All experiments were conducted in accordance with the European Union guidelines for the care and use of laboratory animals (2010/63/EU) and were reviewed and approved by the local Bioethics Committee. The authors attest that all efforts were made to minimize the number of animals used and their suffering.

We used 40 male Sprague Dawley rats (Charles River, Germany), weighing between 175 and 200 g upon arrival. We housed the rats in groups (4 animals/cage) in a temperature (21 ± 1°C) and humidity (40–50%) controlled room. The rats were kept under a 12/12-h light/dark (L/D) cycle (lights on at 07:00 h). The animals were mildly food restricted to approximately 85% of their free-feeding weights what was achieved by providing 15–20 g of food per rat per day (standard laboratory chow, Labofeed, Kcynia, Poland). Food was restricted beginning at 1 week prior to training. Water was freely available, with the exception of during the test sessions. All the behavioral procedures were performed during the light phase of the L/D cycle.

The operant tests were performed in 16 computer-controlled boxes for operant-conditioning (Med Associates, St Albans, VT, USA); each box was equipped with a speaker, a light, a liquid dispenser (set to deliver 0.1 ml of 5% sucrose solution (ACI task), a pellet dispenser (set to deliver 45 mg pellets; Bioserv, Frenchtown, NJ, USA (PD task)), a grid floor through which scrambled electrical shocks (0.5 mA (ACI task)) could be delivered, and two retractable levers. The levers were located on opposite sides of the feeder. All of the behavioral protocols, including the data acquisition and recordings, were programmed in Med State notation code (Med Associates). The experimental procedures for the PD test were modified from the procedures described by St Onge and Floresco (2009). The experimental procedures for the ACI test used in this study were performed according to Enkel et al. (2010) with modifications, and they have been described in detail elsewhere (Rygula et al., 2012, 2015).

The training and testing protocols used have been previously described elsewhere (Floresco et al., 2008; St Onge and Floresco, 2009). Briefly, the rats were first trained to press both levers under a fixed-ratio schedule (1:1) to a criterion of 60 presses in 30 min. Half of the animals were trained first on the left lever and later on the right lever, and the other half were trained in the reverse order. Subsequently, we trained the rats on a simplified version of the full task. Each training session began in darkness, with the levers retracted. Each trial began with the illumination of the house lights and extension of one of the two levers into the operant box. If the animal responded within 10 s, the lever retracted and a single sucrose pellet (45 mg, Bio-Serv, Frenchtown, NJ, USA) was delivered with 50% probability. If the animal did not respond to the lever within 10 s, the lever retracted, the house light was switched off, and the trial was scored as an omission. The trials lasted 40 s each and the training session consisted of 90 trials. We used this procedure to familiarize the animals with the probabilistic nature of the full task. The training took approximately 4–5 days (to a criterion of 80 or more successful trials).

During PD testing, the rats received daily sessions of 72 trials that were separated into four blocks of 18 trials. The entire session took 48 min to complete, and the animals were trained 5 days per week. At the beginning of each session the levers were retracted and the house light was switched off (the intertrial state). Each trial began with the illumination of the house light and extension of one or both levers (the schedule of a single trial is shown in Figure 1B). One lever was designated as the small/certain lever and the other was designated as the large/risky lever; these designations were counterbalanced left/right and remained consistent throughout the training phase. When a lever was chosen, both levers were retracted. Selection of the small/certain lever always resulted in the delivery of one sucrose pellet; however, selection of the large/risky lever resulted in the delivery of four sucrose pellets with a particular probability (see below). Multiple sucrose pellets were delivered 0.5 s apart. If the animal failed to press a lever within 10 s, the operant conditioning box was reset to the intertrial state until the next trial and scored as an omission. Each of the four blocks began with eight forced choice one-lever trials (four trials randomized in pairs for each lever), which allowed the animals to learn the probability of receiving reinforcement during each block and the amount of food associated with each lever press. Next, there were 10 free-choice 2 lever trials, and the rats chose either the large/risky or the small/certain lever. The probability of obtaining four sucrose pellets after pressing the large/risky lever successively decreased across the blocks; the probabilities averaged 100% in block 1, 50% in block 2, 25% in block 3, and 12.5% in block 4. For each session and trial block, the probability of receiving the large reward was drawn from a set probability distribution. Therefore, on any given day, the probabilities in each block could vary, but on average across many training days, the actual probability experienced by the rat approximated the set value. Using these probabilities, selection of the large/risky lever was advantageous in the first two blocks, balanced during the 25% block and disadvantageous in the last block.

