Self-administration

Rats successfully acquired SA behavior in both the sucrose and cocaine groups (Fig. 1B, C). Overall, responding on the active lever significantly increased over sessions, but not on the inactive lever (significant main effect of Session, \(F\left(13,106.46\right)=9.18\), \(p < .001\); Lever, \(F\left(1,60.07\right)=98.61\), \(p < .001\); and Lever × Session interaction, \(F\left(13,112.30\right)=13.11\), \(p < .001\); positive linear trend for the Active Lever, \(t\left(264.63\right)=9.77\), \(p < .001\); linear trend for the Inactive Lever, \(t\left(264.63\right)=0.18\), \(p=.854\)). Acquisition of sucrose SA responding was faster than cocaine SA (main effect of SA, \(F\left(1,60.08\right)=54.39\), \(p < .001\); SA × Lever interaction, \(F\left(1,60.07\right)=20.98\), \(p < .001\); SA × Lever × Session interaction, \(F\left(13,112.30\right)=3.17\), \(p < .001\)). Active lever responding increased significantly more quickly for the sucrose than the cocaine group (linear trend, Sucrose: Active Lever, \(t\left(263.89\right)=9.20\), \(p < .001\); Cocaine: Active Lever, \(t\left(265.33\right)=4.66\), \(p < .001\); significant SA × Session linear trend interaction on Active Lever, \(t\left(264.63\right)=9.77\), \(p < .001\)); inactive lever responding did not differ (linear trend, Sucrose: Inactive Lever, \(t\left(263.89\right)=0.66\), \(p=.513\); Cocaine: Inactive Lever, \(t\left(265.33\right)=-0.39\), \(p=.699\); non-significant SA × Session linear trend interaction on Inactive Lever, \(t\left(264.63\right)=0.18\), \(p=.854\)). Males acquired responding on the active lever faster than females in the Sucrose SA group, but no significant sex differences were evident in the Cocaine SA group (see Supplemental Analysis 1 for full analysis of sex differences). Importantly, SA acquisition was similar for animals assigned to the Vehicle and D3a treatment in the next stage of the experiment (analysis including Treatment as a factor; main effects and interactions with Treatment, all p > 0.118).

Stage 1 - Preconditioning

Overall, responding during exposure to the non-reinforced cue pairs in preconditioning was low (Fig. 2). Sucrose animals spent more time in the port than Cocaine SA animals (main effect of SA: Suc > Coc, \(F\left(1,56\right)=6.69\), \(p=.012\); no main effect of Treatment: Veh vs D3a, \(F\left(1,56\right)=2.60\), \(p=.113\); no SA × Treatment interaction, \(F\left(1,56\right)=0.65\), \(p=.422\)). Responding was also higher to the first stimulus in each pair (main effect of Stimulus: S1 > S2, \(F\left(1,56\right)=11.09\), \(p=.002\)), and slightly higher to the AB than the CD cues (main effect of Cue: AB > CD, \(F\left(1,56\right)=4.19\), \(p=.045\); but no Cue × Stimulus interaction, \(F\left(1,56\right)=0.06\), \(p=.811\)). Importantly, there were no significant interactions between SA, Treatment, and Stimulus order or Cue pairs (all p > 0.060).

Fig. 2: A history of cocaine self-administration disrupts sensory preconditioning (SPC) in rats treated with vehicle prior training sessions.

