Study population and housing
In November 2019 and February 2020, we used 30 sexually mature female junglefowl (36 weeks old in November and 50 weeks old in February). We only used females because adult males are not very willing to work for food rewards (Zidar et al. 2018) which was required in some of our tests. These females (hereafter referred to as ‘subjects’) came from a captive, pedigree bred, population maintained by Linköping University, Sweden (described in Sorato et al. 2018). All subjects were, as chicks, marked with unique numbered wing tags, enabling individual identification. The subjects used in this study differed slightly in early experiences due to a previous experiment (Garcia et al. in prep). Until they were five weeks old, these subjects had been raised at Linköping University. During this time, some subjects (n = 11) were raised in smaller groups (nreplicates = 4) , each consisting of seven individuals, whereas others (n = 19) were raised in larger groups (nreplicates = 3), each consisting of 16 individuals (more details in supplementary information). Twenty-five of the subjects used here experienced a battery of cognitive and personality tests during their first five weeks of life, in this earlier study (more than seven months prior to the current study), while five of them did not. At five weeks old, subjects were moved from Linköping University to an agricultural college (Vreta Gymnasiet) and housed in adult chicken facilities, where they were housed during this study. Specifically, during this study, all subjects were housed in an enclosure (6 m3) with sawdust for substrate and access to an outdoor area (250 × 260 × 400 cm; H x W x L) during the day time. We kept subjects on a 12:12 light cycle with lights on from 7am to 7 pm. While at adult facilities, all subjects lived in the same enclosure and were all exposed to the same husbandry practices. Therefore, from more than seven months prior to the current study, all subjects had similar experiences. Due to logistical constraints, our red junglefowl were housed with 11 white leghorn females from 5 weeks of age. As our junglefowl and the white leghorns had been together for over 30 weeks before testing commenced, their social group should have been stable at the time of testing. At the end of the study, all subjects continued to be maintained by Linköping University.
General testing procedures
We conducted all testing between 9am and 6 pm and tested all subjects individually. Subjects took part in tests in this order: detour test, foraging test, simulated predator attack test, and staged contests (jn November 2019), and repeat testing in a detour test (in February 2020). For all tests, we collected data by direct observation. All tests were performed in the same way, and used the same equipment and set up for all subjects. Thus, there were no differences in how subjects experienced the tests that could have affected how they performed during testing. Further, no subjects experienced disturbances during testing. Testing was conducted outside of the subjects’ home pen, in familiar indoor rooms containing test arenas, which subjects were taken to by the observers, to prevent focal subjects being disturbed by other subjects in the home pen. To prevent observer bias, the subjects’ levels of impulsive action and persistence were not known to the observers in other tests. Food motivation can affect performance on cognitive tests (e.g., Rowe and Healy 2014; Smulders 2019; van Horik et al. 2018), thus we took steps to avoid variation in food motivation. First, we provided subjects with ad libitum food in their home pen, so none were food deprived at the time of testing. Second, if subjects repeatedly showed low food motivation in any test, we removed them from further testing and from our analyses. In total, this happened for four subjects in November (in the original inhibitory control tests) and one subject in February (in the repeat inhibitory control tests).
We used a detour test (Boogert et al. 2011; MacLean et al. 2014) to measure impulsive action and persistence, therefore, assessing inhibitory control. We conducted pre-training for the detour test, and the test itself, in an arena (82 × 53 × 50 cm; L x W x H). Before testing, we trained subjects (n = 30) to use a detour (i.e., to insert their head into the center of an opaque tube, 15 L × 7.5 Ø cm, from the side, to obtain a reward, one mealworm, sensu MacLean et al. 2014; Ryding et al. 2021). We initially helped subjects to learn this detour by tapping on the opaque tube, or guiding them to the reward in the tube, using extra rewards. The training criteria subjects had to reach before we assumed they had learnt the detour, was the retrieval of the reward from the tube five consecutive times without any help. It was easy for observers to notice when a subject had reached the training criteria, as doing so involved the subject performing five consecutive, specific, easily recognisable actions (sensu Garnham et al. 2019; Ryding et al. 2021). There was little variation in the number of training trials subjects needed to learn the detour, and all (bar one subject) learnt this relatively fast (most within 10 presentations of the opaque tube). Thus, we do not believe that variation in exposure to detour training influenced performance in the detour test. Further, as all but one subject (same as mentioned earlier) reached the training criterion quickly, this training did not appear difficult for our subjects. The one subject that did not learn the detour within three training sessions did not appear to struggle with the test but rather was not interested in obtaining the reward due to low food motivation. This subject was excluded from further testing.
