2.1 Animals

Male Wistar Han rats were used in this study (n = 12) and in a prior pilot study (additional n = 6). Rats were obtained from a commercial breeder (Harlan, Rossdorf, Germany) and were six weeks old upon arrival. Preparation for training was given according to the protocol developed by Bussey and colleagues [13]. During the first week, the rats were habituated to the lab environment (22 ± 2°C, 55% humidity), with an inverted 12–12 h dark/light cycle (lighting from 14:00 to 02:00), and were provided food and water ad libitum. At seven weeks of age, each rat was subcutaneously implanted with a radio-frequency identification (RFID) transponder in the groin area (under isoflurane anaesthesia). The RFID tag has a unique identification code allowing the automated individual identification of each animal using an RFID transponder reader. Two days after transponder implantation, the rats were handled and weighed daily for three consecutive days, after which they were introduced to sugar pellets (45 mg, AIN-76A, TestDiet) for two days in their home cages. The same sugar pellets were used throughout the experimental period. In addition to the experimental pellets, at 12:00 each day, six rats received 90–108 g of pellets (15–18 g per rat; chow V1535-3, Ssniff, Soest, Germany) which maintained rats at 85–90% of their free-feeding weight. Body weight was measured three times per week and growth was carefully monitored. Each experimental group in both the pilot and the main study consisted of six animals. Three groups were tested in succession; first the pilot group, then the two experimental groups. The rats were nine weeks old at the beginning of the training phase.

The novel experimental procedures described here were designed to allow for maximal animal welfare. Animals lived undisturbed as a group within their home cages. Briefly, data collection was performed using automated observational methods applied to undisturbed group-housed animals that voluntarily decided to visit the experimental chamber for the collection of rewards. The health of the animals was monitored daily. The experimental procedure did not cause any damage, pain or suffering to the animals. The animals were not sacrificed at the conclusion of the study. This study was performed under the supervision of the animal welfare officer (Tierschutzbeauftragter) at the Humboldt University. Experiments followed national regulations in accordance with the European Communities Council Directive 10/63/EU.

2.2. Apparatus

The home cage consisted of two regular rat cages connected by two transparent tubes and holes (Ø 6.5 cm) in the long side of the wall of each cage (Type IV, 59.5 cm x 38 cm x 20 cm; Fig 1). This dual-home cage provided a spacious cage environment for six adult rats. The standard rat’s food was simultaneously delivered into the food trays of both cages to prevent competition for food. The two connected cages could be placed onto a regular cage rack. The dual home cage was connected to an operant touchscreen chamber via a separate tube that also functioned as an automated animal sorter (ID Sorter, PhenoSys; Fig 1). The automated animal sorter consisted of a linear tube arrangement, three RFID transponder readers and two electronically-operated guillotine gates. The detection of an animal by the RFID readers and the movement of the gates were controlled by PhenoSoft software (PhenoSys). In principle, this operated similarly to an equivalent system for mice [17]. PhenoSoft recorded the time that each individual rat entered or left the sorter and the operant chamber, the start time of each session and the total duration of the animal’s stay in the operant chamber.

The operant chamber was a modified Bussey-Saksida Rat Touch System (Model 80604, Campden Instruments). The chamber was trapezoidal and composed of two black Perspex walls, an operant touchscreen facing the food magazine, a house light and a speaker situated above the chamber. A black Perspex board (mask) containing 15 windows (three rows x five columns, 3.3 × 3.3 cm) was positioned 0.9 cm in front of the touchscreen (Nexio 150A, iNEXIO CO., LTD.) in order to restrict the area available for potential animal responses during the TUNL task. The food magazine could be illuminated and included an infrared beam nose poke detector. Rats could enter and leave the operant chamber through a 6.5 cm (Ø) hole in the wall to which the sorter was connected. We used ABET II software (Campden Instruments) to control the operant task (e.g., maximum number of trials, maximum session time, inter-trial interval, delay and separation conditions).

