Discriminating goal-directed and habitual cocaine seeking in rats using a novel outcome devaluation procedure

Abstract

Habits are theorized to play a key role in compulsive cocaine seeking, yet there is limited methodology for assessing habitual responding for intravenous (IV) cocaine. We developed a novel outcome devaluation procedure to discriminate goal-directed from habitual responding in cocaine-seeking rats. This procedure elicits devaluation temporarily and requires no additional training, allowing repeated testing at different time points. After training male rats to self-administer IV cocaine, we devalued the drug outcome via experimenter-administered IV cocaine (a “satiety” procedure) prior to a 10-min extinction test. Many rats were sensitive to outcome devaluation, a hallmark of goal-directed responding. These animals reduced responding when given a dose of experimenter-administered cocaine that matched or exceeded satiety levels during self-administration. However, other rats were insensitive to experimenter-administered cocaine, suggesting their responding was habitual. Importantly, reinforcement schedules and neural manipulations that produce goal-directed responding (i.e., ratio schedules or dorsolateral striatum lesions) caused sensitivity to outcome devaluation, whereas reinforcement schedules and neural manipulations that produce habitual responding (i.e., interval schedules or dorsomedial striatum lesions) caused insensitivity. Satiety-based outcome devaluation is an innovative new tool to dissect the neural and behavioral mechanisms underlying IV cocaine-seeking behavior.

Instrumental behavior is controlled by two distinct neural systems: a goal-directed system and a habitual system. The goal-directed system guides instrumental behaviors that are rapidly acquired, flexible, and performed in direct pursuit of the desired outcome, whereas the habitual system guides instrumental behaviors that are automatically elicited by conditioned stimuli (Dickinson 1985; Balleine and Dickinson 1998; Yin and Knowlton 2006). Habitual responding has been implicated in drug addiction and compulsive drug seeking (Ostlund and Balleine 2008; Belin et al. 2013; Barker and Taylor 2014; Everitt and Robbins 2016; Torregrossa and Taylor 2016; Smith and Laiks 2018). However, the extent to which goal-directed and habitual response strategies contribute to cocaine seeking remains unclear. In animal models of addiction, this has been difficult to examine due to limited methodology for determining whether intravenous (IV) cocaine seeking is goal-directed or habitual.

For food-seeking behavior, outcome devaluation or contingency degradation methods have been pivotal in assessing whether responding is goal-directed or habitual (Balleine and Dickinson 1998; Adams and Dickinson 1981; Colwill and Rescorla 1985). Outcome devaluation entails diminishing the value of the outcome (e.g., food) via satiety or lithium chloride (LiCl)-induced illness and then monitoring the rate of instrumental responding under extinction conditions. Because goal-directed actions are sensitive to changes in outcome value, reduced responding following outcome devaluation indicates goal-directed behavior, whereas insensitivity to outcome devaluation indicates habitual behavior (Dickinson 1985; Balleine and Dickinson 1998). These methods have been used to establish a role for the dorsomedial striatum (DMS) in goal-directed behavior and the dorsolateral striatum (DLS) in habitual behavior (Yin et al. 2004, 2005a,b, 2006; Yin and Knowlton 2006; Balleine and O'Doherty 2010; Corbit and Janak 2010; Gremel and Costa 2013).

For cocaine-seeking behavior, outcome devaluation has not commonly been used to assess response strategy, partially due to potential difficulties with devaluing an IV reward (Ostlund and Balleine 2008; Everitt and Robbins 2013; Everitt 2014). Methods developed previously for cocaine outcome devaluation include (1) using orally consumed cocaine for self-administration and devaluation (Miles et al. 2003; Swanson et al. 2017), (2) devaluing IV cocaine with LiCl-induced illness (Root et al. 2009; Leong et al. 2016), and/or (3) extinguishing the taking link in a seeking–taking chained schedule of IV self-administration (Olmstead et al. 2001; Zapata et al. 2010). However, despite the powerful uses of these outcome devaluation methods, they have limitations in their use because they require oral consumption, permanent devaluation (in the case of LiCl), or extended extinction training. We sought to develop a method that is free of these limitations and directly evaluates whether IV cocaine seeking is goal-directed or habitual. Here, we present an outcome devaluation method for IV cocaine that elicits devaluation temporarily and requires no additional training, allowing repeated testing at different time points.

We hypothesized that experimenter-administered IV cocaine could be used to produce outcome devaluation via satiety. Numerous studies have demonstrated that rats self-administer IV cocaine in a manner that leads to a consistent concentration of cocaine in the brain, indicating that cocaine self-administration is guided by satiety (Tsibulsky and Norman 1999; Lau and Sun 2002; Norman and Tsibulsky 2006; Zimmer et al. 2011, 2013). In fact, when given experimenter-administered infusions of cocaine during a self-administration session, rats halt self-administration temporarily and then resume again as cocaine levels fall in the brain (Pickens and Thompson 1968; Gerber and Wise 1989; Markou et al. 1999; Tsibulsky and Norman 1999; Norman and Tsibulsky 2006). These studies did not evaluate responding in the context of discriminating goal-directed and habitual response strategies, but they indicate that cocaine satiety can be achieved via self-administered or experimenter-administered infusions. Therefore, we developed an outcome devaluation procedure for IV cocaine using experimenter-administered cocaine.

We tested the efficacy of this outcome devaluation procedure by evaluating two factors that have been established to bias toward goal-directed versus habitual responding in food studies. First, we tested random ratio (RR20) versus random interval (RI60) schedules of reinforcement, because previous studies with food rewards have shown that RR schedules encourage the use of goal-directed strategies, whereas RI schedules encourage the use of habitual strategies (Dickinson et al. 1983; Dickinson 1985; Yin et al. 2004, 2005b, 2006; Yin and Knowlton 2006; Gremel and Costa 2013). Second, we tested bilateral lesions of DMS versus DLS based on previous work using lesions to establish roles for these striatal areas in goal-directed versus habitual responding, respectively (Yin et al. 2004, 2005b; Gremel and Costa 2013). Altogether, the results support the validity of using outcome devaluation via experimenter-administered IV cocaine for discriminating goal-directed and habitual cocaine responding.

Results

Rats were trained to self-administer IV cocaine via a seeking–taking chained schedule of reinforcement (Fig. 1A) under RR20 or RI60 schedules for the seeking lever, either with or without a 4-min time-out between trials and with a stimulus light signaling availability of the seeking lever (S+). This chained schedule allows separate assessment of cocaine-seeking and -taking behaviors, which have distinct motivational processes (Olmstead et al. 2000, 2001). For outcome devaluation testing, we gave experimenter-administered IV cocaine prior to a 10-min extinction session and compared responding with a session without cocaine (Fig. 1B).

Figure 1.

Experimental design for cocaine self-administration training and outcome devaluation testing. (A) Time line of training for seeking–taking chained schedule of reinforcement, with criteria for each stage prior to progressing to the next stage. (FR) Fixed ratio, (RI) random interval, (RR) random ratio, (S+) cue light indicating lever is active, (TO) time-out. (B) Time line of outcome devaluation and nondevaluation sessions, including a 5-min waiting period in the operant conditioning chamber, followed by experimenter-administered IV infusions of cocaine (for devaluation; 20 sec between each infusion; 60 sec after last infusion) or no cocaine (for nondevaluation; no infusions were given). The seeking lever (with S+) was then available for 10 min under extinction conditions, after which it was retracted and a 5-min waiting period occurred before the start of a self-administration session.

