Introduction

Conditioned flavor preferences can be established by pairing a neutral flavor-conditioned stimulus (CS) with a nutrient unconditioned stimulus (US). They are learned when the animal consumes the flavor cue mixed in solution with a nutrient, and also when the nutrient US is presented intragastrically as the animal consumes the flavor CS (or even when presented after a delay; Ackroff et al., 2012). Because conditioned flavor preferences have been shown to be extremely resistant to extinction (e.g., Albertella & Boakes, 2006; Drucker et al., 1994; Elizalde & Sclafani, 1990; Harris et al., 2004) some authors have suggested that at least some instances of flavor preference learning may reflect a distinct form of conditioning with rules that depart from traditional associative principles (e.g., Boakes, 2005; Higgins & Rescorla, 2004; Pearce, 2002). Similar suggestions have been made in the evaluative conditioning and also incentive salience attribution literatures, where similar resistance to extinction effects have been reported (e.g., Ahrens et al., 2016; Baeyens et al., 1988; De Houwer et al., 2001; Fitzpatrick et al., 2019; Hoffman et al., 2010). A similar distinction was made earlier by Garcia et al. (1970) in their discussion of the uniqueness of conditioned taste aversion learning.

Delamater (2007a) suggested that many of the demonstrations of resistance to extinction in the flavor preference learning literature may have been related to relatively insensitive test procedures. Typically, animals are trained with a neutral flavor (e.g., almond) mixed in solution with a nutrient US, for example, sucrose, and a preference is later assessed by pitting the unsweetened almond flavor CS against plain water in a two-bottle test. Delamater (2007a) pointed out that investigators have rarely assessed the magnitude of a learned preference to a flavor cue that had undergone extinction against one that had not. When Delamater (2007a) employed such a procedure, significant preferences were consistently obtained for the non-extinguished flavor over a second flavor that had undergone extinction prior to the test. Moreover, this work was generalized to other procedures that entail non-reinforced presentations of the CS. In particular, Delamater (2011) reported that rats preferred a flavor cue that had been consistently paired with sucrose to another flavor cue that had received the same number of flavor + sucrose pairings but that had also received non-reinforced presentations of the flavor cue alone either before (latent inhibition) or interspersed (partial reinforcement) during the conditioning phase.

Another important feature of the experiments just described was that Delamater (2007a, 2011) focused on the status of the association following non-reinforced training between the flavor cue and the specific sensory aspects of the nutrient US (e.g., the sweet taste of sucrose). In order to accomplish this, a selective US devaluation task was employed (e.g., Delamater et al., 2006; Dwyer, 2005; Garcia-Burgos & Gonzalez, 2012; Harris et al., 2004). In one experiment, Delamater (2007a) trained two different flavor cues with one nutrient US (e.g., almond-sucrose, banana-sucrose) and two additional flavor cues with a second distinct nutrient US (e.g., vanilla-Polycose, strawberry-Polycose). Following such training, one flavor cue from each nutrient set had undergone extinction (e.g., almond, vanilla). Then, all rats were given selective nutrient devaluation training by pairing one of the nutrients (unflavored) with LiCl injections and also presenting the other nutrient without such injections (e.g., sucrose-LiCl, Polycose alone). Finally, the rats were given choice tests pitting the two flavor cues paired with either the devalued nutrient (e.g., almond vs. banana) or the non-devalued nutrient (e.g., vanilla vs. strawberry). These tests revealed that the animals preferred the non-extinguished flavor to the extinguished flavor when the associated nutrient had not been devalued, but the very same rats preferred the extinguished to the non-extinguished flavor when they had been associated with the devalued nutrient. This pattern of results was interpreted to mean that extinction had weakened the integrity of the sensory specific association initially established between the flavor cue and the taste of the nutrient, that was itself later devalued (see also Dwyer, 2005; Harris et al., 2004). Had this been the case, the non-extinguished flavor should more strongly activate its associated nutrient representation (presumably it’s taste), compared to that of the extinguished flavor. As a result, when the associated nutrient is valuable, the animals should prefer the non-extinguished flavor over the extinguished flavor. But when the associated nutrient is devalued, the non-extinguished flavor should now cause an avoidance of that flavor and a preference towards the extinguished flavor.

Gonzalez et al. (2016) suggested that an important aspect of the experimental designs used by Delamater (2007a, 2011) was the use of a selective nutrient devaluation procedure administered post-extinction. These investigators assessed the role of motivational state in the extinction of conditioned flavor preferences. They reported that animals trained under conditions of food and water deprivation also preferred a non-extinguished sucrose-paired flavor to an extinguished one, as had Delamater (2007a) in water-deprived rats. However, animals trained with restricted access to water, and with food freely available outside the conditioning session, failed to show a preference for a non-extinguished to an extinguished flavor, except when the animals were exposed to unpaired presentations of sucrose and LiCl following the extinction phase. Furthermore, they replicated Delamater’s (2007a) finding that water-restricted rats preferred the extinguished to the non-extinguished flavor cues when the associated nutrient had been devalued. Gonzalez et al. (2016) suggested that their findings put into question the interpretation that extinction weakened the sensory-specific flavor-nutrient associations presumed to be at work in the Delamater (2007a) studies.

In order to assess the generality of the Delamater (2007a) extinction results, the present studies explored extinction of conditioned flavor preferences without using US devaluation methods in two additional ways over prior studies. First, Delamater et al. (2017) recently suggested that one variable that may affect sensitivity to extinction is the amount of CS-US training given prior to the introduction of extinction, with extensively trained CSs being less sensitive to extinction. Experiment 1 assessed the effects of extinction after different numbers of flavor-nutrient pairings. The impact of extinction on flavor preference learning was then assessed by preference tests conducted after flavor extinction training without any post-training US devaluation methods employed (see also Gonzalez et al., 2016). The results of Experiment 1 indicated clear extinction effects in thirsty animals and Experiments 2 and 3 went on to explore the possibility of spontaneous recovery following extinction of these weakened flavor preferences in order to assess the durability of the extinction effects obtained (see also Diaz & De la Casa, 2011; Garcia-Burgos & Gonzalez, 2012; Tarner et al., 2004).