ABET II and PhenoSoft directly communicated via a custom programmed software module. PhenoSoft ensured that only one animal entered the chamber at any one time and remained there for the entire session. ABET II assigned the correct training schedule to each individual and a session commenced when PhenoSoft signalled the entry of a rat. ABET II communicated the end of a session to PhenoSoft, which then allowed the animal to return to the home cage. The ABET II parameters included the number of trials completed, number of correct trials, number of incorrect trials, percentage of correct trials, correction trials and latencies of different types. Parameters provided by PhenoSoft included the number of sessions per training phase, the latency before entering the operant chamber and the duration of a visit to the chamber.

2.3 Fine-tuning of automated procedures

The goal of the pilot study was to develop the adapted conditions for training and testing, as described below. The pilot group of rats did not experience the same experimental conditions as those used later; therefore, these data were excluded from the formal analysis. Observations and experimental adjustments drawn from the pilot study are summarised in the following section.

During the pilot phase, PhenoSoft was fine-tuned to ensure that only one animal was sorted at any one time. From the home cage, a rat was identified (Reader 1) and, if the operant system was unoccupied and its individual experimental schedule permitted it, Gate 1 opened so that the animal could enter the sorter. Detection by Reader 2 and Reader 3 led to the closing of Gate 1 (Fig 1). For 30 s, both gates remained closed and, if a second animal was detected by Reader 2 or 3, the sorting sequence was aborted and Gate 1 opened to allow the rats to return to the home cage. Subsequently, if only one rat was detected in the sorter, Gate 2 opened and the rat could enter the operant chamber.

During the training phases of the pilot study, the animals could enter the operant chamber without restriction or delay between two sessions (when the operant chamber was unoccupied). During such unrestricted access to the operant chamber, we observed heterogeneous types of behaviour between animals, and even between two sessions of the same animal. An animal would, in some sessions, not interact with the screen at all or stay beyond the end of the session in the operant chamber (especially during the light phase), indicating a lack of interest in performing the task, and a propensity to enter the operant chamber for reasons other than performing the task. However, during other sessions, the rat would regularly poke the screen for rewards and reach the maximum number of trials within the session. We therefore implemented restricted access to the operant chamber to ensure that all animals had the same level of motivation to enter the chamber and obtain a maximum number of rewards during each experimental session. To accomplish this, we implemented a minimum time interval between two consecutive sessions, during which an animal was denied access to the operant chamber. This inter-session interval began when the animal exited the sorter at the end of a session. Furthermore, to maintain spontaneous explorative behaviour and sustain animal motivation to enter the operant chamber, we limited the number of trials per session. During the first training phase with the activated sorter (Training 2, see below), we combined a short session duration (10 min or 17 trials) with a short inter-session interval (30 min) to ensure frequent entries to the operant chamber and to provide further opportunities for training. Training 2 is a critical learning phase. However, once the animals had learned the association between the instrumental response and the delivery of the reward, the inter-session interval during the phases from Training 3 onwards (see below) increased. For the automated procedure, the two parameters of inter-session interval and maximum number of trials per session proved to be important for homogenising the behaviours of the animals when in the operant chamber, and to motivate regular visits to the chamber.

We also modified the inside of the operant chamber after considering the conditions in Talpos et al. (2010). A small moveable shelf was originally mounted in front of the screen. This shelf had been used in other studies with the aim of helping the animal focus on the task by supporting only their forepaws and thereby preventing impulsive responding. After removal of the shelf during the pilot study, rats that had not previously poked the screen began spending more time closer to the touchscreen and eventually began to poke it. The mask used throughout our study was a standard rat mask (3 x 5 cm) supplied by Campden Instruments, and was identical to those used by Oomen et al. (2013; S1B Table). This differed from Talpos et al. (2010), who altered window size between the training and test phases (S1B Table).

