Dissociation of decision-making from escalated drug intake within a long-access model

Abstract

General decision-making deficits induced by chronic drug intake have been suggested to underlie the development of substance use disorder (SUD). To determine the effects of differential drug self-administration (cocaine or fentanyl) history on decision making, a procedure built upon choice theory was run alongside a drug (cocaine or fentanyl) escalation procedure in rats. Isomorphic food choice was measured each day under unpredictable probabilistic conditions that required constant engagement of decision systems, followed by self-administered drug (cocaine or fentanyl) for 1or 6 hours. Despite clear intake escalation (a measure widely used as indication of dysregulated decision-making and posited as a preclinical model of SUD in which animals take increasing amounts of drug compared to their 1-hr access counterparts), isomorphic food choice did not differ between drug access groups during self-administration or subsequent forced abstinence, suggesting that decision-making remained intact and that general drug-induced deficits in decision-making are not necessarily an underlying contributor in intake escalation within the long-access preclinical model of SUD.

Introduction

A growing literature suggests that substance use disorder (SUD) is, at least partially, mediated by decision-making mechanisms [1–8]. Among a multitude of possible mechanisms, the behavioral shift in drug preference over other available alternatives has been posited to be an expression of drug-induced degradation of basic decision-making mechanisms, or a change in the relative subjective valuation of available reinforcers. Toward teasing apart competing hypotheses, concurrent choice procedures can help identify changes in decision-making mechanisms, like perseverative or habitual choice from changes in the relative value of concurrently available alternatives (c.f. [9]), through the application of quantitative modeling to choice data. Thus, quantitative modeling of decision-making allows for a more detailed and direct examination of specific predictions associated with various theories of SUD, like those of habit and relative subjective value theory.

For example, in habit theory of decision deficit, SUD is viewed as a transition from initial voluntary drug use, driven by the reinforcing effects of the drug, to habitual actions mediated by elicited stimulus-response relationships, independent of the associated outcomes [10, 11]. Actions are said to be habitual in the sense that drug-related behavior (e.g., drug choice) is no longer mediated by the value of the drug outcome; therefore, changes in reinforcer value are predicted to have no effect on responding governed by habit. Perhaps most importantly, habit theory and other decision deficit theories also posit that behavioral changes are mediated by drug-induced modulation of the mesocorticolimbic system, resulting in chronic habitual control [12–20] and enhanced incentive saliency [21]. These same neural modulations are often hypothesized to produce global decision-making dysfunction beyond the context of drug choice. For example, “frontal cortex-derived executive function disorders”, including SUD, are said to detrimentally influence the maintenance of spatial information, decision-making, behavioral inhibition, risky behavior, drug craving, and delayed gratification [18,19, 22]. Furthermore, these same neurobiological changes are posited to be driving factors in non-drug addictions [23–25] such as gambling [23, 26, 27] and obesity or food addiction [23, 28, 29], positing a unified understanding of maladaptive decision-making disorders. In this way, habit theory specifically suggests that maladaptive decisions are driven by a shift away from their original value-based mechanisms to habitual actions that should continue as long as the associated mesocorticolimbic alterations remain present. Thus, rendering the behavior habitual and inflexible for an indefinite amount of time as the neuroadaptations are also associated with the persistent risk of relapse seen in SUD [30]. Consequently, habit theory of SUD predicts that decision-making is altered after chronic substance use, namely as a long-lasting shift away from behavior sensitive to alterations in reinforcer value.

In contrast to deficit-based theories like habit theory, there are theories of SUD that do not rely on positing decision-making deficits via any accompanying neurobiological dysfunction. For example, relative subjective value theory suggests that both initial drug use as well as prolonged use are reflective of changes in relative subjective value within a specific decision context. As such, continued choice for drug alternatives is not accounted for by a disruption or failing of decision-making mechanisms per se, but instead by a shift in relative subjective value of the drug, such that it attains greater subjective value relative to other available alternatives within the specific decision-making context [6, 31–35]. Behavioral economic reviews have demonstrated that there is an increase in the relative value of drugs after continued use [36, 37]. Accordingly, relative value theory seeks to account for continued choice for drug by appealing to the gap between the relative value of drug and present non-drug alternatives, and predicts that because drug choices are based on shifts in the relative subjective value of reinforcers, rather than the failure of decision-making mechanisms; decision-making systems would remain largely intact after chronic drug use. Consequently, intact relative subjective value-based decision-making would allow for continued sensitivity to changes in the environment that alter the relative subjective value of drug reinforcement, such as a change in magnitude (e.g., drug dose).

