Animals and husbandry
We used eight adult gidgee skinks (Egernia stokesii) of undetermined sex in this study. Lizards were collected from the wild around Fowlers Gap Arid Zone Research Station (− 31.086972 S, 141.704836 E), New South Wales, Australia during March 2018. Individuals were transported by car in cloth bags within a cooler box to Macquarie University, Sydney within a week of capture and were individually housed in plastic tubs (683 L × 447 W × 385 H mm). Lizards were housed in a temperature-controlled environment (24 °C ± 2 SD), with relative humidity between 30 and 60% and a light cycle of 12 h (06:00–18:00 h). In addition to the room lighting, UVB light (URS® Outback Max 10.0 UVA & UVB tube) was provided approximately 800 mm above the enclosure floor. A heat cord underneath one side of the enclosures ensured that animals were able to thermoregulate by increasing the temperature to up to 33 °C (± 2 °C SD), thereby creating a thermal gradient which lizard readily used. iButtons (Thermochron iButton model DS1921) recorded temperature hourly within enclosures. Each enclosure was lined with paper and included a refuge for shelter (upside down, brown plant saucer 200 mm in diameter; 40 mm high), a water bowl (heavy, poly resin reptile water bowl, 130 L × 110 W × 40 H mm) and a wooden ramp, a stone, some bark, leaves, and a 150 mm long PVC tube as enrichment.
Lizards were fed on Monday, Wednesday and Friday with an assortment of small cut fruits (e.g. apple, banana, pear, tomato, strawberry) and vegetables (e.g. carrots, zucchini, capsicum, celery, broccoli, different leafy greens such as lettuce, kale, pak choi, choisum, beet root greens). On Fridays, they received 2–3 crickets powdered with aristopet Repti-vite and URS Ultimate Calcium in addition to the fruits and vegetables. On days on which test sessions were conducted, lizards did not receive their regular diet but were only given food as reward when making a correct choice (1–10 times 0.065 g ± 0.021 SD of carrot) except for Fridays, when they were fed their regular diet (to provide optimal nutrition) as well as any reward obtained during test sessions. On Fridays, animals were fed only after all test sessions were completed. Lizards had ad libitum access to water.
All lizards were tested within their home enclosure to avoid stress caused by handling (Langkilde and Shine 2006). Before the start of a test session, a lizard was gently carried (within its enclosure) to a test area approximately 3 m away from the housing area within the same room. Lizards were given 5 min to acclimate before the first test trial started. A grey curtain surrounded the test area and obscured the researcher during trials. Similar to the housing set-up, a heat cord installed under part of the enclosure ensured that lizards were able to thermoregulate during test sessions.
Apparatus and stimulus cards
The wooden apparatus consisted of two wooden ramps (170 L × 65 W × 50 H mm) glued together back-to-back with a wooden coaster (3 mm L × 93 W × 113 H mm) in-between, using non-toxic silicon (Fig. 1b). Each lizard was tested with only its own apparatus to prevent any effect of scent on behaviour. Stimulus cards (60 L × 90 H mm) were created in Microsoft PowerPoint and then printed, laminated, and cut out. The squares depicted on the cards used for the pattern discrimination test were 1 cm2 (10 mm × 10 mm) in area. Each lizard received their own sets of cards (i.e. not interchangeable), which were cleaned with 70% ethanol after each session.
Each lizard participated in one session of 10 trials (in the target training: 10 training trials and in the discrimination test: 1 training trial and 9 test trials) per day between 7:30 and 10:30 h, every day for 5 days a week, Monday to Friday. The order in which the test subjects were tested each day was randomised to avoid order effects. First, all enrichment items and the water bowl were removed from the enclosure and the lizard gently covered with the refuge to prevent it from watching the set-up. Next, the lizard was slowly moved as far back as possible while under the refuge (Fig. 1a) and a wooden apparatus (Fig. 1b) was placed at the opposite end of the enclosure nearest to the experimenter. The lizard was left undisturbed under its’ refuge for 30 s before the first stimulus presentation. A trial started by removing the refuge and presenting the lizard with a single stimulus card (target training, Fig. 1c) or two cards attached to the wooden apparatus (discrimination tests, Fig. 1b, c). A trial lasted until the lizard had either touched a card or a maximum of 5 min had elapsed after which the trial was terminated. If a lizard did not touch a card (i.e. make a choice) in two consecutive trials the whole session was terminated. At the end of a trial the lizard was again gently covered by the refuge and moved backwards within the enclosure for an inter-trial interval (ITI) of 30 s.