Experiment 1

The present study explored the possibility that conditioned flavor preferences might be more sensitive to extinction when it is introduced following more limited verses extensive flavor + nutrient training. Spence et al. (1963), Eisenberg et al. (2003), and Delamater et al. (2017) presented data to suggest that relatively strong CS-US associations might be more immune to extinction than are relatively weak associations (see also Delamater, 1996; Rescorla, 1996). Scarlet et al. (2012), however, presented data suggesting that flavor preference learning may have underlying neural circuits that are distinct from those involved in learning to predict nutrient rewards on the basis of exteroceptive cues (like visual and auditory CSs). Higgins and Rescorla (2004) also provided evidence to suggest that conditioned flavor preferences based upon simultaneous flavor + nutrient pairings may differ, in kind, from those normally recruited during Pavlovian conditioning. Thus, the present study asked whether conditioned preferences established to flavor CSs given two versus ten pairings with a nutrient US would differ in their sensitivity to extinction.

The experimental design is depicted in Table 1. One set of flavor CSs, F1 and F2, were each mixed in solution with a 10% sucrose US and presented to the animals for ten conditioning exposures to each. Towards the end of this training phase a second set of flavor CSs, F3 and F4, were given only two flavor-sucrose pairings. These were interspersed with the final two training exposures to F1 and F2. One flavor CS from each set, F1 and F3, were then presented alone during an extinction phase for a total of 10 extinction trials given to each CS. Finally, choice tests were conducted pitting the extinguished and non-extinguished flavor CSs against one another for both the sets given extensive, F1 versus F2, and limited training, F3 versus F4.

Table 1 Experimental designs
Full size table

Methods

Subjects

The subjects were 32 naïve Long Evans rats (16 male and 16 female), whose weights ranged between 502 and 611 g (males) and 295 and 384 g (females). The subjects were bred at Brooklyn College and derived from rats obtained from Charles River laboratory. They were group-housed (between two and four rats per cage) in standard plastic cages (17 × 8.5 × 8 in., l × w × h) and maintained on a 14:10 light: dark cycle throughout the experiment. Standard chow (Purina 5001) was available ad libitum but fluids were restricted to two 30-min drinking sessions per day spaced 5 h apart (10 am, 3 pm) during the light cycle (which began at 7 am). For all drinking sessions rats were individually placed in suspended stainless-steel wire-mesh cages (24 cm × 18 cm × 17.5 cm) and then returned to their group housed home cages. All procedures were in accordance with institutional IACUC and NIH-recommended regulations as outlined in the Guide for the Care and Use of Laboratory Animals, eighth edition.

Solutions

The CS solutions used were 1% McCormick’s Imitation flavor extracts (banana, almond, strawberry, vanilla), and the US solution was 10% Sucrose (Domino Sugar). All Flavor + Sucrose and Flavor-alone solutions were prepared with room temperature tap water and were presented in 50-mL drinking tubes that were clipped to the front of the drinking cages during drinking sessions. These bottles were weighed individually (to the nearest 0.1 g) before and after each session to provide a measure of fluid intake.

Acquisition and extinction

The rats were initially adapted to the water restriction schedule for three days during which time plain water was presented twice daily for 30 min (at 10 a.m. and again at 3 p.m.). Over the next 12 days, two of the flavors, F1 and F2, were each paired with sucrose ten times and two others, F3 and F4, were each paired with sucrose twice. Here, and throughout, each drinking period lasted for 30 min. The specific flavors designated as the extensively trained F1 and F2 or the minimally trained F3 and F4 were counterbalanced across animals such that almond and banana were used for the F1/F2 set and strawberry and vanilla for the F3/F4 set for half the animals with the remaining animals receiving the opposite. F1 and F3 flavors were designated as the flavors to be extinguished, and their specific identities were also counterbalanced across animals.

Over the first 8 days, each F1 and F2 flavor was presented, mixed in solution with 10% sucrose, once each day (counterbalanced across days for time of day). From Days 9–12 conditioning trials with all four flavors occurred. On each day, rats received one conditioning trial with each flavor from the F1/F2 set (counterbalanced for time of day) or the F3/F4 set in an ABBA or BAAB sequence over days. By the end of Day 12, all rats had received ten conditioning trials with the F1 and F2 flavors and 2 trials with the F3 and F4 flavors. Note that each flavor was presented to each rat equally often in the morning and afternoon sessions across days.

Extinction training then took place over 10 days. On each of these days, rats were exposed to F1 and F3 flavors, counterbalanced for time of day, i.e., each flavor was presented equally often at each time across different sessions. On each drinking session rats were given the appropriate 1% extract solution mixed in plain water without sucrose.

Test procedure

Following training, the rats received 1 day of familiarization with the two-bottle choice test procedure with tap water in both bottles. At the beginning of each morning and afternoon session, each rat sampled from bottle 1 (for approximately 1 s), which was then removed and replaced with bottle 2 (again, with the rat sampling for approximately 1 s). Then, bottle 2 was removed and both bottles were simultaneously presented on the right and left side of the front of the cage, with their spouts separated by approximately 3 cm, for the remainder for the 30 min familiarization test period.

Testing then occurred over the next 2 days. On each day rats were given one F1 versus F2 and one F3 versus F4 choice test following an ABBA or BAAB sequence with both time of day and left/right spout location counterbalanced. Note that for each rat each set of flavors was tested once in the morning and once in the afternoon session, with side counterbalanced across days.

Statistical analysis

Intakes were monitored throughout and means and standard errors of the mean were computed suitable for within-subjects designs following the procedures of Cousineau and O’Brien (2014). The data of primary interest was from the choice test sessions. Intakes were averaged across the two choice test sessions for the extensively trained (F1 vs. F2) and minimally trained (F3 vs. F4) sets of flavors, and these were subjected to a repeated measures ANOVA followed by post hoc tests following the procedures of Rodger (1974, 1975a, b). Rodger’s method defines Type 1 error rate in terms of the mean expected proportion of true mutually orthogonal null contrasts that are rejected in error, Eα. A criterion of Eα = 0.05 was chosen for all analyses reported here, this was based on Rodger’s table of critical F values with ν1, ν2 degrees of freedom (Rodger, 1975a). With the present fairly large sample size (n = 32), use of these procedures is very powerful at detecting moderately sized effects (power > 0.99, Rodger, 1975b; Rodger & Roberts, 2013).