The goal of the present study was to investigate changes in decision-making over the development of escalated stimulant (cocaine) and opioid (fentanyl) intake, the current standard for preclinical modeling of SUD [38]. To do so, an isomorphic (food-food) concurrent choice task with varying unpredictable probabilities of reinforcement was run in tandem with a drug escalation procedure, where animals had drug access for either 1 or 6 hours per day. While certainly not the sole hypothesis (e.g., allostatic modulation and incentive salience sensitization), escalated drug intake associated with 6-hr access has been posited widely to be the result of disrupted or dysregulated decision-making, where drug intake is no longer controlled by or sensitive to outcomes [39–41] and accompanied by mesocorticolimbic changes [42–44], partially giving rise to the widespread use of the procedure as a preclinical model of SUD. As such, this study aimed to compare decision-making processes in rats throughout chronic self-administration of a stimulant (cocaine) or an opioid (fentanyl) in 1-hr or 6-hr drug access conditions. If escalated drug intake within 6-hr access conditions is reflective of global dysregulated decision-making, a deficit in decision-making metrics, relative to 1-hr access, should be present, exemplified as reduced effect of reinforcer value on choice and choice independent of outcome in quantitative choice modeling. Additionally, given that relatively permanent neurobiological changes associated with 6-hr intake have been posited as the causal mechanisms underlying escalated drug intake, any disruption in decision-making metrics associated with 6-hr access should be long-lasting and impair integration of new reinforcer value information following escalated intake.

Methods

Subjects

Twenty-six male and twenty-six female Sprague-Dawley Rats (Harlan Inc.; Indianapolis, IN, USA), weighing approximately 250–275 g on arrival were used. Rats were individually housed (12:12hr light: dark cycle) with ad libitum access to water in their home cage. During food restriction, rats were maintained at approximately 85% of their free-feeding body weights. All experimentation took place during the light phase, conducted according to the 2010 NIH Guide for the Care and Use of Laboratory Animals (8^th Edition), and approved by the Institutional Animal Care and Use Committee at the University of Kentucky.

Apparatus

All experiments were conducted in Med Associates conditioning chambers (ENV-008, MED Associates, St. Albans, VT) and operated using MED-PC (see supplementary information [SI 1.1] for further information)

Initial Training

Prior to the experiment proper, rats were trained to retrieve food from the food receptable and trained to respond in a single illuminated nosepoke port following an orienting response (see Supplemental Information, SI 1.2 for further information).

Concurrent Choice task

The probabilistic choice procedure (c.f., [45–46]) was similar to initial training except completion of the orienting response illuminated both left and right nosepoke ports, allowing for a choice between options. Once a selection was made, both ports went dark, and the reward was delivered or not based upon scheduled reinforcement probability. On rewarded trials, a pellet was dispensed into the magazine, the magazine light was illuminated, and a tone was played for 5-s. On non-rewarded trials, there was a blackout of all stimuli in the chamber for 5-s (i.e., all stimuli off). Trials were separated by a 5-s intertrial interval regardless of reward status. The reward was set up on either the left or right nosepoke port based on 4 probability blocks. The left: right nosepoke ratio of reward across options for the trial blocks were 6:1 (Scheduled reward probabilities of 0.66 and 0.11 respectively), 2:1 (0.22 and 0.11), 1:2 (0.11 and 0.22), and 1:6 (0.11 and 0.66). The block order was pseudo-randomly shuffled each session. For example, if the first block was 1:6 then the second block could be either 2:1 or 1:2. Each block consisted of 30 trials, for a total of 120 trials per session. Blocks were separated by a 2-minute blackout period. Rats were trained on this task until performance was stable (no linear trend over 3 days) before surgery and self-administration began.

