Previous learning changes the perception of a new learning event

How does prior strong learning enhance new memory formation? Given the strict order of training needed (strong followed by weak), we hypothesized that the strong memory might be changing the animal’s perception of the combination of the CS + US during subsequent weak training, increasing the positive value of the weak training such that the association becomes enhanced. To test this idea, we took advantage of the fact that the perception of a presented stimulus, from positive to negative, can be read out directly in Lymnaea by quantifying the number of ingestion and egestion events that occur (Fig. 2A) (16). During the pairing of the CS + US in the weak training protocol, animals performed a mixture of ingestion and egestion bites, often flip-flopping between bouts of the two behaviors (Fig. 2B). By contrast, when this was preceded by strong training, the ratio of ingestion to egestion was significantly higher (Fig. 2, C and D) and more animals performed no egestion bites at all in response to the weak training (Fig. 2E). Furthermore, to examine perceptual stability during weak training, we compared pairs of bites elicited during the CS + US presentation to determine whether they switched between states (ingest/egest and egest/ingest) or remained stable (ingest/ingest and egest/egest). This revealed a significant reduction in the transition probability between states, suggesting that prior strong training shifts, and stabilizes, the perception of the training (fig. S4, A to E). Presentation of the strong training US alone, 4 hours before weak training, did not cause any change in the ingestion/egestion behaviors compared to animals that did not receive the strong training US, demonstrating that the energetic value of high sucrose alone during strong training was not the source of the bias toward ingestion behavior (fig. S5, A to D). Furthermore, prior weak training (AA + s) 4 hours earlier also did not change the perception of subsequent weak training, confirming that the acquisition of a past strong memory is necessary to alter the perception of the weak training (fig. S5, A to D). We also ruled out the possibility that previous strong or weak training was simply altering the responsiveness to the CS or US used in the weak training (fig. S6, A to D). Together, our findings demonstrate that naïve animals perceive the CS + US used during weak training as an ambiguous/bistable stimulus and that previous strong training shifts the animal’s perception of the weak training to a stable state, biasing behavioral expression toward positive (ingestion) versus negative (egestion) behavior.

Open in a separate window

Fig. 2.

Previous learning alters the perception of future training by shifting the feeding network state.

(A) Example frames showing mouth movements during ingestion or egestion. Frames are color-matched to (B) (red: ingestion, blue: egestion). White dots indicate distal tip of the radula tracked during bite classification. Scale bar, 0.5 mm. (B) Heat plots of radula movements during the first 15 bites in response to the CS + US during weak training in animals receiving weak training only versus animals receiving strong training 4 hours earlier. Red-white-blue lookup table represents radula movements. Positive (blue) is egestion; negative (red) is ingestion. (C) Statistical summary of (B) shows a significant change in the mean difference in radula movements between conditions (two-tailed t test, P < 0.01, t = 3.1). (D) Plot of fraction of ingestion/egestion bites produced during weak training showing a significant difference between conditions (Fisher’s exact test, P < 0.01). (E) Plot of fraction of animals performing no egestion bites versus >0 egestion bites showing a significant difference between conditions (Fisher’s exact test, P < 0.001). (F) B11 and N2v activity in an in vitro preparation during ingestion and egestion cycles. B11 is predominantly active in the retraction phase during ingestion and the protraction phase during egestion. N2v activity does not change during ingestion and egestion cycles. B11 is therefore a readout of ingestion versus egestion. Gray lines represent retraction phase onset. (G) Heat plots of B11 activity during fictive feeding cycles. Lookup table colors are normalized B11 spike differences. Positive (blue) is egestion; negative (red) is ingestion. (H) Statistical summary of (G) shows a significant change in B11 spike difference between conditions (Mann-Whitney test, P < 0.01, U = 70). (I) Plot of fraction of ingestion/egestion cycles showing a significant difference between conditions (Fisher’s exact test, P < 0.001). (J) Plot of fraction of preparations producing no egestion cycles versus >0 egestion cycles showing a significant difference between conditions (Fisher’s exact test, P < 0.001). h, hours.

