Prelimbic inactivation impairs action-consequence memory

To begin these investigations, we utilized an action–outcome contingency degradation procedure (Fig. 1a). In this task, modestly food-restricted mice are trained to generate two food-reinforced operant responses equally, then the likelihood that one response will be reinforced is reduced; instead, food pellets associated with that response are provided non-contingently. Meanwhile, the other response remains reinforced. In a probe test conducted in extinction, both response operandi are available, and preferential engagement of the behavior that is likely to be reinforced provides evidence of knowledge of the action–outcome relationship, whereas a failure to modify response strategies reflects a deferral to familiar, habit-based behaviors22.

Fig. 1 Prelimbic prefrontal cortex inactivation impairs the ability of mice to select actions based on their consequences. a Experimental timeline: Food-restricted mice are placed in operant-conditioning chambers and trained to respond on two apertures for food pellets. Following training, responding on one aperture continues to be reinforced (“non-degraded”), whereas food pellets associated with the other response are delivered non-contingently, “degrading” the action–outcome contingency. During a probe test, preferential engagement of the behavior that is most likely to be reinforced is considered goal-directed, whereas engaging both responses non-selectively is considered a failure in action–outcome conditioning, a deferral to familiar habit-based behavior. b The largest and smallest viral vector spread is represented on images from the Mouse Brain Library70. c Representative GFP expression is also shown. d Following instrumental contingency degradation and CNO injection, control GFP-expressing mice (n = 12) preferentially engaged the response most likely to be reinforced. Gi-DREADDs-expressing mice did not differentiate between the responses, deferring instead to habit-based responding (n = 7). Bars and symbols = means + s.e.m., *p < 0.05. This experiment was conducted twice, with concordant outcomes; a single cohort is represented. We thank A. Allen for generating the illustration used in this figure Full size image

To confirm that the prelimbic cortex in mice is involved in action–outcome-based decision making, we delivered viral vectors expressing either CaMKII-driven GFP or G i -coupled Designer Receptors Exclusively Activated by Designer Drugs (DREADDs23) to the prelimbic cortex (Fig. 1b, c). Following recovery, all mice acquired the food-reinforced operant responses, with no differences between groups (Supplementary Fig. 1). Given that the prelimbic cortex is necessary for forming, but not expressing, new memory regarding the predictive relationship between actions and their consequences15, G i -DREADDs were next pharmacologically activated in conjunction with action–outcome contingency degradation. Then, mice were subsequently tested for response preferences during a probe test, drug-free. Control GFP-expressing mice preferentially performed the response most likely to be reinforced in a goal-directed fashion, whereas G i -DREADDs-expressing mice responded non-selectively, deferring to a habit-based strategy (ANOVA: interaction F (1,17) = 4.370, p = 0.05) (Fig. 1d). Thus, chemogenetic silencing of the prelimbic prefrontal cortex impaired action–outcome learning and memory.

Dendritic spine densities predict decision making strategies

Multiple types of learning and memory are thought to require activity-dependent dendritic spine proliferation on the one hand, or dendritic spine elimination, which can also be activity-dependent, on the other24. To dissect whether dendritic spine dynamics influence action–outcome learning and memory, we next manipulated the cytoskeletal regulatory factor ROCK, a major substrate of the RhoA GTPase that significantly impacts activity-dependent structural remodeling25, 26.

In this experiment, mice were given extensive response training using random interval (RI) schedules of reinforcement to induce habit-based responding, which is by definition insensitive to the predictive relationship between an action and an outcome27 (Fig. 2a). Groups (to be vehicle vs. to be fasudil) were designated by matching response rates during training (ANOVA: no interaction F (15,450) = 1.217, p = 0.255, main effect of session F (15,450) = 108.956, p < 0.001, no effect of group F < 1) (Fig. 2b).

