Differential Effects of Glutamate N-methyl-D-aspartate Receptor Antagonists on Risky Choice as Assessed in the Risky Decision Task

Abstract

Rationale.

Risky choice can be measured using the risky decision task (RDT). In the RDT, animals choose between a large, risky option that is paired with probabilistic foot shock and a small, safe option that is never paired with shock. To date, studies examining the neurochemical basis of decision making in the RDT have focused primarily on the dopaminergic system but have not focused on the glutamatergic system, which has been implicated in risky decision making.

Objectives.

Because glutamate is known to play a critical role in decision making, we wanted to determine the contribution of the glutamatergic system to performance in the RDT.

Methods.

In the experiment, 32 rats (16 male; 16 female) were tested in the RDT. The probability of receiving a foot shock increased across the session (ascending schedule) for half of the rats but decreased across the session (descending schedule) for half of the rats. Following training, rats received injections of the N-methyl-D-aspartate (NMDA) receptor competitive antagonist CGS 19755 (0, 1.0, 2.5, 5.0 mg/kg; s.c.) and the GluN2B-selective antagonist Ro 63–1908 (0, 0.1, 0.3, 1.0 mg/kg; s.c.).

Results.

CGS 19755 (2.5 and 5.0 mg/kg) increased risky choice in males and females trained on the ascending schedule. Ro 63–1908 (1.0 mg/kg) decreased risky choice, but only in male rats trained on the ascending schedule.

Conclusions.

Although NMDA receptor antagonists differentially alter risky choice in the RDT, the current results show that NMDA receptors are an important mediator of decision making involving probabilistic delivery of positive punishment.

Introduction

Risky choice reflects decisions that involve uncertainty. In some situations, individuals have to make a choice between a small, guaranteed reinforcer and a large, uncertain reinforcer. In other situations, risk entails receiving positive punishment for engaging in a behavior (e.g., an individual using a drug may contract a sexually transmitted infection by sharing needles). Not surprisingly, risky choice has been linked to several maladaptive behaviors, including pathological gambling (Brand et al. 2005; Madden et al. 2009) and substance use disorders (Brevers et al. 2014; Schutter et al. 2011). As such, understanding the neurochemical underpinnings of risky choice is important for helping those that have a disorder characterized by excessive risk.

Preclinical studies often use one of three tasks to measure risky choice: the probability-discounting task (PDT), the rat gambling task (rGT), and the risky decision task (RDT) (see Winstanley and Floresco 2016; Yates 2019 for reviews). In the PDT (Cardinal and Howes 2005; St Onge and Floresco 2009), animals choose between a small magnitude reinforcer that is always delivered and a large magnitude reinforcer whose delivery is probabilistic. In the rGT (Zeeb et al. 2009), subjects choose between four alternatives that differ across three parameters: magnitude of reinforcer, probability of receiving reinforcement, and the duration of a “time-out” period in which animals have to wait until they can respond. In the RDT (Simon et al. 2009), subjects choose between a small magnitude reinforcer that is delivered safely and a large magnitude reinforcer that is paired with a mild foot shock. The rGT and the RDT are of particular interest, as increased risky choice in these tasks is associated with increased drug self-administration in animals (Ferland and Winstanley 2017; Mitchell et al. 2014; Orsini et al. 2020).

Studies examining the neurochemical basis of risky choice have focused primarily on the dopaminergic system (see Yates 2019). Because the glutamatergic system has been implicated in disorders characterized by excessive risk (Kalivas 2009; Pettorruso et al. 2014), there is utility in studying the contribution of glutamatergic receptors to risky choice. To date, the effects of glutamate receptor ligands on risky choice have largely been characterized in the PDT (Yates et al. 2015; Yates et al. 2016; Yates et al. 2018; Yates et al. 2019a). In the rGT, the N-methyl-D-aspartate (NMDA) receptor channel blocker MK-801 and, to a lesser extent, the GluN2B-selective subunit antagonist Ro 63–1908 increase preference for the least risky option, but they have no effect on choice of the optimal option (Higgins et al. 2018). No studies have examined the contribution of the glutamatergic system to performance in the RDT. Because the RDT incorporates positive punishment as opposed to negative punishment as observed in the PDT or in the rGT, testing the contribution of NMDA receptor ligands on this task can further elucidate how the glutamatergic system mediates decision making in the face of uncertainty.

The main goal of the current experiment was to determine if the NMDA receptor mediates risky choice in the RDT. Because risky choice, as assessed in the RDT, predicts drug self-administration (Mitchell et al. 2014; Orsini et al. 2020), examining the contribution of the glutamatergic system can allow us to potentially uncover the potential shared neural mechanisms of risky choice and substance use disorders. In the current experiment, male and female rats were first trained in a version of the RDT before receiving injections of the NMDA receptor competitive antagonist CGS 19755 and the GluN2B-selective antagonist Ro 63–1908. CGS 19755 was chosen because it is highly selective for the NMDA receptor (Lehmann et al. 1988), which contrasts with NMDA receptor channel blockers like MK-801 and ketamine that have high affinity for non-NMDA receptors such as nicotinic acetylcholine receptors (Amador and Dani 1991) and serotonin receptors Kapur and Seeman 2002; Rammes et al. 2001). Ro 63–1908 was chosen because it lacks the psychotomimetic side effects (Jiménez-Sánchez et al. 2014; Lima-Ojeda et al. 2013) that are characteristic of NMDA receptor channel blockers (Li et al. 2011; Tang and Franklin 1983; Ranganathan et. al. 2017) and CGS 19755 (Bennett et al. 1989; Kawabe et al. 2007; Li et al. 1997). Additionally, by using a GluN2B-selective antagonist, we can further characterize the role of the NMDA receptor to risky choice as assessed in the RDT. Because research has shown that the order in which probabilities are presented can modulate drug effects in the RDT (Orsini et al. 2018), we tested four groups: males and females tested on an ascending schedule (i.e., the probability of receiving foot shock increases across the session) and males and females tested on a descending schedule.