Carrot strips (created using a grater and then cut into equally sized pieces, 0.065 g ± 0.021 SD each) were used as a reward for a correct response both during target training and discrimination learning. Carrot is a favoured food item for these lizards in captivity (personal observation made during regular husbandry; also see Szabo et al. 2021c) and were prepared fresh each day.
Target training was used to teach lizards to touch a target card attached to the wooden apparatus with the goal to use this behaviour in future simultaneous two-choice discrimination tasks (similar target training procedures were used in Hellmuth et al 2012).
In the first step (Pre1) we taught the lizard to associate touching a card with receiving food (Supplementary Video M1). To this end, we presented the lizard with the single grey stimulus card in front of its head, 15 mm from its snout (Fig. 2) after the refuge was removed at the beginning of a trial. The stimulus card was attached to a pair of forceps using Bostik Blu-Tack® adhesive putty for easy presentation. To initiate approach of the card, the experimenter presented a strip of carrot held in a second pair of forceps directly in front of the stimulus card for 1 s, after which the carrot was hidden behind the stimulus card. This was repeated every 5 s until the lizard touched the card with any body part, which resulted in the lizard receiving the reward. This step was repeated for as many trials as it took until the lizard touched the card without the presentation of the carrot. A lizard moved on to the next step after touching the card without reward presentation in every trial for at least three consecutive sessions (i.e. 30 trials).
The next two steps were designed to teach the lizard to approach the card from a distance. To this end, we presented the stimulus card 50 mm away from its snout (Pre2; Fig. 2; Supplementary Video M1). If the lizard did not approach the card it was shown the carrot, as in the previous step. The criterion to move on was, again, to touch the card without reward presentation in every trial for at least three consecutive sessions (i.e. 30 trials). From this point on, we presented the card 50 mm away from the lizard (Pre2 procedure) in every first trial of a session (in Pre3, colour and pattern discrimination) to keep reinforcing the touching of the stimulus card throughout the whole experiment.
In the third and final step (Pre3), the cue card was held in front of the wooden apparatus (Fig. 2; Supplementary Video M1) from the start of a trial (except for trial 1 in each session) on the left or right side in a predetermined pseudorandom fashion no more than twice consecutively on the same side. Again, if the lizard did not approach the card immediately, the carrot strip was shown. For a lizard to move on to the visual discrimination test, they had to approach and touch the card without reward presentation in every trial for at least three consecutive sessions (i.e. 30 trials).
We used two lizards in the pilot. In the visual discrimination test (T1), two cards, one light and one dark blue were attached to either side of the apparatus (Figs. 1b, and 2; Supplementary Video M1). Light and dark blue were chosen as stimuli for the pilot because gidgee skinks had shown an ability to discriminate between these two colours in a previous study (Szabo et al 2021b). One of the two lizards used in the pilot was assigned light blue as the correct stimulus, while the other was assigned dark blue as the correct stimulus. Trials were run as follows (except for trial 1 in each session; see above): first, the cards were simultaneously attached to the apparatus after the lizard was already under the refuge. Second, the refuge was removed and the experimenter moved behind the curtain. Third, the experimenter observed the lizards behaviour live on a video screen. Lizards were filmed from above using a CCTV system (3-Axis Day & Night Dome Camera recorded with a H.264 Digital Video Recorder). If the lizard touched the correct card the experimenter emerged from behind the curtain and rewarded the individual with a carrot strip presented in forceps. If, however, the lizard touched the incorrect card, the lizard was covered with the refuge and moved gently to the back of the enclosure in preparation for the next trial (for an ITI of 30 s). If the last choice within a session was incorrect, we conducted another Pre2 trial in which the grey target card was presented 50 mm in front of the lizard to ensure a session ended on a positive note.
Each lizard received one target training (Pre2 procedure) plus nine discrimination trials within one session per day, and had to complete at least three sessions before the learning criterion of 8/9 correct choices or better in each of two consecutive sessions was applied. The side (left/right) that the correct card was presented was predetermined and pseudorandomized to never appear on the same side more than twice in a row. To confirm that a lizard had learned the discrimination they were tested on a reversal session (T2) in which the previously incorrect stimulus became correct and vice versa. The pilot was conducted from the end of April to the beginning of June 2019.