Fig. 2 ROCK inhibition enhances action–outcome decision making. a Experimental timeline: mice were trained using random interval (RI) schedules of reinforcement to induce habit-based behavior. Immediately following instrumental contingency degradation, mice were given a systemic injection of either vehicle (n = 15) or the ROCK inhibitor fasudil (n = 17). Following a probe test, brains were either collected or mice were tested further for sensitivity to outcome value using lithium chloride (LiCl)-based conditioned taste aversion (CTA). b Groups did not differ during instrumental response acquisition. c Following instrumental contingency degradation, vehicle-treated mice did not differentiate between the responses that were likely, vs. unlikely, to be reinforced, behaving habitually. By contrast, fasudil-treated mice preferentially engaged the response most likely to be reinforced. d The same data are represented as preference ratios (non-degraded/degraded contingencies), with values > 1 indicating goal-directed responding. Vehicle-treated mice responded at chance levels (habit), whereas fasudil-treated mice utilized a goal-directed response strategy. Right: although we focus our report on males, we found that female mice were also sensitive to fasudil in the same task (n = 7, 8 per group). e When fasudil injections were delayed by 4 or 18 h, no groups responded preferentially, instead relying on habitual response strategies (n = 8 per group). f We next induced CTA (n = 11, 12 per group), reducing the amount of food consumed over two pairings. g Despite CTA, previously vehicle-treated mice failed to reduce responding, consistent with habitual behavior following instrumental contingency degradation (in c). Meanwhile previously fasudil-treated mice reduced responding, sensitive to the now-reduced value of the reinforcer. h Representative prelimbic cortical dendritic segments with corresponding digital reconstructions below. i Dendritic spine densities correlated with response selection strategies, although spine densities did not differ between groups (inset). Scale bar = 2 µm. Bars and symbols = means + s.e.m. except in i where diamonds represent individual mice, *p < 0.05. For each experiment, > 2 cohorts of mice were tested. We thank A. Allen for generating the illustration used in this figure Full size image

We hypothesized that neuronal structural plasticity could be associated with the formation of new memory, so we administered fasudil immediately following instrumental contingency degradation, during the presumptive period of new memory consolidation. During the probe test the following day, the vehicle group showed no response preferences, i.e., habit-based responding, as expected. In contrast, the fasudil group preferentially generated the response that was likely to be reinforced, a goal-directed strategy (ANOVA: interaction F (1,30) = 4.560, p = 0.04) (Fig. 2c). Comparing response preference ratios revealed that the fasudil group responded well above chance (habit) levels in its preference for the behavior that was most closely associated with reinforcement, whereas vehicle-treated mice generated non-selective, habitual response strategies, at a ratio of 1 (Mann–Whitney U = 53, p = 0.005) (Fig. 2d). Although we focus in this report on male mice, we also found that fasudil had the same effects in gonadally intact female mice (Mann–Whitney U = 3, p = 0.008) (Fig. 2d, right).

In this experiment, fasudil was administered immediately following action–outcome contingency degradation, with the hypothesis that ROCK inhibition could enhance the consolidation of action–outcome learning and memory. This hypothesis predicts that fasudil treatment that is unpaired from a learning opportunity should have no behavioral effects. To test this, we trained separate mice to respond for food reinforcement (Supplementary Fig. 2). Next, vehicle or fasudil was delivered, but injections were delayed 4 or 18 h after action–outcome contingency degradation. In a subsequent probe test, no mice showed a preference for the response that was likely to be reinforced (ANOVA: no interaction F < 1, no effect of response F (1,42) = 1.538, p = 0.222, no effect of group F < 1) (Fig. 2e). Thus, ROCK inhibition appears to enhance the consolidation of action–outcome conditioning during a < 4-h time window, facilitating subsequent goal-directed action.

To rule out the possibility that the effects of fasudil could be attributable to enhancing extinction, these mice were next given 3 days of extinction training, with vehicle or fasudil administered concurrent with each session. All mice extinguished responding, and fasudil had no within- or between-sessions effects (Supplementary Fig. 2), suggesting that ROCK inhibition enhanced new action–outcome learning, rather than response extinction per se.