Methods

Subjects

Sixteen adult, male (200–224 g upon arrival) and 16 adult, female (150–174 g upon arrival) Sprague Dawley rats were obtained from Envigo (Indianapolis, IN) and were individually housed upon arrival to the lab. They were acclimated to an animal housing room and handled for 6 days before any behavioral testing began. The housing room was maintained on a 12:12-h cycle (lights on at 700 h), and rats were tested in the light phase (approximately 1100–1800 h). Rats were individually housed in clear plastic cages (30 cm wide × 24.6 cm high × 41.2 cm long) with a wire top lid to hold food and a plastic external bottle top to hold two water bottles. Following each RDT session, male rats were fed 15 g, and female rats were fed 12 g. Each cage contained one plastic nylon bone. All experimental procedures were carried out according to the Current Guide for the Care and Use of Laboratory Animals (National Research Council, 2011) under a protocol approved by the Northern Kentucky University Institutional Animal Care and Use Committee.

Before being tested in the RDT, all rats were first tested in a 10-day conditioned place preference (CPP) paradigm that was unrelated to the current experiment. Therefore, the details of this experiment will not be described here. One important aspect to note is that each rat received four injections of methamphetamine (0.5 mg/kg) and four injections of saline during the CPP experiment.

Drugs

Cis-4-[phosphomethyl]-piperidine-2-carboxylic acid (CGS 19755; obtained from the NIMH’s Chemical Synthesis and Drug Supply Program) was mixed in saline, and 1-[2-(4-hydroxyphenoxy)-ethyl]-4-[(4-methylphenyl)methyl]-4-piperidinol hydrochloride (Ro 63–1908) (Tocris Bioscience) was mixed in 5% Tween 80 in saline. All injections were delivered subcutaneously (s.c.) at a volume of 1.0 ml/kg. The doses were calculated based on salt weight.

Apparatus

Sixteen operant conditioning chambers (28 × 21 × 21 cm; ENV-008; MED Associates, St. Albans, VT) located inside sound attenuating chambers (ENV-018M; MED Associates) were used. The front and back walls of the chambers were made of aluminum, while the side walls were made of Plexiglas. There was a recessed food tray (5 × 4.2 cm) located 2 cm above the floor in the bottom-center of the front wall and was located between two retractable levers. Each lever (4.8 × 0.55 × 1.9 cm) was located 2.1 cm above the floor and required a force of 0.245 N to depress. An infrared photobeam was used to record head entries into the food tray. A 28-V white stimulus light (2.54 cm diameter) was located 6 cm above each response lever. A 28-V white house light was mounted in the center of the back wall of the chamber. The floor of the operant chamber was composed of steel rods connected to a shock generator (ENV-414S; MED Associates) that delivered foot shocks. All responses and scheduled consequences were recorded and controlled by a computer interface. A computer controlled each experimental session using Med-V software.

Procedure

During two sessions of magazine training, rats received 45-mg food pellets (F0021; Bio-Serv, Frenchtown, NJ) non-contingently according to a variable-time (VT) 30 s schedule of reinforcement. The session ended after rats earned 20 pellets, which took 10 min. Rats then received three sessions of lever-press training. Each session began with illumination of the house light. A head entry into the food tray resulted in presentation of one lever; each lever was presented pseudo-randomly, with no more than two consecutive presentations of the same lever. A single response (fixed ratio [FR] 1) on the extended lever resulted in the following events: extinguishment of the house light, retraction of the lever, and delivery of one food pellet. After a 5-s intertrial interval (ITI), the house light was illuminated, signaling the start of the next trial. Each lever-press training session ended after a rat earned 40 reinforcers or after 30 min, whichever came first. Each subject earned all 40 reinforcers by the third session of lever-press training, with the exception of one rat that needed one extra session to earn all 40 reinforcers.

Following lever-press training, rats were then trained in the RDT. The first five sessions served as magnitude discrimination training (i.e., rats did not receive foot shocks during these sessions). In our pilot studies, we attempted to use the RDT that was developed by Setlow and colleagues (e.g., Simon et al. 2009), but had to make some modifications to the RDT in order to maintain stable responding. First, we omitted the 10-s limited hold that is common in the RDT. When conducting our pilot studies, we observed that some rats failed to respond during forced-choice trials in which the large, risky option was made available. Consequently, during free-choice trials, these rats showed exclusive preference for the small, safe option. This is problematic as we could not ensure that animals learned that the probabilities of receiving foot shock changed across blocks of trials. Second, to ensure animals responded during the entire session, we reduced the number of trials from 90 to 70. With these changes, we were able to maintain stable responding in 28 of 32 subjects during the course of the experiment.

Each RDT session was composed of five blocks of 14 trials. The first six trials of each block were forced-choice trials, in which only one lever was made available. The final eight trials of a block were free-choice trials, in which both levers were made available. For half of the rats, a response on the left lever was associated with delivery of one food pellet, and a response on the right lever was associated with delivery of two food pellets. These contingencies were reversed for the other half of the rats. Each trial began with illumination of the house light. A head entry into the food tray resulted in the presentation of one lever (if forced-choice trial) or both levers (if free-choice trial). During forced-choice trials, each lever was presented pseudo-randomly, with no more than two consecutive presentations of the same lever. A response (FR 1) on the extended lever resulted in the following events: extinguishment of the house light, retraction of the lever, and delivery of reinforcement. Additionally, responses on the lever associated with the large magnitude reinforcer resulted in delivery of probabilistic foot shock that lasted for 1 s (see below for more details). After a 10-s ITI, the house light was illuminated, signaling the beginning of the next trial. Sessions ended after 70 trials or after 60 min, whichever occurred first. Even though no limited hold was implemented, it is important to note that all rats completed all 70 trials by the end of magnitude discrimination training and by the end of RDT baseline training.