2.4 Training and Testing

Training and testing followed established TUNL touchscreen protocols [13,14] with adjustments for our automated system (S1A and S1B Table). During a training session, the house light was off. However, upon delivery of a reward pellet, the food tray was illuminated and a brief high-pitched tone sounded (0.5 s). At the conclusion of a session, all windows on the screen turned off, and the magazine light remained off. No other stimulus signalled the end of the session. Once a rat had fulfilled the acquisition criterion (unless more than half the group still required sessions to reach the criterion), the rat could immediately begin the following training phase. If at least half of the animals were still required to complete sessions in order to fulfil the criterion, access to the operant cage was permitted to the late-learners only for a certain period (e.g., the following 4 h). This procedure aided the rats in reaching the level of the remaining group members by increasing their chances to access the operant chamber and to perform the remaining sessions to meet the criterion. Sessions were voluntarily commenced by the animals during dark or light phases.

2.5 Data analysis

Statistical tests were performed using R [19] and open access macros for non-parametric tests (Anastat.fr). We chose to perform nonparametric tests as the sample size of the experimental groups (n = 6, or n = 12 when the two groups of six rats were pooled) was too small to confidently assume that each dataset was normally distributed. We first tested whether the results obtained for our two experimental groups (each n = 6) differed to determine whether the data could be pooled for subsequent analyses. For each parameter measured in each training and testing phase, we compared the results between the two experimental groups using the Mann-Whitney test for independent samples (U) and then applied Bonferroni correction to control for the effects of multiple testing (S3 Table).

As the two experimental groups did not differ, the results for the 12 animals were pooled for the main analyses. The non-parametric Friedman test for repeated measures was performed to test for changes in dependent measures across conditions or blocks of sessions. The pairwise comparison between blocks (or conditions) was completed using the Wilcoxon signed-rank test. The results are illustrated as boxplots; to complement the description of our results the values given in the Results section are provided as mean ± standard deviation. As a rule, no statistical analysis was performed when the position of the median and of the first and third quartiles of the boxplots overlapped, as this indicated similar distributions of the datasets (as an example, see Fig 5B). Data were excluded from time periods in which technical problems were experienced.

3.1 Repeatability of experimental results between groups

The two groups of six rats were then tested to determine whether there was a group effect on the results from the training phases and the TUNL task (Mann-Whitney test and Bonferroni correction for multiple testing). After Bonferroni correction, no difference was observed between the two experimental groups (S3 Table); therefore, we pooled the data from both experimental groups for subsequent analyses.

3.2 Rapid habituation to operant cage and sorter

During Training 1, all rats entered the unfamiliar operant chamber within the first 15 min (4 min 46 s ± 4 min 59 s; n = 11 rats; one rat entered after 73 min) by passing from the home cage through the sorter tubes. After one rat had entered the sorter and operant chamber, all other rats quickly followed. At the commencement of this phase, more than two rats could simultaneously be seen in the tube. Individuals made 211 ± 28 visits to the operant chamber during the first 24 h, with 88 ± 3.7% of these occurring during the dark phase (Fig 2). A visit during the dark phase lasted 62 ± 15 s, which was on average four times shorter than during the light phase (262 ± 208 s). Almost all pellets (98%) delivered during the dark phase were triggered by touching a lit window on the touchscreen.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The number of visits to the operant chamber and trials made by each rat during the initial 24 h of the habituation/exploration phase.

Each vertical black line indicates a visit to the operant chamber by an individual rat (R). Grey lines (upper: rats 1–6, lower: rats 7–12) signify completed trials and delivered rewards. Rats 1–6 and 7–12 were in separate groups. The vertical black line at 11 h indicates the shift from the dark to light phase. On the x-axis, “00” represents the beginning of the training phase for all rats (all introduced simultaneously to the system). The time of day corresponding to experimental time “00” was 15:00 h.

https://doi.org/10.1371/journal.pone.0169476.g002

In Training 2, the sorter was active and each rat was mechanically sorted (gates moved regularly during the sorting procedure, as described in section 2.2) to enter the operant chamber and start a session. Animals required an average of four attempts (4.10 ± 1.45; n = 10) to enter the sorter until they were correctly sorted; they then proceeded to their first individual training session (two rats required more than four attempts). Once the sorting procedure concluded and an animal arrived at the operant end of the sorter, the first session commenced within 41 ± 10 s.