The rats were trained on the task until the following criteria were met by the group: (1) In at least 80% of the trials they chose the large/risky lever during the first trial block (100% probability); (2) In less than 60% of the trials they chose the large/risky lever during the final trial block (12.5% probability); and (3) the consistent choice patterns fulfilling criteria 1 and 2 were demonstrated on three consecutive days. Individual patterns of PD behavior over the three criterion days were used to compare risk-taking behavior between the rats displaying “optimistic” and “pessimistic” traits.

The primary dependent measure of interest was the proportion of choices directed towards the large/risky lever during each block of free-choice trials, taking into account trial omissions. For each block, this proportion was calculated by dividing the number of choices of the large/risky lever by the total number of successful trials.

After training and testing on the PD task, the rats were re-trained and re-tested on the ACI paradigm. As mentioned previously, the experimental training and testing procedures for the ACI paradigm used in this study have been described previously (Enkel et al., 2010; Rygula et al., 2012, 2015).

Briefly, the animals were initially trained to press one lever when a “positive” tone (2000 Hz at 75 dB, counterbalanced) signaled reward (20% sucrose solution) availability and to press a second lever when another “negative” tone (9000 Hz at 75 dB, counterbalanced) signaled a forthcoming punishment (electric foot shock: 0.5 mA, 10-s). By pressing the appropriate levers, the rats could either avoid the punishment or receive a reward. 10 s intertrial intervals (ITIs) separated the tone presentations, and training sessions lasted for 30 min. The rats had to reach the criteria of at least 90% correct responses to the tone signaling reward availability over three consecutive training sessions and at least 60% correct punishment-prevention responses over three consecutive training sessions to proceed to discrimination training.

During the discrimination-training phase, the animals were trained to discriminate between 20 negative and 20 positive tones presented in pseudorandom order, by pressing the appropriate levers (as learned in the previous training stage) to maximize reward delivery and minimize punishment. Training sessions lasted 40 min. The animals were required to do a minimum of 70% correct responses for each lever over three consecutive discrimination sessions to qualify for the ACI testing.

During the ACI testing sessions, the rats were exposed to 20 positive, 20 negative and 10 ambiguous (5000 Hz at 75 dB) tone presentations. Each test session lasted for 50 min. The tones were played in a pseudo-randomized order and were separated by 10-s ITIs. The responses to each tone (positive, ambiguous and negative) presented during the ACI testing were analyzed as a proportion of the overall number of responses to a given tone. To calculate the cognitive bias index, we subtracted the proportion of negative responses to the ambiguous cues from the proportion of positive responses to the ambiguous cues, resulting in values of between −1 and 1. Values of greater than or equal to 0 indicated an overall positive judgment and “optimistic” interpretation of the ambiguous cue, whereas values of below 0 indicated an overall negative judgment and “pessimism”.

The CBS has been described in detail in Rygula et al. (2013). Briefly, to assess the “optimistic” and “pessimistic” traits, we subjected the rats to a series of ten consecutive ACI tests, conducted at 1-week intervals. Based on the average values of cognitive bias index data obtained from these 10 ACI tests, the animals were divided into two experimental groups: “optimistic” and “pessimistic”.

The APO sensitivity test that was applied in our study was modified from the procedures previously described by Havemann et al. (1986) and Surmann and Havemann-Reinecke (1995). Briefly, immediately after the animals were injected with APO (2 mg/kg s.c.), they were placed in an open field (66 × 56 × 30 cm) for 1 h. Next, scoring was performed in 30-s periods with 5-min intervals. Within each scoring period, we recorded all of the stereotyped behaviors displayed by the animals. Each 30-s scoring period was classified according to the stereotyped behavior exhibited during the longest period (e.g., when gnawing, licking and sniffing lasted for 20, 7 and 3 s, respectively, the scoring period was classified as “gnawing”). When two stereotyped behaviors were displayed for equal amounts of time (e.g., 15 s each), we scored both behaviors. Stereotyped sniffing was defined as rhythmic movement of the snout and head along the cage wall or floor, accompanied by rapid vibrissae movements. Stereotyped licking was characterized by protrusion of the tongue against the cage floor or wall, and stereotyped gnawing was scored when the animal gripped the cage wall or floor between its teeth. We also scored stereotyped climbing, characterized by vertical scratching and rhythmic, alternating movements of both forelegs on the cage wall (Havemann et al., 1986).