Insensitivity to action–outcome contingencies—as in extensively trained mice here—is often accompanied by insensitivity to outcome value27. To assess whether ROCK inhibition could also restore value-based responding (and thus, block habits), we used a devaluation procedure. The male mice in these experiments were trained for two additional sessions to reinstate responding and ensure comparable responding between groups. Next, mice were placed individually in an empty cage and allowed free access to the reinforcer pellets used in instrumental conditioning experiments for 1 h. Immediately following, mice were injected with lithium chloride (LiCl), inducing transient malaise and conditioned taste aversion (CTA), as evidenced by reduced food consumption across two sessions (ANOVA: no interaction F < 1, main effect of session F (1,21) = 16.364, p < 0.001, no effect of group F (1,21) = 2.080, p = 0.164) (Fig. 2f). When returned to the conditioning chambers drug-free, mice previously treated with vehicle did not reduce responding relative to the last day of training, despite reinforcer devaluation; in other words, control mice responded habitually, as expected. By contrast, a history of fasudil treatment reduced responding, evidence that fasudil-treated mice used the value of the reinforcer to guide their behavior (ANOVA: interaction F (1,21) = 4.794, p = 0.04) (Fig. 2g). Altogether, these findings further indicate that the ROCK inhibitor fasudil enhances action–outcome conditioning, blocking habits in favor of goal-directed response strategies.

One cohort of mice tested in the instrumental contingency degradation experiment (in Fig. 2b–d) was killed 72 h after injection. Thy1-YFP expression in these mice enabled us to image and quantify prelimbic cortical dendritic spines (Fig. 2h). Somewhat surprisingly, densities did not differ between groups (Fig. 2i, inset), nor did dendritic spine length or head diameter (compared by Kolmogorov–Smirnov tests throughout: Supplementary Table 1). However, densities correlated with decision making strategies, such that lower densities were associated with selecting actions that were likely to be reinforced in a goal-directed fashion (linear regression: r 2 = 0.725, p < 0.05) (Fig. 2i).

Enrichment of action–outcome memory

We next determined whether the behavioral effects of fasudil were attributable to efficacy in the prelimbic prefrontal cortex using intracranial fasudil delivery (Fig. 3a, b). Naive mice were trained to nose-poke for food reinforcers, and groups (to be vehicle vs. to be fasudil) were designated by matching response rates (ANOVA: no interaction F < 1, main effect of session F (6,144) = 39.253, p < 0.001, no effect of group F (2,24) = 1.585, p = 0.226) (Fig. 3c). Here, the training period was shorter because the stress of intracranial surgery would be expected to bias responding towards habits28. Immediately following instrumental contingency degradation training, vehicle or fasudil was intracranially infused. Subsequently, vehicle-infused mice generated habit-based response patterns (no preference for the response most likely to be reinforced), as expected. Meanwhile, prelimbic cortical-targeted fasudil infusions blocked habit-based responding, inducing a preference for the response most likely to be reinforced. By contrast, infusions that had unintentionally terminated in the adjacent anterior cingulate cortex had no effects (ANOVA: with control groups combined, interaction F (2,24) = 1.659, p = 0.007) (Fig. 3d). These findings suggest that inhibiting ROCK within the prelimbic cortex enhances action–outcome conditioning.

