for a significant part of the amplitude of this reflex, is the monosynaptic component consisting of the sensory neurons that innervate the siphon skin, the motor neurons that innervate the gill and the siphon, and the connections between them. The siphon is innervated by about 24 sensory cells, whose cell bodies are located in the abdominal ganglion (Byrne et al. 1974). In addition to their connections with motor neurons that produce the withdrawal reflex, the sensory neurons also make excitatory synaptic connections with excitatory and inhibitory interneurons. We will focus exclusively on the monosynaptic portion of the reflex circuit, consisting of the sensory neurons, the motor neurons, and their connections (for a discussion of these parallel changes, see Frost and Kandel 1984; Frost et al. 1985).

Physiological analyses of short- and long-term changes in the sensory neurons, in response to short- and long-term sensitization training, illustrate five changes in the sensory neuron during long-term facilitation that resemble those found in short-term facilitation. First, the change occurs at the same loci, the connections between the sensory neurons and the motor neurons (Frost et al. 1985). Second, in both cases it involves a strengthening of the connection (Frost et al. 1985). Third, in each case the strengthening is due to an enhancement of transmitter release by an increase in number of transmitter quanta released per impulse (Dale et al. 1988). There is, as far as we can detect, no change in the sensitivity of postsynaptic receptors. Fourth, the enhancement of transmitter release is accompanied, in both cases, by an increase in excitability associated with a depression in outward S-K+ current (Dale et al. 1987; Scholz and Byrne 1987). Finally, in each case the change can be produced either by a repeated presentation of the same facilitating transmitter (5-HT) or by prolonged exposure to the same (cAMP) second messenger (Montarolo et al. 1986; Schacher et al. 1988) that contributes to the short-term process.

On a cellular level, therefore, long-term and short-term memory do not appear to involve two fundamentally different mechanisms but rather resemble a mechanism that varies in duration. How is the short-term process prolonged? How is it maintained?

What Underlies the Similarities Between Short- and
Long-Term Facilitation?

At first glance, it would appear that the same phosphorylation machinery involved in setting up the short-term facilitation may also be involved in setting up the long-term facilitation. Since the short-term effect involves cyclic AMP-dependent protein phosphorylation, Sweatt and Kandel developed an in vivo phosphorylation assay to examine the possibility that the phosphorylation present in the short term, following a single exposure to 5-HT, can be maintained after repeated exposure to 5-HT.

To obtain a profile of the substrate proteins phosphorylated by serotonin and cyclic AMP, Sweatt and Kandel used quantitative two-dimensional gels. By preincubating the sensory neurons with 32P, Sweatt and Kandel found that one brief exposure to serotonin or to cyclic AMP analog leads to an increased level of phosphorylation of 17 specific substrate proteins. This phosphorylation exhibits several properties characteristic of the short-term cellular and behavioral changes. First, it is transient and is not detected 24 hours later. Second, this transient phosphorylation does not require new protein synthesis, as evidenced by the fact that it is not affected by the presence of anisomycin.

By contrast, five repeated pulses of serotonin or exposure for 11/2 to 2 hours to either serotonin or cyclic AMP analog leads to a persistent increase in phosphorylation of the same 17 substrate proteins. This pattern of phosphorylation is similar to that for the short term but has the three characteristics of long-term cellular and behavioral change. First, the increase is maintained for at least 24 hours beyond washout of the 5-HT or cAMP analog. Second, unlike the transient phosphorylation, this persistent phosphorylation is specifically blocked by translational inhibitors. Third, the long-term change is also blocked by inhibitors of transcription. These results suggest that long-term memory may resemble short-term memory in part because repeated exposure to 5-HT and cAMP leads to a persistent increase in phosphorylation of the same substrate proteins involved in the short term. But these long-term effects differ from the short-term changes in not being simply dependent on covalent modifications of preexisting proteins. Rather, the long-term changes are translationally and transcriptionally dependent.

What underlies the persistent increase in protein phosphorylation? Does it represent a depression of a phosphatase or enhanced activity of a kinase? If so, what kinase is involved? The fact that the long-term change leads to phosphorylation of the same substrates involved in the short term suggests that the cAMP-dependent kinase might also be involved in the long term. Moreover, recent work by Greenberg, Castellucci, Bayley, and Schwartz (1987) suggest that prolonged exposure to cAMP could lead to a reduction in the level of the regulatory subunits of the cAMP-dependent protein kinase. Our data suggest that this reduction of the regulatory subunit could reflect either the induction of a specific protease or down-regulation of the regulatory subunit. Either mechanism could make the operation of the catalytic subunit of cAMP-dependent protein kinase autonomous and independent of 5-HT receptor activation.

Long-Term Memory Differs From Short-Term Memory in Two Ways

In addition to the important similarities between short- and long-term effects described above, we found that the neuronal changes associated with

long-term sensitization also involve two fundamental differences from shortterm sensitization. First, the long-term changes require the synthesis of new proteins. Second, the long-term change involves a growth process.

New Protein Synthesis

To determine whether short- and long-term memory can be distinguished on a cellular level, we examined whether long-term memory in this invertebrate synapse requires new protein synthesis (Montarolo et al. 1986). We found a clear separation between these two forms of memory. Inhibitors of protein synthesis (anisomycin and emetine) and of RNA synthesis (actinomycin D and alpha-amanitin) selectively blocked the long-term facilitation of the synaptic connection measured 24 hours after five applications of 5-HT, without in any way interfering with the short-term facilitation that results from a single 5-HT exposure. Most striking is the finding that this single synapse shows a time window in its requirement for protein and mRNA synthesis for long-term facilitation similar to that seen in vertebrates (Davis and Squire 1984). The long-term facilitation induced by 5-HT requires protein synthesis only during the time window in which 5-HT is applied. Inhibiting protein and RNA synthesis after the period of 5-HT application fails to block long-term facilitation.