An additional exploratory analysis was performed on the data broken down by sex. For this analysis, % preference scores for the non-extinguished flavor were computed for each test condition and collapsed across test conditions and analyzed with between group t-tests. These were computed by expressing intake of the non-extinguished flavor as a percentage of total intake.

Use of Rodger’s method also provides an estimate of effect sizes whereby all the population means in the analysis implied by statistical inference are expressed in standard σ units but relative to one another (i.e., μj – μ.., a difference between each mean relative to the grand population mean). The approach is analogous to but much more general than Cohen’s d effect size estimate because it includes all of the sample means in the analysis as opposed to being restricted to a single comparison between two means (see Rodger, 1974). In addition, an estimate of the overall non-centrality parameter is also provided for all significant ANOVA tests. This estimate is based on Perlman and Rasmussen’s (1975) uniformly minimum variance unbiased estimator of non-centrality. Monte Carlo simulations (available upon request), indeed, confirm that this estimate is highly accurate (more than the standard estimator, ν1*F) at reflecting the true amount of variation that may exist among different population means.

An additional Bayesian analysis was performed to determine whether % preference for the non-extinguished flavor (100 * non-extinguished flavor intake / total intake in the test) differed between the two training conditions (two vs. ten pairings). Strong evidence in favor of the null hypothesis is provided by a Bayes Factor (BF) < 1/3 (Dienes, 2014).

The critical choice test data from all experiments can be accessed from the osf.io repository: DOI 10.17605/OSF.IO/MH3B5.

Results

Figure 1 shows intakes of the different flavor + sucrose solutions (F1–F4) during the conditioning phase and the flavor-alone solutions (F1, F3) during the extinction phase. Intakes of the two extensively trained F1 and F2 flavor + sucrose solutions increased over their ten presentations during the acquisition phase, from a mean of 12.1 g on session 1 to 18.3 g over the last two sessions. The minimally trained F3 and F4 flavor + sucrose solutions averaged 18.7 g over their two training sessions. Intake of the F1 and F3 extract + water solutions decreased and changed little throughout their ten extinction sessions, averaging 14.3 g.

Fig. 1
figure 1

Mean intakes (+/- SEM) of the four flavor+sucrose solutions over the 10 (F1, F2) or 2 (F3, F4) exposures during training, and to the 10 flavor alone extinction exposures to F1 and F3 in Experiment 1

Full size image

The data of most interest came from the two choice test sessions that pitted the extinguished and non-extinguished flavors against one another for the cues given minimal (F3 vs. F4) and extensive (F1 vs. F2) training. Figure 2 presents these data averaged across all subjects. A clear preference was observed for the non-extinguished over the extinguished flavor, and this was equally true when extinction had taken place after minimal or extensive training. The data were analyzed with a repeated-measures ANOVA that revealed significant differences between the four flavors, F(3,93) = 8.43, MSE = 18.167, p < 0.01, Δ = 21.8. Post hoc tests with orthogonal contrasts (Rodger, 1974) revealed no differences between the two extinguished flavor cues (F1, F3), or between the two non-extinguished flavor cues (F2, F4). However, intake of both non-extinguished flavor cues was greater than intake of both extinguished flavor cues, F(3,93) = 8.19, p < 0.01. Figure 2 also displays the magnitude of the extinction effect estimated from statistical analysis to be a moderate-to-large effect size (0.86σ units = 0.43σ – (-0.43σ)).

Fig. 2
figure 2

Mean intakes (+/- SEM) of the extinguished and non-extinguished flavors from the test sessions in Experiment 1. In addition, the population means implied by statistical analysis are indicated in σ units

Full size image

These statistical conclusions were supported by an additional analysis in which the BF was calculated on the observed differences in the % preference scores for the two and ten training conditions. For this analysis, we assumed a maximal preference for the non-extinguished flavor of 75% based on the data from Delamater (2007a, 2011). The BF for the differences between the two and ten training test data was calculated to be 0.145. This provides strong evidence in favor of the null hypothesis of no difference in the preferences for the two and ten pairing conditions (mean % preference = 58.7 and 61.8, respectively).

A further analysis of the data explored the sex variable. For this analysis, % preference scores for the non-extinguished flavor were computed for female and male rats in each of the two training conditions (two vs. ten flavor-nutrient pairings, see Table 2). Although male rats displayed a numerically larger preference in these tests, separate t-tests showed no significant differences between males and females at either training condition or when computed collapsed across training conditions.

Table 2 % preference for non-extinguished flavor
Full size table

Discussion

The data from the present study showed a clear and sizable extinction effect similar to those initially reported by Delamater (2007a), but we also extended the finding in two ways. First, we found no evidence that the amount of initial flavor + nutrient training influenced sensitivity of the flavor cues to extinction. While some research suggests that stronger CS-US associations are more immune to extinction (e.g., see Delamater, 2012; Delamater et al., 2017), we found no evidence to support that hypothesis here. Second, we demonstrated that these clear extinction effects did not depend upon the use of a post-conditioning US devaluation treatment, unlike the conclusion reached by Gonzalez et al. (2016). Further discussion of this discrepancy is deferred to the General discussion section. For now, having demonstrated that nonreinforced presentations of a flavor cue can weaken expression of a learned preference in thirsty, and not hungry, rats in the absence of a post-conditioning US devaluation procedure, we next asked if the effect could be replicated in the context of a spontaneous recovery experiment.

Experiment 2

One explanation for a weakened conditioned flavor preference is that extinction might actually weaken learning that had occurred during the initial flavor + nutrient pairing phase. This “unlearning” explanation of extinction is generally less popular than other “new learning” accounts of extinction (e.g., Bouton, 2004). In particular, if extinction entails new inhibitory learning, such learning could temporarily mask the expression of the initially acquired preference to weaken the expression of that preference. It is noteworthy that Higgins and Rescorla (2004) attempted to reestablish a flavor – nutrient association following extinction and observed that additional flavor + nutrient pairings reestablished a preference only when that training consisted of sequential, but not simultaneous, flavor + nutrient pairings. This result led them to question whether similar learning mechanisms operated in the two situations. If simultaneous flavor-nutrient learning obeys different principles, then it is not obvious that spontaneous recovery would be expected to occur following extinction in the present situation and, indeed, extinction might be more permanent.