Catheter Surgery and Maintenance

Rats then underwent surgery for implantation of a chronic indwelling jugular catheter (see SI 1.3 for further information). Animals were given a week to recover after surgery. After each self-administration session, catheters were flushed with 0.2ml of a gentamicin-heparin-saline solution to prevent infection and preserve patency. Catheter patency was checked periodically and at the end of the self-administration portion of the experiment by muscle tone loss following 0.2ml administration of 10 μg/kg remifentanil.

Drug Self-administration and escalation

Methods were similar to those published previously [47] with minor changes. Following recovery from indwelling jugular catheter surgery, rats were trained to self-administer drug (Cocaine: 0.3 mg/kg/infusion; Fentanyl 0.25 μg/kg/infusion) on a fixed ratio 1 (FR-1) for 1 hour. Drug doses were selected as they have previously produced robust responding as a midpoint dose in cocaine choice procedures [2, 48] and in fentanyl escalation procedures [47]. Presses on the active lever resulted in a 0.1ml infusion over 5.9 s, retraction of levers, and illumination of the cue light above the active lever for 5.9 s. After the cue light turned off, the levers immediately came out again to start the next trial. Presses on the inactive lever had no consequence. Active/inactive levers were counterbalanced for side across individuals.

Animals were then assigned to one of two groups, matched for 1-hr intake during acquisition. The short-access group (Cocaine: n = 16, 7 males and 9 females; Fentanyl: n = 9, 5 males and 4 females) continued to self-administer cocaine under the 1-hr access conditions, whereas the long access group (Cocaine: n = 15, 8 males and 7 females; Fentanyl: n = 12, 6 males and 6 females) now had 6-hr access to drug (Cocaine: 21 days; Fentanyl: 10 days – escalation was robust and resulted in significant negative opioid-related side effects; as such, a 10-day timeline was used, instead of 21 to maximize animal welfare). After the escalation period, the cocaine 6-hr access animals returned to 1-hr access for a week to determine if the self-administration behavior would return to pre-escalation levels and to allow for the testing of any long-lasting changes due to escalation history before moving to forced abstinence, while fentanyl animals moved directly to forced abstinence after escalation on account of significant opioid-related side effects.

Magnitude Manipulation

To determine if there were changes in the decision-making processes after fentanyl escalation was complete, a magnitude manipulation was completed after escalation. First, animals were placed into forced abstinence and re-baselined to the choice program for 1 week. Next, the magnitude of one option was increased to 2 pellets for 1 week and finally increased again to 4 pellets.

Daily/Overall Timeline

Each day the concurrent choice task was run in the morning, before self-administration sessions. Once the rats had completed the task, they were returned to their home cages for 1 hour before they were run in self-administration sessions. Choice was measured via nosepokes, self-administration was measured via lever presses, and rats were run in different boxes for choice and self-administration sessions to avoid carryover and context effects. For additional description of the daily and overall timelines see SI 1.4.

ANALYSIS

Self-administration

Self-Administration data were calculated as the number of infusions per day. Then data were fit and compared with hierarchical linear models comparing Sex and access as predictors of the intercept and slope. Models were compared using ΔBIC values [49, 50] Only statistics from best fit models were reported.

Hierarchical linear models of global level analyses

The logarithmic form of Baum’s generalized matching equation [51] details how choice between available options is affected by different reinforcer dimensions, like probability (or reinforcement rate) and magnitude as follows [52–53]

$$log\left(\frac{B_1}{B_2}\right)=s_{R}log\left(\frac{R_1}{R_2}\right)+s_{M}log\left(\frac{M_1}{M_2}\right)+logb$$ (1)

In Equation 1, the ratio of choices for available options (B1/B2) is determined by the ratio of reinforcer probability ($\mathrm{R}$) and the reinforcer magnitude ($\mathrm{M}$) associated with each option (1/2). The effect of each reinforcer dimension on choice is scaled with sensitivity parameters ($S_R$ and $S_M$) that determine to what extent choices redistribute as the ratios associated with each reinforcer dimension change. Thus, Equation 1 can help isolate the influence of specific reinforcer dimensions on choice.

All models were compared using ΔBIC values [49,50]. Only statistics from best fit models were reported. Aggregate choice data were calculated as the average Option A/Option B choice ratio as a function of Option A’s relative reward rates and analyzed using Eq 1.