Fig. 3 Bidirectional coordination of action–outcome memory. a Experimental timeline: mice were trained to respond for food reinforcement. Immediately after instrumental contingency degradation, vehicle or fasudil was infused into either the prelimbic or anterior cingulate cortex. b Infusion sites, with coordinates relative to Bregma, are indicated on images adapted from the Allen Brain Atlas. c All mice acquired the responses, with no differences between groups. d Subsequently, vehicle-infused mice failed to differentiate between the responses that were more, vs. less, likely to be reinforced (non-degraded vs. degraded), deferring to habit-based behavior. By contrast, prelimbic cortical infusions of fasudil produced a preference for the response most likely to be reinforced, whereas fasudil in the anterior cingulate cortex had no effects (n = 12 combined control, 8 prelimbic, 7 anterior cingulate). e The same data are represented as preference ratios (non-degraded/degraded), with values >1 indicating goal-directed responding. Vehicle-infused mice responded at chance levels (habit), whereas prelimbic fasudil-infused mice utilized a goal-directed response strategy. Right: by contrast, in mice trained to be sensitive to action–outcome relationships, knockout of the endogenous ROCK inhibitor Abl2/arg or local infusion of an Arg inhibitor STI-571 ablated response preference, indicating impaired action–outcome decision making (n = 9 combined control, 6 arg−/−, 10 STI-571). Raw data are reported in Supplementary Fig. 3. Bars and symbols = means + s.e.m., *p < 0.05. For each experiment, >2 cohorts of mice were tested. We thank A. Allen for generating the illustration used in this figure Full size image

ROCK is endogenously suppressed by Abl2/Arg kinase, such that Arg ablation disinhibits RhoA GTPase-ROCK interactions29 and blocks dendritic spine plasticity in response to stimuli such as cocaine30. We obtained arg knockout mice (arg−/−) and an Abl-family (Abl and Arg) kinase inhibitor, STI-571. Using a training procedure that biases responding in typical mice towards action–outcome-sensitive goal-directed strategies, we found that arg deficiency and prelimbic cortex-targeted STI-571 infusions weakened action–outcome learning, inducing a deferral to familiar, habit-based behaviors (ANOVA: F (2,22) = 6.170, p = 0.007) (Fig. 3e, right; see also Supplementary Fig. 3). Thus, silencing the endogenous ROCK inhibitor Arg produced the opposite behavioral effect relative to the ROCK inhibitor fasudil, which stimulated goal-directed response strategies (same data as in Fig. 3d represented as preference ratios, compared by Student’s t-test: t(18) = 3.311, p = 0.004) (Fig. 3e, left).

Transient, conditioning-dependent dendritic spine remodeling

Our findings suggest that mice can use action–outcome relationships to guide decision making strategies, and this process can be facilitated by the ROCK inhibitor fasudil. This could imply that activity-dependent dendritic spine plasticity is associated with action–outcome learning. To test this hypothesis, we again trained mice to respond for food reinforcement, followed by instrumental contingency degradation training and systemic fasudil treatment (Fig. 4a). Groups were designated by matching mice based on response rates during training (ANOVA: no interaction F < 1, main effect of session F (15,225) = 111.846, p < 0.001, no effect of group F < 1) (Fig. 4b). Instead of testing response preference in a probe test the next day, as in our experiments above, brains were collected 1 h after injection. Dendritic spine imaging and enumeration revealed that fasudil caused a 9% reduction in spine densities on excitatory prelimbic cortical neurons (Student’s t-test: t(15) = −2.608, p = 0.020) (Fig. 4c, d; morphology values in Supplementary Table 1), but had no effects on densities in the anterior cingulate cortex (Supplementary Fig. 4a; morphology values in Supplementary Table 1). Further, the densities of prelimbic cortical mushroom-shaped dendritic spines—those likely to contain synapses—were reduced by 16% in the fasudil group (Student’s t-test: t(15) = −2.357, p = 0.032) (Fig. 4e). Other spine subtypes (thin and stubby) were not affected (Supplementary Fig. 4b).