In contrast to these findings, the possibility that spontaneous recovery might be expected to occur in the present situation is supported by a related finding in our laboratory. Scarlet et al. (2009) observed that animals trained initially with two distinct flavor-nutrient pairs (e.g., A-Nutr1, B-Nutr2) and then given reversal training (e.g., A-Nutr2, B-Nutr1) displayed preferences that were controlled by the reversal phase associations when tested soon after reversal training but controlled by the initial phase associations when tested after a 3-week delay. These results agree with others in the reversal and counterconditioning literatures that have shown similar recovery of phase 1 learning after 3-week delays but in very different learning paradigms (e.g., Bouton & Peck, 1992; Lipatova et al., 2006; Urushihara et al., 2004). The data support the view that these effects reflect the operation of a masking process whereby the initially learned associations are only temporarily suppressed but can spontaneously recover with time.

It is also worth noting, however, that the literature is mixed on the possibility of spontaneous recovery following extinction in a flavor preference learning paradigm. Diaz and De la Casa (2011) reported that following citric acid–saccharin pairings a preference was established to citric acid over water, and that this preference extinguished over repeated test sessions but recovered following a 3-week delay. Similarly, Tarner et al. (2004) reported that a sucrose-paired flavor was preferred to a saccharin-paired flavor, but that extinction exposures to the sucrose-paired flavor weakened this preference. Further, when the animals were tested following 7, 14, or 21 days following extinction, although the percentage intake of the conditioned target flavor increased, a preference for the sucrose-paired flavor over the control flavor was not restored. Thus, it is questionable whether an actual spontaneous recovery of the learned preference actually occurred in this study. Finally, Garcia-Burgos and Gonzalez (2012) gave thirsty rats almond-sucrose pairings followed by repeated almond alone versus water tests conducted while the animals were both thirsty and hungry. They observed that a reliable almond preference extinguished over repeated tests, but also failed to recover when a final test occurred 14 days later. It is difficult to interpret this finding in terms of a lack of spontaneous recovery, though, because the motivational conditions of training differed from those present during extinction and testing, and so the study could be construed as an ABB renewal experiment. Together, the published studies to date present a mixed view on the possibility of spontaneous recovery in flavor preference learning, and it is unclear whether spontaneous recovery might occur using the within-subject procedures developed here.

Table 1 shows the experimental design we employed in Experiment 2 to address this question. Rats were initially trained with two flavor + sucrose pairs, F1+ and F2+, before one of these was then extinguished, F1-. A second set of flavor + sucrose pairs was then trained after a gap of time, F3+ and F4+, before one of these was extinguished, F3-. Soon after this second extinction phase, all rats were given tests between the extinguished and non-extinguished flavor cues that had been remotely trained, F1 versus F2, and more recently trained, F3 versus F4. It is worth noting that these tests occurred 3 weeks following the final F1 extinction trial, as we, and other researchers, had shown this interval to be effective in promoting recovery in reversal learning, counterconditioning, and spontaneous recovery experiments. We expected that the non-extinguished F4 would be preferred to the recently extinguished F3, as this would replicate the result obtained in Experiment 1 and in Delamater (2007a, 2011). However, if spontaneous recovery occurred over a 3-week delay, then there should be no detectable preference for F2 over the remotely extinguished F1.

Methods

Subjects

The subjects were 28 naïve Long Evans rats (13 male and 15 female), whose weights ranged between 376 and 654 g (males) and 248 and 382 g (females). The subjects were bred at Brooklyn College (derived from Charles River stock) and maintained as in Experiment 1.

Solutions

The CS and US solutions used were as in Experiment 1, except that McCormick’s natural banana and raspberry extracts were substituted for imitation banana and strawberry, respectively, due to their availability.

Acquisition and extinction

The rats were run 6 days a week throughout the study. They were initially adapted to the water restriction schedule for 3 days during which time plain water was presented twice daily for 30 min (at 10 a.m. and again at 3 p.m.). On the off day each week all rats were given water for 1 h at 12:30 p.m. each day. Over the next five experiment days, two of the flavors, F1 and F2, were each paired with sucrose (i.e., mixed in solution, 1% extract + 10% sucrose) five times. Each solution was presented once per day, counterbalanced within each rat for time of day. Over the next five experiment days, the F1 flavor was presented alone (mixed in plain water without sucrose) twice daily, i.e., once at the morning and once at the afternoon times, for a total of ten extinction trials. Over the next eight experiment days all rats were given plain water twice per day to ensure that a 3-week period elapsed between the final flavor F1 extinction trial and the onset of testing. On the final two of these water-only days, rats were familiarized with the two-bottle procedure to be used during testing, but they were given a choice between water in both bottles in each morning and afternoon session on each of these days (as described in Experiment 1). Conditioning then proceeded with flavors F3 and F4 each paired with sucrose over the next five experiment days (counterbalanced across time of day within rats as had been done earlier with flavors F1 and F2). This was followed by extinction training with flavor F3 presented alone twice daily over the subsequent five extinction days (ten total F3 extinction trials).

The specific flavors designated as F1 and F2 were counterbalanced across animals such that almond and banana were used as the F1/F2 set and raspberry and vanilla for the F3/F4 set for half the animals with the remaining animals receiving the opposite. The F1 and F3 flavors were the extinguished flavors, and their specific identities were also counterbalanced across animals. This meant that each specific flavor equally played the roles of F1, F2, F3, and F4 across animals.

Test procedure

For the 2 days following F3 extinction, the rats were given two choice tests pitting flavors F1 versus F2 and two pitting flavors F3 versus F4. These tests were counterbalanced for time of day and also for left/right cage position. Unfortunately, the data from the second morning test session was lost due to experimenter error and so this test was repeated on the third day. Separate analyses on the data that omitted this extra test session did not change the pattern of results, but since the procedures are fully counterbalanced for side with those data included the results to be reported include the results from this additional test session.