Shifts in the perception of weak training are driven by changes in network state

We next set out to identify the neural changes that underlie the shift in perception of the CS + US used during weak training that is correlated with new memory acquisition. This is readily achievable in Lymnaea thanks to the highly accessible and well-defined neural circuits that control its feeding behavior. Specifically, we made intracellular recordings from the phase-switching feeding motoneuron, B11, which provides a readout of the relative activity in both the ingestion and egestion circuits and thus serves as an in vitro measure of the internal network state (Fig. 2F). In isolated brains, where sensory pathways were absent, we compared this readout in both naïve animals versus those that had previously received strong training. Matching our findings in vivo, preparations from naïve animals showed many rhythmic activity patterns known to underlie egestion cycles (fictive egestion), while those that had experienced strong training were almost exclusively biased toward the electrophysiological expression of ingestion cycles (fictive ingestion; Fig. 2, G to J, and fig. S7A). Furthermore, by examining pairs of cycles, we found that naïve preparations transition between ingestion and egestion cycles at a higher rate than preparations that have received the strong training (fig. S7, B to E). Preparations that received prior weak training showed no significant difference in their fictive motor programs compared to naïve preparations (fig. S7, F to K). This provides compelling evidence to suggest that strong training elicits a learning-induced shift and stabilization in the network state.

Next, we tested whether strong training altered the “responsiveness” of the feeding circuitry. To do this, we stimulated the medial lip nerve (MLN), a projection pathway carrying chemosensory input from the lip structures to the central nervous system (CNS) (28), which is known to drive strong fictive feeding (29). This stimulation does not encode specific information about the nature of the stimulus but activates all chemosensory projections, allowing us to test whether strong training induced a generalized sensitization effect regardless of the appetitive cue. We monitored this on B9, a motoneuron that provides a readout of feeding cycles in the circuit, and the cerebral giant cell (CGC), a key identified modulatory neuron that receives excitation from appetitive cues (30). We found that neither the number of feeding cycles nor the CGC spike frequency were significantly different between naïve and trained preparations (fig. S8, A to D) following lip nerve stimulation.

Together, these results demonstrate that strong training is not eliciting a generalized sensitization of the animal that enhances appetitive cue responses. Rather, there is a learning-induced shift in the network state after strong training that biases the network toward the generation of ingestion cycles. We propose that this underlies the altered perception that is observed in these animals, in turn enhancing their capability to acquire new memories.

Elucidating the control circuit which drives the shift in perception

Next, to characterize the circuit mechanisms that mediate this learning-induced shift in perception, we measured activity in two higher-order interneurons, cerebral ventral 1a (CV1a) and pattern reversing neuron (PRN). These cells are command-like neurons in the ingestion and egestion circuits, respectively, and, when active, are sufficient to drive their respective motor programs (16, 31). Intracellular recordings showed that CV1a activity was significantly up-regulated at 4 hours after strong training versus naïve animals, while PRN activity was significantly down-regulated (Fig. 3, A to E). We reasoned that this learning-driven antagonistic activity relationship could be explained by a common circuit motif, namely, an inhibition between competing circuits (32). In this scenario, an up-regulation of activity in one network would serve to drive a suppression in the antagonistic one. In support of this, CV1a receives large inhibitory inputs throughout the protraction phase during both spontaneous and PRN-driven egestion cycles (Fig. 3F). These inhibitory inputs did not arise directly from PRN, suggesting that downstream neurons were recruited.

Open in a separate window

Fig. 3.

Learning-induced shift recorded in command-like interneurons controlling antagonistic behaviors.

(A and B) Example traces showing activity of the ingestion command-like interneuron CV1a (A) and the egestion command-like interneuron PRN (B) during spontaneous cycles from naïve preparations and preparations from animals that received strong training 4 hours earlier. Gray lines represent retraction phase onset. (C and D) Heat plots of CV1a (C) and PRN (D) spike activity during multiple trials during in vitro cycles from naïve and trained preparations. Color code shows spike frequency. (E) Statistical analysis of (C) and (D). Four-hour trained preparations had more CV1a activity per cycle versus naïve (two-tailed t test, P < 0.05, t = 2.5), whereas PRN activity was significantly higher in naïve versus trained preparations (Mann-Whitney test, P < 0.01, U = 33.5). (F) Example traces of CV1a, PRN, and B11 activity during a spontaneous (left) and PRN-driven (right) egestion cycle. Red arrows indicate periods of inhibition on CV1a. h, hours.