Trial level analysis

In addition to more global analytics like matching, Reinforcement Learning (RL) models can also help isolate specific facets of decision-making at a choice-by-choice resolution as shown below [54,55]:

$${\delta }_t=V_A^t-{\lambda }_A^t$$ (2)

$$V_A^{t+1}=V_A^t+\alpha {\delta }_t$$ (3)

Using the reinforcement history of a commodity, RL models construct an expected value for each option to guide decisions (e.g. ${\mathit{V}}_{\mathit{A}}^{\mathit{t}}$ 𝐟𝐨𝐫 𝐨𝐩𝐭𝐢𝐨𝐧 𝐀). Upon each choice, the difference between the expected value and realized outcome (${\mathit{\lambda }}_{\mathit{A}}$ 𝐟𝐨𝐫 𝐨𝐩𝐭𝐢𝐨𝐧 𝐀) is computed as a reward prediction error (RPE, Equation 2). The change in expected value for a choice option is then updated according to the RPE crossed with a learning rate parameter, $\mathrm{\alpha }$, which scales the extent to which expected value is updated per trial and then added to the previous expected value estimate as an updated expect value for the next trial (Equation 3). Choices for either option are predicted to fluctuate according to relative increases or decreases in their subjective value (see Equation 4, below). Thus, an RL model can detect changes in the reliance on relative valuation in the decision-making process over the course of SUD development. To model choices between options, choices for option A can be determined according to the following softmax equation(s):

$$pC{h}_A^t=\frac{1}{1+exp\left(-\left[\beta \left(V_A^t-V_B^t\right)\right]\right)}$$ (4)

$$pC{h}_A^t=\frac{1}{1+exp\left(-\left[\beta \left(V_A^t-V_B^t\right)\right]+C\left(C{h}_A^{t-1}-C{h}_B^{t-1}\right)\right)}$$ (5)

In Equation 4, the probability of choosing option A is determined by the relative difference in model-derived expected value (determined via Equations 2 and 3) and scaled according to the inverse temperature parameter, $\beta$. Additionally, Equation 5 assesses the inclusion of a perseveration parameter, C. The perseveration parameter weighs the tendency to repeat or alternate from the previous choice (Ch) independently of model-derived expected value. Together, RL models can provide a choice-by-choice view into the value-based decision-making of the organism to determine where, if at all, deviations in decision-making process occur. To further test the matching assumption of relative valuation at the trial-by-trial level, RL models were fit to individual rat concurrent choice data using the fmincon optimization algorithm in MATLAB with maximum likelihood estimation. The models (A-F) were assessed through model comparison of the concurrent choice data using ΔBIC values. Only the best fit model is detailed below. For a description of the other models and their parameter values see the Supplemental 2.1. After model selection, parameter estimates between short- and long-access groups were compared using a Wilcoxon rank-order test with a Hochberg correction.

The best fit model (E) was a simple RL model utilizing equations 2, 3, and 4 to determine the Choices for Option A according to the relative differences in value between the options. The learning rate, $\mathrm{\alpha }$ (0–1), scales RPE updates the magnitude on each trial. The estimated expected value differences across options are then scaled with the inverse temperature parameter, $\beta$. Additionally, it includes a perseveration parameter, c. The perseveration parameter weighs the tendency to repeat or alternate from the previous choice (Ch) independently of model-derived reward value. Values of 1 and −1 were assigned, with 1 representing a prior choice for Option A and −1 for option B.

RESULTS

Overall, all groups had low, stable responding during the self-administration acquisition period, and models revealed no main effects of access on acquisition. Figure 1 shows the infusions of drug throughout self-administration periods. During the escalation period, the 6-hr access groups showed an escalation in drug-taking that increased over time, whereas the 1-hr access groups remained low and stable. For the cocaine animals, the best fit model included sex as a predictor but revealed no main effects of sex, but a main effect of access [F_1,28 = 14.41, p < 0.001], where the intercept for the 6-hr access group was significantly higher than that of the 1-hr access group (80.26 and 33.34 infusions, respectively; Fig. 1A). As the sex effects were insignificant, they are not included in Figure 1 for ease of interpretation. There was also a main effect of Day [F_1,28 = 5.73, p = 0.024] and an interaction between access and day [F_1,28 = 4.59, p=0.041], such that the slope estimate for the 6-hr access group was significantly positive (1.81) while the slope for the 1-hr access group was not significantly different from 0 (0.19). Finally, when the cocaine 6-hr access rats were returned to the 1-hr access condition, the model revealed no main effect of access, such that the 6-hr access rats returned to the same level as that of the 1-hr access group [F_1,28 = 0.48, p = 0.494]. In summary, the 6-cocaine access rats took more infusions and increased their intake over the escalation period, while intake for 1-hr access rats was low and did not change over the escalation period.