Results

One female rat died towards the end of the F3 extinction phase, leaving a total n = 27. The data from this animal were excluded throughout. Figure 3 presents intake data from the F1/F2 acquisition, F1 extinction, F3/F4 acquisition, and F3 extinction phases. Intakes of the flavor + sucrose solutions rapidly increased to a mean of approximately 20 g during the F1/F2 and F3/F4 conditioning phases, and intakes rapidly dropped to the flavor extract alone solutions to a mean of approximately 12 g throughout the F1 and F3 extinction phases.

Fig. 3
figure 3

Mean intakes (+/- SEM) over the course of 5 acquisition exposures to F1+sucrose and F2+sucrose followed by 10 extinction exposures to F1 alone and then 5 acquisition exposures to F3+sucrose and F4+sucrose followed by 5 extinction exposures to F3 alone in Experiment 2

Full size image

The data of most interest came from the F1 versus F2 and F3 versus F4 choice tests. These have been combined across all subjects and test sessions and the data are presented in Fig. 4. The data shows intakes of the extinguished and non-extinguished flavors in the tests involving the remotely trained F1 versus F2 flavors and the more recently trained F3 versus F4 flavors. The rats generally preferred the non-extinguished to the extinguished flavor, and this was equally true for the remotely extinguished F1 and recently extinguished F3 flavors. In other words, there was an extinction effect but no evidence for spontaneous recovery. The data were analyzed with a repeated-measures ANOVA that showed overall intake differences among the four flavors, F(3,78) = 4.87, MSE = 19.455, p < 0.05, Δ = 11.2. Subsequent orthogonal post hoc contrasts showed no differences between the two non-extinguished flavors or between the two extinguished flavors, but that both of the non-extinguished flavors were consumed more than the two extinguished flavors, F(3,78) = 4.53, p < 0.05. Moreover, the effect size was estimated to be 0.64σ (0.32σ – (-0.32σ)), an effect size that was somewhat smaller but similar to that seen in Experiment 1.

Fig. 4
figure 4

Mean intakes (+/- SEM) of the extinguished and non-extinguished flavors for the set of flavors trained remotely or recently from the test sessions in Experiment 2. In addition, the population means implied by statistical analysis are indicated in σ units

Full size image

These statistical conclusions were supported by an additional BF analysis on the % preference data. The same prior assumptions were used as in Experiment 1. The % preference scores for the remotely and recently tested pairs of flavors were, if anything, slightly in the wrong direction if spontaneous recovery were to have occurred (preference for the non-extinguished flavor was 62.3% and 59.4%, respectively, for the remotely and recently trained pairs of flavors). The BF = 0.199 for a comparison between these two preference scores, and the finding supports the null hypothesis that the extinction effect is equivalent across the two test intervals.

The data were also analyzed according to sex (Table 2). In this experiment, unlike Experiment 1 (or 3), female rats displayed larger preferences for the non-extinguished flavor than males. This was seen in the test of remotely, but not recently, trained and extinguished flavors, and also in the data collapsed across tests (ps < 0.05). The data are consistent with spontaneous recovery having occurred in the males in the test with remotely trained flavors, but further separate repeated-measures ANOVAs performed on the male and female rats, using a pooled error term (MSE = 18.657) revealed no significant differences in intakes among the four flavors in males, F(3,75) = 1.80, p > 0.05, although the difference in females was significant, F(3,75) = 5.39, p < 0.05, Δ = 12.7, and, if anything, females displayed a slightly larger extinction effect in the remotely trained flavors (the opposite of what would be expected if spontaneous recovery had occurred). Thus, the data cannot be taken as strong support for spontaneous recovery in males, together with a lack of spontaneous recovery in females.

Discussion

The results of the present experiment again revealed that thirsty (and not hungry) rats displayed a clear preference for a non-extinguished to an extinguished flavor cue in a procedure that did not involve use of post-conditioning US devaluation treatments. The new result was that this extinction effect survived a 3-week delay interval, suggesting that spontaneous recovery might be unlikely to occur when extinction follows simultaneous flavor + nutrient pairings (see also Garcia-Burgos & Gonzalez, 2012; Higgins & Rescorla, 2004). Perhaps extinction relatively permanently weakens a previously learned flavor preference.

One reason to express some degree of caution with this interpretation is that we observed a sex difference in the present study whereby female rats appeared to be more sensitive to extinction than were male rats who, themselves, displayed a pattern of results that, although not statistically reliable, was more consistent with the possibility of spontaneous recovery. This issue is taken up in the General discussion section. A second reason, though, to express some caution is that the present experimental design may have encouraged similar performance between remotely and recently trained stimulus sets. Because all flavor CSs were trained with the same sucrose US, it is possible that acquired equivalence was established between the cues receiving similar training histories (e.g., see Delamater, 1998, 2012). If the animals had actually coded F1 and F3 flavors, on the one hand, and F2 and F4 flavors, on the other, as equivalent sets of flavors, then similar obtained extinction effects might be expected. The next experiment addressed this issue by adopting a procedure that would discourage the development of acquired equivalence learning in the present situation, and encourage acquired distinctiveness, instead.

Experiment 3

Prior research has established that when stimuli are given common training histories an acquired equivalence relation can be established between them (e.g., Hall, 1996; Sidman & Tailby, 1982; Vaughan Jr., 1988; Wasserman, 1994; Zentall et al., 1993). In Experiment 2, such acquired equivalence between the remotely and recently extinguished flavor cues and between the remotely and recently trained non-extinguished flavor cues could have been responsible for the lack of a difference in extinction effects during the choice tests with remotely and recently trained sets of stimuli. To overcome this potential problem, Experiment 3 used a procedure designed to encourage acquired distinctiveness between the remotely and recently trained flavor cues (e.g., Delamater, 1998, 2012).

Table 1 illustrates the design. The same general within-subject experimental procedure was used to assess the possibility of spontaneous recovery from extinction after a 3-week recovery interval as was used in Experiment 2. However, Flavors 1 and 2, in the present study, were paired with one nutrient (e.g., sucrose), while Flavors 3 and 4 were paired with a second, distinctive, nutrient (e.g., Polycose). Because training with distinctive outcomes is known to result in acquired distinctiveness among cues (e.g., Delamater, 1998; Honey & Hall, 1989), this training procedure should more effectively reveal a spontaneous recovery effect to the remotely trained and then extinguished Flavor 1 stimulus if such an effect exists.