The experimental design and schematic representations of the procedures are presented in Figures 1A–D. The animals that met the criteria and successfully completed the training were tested on the PD task three times over three consecutive days and were then re-trained on the ACI task. Each animal was subjected to the CBS procedure as previously described. After the rats were evaluated to determine their degree of “optimism” or “pessimism”, they were divided into “optimistic” and “pessimistic” experimental groups. To assess whether the traits of “pessimism” and “optimism” interacted with risky behaviors, we compared the average performances of the “optimistic” and “pessimistic” animals on the three PD tests.

Notably, the PD testing preceded the ACI testing because the levers remained unbiased during the second test, only if the tests were conducted in that order. If the animals had been tested first on the ACI paradigm, then the levers would have acquired negative and positive valences via their associations with a punishment or palatable reward, and they would not have been able to be used as reliable operands in the PD paradigm.

Because DA has been implicated in both: risky choices and optimism, in the final experiment, we compared the reactivity of the DA system between the “optimistic” and “pessimistic” animals using the APO sensitivity test.

The data were analyzed using SPSS (version 21.0, SPSS Inc., Chicago, IL, USA). The Kolmogorov-Smirnov test was used to assess the distribution of the cognitive bias index data. Differences in PD between the “optimists” and “pessimists” were analyzed by 2-way analysis of variance (ANOVA), with the cognitive bias index as a between-subject factor (2 levels: “optimism” and “pessimism”) and the trial block as a within-subject factor (4 levels: 100, 50, 25 and 12.5%).

The differences in the frequency of “optimism” and the speed and distance traveled following APO administration between the “optimists” and “pessimists” were analyzed using t-tests. Differences in the processing of the positive and negative tones and ambiguous cues between the “optimistic” and “pessimistic” animals were investigated by 4-way ANOVA, with cognitive bias as the between-subject factor (2 levels: optimistic and pessimistic) and lever (2 levels: positive and negative), tone (3 levels: positive, ambiguous and negative) and test (10 levels: baseline test 1–10) as the within-subjects factors.

Differences in the length of training for the ACI test and insensitivity to the APO challenge (number of different stereotyped behaviors) between the “optimists” and “pessimists” were analyzed separately using the Mann-Whitney test because the data were not normally distributed. Finally, the Spearman correlation coefficient between the cognitive bias index and the frequency of gnawing was determined. For pair-wise comparisons, the values were adjusted using Sidak’s correction factor for multiple comparisons (Howell, 1997). All of the tests of significance were performed at α = 0.05. Homogeneity of variance was confirmed using a Levene’s test. For repeated-measures analyses, sphericity was also verified using Mauchly’s test. The data are presented as the mean ± SEM.

All of the trained animals reached the criteria for stable PD as a group (defined as a selection of the large/risky lever during the first trial block (100% probability) in at least 80% of trials and selection of the large/risky lever during the last trial block (12.5% probability) in less than 60% of trials, maintained over three consecutive days) after 15 sessions. The average proportion of risky lever choices made by all of the rats in each block of free-choice trials during the 12 training and three test sessions is presented in Figure 2. Repeated-measures ANOVA of the average choice data from the three PD tests revealed a significant main effect of trial block (F_(3,84) = 30.55, p ≤ 0.001). Post hoc multiple comparisons revealed significant differences in the proportion of choices of the large/risky lever between the 100% vs. 25%, 100% vs. 12.5%, 50% vs. 12.5%, and 25% vs. 12.5% trial blocks (p ≤ 0.001), as well as between the 50% vs. 25% trial blocks (p ≤ 0.01).

To compare the risk-taking behaviors between the “optimistic” and “pessimistic” animals, after the PD testing, the rats were trained for the ACI paradigm and on the basis of the CBS procedure, classified as “optimistic” or “pessimistic”. Only 30 out of 40 trained rats, reached the criteria and qualified for the CBS. The animals that were further classified as “optimistic” reached the criteria of the positive tone, negative tone and discrimination training after 4.8 ± 0.3, 30.1 ± 5.7, and 60.1 ± 8.2 days, respectively, whereas those classified as “pessimistic” reached the same criteria after 4.2 ± 0.17, 32.1 ± 3.7 and 49.7 ± 5.7 days, respectively. The Mann-Whitney test revealed that the total length of training did not significantly differ between the “pessimistic” and “optimistic” animals (medians = 74 and 94, respectively; U = 83, NS).