Fig. 4 Selective remodeling of prelimbic cortical dendritic spines. a Experimental timeline: mice were given extensive response training to promote habit-based responding. A systemic injection of vehicle or fasudil was administered immediately following instrumental contingency degradation, with brains collected 1 h after the injection (n = 8, 9 per group). b Mice acquired the instrumental responses, with no group differences. c Prelimbic cortical dendritic segments with corresponding digital reconstructions below. d Fasudil reduced prelimbic cortical dendritic spine density. e Dendritic spine classification revealed that specifically, mushroom-shaped spines were pruned, with fasudil-exposed segments expressing as few as 0.3 per µm (represented segment above). Densities of all spine types are represented in Supplementary Fig. 4. f Dendrite diameter did not differ between groups. g Experimental timeline: Another group of mice was trained and tested identically except that injections were administered 1 day after instrumental contingency degradation (n = 7, 8 per group). h All mice acquired the responses without differences between groups. i Representative prelimbic cortical dendritic segments with corresponding digital reconstructions below. j A delayed injection of fasudil did not alter prelimbic cortical dendritic spine density, suggesting that fasudil-induced dendritic spine elimination is activity (learning)-dependent. Scale bar = 2 µm. Bars and symbols = means + s.e.m., *p < 0.05. We thank A. Allen for generating the illustration used in this figure Full size image

We next asked: Is fasudil simply “damaging” neurons? One marker of structural damage is dendritic blebbing, causing the dendrite to enlarge. 3-D dendrite reconstruction revealed that fasudil did not alter dendrite diameter (Student’s t-test: t(15) = −0.979, p = 0.343) (Fig. 4f), suggesting that neurons were not damaged.

We next trained another group of mice to nose-poke for food reinforcers (ANOVA: no interaction F < 1, main effect of session F (15,195) = 119.827, p < 0.001, no effect of group F < 1) (Fig. 4g, h). In this case, fasudil was delayed, administered 1 day after contingency degradation training, with brains again collected 1 h after injection. Here, we found no differences between groups in prelimbic cortical dendritic spine densities (Student’s t-test: t(13) = −0.834, p = 0.419) (Fig. 4i, j; Supplementary Fig. 4c), spine length, or head diameter (Supplementary Table 1). Altogether, these findings suggest that fasudil prunes mushroom-shaped prelimbic cortical dendritic spines in an experience-dependent manner.

Notably, dendritic spine densities were higher in these analyses than those in Fig. 2. This may be due to the recentness of injection prior to killing, given that injection stress can trigger dendritic spine proliferation on apical dendrites in the prelimbic cortex31. In addition, we preferentially imaged dendritic segments within our imaging window that were distal to the soma, since these dendrites are considered more labile, and thus, potentially more likely to change following drug treatment.

ROCK inhibition blocks habitual responding for cocaine

We report here that inhibiting ROCK using fasudil enhances action–outcome learning and memory, blocking habit-based responding. These experiments were conducted using food reinforcement. To build on this finding, we next assessed whether fasudil can block drug (cocaine) habits as well. Mice were trained to respond on a single operandum for a liquid solution containing cocaine and sucrose that was consumed orally (adapted from ref. 1) (Fig. 5a). Groups were assigned by matching mice based on response rates during training (ANOVA: no interaction F (16,496) = 1.175, p = 0.284, main effect of session F (16,496) = 38.453, p < 0.001, no effect of group F < 1) (Fig. 5b). Next, the cocaine solution was paired in a separate context with LiCl, inducing CTA (ANOVA: no interaction F < 1, main effect of session F (2,62) = 23.005, p < 0.001, no effect of group F < 1) (Fig. 5c). Next, mice were briefly returned to the operant-conditioning chambers, providing them with an opportunity to update the association between the cocaine-reinforced response and the now-devalued cocaine32, 33. Immediately after this session, mice received an i.p. injection of vehicle or fasudil. During a probe test the next day, the vehicle-treated group continued to respond, indicating habitual behavior. By contrast, fasudil reduced responding, indicating goal-directed, value-based action selection (ANOVA: interaction F (1,31) = 4.322, p = 0.046) (Fig. 5d). A post-probe consumption test revealed no differences in cocaine intake between groups (Mann–Whitney U = 121, p = 0.624) (Fig. 5c) and re-confirmed that both groups acquired the CTA.