For half of the rats, the probability of receiving a foot shock increased across the session (0, 25, 50, 75, 100%). For the other half of the rats, the probability of receiving a foot shock decreased across the session (100, 75, 50, 25, 0%). On each individual trial within a block of trials, the probability of receiving foot shock was the same (e.g., during the 25% block, the probability of receiving shock was always set to 0.25). Initially, each rat started with a shock intensity of 0.10 mA. If a rat showed near exclusive preference for the large magnitude reinforcer across each block of trials during a 3-day period, the shock intensity increased by 0.05 mA. For some rats, increasing the shock intensity by 0.05 mA led to suppression of behavior; thus, we had to adjust the shock intensity by 0.01–0.03 mA. Shock intensities were adjusted as needed until rats met stability criteria, as defined one of two ways. The first method of determining stability required that rats meet the following criteria: 1) area under the curve (AUC; see Borges et al. 2016; Myerson et al. 2001 on how to calculate AUCs) did not vary by more than 20% across three sessions (note: we used an ordinal scale transformation of probability as described in Borges et al. 2016); 2) averaged across three sessions, rats responded for the large, risky option more than 75% of the time when delivery of this alternative was not associated with shock (i.e., 0% block of trials); and 3) averaged across three sessions, rats responded for the large, risky option less than 50% of the time when this alternative was associated with guaranteed shock (i.e., 100% block of trials). Because some rats had difficulty meeting stability criteria, we used a second method of determining stability, which required that rats meet the following criteria across a 10-day period: 1) the average proportion of choices for the large, risky option was above 75% during the 0% block of trials; 2) the average proportion of choices for the large, risky option was at or below 50% during the 100% block of trials; and 3) there were no significant increasing or decreasing trends in AUCs. This second requirement was needed for five rats before the first set of injections and was needed for nine rats before the second set of injections. Only two rats needed the second set of criteria requirements to receive both sets of injections.

Once stable responding was achieved, rats received injections of CGS 19755 (0, 1.0, 2.5, 5.0 mg/kg; s.c.) and Ro 63–1908 (0, 0.1, 0.3, 1.0 mg/kg; s.c.), with half of the rats receiving CGS 19755 first and half of the rats receiving Ro 63–1908 first. Dose order of each drug was counterbalanced across rats, and each injection occurred 30 min before the session as described previously (Higgins et al. 2016; Yates et al. 2017a; Yates et al. 2018). In our previous work (Yates et al. 2017a), we found that CGS 19755 doses above 5.0 mg/kg led to complete suppression of behavior; thus, we lowered the doses for the current experiment. The doses of Ro 63–1908 were chosen based on previous research (Higgins et al. 2016; Yates et al. 2018). Rats were tested for an additional three sessions in the RDT before receiving the next injection. Using three sessions between injections ensured responding returned to baseline levels before receiving the next injection. Following the first set of injections, rats were trained until they met stability as described above.

After 120 sessions, two rats failed to meet the stability criteria and were removed from the experiment (one male and one female trained on the descending schedule). One rat (female trained on descending schedule) met stability but lost stimulus control after the first injection (e.g., no longer showed preference for the large magnitude reinforcer even when its delivery was not paired with shock). After 60 additional sessions, this rat never reacquired stable responding and was excluded from the experiment. One rat (female trained on ascending schedule) met stability criteria but lost stimulus control after the third injection of Ro 63–1908. After 60 additional sessions, this rat never reacquired stable responding. Because this animal received three injections of Ro 63–1908 (0, 0.1, 1.0 mg/kg), we kept this subject’s data in the statistical analyses. Two rats (females trained on the descending schedule) lost stimulus control after the second injection during the second set of injections. These subjects were retrained until they met stability criteria. Because AUC values differed by more than 20% when rats reacquired stable responding, they received the two drug doses that they had originally received in addition to the two doses they had not received previously (i.e., these rats received six total injections during the second set of injections as opposed to four).

Statistical Analyses

Baseline performance in the RDT was analyzed by fitting an exponential discounting function to individual subject data via nonlinear mixed effects (NLME) modeling using the NLME package in R (Pinheiro et al. 2019). NLME offer several advantages over ANOVA (see Gueorguieva and Krystal, 2004; Young et al., 2009 for full discussions on the benefits of using mixed effects models), which is typically used to analyze RDT data (e.g., Orsini et al. 2016; Simon et al. 2009). Traditionally, discounting functions have been applied to delay and probability discounting, but this analysis can also be applied to the current data, as animals discount the large, risky option as a function of the probability of receiving shock. The exponential discounting function is defined as V = Ae^−hθ, where V is the subjective value of the reinforcer, A refers to discriminability of reinforcer magnitude¹ (i.e., how much the animal responds for the large, risky option when its delivery is never paired with shock), e is Euler’s number, h represents the rate at which animals discount the large, risky option as shock becomes more certain (i.e., measure of risky choice), and θ is the odds against safe delivery of the large, risky option. Adopting the odds against calculation described for probability discounting (Rachlin et al. 1991), the odds against safe delivery of the large, risky option is calculated as (1-p)/p, where p is the probability of receiving the large, risky option safely. For example, during the 0% block of trials, the probability of receiving the large, risky option safely is 1 (i.e., rats never get shocked during this block of trials); thus, the odds against safe delivery of reinforcement is (1–1)/1 = 0. For the first four blocks of trials, the corresponding odds against are 0, 0.33, 1, and 3. For the final block of trials, the odds against had to be arbitrarily set to 10 (due to [1–0]/0 being undefined). In the NLME models, A and h were defined as free parameters, Sex and Schedule were defined as nominal, between-subjects factors, and Subject was defined as a random factor, with the A and h parameters being allowed to vary across subjects.