The D3R antagonist VK4-116 effectively treats sensory preconditioning deficits in rats with a history of cocaine self-administration. Following SA training and 4 weeks of withdrawal, rats were trained on the SPC task. Prior to each session, rats were treated with either Vehicle or the D3R antagonist VK4-116 (15 mg/kg; i.p.). A Suc_Veh: Sucrose SA rats given Vehicle. B Coc_Veh: Cocaine SA rats given Vehicle. C Suc_D3a: Sucrose SA rats given VK4-116. D Coc_D3a: Cocaine SA rats given VK4-116. Stage 1 - Preconditioning (A- > B, C- > D), Stage 2 - Conditioning (B- > Pellets, D- > No-outcome), and Probe test (A, C, B, D: in extinction) behavior are presented for each group (left, middle, right). Discriminative responding to the cues is depicted as the duration of time spent in the food cup during the CS, above the pre-CS baseline (mean ± SEM). E The Conditioning effect: The difference in responding to cues B and D during the probe test provides an index of the conditioning effect such that scores above 0 reflect successful conditioning. # = Significant Conditioning effects: Responding to B was greater than D in vehicle (main effect of Cue, Vehicle: Cue B > D, \(t\left(56\right)=8.50\), \(p < .001\)) and D3a treated groups (D3a: Cue B > D, \(t\left(55.90\right)=7.43\), \(p < .001\)). N.S. = Magnitude of Conditioning effects were not significantly different: Magnitude of Cue (B > D) differences did not interact with SA experience in Vehicle (Vehicle: SA × Cue \(t\left(56\right)=0.66\), \(p=.512\)) or D3a treated groups (D3a: SA × Cue, \(t\left(55.90\right)=-1.46\), \(p=.149\)). F The SPC effect: The difference in responding to cues A and C during the probe test provides an index of the SPC effect such that scores above 0 reflect successful SPC. Sex and individual differences in the conditioning and SPC effect are depicted in Supplementary Fig. 2. Difference scores were calculated as the difference in discriminative responding to each cue from the Probe test in A–D (mean ± SEM). # = The SPC effect (A > C) was significant in Suc_Veh (Suc_Veh: A > C, \(t\left(56\right)=2.58\), \(p=.013\)) but not Coc_Veh rats (Coc_Veh: A vs C, \(t\left(56\right)=-1.05\), \(p=.297\)), reflecting a significant SA × Cue interaction for S1 stimuli (Veh: SA × Cue, \(t\left(56\right)=-2.51\), \(p=.015\)). This reflected a significant SPC effect across both Suc_D3a and Coc_D3a groups (D3a: SA × Cue, A > C, \(t\left(55.90\right)=2.02\), \(p=.048\)). N.S. = The SPC effect was of similar magnitude in the Suc_D3a and Coc_D3a treated groups (D3a: SA × Cue, \(t\left(55.90\right)=0.26\), \(p=.797\)). Overall, the depicted pattern of differences in E and F are supported by significant SA × Treatment × Cue × Stimulus interaction (\(F\left(1,112\right)=5.90\), \(p=.017\)), and this interaction was decomposed using separate ANOVAs for each Vehicle and D3a Treatment conditions. Alternatively, separate Treatment × SA × Cue analyses of the stimuli in E and F were also consistent with the reported pattern of significance. The Conditioning effect (E) did not differ between groups (main effect of Cue: B > D \(F\left(1,56\right)=124.61\), \(p < .001\); no main or interaction effects between SA, Treatment, and Cue, p > 0.130). The SPC effect was also consistent with the reported pattern (SA × Treatment × Cue interaction approached significance, \(F\left(1,56\right)=3.76\), \(p=.058\)). See Supplementary Analysis 4 for additional statistical support and considerations.

A full analysis of any sex differences in SPC is reported in the supplementary analyses; however, any observed significant sex differences were transient, and were not observed by the end of conditioning in stage 2, or during the critical probe tests.

Stage 2 - Conditioning

All four groups successfully increased responding to cue B more than cue D by the end of conditioning (main effect of Cue, \(F\left(1,56\right)=726.65\), \(p < .001\); Session, \(F\left(5,560\right)=79.20\), \(p < .001\); and significant Cue × Session, \(F\left(5,560\right)=13.42\), \(p < .001\)). However, it is important to note that pellets were delivered during presentations of cue B, and responding in this stage reflects both anticipation and consumption of the pellets. Responding to cues B and D during the probe test (i.e. in extinction) can provide a test of any differences in anticipatory responding without this confound.