Sensitivity to experimenter-administered cocaine

We evaluated individual levels of cocaine satiety by analyzing the final four self-administration sessions prior to testing (Fig. 2A). We used pharmacokinetic modeling to estimate cocaine concentrations in the brain during self-administration sessions and found that levels remained relatively consistent across a 2-h session (Fig. 2B, representative rats). Furthermore, within-subject comparisons (n = 37) showed that rats were consistent in their self-administration across sessions (final session 1 vs. 4 in the final days prior to testing) according to the number of infusions (r = 0.91, P < 0.0001), average estimated cocaine concentration in the brain (r = 0.92, P < 0.0001), maximum estimated cocaine concentration in the brain (r = 0.79, P < 0.0001), and average of the estimated cocaine concentration peaks in the brain (r = 0.91, P < 0.0001) (Fig. 2C, peaks represented by orange tick marks in B). The latter was the most consistent measure across all four final self-administration sessions (one-way ANOVA: n.s., F_(3,108) = 0.53, P = 0.66), so we used this as a representative measure of satiety. Satiety was consistent within rats, but differed among rats (subject comparison: F_(36,108) = 25, P < 0.0001).

Figure 2.

Satiety via self-administered or experimenter-administered IV cocaine. (A) Time line showing the final four self-administration sessions, followed by outcome devaluation testing with experimenter-administered cocaine. (B) During self-administration sessions, estimated cocaine concentrations in the brain are relatively constant (representative sessions with and without an intertrial time-out). Orange dots represent concentration peaks. (C) Across self-administration sessions (final session 1 vs. 4), rats showed consistent satiety levels, as measured by the average of the brain cocaine concentration peaks across a session. (****) P < 0.0001; n = 37. Dots represent individual rats. (D) Representative data from two rats (trained on RR20 vs. RI60) showing raw lever presses during 10-min tests after no infusion or IV cocaine (1.0–2.5 mg/kg). (E) Normalized data from the two representative rats shown in D. (F) Experimenter-administered IV cocaine (1.0–2.5 mg/kg) reduced responding during a 10-min extinction test in a subset of rats (blue, n = 9), while other rats were insensitive regardless of the dose (orange, n = 2). (*) P = 0.01 for group effect. Mean (±SEM) normalized lever presses for the devalued sessions are shown in larger, darker symbols; individual data are shown in lighter symbols/lines. (G) Difference scores for normalized responding with each increasing dose of IV cocaine (in F) typically fell within a ±0.2 range (black dots in gray box), but some difference scores were noticeably larger and less than −0.2 (blue dots). (H) Normalized responding for the session prior to and subsequent to the large difference score (less than −0.2, shown in G) shows that responding in all rats fell below 0.4 ([****] P < 0.0001), indicating a sensitivity threshold. Mean (±SEM) and individual data are shown. (I) Self-administered cocaine (final four sessions) correlated with experimenter-administered cocaine (the lowest dose causing sensitivity). The estimated maximum brain concentration for experimenter-administered cocaine correlated with the average of the brain cocaine peaks during self-administration. (****) P < 0.0001; n = 26. (J) The dose of experimenter-administered cocaine (milligrams per kilogram) correlated with daily intake during self-administration (milligrams per kilogram in 2 h). (****) P < 0.0001; n = 26.

To determine whether experimenter-administered cocaine could be used to produce satiety and thus outcome devaluation, we gave IV cocaine prior to 10-min extinction tests and compared responding with a separate test with no infusion (Fig. 2A). Data were normalized within-subject, such that responding on a cocaine test session was divided by total responding on both the cocaine and no-infusion test sessions (i.e., lever presses after cocaine/[lever presses after cocaine + lever presses after no infusion], with 0.5 meaning equal responding on both test sessions). Representative data from two rats are shown to illustrate how normalization reduces the variability among rats (Fig. 2D,E). We tested multiple doses of experimenter-administered IV cocaine (1.0–2.5 mg/kg) on different test days in a subset of rats. Satiety was verified to be consistent within rats across self-administration sessions (r = 0.93, P < 0.0001). We found that many rats were sensitive to experimenter-administered IV cocaine and reduced responding during the 10-min extinction test, particularly with higher doses (Fig. 2F, blue lines, n = 9/11). However, some rats were insensitive to experimenter-administered IV cocaine and had little change in responding, regardless of dose (Fig. 2F, orange lines, n = 2/11). K-means cluster analysis identified two groups of rats based on responding at the highest dose (2.5 mg/kg; t₍₉₎ = 9.7, P < 0.0001). These two groups were significantly different from each other across doses (two-way ANOVA of group × dose; group effect: F_(1,9) = 10, P = 0.01; dose effect: F_(3,27) = 3.2, P = 0.04; interaction: F_(3,27) = 1.6, P = 0.20). To identify the lowest IV cocaine dose that resulted in sensitivity for each rat, we calculated the change in normalized responding for each increasing dose of cocaine, giving three difference scores (i.e., 1.5 mg/kg vs. 1.0 mg/kg, 2.0 mg/kg vs. 1.5 mg/kg, and 2.5 mg/kg vs. 2.0 mg/kg). Although the majority of difference scores were small (±0.2) (Fig. 2G, black dots), some difference scores were noticeably larger (falling below −0.2) (Fig. 2G, blue dots), indicating heightened sensitivity at a particular dose. At this particular dose, normalized responding consistently fell below 0.4 (t₍₈₎ = 10.24, P < 0.0001, as compared with the dose below) (Fig. 2H), indicating that 0.4 can be used as a threshold to determine sensitivity to experimenter-administered IV cocaine. Importantly, if a rat showed sensitivity to a particular dose of cocaine, then it continued to show sensitivity to higher doses of cocaine (i.e., normalized responding remained below 0.4). All three analyses revealed the same two groups of sensitive versus insensitive rats: K-means cluster analysis, difference scores lower than −0.2, and normalized responding <0.4. Although these two groups differed in how they responded to experimenter-administered cocaine, they did not differ in self-administered cocaine in terms of infusions (t₍₉₎ = 0.39, P = 0.7), satiety levels (average of estimated cocaine concentration peaks in the brain, t₍₉₎ = 0.17, P = 0.9), or number of sessions prior to testing (t₍₉₎ = 0.55, P = 0.6); all rats received a minimum of 3 wk of training prior to testing.

To test the reproducibility of these findings, we replicated the experiments in two additional groups of rats that received similar experimental testing but different ranges of experimenter-administered cocaine (0.5–1.5 mg/kg, n = 10; or 2.0–6.0 mg/kg, n = 16) (Supplemental Fig. S1a–e). Ranges of cocaine were based on satiety levels during self-administration, which were either lower due to the addition of an intertrial time-out period or higher due to using an FR1 reinforcement schedule. For these two groups, we verified that satiety was consistent within rats across self-administration sessions (0.5–1.5 mg/kg group: r = 0.90, P = 0.0004; 2.0–6.0 mg/kg group: r = 0.74, P = 0.0012). For these two groups, we again observed that many rats were sensitive to experimenter-administered IV cocaine on the 10-min extinction test and reduced responding below 0.4 at a particular dose of cocaine as well as higher doses (Supplemental Fig. S1a,c,d, blue lines, n = 7/10 and n = 10/16). In contrast, some rats were insensitive to experimenter-administered cocaine (Supplemental Fig. S1a,c,d, orange lines, n = 3/10 and n = 6/16). When evaluating the cocaine dose that caused sensitivity, we again observed that normalized responding consistently fell below 0.4 and was significantly different than the dose below (0.5–1.5 mg/kg group: t₍₄₎ = 5.3, P = 0.006; 2.0–6.0 mg/kg group: t₍₉₎ = 6.2, P = 0.0002; two rats were excluded from this analysis because they were sensitive at the lowest dose tested) (Supplemental Fig. S1b,e).

