required a presynaptic increase in [Ca2+]i; see the preceding section.) Under most conditions, postsynaptic [Ca2+]i levels rise during a tetanic stimulus because of the activation of NMDA receptors, the only type of glutamate-activated channel that is usually permeable to Ca2+ (see Fig. 13-16). Recall also that the NMDA-type glutamate receptor channel is voltage dependent; to open, it requires Vm to be relatively positive. The cooperativity requirement of LTP is really a requirement for the activation of NMDA receptors so that Ca2+ can enter—if the postsynaptic Vm is too negative (as it is when synaptic activation is weak), NMDA channels remain mostly closed. If activation is strong (as it is when multiple inputs cooperate or when tetanus occurs), Vm becomes positive enough to allow NMDA channels to open. The stimulus-induced rise in postsynaptic [Ca2+]i activates at least one essential kinase: CaMKII (see p. 60). Blocking this kinase with drugs prevents induction of LTP. Several other types of kinases are implicated, but the evidence that these mediate—as opposed to modulate—LTP is equivocal.
The molecular pathways leading to the expression and maintenance of LTP are more obscure after the Ca2+-induced activation of CaMKII. Evidence has been presented both for and against postsynaptic and presynaptic changes as explaining the increase in synaptic strength. For the most commonly studied form of LTP described here (i.e., dependent on NMDA receptors), compelling evidence demonstrates a postsynaptic mechanism involving the recruitment of more AMPA receptors to the postsynaptic membrane. Sometimes, postsynaptic AMPA receptors appear to be functionally “silent” until LTP mechanisms activate them or insert them into the membrane. Evidence also suggests a presynaptic mechanism for enhancing transmitter release. This presynaptic hypothesis requires the presence of some unknown, rapidly diffusing retrograde messenger that can carry a signal from the postsynaptic side (where rising [Ca2+]i is clearly a trigger for LTP) back to the presynaptic terminal.
Long-term depression exists in multiple forms
Memory systems may have mechanisms not only to increase synaptic strength but also to decrease it. In fact, long-term depression (LTD) can be induced in the same synapses within the hippocampus that generate the [Ca2+]i-dependent LTP described in the preceding section. The critical feature that determines whether the synapses will strengthen or weaken is simply the frequency of stimulation that they receive. For example, several hundred stimuli delivered at 50 Hz produce LTP, the same number delivered at 10 Hz has little effect, and at 1 Hz they produce LTD. One set of synapses can be strengthened or weakened repeatedly, which suggests that each process (LTP and LTD) acts on the same molecular component of the synapses. LTD induced in this way shows the same input specificity as LTP—only the stimulated synapses onto a cell are depressed.
Multiple forms of LTD differ according to their molecular mechanisms. One type depends on activation of mGluRs; another apparently requires activation of cannabinoid receptors. We describe the type of LTD that has been studied most extensively. The induction requirements of this form of hippocampal LTD are paradoxically similar to those of LTP: LTD induced by low-frequency stimulation depends on the activation of NMDA receptors, and it requires an increase in postsynaptic [Ca2+]i. The key determinant of whether a tetanic stimulus induces LTP or LTD may be the level to which postsynaptic [Ca2+]i rises. Figure 13-21 illustrates a simple model of the induction mechanisms for LTD and LTP. Synaptic activation releases glutamate, which activates NMDA receptors, which in turn allow Ca2+ to enter the postsynaptic cell. In the case of high-frequency stimulation, postsynaptic [Ca2+]i rises to very high levels; if stimulation is of low frequency, the rise in postsynaptic [Ca2+]i is more modest. High levels of [Ca2+]i lead to a net activation of protein kinases, whereas modest levels of [Ca2+]i preferentially activate protein phosphatases, perhaps calcineurin. The kinases and phosphatases in turn act on synaptic proteins or phosphoproteins that somehow regulate synaptic strength.
FIGURE 13-21 Proposed molecular mechanism for LTP and LTD. Glutamate release from the presynaptic terminal activates NMDA receptor channels, which allow Ca2+ to enter the postsynaptic cell. The glutamate also activates AMPA receptor channels. Whether the ultimate effect is LTP or LTD appears to depend on the extent to which [Ca2+]i rises. High levels of [Ca2+]i—produced by high-frequency stimulation—lead to a net activation of protein kinases and thus phosphorylation of one or more synaptic proteins that regulate synaptic strength. One hypothetical pathway has the phosphorylated/dephosphorylated synaptic proteins modulating the AMPA receptor channel. The result is LTP. The postsynaptic neuron may also be able to influence the presynaptic terminal. A moderate increase in [Ca2+]i—produced by low-frequency stimulation—preferentially activates protein phosphatases, which presumably dephosphorylate the same synaptic proteins as in the previous example. The result is LTD.