3.3 Automated touchscreen training and exiting the operant cage

All rats quickly reached a high level of performance in each training phase. The rats achieved more than 80% correct trials after four sessions in Training 2 (in three days), three sessions in Training 3 (in one day), three sessions in Training 4 (in one day), four sessions in Training 5 (in one to two days) and five sessions in Training 6 (in one to two days; Fig 3A). In Training 5, a poke into the food tray was required to initiate the trial. In Training 6, incorrect responses were punished. Incorrect touches to the screen significantly decreased from Training 4–6 and Training 5–6 (Fig 3B; Friedman test on the repeated (averaged) data of each training, Q(2) = 9.21, p < 0.01 and pairwise comparison using Wilcoxon signed-rank test, between t4 and t6, and t5 and t6, p < 0.05). Importantly, the rats did not spend unnecessary time in the operant chamber after the conclusion of a session. The mean latency time prior to exiting the operant chamber after the conclusion of a training session (all trainings included) was 2.6 ± 3.1 min.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Automated touchscreen training.

(A) Percentage of individual correct trials across Training phases 2 (t2) to 6 (t6). White circles show individual performance during a session. Circles can overlap. In t2, the percentage of correct trials is also referred to as the percentage of active trials (see Methods). Difficulty increases from t4 to t5 when rats learn to self-initiate a trial by poking into the pellet magazine. In t6, incorrect trials are punished with a time out. Data are based on n = 12 individuals. Due to technical problems, six rats continued with t2 for a further eight sessions although they already had achieved > 80% correct trials (grey dotted line; n = 12 sessions 1–5 and n = 6 sessions 8–15). (B) The number of incorrect nose pokes (nose pokes to unlit windows) during the final three phases of training. Boxplots show median, quartiles and 10^th/90^th percentiles. *p < 0.05 between t4 and t6, and t5 and t6

https://doi.org/10.1371/journal.pone.0169476.g003

3.4 Influence of automation on performance in the TUNL task

During the TUNL task, the rats achieved the learning criterion of 70% correct responses within a session, in 10–19 sessions (median = 16) and 5–10 days (median = 8). During the TUNL task, the number of correct trials (%) increased with experience (Friedman test, Q(3) = 11.34, p < 0.01 and pairwise comparison using Wilcoxon signed-rank test between blocks 1 and 4, and 2 and 4, p < 0.05; n = 12 rats; Fig 4A), while the number of correction trials decreased with experience (Friedman test, Q(3) = 11.34, p < 0.01 and pairwise comparison using Wilcoxon signed-rank test between blocks 1 and 3, and 2 and 3, p < 0.05; Fig 4B). These data were compared with data published by Talpos et al. (2010), who used a non-automated procedure with rats manually introduced into operant chambers for daily sessions (Fig 4C). During task acquisition, there was no difference in the level of performance between our study and the control group of Talpos et al. (2010; Mann-Whitney tests). However, the 20 sessions were completed within 10–11 days (10.8 ± 0.4 days) in our system, which was half of the days necessary to learn the task in the group of Talpos et al. (2010; one session per day and 20 days for all animals). Interestingly, rats reached 70% correct responses within 5.0 ± 3.2 days, which is similar to the 5.0 ± 3.5 days in Talpos et al. (2010). In our study, the mean latencies before making a choice (5.0 ± 2.5 s) and collecting the reward (2 ± 12 s) were very short (n = 6; for 6 other animals these parameters were incorrectly recorded). A trial was completed within an average of 47 ± 7.0 s, and this duration was stable across all sessions of the TUNL acquisition task.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Automated TUNL task.