Analysis of all three groups together indicated that satiety was reached with an experimenter-administered dose of cocaine that matched or exceeded self-administered levels of cocaine on previous sessions. We found significant correlations between experimenter-administered cocaine (i.e., the minimum dose that caused normalized responding to fall below 0.4) and previous cocaine self-administration in sensitive rats. In particular, the estimated maximum brain concentration achieved from experimenter-administered cocaine (Supplemental Fig. S1f) correlated with previous self-administered cocaine in terms of the average of the peak brain concentrations (r = 0.88, P < 0.0001) (Fig. 2I), the maximum brain concentrations (r = 0.81, P < 0.0001), and the average brain concentrations (r = 0.80, P < 0.0001). Additionally, a significant correlation was observed when comparing experimenter-administered cocaine dose (milligrams per kilogram) with previous daily intake during self-administration (r = 0.86, P < 0.0001) (Fig. 2J). Sensitivity to experimenter-administered cocaine did not correlate with intake during early training (r = 0.047, P = 0.82), indicating that satiety level was related to recent experience and often was much lower than the satiety level observed under conditions of free access.

When we analyzed the time course of responding during the 10-min test (first and second 5 min), we found that sensitive rats significantly reduced responding in the first 5 min after cocaine was given, as compared with the no-infusion test (two-way ANOVA of cocaine × time; time: F_(1,25) = 7.7, P = 0.01; cocaine: F_(1,25) = 31, P < 0.0001; interaction: F_(1,25) = 24, P < 0.0001) (Supplemental Fig. S2a). However, insensitive rats did not significantly reduce responding after cocaine (two-way ANOVA of cocaine × time; interaction: F_(1,10) = 3.5, P = 0.09) (Supplemental Fig. S2b). Sensitive and insensitive rats were not different from each other when comparing responding on the no-infusion test (two-way ANOVA of group × time; group: F_(1,35) = 1.8, P = 0.18; time: F_(1,35) = 13, P = 0.0013; interaction: F_(1,35) = 29, P = 0.09). To counteract potential extinction learning that may have occurred during the 10-min test, rats were given a typical 2-h self-administration session following the test, during which normal responding resumed and cocaine was self-administered at levels similar to those of the sessions before testing (t₍₃₆₎ = 0.13, P = 0.90). Together, these data show that experimenter-administered IV cocaine produces cocaine satiety, outcome devaluation, and, consequently, a temporary reduction in responding when behavior is goal-directed but not habitual. Hence, we refer to this procedure of administering experimenter-administered cocaine as “outcome devaluation.”

Supporting the validity of the outcome devaluation procedure, we found consistent results for sensitivity to experimenter-administered IV cocaine when animals were given repeated tests on consecutive weeks. Sensitivity to experimenter-administered cocaine on the first test was correlated with sensitivity on the second test, during which rats were given access to the seeking lever once again (r = 0.69, P = 0.018; n = 16) (Supplemental Fig. S3a) or were given access instead to the taking lever under extinction conditions (r = 0.80, P = 0.0002; n = 16) (Supplemental Fig. S3b).

To ensure that reduced responding after experimenter-administered cocaine was not due to a general suppressive effect of cocaine on reward seeking or instrumental responding, we gave experimenter-administered IV cocaine (1.0 or 2.5 mg/kg) to separate groups of rats that were trained on a similar seeking–taking chained schedule of reinforcement but with food reward instead (n = 13), so that they were naïve to cocaine. Prior to 10-min extinction sessions on 2 d, rats were given either experimenter-administered IV cocaine or no infusion (counterbalanced order for within-subject comparison, with between-subject comparison of dose). We found no effect of cocaine on normalized responding (two-way ANOVA of cocaine × dose; cocaine: F_(1,11) < 0.0001, P > 99; dose: F_(1,11) = 1.7, P = 0.2; interaction: F_(1,11) = 0.29, P = 0.6) (Fig. 3A) or raw lever presses (cocaine: F_(1,11) = 0.56, P = 0.5; dose: F_(1,11) = 0.48, P = 0.5; interaction: F_(1,11) = 1.8, P = 0.2) (Fig. 3B). However, we did observe a significant order effect: Rats consistently responded less on the second test as compared with the first test, regardless of whether they received cocaine (two-way ANOVA of dose × test order; test order: F_(1,11) = 10, P = 0.009) (Fig. 3C). This emphasizes the importance of counterbalancing the test order. Finally, when we analyzed the time course of responding (first and second 5 min), we found a significant effect of time but no differences when comparing responding after cocaine versus no infusion (two-way ANOVA of cocaine × time; time: F_(1,12) = 10, P = 0.008) (Supplemental Fig. S2c).

Figure 3.

Experimenter-administered IV cocaine does not reduce responding in rats trained to self-administer food. (A) IV cocaine (1.0 or 2.5 mg/kg) did not alter food seeking during a 10-min extinction session, as compared with a session with no infusions given (counterbalanced test order). Mean (±SEM) normalized lever presses are shown; n is shown in bars. (B) Raw lever presses for data in A (mean ± SEM); dots represent individual rats. (C) Significant effect of test order ([**] P < 0.01) regardless of whether rats received cocaine on the first or second test session, emphasizing the importance of counterbalanced testing.

RR20/RI60 schedule effects on outcome devaluation

To test the efficacy of the outcome devaluation procedure, we evaluated whether responding was biased when animals self-administered cocaine under RR20 or RI60 schedules (Fig. 4). We also evaluated the influence of a 4-min time-out on the biasing effects of these schedules due to the fact that many cocaine self-administration schedules include an intertrial time-out. A time-out period might reduce the effects of short-term satiety on cocaine seeking and therefore influence the motivational processes underlying responding (Olmstead et al. 2000). A comparison of normalized responding following outcome devaluation versus nondevaluation for cocaine (Fig. 4A) revealed a significant effect for devaluation but not for schedule or time-out (three-way ANOVA; devaluation: F_(1,31) = 13, P = 0.0009). A comparison of the raw data for lever presses revealed a significant effect for devaluation and schedule and an interaction between devaluation and schedule, with RR20-trained rats showing sensitivity to outcome devaluation with or without a time-out, and RI60-trained rats showing insensitivity to outcome devaluation (devaluation: F_(1,31) = 16, P = 0.0003; schedule: F_(1,31) = 12, P = 0.002; devaluation × schedule interaction: F_(1,31) = 13, P = 0.001; time-out n.s.) (Fig. 4B). These data indicate that RR20 and RI60 schedules bias cocaine responding to be goal-directed or habitual, respectively. However, individual data (Fig. 4C) clearly show goal-directed and habitual responding under both schedules, indicating that schedule only biases strategy but does not require the use of a particular strategy. During self-administration, we observed no difference between schedules in terms of cocaine infusions (Supplemental Fig. S4a), although there were differences in the number of seeking lever presses (two-way ANOVA of schedule × time-out; schedule: F_(1,31) = 4.4, P = 0.045; time-out n.s.) (Supplemental Fig. S4b) and responding during the time-out (Mann–Whitney test; U = 28, P = 0.047, n₁ = 17, n₂ = 7) (Supplemental Fig. S4c). Due to differences in seeking lever presses between the schedules, we reanalyzed the outcome devaluation data using an ANCOVA with baseline seeking lever presses as a covariate (two-way ANCOVA of schedule × devaluation) and observed similar significant effects for normalized and raw responding during devaluation.