Next, to identify the source of these inputs, we carried out an extensive search for a neuron type that would fulfil the following criteria: (i) It should inhibit CV1a when active, (ii) it should be active during the protraction phase of egestion cycles, and (iii) it should be excited by PRN activity. Using a fluorescence labeling approach to reveal neurons projecting from the buccal ganglia where most of the feeding circuitry is housed (33), we identified a single candidate neuron type, pattern switch 1 (PS1) (Fig. 4A), that satisfied all three criteria. First, artificial stimulation of PS1 monosynaptically inhibited the ipsilateral CV1a (Fig. 4, B and C). Second, this neuron was strongly active during the protraction phase of both PRN-driven and stimulus-driven egestion cycles (Fig. 4D and fig. S9A) when CV1a is inhibited. Third, PRN activity excited PS1 monosynaptically, observed as 1:1 excitatory postsynaptic potentials (EPSPs) (Fig. 4, E and F). To ascertain whether PS1 was the sole source of inhibition to CV1a during egestion feeding behavior, we artificially manipulated its activity during PRN-driven cycles. When hyperpolarized, there was now a significant increase in CV1a activity (Fig. 4, G and H) with no evidence for the phasic inhibitory synaptic inputs it normally receives. Moreover, PS1 hyperpolarization was also sufficient to increase CV1a activity during sensory-driven egestion cycles (fig. S9B). Thus, this pivotal inhibitory neuron type acts as a switch during action selection, preventing disruptive activation of the ingestion command centers during egestion.

Open in a separate window

Fig. 4.

Characterization of perceptual control circuit that mediates competitive interactions between ingestion and egestion.

(A) Identification of candidate egestion interneurons that inhibit ingestion. Left: Schematic and image of the Lymnaea central nervous system showing the main sites of the feeding circuitry. Middle: Strategy to reveal candidate buccal neurons projecting to cerebral ganglia (where CV1a is located) relies on back-filling the cerebrobuccal connective (CBC) with the fluorescent dye 5(6)-carboxyfluorescein. Right: Morphology and schematic of PS1 interneuron in the buccal ganglia after iontophoretic dye filling confirms single projection leaving ganglion via the ipsilateral CBC. Morphology was confirmed in n = 3 cells. Scale bars, 100 μm. (B and C) Artificial activation of PS1 hyperpolarizes membrane potential of CV1a (B) and can prevent spiking activity when CV1a is artificially depolarized to tonically spike (C). (D) PS1 is strongly activated during a PRN-driven egestion cycle, corecorded with B11. Gray lines represent retraction phase onset. (E and F) Monosynaptic excitatory connection between PRN and PS1; evoked spikes in PRN excite PS1 and generate 1:1 EPSPs (F). (G) Representative traces of a PRN-driven egestion cycle. PS1 is strongly recruited, and CV1a shows no spiking activity (left). Hyperpolarizing PS1 throughout the PRN-driven cycle causes CV1a to be strongly recruited during the protraction phase (right). (H) Summary plot. CV1a spike frequency was significantly greater during PRN-driven cycles where PS1 was artificially hyperpolarized (n = 6, two-tailed paired t test, P < 0.001, t = −7.2).

Since CV1a does not have a monosynaptic connection with PRN, how does it ensure the suppression of egestion when it is active? We were able to answer this question by identifying the second component of the control circuit: a buccal interneuron type, PS2, which was strongly electrically coupled to PRN and sufficient to drive robust egestion cycles (fig. S9, C and D). It receives strong facilitating inhibition from CV1a and thus leads to the suppression of egestion when ingestion cycles are generated (fig. S9, E and F). Moreover, artificial activation of PS2 caused polysynaptic inhibitory inputs on CV1a, arising through its monosynaptic excitatory connection onto PS1 (fig. S9, G and H). Together, these results demonstrate that mutual inhibition is used to prevent coactivation of competitive circuits and that this circuit motif provides a control point on which plasticity could act (fig. S9I). Next, we explored the possible role for suppression of the egestion circuit in biasing the perception of learning events.