The number of sessions required to reach stability was analyzed with linear mixed effects (LME) models, with Schedule and Sex defined as nominal, between-subjects factors and Subject as a random factor. A similar model was used to determine if the shock intensity required to reach stable responding differed across groups. Significant main effects/interactions were probed using contrasts via the emmeans package in R (Lenth et al., 2019). The Tukey method was used to adjust p values. Statistical significance was defined as p < .05.

To assess drug effects on performance in the RDT, we fit an exponential discounting function to individual subject data via NLME modeling using the NLME package in R (Pinheiro et al. 2019). The NLME models defined A and h as free parameters, Dose as a nominal, within-subjects factor, Schedule and Sex as nominal, between-subjects factors, and Subject as a random factor (each free parameter was allowed to vary across subjects). Separate NLME models were used for each drug. Significant main effects/interactions were probed using contrasts via the emmeans package in R (Lenth et al. 2019). In the event of a significant 3-way interaction, separate NLME models were conducted for each sex. The Tukey method was used to adjust p values. Statistical significance was defined as p < .05.

Poisson generalized linear models were used to analyze the number of completed trials in Jamovi. Each model included Dose as a nominal, within-subjects factor and Schedule and Sex as a nominal, between-subjects factors. Statistical significance was defined as p < .05.

Response latencies were analyzed with LME, with Dose, Trial Block, and Lever defined as nominal, within-subjects factors, Schedule and Sex defined as nominal, between-subjects factors, and Subject defined as a random factor. Significant main effects/interactions were probed using contrasts via the emmeans package in R (Lenth et al. 2019). The Tukey method was used to adjust p values. Statistical significance was defined as p < .05.

Results

The exponential discounting function was used to quantify performance in the RDT. This analysis accounted for a large proportion of variance in choice for the large, risky option (median R² at the end of baseline training = .920). When examining baseline performance in the RDT for males (Fig. 1a) and females (Fig. 1b), the NLME analysis showed that A parameter (i.e., discriminability of reinforcer magnitude) did not differ as a function of sex or schedule, all F’s ≤ 1.194, all p’s ≥ .277 (Fig. 1c). However, female rats had higher h parameter estimates (i.e., decreased risky choice) compared to males, F(1, 109) = 8.911, p = .004 (Fig. 1d). Rats trained on the ascending schedule tended to show decreased risky choice compared to rats trained on the descending schedule, F(1, 109) = 3.906, p = .051. There was no significant Sex × Schedule interaction, F(1, 109) = 0.195, p = .660. Rats trained on the descending schedule took longer to meet stable responding before the first set of injections, F(1, 25) = 6.717, p = .016 (Fig. 1e); however, the shock intensity required to meet stable responding did not differ across groups, all F’s ≤ 1.287, all p’s ≥ .267 (Fig. 1f).

Figure 1.

Mean (±SEM) proportion of responses for the large, risky option as a function of the probability of receiving foot shock in male (a) and female (b) rats at the end of baseline training. Mean (±SEM) A parameter (c) and h parameter (d) estimates derived from the exponential discounting function. Mean (±SEM) number of sessions required to meet stable responding (e). Mean (±SEM) shock intensity (mA) at the end of baseline training (f). *p < .05, compared to rats trained on the ascending schedule. #p < .05, compared to male rats.

Figure 2 shows the raw proportion of responses for the large, risky option following CGS 19755 administration. CGS 19755 (2.5 mg/kg) significantly increased A parameter estimates (i.e., increased discriminability of reinforcer magnitude), F(3, 499) = 2.634, p = .049. This effect appears to be driven primarily by female rats, regardless of which schedule they were trained on. There was a trend for females to have larger A parameter estimates relative to males, F(1, 499) = 3.850, p = .050 (Fig. 3a). Concerning h parameter estimates, CGS 19755 (2.5 and 5.0 mg/kg) significantly increased risky choice, F(3, 499) = 5.884, p = .001. However, the effect of CGS 19755 was observed in rats trained on the ascending schedule only, F(3, 499) = 5.567, p = .001 (Fig. 3b). None of the other main effects/interactions were statistically significant, all F’s ≤ 2.456, all p’s ≥ .118. The number of completed trials did not differ across schedule, sex, or dose, all χ²’s ≤ 0.330, all p’s ≥ .716 (Fig. 3c).

Figure 2.

Mean (±SEM) proportion of responses for the large, risky option as a function of the probability of receiving foot shock in male rats trained on the ascending schedule (a) and the descending (b) schedule and in female rats trained on the ascending schedule (c) and the descending (d) schedule following CGS 19755 administration.

Figure 3.

Mean (±SEM) A parameter (a) and h parameter (b) estimates derived from the exponential discounting function following CGS 19755 administration. Mean (±SEM) number of completed trials (c). *p < .05, relative to saline.

Figure 4 shows the raw proportion of responses for the large, risky option following Ro 63–1908 administration. Ro 63–1908 did not affect A parameter estimates; however, rats trained on the ascending schedule had larger A parameter estimates compared to rats trained on the descending schedule, F(1, 505) = 4.572, p = .033 (Fig. 5a). When examining risky choice (h parameter estimates; Fig. 5b), there was a significant Schedule × Sex interaction, F(1, 505) = 5.017, p = .026. This effect was driven by the finding that male rats trained on the ascending schedule had higher h parameter estimates compared to male rats trained on the descending schedule, whereas h parameter estimates were similar for females trained on the ascending and the descending schedules. Ro 63–1908 affected risky choice, as evidenced by a main effect of Dose, F(3, 505) = 4.539, p = .004, a significant Sex × Dose interaction, F(3, 505) = 3.261, p = .021, and a significant Schedule × Sex × Dose interaction, F(3, 505) = 3.832, p = .010. To explore the 3-way interaction, separate NLME models were conducted for each sex. For males, rats trained on the ascending schedule showed more risk aversion compared to rats trained on the descending schedule, F(1, 261) = 6.840, p = .009. Ro 63–1908 (1.0 mg/kg) significantly decreased risky choice, F(3, 261) = 4.433, p = .005, although this effect was only observed in rats trained on the ascending schedule, F(3, 261) = 3.451, p = .017. For females, there were no differences between rats trained on the ascending and the descending schedules, and there was no effect of Ro 63–1908 on performance in the RDT, all F’s ≤ 0.700, all p’s ≥ .553. The number of completed trials did not differ across schedule, sex, or dose, all χ²’s ≤ 2.404, all p’s ≥ .226 (Fig. 5c).