Pattern discrimination test
We used six naïve lizards to test pattern discrimination using the same training and test procedure developed and verified in the pilot. The whole experiment (including target training and pattern discrimination) was conducted from June to September 2019.
Target training (Pre1–Pre3)
We made small changes to the first two target training session to facilitate learning of the target-trained behaviour. In the first five trials of the first training session we presented a carrot strip in forceps to the lizard without the target card then placed the carrot on the enclosure floor (1–2 cm away from the lizard) for the lizard to eat (pre-pre). In the following five trials of the same session, the carrot was presented again in forceps but the lizard had to eat the carrot from the forceps held out by the experimenter (preT). All lizards ate all carrots in the first training session.
In the first five trials of the second training session the target card was presented 15 mm away from the lizard’s snout and the carrot strip was presented in front of the card and not hidden behind the card (Supplementary Video M1). This resulted in the lizard touching the card while eating the carrot. The following 5 trials of the same session were conducted as described above (Pre1).
Finally, instead of holding the target card in front to the apparatus in Pre3 of the training, it was attached to the apparatus from the start of a trial. The rest of the target training was performed exactly as described above (Pre1–Pre3).
In the pattern discrimination test, lizards had to learn to discriminate between a grey card depicting two squares and a grey card depicting eight squares. We followed the procedure described above for the visual discrimination test (T1): lizards were each tested in one session of 10 trials (1 training and 9 test trials) per day for 5 days a week until they reached a learning criterion of 8/9 correct choices or better in each of two consecutive sessions (after completing at least three sessions). For three of the six test lizards (randomly chosen) the card depicting two squares was assigned as correct (stimulus group 2), while for the other three lizards the card depicting eight squares was assigned as correct (stimulus group 8). The side a stimulus card was presented was predetermined for each session and followed a pseudorandom order in which the same card was never presented more than twice on the same side. As described above, the first trial of each session was conducted as a Pre2 training trial. This ensured that, even when a lizard made many incorrect choices during test trials (not receiving food for touching a card), they would continue to reliably perform this behaviour throughout the whole experiment.
Based on the data collected in the pilot we expected lizards to acquire the pattern discrimination within approximately 10 sessions (90 trials). However, we did not find the expected performance and decided to implement some minor changes in the test procedure to investigate the reason for the lizards’ poor performance:
Starting from the 12th test session, we replaced the single grey card presented in the first trial (Pre2) of each session to reinforce the target behaviour (touching the card) with the stimulus card that was assigned as correct for each lizard (similar to a matching-to-sample test) (Supplementary Video M1). We hoped that reinforcing the correct stimulus card in this way would improve performance, but it did not (see "Results").
After the 21st session, we conducted a whole Pre2 session (target training) but we used the correct stimulus card (either showing two or eight squares depending on test group) instead of the empty grey card. For a whole session of 10 trials, we presented the correct card 50 mm away in front of the snout of each lizard reinforcing touching of the stimulus cards with a carrot (all lizards reliably approached and touched the card in all trials without the presentation of the carrot). We hoped that this would further reinforce choosing the correct card, but it did not lead to an improvement in performance (see "Results").
Starting from the 32nd test session, we moved the apparatus instead of covering the lizard with the refuge and moving it backwards. We hypothesised that stress might negatively affect the lizards’ performance and wanted to reduce physical handling time. After a lizard had made a choice (correct or incorrect), the apparatus with the stimulus cards attached was slowly lifted out of the enclosure. Only thereafter, was the lizard gently covered with the refuge but not moved. We attached the cards in the configuration needed for the next test trial to the apparatus before placing it back inside the enclosure at the opposite end, furthest away from the lizard (Supplementary Video M1). This change had a significant but small effect on trial choice in one group and a strong effect on latency in both groups (see "Results").
Starting from session 42, we stopped cleaning the stimulus cards after each session to facilitate odour accumulation on the correct card because it was touched by the lizard more often than the incorrect card (first trial in each session). This change did not improve the lizards’ performance (see "Results").
Finally, in sessions 52–54, we replaced the incorrect card with an empty grey card to increase discriminability between the two stimulus cards. This change had an effect on the lizards’ choice performance (see "Results").