Figure 4.

Effects of RR20 and RI60 schedules of reinforcement on sensitivity to outcome devaluation for IV cocaine or food, and the influence of an intertrial time-out. Cocaine: Mean (±SEM) normalized responding (A) and raw lever presses (B) following outcome devaluation and nondevaluation for cocaine (significant effect of devaluation for both, but a devaluation × schedule interaction only for raw lever presses; no significant effect of time-out; n is shown in bars). (**) P < 0.01, (*) P < 0.05. (C) Individual data for cocaine normalized responding show that rats can be sensitive (<0.4) or insensitive (≥0.4) to outcome devaluation with either an RR20 and RI60 schedule. Food: Mean (±SEM) normalized responding (D) and raw lever presses (E) following outcome devaluation and nondevaluation for food (significant effects of devaluation and devaluation × schedule interactions for both; no significant effect of time-out). (**) P < 0.01, (***) P < 0.001. (F) Individual data for food normalized responding show variability across rats on RR20 and RI60 schedules.

Separate groups of rats were trained to self-administer food pellets using a seeking–taking chained schedule of reinforcement similar to cocaine studies, with or without a 1-min time-out. Outcome devaluation was tested using sensory-specific satiety for food, comparing responding following home cage prefeeding with the food pellets earned during self-administration (devaluation) or with 15% sucrose solution (nondevaluation). We found that RR20-trained rats showed sensitivity to outcome devaluation with or without a time-out, while RI60-trained rats showed insensitivity, when comparing normalized responding (three-way ANOVA; devaluation: F_(1,29) = 31, P < 0.0001; devaluation × schedule interaction: F_(1,29) = 5.7, P = 0.024) (Fig. 4D). Raw data for lever presses indicated similar results (devaluation: F_(1,29) = 15, P = 0.0006; schedule: F_(1,29) = 31, P < 0.0001; schedule × devaluation interaction: F_(1,29) = 10, P = 0.004) (Fig. 4E). Individual data show goal-directed and habitual responding under both schedules (Fig. 4F), like we observed with cocaine. We observed no difference between schedules for the number of food pellets earned, but a significant effect of time-out (maximum of 30 pellets; two-way ANOVA; schedule: F_(1,29) = 1.0, P = 0.32; time-out: F_(1,29) = 16, P = 0.0005; schedule × time-out interaction: F_(1,29) = 1.0, P = 0.32). Both schedule and time-out affected the time taken to earn 30 pellets (schedule: F_(1,29) = 60, P < 0.0001; time-out: F_(1,29) = 74, P < 0.0001; schedule × time-out interaction: F_(1,29) = 21, P < 0.0001; maximum session time of 1 h if rats did not reach the 30-pellet maximum) (Supplemental Fig. S4d). We observed no schedule difference for the number of seeking lever presses (two-way ANOVA; schedule: F_(1,29) = 3.6, P = 0.07; time-out: F_(1,29) = 0.30, P = 0.59) (Supplemental Fig. S4e), but a difference for responding during the time-out (Mann–Whitney test; U = 1.0, P < 0.0001, n₁ = 11, n₂ = 13) (Supplemental Fig. S4f).

DMS/DLS lesion effects on outcome devaluation

To further test the efficacy of the outcome devaluation procedure, we evaluated whether responding was affected by pretraining bilateral NMDA lesions of the DMS or DLS (Fig. 5A,B). Rats were trained to self-administer cocaine via RR20 or RI60 seeking–taking chained schedules with a 4-min time-out. When tested for outcome devaluation, rats with DMS lesions showed insensitivity under both RR20 and RI60 schedules, whereas rats with DLS lesions showed sensitivity under both schedules (three-way ANOVA of devaluation × lesion × schedule; devaluation: F_(1,34) = 16, P = 0.0003; devaluation × lesion interaction: F_(2,34) = 10, P = 0.0004) (Fig. 5C). In rats with sham lesions, we did not observe a significant devaluation effect with RR20, emphasizing that schedule only biases strategy (as we observed in Fig. 4C). Raw lever press data for these lesion groups showed similar effects (devaluation: F_(1,34) = 4.8, P = 0.03; schedule: F_(1,34) = 16, P = 0.0003; devaluation × lesion interaction: F_(2,34) = 4.7, P = 0.016) (Supplemental Fig. S5a).

Figure 5.

Outcome devaluation for IV cocaine in rats with bilateral NMDA lesions of the DMS or DLS and trained on either RR20 or RI60 schedules of reinforcement. (A) Location and spread of NMDA lesions in the DMS and DLS across all rats. Each outline represents a different animal at that AP level, and lesions typically spread ∼1.5 mm in the AP axis. (B) Representative images showing DMS and DLS lesions, visualized using NeuN immunohistochemistry. (C) Mean (±SEM) normalized lever presses on nondevalued and devalued sessions according to lesion type and schedule of reinforcement (n is shown in bars). (*) P < 0.05, (****) P < 0.0001. (D) Individual data from the devalued session in C, collapsed across schedules, show that DLS lesions were significantly different than sham and DMS lesions. (***) P < 0.001, post hoc from one-way ANOVA. The numbers of goal-directed rats (blue, <0.4) and habitual rats (orange, ≥0.4) per lesion group were significantly different. χ² test: (#) P < 0.05, (##) P < 0.01, (####) P < 0.0001. (E) Mean (±SEM) number of cocaine infusions earned during self-administration was not different among lesion groups during early training (FR1 take) or late training (seek–take, RR20/RI60), but differed according to training stage. (****) P < 0.0001.

After observing no main effect for schedule for normalized data, we collapsed the data across schedules and found a significant difference among the lesion groups when evaluating responding after devaluation (one-way ANOVA across lesion groups; F_(2,37) = 13, P < 0.0001) (Fig. 5D). Although rats with DLS lesions were significantly more sensitive than rats with sham lesions (post hoc P < 0.001) or DMS lesions (P < 0.001), there was no difference between sham lesions and DMS lesions (P = 0.53). However, a large degree of variability can be seen in the individual devaluation scores of rats with sham lesions (Fig. 5D, similar variability is seen in Fig. 4C), with some rats showing goal-directed responding (<0.4) (Fig. 5D, blue circles) and others showing habitual responding (>0.4) (Fig. 5D, orange circles). In contrast, almost all rats with DMS lesions showed habitual responding, and all rats with DLS lesions showed goal-directed responding (Fig. 5D). An analysis of the distribution of goal-directed versus habitual rats revealed significant differences among all three lesion groups (χ² test; sham vs. DMS: X_(1,30) = 3.6, P = 0.03; sham vs. DLS: X_(1,29) = 9.3, P = 0.001; DMS vs. DLS: X_1,21 = 17, P < 0.0001). Importantly, there were no differences among these groups for self-administered infusions (two-way ANOVA; lesion: F_(2,37) = 0.60, P = 0.6; training stage: F_(1,37) = 141, P < 0.0001) (Fig. 5E) or lever presses during self-administration (two-way ANOVAs of lesion × schedule; seeking presses: F_(2,34) = 0.46, P = 0.6; presses during time-out: F_(2,34) = 1.5, P = 0.2) (Supplemental Fig. S5b,c), although we observed differences in seeking presses according to schedule (schedule: F_(1,34) = 45, P < 0.0001). Additionally, when we separated rats according to whether their lesions were more anterior or posterior for the DMS or DLS (Supplemental Fig. S6a,b), we found no anterior/posterior differences in sensitivity to outcome devaluation for normalized responding (Supplemental Fig. S6c,d) or raw lever presses (Supplemental Fig. S6e,f).