Figure 1.

A) Mean (± SEM) infusions of 0.3mg/kg cocaine over acquisition (days 1–7), escalation (days 8–28), and return to 1-hr access (days 29–35). B) Mean (± SEM) infusions of cocaine fit with a simple linear regression for each group. C) Mean (± SEM) infusions of 0.25 μg/kg fentanyl over acquisition (days 1–7) and escalation (days 8–17). D) Mean (± SEM) infusions of fentanyl fit with a simple linear regression for each group. *Note the change in the Y axis for the different drugs. Error bars may be smaller than the symbol size in some cases

For the fentanyl animals, the model revealed no effect of sex, but a main effect of Access [F_1,18.6 = 22.47, p < 0.001], where the intercept for the 6-hr access group was significantly higher than that of the 1-hr access group (35.34 and 14.03 infusions, respectively; Fig. 1B). There was also a main effect of Day [F_1,18.6 = 21.06, p < 0.001] and an interaction between Access and Day [F_1,18.6 = 12.74, p = 0.002] such that the slope estimate for the 6-hr access group was significantly positive (2.65) while the slope for the 1-hr access group was not significantly different from 0 (0.59). In summary, the 6-hr fentanyl access rats took more infusions and increased their intake over the escalation period, while intake for 1-hr access rats was low and did not change over the escalation period. Overall, escalated intake of both cocaine and fentanyl was observed in the 6-hr conditions, while both cocaine and fentanyl intake remained lower and stable across the escalation period within the 1-hr conditions.

Figure 2 shows log choice ratios for Options A:B as a function of log reinforcer rates for Options A:B. Overall, both groups tended to prefer the option providing the greater reinforcer odds (or the “richer” option), with preferences being stronger the more disparate the relative reinforcer rates.

Figure 2.

A, B, C- Cocaine Experiment: Mean (± SEM) log raw ratio for choice for Option A over Option B as a function of Option A’s log relative reward rate( 1:6 as −0.77, 1:2 as −0.33, 2:1 as 0.33, 6:1 as 0.77) over four 7-day averages A) 1-hr rats (n = 16) B) 6-hr rats(n =15) C) Comparison during final week of escalation. D,E,F- Fentanyl Experiment: Mean (± SEM) log raw ratio choice for Option A over three- 5-day averages. D) 1-hr rats (n = 9) E) 6-hr rats (n = 12) F) Comparison during the final 5 days of escalation.

In the comparison of choice behavior during the escalation period, the best fit models revealed a main effect of reinforcer ratio (slope) [Cocaine: F_1,29 = 874.89, p < 0.001, Fentanyl: F_1,20 = 411.3725, p < 0.001], such that both groups had a significantly positive slope of 0.33 and 0.30 respectively, with no differences between the access groups (Fig. 2 C & F). Additionally, comparison of choice behavior over the different timepoints throughout the study, (Baseline, 1-hr access, Escalation, Return to 1-hr access, and Forced Abstinence) revealed no significant differences for the fentanyl animals (Fig. 2 D & E). However, for the cocaine animals, there was an interaction between reinforcer ratio (slope) and condition (timepoint; [F_1,29 = 11.44, p = 0.002]) such that the slope positively increased with each subsequent week, indicating improved sensitivity to relative reinforcer ratio over time (Fig. 2 A & B). In summary, there were no differences in decision-making behavior due to sex or access, and all groups were sensitive to the relative reinforcer probabilities throughout the escalated intake and through forced abstinence.