Figure 4.

Mean (±SEM) proportion of responses for the large, risky option as a function of the probability of receiving foot shock in male rats trained on the ascending schedule (a) and the descending (b) schedule and in female rats trained on the ascending schedule (c) and the descending (d) schedule following Ro 63–1908 administration.

Figure 5.

Mean (±SEM) A parameter (a) and h parameter (b) estimates derived from the exponential discounting function following Ro 63–1908 administration. Mean (±SEM) number of completed trials (c). *p < .05, relative to vehicle. #p < .05, compared to rats trained on the descending schedule.

When examining the effects of CGS 19755 on response latencies (Table 1), results of the LME model showed that response latencies were higher for forced-choice trials relative to free-choice trials, F(1, 356) = 4.325, p = .038, and that response latencies were higher for the lever associated with the large, risky option, F(1, 356) = 11.605, p = .001. However, response latencies were higher for the lever associated for the large, risky option for rats trained on the descending schedule only, F(1, 356) = 4.077, p = .044. There was a significant Trial × Lever interaction, F(1, 356) = 15.442, p < .001, which can be explained by the finding that response latencies were longer for forced-choice trials compared to free-choice trials, but only for the lever associated with the large, risky option. There were also Schedule × Sex × Trial, F(1, 356) = 4.069, p = .044, and Schedule × Sex × Lever, F(1, 356) = 6.437, p = .012, interactions. When probing the first interaction, there were no significant contrasts. When probing the second 3-way interaction, we found that females trained on the descending schedule had longer response latencies compared to females trained on the ascending schedule, but only for the lever associated with the large, risky option. Females trained on the descending schedule had longer response latencies for trials associated with the large, risky option compared to the small, safe option. None of the other main effects/interactions were statistically significant, all F’s ≤ 3.752, all p’s ≥ .054.

Table 1.

Mean (±SEM) response latencies following CGS 19755 administration.

Male - Ascending Schedule
Dose	Large Reinforcer - Free Choice	Small Reinforcer - Free Choice	Large Reinforcer - Forced Choice	Small Reinforcer - Forced Choice
0 mg/kg	1.394 (0.289)	5.841 (2.267)	13.854 (9.672)	0.903 (0.143)
1.0 mg/kg	1.650 (0.332)	5.289 (1.885)	18.137 (9.972)	0.725 (0.078)
2.5 mg/kg	1.374 (0.568)	1.828 (0.726)	4.514 (3.816)	0.842 (0.077)
5.0 mg/kg	1.782 (0.538)	1.735 (0.475)	0.856 (0.089)	1.035 (0.148)
Male - Descending Schedule
Dose	Large Reinforcer - Free Choice	Small Reinforcer - Free Choice	Large Reinforcer - Forced Choice	Small Reinforcer - Forced Choice
0 mg/kg	2.214 (0.574)	2.342 (0.836)	1.365 (0.490)	0.915 (0.092)
1.0 mg/kg	1.402 (0.224)	3.201 (1.180)	6.187 (3.600)	0.746 (0.053)
2.5 mg/kg	1.401 (0.308)	3.186 (1.456)	5.565 (4.290)	0.876 (0.093)
5.0 mg/kg	1.926 (0.430)	2.951 (0.919)	10.836 (6.757)	1.148 (0.198)
Female - Ascending Schedule
Dose	Large Reinforcer - Free Choice	Small Reinforcer - Free Choice	Large Reinforcer - Forced Choice	Small Reinforcer - Forced Choice
0 mg/kg	5.117 (2.538)	9.943 (5.802)	5.884 (4.220)	1.183 (0.108)
1.0 mg/kg	1.378 (0.257)	1.805 (0.401)	0.876 (0.046)	1.237 (0.072)
2.5 mg/kg	1.427 (0.517)	1.140 (0.146)	0.693 (0.056)	1.365 (0.145)
5.0 mg/kg	1.433 (0.130)	2.017 (0.771)	3.882 (3.045)	1.376 (0.223)
Female - Descending Schedule
Dose	Large Reinforcer - Free Choice*	Small Reinforcer - Free Choice#	Large Reinforcer - Forced Choice*	Small Reinforcer - Forced Choice#
0 mg/kg	5.299 (2.524)	1.568 (0.315)	3.814 (2.111)	2.143 (1.207)
1.0 mg/kg	2.494 (1.514)	1.312 (0.209)	2.167 (0.835)	0.807 (0.068)
2.5 mg/kg	1.487 (0.506)	1.290 (0.401)	1.136 (0.203)	0.873 (0.093)
5.0 mg/kg	3.212 (1.121)	2.783 (0.793)	2.128 (1.129)	0.998 (0.165)

^*p < .05, compared to female rats trained on the ascending schedule.

^#p < .05, compared to response latencies for the large reinforcer.

When examining the effects of Ro 63–1908 on response latencies (Table 2), results of the LME model showed that response latencies were higher for forced-choice trials relative to free-choice trials, F(1, 369) = 5.405, p = .021. There was also a Trial × Lever interaction, F(1, 369) = 4.044, p = .045, which can be explained by the finding that response latencies were longer for forced-choice trials compared to free-choice trials, but only for the lever associated with the large, risky option. Finally, there was a Schedule × Sex × Trial interaction, F(1, 369) = 4.537, p = .034. Post hoc tests showed that the increased response latencies during forced-choice trials was observed in male rats only. None of the other main effects/interactions were statistically significant, all F’s ≤ 3.036, all p’s ≥ .082.

Table 2.