Manipulation of perceptual control circuit enables new learning in vivo

Given that it favors the expression of egestion behavior, we reasoned that a reduction in PRN → PS1 activity might be sufficient to alter an animal’s perception of weak training and thus enhance memory acquisition. To investigate this, we developed a pharmacological strategy that allowed us to manipulate the PRN output pathway. We have previously shown that this neuron is dopaminergic and that the D2 receptor blocker, sulpiride, is highly effective in inhibiting its action on follower motoneurons (16). Here, we confirmed that the PRN → PS1 connection is also sulpiride sensitive (Fig. 5, A and B), causing a significant reduction in the amplitude of the PRN → PS1 EPSP. Next, we tested whether blocking this connection could mimic the increase in CV1a activity observed 4 hours after strong training. We found that sulpiride caused a robust elevation in CV1a cycle activity versus pretreatment (Fig. 5, C and D), consistent with our previous work showing that sulpiride application biases activity toward ingestion events (16). Thus, sulpiride can robustly alter the network state, substituting for the effects of strong training. As such, this agent provides the opportunity to test in vivo whether this pathway underlies the animal’s altered perception of the weak training. Animals were injected with either sulpiride or normal saline and underwent the weak training protocol with feeding responses measured as in Fig. 2 (A and B). We found that sulpiride-injected animals performed significantly more ingestion events in response to the weak training than saline-injected animals (Fig. 5, E to G), and significantly more of the sulpiride-injected animals performed no egestion responses at all (Fig. 5H). Moreover, there was a significant decrease in the transition probability between states after sulpiride injection (fig. S10, A to D). Therefore, both strong training and sulpiride injection shift the network state in vitro and stabilize the perception of weak training in vivo.

Open in a separate window

Fig. 5.

Pharmacological manipulation of perceptual control circuit substitutes for strong training and facilitates new memory acquisition in vivo.

(A) Sulpiride (10⁻⁴ M) reduces EPSP amplitude in PS1 arising from PRN spikes. High divalent (HiDi) saline was used to reveal monosynaptic connections. (B) EPSP size was significantly reduced by sulpiride (n = 7, Wilcoxon test, P < 0.05, W = 28). (C) Heat plots of CV1a activity during in vitro cycles from naïve preparations in saline and sulpiride. Gray lines represent retraction phase onset. (D) CV1a activity was significantly higher in sulpiride versus saline (n = 8, two-tailed paired t test, P < 0.001, t = −7.2). (E) Sulpiride shifts perception of weak training. Heat plots of radula movements during the first 15 bites in response to the CS + US during weak training in saline or sulpiride-injected animals. (F) Summary analysis of (E) showing a significant shift toward ingestion bites in sulpiride versus saline-injected animals (Mann-Whitney test, P < 0.05, U = 81). (G) Comparison of fraction of ingestion/egestion bites in response to the weak training showing a significant difference between conditions (Fisher’s exact test, P < 0.001). (H) Comparison of fraction of animals performing no egestion bites versus those performing >0 egestion bites in response to weak training shows a significant difference between conditions (Fisher’s exact test, P < 0.001). (I) Sulpiride injection 2 hours before weak training significantly increases the response to the CS (tested 1 day later) versus naïve animals, saline-injected weak trained animals, and sulpiride-injected naïve animals {naïve, n = 21; saline and weak trained, n = 19; sulpiride and weak trained, n = 20; sulpiride only, n = 22; one-way ANOVA, P < 0.001 [F(3,78) = 8.3]; Tukey’s test: sulpiride and weak trained versus naïve, P < 0.001, versus saline and weak trained, P < 0.01, versus sulpiride only, P < 0.001; all other conditions, P > 0.05}. (J) Sulpiride injected 2 hours earlier produced no significant increase in response to AA versus saline-injected and naïve animals {naïve; n = 19, saline-injected; n = 24, sulpiride-injected; n = 22, one-way ANOVA, P > 0.05 [F(2,62) = 0.015]}.