Figure 3 shows the magnitude manipulation after fentanyl escalation was completed, where the magnitude of one option increased from 1 to 2, then 4 pellets each subsequent week. The best-fit model revealed a main effect of reinforcer ratio [F_1,20.11 = 288.37, p < 0.001] and a main effect of reinforcer magnitude [F_1,20.42 =161.12, p < 0.001] but no effect of access or sex. In summary, following fentanyl escalation, all animals remained sensitive to both reinforcer probability and the novel magnitude manipulation.

Figure 3.

Mean (± SEM) Log relative reward ratio for Option A over Option B as a function of relative reward ratio for increasing magnitudes of pellet reward on one option. *Note error bars may be smaller than the symbols for some points.

Figure 4 shows an example RL model fit for individual animals and parameter estimates from the best-fit model for the final period of the escalation timepoint. RL models successfully parameterized choices and model comparisons corroborated the assumptions of the matching equation for both groups. Model comparison showed strong support for the Perseverative Rescorla-Wagner model (Model E; Equations 2, 3 and 5) with an average ΔBIC of 47.14 in the cocaine data and an average ΔBIC of 35.53 in the fentanyl data (Table S1 and S2 for full model comparison data). Visualization of the individual choice data and the model fit demonstrate the success of Model E in capturing choice behavior across trials and over days. Parameter estimate comparison between the 1-hr and 6--access revealed that there was no difference in $\mathrm{\alpha }$ estimates for cocaine or fentanyl, with a mean of 0.665 and 0.612, respectively (Fig. 4 C, See Table S3 for full comparisons). There was also no difference between 1-hr and 6-hr groups in $\beta$ estimates for cocaine or fentanyl, with a mean of 1.951 and 2.17, respectively (Fig. 4D and Table S3). And finally, there was no significant difference in the in the c parameter estimates [Cocaine: t_1,29=1.988, p = 0.056, Fentanyl: t_1,19 = 0.168, p = 0.868] with a mean of 0.541and 0.579, respectively (Fig. 4E and Table S3).

Figure 4.

Example RL model fit for the choice procedure from the final period of escalation for A) Cocaine animals (7 days) B) Fentanyl animals (5 days). C-D Mean (±SEM) parameter estimates. No significant differences were found between access group or sex. C) Alpha-learning rate. D) Beta- inverse temperature. E) C- perseveration.

DISCUSSION

The present study used an escalation procedure in tandem with a probabilistic food choice procedure to investigate changes in decision-making over the course of both stimulant (cocaine) and opioid (fentanyl) intake escalation. The results indicate that despite drastic differences in escalated drug intake (Fig. 1), there were no differences in decision-making between the two access groups. Rather, choice behavior remained largely governed by reward-associated value (Fig. 2). This was true for both drug classes, supporting previous research findings that although stimulants and opioids have different mechanisms of action and influence the dopamine system through different neural mechanisms, consistent with metanalytic analyses of decision-making (e.g., delay discounting measures) and substance use severity (alcohol, tobacco, cannabis, opiates and stimulants; c.f. [56]) and in group comparisons (alcohol, opioids, stimulants, and polysubstance use; c.f. [57]). Both matching and reinforcement learning modeling indicate animals remained sensitive to relative reward probability and magnitude. Both the longer (6-hr) and shorter (1-hr) access rats favored Option A when the relative reward rates also favored Option A and gradually shifted their preferences to Option B when relative reward rates also switched; these systematic changes in choice can be seen in aggregate within the matching analyses (Figures 1 and 2) and more precisely within the trial-by-trial analysis of choice (Fig. 4 A & B), indicating that despite escalated drug intake, rats showed no change in sensitivity to reinforcement within the decision-making procedure. Additionally, the magnitude manipulations after fentanyl escalation (Fig. 3), showed that rats retained sensitivity to changes in reward magnitude, even after escalated intake of fentanyl, indicating no deficit in integrating new information into the decision-making process post-escalation.