Mean (±SEM) response latencies following Ro 63–1908 administration.

Male - Ascending Schedule
Dose	Large Reinforcer - Free Choice	Small Reinforcer - Free Choice	Large Reinforcer - Forced Choice#	Small Reinforcer - Forced Choice
0 mg/kg	2.892 (1.156)	8.193 (4.595)	45.790 (30.262)	0.883 (0.156)
0.1 mg/kg	0.790 (0.130)	6.252 (2.658)	37.850 (23.294)	1.398 (0.610)
0.3 mg/kg	1.680 (0.487)	5.199 (2.692)	16.186 (12.112)	3.751 (2.050)
1.0 mg/kg	1.437 (0.271)	5.829 (0.471)	28.889 (10.109)	85.768 (84.677)
Male - Descending Schedule
Dose	Large Reinforcer - Free Choice	Small Reinforcer - Free Choice	Large Reinforcer - Forced Choice#	Small Reinforcer - Forced Choice
0 mg/kg	12.253 (10.160)	9.572 (4.475)	8.081 (4.360)	0.859 (0.133)
0.1 mg/kg	1.440 (0.231)	12.803 (8.037)	18.650 (14.200)	3.650 (2.910)
0.3 mg/kg	3.730 (1.340)	2.596 (0.428)	10.970 (4.488)	0.923 (0.096)
1.0 mg/kg	2.966 (1.040)	4.344 (0.737)	16.651 (6.362)	1.014 (0.194)
Female - Ascending Schedule
Dose	Large Reinforcer - Free Choice	Small Reinforcer - Free Choice	Large Reinforcer - Forced Choice	Small Reinforcer - Forced Choice
0 mg/kg	18.627 (17.217)	5.786 (2.418)	24.173 (21.296)	1.496 (0.263)
0.1 mg/kg	1.532 (0.487)	2.002 (0.771)	3.737 (1.707)	1.015 (0.126)
0.3 mg/kg	3.884 (1.257)	3.539 (0.912)	4.651 (1.957)	1.039 (0.143)
1.0 mg/kg	1.936 (0.674)	4.236 (1.048)	6.547 (3.641)	1.884 (0.429)
Female - Descending Schedule
Dose	Large Reinforcer - Free Choice	Small Reinforcer - Free Choice	Large Reinforcer - Forced Choice	Small Reinforcer - Forced Choice
0 mg/kg	4.336 (1.043)	4.419 (1.597)	29.956 (25.801)	0.890 (0.115)
0.1 mg/kg	1.748 (0.346)	2.712 (1.135)	8.397 (5.573)	0.938 (0.065)
0.3 mg/kg	5.939 (2.825)	4.759 (1.553)	37.245 (15.467)	1.149 (0.157)
1.0 mg/kg	4.652 (0.897)	3.255 (0.639)	25.885 (18.649)	0.942 (0.087)

^#p < .05, compared to response latencies during free-choice trials for the large reinforcer.

Discussion

The results of the current study show that NMDA receptor antagonists have dissociable effects on performance in the RDT, as the competitive antagonist CGS 19755 (2.5 and 5.0 mg/kg) selectively increased risky choice in male and female rats whereas the GluN2B-selective antagonist Ro 63–1908 (1.0 mg/kg) selectively decreased risky choice, but only in male rats. Another finding is that neither CGS 19755 nor Ro 63–1908 affected performance on the RDT in rats trained on the descending schedule, showing that probability presentation order can modulate the effects of pharmacological manipulations on this task. Overall, these results demonstrate that glutamate NMDA receptors are an important mediator of risky decision making when positive punishment is involved, but only when the probability of receiving punishment increases across the session.

In the current study we had to make several alterations to the RDT that is used by Setlow and colleagues (e.g., Simon et al. 2009), with one major change being the removal of the 10-s limited hold. By removing the limited hold, one challenge we encountered early in RDT training is that rats would stop responding and wait for the 60-min session to end. By adjusting the shock intensity for individual rats, we were able to ensure that rats completed all 70 trials of the session. Importantly, neither CGS 19755 nor Ro 63–1908 significantly altered the number of completed trials. In fact, out of the 232 injections² we delivered to rats in the experiment, there were only eight cases in which a rat failed to complete all 70 trials (3.448% of all drug injection sessions). In addition to the procedural modifications we made to the RDT, we also altered the analysis used to quantify risky choice. Previous studies have used ANOVA to compare the raw proportion of responses for the large, risky option across drug doses (e.g., Orsini et al. 2016; Orsini et al. 2018; Simon et al. 2009). ANOVA is problematic for several reasons (see Gueorguieva and Krystal, 2004; Young et al., 2009 for full discussions), the most problematic being that this analysis cannot handle partially missing data. In the current experiment, some subjects did not respond during a block of trials following NMDA receptor blockade. Because ANOVA uses listwise deletion when there are missing data, a subject’s entire dataset is excluded from the analysis if they fail to respond during a single block of trials following one injection. By using NLME, we are able to account for partially missing data and are able to include data from each subject. Furthermore, because the RDT resembles delay/probability-discounting procedures, using an exponential discounting function allows us to derive two parameter estimates (A and h) that can better capture how drug administration alters decision making in this task (see Yates 2019 for an example of how using NLME analyses can lead to different interpretations of a drug’s effects on RDT performance compared to ANOVA). It is important to note that the exponential discounting function accounted for a substantial proportion of variance in responding (median R² = .920), which makes it a suitable analysis for the current data³. It is also important to note that we prefer using an exponential function over a hyperbolic function, which historically has been used to model discounting (see Green and Myerson 2004), because exponential functions provide a better fit to discounting data in procedures modeled after the task developed by Evenden and Ryan (1996) (see Yates et al. 2019b for a recent example).