LTP-inducing stimuli phosphorylate specific residues on AMPA receptors and, conversely, dephosphorylate other residues on AMPA receptors. These post-translational changes are, however, only part of the story of long-term synaptic plasticity. LTP increases—and LTD decreases—the numbers of AMPA receptors in the postsynaptic membrane by modulating receptor trafficking into and out of the surface membrane. Longer lasting LTP- and LTD-induced changes seem to involve mechanisms that depend on protein synthesis, including structural changes in synapses and spines.
The simple scheme in Figure 13-21 leaves unidentified many of the steps between the rise in postsynaptic [Ca2+]i and the change in synaptic strength; most of the molecular details remain to be determined. However, if the model is correct, it means that synaptic strength (and, by implication, some memory) is under the dynamic control of cellular processes that determine postsynaptic [Ca2+]i. Once again in physiology, [Ca2+]i has been assigned a pivotal role in a vital process.
Long-term depression in the cerebellum may be important for motor learning
A variety of other types of LTP and LTD have been described in other synapses and even within the same synapses of the CA1 region in the hippocampus. Clearly, multiple means and mechanisms may be used to strengthen and to weaken synapses in the brain over a range of time courses. We briefly describe one other well-studied type of synaptic modification in the mammalian brain, LTD in the cerebellum.
The cerebellum is a large brain structure that is important in motor control and strongly implicated in motor learning. The cortex of the cerebellum is a thin, multiply folded sheet of cells with an intricate but highly repetitious neural structure. The principal cell of the cerebellum is the Purkinje cell, a large neuron that uses GABA as its transmitter and whose axon forms the sole output of the cerebellar cortex. Purkinje cells receive two types of excitatory synaptic input: (1) each Purkinje cell receives powerful synaptic contact from just a single climbing fiber, which comes from a cell in the inferior olivary nucleus (see Fig. 13-1B), and (2) each Purkinje cell also receives synaptic input from ~150,000 parallel fibers, which originate from the tiny granule cells of the cerebellum itself. This remarkable conjunction of synaptic inputs is the basis for a theory of motor learning that was proposed by David Marr and James Albus in 1970. They predicted that the parallel-fiber synapses should change their strength only if they are active at the same time as the climbing fiber onto the same cell. This idea received important experimental support from the laboratory of Masao Ito.
Ito and colleagues monitored EPSPs in a Purkinje cell while stimulating some of its inputs from parallel fibers and its single input from the climbing fiber. They found that the EPSPs generated by the parallel fibers became smaller when both the parallel fibers and the cell's climbing fiber were coactivated at low frequencies. Stimulating either input alone did not cause any change. Cerebellar LTD can last at least several hours. As with the hippocampal LTP and LTD described above, cerebellar LTD showed input specificity: only those parallel fibers coactivated with the climbing fiber were depressed, whereas others with synapses onto the cell were unchanged.
The mechanism of cerebellar LTD has some similarities to that of LTD in the hippocampus, but it is also distinctly different. Parallel-fiber synapses weaken because of reduced effectiveness—a reduction in either number or sensitivity—of postsynaptic AMPA-type glutamate receptors. As in the hippocampus, induction of cerebellar LTD requires an increase in postsynaptic [Ca2+]i. However, unlike in the hippocampus, no NMDA-type glutamate receptors are present in the mature cerebellum to mediate Ca2+ flux. Instead, Ca2+ can enter Purkinje cells through voltage-gated Ca2+ channels that are opened during the exceptionally powerful EPSP that the climbing fiber generates. In addition to a rise in postsynaptic [Ca2+]i, cerebellar LTD induction seems to require the activation of mGluRs and protein kinase C by the parallel fibers. Increases in postsynaptic [Na+]i and NO have also been implicated in LTD induction. At present, the relationships between these putative induction factors and the molecular pathways leading to the expression of cerebellar LTD are obscure.
As we pointed out for the hippocampus, most efficient memory systems need mechanisms for both weakening and strengthening of their synapses. It turns out that parallel-fiber synapses of the cerebellum can be induced to generate LTP as well as LTD by stimulating them at relatively low frequencies (2 to 8 Hz). Cerebellar LTP, unlike hippocampal LTP, requires presynaptic but not postsynaptic increases in [Ca2+]i. Potentiation seems to be a result of increased transmitter release from the presynaptic terminal.