For the target training we recorded if the reward was shown to the lizard, how often it was shown to the lizard, and if a lizard made a correct response (touching the cards) thereby receiving the reward for each trial. For each test trial (colour and pattern discrimination) we recorded if the response was correct or incorrect (1—correct choice, 0—incorrect choice), the latency to choice (from the removal of the refuge up to the point when a lizard touched a card regardless of if the response was correct or incorrect) in addition to the above described measurements. Furthermore, for each trial we recorded the date a session was performed, the start time of each session, and the initials of the researcher conducting the trial (all trial were conducted by the first author). We also recorded which stimuli were used in each trial (e.g. g—empty grey card, l/r—left or right position of the grey card in Pre3, lb/db—light/dark blue card presented on the left from the experimenters perspective in T1/T2, 2/8—card showing two or eight squares presented on the left in T1). Enclosure temperature was recorded with Thermochron iButtons (model DS1921) and added to the raw datafile based on date after data collection had finished.
We were primarily interested in analysing if any of the five changes we made to the procedure had an effect on lizards’ choice performance and latency to choice. To this end, we assigned a unique letter (b–f, a representing the original procedure used in the first 11 sessions) to the sessions representing a change in procedure (= stages of the test). We used Bayesian generalised linear mixed models (GLMM; R package MCMCglmm, Hadfield 2010) to compare the performance following each change with the previous sessions: stage a was compared to b (Pre2 trial 1 with a card showing the correct stimulus), b compared to c (additional target training with a card showing the correct stimulus), c compared to d (reducing physical contact with the lizards), d compared to e (no cleaning of the cue cards with ethanol), and e compared to f (replacing the incorrect card with an empty grey card).
To analyse choice behaviour of the whole group (not considering stimulus group) we used choice made in each trial (1—correct, 0—incorrect, Bernoulli variable) as the response variable and both stimulus group and stage in interaction with session as the fixed effects. To analyse choice behaviour of each stimulus group, we used choice made in each trial (1—correct, 0—incorrect, Bernoulli variable) as the response variable and the interaction between stage (a–f) and session as the only fixed effect. Session (scaled and centred) was included as a fixed effect because we were not just interested in the overall effect but also in the possible effects on the rate of change (e.g. learning). Additionally, we wanted to know if choice performance increased across all sessions (excluding the last three sessions in which the incorrect cards were replaced). To this end we ran a model with choice made in each trial (1—correct, 0—incorrect, Bernoulli variable) as the response variable and session (scaled and centred) as the only fixed effect. In all models we included a random intercept of ID interacting with a random slope of trial nested in session as the random effect (random intercept and slope model). This way, we were able to account for non-independence and autocorrelation across successive choices (repeated measures of trial and session across individuals).
We ran similar models to analyse differences across stages (a–f) in latency to choice, but instead used the log transformed latency in seconds as the response variable. Log transformation was used because latency data generally are log normal distributed and the DIC of the model using log transformed latency was much smaller than that of the model run without transformation (DICnon-log = 16,650.5; DIClog = 3334.9). Using the posterior of the models we calculated mean estimates and Higher Posterior Density intervals (CIs—confidence intervals) for each stage comparison. We assumed statistical significance if the confidence intervals did not cross 0. Finally, we were interested if performance (choice and latency) were associated with lizard body size or room temperature. We added the lizards SVL (snout-vent length in mm) and room temperature and their interaction as additional fixed effects to the models looking at general patterns across sessions. In all cases, trial 1 (Pre2 trial) was removed before analysis.
As a prior we used a common weak prior (for all models) as we had no specific prior knowledge regarding the lizards’ performance using this testing procedure (for details see R code provided on OSF). We used binomial models with a logit link function when choice was used as the response variable and gaussian models with identity link function when latency was used as the response variable. For all models, we confirmed that no autocorrelation (correlation between lags < 0.1; Hadfield 2010) was present, that sufficient mixing (by visually inspecting plots of MCMC chains; Hadfield 2010) was achieved and that the Markov chain was run for long enough (Heidelberg and Welch diagnostic tests; Hadfield 2010). All analyses were conducted in R version 4.0.3 (R Core Team 2020) and all raw data sets generated during this study and code for analysis are available on the Open Science Framework (https://doi.org/10.17605/OSF.IO/SDUX7).