(A) Percentage of correct trials and (B) number of correction trials averaged for blocks of five sessions. Data from n = 12 individuals. (C) Comparison of performance (correct trials across blocks in the TUNL task) between this study (white symbols; identical data as in A) and the published data from a study by other authors using a non-automated setup (Talpos et al., 2010; grey symbols). Boxplots show median, quartiles, and 10^th/90^th percentiles. *p < 0.05

https://doi.org/10.1371/journal.pone.0169476.g004

3.5 Performance during the probe and interference tests

In the TUNL task, the ability to cope with various levels of difficulty was evaluated in probe sessions in which two parameters varied: the spatial distance between the S+ and S- locations from maximum (easy) to adjacent (difficult to discriminate), and the delay between sample and choice phases (2–6 s), which increased working memory demand. Performance decreased as discrimination of the two stimulus locations increased in difficulty (separation distance from maximum to adjacent) for both delays (Friedman test, Q(2) = 13.82, p < 0.001; Fig 5A). Performance was lowest for the adjacent condition, and in this condition, rats performed better with the shorter, 2 s, delay (Wilcoxon test for paired samples with Bonferroni correction for multiple testing w(12, 0.025) = 14, p’ < 0.05; Fig 5A). In the interference test, which tested for a potential interference effect between trials, the shorter ITI of 5 s instead of 20 s did not affect the level of performance (Fig 5B). The motivation to collect reinforcement measured as latency in reaching the food tray after a correct choice did not differ between probe test conditions. These latencies were on average 1.0 ± 0.64 s. The latency before making a correct choice in the choice phase was also unaffected by the different delay and stimulus distance separation levels (an average of 5.0 ± 2.4 s).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Performance during the automated TUNL probe test.

(A) Performance with two different time delays (2 or 6 s) and three spatial separation conditions (maximum, medium or adjacent). (B) Performance during maximum spatial separation and a 2 s delay with different ITIs (interference test). Boxplots show median, quartiles, and 10^th/90^th percentiles. n = 12 rats per condition. *p < 0.05

https://doi.org/10.1371/journal.pone.0169476.g005

In the probe test, the performance of rats in the current study differed from the results of Talpos et al. (2010; both studies using a 6 s delay condition) regarding the two more challenging conditions of separation (medium and adjacent; Mann-Whitney with Bonferroni correction for multiple testing U = 15, p’ < 0.05 and U = 16, p’ < 0.05, respectively; Fig 6). In fact, the performance of our experimental rats was higher during the medium condition compared to the results in Talpos et al. (2010). However, in the most difficult condition, the performance of our rats was lower than that previously observed.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Comparison of automated versus non-automated procedures of the TUNL probe test.

Both studies used a delay of 6 s. Data were obtained using an automated (current study: n = 12, striped boxes) versus a non-automated (Talpos et al., 2010: n = 7, grey boxes) system. The number of windows differed between studies (current study: three rows of five; Talpos et al., 2010: two rows of seven). The classification of the distance separations by the number of stimuli separating the lit windows during the sample and the choice phases of the test also differed between the studies (current study, Adj (0), Med (1, 2), Max (3) and Talpos et al., 2010, Adj (< 3), Med (3 ≥ n < 6), Max (≥ 6)). Boxplots show median, quartiles, and 10^th/90^th percentiles. *p < 0.05

https://doi.org/10.1371/journal.pone.0169476.g006

4.1 Specificities of the automated procedure and data collection

The experimental conditions on the first training day (unrestricted access to the operant chamber for 24 h) proved efficient at overcoming any hesitation by the animals to enter the sorter and operant cage. Taking advantage of the rats’ natural tendency to follow each other, all animals entered the system as soon as they were given access to it, interacted with the touchscreen and retrieved pellets from the food tray (Fig 2). Familiarisation of the animals with the system was most likely facilitated by the possibility of making high frequency visits of short duration to the operant chamber. A further indicator of the low level of stress toward the automated system and the sorter was in Training 2, where a short latency period prior to entering the sorter was observed the first time it was activated, and a negligible time lapse was seen before entering the operant chamber from there. This was confirmed in all subsequent training phases, with the persistence of regular, voluntary individual visits to the operant chamber.

