Technical Comments

Learning and the Sensorimotor Synapse in Aplysia

Science  31 Jul 1998:
Vol. 281, Issue 5377, pp. 619
DOI: 10.1126/science.281.5377.619a

One of the key questions in neuroscience is the neuronal basis of learning and memory. Several neuronal mechanisms have been described that could plausibly contribute to learning and memory in mammals, but it has been difficult to relate them directly to behavior. The gill- and siphon-withdrawal reflex of the marine snail Aplysia is advantageous for such studies because the neuronal circuit for the reflex is partly monosynaptic, and the reflex has been shown to undergo several simple forms of learning, including classical conditioning (1). Moreover, a cellular analog of the conditioning has been demonstrated at synapses between identified sensory and motor neurons in the isolated nervous system (2). This analog has been thought to be a result of activity-dependent enhancement of presynaptic facilitation of the sensory neurons by modulatory transmitters, including serotonin (2, 3). Recent evidence suggests, however, that the analog may also be a result, in part, of Hebbian long-term potentiation (LTP) similar to that observed in mammalian hippocampus (4). The sensory neurons have been shown to use glutamate as a transmitter that acts onN-methyl-d-aspartate (NMDA)–like receptors on the motor neurons (5), and sensorimotor neuron synapses in isolated cell culture undergo Hebbian LTP that is blocked by the NMDA receptor blocker dl-2-amino-5-phosphonovalerate (APV) (6). Murphy and Glanzman (7) report that APV also reduces the cellular analog of conditioning at sensorimotor neuron synapses in the isolated nervous system, suggesting that LTP is involved. Their experimental design, however, appears not to support strong conclusions about the relative contributions of Hebbian LTP and activity-dependent presynaptic facilitation by a modulatory transmitter.

The conditioning analog includes both associative effects of paired training and nonassociative effects of unpaired training (2). If APV selectively eliminated the associative effect, as proposed by Murphy and Glanzman (7), then one would predict that it should reduce the facilitation by paired training to the level of facilitation by unpaired training. Murphy and Glanzman (7) report that when tested 15 min after training, facilitation by paired training was reduced by perfusion with APV during training, and facilitation by unpaired training was not reduced by APV, consistent with this prediction. However, in the control experiments without APV, there was no difference between paired and unpaired training (that is, there was no associative effect). According to Murphy and Glanzman's hypothesis, these statements cannot be simultaneously correct. The results 60 min after training were more internally consistent, but the data from 15-min after training illustrate the problem.

The most likely explanation of these results is that some or all of the experiments with paired training appear to have been run at different times (even different years) than the experiments with unpaired training. In a 1995 abstract, Murphy and Glanzman stated that facilitation by paired training is reduced by APV (8), and in a 1996 abstract, they stated that facilitation by unpaired training is not blocked by APV (9). In their report (7), they presented the results on paired and unpaired training in separate figures [figures 1 and 2 in (7)]. Moreover, they analyzed the results on paired training with parametric statistics (ANOVA) and the results on unpaired training with nonparametric statistics (Kruskal-Wallis ANOVA), because of differences in homogeneity of variance. The use of different types of statistical tests would not have been appropriate if the experiments on paired and unpaired training were run at the same time.

The problem with running some or all of the experiments on paired and unpaired training at different times is that the nonassociative effects of unpaired training may fluctuate dramatically from week to week and month to month in wild-caught animals such as Aplysia. Thus, a likely explanation of the 15-min data in the report (7) is that the nonassociative effect was stronger when the unpaired experiments were run than when the paired experiments were run. One could not then separate the associative effects of paired training from the nonassociative effects of unpaired training, and thus one could not tell to what extent APV reduced the associative effect.

Murphy and Glanzman imply that the associative effect is largely a result of Hebbian LTP. However, activity-dependent presynaptic facilitation could contribute to the cellular analog of conditioning (2, 3). For example, Abrams and Galun (10) have found that injection of an inhibitor of cAMP-dependent protein kinase into the sensory neuron, which blocks the presynaptic effects of serotonin, also blocks the cellular analog of conditioning. It is also possible that Hebbian postsynaptic mechanisms and activity-dependent presynaptic facilitation interact in some way, because Bao et al. (11) and Schacher et al. (12) have found that manipulations that block Hebbian LTP also block activity-dependent presynaptic facilitation by serotonin in isolated cell culture. Disentangling the relative contributions of these different mechanisms and their possible interactions will require experiments designed to distinguish between the associative and nonassociative effects of training.