Fig. 5 ROCK inhibition blocks habitual responding for cocaine. a Experimental timeline: Mice were trained to respond for an orally ingested cocaine–sucrose solution. Then, mice were subject to LiCl-induced CTA. Mice were then placed in the conditioning chambers for a brief “reminder session,” which served as an opportunity for mice to update the association between the now-devalued outcome and responding. Mice were administered either vehicle or fasudil, followed by a probe test and finally, a post-probe consumption test. These mice were then implanted with indwelling jugular catheters for i.v. cocaine self-administration in contextually distinct conditioning chambers. b All mice responded for the cocaine–sucrose solution, without differences between groups (n = 15, 18 per group). c During CTA, consumption diminished. d During a probe test, however, vehicle-treated mice did not reduce responding, despite CTA, indicating insensitivity to the now-reduced value of the cocaine reinforcer. In contrast, fasudil-treated mice reduced responding, indicating sensitivity to outcome value. e Mice were trained in contextually distinct chambers to self-administer i.v.-delivered cocaine. Although all mice ultimately acquired the response, mice with a prior history of fasudil treatment responded less throughout and required more sessions to ingest 20 mg/kg (inset). This outcome indicates that fasudil enhanced sensitivity to devaluation of the cocaine reinforcer, and not simply the sucrose that was part of the orally ingested solution. Bars and symbols = means + s.e.m., *p < 0.05. Three cohorts of mice were tested. We thank A. Allen for generating the illustration used in this figure Full size image

Given that systemic fasudil treatment had persistent effects in inhibiting habitual responding for food (Fig. 2f, g), we next assessed whether fasudil could persistently mitigate habitual responding for cocaine as well. We also aimed to confirm that the effects of fasudil in the present experiment could be attributed to sensitivity to the reduced value of cocaine, as opposed to the sucrose included in the cocaine–sucrose solution. To accomplish these goals, a cohort of the mice tested in Fig. 5b–d was surgically implanted with indwelling jugular catheters for intravenous cocaine self-administration. After recovery, mice were trained in a different room, in different operant-conditioning chambers, to respond for intravenous cocaine delivery. Mice acquired the cocaine-reinforced response (ANOVA: main effect of session F (6,96) = 7, p < 0.001), however mice with a history of fasudil treatment generated lower cocaine-reinforced response rates (ANOVA: main effect of group F (1,16) = 6.100, p = 0.025) (Fig. 5e). Responding on the inactive nose-pokes (i.e., responses that did not result in reinforcement) was also reduced by fasudil, but unlike with cocaine-reinforced response, only during the first session (ANOVA: interaction F (6,96) = 4, p < 0.001). Further, the fasudil group required more than twice as many test sessions to ingest 20 mg/kg cocaine in a single session (Mann–Whitney U = 16.501, p = 0.041) (Fig. 5e, inset). This outcome indicates that ROCK inhibition can enhance sensitivity to cocaine devaluation, reducing cocaine self-administration in an i.v.-delivered setting.

To address the potential concern that, rather than promoting sensitivity to action–outcome relationships, ROCK inhibition could be having a generalized inhibitory effect on the acquisition of new instrumental behaviors, a cohort of the mice generated in Fig. 5b–d was not subjected to catheter implantation, but instead subsequently trained in different chambers to respond on a different nose-poke aperture for sucrose pellets, a novel reinforcer that had not been devalued (Supplementary Fig. 5a). Groups did not differ in the acquisition of this new sucrose-reinforced response (Supplementary Fig. 5b). Thus, the response-inhibitory effects of fasudil were selectively associated with a devalued reinforcer.