Frequent visits to the operant chamber during training were also sustained by the combination of short training sessions and intersession intervals (section 2.3). Under these automated conditions, the animals rapidly learned the requirements of each training phase (Fig 3). In Training 5, a poke into the food tray was required to initiate the trial. During Training 6, incorrect pokes were punished. These additional steps slightly complicated the instrumental sequence of actions and may explain the higher number of sessions that were required to meet the criteria in Training 5 and 6 (Fig 3A). All rats achieved the acquisition criterion of each training phase in one or two days (three days in Training 2). This was remarkably fast considering that all rats in a group (each n = 6) only had access to one operant chamber.

In addition to their strong motivation to enter the system, several parameters (e.g., short latency prior to making a choice and collecting a reward, short duration of a trial and short exit latency), proves that the self-motivated start of a cognitive task did not affect the animal’s efficiency during the task. It also indicates that the motivation between animals and sessions was stable. These points are important in establishing the validity of this automated and self-initiated, experimenter-free method. The rats mostly used the dark phase (as opposed to the light phase) of the day, which can be expected for a nocturnal animal (Fig 2).

4.2 Adaptations of the TUNL training procedure

Training 2 (initial touch) of the TUNL training procedure differed from Training 3 (must touch, all lit) only in that a pellet was given after 30 s of inactivity in Training 2. The animals had already learnt to touch a window at the end of Training 2 and were no longer using the “free” pellet option (Fig 3). Thus, they were already fully conforming to Training 3 requirements at the end of Training 2. In such a situation, one might consider keeping phase 3 of TUNL training very brief or omitting it. In Training 6, as described in Oomen et al. (2013), punishing incorrect choices with timeouts should have increased efficiency during the TUNL acquisition phase. We found no difference in performance levels at the beginning (first block) of the TUNL test between our current study (which included a training stage to punish incorrect choices) and the TUNL data of Talpos et al. (2010), which did not include such a training phase (Fig 4; S1A Table). However, this lack of facilitation effect on the TUNL performance (first block) could also be due to method differences between our and Talpos et al. (2010) study. It might then be considered to keep Training 6 (punish incorrect) very brief or even omit it from our automated setup. However, one should keep in mind that the existing protocols developed over the course of 20 years consist of different phases based on fundamental psychological processes. The important experience an animal makes during Training 6 is to learn about the consequence of an ‘incorrect response’. It should be avoided that an animal makes such an experience for the first time during the actual cognitive testing since this would introduce a confounding factor.

4.3 Comparison of results within and between laboratories, and advantages of the automated and standardized approaches

Rat performance during training and testing phases in our fully-automated system was comparable between the two experimental groups (S3 Table), and was also comparable to results obtained in a less-automated system (Figs 4 and 6) [14].

The level of performance observed here in the TUNL task acquisition (Fig 4) and TUNL probe tests (Figs 5 and 6) was as expected. However, performance was higher when choice and sample stimuli were further apart, both with shorter and longer delays between sample and choice phases (Fig 5A). The interference test, which evaluates the influence of the previous trial on the performance of the current trial (Fig 5B), showed that the position of the previously correct window did not affect the following choice, in agreement with previous finding [14,20]. While the majority of training sessions in previous studies had a duration of 60 min (S1A Table), most of our sessions had a duration of 30 min, and only 10 min for Training 2. Under manual training conditions, an animal usually had only one training session per day, requiring 10 to 30 days to complete the training [13]. In contrast, in our automated system, only seven to nine days were required to complete a comparable amount of training. Likewise, TUNL training required 20 [14] to 30 or 35 [13] days with manual transfer of the rat; only 10 to 11 days was required in our automated system. In comparison with previously published data [14,20], our results confirm the validity of this fully-automated approach for rapid and efficient cognitive testing of undisturbed groups of rats. TUNL task acquisition was achieved in less days than reported in previous studies [14], and training the animals took a little longer than a week, which is a very short length of time for this type of instrumental learning procedure.