Learning and the Sensorimotor Synapse in Aplysia

Response: We reported (1) that the specific NMDA receptor antagonist APV blocks associative enhancement of sensorimotor synapses in a cellular analog of classical conditioning of the siphon-withdrawal reflex of Aplysia (2). Earlier, we found that this cellular analog of conditioning is also blocked by the presence in the postsynaptic motor neuron of the rapid calcium chelator BAPTA (3). These two studies extend earlier in vitro results from our laboratory, which indicated that sensorimotor synapses in culture exhibit a form of LTP that resembles LTP of CA1 hippocampal synapses (4). Like LTP of CA1 synapses, LTP of sensorimotor synapses depends on postsynaptic depolarization, a rise in postsynaptic calcium, and activation of NMDA-type receptors, because it can be blocked by postsynaptic hyperpolarization, by postsynaptic BAPTA, or by APV (5, 6). Furthermore, LTP of sensorimotor synapses can be induced in Hebbian fashion, by pairing brief stimulation of a presynaptic sensory neuron with strong depolarization of a postsynaptic motor neuron (6,7). On the basis of the similarities between the cellular analog of conditioning and LTP—and also because paired training in classical conditioning of the withdrawal reflex, like Hebbian LTP of sensorimotor synapses, involves the conjoint activation of sensory and motor neurons (8)—we concluded (1) that LTP contributes to classical conditioning of the withdrawal reflex (9).

In our report (1), we asked two questions: (i) Does APV disrupt associative synaptic enhancement in a cellular analog of classical conditioning? (ii) Does APV disrupt nonassociative, sensitization-related synaptic enhancement? In answer to the first, we found that APV blocked conditioning-related synaptic enhancement [figure 1C in (1)]; in answer to the second, we found that APV did not affect sensitization-related synaptic enhancement [figure 2C in (1)]. These are two separate experimental questions, and it was therefore appropriate to carry out the experiments that address these questions separately (10). Our experimental design would be problematic only if, as Hawkins states, there were significant variability in the nonassociative effects of training across the two sets of experiments; such variability would make it difficult to interpret the effect of APV as being strictly disruption of associative enhancement. In our report, the CS+-APV EPSP at 15 min was somewhat less than the CS EPSP [figures 1C and 2C in (1)]. Moreover, we found no difference between the CS+ and CS results at 15 min, contrary to what one would expect if there were an associative effect of paired training [figure 2D in (1)]. The 60-min results, in contrast, are consistent with our interpretation of the data, which Hawkins does not dispute. Thus, the amplitudes of the CS+-APV EPSP and CS EPSP are almost identical at 60 min. Also, the CS+ EPSP is significantly larger than the CS EPSP at 60 min, which indicates that the paired training produced an associative effect at 60 min, although not at 15 min, posttraining.

Hawkins' critique rests on his interpretation of our 15-min results, which appear to conflict with results from an earlier study by Hawkins and his colleagues (11; see also 12). That study, unlike ours, used a differential conditioning paradigm in which paired and unpaired training were delivered to different synapses made by two presynaptic sensory neurons with the same target motor neuron. (In our experiments only a single synapse was trained per preparation.) In the experiments by Hawkins et al., stimulation of one presynaptic sensory neuron was paired with tail nerve (or tail) shock (the US), while stimulation of the other presynaptic neuron was unpaired with the US. (In some experiments the nonassociative condition involved a “US alone” protocol, in which one of the sensory neurons received no stimulation during the training.) Hawkins et al. observed a significant difference between the paired and unpaired data 5 to 15 minutes after training [figure 1C in (11)].