There was greater cue discrimination (B > D) in the sucrose than the cocaine SA groups (SA × Cue interaction, \(F\left(1,56\right)=6.35\), \(p=.015\); significant cue discrimination in both SA conditions, Sucrose: B > D, \(t\left(56\right)=21.08\), \(p < .001\); Cocaine: B > D, \(t\left(56\right)=17.09\), \(p < .001\)), however there was also a greater linear increase in responding to both cues in the cocaine than sucrose groups (SA × Session interaction, \(F\left(5,560\right)=5.39\), \(p < .001\); significant positive linear trend in both SA conditions: Sucrose, \(t\left(560\right)=9.60\), \(p < .001\); Cocaine, \(t\left(560\right)=14.92\), \(p < .001\); Overall responding main effect of SA: Suc > Coc, \(F\left(1,56\right)=14.42\), \(p < .001\)).

When looking at the effects of Treatment groups, there was greater cue discrimination (B > D) in the Vehicle than the D3a groups (Treatment × Cue interaction, \(F\left(1,56\right)=8.21\), \(p=.006\); significant cue discrimination in both Treatment conditions, Vehicle: B > D, \(t\left(56\right)=21.04\), \(p < .001\); D3a: B > D, \(t\left(56\right)=17.07\), \(p < .001\)), however there was also greater linear increase in responding to all cues in the D3a than vehicle Treatment groups (Treatment × Session interaction, \(F\left(5,560\right)=11.27\), \(p < .001\); significant positive linear trend in both SA conditions, Vehicle: linear trend, \(t\left(560\right)=7.19\), \(p < .001\); D3a: linear trend, \(t\left(560\right)=17.41\), \(p < .001\); and an overall main effect of Treatment: Veh > D3a, \(F\left(1,56\right)=12.40\), \(p=.001\)).

Overall, the effects of SA and Treatment did not interact significantly (SA × Treatment and higher order interactions with Cue and Session, p > 0.054).

Final session

Focusing on the final day of conditioning (Session 6), all groups responded significantly more to cue B than D (main effect of Cue, \(F\left(1,56\right)=420.36\), \(p < .001\)). Between groups, there were no differences in responding to cue B, however responding to cue D was higher in the Cocaine than Sucrose SA groups (SA × Cue interaction, \(F\left(1,56\right)=11.15\), \(p=.002\); Cue B: Suc vs Coc, \(t\left(102.22\right)=-1.47\), \(p=.144\); Cue D: Suc < Coc, \(t\left(102.22\right)=2.45\), \(p=.016\)), and greater in the D3a than the Vehicle Treatment groups (Treatment × Cue interaction, \(F\left(1,56\right)=6.45\), \(p=.014\); Cue B: Veh vs D3a, \(t\left(102.22\right)=-0.37\), \(p=.714\); Cue D: Veh < D3a, \(t\left(102.22\right)=2.62\), \(p=.010\)). All remaining main effects and interactions between SA, Treatment, Cue, and Sex failed to reach significance (p > 0.100).

Stage 3 - Probe test

During the probe test (Fig. 2A–D), all groups demonstrated successful stage 2 conditioning (i.e., the conditioning effect, B > D; Fig. 2E). In contrast, all groups except for the untreated cocaine group (Coc_Veh), responded more to cue A than C (i.e. the SPC effect, A > C; Fig. 2F). This pattern of group differences was supported by a significant SA × Treatment × Cue × Stimulus interaction (\(F\left(1,112\right)=5.90\), \(p=.017\); a main effect of Cue pair, AB > CD, \(F\left(1,112\right)=89.16\), \(p < .001\); and a Cue × Stimulus interaction, \(F\left(1,112\right)=41.78\), \(p < .001\); all remaining effects, including sex differences did not reach significance, p > 0.098; Sex and individual subject data points corresponding to Fig. 2E, F are presented in Supplementary Fig. 2), and explored with separate follow-up analyses within each Treatment condition (additional analysis within each SA condition presented in Supplementary Analysis 4).