We also assessed cocaine self-administration in mice with a history of food-reinforced responding (mice from Fig. 2b–d). This procedure allowed us to further confirm that fasudil interfered with cocaine self-administration by enhancing behavioral sensitivity to cocaine devaluation, rather than by non-specifically reducing cocaine self-administration in general. Here, a history of fasudil resulted in less responding for cocaine on day 1, likely because nose-poking had recently generated no reinforcement, but this difference was quickly lost (Supplementary Fig. 5c, d). Thus, durable cocaine avoidance results from pairing fasudil with the devaluation of cocaine.

Effects of inhibiting F-actin polymerization

ROCK inhibition can increase actin turnover, which is presumably the mechanism by which fasudil eliminates prelimbic cortical dendritic spines and promotes goal-directed decision making. To test this perspective, we sought to disrupt actin turnover with latrunculin A, which blocks F-actin polymerization. In the presence of latrunculin A, fasudil should be ineffective. We again trained mice to respond for an orally available cocaine–sucrose solution, followed by LiCl-induced CTA, and then a “reminder session” paired with a systemic injection of vehicle or fasudil. Vehicle or latrunculin A was then infused intracranially into the prelimbic prefrontal cortex 30 min later (Fig. 6a, b). Subsequently, mice were given a probe test to assess decision making strategies and finally, a post-probe consumption test to re-confirm that all mice acquired the CTA.

Fig. 6 Blocking F-actin polymerization in the prelimbic cortex prevents ROCK inhibition from promoting goal-directed action selection. a Experimental timeline: As in Fig. 5, mice were extensively trained to respond for an orally ingested cocaine–sucrose solution, followed by LiCl-mediated CTA and a brief “reminder session” paired with vehicle or fasudil. Thirty min after this injection, vehicle or latrunculin A was infused into the prelimbic cortex. Probe and post-probe consumption tests followed. b Prelimbic cortical infusion sites are indicated on images adapted from the Allen Brain Atlas. c All mice acquired the cocaine-reinforced response, with no group differences. d CTA reduced cocaine–sucrose consumption. e During the probe test, however, vehicle–vehicle-treated mice did not reduce responding relative to the last day of training (“before CTA”), indicating habitual response strategies. As expected, fasudil–vehicle-treated mice reduced responding, showing sensitivity to the now-reduced value of the reinforcer. This effect was blocked by prelimbic cortical-selective latrunculin A infusion. f The same data are represented as the ratio of response rates before vs. after CTA. Scores of 1 represent no change from baseline, whereas scores < 1 represent response inhibition (*compared to vehicle–vehicle group) (n = 13 vehicle–vehicle, 6 fasudil–vehicle, 8 vehicle–latrunculin A, 6 fasudil–latrunculin A). Bars and symbols = means + s.e.m., *p < 0.05. Four cohorts of mice were tested Full size image

Groups were matched based on response rates during training (ANOVA: no interaction F < 1, main effect of session F (16,448) = 5.813, p = 0.001, no effect of injection by infusion F < 1) (Fig. 6c). Over the course of three pairings of LiCl with the cocaine–sucrose solution, all mice reduced consumption (ANOVA: no interaction F < 1, main effect of session F (2,56) = 94.125, p < 0.001, no effect of injection by infusion F (1,28) = 1.301, p = 0.264) (Fig. 6d). A post-probe consumption test re-confirmed that all mice acquired the CTA (ANOVA: no interaction F < 1, no effect of injection F < 1, no effect of infusion F < 1) (Fig. 6d). Nevertheless, when response rates during the probe test were compared against the last day of training, vehicle–vehicle mice were insensitive to the change in outcome value, as expected, whereas fasudil–vehicle-treated mice were sensitive to the change in outcome value and reduced responding in the probe test, replicating the results of Fig. 5d. As predicted, local infusion of latrunculin A into the prelimbic prefrontal cortex blocked the effects of fasudil, resulting in response rates that did not change relative to the last day of training (ANOVA: three-factor interaction F (1,29) = 6.086, p = 0.020) (Fig. 6e). Surprisingly, mice that received a systemic injection of saline and locally infused latrunculin A were also sensitive to the change in outcome value and reduced responding.