Discussion

We developed a novel procedure for assessing goal-directed versus habitual responding for IV cocaine using outcome devaluation via satiety. We observed that satiety could be achieved by giving experimenter-administered IV cocaine at a dose that matched or exceeded satiety levels observed with self-administered cocaine in previous sessions (Fig. 2). While some rats showed sensitivity to outcome devaluation via experimenter-administered IV cocaine, other rats showed insensitivity to outcome devaluation, indicating the use of goal-directed and habitual responding, respectively (Fig. 2). We found that the effects of experimenter-administered IV cocaine on responding were consistent with repeated testing (Supplemental Fig. S2) and were specific to rats trained to self-administer cocaine rather than food (Fig. 3). Finally, we observed a biasing effect with RR20 or RI60 schedules of reinforcement (Fig. 4) and following DMS or DLS lesions (Fig. 5).

Cocaine satiety

The outcome devaluation procedure we developed is based on cocaine satiety, a concept that has long been demonstrated in IV cocaine self-administration studies, and is evidenced by rats making consistent pauses after infusions and adjusting infusion frequency inversely with dose (Pickens and Thompson 1968; Dougherty and Pickens 1973; Gerber and Wise 1989; Tsibulsky and Norman 1999; Tornatzky and Miczek 2000; Lau and Sun 2002; Panlilio et al. 2003; Norman and Tsibulsky 2006; Zimmer et al. 2011, 2013). Even experimenter-administered infusions of cocaine have been shown to result in a temporary pause in responding or complete cessation of responding, depending on the dose and rate of delivery (Pickens and Thompson 1968; Gerber and Wise 1989; Markou et al. 1999; Tsibulsky and Norman 1999; Norman and Tsibulsky 2006). Whereas higher doses of experimenter-administered cocaine cause satiety, lower doses cause priming, or reinstatement of extinguished responding (Norman et al. 1999; Tsibulsky and Norman 1999; Norman and Tsibulsky 2006). Although higher doses of cocaine can also cause priming, the response latencies are increased, reflecting an initial period of satiety followed by priming once brain cocaine levels have fallen below a critical level (de Wit and Stewart 1981; Markou et al. 1999; Norman and Tsibulsky 2006). Pharmacokinetic modeling has been used to show consistent brain cocaine concentrations for satiety across dosing and contingency (Tsibulsky and Norman 1999; Lau and Sun 2002; Norman and Tsibulsky 2006; Zimmer et al. 2011, 2013), and we also observed consistent within-subject brain concentrations across self-administration sessions (Fig. 2C). Thus, cocaine self-administration is guided by satiety, and satiety can be achieved via self-administered or experimenter-administered cocaine.

Satiety for IV cocaine differs considerably from satiety for food. IV cocaine has almost immediate central effects and faster metabolism as compared with food. Thus, the attainment of satiety is more rapid but also more fleeting. After the initial loading phase of a self-administration session, each cocaine infusion is sufficient to cause a satiety period, while during food self-administration, each pellet constitutes a fraction of a satiating meal (Wise 1987). With IV cocaine self-administration, animals make uniform pauses after responding (as discussed above), reflecting the satiety reached by each infusion. However, with food self-administration, nonsated animals respond many times without pausing (except to eat the acquired food) and then make a long pause only after extended intake, reflecting the satiety reached by the accumulated feeding (Wise 1987). Due to the more rapid onset and offset of cocaine satiety, we determined that the most reliable way to produce satiety, and thus outcome devaluation, would be experimenter-administered delivery of IV cocaine in the self-administration chamber.

A potential concern with administering IV cocaine in the self-administration chamber is that responding is reduced due to contingency degradation rather than satiety. However, the procedure used here for outcome devaluation is not consistent with methods used for contingency degradation in food studies, where rewards are delivered with equal probability after responding and not responding across several sessions (Hammond 1980; Dickinson and Mulatero 1989; Corbit and Balleine 2000), disrupting response–outcome (R-O) contingency but not R-O contiguity. This is used because goal-directed responding is sensitive to disruptions in R-O contingency, but both goal-directed and habitual responding may be sensitive to disruptions in R-O contiguity (e.g., after extended extinction training). In contrast, for the current studies, experimenter-administered IV cocaine is given at the start of session with no levers present, and then the levers are extended during a 10-min extinction test. We found that rats with goal-directed behavior showed an immediate reduction in responding at the start of the 10-min extinction test following experimenter-administered cocaine (Supplemental Fig. S2a), supporting a satiety-induced effect rather than a contingency degradation effect, because there was little opportunity to perceive a change in R-O contingency. Additionally, we found that responding was reduced only temporarily during the 10-min extinction test but then resumed again during the subsequent 2-h self-administration session, consistent with previous observations of temporary satiety achieved by experimenter-administered cocaine (Tsibulsky and Norman 1999).

Another way that the outcome devaluation method described here for IV cocaine differs from outcome devaluation for food is the utilization of specific satiety. Specific satiety is used to control for Pavlovian associations that are formed during instrumental training, including associations between the context and reinforcer (i.e., stimulus–outcome [S-O] associations). Both R-O and S-O associations may be sensitive to outcome devaluation (Adams and Dickinson 1981; Jonkman et al. 2010). Therefore, specific satiety for food has been used to ensure that reduced responding following outcome devaluation is mediated by instrumental (R-O) and not Pavlovian (S-O) contingencies. However, Pavlovian processes are not expected to influence outcome devaluation when the rate of reinforcement is matched across treatments (Jonkman et al. 2010). In the current set of cocaine studies, comparison groups experienced similar rates of reinforcement and S-O exposure in terms of the number of training days and cocaine infusions when comparing goal-directed versus habitual, RR20 versus RI60, and DMS versus DLS lesions. Thus, this minimizes the need to control for the influence of Pavlovian associations.

It is important to note that the dose of experimenter-administered cocaine needed to achieve satiety varies across rats because the rate of self-administered cocaine varies due to individual differences and experimental parameters used in a given study (e.g., infusion dose, schedule requirements, and time-out duration) that affect peak cocaine concentrations experienced by the animal. The dose used during these studies was calculated specifically to mimic brain concentrations reached during self-administration (satiety) and was developed and tested with particular reinforcement parameters (0.5 mg/kg/infusion, 2-h session, RR20 and RI60 schedules, <25 mg/kg total intake during the final stages of self-administration). In contrast, if high intake is permitted during cocaine self-administration, then a larger dose of IV cocaine might be required for outcome devaluation via satiety. However, larger doses have more rapid elimination (Supplemental Fig. S1f) due to metabolism via first-order kinetics, and thus cocaine concentrations in the brain may reduce below the required satiety levels across the 10-min devaluation testing session. Further studies are needed to fully understand cocaine satiety across a variety of schedules of reinforcement.