We next tested whether the shift in perception induced by sulpiride was sufficient for acquisition of LTM after weak training, as we have shown in the case of strong training. Animals were injected with sulpiride or saline and then underwent weak training with LTM tested 1 day later. Consistent with the effects of strong training, we found that animals injected with sulpiride before weak training had a significantly greater response to GNL compared to naïve or saline-injected trained animals (Fig. 5I). Furthermore, sulpiride injection in the absence of weak training did not increase the feeding response to GNL when tested 1 day later (Fig. 5I). Therefore, pharmacologically manipulating the network state causes a change in the perception of the weak training that is sufficient for the animal to acquire and fully consolidate a memory.

Next, we examined whether the identified learning-induced change in network state after strong training was involved in the expression of the original strong memory or alternatively whether this was a parallel process that served the purpose of enhancing future learning events. To test this, we injected animals with sulpiride or saline and recorded their response to the strong training CS (AA) but in the absence of prior strong training. We hypothesized that if the learning-induced change in network state was involved in the expression of the original memory, then feeding behavior in response to the CS (AA) used for strong training would be increased by artificially inducing the same change in network state with sulpiride, but in the absence of prior strong training. However, we found that sulpiride injection did not cause an increase in the response to AA compared to naïve or saline-injected animals (Fig. 5J). Therefore, although the strong training causes a shift in the network state, this learning-induced change does not actively participate in the expression of the strong memory itself, suggesting that distinct mechanisms are involved. Together, these results demonstrate that strong learning causes parallel changes in neural activity: one for the expression of the memory itself and one for altering the perception of future appetitive learning.

Single-trial appetitive training and testing procedures

Strong single-trial appetitive conditioning was performed by pairing AA (0.004%) as the CS with sucrose (0.33%) as the US using a previously well-described method (23, 24). Weak single-trial appetitive conditioning was performed by pairing GNL (0.004%) as the CS with sucrose (0.11%) or l-serine (0.11%) as the US. In a counterbalance control experiment, strong conditioning was performed by pairing GNL as the CS with sucrose (0.33%) as the US, and weak conditioning was performed by pairing AA as the CS with sucrose (0.11%) as the US. Briefly, animals were placed individually in petri dishes containing 90 ml of Cu²⁺-free water for 10 min to acclimatize to the new environment before the training procedure began. Five milliliters of the CS was added to the water, and 30 s later, 5 ml of the US was applied. Animals were left in the solution containing the CS and US for 2 min, then rinsed in Cu²⁺-free water, and returned to their home tanks. To test for CS responses 1 day after conditioning, animals from trained and naïve groups were transferred from their home tanks to a petri dish filled with 90 ml of Cu²⁺-free water and allowed to acclimatize for 10 min. Five milliliters of water was then added to the dish and the number of feeding responses (bites) in the following 2 min was counted. Next, 5 ml of the CS was added to the dish, and the number of feeding responses was counted in the subsequent 2 min. CS responses were then assessed using a “difference score” ( bite number). This was obtained by subtracting the number of feeding cycles observed during 2 min after water application from the number of feeding cycles in 2 min after CS application. During dual conditioning experiments, animals received both the strong and weak appetitive conditioning separated by time intervals as described in the results. To test whether preexposure to the US used during strong training (0.33% sucrose) enhanced weak training learning, we performed the strong training as described above but in the absence of presentation of the CS (AA). Animals then received the weak training of GNL and sucrose (0.11%) 4 hours later and were tested for their response to GNL 1 day later.