Although we adjusted the shock intensity for each individual rat, we found that female rats demonstrated lower baseline levels of risky choice relative to male rats. Even though the shock intensity needed to achieve stable responding did not significantly differ between males and females, females showed some evidence that they were more affected by foot shocks compared to males. There were four female rats that lost stimulus control after meeting the criteria for stable responding. This loss of stimulus control primarily manifested itself as a decrease in preference for the large magnitude reinforcer even when its delivery was not paired with shock (i.e., 0% block), thus suggesting an exaggerated avoidance of the lever associated with shock. The current results mirror those examining sex differences in RDT performance (Orsini et al. 2016). In addition to observing a sex difference in baseline risky choice, there was a trend for increased risky choice in rats trained on the descending schedule relative to rats trained on the ascending schedule, although in females, the decreased h parameter estimates observed in those trained on the descending schedule is most likely an artifact of the decreased responding observed during the 0% and 25% blocks of trials. Previous research has shown no baseline differences in performance in rats trained on an ascending and a descending schedule in the RDT (Simon et al. 2009; note: Orsini et al. 2018 used ascending and descending schedules, but they did not directly compare these schedules to one another). Although there was no statistically significant difference between rats trained on the ascending and the descending schedules, the direction of the trend was consistent with our findings with delay discounting (Yates et al. 2017b; Yates et al. 2018) and probability discounting (Yates et al. 2016; Yates et al. 2018) showing that animals trained on a descending schedule discount a large, delayed/probabilistic reinforcer less compared to rats trained on an ascending schedule.

CGS 19755 (2.5 and 5.0 mg/kg) selectively increased risky choice (h parameter estimates) in both males and females without altering discriminability of reinforcer magnitude (A parameter estimates) or the number of completed trials or response latencies. Although there was no statistically significant interaction between sex and drug treatment, females appeared to be more sensitive to the effects of CGS 19755 (5.0 mg/kg), as h parameter estimates decreased by 90.719% compared to 73.469% for males. This greater percentage decrease may have been driven by the finding that females responded less for the large, risky option even when its delivery was not paired with shock compared to males (76.857% vs. 89.063%). We need to note that the ability of CGS 19755 to increase preference for the large, risky option does not appear to be mediated by increased food motivation, as previous work has shown that CGS 19755 (5.0 mg/kg) increases impulsive choice in a delay-discounting procedure (Cottone et al. 2013; Yates et al. 2017a), which results in decreased food intake during the session. Although performance in the RDT is unrelated to impulsive choice (Shimp et al. 2015; Simon et al. 2009), the current results, in conjunction with previous research, show that blockade of NMDA receptors with CGS 19755 leads to suboptimal choice in decision-making tasks. The ability of CGS 19755 to increase risky choice in the RDT may be due to its actions on NMDA receptors located within the frontal cortex. NMDA receptors are found in the frontal cortex (Monaghan and Cotman 1985), a region that has been implicated in RDT performance (Deng et al. 2018; Orsini et al. 2015). Specifically, lesions to orbitofrontal cortex (OFC) decrease risky choice in the RDT (Orsini et al. 2015), and MeCP2 expression (epigenetic factor) within medial prefrontal cortex (mPFC) is inversely related to risky choice in this task (Deng et al. 2018), but pharmacological inactivation of mPFC has been shown to alter behavioral flexibility as opposed to risky choice in the RDT (Orsini et al. 2018). Future studies using direct infusions of CGS 19755 or using siRNA to reduce the number of NMDA receptors within distinct regions of the brain will provide additional insights into how NMDA receptors mediate risky choice in the RDT.

In contrast to CGS 19755, Ro 63–1908 (1.0 mg/kg) selectively decreased risky choice, but only in male rats trained on the ascending schedule. Interestingly, Ro 63–1908 (0.1 mg/kg) increased preference for the large, risky option in female rats trained on the descending schedule, particularly during the 25% and 50% blocks, but the NLME analysis showed no corresponding change in h parameter estimates. The null effect observed for females could be a result of reduced statistical power (due to having only six rats in this condition) or could be an artifact of using NLME (which uses maximum likelihood estimation) instead of ANOVA (which uses error minimization). The ability of Ro 63–1908 to decrease risky choice in the current experiment is somewhat inconsistent with what we have previously observed following Ro 63–1908 administration in delay discounting and in the PDT (Yates et al. 2018). In delay discounting, Ro 63–1908 decreases preference for a large, delayed reinforcer (similar to what is observed in the current experiment), but this is interpreted as an increase in impulsive choice. In the PDT, Ro 63–1908 increased risky choice when the probabilities of obtaining the large, risky option decreased across the session but decreased risky choice when the probabilities increased across the session, suggesting that blocking the GluN2B subunit increases response perseveration in this task. Given that the PDT and the RDT measure distinct aspects of risk taking and are differentially mediated by neurotransmitter systems (see Winstanley and Floresco 2016; Yates 2019), the discrepancy observed here is not too surprising. Somewhat surprising was the finding that Ro 63–1908 had no effect on female rats. Similar to the results of the current study, our laboratory has found that Ro 63–1908 (3.0 mg/kg) attenuates the conditioned rewarding effects of methamphetamine in male rats, but not in female rats (unpublished results). Even though females had lower baseline levels of risky choice, h parameter estimates were similar following vehicle treatment; thus, the discrepant results obtained for males and females does not appear to be due to differences in baseline responding. One potential explanation for the discrepant results is that males have decreased GluN2B subunit expression relative to females, but this decrease is only in the hippocampus (Wang et al. 2015), a region not associated with risky decision making (Abela and Chudasama 2013; Kwan et al. 2013; Mendez et al. 2013). Although the current experiment does not explain why males and females are differentially sensitive to the effects of Ro 63–1908, these results provide some evidence that the GluN2B subunit may be an important molecular target for treating disorders characterized by excessive risk in males.