Similarities between cocaine and food response strategies

We found that the RR20 schedule biased toward sensitivity to outcome devaluation for cocaine, while the RI60 schedule biased toward insensitivity. This parallels our observations with food reinforcement and what has been observed in numerous food studies (Dickinson et al. 1983; Yin et al. 2004, 2005b, 2006; Gremel and Costa 2013). However, we observed goal-directed and habitual responders under both schedules, indicating that schedule only biases strategy but does not require the use of a particular strategy. This is similar to previous studies showing sensitivity to outcome devaluation for cocaine or alcohol despite training with the RI seeking–taking chained schedule (Olmstead et al. 2001; Zapata et al. 2010; Giuliano et al. 2021).

We found that bilateral lesions of the DMS or DLS had a strong influence on sensitivity to outcome devaluation for cocaine regardless of reinforcement schedule. First, this further emphasizes that schedule has only a biasing influence over the response strategy that guides behavior. Second, this indicates that similar neural systems underlie both food and cocaine responding. Previous work with food rewards demonstrated roles for the DMS and DLS in the acquisition and expression of goal-directed and habitual responding, respectively (Yin et al. 2004, 2005a,b, 2006; Corbit and Janak 2010; Gremel and Costa 2013). In studies using drug rewards, extended training for alcohol and cocaine has been shown to increase habitual behavior and involvement of the DLS (Zapata et al. 2010; Corbit et al. 2012, 2014b; Giuliano et al. 2019, 2021). Similarly, while the posterior DMS was involved in cocaine seeking during the early stages of self-administration, the anterior DLS was involved in cocaine seeking after well-established self-administration on a second-order schedule, further supporting roles for the DMS and DLS in goal-directed and habitual cocaine seeking (Murray et al. 2012). Although some previous work has shown that the posterior DMS, and not the anterior DMS, is involved in R-O learning for food (Yin et al. 2005b; Bradfield et al. 2013), other work has shown clear involvement of the anterior DMS in R-O learning for food and alcohol (Corbit and Janak 2010; Corbit et al. 2012). We found that lesions of either the anterior or posterior DMS led to insensitivity to outcome devaluation for cocaine (Supplemental Fig. S6), indicating that both are involved in goal-directed responding for cocaine.

We observed similar sensitivity to outcome devaluation on both the seeking lever (the distal link in the chain) and the taking lever (the proximal link in the chain) for cocaine responding (Supplemental Fig. S3). In contrast, previous food studies found that the proximal link was sensitive to outcome devaluation, while the distal link was insensitive; the distal link only showed sensitivity to devaluation when there was an opportunity for incentive learning via experiencing the reward in the devalued state (Balleine 1992; Balleine et al. 1995, 2005). Interestingly, we observed sensitivity to outcome devaluation on the distal link for both cocaine and food responding. Although it is unclear why our results differ from previous work, one potential reason is that in previous food studies, the seeking and taking links in the chain were both present at all times. In contrast, in our study, the seeking and taking levers were presented separately, with only one lever extended at any time (similar to Olmstead et al. 2000, 2001). This may be related to why we observed sensitivity to outcome devaluation on the seeking lever.

Finally, although we showed increased habitual responding for cocaine with the RI60 schedule and DMS lesions, we did not test whether habitual responding was increased after overtraining. Previous work with food rewards has shown goal-directed responding after limited training and habitual responding after extended training (Adams 1982); however, the development of habits seems more heavily biased by schedule of reinforcement than by length of training (Dickinson et al. 1983; Vandaele et al. 2017). Additionally, given that a history of cocaine exposure has been shown to bias toward habits even after limited training (LeBlanc et al. 2013; Corbit et al. 2014a), overtraining did not seem like a suitable method for testing the efficacy of the outcome devaluation procedure.

Conclusions

We developed an outcome devaluation procedure for IV cocaine using experimenter-administered cocaine and showed that this simple method can be used for discriminating cocaine seeking as goal-directed or habitual. We found that both goal-directed responding and habitual responding are observed for cocaine self-administration, and that this is influenced by schedule of reinforcement but also other factors as well. This novel outcome devaluation method will be valuable for investigating the role of habits in cocaine addiction and compulsive seeking.

Materials and Methods

Animals

Male Sprague Dawley rats (initial weight 225–250 g; Charles River) were single-housed in a temperature- and humidity-controlled facility at Texas A&M University accredited by AAALAC. Rats were housed under a reversed 12-h light/dark cycle (lights off at 6:00 a.m.) with food and water access ad libitum, except when noted below. All experiments were approved by the IACUC at Texas A&M University and conducted according to specifications of the National Institutes of Health as outlined in the Guide for the Care and Use of Laboratory Animals.

Surgery

For cocaine self-administration studies, rats were anesthetized via isoflurane (induction 5%, maintenance 1%–3%), given a nonsteroidal anti-inflammatory analgesic (2 mg/kg ketoprofen, s.c.), and implanted with chronic indwelling IV jugular catheters, as previously described (Smith et al. 2009). Beginning 3 d after surgery, catheters were flushed once daily with 0.1 mL of 100 mg/mL cefazolin and 0.05 mL of 100 U/mL heparin. Self-administration sessions began after at least 5 d of recovery from surgery.

For lesion experiments, rats were placed into a stereotaxic frame (Kopf Instruments) immediately following IV catheter implantation. Atropine (0.5 mg/kg, s.c.) was administered to reduce salivary and bronchial secretions associated with postsurgical pentobarbital anesthesia. A skin incision was made over the skull, and holes were drilled into the skull over the targeted injection sites. Pulled glass micropipettes attached to motorized injectors (Nanoliter 2010, World Precision Instruments) were bilaterally inserted into the DMS (anterior: AP +0.9, ML +2.9 with 6° angle, DV −5.3 from bregma; posterior: AP +0.4, ML +2.9 with 6° angle, DV −5.3 from bregma) or DLS (anterior: AP +0.7, ML +3.7 with 6° angle, DV −4.8 from bregma; posterior: AP -3.5, ML +3.5 with 6° angle, DV −4.5 from bregma). A volume of 400 nL of 100 mM NMDA was injected per hemisphere over 5 min, after which the micropipette remained in place for 5 min to allow for diffusion and then was slowly raised over 5 min. The incision site was closed with skin staples. Following surgery, animals were kept anesthetized for 2–3 h via pentobarbital (39 mg/mL solution, 0.3-mL initial injection, 0.1-mL boosters as needed, i.p.) to protect them from the acute distress and potentially harmful effects experienced during NMDA-induced excitotoxicity, including involuntary and convulsive movements. For sham lesions, rats underwent identical surgical and postoperative procedures except that no micropipette was inserted into the brain.