These findings suggest that the sensitive time window evident in vertebrate (and perhaps even in human) learning and memory can be demonstrated on an elementary synaptic level and reflects a specific neuronal process involved in the storage of long-term memory. Presumably in vertebrates, as in invertebrates, long-term memory depends on the synthesis of protein and RNA, during or briefly after learning, in specific neurons whose synaptic connections are responsible for memory. From a molecular perspective, these studies indicate that long-term sensitization, which lasts more than 24 hours, requires gene products that are not required for short-term memory. Whereas the gene products required for short-term memory are preexisting and must be turned over relatively slowly, some of the gene products required for long-term memory must be newly synthesized. Thus, reduction of a characteristic of memory found in vertebrates (and perhaps in humans) to a single learningrelated synapse in culture should make it possible, at least in principle, to examine an aspect of long-term memory in terms of the specific gene products and the molecular mechanisms critical for acquisition and long-term retention of information.

Neuronal Growth

The morphological studies of Bailey and Chen (1983, 1988) indicated that this long-term alteration in synaptic strength involves significant structural

changes in the sensory neurons. Using horseradish peroxidase (HRP) to visualize the terminals of the sensory neurons, they analyzed the changes in the number of synaptic varicosities, the number and size of the active zones, and the distribution of the synaptic vesicles. These studies led to three significant findings. First, Bailey and Chen (1988) found that, following long-term sensitization, the number of varicose expansions doubled, from an average 1300 varicose expansions per sensory neuron in control animals to about 2600 in sensory neurons in sensitized animals. Second, they found that, in sensitized animals, a larger percentage of varicosities had an active zone (Bailey and Chen 1983). The mean ratio of active zones to varicosities increased from 41 percent in control animals to 65 percent in long-term sensitized animals. Third, there was an increase in the surface area of the active zones and in the total number of vesicles associated with each release site.

These morphological changes could represent an anatomical substrate for memory consolidation. The findings of Bailey and Chen also suggest that varicosities and active zones are not immutable structures and that learning may modulate these sites to alter synaptic effectiveness. (For parallel studies in mammals, see Greenough 1984.) Finally, the evidence that long-term memory involves a significant growth process provides a rationale for thinking about the possible role of macromolecular synthesis in long-term memory.

How Does the Structural Change Come About?

To determine whether synaptic changes that occur during learning and development share common mechanisms and whether they require interaction between the presynaptic cell and its postsynaptic target, Glanzman, Kandel, and Schacher applied to sensory neurons of Aplysia in dissociated cell culture the low-light video microscopy method recently developed by Kater and Hadley (1982) and Purves and Hadley (1985). This methodology permits the visualization of the structure of a living neuron repeatedly over the course of days. Glanzman et al. found that sensory neurons cultured alone are significantly less complex structurally than sensory neurons that are cocultured, and have formed synaptic connections, with an L7 motor cell. Thus, sensory neurons cultured alone (or with inappropriate target cells) have fewer processes and varicosities than sensory neurons cocultured with L7 motor cells. These data suggest that some interaction with the target motor neuron - an interaction related to synaptic contact - may regulate the growth of sensory neurons. Moreover, Glanzman et al. found, for sensory neurons cocultured with L7 cells, a good correlation between the number of varicosities on a sensory neuron and the strength of its synaptic connection with the motor neuron. This correlation suggests that the varicosities, par

ticularly those associated with the initial segment of the motor neuron's major neurite, represent sites of synaptic release, a suggestion supported by preliminary studies in which the same sensorimotor cultures have been inspected first with fluorescence video microscopy and then with electron microscopy. Glanzman, Kandel, and Schacher have now begun experiments with video microscopy to determine whether repeated or prolonged application of serotonin, which produces long-term facilitation of the in vitro sensorimotor synapse, can also produce an increase in varicosities and processes on sensory neurons. Preliminary results in culture suggest that repeated or long-term exposure to 5-HT can produce structural changes in sensory neurons cocultured with L7 cells. Serotonin treatment produces an increase in varicosities and processes that contact the initial segment of L7's major neurite. Interestingly, structural changes in the sensory neurons also appear to require the presence of the postsynaptic cell, since the serotonin treatment does not appear to cause growth of sensory neurons cultured alone. It will now be important to explore the mechanisms whereby the postsynaptic target neuron communicates with the presynaptic cell and how it induces and directs the presynaptic cell's growth.

The Search for Proteins Important for Long-Term
Facilitation

Since, in both the intact animal and the dissociated cell culture, long-term memory for sensitization is accompanied by neuronal growth (Bailey and Chen 1985) and is blocked by inhibitors of protein synthesis (Montarolo et al. 1986), we searched for proteins whose rates of synthesis were altered during both the acquisition and the maintenance phase of long-term memory for sensitization by using computer-assisted quantitative two-dimensional gel analysis (Protein Databases, Inc., NY). Castellucci, Kennedy, Kandel, and Goelet (1988) focused on the maintenance phase and identified four learning-associated proteins whose rates of synthesis were altered 24 hours after either 1 day or 4 days of training. By mixing these labeled samples with protein extracts from the total nervous system, Castellucci et al. found, by silver staining, that three of these proteins are present at levels that may permit their isolation and characterization in molecular terms. Indeed, Kennedy, Kandel, and Sweatt succeeded in isolating one of these proteins from preparative 2-D gels and were able to determine 17 residues of amino acid sequence for a proteolytic fragment. Based on this sequence, they synthesized a set of oligonucleotides and are screening an Aplysia cDNA library in collaboration with Michael Knapp.

Recently, Barzilai, Sweatt, Kennedy, and Kandel began to also focus on the initiation phase of long-term memory and have identified proteins whose