Regardless of which drug was administered, we found that rats trained on the descending schedule were insensitive to the effects of CGS 19755 and Ro 63–1908 on risky choice. Specific to the RDT, past research has shown that the order in which the probability of foot shock is presented can modulate the effects of drugs on performance in this task (Orsini et al. 2018). One potential explanation for the modulatory effect of probability presentation order on drug effects may be related to baseline differences in responding between rats trained on the ascending and the descending schedules. Considering that rats trained on the descending schedule tended to show less discounting of the large, risky option, CGS 19755 may not have been able to further shift these subjects’ preference for this alternative relative to baseline levels. However, this explanation fails to explain the discrepant findings observed for Ro 63–1908, as rats trained on the ascending schedule (which tended to show higher baseline levels of risk aversion) showed a decreased preference for the large, risky option following Ro 63–1908 administration. Another potential explanation for the null effects observed for rats trained on the descending schedule is that these animals were trained for a significantly longer time (~50 sessions) compared to animals trained on the ascending schedule (~35 sessions). This extended training may have negated the effects of NMDA receptor blockade on performance in the RDT. Somewhat related to the current results, research has shown that intra-accumbens ionotropic glutamate receptor blockade, as well as dopamine D₁ receptor blockade, impairs cued approach behavior, but only in rats early in training (Dobrovitsky et al. 2019).

In addition to the differences described above, the discrepant findings observed for rats trained on the ascending and the descending schedules may be related to distinct learning mechanisms associated with each variant of the RDT. Because rats trained on the ascending schedule encounter increasing probabilities of receiving foot shock, they experience a form of successive negative contrast, whereas rats trained on the descending schedule experience a form of positive successive contrast as the probability of receiving shock decreases. At the neurochemical level, dopamine levels within the nucleus accumbens (NAc), a region associated with RDT performance (Mitchell et al. 2014), are elevated during positive contrast but attenuated during negative contrast (Phillips et al. 2008). Blocking NMDA receptors with CGS 19755 leads to increased dopamine levels in the medial prefrontal cortex (mPFC) (Nishijima et al. 1994), an area known to regulate NAc dopamine release via glutamatergic projections (Carlezon Jr and Thomas 2009). Furthermore, competitive antagonists like CGS 19755, which tend to have greater affinity for the GluN2A subunit relative to the other NMDA receptor subunits (Laurie and Seeburg 1994), increase dopamine D₁ potentiation of NMDA responses in the striatum (Moscarello et al. 2007). Therefore, the discrepant effects of CGS 19755 on RDT performance in the ascending and the descending conditions may be due to a complex interaction between baseline levels of intra-NAc dopamine and inhibition of NR2A-containing NMDA receptors within the mesocorticolimbic pathway. Related to this point is that the differential results obtained for CGS 19755 and Ro 63–1908, at least in males trained on the ascending condition, may arise from the finding that GluN2B subunit antagonists reduce dopamine D₁ potentiation (Moscarello et al. 2007), an effect that is opposite of what is observed with GluN2A blockade.

One major challenge to studying the contribution of the NMDA receptors to risky choice as assessed in the RDT is the finding that NMDA receptor antagonists have analgesic effects (Bennett et al. 1989; France et al. 1990; Zhuo 2017). The increased responses for the large, risky option following CGS 19755 administration may be caused by the anti-nociception effects of NMDA receptor blockade as opposed to an increase in risky choice. However, there are two arguments against this notion. First, if the effects of CGS 19755 on choice were due to changes in pain sensitivity, one would expect to see a significant increase in preference for the large, risky option in rats trained on the descending schedule. Second, Ro 63–1908, which also has analgesic effects (Zhuo 2017), should have increased preference for the large, risky option. However, Ro 63–1908 decreased preference for this alternative in male rats.

There were other limitations to the study that need to be addressed. Our examination of the glutamatergic system to risky choice focused exclusively on NMDA receptors because our previous work has shown that AMPA receptors and Type I metabotropic glutamate receptors (mGluRs) do not mediate risky decision making in the PDT (Yates et al. 2015; Yates et al. 2019a). Because the PDT and the RDT do not measure isomorphic forms of risky choice (Winstanely and Floresco 2016), examining the contribution of AMPA receptors and mGluRs to risk-based decision making involving positive punishment is of interest. Another limitation to the study is that response latencies following Ro 63–1908 treatments (including vehicle) were more variable compared to latencies following CGS 19755 treatments. The discrepancy observed between response latencies is not due to an order effect, as the order of CGS 19755 and Ro 63–1908 were counterbalanced across rats. Instead, the high variability associated with Ro 63–1908 treatments was most likely due to outlier effects. For example, when examining the data for male rats trained on the descending schedule, the average response latency for free-choice trials for the large, risky option following vehicle treatment was 12.253 s, with a standard error of the mean (SEM) of 10.160. These values were influenced by one rat that had an average response latency of 73.186, compared to a range of 1.268–3.534 s for the other six rats. If this rat’s data are excluded, the average response latency becomes 2.097 s with a SEM of 0.352. One advantage to using LME models over ANOVA to analyze response latencies is that LME, like NMLE, uses maximum likelihood estimation. This is advantageous because LME models do not allow outliers to significantly impact the results of the analysis (this is known as the shrinkage effect; see Young 2017 for specific details).

In conclusion, the results of the current study highlight the complex contribution of NMDA receptors to risky choice. Blocking NMDA receptors with a competitive antagonist such as CGS 19755 increases risky choice and impulsive choice (Yates et al. 2017a), whereas selectively blocking GluN2B-containing NMDA receptors decreases risky choice in the RDT. However, parametric manipulations impact the effects of these drugs on performance in the RDT, which is consistent with our past research with delay discounting and the PDT (Yates et al. 2016; Yates et al. 2017b; Yates et al. 2018; Yates et al. 2019b). More research is needed to further understand how other parametric manipulations (e.g., changing the schedule of reinforcement, signaling the delivery of punishment, etc.) can alter how glutamatergic ligands mediate decision making when positive punishment is involved. This knowledge will allow us to better examine the neurochemical basis underlying risky choice.