Methods

Subjects

The subjects were 28 naïve Long Evans rats (14 male and 14 female), whose weights ranged between 451 and 708 g (males) and 284 and 356 g (females). The subjects were bred at Brooklyn College (derived from Charles River stock) and maintained as in Experiments 1 and 2.

Solutions

The CS solutions used were as in Experiment 2. However, 10% Polycose (Ross Laboratories) and 10% sucrose now served as the two nutrient USs. Note: Polycose is no longer manufactured by Ross laboratories, but SolCarb (Medica Nutrition) is a close substitute. Both of these have a dextrose equivalency of ~18. The Polycose profile includes 2% glucose, 7% maltose, ~55% maltooligosaccharides (degree of polymerization, DP, 3-10), and ~36% maltopolysaccharides (DP 11+). This profile is very similar to that of SolCarb.

Acquisition, extinction, and test

The same procedures were used as in Experiment 2, except that one of the nutrient USs was used with the F1 and F2 flavor set and the other was used with the F3 and F4 flavor set. This was fully counterbalanced across animals. Once again, all flavors were presented, across days, in both morning and afternoon sessions throughout training, extinction, and testing in each subject.

Results

Figure 5 presents intake data from the F1/F2 acquisition, F1 extinction, F3/F4 acquisition, and F3 extinction phases. Intakes of the flavor + nutrient solutions rapidly increased to a mean of approximately 15.5 g during the F1/F2 and F3/F4 conditioning phases, and intakes rapidly dropped to the flavor extract alone solutions to a mean of approximately 12 g throughout the F1 and F3 extinction phases.

Fig. 5
figure 5

Mean intakes (+/- SEM) over the course of 5 acquisition exposures to F1+nutrient 1 and F2+nutrient 1 followed by 10 extinction exposures to F1 alone and then 5 acquisition exposures to F3+nutrient 2 and F4+nutrient 2 followed by 5 extinction exposures to F3 alone in Experiment 3

Full size image

The data of primary interest came from the choice test phase and the intake data are presented in Fig. 6. Once again, the animals preferred the non-extinguished flavor to the extinguished one, but, importantly, this was equally true of the set of flavors trained remotely with US1 and recently with US2. The data were analyzed with a repeated-measures ANOVA that revealed differences among the four flavors, F(3,81) = 3.27, MSE = 11.017, p < 0.05, Δ = 6.6. Post hoc tests using orthogonal contrasts revealed no differences between the two extinguished flavors or between the two non-extinguished flavors, but that both non-extinguished flavors were preferred to the extinguished ones, F(3,81) = 3.19, p < 0.05. The effect size implied by this statistical analysis is 0.48σ, an effect that is somewhat smaller than in Experiments 1 and 2 but of a moderate size, nonetheless.

Fig. 6
figure 6

Mean intakes (+/- SEM) of the extinguished and non-extinguished flavors for the set of flavors trained remotely with US1 or recently with US2 from the test sessions in Experiment 3. In addition, the population means implied by statistical analysis are indicated in σ units

Full size image

As in Experiment 2, the % preference for the non-extinguished flavors was also evaluated with a Bayesian analysis contrasting the two test intervals. The % preference for the non-extinguished flavors was 57.0% and 58.5%, respectively, for the remotely and recently tested pairs of flavors. Using the same assumptions as in Experiments 1 and 2, the BF = 0.34, which provides moderate to strong support for the null hypothesis of no difference in the two preference tests.

Table 2 shows the % preference data for the non-extinguished flavor broken down by sex. Although males tended to display a somewhat larger preference than females, there were no significant differences between the preferences for males and females in this study and, generally, the data were more in line with those seen in Experiment 1 than in Experiment 2.

Discussion

The data from the present study, once again, revealed consistent extinction effects on preference learning when using a procedure that did not include a post-training US devaluation treatment. In particular, thirsty animals preferred a non-extinguished flavor to an extinguished one, and, importantly, this effect did not spontaneously recover over a 3-week period following extinction. The present results do not support an acquired equivalence interpretation of the similar findings in Experiment 2 because in the present study different nutrient USs were used to train the remote and recent sets of flavor cues and this should have encouraged acquired distinctiveness among those sets of flavor cues, and that would have better enabled the expression of spontaneous recovery to the remotely extinguished flavor cue if such an effect exists. Because our prior work had established that a 3-week delay is sufficient to produce recovery of previously learned associations in a flavor preference reversal learning paradigm, we think that our present findings suggest that simple extinction leads to a more durable loss of control by the originally learned association – a conclusion also supported by the findings of Garcia-Burgos and Gonzalez (2012) and Higgins and Rescorla (2004), but at odds with Diaz and De la Casa (2011).

General discussion

The present studies replicate and extend the earlier findings of Delamater (2007a, 2011). In particular, conditioned flavor preferences established by simultaneous pairings with a nutrient US were weakened by an extinction procedure in which the flavor CS was presented in the absence of the nutrient US. This extinction effect was assessed in preference tests that pitted the extinguished flavor CS against another CS that had received equivalent conditioning but was not also extinguished. Delamater (2007a) suggested that this would be an especially sensitive test procedure that might reflect extinction learning even when other less sensitive tests might fail to do so, e.g., when pitting the flavor CS against plain water or a CS- flavor. Moreover, the present studies also demonstrated that flavor extinction training (1) weakens preference learning after the flavor CS had received limited or extensive flavor-nutrient training, and (2) produces a long-lasting reduction in preference learning, that is, that does not spontaneously recover with time. These results point to several important issues with reference to the literature on flavor extinction learning and Pavlovian extinction learning, more generally.