The average cognitive bias index of all of the tested rats, which was determined based on CBS, was 0.09 ± 0.06. The cognitive bias index data collected during the CBS were normally distributed (Z = 0.12, N = 30, Kolmogorov-Smirnov test).

Analysis of the animals’ responses to the negative and positive levers following presentation of the reference and ambiguous tones across the CBS indicated the absence of test-retest effects. Despite the significant lever × test × tone interaction (F_(18,504) = 2.07, p < 0.01), post hoc pair-wise comparisons revealed the absence of unequivocal patterns in between-test differences.

The CBS results enabled the animals to be separated into two groups, “pessimistic” (N = 20, average (AVG) cognitive bias index <0) and “optimistic” (N = 10, AVG cognitive bias index > 0), that differed significantly in their interpretation of the ambiguous cues over time (Figure 3A). The “pessimistic” animals had an average cognitive bias index, ranging from −0.02 to −0.80, whereas the “optimists” displayed cognitive bias index in the range from 0.00 to 0.48.

Further analysis revealed significant differences in the response patterns of the “pessimistic” and “optimistic” groups (there was a significant lever × tone × cognitive bias interaction (F_(2,56) = 14.40, p < 0.001)). In response to the ambiguous tone, the “optimistic” animals responded significantly more often (p < 0.001) to the positive lever (Figure 3B) and significantly (p < 0.001) less often to the negative lever (Figure 3C) than their “pessimistic” counterparts. The “optimistic” animals also responded significantly (p < 0.01) more often to the positive lever in response to the reference positive tone (Figure 3B) and significantly (p < 0.01) less often to the negative lever in response to the reference negative tone (Figure 3C). Analysis of the trials in which the animals did not respond (omissions) revealed a significant cognitive bias × tone interaction (p = 0.018) and a trend (p = 0.051, Sidak post hoc test) towards a higher proportion of omitted trials in the “pessimistic” animals compared to the “optimistic” animals following presentation of the positive reference tone (Figure 3D).

Analysis of the “optimism” frequency (number of tests in which the cognitive bias index of an individual animal was greater than zero, out of the 10 CBS sessions) revealed that on average, the rats classified as “pessimists” were significantly (p < 0.001) less frequently “optimistic” than their “optimistic” conspecifics (t₍₂₈₎ = 8.09, Figures 4A,C).

As mentioned previously, although the cognitive judgment bias index fluctuated in both groups of animals (there was a significant lever × test × tone interaction (F_(18,504) = 2.07, p < 0.01)), the differences between the “pessimistic” and “optimistic” remained constant across the screening period (there was no significant cognitive bias × test interaction; F_(9,252) = 1.38, NS), indicating that the traits were stable (Figure 4B).

When the probability of reward was high (100%, 50%) or intermediate (25%), the proportion of time that the large/risky lever was chosen did not differ between the “optimists” and “pessimists” (Figure 5). However, compared to “pessimism”, “optimism” was associated with a significantly (*p ≤ 0.05) higher proportion of selection of the large/risky lever in the last block, when the probability of receiving a reward was the lowest (12.5%, Figure 5). Two-way repeated-measures ANOVA of the average choice data from the 3 PD tests revealed a significant cognitive bias × trial block interaction (F_(3,84) = 3.53, p ≤ 0.05).

After the administration of APO, the intensity of stereotyped climbing (Mann-Whitney U = 82) and licking behaviors (Mann-Whitney U = 85) did not differ between the “optimists” and “pessimists”; however, “optimism” was associated with significantly (Mann-Whitney U = 62.5, p ≤ 0.05) more intense gnawing behavior (Figure 6A) and a statistical trend (Mann-Whitney U = 66.5, p = 0.088) towards less intense sniffing (Figure 6A). Correlation analysis revealed that the cognitive bias index was significantly positively correlated (r = 0.39, N = 30, p < 0.05) with the frequency of stereotyped gnawing (Figure 6B). We observed no significant differences in APO-induced locomotor activity, as expressed by the speed of locomotion or the distance traveled in the open field (t₍₂₈₎ = 0.93 and 0.91, respectively, NS, Figure 6C).