Vehicle Treatment

In the Vehicle-treated groups, sucrose SA rats showed a significant SPC effect that was abolished in cocaine SA rats. This was supported by a significant SA × Cue × Stimulus interaction (\(F\left(1,56\right)=5.01\), \(p=.029\)), such that the SPC effect (A > C) was significant in Suc_Veh (Suc_Veh: A > C, \(t\left(56\right)=2.58\), \(p=.013\)) but not Coc_Veh rats (Coc_Veh: A vs C, \(t\left(56\right)=-1.05\), \(p=.297\); reflecting a significant SA × Cue for S1 stimuli, S1: SA × Cue \(t\left(56\right)=-2.51\), \(p=.015\)). Importantly, both groups showed significant evidence of successful discriminative responding to the previously rewarded cue B compared to the non-reinforced cue D (Suc_Veh: B > D, \(t\left(56\right)=5.88\), \(p < .001\); Coc_Veh: B > D, \(t\left(56\right)=6.15\), \(p < .001\)), and this effect was of similar magnitude in both groups (no SA × Cue interaction for S2 stimuli, S2: SA × Cue \(t\left(56\right)=0.66\), \(p=.512\)). This finding successfully replicates our earlier report that a history of cocaine SA disrupts SPC in rats, and it further extends this by suggesting that there are no significant sex differences in this effect (Supplementary Fig. 2).

D3a treatment

In D3a treated groups, the SPC effect was significant and did not differ between sucrose and cocaine SA rats. This was supported by a significant Cue × Stimulus interaction (\(F\left(1,28\right)=15.25\), \(p=.001\); and significant main effect of Cue pair, AB > CD, \(F\left(1,28\right)=42.89\), \(p < .001\)) that did not differ between Suc_D3a and Coc_D3a groups (no Treatment × Cue × Stimulus interaction, \(F\left(1,28\right)=1.55\), \(p=.224\)). This reflected a significant SPC effect across both Suc_D3a and Coc_D3a groups (A > C, \(t\left(55.90\right)=2.02\), \(p=.048\)), as well as a significant, albeit larger, effect of discriminative responding (B > D, \(t\left(55.90\right)=7.43\), \(p < .001\)). Notably, both of these effects were of similar magnitude in both groups (no SA × Cue interaction for S1 or S2 stimuli; SPC effect, S1: SA × Cue, \(t\left(55.90\right)=0.26\), \(p=.797\); conditioning effect, S2: SA × Cue, \(t\left(55.90\right)=-1.46\), \(p=.149\)). Therefore, treating cocaine-experienced rats with the D3-receptor antagonist VK4-116 successfully recovered deficits in SPC (see also Supplementary Analysis 4).

Group differences in SPC task interpretation

Although no cues were used in our self-administration sessions, it is nevertheless plausible that different histories of cocaine or sucrose SA (i.e. Suc_Veh and Coc_Veh) could change the nature of how the cues are represented, integrated, or generalized across the stages of the SPC task, i.e. how the task is “solved”. In control animals, the SPC effect can be supported by a number of mechanisms that are likely to be a parameter dependent [41, 42]. In the present version of the task, the predominant mechanism in controls is likely to be associative inference during the probe test [9, 43] i.e. at test, responding to cue A is driven by recall of the A- > B (stage 1) and B->Pellet (stage 2) associations that are integrated to infer that A- > B->Pellet. This is suggested by several lines of evidence, including the OFC-dependence of responding in the probe test [8, 44], as well as the failure of the preconditioned cue to support conditioned reinforcement [43], both of which are contrary to other mechanisms, especially mediated learning, in which value would accrue directly to the preconditioned cue [41]. This solution to SPC is disrupted by a history of cocaine use, both here in the Coc_Veh group as well as in our prior study [10]. If VK4-116 (D3a) is successfully “treating” the effect of Cocaine SA, then the Coc_D3a group should also have a task solution that is more similar to the Suc_Veh than the Coc_Veh groups. We tested the similarity of the group solutions using two complementary approaches: (1) testing specific predictions about the A~B and C~D relationships in each group, and (2) comparing the similarity of the overall behavioral task solution between groups.