The present experiments were motivated by Gonzalez et al. (2016) who performed conceptually similar studies but found some discrepant results. In particular, while Gonzalez et al. (2016) generally replicated Delamater’s (2007a) finding of a basic extinction effect in animals trained while both food and water restricted, they failed to find that animals trained just water restricted were sensitive to flavor extinction except when tested after having received post-extinction US devaluation or control treatments. These findings are in stark contrast to the present results in which we observed clear extinction effects in each of three experiments (six different conditions in all), as well as those reported by Delamater (2007a). There are a number of procedural differences between our experiments that could account for the different results. First, the animals used here were Long Evans rats and Gonzalez et al. (2016) used Wistar rats. Second, the studies employed here used fully within-subject designs in which two different sets of extinguished and non-extinguished flavors were used, as opposed to a single set of flavors. Third, the specific training parameters also differed between our studies, with the present experiments generally administering more extensive extinction relative to the number of flavor-nutrient pairings than were given by Gonzalez et al. (2016). In addition, the specific test procedures employed differed. In the present studies, animals were given two, 30-min choice tests that pitted an extinguished flavor against a non-extinguished flavor. The left/right position of the bottles in these choice tests, therefore, were counterbalanced within and across animals in the present studies. Gonzalez et al. (2016) administered a single, 15-min extinguished versus non-extinguished choice test with bottle position counterbalanced across, but not within, animals. In addition, we familiarized the animals to the two-bottle test procedure prior to the critical test sessions with water versus water exposure so as to diminish the influence of novelty at the time of testing. It seems possible that some or all of the above-mentioned variables could have contributed to the obtained differences. It is noteworthy, though, that Gonzalez et al. (2016) generally found extinction effects similar to those reported by Delamater (2007a), except in the one condition in which thirsty rats were trained and tested. The data pattern they found for those animals was generally in the right direction of an extinction effect having occurred, although those differences were not statistically reliable. It seems possible that the many procedural differences noted above could have added uncontrolled error variability to the Gonzalez et al. (2016) procedures that would have made it especially difficult to detect the extinction effect in that particular experimental condition. Consistent with this is the fact that Gonzalez et al. (2016) did report a statistically reliable extinction effect in their final experiment in which two, 2-bottle tests were conducted as in Delamater (2007a). Although these authors attributed that finding to the importance of the use of a post-conditioning treatment, the obvious alternative is that increased sampling (with two tests instead of just one) provided greater stability in the data and within-subject side counterbalancing that could both allow a reliable extinction effect to be obtained. Nonetheless, because of the many procedural differences, it is difficult to know the source of the difference, but this should not change the fact we have easily observed reliable extinction effects without using post-conditioning devaluation treatments and this puts into question the generality of the claim of the importance of post-conditioning nutrient devaluation treatments in mediating the effect.

The present studies also included male and female subjects and, although the purpose of the present studies was not to focus on potential sex differences, an exploratory analysis revealed a difference in Experiment 2 with females generally showing greater sensitivity to extinction than males. This result is difficult to interpret because although there were no statistically reliable sex differences obtained in Experiments 1 and 3, the males tended to show stronger extinction sensitivity in these studies. Nonetheless, it may be fruitful for future studies to explore potential sex differences more directly, for example, by controlling estrus variables.

The lack of any spontaneous recovery in the present studies is potentially revealing. On the one hand, one might argue that our recovery interval was simply not long enough to permit for an adequate recovery effect in this preparation. However, as noted above, Scarlet et al. (2009) demonstrated recovery of phase 1 learning after a 3-week delay in a flavor preference reversal learning task not very different from that employed here. Thus, we have previously found this interval, in this laboratory, to be sufficient for recovery to occur in a similar flavor preference learning paradigm. On the other hand, Higgins and Rescorla (2004) demonstrated that simultaneous flavor-nutrient pairings resulted in a form of flavor conditioning that once extinguished could not be reestablished even with explicit further flavor-nutrient pairings (see also Garcia-Burgos & Gonzalez, 2012), although sequential flavor-nutrient pairings did result in reconditioning following extinction. These authors suggested that simultaneous flavor-nutrient pairings may result in a form of configural encoding, distinct from that which normally occurs when associations have been established between separate CS and US representations. In this case, separate flavor-alone presentations during extinction training could establish a new configural representation of the flavor-alone (see also Pearce, 2002) distinct from the flavor+nutrient configuration acquired during the initial training phase. If expression of a flavor preference depends upon the ability of the flavor stimulus to reactivate the flavor + nutrient configural representation, that process could have been permanently disrupted by flavor extinction as was originally suggested by Rescorla (1981). Thus, there is reason to expect that spontaneous recovery might not occur following simple flavor extinction, as even explicit retraining seemingly fails to restore simultaneous flavor preference learning (Garcia-Burgos & Gonzalez, 2012; Higgins & Rescorla, 2004).

The above considerations raise the question of why Scarlet et al. (2009) should have observed recovery of phase 1 learning in a flavor preference reversal learning task. A potentially important difference is that flavor extinction entails presentation of the flavor alone following flavor-nutrient pairings. However, in reversal learning flavor-nutrient 1 pairings are replaced with flavor-nutrient 2 pairings. During test, responding to the flavor alone is assessed for the first time, unlike in extinction procedures. Thus, in that case, it may be presumed that two distinct flavor-nutrient configural representations were encoded and memory competition for one over the other could result when the flavor is presented alone during test. However, during a test conducted after extinction, memory for the flavor alone configural representation would presumably dominate because the test stimulus most resembles it. This idea is similar to that proposed by Pearce (2002); however, Pearce suggested that the flavor cue would not lose its ability to also activate the flavor + nutrient configural representation. The basis for that claim was that flavor preference learning was presumed to be insensitive to extinction. Our results suggest that this formulation may not be correct. Regardless of these speculations, the present results suggest that recovery effects may differ in interesting ways between extinction and reversal and/or counterconditioning flavor conditioning procedures.

Our lack of a spontaneous recovery effect is consistent with some but not all studies in the flavor preference learning literature. As noted above, Garcia-Burgos and Gonzalez (2012) failed to find evidence for spontaneous recovery, but their experimental design was partly compromised by the fact that the motivational conditions of training differed from those of extinction and test. Therefore, renewal processes could have played a role since testing occurred under the same motivational conditions of extinction. The Tarner et al. (2004) study reported evidence consistent with spontaneous recovery, but, unfortunately, a significant preference for the sucrose-paired flavor was not reported in this study when testing occurred following any delay interval. The Diaz and De la Casa (2011) study, however, did observe that a preference for citrus acid over water was extinguished over repeated tests but then returned after a 3-week delay. It is difficult to know how to reconcile this result with the present findings. However, one consideration is that in the Diaz and De la Casa (2011) study, over the course of repeated test sessions, experimental animals could have lost their preference because of habituation to the taste of citrus acid. If that habituation was short term, then a conditioned effect may have been revealed after 3 weeks. Consistent with this possibility was their finding that control rats who normally preferred water to citrus acid lost that preference over repeated tests, but once again regained it when testing occurred after a 3-week delay. Delamater (2007a) described the possible role of habituation in extinction and used a within-subject experimental design, similar to those used in the present studies, that could not be readily interpreted in terms of habituation. It remains to be understood whether the spontaneous recovery effect reported by Diaz and De la Casa (2011) would generalize to procedures using alternative flavor cues and within-subject designs.