Why did our 15-min results (1) differ from those of Hawkins and colleagues (11)? In part to answer this question, we have recently carried out a series of conditioning experiments involving differential training. The results indicate that the absence of an associative effect in our 15-min data is related to the use of a nondifferential training procedure. We have found a significant difference between associative and nonassociative effects at both 15 min (Fig. 1) and 60 min after differential training (13). Why the results of nondifferential and differential conditioning should differ at 15 min after training is unclear. Possibly differential training—because it involves two synapses on the same postsynaptic target cell—reveals a competitive interaction between the synapses that is not apparent in the nondifferential paradigm, in which only a single synapse is trained (14, 15). In any case, our new experiments provide an alternative explanation for the 15-min data of our report (1), one that is consistent with our interpretation of APV's effect.

Figure 1

Comparison of the effects of nondifferential and differential conditioning 15 min after the end of training. Nondifferential data are from our report (1). Differential data are from experiments (n = 7) that used two sensory neuron connections with the same siphon motor neuron (13). One connection received paired training, while the other received unpaired training. In the unpaired condition the CS and US were separated by 2.5 min. Otherwise, the experimental methods were like those used in our report (1). Difference between the CS+ and CS EPSPs is statistically significant for the differential training (P < 0.02), but not for the nondifferential training (P > 0.05) (unpaired ttests).

One must ask whether Hawkins's interpretation of the data provides a strong explanation for our APV results. Variability in the nonassociative effects of training could account for the effect of APV in our experiments only if (i) APV disrupted the nonassociative component of conditioning when it was “weak” (paired experiments), but not when it was “strong” (unpaired experiments); and if (ii) APV had little or no effect on the associative component of conditioning (because otherwise the CS+-APV EPSP should have been even smaller than it was) [see figure 1C in (1)]. Besides being ad hoc, such an argument would not account for our previous finding that APV blocks associative LTP of sensorimotor synapses in vitro (6). Moreover, this argument is not strongly supported by our data. A posthoc analysis of our 15-min data indicates that the difference between the CS and CS+-APV groups is not statistically significant ( P > 0.1, unpaired t test), whereas the difference between the CS+ and CS+-APV groups is statistically significant [ P < 0.05, unpaired t test; see also note 18 of (1)].

The results of our recent differential conditioning experiments make a strong case. If the effect of APV in the paired experiments of our study (1) were a result, in part, of fluctuation in the nonassociative effects of training, then one would predict that APV's disruptive effect would disappear, or at least significantly diminish, if a differential conditioning paradigm were utilized. But this prediction fails. As we have reported (16), APV blocks the associative component of differential conditioning, but does not effect the nonassociative component.

Although our data indicate that classical conditioning of the withdrawal reflex in Aplysia involves Hebbian LTP, we agree with Hawkins that other processes could contribute to this form of learning. The model of associative enhancement of sensorimotor synapses in figure 3 of our report (1) includes a role for facilitatory interneurons, and in the legend to figure 3 we specifically state that facilitatory transmitters, such as serotonin, might interact with LTP during classical conditioning.

The conclusion that classical conditioning in Aplysia is mediated, in part, by Hebbian LTP was unexpected because an earlier study by Hawkins and his colleagues (18) specifically ruled out the involvement of Hebbian modulation of sensorimotor synapses. How can the data from this earlier study be reconciled with the results from our laboratory? We (3, 6, 17) have previously suggested that the intrasomal injections of current used in the 1984 study were insufficient to significantly alter the membrane potential of the siphon motor neurons at postsynaptic sites in the central nervous system. Hawkins now appears to accept this explanation for his group not detecting a postsynaptic contribution to classical conditioning in Aplysia (19). Hawkins and his colleagues have recently confirmed our findings (3, 5,6) that associative enhancement of sensorimotor synapses is blocked by postsynaptic hyperpolarization and postsynaptic BAPTA (19).

Recently, two papers appeared that presented evidence that LTP mediates fear conditioning in rats (20). Anthropomorphic considerations aside, one can view classical conditioning of the withdrawal reflex in Aplysia as a type of fear conditioning. It is therefore intriguing that the synapses that appear to undergo LTP during fear conditioning in rats are synapses that transmit CS information. The parallels between the invertebrate and vertebrate data are striking. They suggest the possibility that NMDA receptor-dependent LTP of CS pathways may be a general feature of fear conditioning (see also 21).


Navigate This Article