Interestingly, while the albino strains of rats are commonly considered to perform less efficiently in cognitive tasks than pigmented strains, the albino Wistar Han rats (WH) used in this study performed at the same level as the pigmented strain used by Talpos et al. (2010; Lister Hooded, LH; Fig 4C). WH rats perform worse than LH rats when detection of visual detail is required, due to their poor visual acuity [21]; however, WH rats equal LH rats in performance in a spatial discrimination task [21]. Further studies are required to determine whether the good WH rat performance is due to the shorter delay used in our study (2 s) compared to that used in Talpos et al. (2010; 6 s) during the TUNL acquisition task, procedural differences (distance between stimuli, total number of windows) or a facilitating effect of our automated conditions. The rats acquired the TUNL task in approximately half the number of days reported in Talpos et al. (2010), with a level of performance comparable between both studies for blocks of five sessions (Fig 4C). However, our rats reached the level of 70% correct choices after an average of five days (approximately 10 sessions), while this level was reached after approximately five sessions in Talpos et al. (2010). These observations could explain the lower median of our results in blocks 1 and 2 (Fig 4C) compared to Talpos et al. (2010). Procedural differences (distance between stimuli, total number of windows) may be the reason for these variations.

Surprisingly, at the medium condition of separation in the TUNL probe test, the performance of the rats was higher in our study than that reported in Talpos et al. (2010; same delay for both experiments), but declined more dramatically when conditions increased in difficulty (Fig 6). Thus, our WH rats may have been more sensitive to the increasing difficulty of test conditions. As our rats reached a somewhat higher level of performance than the pigmented rats in the easier conditions [14], our results exclude the possibility that WH rats consistently perform inferiorly to LH rats. Such an increase in variability between the groups under the more challenging test conditions is not uncommon. It remains to be seen whether there is a true variation in task performance between rat strains.

The development of procedures that do not require human handling is a promising avenue for increasing standardization and repeatability between studies and, thus, in increasing the translational validity of potential therapeutic effects [22,23]. Our automated, experimenter-free approach is therefore a promising tool for drug testing studies that usually require multiple and lengthy testing phases.

The sustained activity of the animals during the TUNL test sessions (the latencies and number of trials per minute) is a further indicator of the suitability of this procedure for drug testing that requires a low variability in behaviour between animals [23]. Our system could easily be expanded with an automated drug delivery system that dispenses drugs within the sorter. A liquid feeder could deliver a drug and the animal would consume its dose before being admitted to the operant cage [24]. This would synchronise the onset of the drug effect with the individual test phase. Admission to the sorter compartment for drug administration alone would also be possible.

With this fully-automated setup, longitudinal and uninterrupted behavioural testing is in synchrony with the animals’ normal activity rhythm. When animals are transferred by hand to the testing apparatus, their motivation to perform the task cannot be standardized. Moreover, whether an animal is awake or asleep shortly before the test influences the test outcome [25,26].

The fully-automated approach is also advantageous for the experimenter. In the current study, a reduction in daily experimental time of up to 80% was experienced, and the cost of equipment was significantly reduced. In the training phase, one rat is usually given one training session (60 min duration) per day and requires between 10 and 30 sessions before entering the testing phase [13]. Considering the time required to handle one rat before and after each session (i.e., 2 x 10 min = 20 min), a group of six rats using only one operant cage would require a total of 2 h of handling-related time and six hours of training (a total of 8 h per day) for 10 to 30 days (1 session per day). With our system (with only one operant cage available), the time was reduced to a maximum of 2 h, including the weighing of the rats, checking the system, altering schedules and exporting data for analysis. Moreover, the added possibility of training the animals for 24 hours per day drastically reduced the total number of sessions and days required as it was also found in single-housed mice using and automated system for testing cognitive flexibility [27]. One touchscreen system was used for six rats; however, this could be increased to 12 rats with a single touchscreen connected to a second home cage system (Fig 1), with the two groups kept under opposing light/dark cycles. Other types of operant chambers may also be easily attached to the sorter, and the training conditions established here could be adapted to a large variety of tests. However, the extent to which automation will increase reproducibility of results between laboratories remains unknown. Its use in an increased number of laboratories would allow such a comparison.