Another line of evidence also points to a potential difference between flavor preference learning and more conventional Pavlovian conditioning. Scarlet et al. (2012) explored the neural substrates of flavor preference learning and observed that disruption of two key areas that have been shown to be critical for the encoding of specific outcome information in standard appetitive Pavlovian conditioning had no impact on the learning of specific flavor-nutrient associations. In particular, prior work had established that the basolateral amygdala and orbito-frontal cortex both are critical for the expression of outcome-specific information as revealed by US devaluation and specific Pavlovian-instrumental transfer (PIT) tests (e.g., Balleine & Killcross, 2006; Delamater, 2007b; Lichtenberg et al., 2017). Scarlet et al. (2012) demonstrated that pretraining lesions to one or the other of these areas had no harmful impact on the sensitivity of nutrient-specific conditioned flavor preferences to selective nutrient devaluation, although both did undermine outcome-specific control in PIT tests using conventional auditory and visual CSs paired sequentially with these nutrients. Furthermore, Scarlet et al. (2012) also showed that flavor-nutrient conditioning was insensitive to these lesions even when trained using a sequential pairing procedure. The results, overall, suggest that conditioning with auditory and visual cues, on the one hand, and flavor cues, on the other, may recruit distinct neural circuits, and possibly psychological ones as well.

Delamater et al. (2017) suggested that whether or not outcome specific CS-US associations are sensitive to an extinction procedure may depend upon the strength of the original CS-US association prior to the introduction to extinction. Delamater (1996) had shown that outcome-specific CS-US associations in standard appetitive conditioning procedures are completely immune to extinction and a variety of other procedures that were otherwise effective at selectively diminishing conditioned magazine approach responses. This preservation of the integrity of the underlying outcome-specific CS-US association was assessed using specific PIT tests, though Rescorla (1996) obtained similar results using selective US devaluation tests. Delamater et al. (2017) pointed out that the earlier studies by Delamater (1996) gave fairly extensive Pavlovian training prior to assessing the effects of extinction. In the Delamater et al. (2017) studies, procedures hypothesized to lead to relatively weak encoding of the CS-US association consistently resulted in sensitivity to extinction, that is, extinction weakened specific PIT. Furthermore, spontaneous recovery of extinction-induced weakened PIT failed to occur even when spontaneous recovery was shown to occur with the extinguished conditioned magazine approach response. In that case, similar to the present studies, it appears as though extinction resulted in a durable loss of outcome selective control. However, while it may be tempting to conclude that similar processes might be at work in these two domains, the evidence suggesting that flavor preference learning may rely on different underlying psychological and neurobiological mechanisms suggests that similar durable extinction effects in these two domains may be more of a coincidence. The present studies provided no evidence, for instance, that the extinction effect was different between flavors given two or ten flavor-nutrient pairings. However, it could be that just two 30-min pairings are sufficient to produce asymptotic levels of preference learning, and that more limited training would render the preference even more extinction-sensitive. Even if this were true, however, it seems remarkable that the present studies established such clear extinction effects after fairly extensive flavor-nutrient training to begin with, given that such effects have not been easy to obtain either in conventional Pavlovian paradigms or given that flavor preference learning is generally thought to be relatively insensitive to extinction.

As noted in the Introduction, one of the reasons why flavor preference learning has often been thought to be a special form of learning is because its property of being resistant to extinction. In support of Delamater (2007a, 2011), the present studies consistently found that nutrient-paired flavors were highly sensitive to extinction. However, especially sensitive test procedures were used, and we did not assess extinction using flavor versus water tests. Gonzalez et al. (2016) did use such tests and sometimes did find extinction effects even with these tests (see also Diaz & De la Casa, 2011; Tarner et al., 2004; but not Harris et al., 2004). Those results, taken together with ours, just point to the conclusion that flavor preferences, under some circumstances, can be shown to be weakened by extinction. Of course, if it turns out that simultaneous and sequential flavor-nutrient training results in fundamentally different forms of associative encoding and that flavor preference learning has underlying mechanisms distinct from Pavlovian learning with visual and auditory CSs, there may be other reasons to believe in the uniqueness of flavor preference learning.

Extinction training is widely thought to result in new inhibitory learning, but there may be reason to suspect that it can result in at least a partial weakening of the underlying association as well (e.g., see reviews by Delamater, 2004; Delamater & Westbrook, 2014). The failure of extinguished learning to spontaneously recover, for instance, points to a more durable form of extinction learning (Delamater et al., 2017). Extinction could, in principle, result in (1) new inhibitory learning, (2) weakening of underlying associative learning, and/or (3) the establishment of new configural representations unique to extinction training. In connection with this last possibility, Garcia-Burgos and Gonzalez (2012) provided some evidence to support the possibility that extinction might even result in the stimulus becoming a net inhibitor. Our studies were not designed to evaluate these different mechanisms of extinction, but it should be mentioned that most theories would predict that the amount of new inhibitory learning accruing to a stimulus during extinction training should only be enough to compensate the amount of excitatory learning the stimulus might initially possess. Explaining how a stimulus could become a net inhibitor remains a challenging theoretical task for such a view.

In summary, the present studies examined the sensitivity of nutrient conditioned flavor preferences to extinction using a procedure that did not involve any post-training US devaluation task. We consistently observed that flavor preferences were weakened by extinction, that this extinction effect did not depend upon the initial amount of flavor-nutrient training, and that the extinguished flavor preference did not spontaneously recovery over a 3-week delay (even after controlling for acquired equivalence). The results generally support those obtained by Delamater (2007a, 2011) and point to the importance of assessing extinction in this paradigm using the very reasonable test of pitting an extinguished flavor against a non-extinguished flavor.