Measurement of perception during weak training

The effect of past learning on the animal’s perception of weak training was tested by performing strong training followed by weak training 4 hours later as above. During the weak CS + US presentation, animals’ feeding responses were videoed (33 frames/s) from below. The direction of movement of the radula and underlying odontophore structure was measured during the first 15 feeding responses as in (16). Briefly, the position of the dorsal mandible was first marked in the frame preceding the first frame in which the radula was visible during each bite using ImageJ software. The radula was then tracked for the entire bite, and the distance from the initial position of the dorsal mandible was calculated. The average difference moved between frames was then measured. A negative score therefore represented the radula and dorsal mandible being apart at the start of the bite and the radula moving toward the dorsal mandible during the bite, whereas a positive score represented the mandible and radula being close together at the start of the bite and the radula moving away as the bite progressed. The criteria to define whether a response was ingestion or egestion were based on whether the difference in movement was negative (ingestion) or positive (egestion). To measure how prior strong training alters the stability of the perception of the CS + US used in weak training, pairs of consecutive bites were analyzed. A stable bite pair was classified as two of the same consecutive bites (ingest-ingest or egest-egest). A switching bite pair was classified as two different consecutive bites (ingest-egest or egest-ingest). Transition probabilities were then calculated by counting the number of switching bites expressed as a fraction of the total number of bite pairs. To test whether preexposure to the US used during strong training (0.33% sucrose) altered the perception of weak training, we performed the strong training as described above but in the absence of presentation of the CS (AA) and measured ingestion/egestion behaviors as above. To test whether prior weak training altered the perception of later weak training, animals first received AA paired with 0.11% sucrose and then their ingestion/egestion responses to GNL and 0.11% sucrose were measured 4 hours later. To test for the effects of strong training on the responsiveness of the animal to either the CS or US used during weak training, animals received the strong training as above. Four hours later, animals were placed in a petri dish of 90 ml of Cu²⁺-free water and allowed to acclimatize for 10 min. They then received 5 ml of water, and their feeding responses were counted. They subsequently received 5 ml of either GNL or 0.11% sucrose, and feeding responses were counted so that the bite number could be calculated as above.

Preparations and electrophysiological methods

Following procedures previously described in (16), we carried out in vitro experiments using an isolated CNS preparation. A small region of the anterior esophagus was kept attached to the CNS by the dorsal buccal nerves. Preparations were perfused with normal saline containing 50 mM NaCl, 1.6 mM KCl, 2 mM MgCl₂, 3.5 mM CaCl₂, and 10 mM Hepes buffer in water. Monosynaptic connections were tested for by bathing the preparation in a high divalent (HiDi) saline, which increases action potential threshold, reducing polysynaptic connections. HiDi saline was composed of 35.0 mM NaCl, 2 mM KCl, 8.0 mM MgCl₂, 14.0 mM CaCl₂, and 10 mM Hepes buffer in water. Intracellular recordings were made using sharp electrodes (10 to 40 megohms) filled with 3 M KAc and 0.5 mM KCl. Signals were collected using NL 102 (Digitimer Ltd.) and Axoclamp 2B (Axon Instrument, Molecular Device) amplifiers, and data were acquired using a micro 1401 Mk II interface and analyzed using Spike2 software (Cambridge Electronic Design, Cambridge, UK).

Neuron identification

The phase switching motoneuron B11 is located in the buccal ganglia and was identified on the basis of its location, spike shape, synaptic inputs from PRN, and ability to switch its activity pattern during ingestion and egestion cycles (16). The ingestion command-like interneuron CV1a is located in the cerebral ganglia and was identified by its electrical properties, characteristic location, and its ability to drive fictive feeding cycles when artificially depolarized to fire spikes (31). The egestion command-like interneuron PRN is located in the buccal ganglia and was identified by its location and monosynaptic excitatory connection to B11. The N2v neuron is a central pattern generator interneuron located on the ventral surface of the buccal ganglia. It can be identified by its characteristic plateau during the retraction phase of a cycle, and artificial activation causes widespread retraction phase activity in many buccal neurons (25). B9 is a retraction phase motoneuron located in the buccal ganglia. The CGCs are large, serotonergic interneurons located in the cerebral ganglia, which can be identified by their size, location, and tonic spiking activity (30). To identify previously uncharacterized candidate members of the egestion network, we backfilled the cerebrobuccal connective (CBC) with the fluorescent dye, 5(6)-carboxyfluorescein (5-CF). Projection interneurons are known to be influential in driving patterned activity in Lymnaea (30, 31), and egestion driving neurons have been identified in the buccal ganglia (16). Backfilling the CBC revealed a population of buccal projection interneurons that we could reidentify in other preparations and test electrophysiologically. Neurons of interest were impaled and corecorded with ingestion and egestion command-like neurons.