To test this, we first examined probe test behavior to see if responding to cues A and C was related to the level of conditioning to cues B and D, i.e. is there an S1- > S2 relationship? Specifically, in the Suc_Veh group we expected a strong positive correlation between learning about reinforced cue B and responding to cue A and no correlation between non-reinforced cue D and cue C during the probe test (reflecting low levels of magazine responding to cue D that are likely driven by factors other than reward expectation). In contrast, in the Coc_Veh group we expected that responding would not be correlated between cues pairs A-B and C-D. Importantly, we predicted that the D3a treatment in the Coc_D3a group would recover the strong positive correlation between responding to cues A and B, but not C and D. The pattern of group differences in slopes between cues A~B and C~D was compared by fitting a linear model predicting responding to the first stimulus in each pair (S1: A/C) with responding to the relevant second cue in each pair (S2: B/D), as well as categorical factors of Cue pair (AB/CD), Treatment, and SA.

Consistent with our predictions (Fig. 3A), there was a significant positive correlation between A and B in the Suc_Veh and Coc_D3a groups but not in the Coc_Veh group or, unexpectedly, in the Suc_D3a group (Suc_Veh: A~B, \(b=0.75\), 95% CI \(\left[0.35,1.15\right]\), \(t\left(16\right)=3.96\), \(p=.001\); Coc_D3a: A~B, \(b=0.57\), 95% CI \(\left[0.20,0.95\right]\), \(t\left(15\right)=3.26\), \(p=.005\); Coc_Veh: A~B, \(b=0.09\), 95% CI \(\left[-0.66,0.83\right]\), \(t\left(12\right)=0.25\), \(p=.805\); Suc_D3a: A~B, \(b=-0.22\), 95% CI \(\left[-0.70,0.26\right]\), \(t\left(13\right)=-0.99\), \(p=.340\)). These slope differences between SA and Treatment groups was supported by a significant SA × Treatment interaction (A~B: SA × Treatment, \(F\left(1,56\right)=9.68\), \(p=.003\)), which reflected significant differences between slopes for Coc_Veh and Suc_Veh groups (A~B: Coc_Veh < Suc_Veh, \(t\left(56\right)=-2.10\), \(p=.040\)), and Coc_D3a and Suc_D3a groups (A~B: Coc_D3a > Suc_D3a, \(t\left(56\right)=2.30\), \(p=.025\)).

Fig. 3: Testing whether probe test responding reflects the relationships between the cue pairs that were presented during stage 1 preconditioning and stage 2 conditioning (A- > B->Pellet, C- > D->No-outcome).

A A–B relationship: Responding to cue A is driven by learning to B in both the control group (Suc_Veh) and treated cocaine group (Coc_D3a), but not in the untreated cocaine (Coc_Veh) and treated control groups (Suc_D3a). B C-D relationship: There was no relationship between responding to cue C and D. Data points indicate probe test responding for individual subjects for the first cue (y-axis) and second cue (x-axis) in each pair, and plotted separately for each group (panels from left to right). Responding was defined as the duration of time spent in the food cup during the CS, above the pre-CS baseline. Lines and error shading from linear model fit, and corresponding model r2 and p-values presented at the bottom of each plot. The sex of individual subjects is represented as: Female = Circle, Male = Triangle.

In contrast to the A~B correlations, there were no significant relationships between cue C and D in any group (Fig. 3B; Suc_Veh: C~D, \(b=-0.81\), 95% CI \(\left[-3.01,1.38\right]\), \(t\left(16\right)=-0.79\), \(p=.443\); Coc_Veh: C~D, \(b=0.74\), 95% CI \(\left[-0.59,2.08\right]\), \(t\left(12\right)=1.21\), \(p=.249\); Suc_D3a: C~D, \(b=-0.19\), 95% CI \(\left[-1.01,0.63\right]\), \(t\left(13\right)=-0.50\), \(p=.625\); Coc_D3a: C~D, \(b=0.38\), 95% CI \(\left[-0.36,1.11\right]\), \(t\left(15\right)=1.09\), \(p=.292\)). The lack of significant group differences in the C-D relationship was consistent with a non-significant SA × Treatment interaction (C~D: SA × Treatment, \(F\left(1,56\right)=0.68\), \(p=.414\); all remaining main effects and interactions with cue D, p > 0.399). Finally, this overall pattern of group differences in slopes between cues A~B but not C~D was supported by a significant SA × Treatment × Cue × S2 interaction (\(F\left(1,112\right)=4.06\), \(p=.046\)).