Cocaine self-administration

Rats were trained on a seeking–taking chained schedule of reinforcement due to this being a commonly used schedule for studying compulsive cocaine seeking despite footshock consequences (Pelloux et al. 2007; Chen et al. 2013; Everitt and Robbins 2016). Additionally, this schedule allows assessment of cocaine-seeking and -taking behaviors separately, which have distinct motivational processes (Olmstead et al. 2000, 2001). At the final stage of training, rats self-administered IV cocaine (0.5 mg/kg per infusion, pump speed of 70 µg/sec) on a seeking–taking chained schedule similar to that developed by Olmstead et al. (2001), in which completion of pressing requirements on a seeking lever gave access to a taking lever during daily 2-h sessions. All infusions were paired with 5-sec tone and light cues (78 dB, 2900 Hz; white stimulus light above the active lever). Operant conditioning chambers were housed in sound-attenuating cubicles and controlled via MED-PC IV (Med-Associates). Cocaine HCl was obtained as a gift through the NIDA Drug Supply Program.

To train animals (Fig. 1), self-administration began with fixed ratio (FR) 1 reinforcement, with only the taking lever available (criterion of five sessions and ≥20 infusions). Rats were food-restricted (85%–90% of free-feeding weight) at the start of the experiment to increase general motivation and were placed back onto free feeding once they had at least two consecutive sessions where they earned ≥20 infusions. Training then progressed to a chained seeking–taking schedule with FR1 (seeking)–FR1 (taking) reinforcement, during which completion of the seeking link of the chain led to retraction of the seeking lever and extension of the taking lever; completion of the taking link of the chain delivered cocaine and led to retraction of the taking lever and the start of the next trial. During the seeking link of the chain, a stimulus light (S+) was presented above the seeking lever and signaled availability. The criterion for this stage before moving to the subsequent stage was 2 d with ≥15 infusions. At the next stage, most experimental groups were given a 4-min time-out between trials, such that completion of the taking link of the chain led to retraction of the taking lever and extension of the seeking lever, but with no S+ and no programmed consequence for responding (criterion of 2 d with ≥15 infusions). Training then progressed to RR or RI seeking schedules, and the taking lever was available for only 60 sec or until an infusion was earned (FR1), whichever occurred first. Each animal was trained on only one schedule (either RR or RI), for between-subject comparisons. For experiment 1, some rats were trained on FR20 seeking–taking instead. For the RI schedule, the first press on the seeking lever initiated the start of the random interval, and then the first press made following the random interval completed the schedule. Training for the seeking lever began at RR3 or RI10 (criterion of 2 d with ≥15 infusions), progressed to RR10 or RI30 (criterion of 2 d with ≥15 infusions), and then to the final schedule of RR20 or RI60 (criterion of 5 d with ≥15 infusions). The MED-PC program determined the random ratio or interval for a given trial via a probability function (i.e., 0.05 probability per lever press for RR20; 0.0166 probability per second for RI60). Animals were removed from studies if they did not meet the minimum criteria after 2 wk at a given stage of training.

Food self-administration

A separate group of rats was trained to self-administer food pellets instead of IV cocaine. Rats were mildly food-restricted for the entire experiment (25 g/d, fed >1 h after the operant conditioning session ended). Rats were trained on a similar seeking–taking chained schedule of reinforcement (RR20 or RI60), except that a press on the taking lever resulted in delivery of a food pellet (45-mg plain purified pellets; Bio-Serv) with tone and light cues. Some rats experienced a time-out between trials (although only 1 min and the seeking lever was retracted), and some rats continued without a time-out. Sessions were limited to 1 h or 30 rewards, whichever occurred first, so that the total trials per session were comparable with cocaine studies.

Outcome devaluation for cocaine

Once animals were trained on the final seeking–taking schedule (criterion of 5 d with ≥15 rewards), outcome devaluation was tested across consecutive days in a within-subject manner (devaluation and nondevaluation days, counterbalanced order). As shown in Figure 1B, on the day of outcome devaluation, rats were placed into the operant conditioning chambers, and after 5 min were given experimenter-administered IV cocaine, consisting of 10 µL (to fill the catheter volume) plus a dose based on peak brain cocaine concentration estimated during self-administration, given in increments of 0.5 mg/kg infusions separated by 20 sec. After a 60-sec waiting period, the seeking lever was available with S+ for 10 min under extinction conditions. On the day of nondevaluation, no infusions were administered, but animals spent a similar amount of time in the chamber prior to starting the 10-min extinction test. Devaluation responding and nondevaluation responding were normalized per rat, such that the number of lever presses on one session was divided by the total lever presses on both sessions (e.g., devaluation lever presses/[devaluation lever presses + nondevaluation lever presses]). Each 10-min devaluation or nondevaluation test was followed by a 5-min period with no levers extended and then the start of a typical cocaine self-administration session. For acclimation purposes, at least 2 d prior to the first devaluation test, rats were given a 10-min extinction session similar to the nondevaluation day, based on preliminary data indicating that this reduced variability on subsequent devaluation test days.

Outcome devaluation for food

Once animals were trained on the final seeking–taking schedule (criterion of 5 d with ≥20 rewards), outcome devaluation was tested via sensory-specific satiety. Rats were allowed to free feed on either 45-mg plain purified food pellets (for the devaluation day; the same pellets earned during self-administration) or 15% sucrose solution (for the nondevaluation day) in the home cage for 1 h prior to being placed into the operant conditioning chamber. After a 5-min waiting period, the seeking lever was available for 10 min under extinction conditions, and then rats were returned to the home cage. A normal self-administration session took place the next day, and then the second test (devaluation or nondevaluation, counterbalanced across rats) took place the following day. Devaluation responding and nondevaluation responding were normalized per rat, as described above for cocaine outcome devaluation. Rats were acclimated to 15% sucrose in the home cage for 3 h/d for ≥3 d prior to the first devaluation test.

Experimental design

Experiment 1: To evaluate sensitivity to experimenter-administered cocaine (Fig. 2), rats were trained on cocaine self-administration and then given repeated testing with different doses of IV cocaine. One group was trained on a seeking–taking chained schedule using RR20 (n = 3), RI60 (n = 7), or FR20 (n = 1) without a time-out and was given five test sessions (1.0, 1.5, 2.0, or 2.5 mg/kg IV cocaine, or no infusion) (Fig. 2F), with the latter representing the nondevaluation session; dose order was counterbalanced in some rats (n = 6) but not in others (n = 5). A second group was trained on a seeking–taking chained schedule using RR20 (n = 3), FR20 (n = 2), or RI60 (n = 5) requirements with a 4-min time-out and was given four test sessions (0.5, 1.0, or 1.5 mg/kg IV cocaine, or no infusion) (Supplemental Fig. S1a); the maximum dose of 1.5 mg/kg was chosen because it yields a brain cocaine concentration higher than that obtained during self-administration with a 4-min time-out. This group was then allowed to self-administer cocaine without a time-out for ≥2 wk and was given an additional six test sessions (0.5, 1.0, 1.5, 2.0, or 2.5 mg/kg IV cocaine, or no infusion) (Fig. 2F). A third group was trained on an FR1 taking schedule for only eight to 11 sessions until they showed consistent cocaine self-administration (four consecutive sessions with ≤20% difference in infusion number as compared with the highest intake session; n = 16). This group was then given three to four test sessions (2.0–6.0 mg/kg IV cocaine, based on cocaine intake during self-administration, and no infusion) (Supplemental Fig. S1c). All three groups were analyzed together for correlations (Fig. 2C,I,J).