Habit theory of decision deficit suggests that disordered drug intake is the product of neurobiological impairments from chronic drug exposure [12–20]. Furthermore, due to the hypothesized relatively permanent nature of drug-induced neurological impairments, global decision-making is predicted to be altered after chronic substance use, these impairments are suggested to continue even after drug access has ceased entirely, and these long-lasting impairments are put forward to account for the chronic nature of SUD. In the context of the present procedures, the neurobiological changes associated with 6-hr intake escalation are posited as the cause of said escalation, itself a reflection of dysfunctional decision systems impaired by those same neurobiological changes. Consequently, chronic global drug-induced impairment of decision-making governed by habit would present as decreased sensitivity to changes in reinforcement and would be evident in flattened slopes in the matching analysis and increased values of c in the RL models. Furthermore, the long-lasting neurological adaptations proposed to accompany escalation behavior are also predicted to interfere with the integration of novel reinforcement information, as tested here via food magnitude manipulations following escalation. However, despite clear and robust escalation in the 6-hr access conditions, none of the present data, modeling, or model comparison analyses support any of the above hypotheses. The continued integration of new information after escalation aligns with findings from Singer et al., 2018 [58] in which rats showed continued behavioral flexibility through solving puzzles to access self-administration after prolonged intermittent access (IntA) to cocaine, another procedure said to induce the addition-like behavioral and neurobiological phenotype [59–61].

In contrast, a relative value perspective posits that it is an increase in the relative value of the drug, rather than neurobiological impairments in decision-making, that leads to the increased consumption seen in the 6-hr access condition (greater relative value of taking the drug vs other alternative rewards, which under single schedule reinforcer schedules are very few); as such, value-based decision-making should remain largely intact. Thus, despite escalated drug intake, decision-making within the food choice procedure herein would be predicted to remain relatively sensitive to changes in the value of choice alternatives. The present data, modeling, and model comparison analyses are consistent with the relative value predictions, and similar results have been found in laboratory decision-making procedures with populations diagnosed with SUD that have shown retained sensitivity to relative reward magnitude, price, and delay [7, 62, 63].

A growing literature looks to account for SUD by appealing to value-based mechanisms [6]. From a relative value position, the development of SUD is a reinforcement pathology, an imbalance in the valuation of drug and non-drug alternatives within the context of drug-associated decisions [cf. 5]. Results from both demand and discounting procedures indicate an increase in the relative value of the drug after continued use [36, 37, 64–67]. A recent study in humans suggests that cocaine demand, rather than anhedonia, used as a proxy for decreased value of alternatives, is a strong predictor of cocaine use severity [68] Furthermore, studies comparing concurrent choice behavior between self-administered cocaine and a species-specific non-drug alternative (money for humans, food for rhesus monkeys and rats) have indicated that increasing the dose of cocaine increased cocaine choice and vice versa [69, 70], such shifts in choice require relative valuation processes to be intact. Additionally, rats have shown context dependent alterations in self-administration during escalation and intermittent access procedures, with the same animals showing “dysregulated” intake and then shortly thereafter returning to “regulated” behavioral states as the context of the decision changed. Further demonstrating that the interaction between the environment and the relative value of the options therein determine behavior rather than merely habitual processes [71, 72]. Thus, a shift in the relative value of a drug compared to non-drug alternatives can explain continued choice for a drug, while the mechanisms for making value-based decisions could remain intact and functional, including hyperbolic discounting promoting its immediate use [64–67]. As such the differences in decision making within a drug context are predicted to be quantitative (e.g. relative value shift), rather than qualitative (e.g. switch or malfunction of decision-making mechanisms).

Importantly, alternative frameworks of SUD, like allostatic dysfunction and incentive salience sensitization, offer distinct mechanisms for escalated drug intake that carry testable predictions for performance under the current experimental procedures. Allostatic models posit that chronic, long-access drug exposure recruits brain stress and anti-reward systems (e.g., corticotropin-releasing factor and dynorphin), creating a persistent negative affective state that shifts the hedonic baseline downward [15]. Under our current experimental context, this homeostatic deficit would predict decreased learning rates and/or decreased sensitivity to reward magnitude differences following escalation [73]. Conversely, incentive sensitization theory attributes escalation to a pathological hyper-sensitization of mesolimbic dopamine circuitry, causing drug-associated cues to over-allocate 'wanting' or incentive motivational magnetism independently of 'liking' [74]. Applied to our current paradigm, incentive sensitization would predict that discrete cues paired with a high probability and/or high magnitude options would induce cue-bound behavior, inducing greater perseveration toward options with high incentive salience following escalation [75]. However, our finding that both short- and long-access rats across cocaine and fentanyl cohorts demonstrated intact matching slopes, stable learning rates, equal sensitivity to magnitude differences, and no differences in perseverative responding challenges these predictions. The current results suggest that if allostatic dysfunction, incentive sensitization, or habitual control are driving escalated intake, their dependent neuroadaptations do not necessarily manifest as a generalized computational decision deficit or cross-modal behavioral rigidity, thereby highlighting the potential importance of contextual control of drug-related behavior [31,71, 76]