Analysis and classification of in vitro cycles

Activity on B11 was measured with respect to the onset of the retraction phase, as determined by the N2v plateau or large excitation on retraction phase interneuron B9. To analyze the relative activity of B11 in a cycle, it was measured 4 s before and 4 s after retraction phase onset. The number of B11 spikes after retraction phase onset was subtracted from the number of B11 spikes before retraction phase onset and then divided by the total number of spikes in the 8-s period to gain a normalized difference score. Using this score, a positive value represents more activity occurring before the retraction phase onset and therefore is classified as an egestion cycle. A negative score represents more activity occurring after retraction phase onset and is therefore classified as an ingestion cycle. To compare the effects of strong training on fictive feeding cycles in vitro, the first 10 spontaneous cycles were analyzed from 19 naïve and 19 trained preparations. To measure how training alters the stability of cycle expression, pairs of consecutive cycles were analyzed. A stable cycle pair was classified as two of the same consecutive cycles (ingest-ingest or egest-egest). A switching cycle pair was classified as two different consecutive cycles (ingest-egest or egest-ingest). Transition probabilities were then calculated by counting the number of switching cycles expressed as a fraction of the total number of cycle pairs. To test whether strong training alters the responsiveness of the preparation to appetitive cues, we stimulated the main chemosensory pathway, the MLN (28), which can drive strong fictive feeding (29). The MLN was stimulated using a glass suction electrode with biphasic pulses of 4 V with 0.5-ms duration at 1 Hz for 120 s. The fictive feeding cycle number was calculated by recording activity in feeding motoneurons, such as B9, counting the number of cycles that occurred in the 120-s period preceding MLN stimulation, and subtracting this from the number of cycles in response to MLN stimulation. CGC activity was measured for the 120 s before and 120 s during the MLN stimulation. To elicit sensory-driven egestion in vitro, a 1-s tactile stimulus was applied to the esophagus, which activates mechanosensory neurons that signal aversive cues to the feeding network in response to overextension of the gut due to an inedible object lodged in the esophagus (16). The tactile stimulus was applied using a mechanical probe controlled by a transistor–transistor logic pulse from the micro 1401 Mk II (CED).

Iontophoretic dye filling of neurons

Following procedures previously described in (16), we filled target neurons with a fluorescent dye (5-CF) using a microelectrode. This was achieved iontophoretically using a pulse generator to apply regular interval negative square current pulses into the neuron for >30 min. Preparations were then left overnight at 4°C. Images of the neurons were taken using a digital camera (Andor Ixon electron multiplying charge-coupled device) mounted on a Leica stereomicroscope.

D2 receptor blocker application in vitro and in vivo

Sulpiride is an effective dopamine antagonist in Lymnaea, blocking the effects of dopaminergic interneurons on follower neurons as well as the focal application of dopamine (16, 54). To test for the effect of sulpiride (±) (Sigma-Aldrich) on the PRN → PS1 connection, preparations were first bathed in HiDi saline (see above). Baseline EPSP amplitudes were recorded before 10⁻⁴ M sulpiride in HiDi saline was perfused into the bath for 10 min and then EPSP amplitudes were recorded again. To test the effects of sulpiride on in vitro cycle generation in naïve preparations, the first 10 spontaneous cycles generated were recorded and then 10⁻⁴ M sulpiride in normal saline was perfused onto the preparation. The first 10 spontaneous cycles generated after 10 min of perfusion were analyzed. To test the effects of sulpiride on the perception of weak training and memory acquisition/recall, animals were injected with 100 μl of 10⁻³ M sulpiride in normal saline. It has previously been shown that the injected concentration of the drug is diluted ~10-fold by the body fluids of the animal (55). Control animals were injected with 100 μl of normal saline alone. Animals were left for 2 hours before behavioral tests were carried out.

Funding: This work was funded by grants from the Biology and Biotechnology Research Council to K.S., M.C., and G.K. (BBSRC/BB/V000233/1); to K.S. (BBSRC/BB/S00310X/1); and to I.K., G.K., and P.R.B. (BBSRC/BB/P00766X/1).