These results support our predictions that responding to cue A was uniquely related to learning about cue B in control rats (Suc_Veh), which was disrupted by a history of cocaine use (Coc_Veh) but successfully recovered by D3a treatment (Coc_D3a). Surprisingly, the D3a treatment disrupted the A~B relationship in sucrose-control rats (Suc_D3a) but not the SPC effect. This result is important because it indicates that the mitigation of the cocaine-related deficits is not due to an independent effect of VK4-116 to somehow improve normal performance in preconditioning. Indeed, it supports the idea that VK-116 is having an effect in the cocaine-trained group that reflects an interaction with changes caused by cocaine use.

Behavioral similarity

Next, we tested the similarity of the overall task solutions in each group using a behavioral similarity analysis, an approach based on representational similarity analysis [45]. How a task is solved is likely to be reflected in the pattern of multivariate relationships both between and within each cue. To capture the pattern of responding within each cue, the 10 s cue period was split into early and late epochs (i.e. 5 s bins), a standard approach [46,47,48] that also reflected the observed pattern of responding within-cue (Supplementary Figs. 3–4). A Pearson cross-correlation matrix was calculated across all 8 (cue [4] × time [2]) epochs to create a behavioral similarity matrix for each group (Fig. 4). These behavioral similarity matrices represent the multivariate patterns of probe test responding within each group, i.e. a group level index of the overall task solution. Finally, we tested whether groups used similar overall task solutions by comparing behavioral similarity matrices between pairs of groups using Spearman’s rank correlations. Evidence of significant similarity was found between Suc_Veh and Coc_D3a groups (Spearman’s rank correlation, Suc_Veh vs Coc_D3a, \(_}}}}}}=.76\), \(S=894.00\), \(p < .001\)), but not between the other groups (Suc_Veh vs Coc_Veh, \(_}}}}}}=.15\), \(S=3,088.00\), \(p > .999\); Suc_Veh vs Suc_D3a, \(_}}}}}}=.29\), \(S=2,594.00\), \(p=.805\); Coc_Veh vs Coc_D3a, \(_}}}}}}=-.02\), \(S=3,716.00\), \(p > .999\); Coc_Veh vs Suc_D3a, \(_}}}}}}=.44\), \(S=2,052.00\), \(p=.123\); Coc_D3a vs Suc_D3a, \(_}}}}}}=.28\), \(S=2,622.00\), \(p=.871\); Holm-Sidak corrected). These findings provide further evidence that D3a treatment returned the performance of cocaine-experienced rats to what is normally observed, and that this effect was selective, representing an interaction with changes caused by cocaine use, and not simply an independent effect of VK4-116 to improve normal inference in the preconditioning task.

Fig. 4: Behavioral similarity analysis of probe test responding.

A Behavioral similarity matrices for Sucrose (left) or Cocaine (right) SA groups, and Vehicle (top) or D3R antagonist (bottom) treatment groups. For each group, the similarity of responding between and within cues was quantified by separating responses within each cue into 5 s time bins (Supplementary Fig. 3) and generating a cross-correlation matrix between all the resultant cue epochs. Color values are plotted to represent the correlation values (Pearson’s r), to provide a visual summary of the group specific pattern of response relationships. The behavioral similarity patterns were similar between the untreated control group (Suc_Veh) and the cocaine group treated with the D3R antagonist (Coc_D3a), but different to the untreated cocaine group (Coc_Veh) and the control group treated with the D3R antagonist (Suc_D3a). B Behavioral similarity analysis: A Spearman correlation was used to test the similarity between the group similarity matrices. Numbers (and corresponding color values) indicate the rank correlation (Spearman’s rho) between the lower diagonal of the cross-correlation matrices above. An analysis of just the between-cue similarities (i.e. not considering within-cue time bins) supported similar conclusions (Supplementary Fig. 4).