It is relevant to note that there have been inconsistencies in studies investigating changes in decision-making following chronic self-administration. Studies have reported deleterious changes in decision-making after cocaine and meth exposure in the rat gambling task (rGT) and a multi-stage decision-making task (MSDM), respectively [77, 78]. However, the Iowa Gambling Task (IGT; [79]), and thus it’s rodent counterpart, rGT, are known to have high interstudy and inter-individual variability ([80; see 81 for further review). This inconsistency may be due to the use of a wide number of procedural variables that differ between laboratories in both the IGT [82] and the rGT [83]. Additionally, IGT, rGT, and MSDM include the initial learning trials in the analysis instead of only those completed after reaching a predefined criterion of accuracy. The inclusion of all trials means that healthy individuals with a slower learning rate are not flagged as a false-positive for impairments [81]. Therefore, it is possible that if the rats did not reach a stable baseline for behavior, the deleterious effects are sample variability in the task rather than negative effects of the drug exposure. In the current choice study, rats reached a stable baseline before the drug was introduced to avoid differences in learning rate to focus on changes in decision-making relevant to the onset of drug self-administration.

As most drug valuation studies have been done using single reinforcer schedule procedures, more studies are needed to investigate the effects of various reinforcer dimensions for drug and non-drug reinforcers in SUD to determine the mechanisms governing drug-related behavior. Thus, the use of drug/non-drug concurrent choice procedures is crucial given that value is defined by the decision-making context in which a reinforcer is made available, including the presence of alternative reinforcers [76]. For example, procedures that ‘devalue’ the drug reward through satiation or increase the choice for the non-drug reward after inducing a rodent model of SUD would allow us to examine how the relative value of both rewards shift over time within a drug context. Interestingly, relative value theories would predict that after chronic use the relative subjective value between alternatives would be shifted toward the drug such that manipulations that favor the drug option would lead to increased drug choice, whereas those that favor the alternative could either decrease the drug choice or increase the choice for the alternative. In support of these predictions, preclinical models, including rats and monkeys in drug choice procedures, have been shown to reduce their preference for drug as the relative subjective value of an alternative reinforcer is increased or as the relative subjective value of the drug reinforcer is decreased [2, 8, 48, 69, 70, 84– 86]. Further, cocaine, marijuana, and heroin choices by people have been demonstrated to decrease with increases in monetary alternatives [86, 87] and therapies like contingency management have been repeatedly demonstrated as among the most efficacious in the treatment of SUD [88, 89] Collectively, there is much existing data demonstrating the interplay between shifts in relative subjective value of drug and non-drug alternatives and subsequent choices between them. Utilizing choice procedures in a drug context along with preclinical SUD models, like extended access [90, 91], and computational modeling of choice behavior could help determine how drastically and how long-lasting any potential shifts in relative value for drug might be, while also serving as a rigorous testbed for competing theories of SUD, including allostatic dysfunction, incentive salience sensitization, and habitual control.

In conclusion, given that SUD is considered a multidimensional behavioral disorder with a complex interplay between individual use, environment, and an ever-evolving understanding of neurobiological mechanisms, the careful comparison of opposing theories of addiction is crucial in order to further the most robust theories of addiction and help reduce the harm caused by SUD and drug misuse. As posited by Redish, Jensen, and Johnson [92], many of the theories of SUD have been noted to be incomplete and unable to explain the full spectrum of behavior seen in SUD, but when combined they may lead to a richer understanding. Such synthesis requires a clear understanding of the boundary conditions associated with different theories of SUD. Based on the present results, global decision deficits alone are unlikely to fully account for escalated intake in preclinical models of SUD development and highlight the importance of contextual control over drug-related behavior.