Experiment 2: To test the consistency and reproducibility of responding after experimenter-administered IV cocaine, two groups of rats were given two outcome devaluation tests on consecutive weeks (Supplemental Fig. S3). Following the first set of tests (devaluation and nondevaluation) on the seeking lever, a second set of tests occurred in the same order the next week (order counterbalanced among animals but maintained within animals). The second set of tests either involved access to the seeking lever for 10 min (mimicking the first set of tests; n = 16; 11 RI and five RR) or instead involved access to the taking lever for 10 min (n = 16; 10 RI and six RR). All rats received 1 mg/kg cocaine for outcome devaluation, based on self-administration satiety levels.

Experiment 3: To test the effects of IV cocaine on food responding (Fig. 3), two groups of rats were trained to self-administer food on a seeking–taking chained schedule (criterion: ≥10 pellets/session on final four sessions) and were then given experimenter-administered IV cocaine (n = 4 RR, n = 4 RI for 1.0 mg/kg cocaine; n = 4 RR, n = 1 RI for 2.5 mg/kg cocaine). In the same manner as used for cocaine-trained rats, food-trained rats had access to the seeking lever for 10 min under extinction conditions on two consecutive days, and one of these days (counterbalanced order) was preceded by administration of IV cocaine. These rats were naïve to cocaine, so this was their first and only exposure to cocaine.

Experiment 4: Separate groups of rats were used to test the influence of schedules and time-out on response strategy for cocaine or food self-administration (Fig. 4). After training on either an RR20 or RI60 seeking–taking chained schedule with or without a time-out, rats were given outcome devaluation testing (cocaine time-out: n = 7 RR, n = 17 RI; cocaine no time-out: n = 4 RR, n = 6 RI; food time-out: n = 11 RR, n = 13 RI; food no time-out: n = 5 RR, n = 4 RI). For cocaine outcome devaluation, the cocaine dose was individualized and based on satiety levels estimated during self-administration.

Experiment 5: Separate groups of rats received sham, DMS, or DLS lesions prior to self-administration training (Fig. 5). Once the rats were on the final RR20 or RI60 seeking–taking chained schedule for cocaine self-administration, they were given outcome devaluation testing for IV cocaine (sham: n = 8 RR, n = 11 RI; DMS: n = 4 RR, n = 7 RI; DLS: n = 3 RR, n = 7 RI). For cocaine outcome devaluation, the cocaine dose was based on satiety levels estimated during self-administration.

Histology

At the end of experiments, rats were deeply anesthetized via inhaled isoflurane and perfused transcardially with 0.9% NaCl and 10% neutral-buffered formalin (NBF) via a peristaltic pump (NaCl at 70 mL/min for 120 sec, and NBF at 140 mL/min for 20 sec and then 70 mL/min for 60 sec). Brains were removed, postfixed overnight in 10% NBF at 4°C, and transferred to 20% sucrose in phosphate-buffered saline (PBS) with 0.02% sodium azide for at least 2 d prior to freezing, sectioning on a cryostat at 40 µm, and storing in PBS-azide.

To assess NMDA lesion spread and location, brain sections were stained using immunohistochemistry for NeuN, which we found provides a more precise assessment of lesion borders as compared with Nissl staining. Free-floating brain sections were incubated overnight at room temperature in mouse anti-NeuN (1:5000; Millipore MAB-377, RRID:AB_2298772) in PBS with 0.25% Triton X-100 (Sigma-Aldrich). Following each incubation step, tissue was rinsed three times in PBS for 1 min each. Tissue was incubated 30 min in biotinylated secondary antibody (donkey antimouse; 1:500; Jackson ImmunoResearch, RRID:AB_2340785), followed by 45 min in VectaStain Elite ABC (1:500; Vector Laboratoriess). The reaction was visualized with 0.025% 3,3′-diaminobenzidine tetrahydrochloride (DAB; Sigma-Aldrich D5637), 0.015% hydrogen peroxide, and 0.5% nickel ammonium sulfate in 2 mL of PBS for 10 min. Following staining, sections were mounted onto glass slides and dried, and then slides were coverslipped with Permount (VWR). Lesion spread and location were plotted using Adobe Illustrator. We delineated the boundaries of the DMS and DLS according to the targeting of cortical afferents from the prelimbic cortex and anterior cingulate described by Berendse et al. (1992). Rats were excluded from behavioral analyses if lesions were too small, mixed (hitting both the DMS and DLS), or asymmetrical bilaterally. To be included, we required lesions to be at least 1 mm² in size in the DV/ML planes and at least 0.5 mm in the AP plane within the boundaries of the desired area (the DMS or DLS) on both hemispheres, without also spreading into the converse area (the DLS or DMS) with a size of 1 mm² in the DV/ML planes and 0.5 mm in the AP plane in either hemisphere.

Pharmacokinetic estimation of cocaine concentration

A two-compartment model was used to estimate cocaine concentration in the brain every 30 sec during the self-administration session. The concentration of cocaine in the brain (c₂) was estimated at the time (t) after cocaine infusion at a specific dose (d) with an apparent volume of distribution (v₂) using the following equations (Pan et al. 1991):

$$c_2={\displaystyle \frac{d{k}_{12}}{v_2\left(\alpha -\text{β}\right)}\left(e^{-\text{β}t}-e^{-\alpha t}\right),}$$

$$\alpha =0.5\left[\left(k_{12}{k}_{21}{k}_{\mathrm{el}}\right)+\sqrt{{\left(k_{12}{k}_{21}{k}_{\mathrm{el}}\right)}^2-4k_{21}{k}_{\mathrm{el}}}\right],\mathrm{and}$$

$$\text{β}=0.5\left[\left(k_{12}{k}_{21}{k}_{\mathrm{el}}\right)-\sqrt{{\left(k_{12}{k}_{21}{k}_{\mathrm{el}}\right)}^2-4k_{21}{k}_{\mathrm{el}}}\right].$$

The constants k₁₂, k₂₁, and k_el represent the rate of transfer from blood to brain and from brain to blood and elimination. Variables were based on values from the same study (Pan et al. 1991) according to brain microdialysis after chronic experimenter-administered IV cocaine: d = 1.471 µM/kg, v₂ = 0.178 L/kg, k₁₂ = 0.332 min⁻¹, k₂₁ = 0.182 min⁻¹, k_el = 0.450 min⁻¹, α = 0.870, and β = 0.094.

Data analyses

Animals were removed from all analyses if they failed to meet the criteria for self-administration (described in the Materials and Methods and in Fig. 1A) or if they failed to meet a criterion of ≥10 presses during the nondevaluation session (nine out of 88 rats were excluded based on the latter). Data were analyzed using t-tests or one-way, two-way, or three-way ANOVAs (with repeated measures when appropriate) as detailed in the Results, with Sidak's multiple comparisons tests used for post hoc analyses; post hoc results are shown in the figures. Statistical results are reported only for effects with a medium or large effect size (for ANOVAs, partial η² values ≥0.06) and significant P-values (<0.05). K-means clustering analysis was used to identify and separate sensitive and insensitive groups in Figure 2F. χ² tests were used for analysis of lesion groups in Figure 5D using one-sided tests. Correlation analyses were evaluated via the Pearson correlation coefficient (r). Normality was tested using the D'Agostino and Pearson test or the Shapiro–Wilk test (when sample sizes were n < 20). Homogeneity of variance was tested using an F-test (one-way ANOVA and t-tests) or Levene's test (two-way and three-way ANOVAs). In cases where homogeneity of variance was a concern, we performed additional analyses, such as χ² tests or Mann–Whitney signed-ranks tests.

Competing interest statement

The authors declare no competing interests.