Immediately after a subject reached our training criteria, she began the detour test. In total, 29 subjects participated in this test. We began each trial of this test by placing the subject and a transparent tube (15 L × 7.5 Ø cm) with a reward inside, at opposite short sides of the arena with the long side of the tube facing the subject (sensu Ryding et al. 2021). All subjects, but five, had already experienced transparent objects prior to this detour test due to participating in detour tests as chicks, more than seven months earlier (Garcia et al. in prep). As experience with transparent objects can affect performance in the detour test (van Horik et al. 2018), we took this difference in previous experience into account in our analyses, by conducting analyses on data both with, and without, these subjects included. In our detour test, each subject had to use the previously learned detour to access the reward without being helped. We gave each subject five trials in succession, and up to one minute per trial. Five trials captured variation among our subjects, while reducing the effect of learning over successive trails (as seen in Kabadayi et al. 2018; Ryding et al. 2021; van Horik et al. 2018). Similar to previous studies (Boogert et al. 2011; Garnham et al. 2019; MacLean et al. 2014; Ryding et al. 2021), we measured impulsive action as the number of these five trials in which a subject attempted to reach the reward directly by pecking at the transparent tube (we called this measure ‘Impulsive action’, and a higher score implied higher impulsive action). We measured persistence as the subject’s total number of pecks at the transparent tube across these five trials (termed ‘Persistence DT’, where DT referred to detour test, and a higher score implied higher persistence). Higher impulsive action and higher persistence imply lower inhibitory control (Adinoff et al. 2007; Dalley et al. 2011; Devos et al. 2014; Schippers et al. 2017; Winstanley et al. 2006). We considered subjects that did not approach the tube within one minute in three consecutive trials to have failed the detour test. Three subjects failed the test, again due to low food motivation, and were excluded from further testing. Two of these subjects had not experienced the detour test as chicks. Therefore only three subjects (of five) that had not experienced the detour test as chicks went on to be included in further testing and analyses.
We assessed the food motivation of all subjects that completed the detour test (n = 26), by seeing how many mealworms, out of five (presented sequentially), they would eat. All the subjects that completed the detour test showed high food motivation (all, but three subjects, ate all five mealworms and those three that did not eat five mealworms ate four mealworms). Thus, we believe that there were no significant differences in food motivation between subjects that completed the detour test and thus that food motivation did not influence the inhibitory control measures used in our analyses.
To determine if our measures of ‘Impulsive action’ and ‘Persistence DT’ were consistent within subjects over time, we repeated the detour test three months later (in February 2020), without any new training beforehand. One subject, which passed the detour test the first time, failed it this time. Therefore, the temporal consistency of impulsive action and persistence, two behaviours influenced by inhibitory control, was tested in 25 subjects.
To investigate whether inhibitory control was linked to a subject’s relative social rank in a group, we staged pairwise contests (sensu Garnham et al. 2019), in which each contesting pair contained one subject with a relatively higher impulsive action score (i.e., 4–5) and one with a relatively lower impulsive action score (i.e., 0–2). As our subjects were already part of a social group, by staging these contests we investigated links between current social rank and inhibitory control, not whether inhibitory control affected the establishment of social rank (as investigated in Garnham et al. 2019). We held contests in a room familiar to our subjects (used for the foraging test and simulated predator attack which were conducted prior to this test) with sawdust as substrate. A piece of apple was provided to give the subjects something to compete over and thus increase our chances of observing agonistic behaviours. To reduce the effect of morphology on contest outcome, we chose opponents that had comb (mm), tarsus (mm) and weight (g) within 10% of each other (sensu Favati et al. 2014; Garnham et al. 2019). We measured comb and tarsus with digital callipers, and weight using an electronic scale, prior to the detour test. Morphologically matched opponents could not be found for seven subjects, so they were excluded from the contests, resulting in 21 pairwise contests between 19 subjects. Most subjects participated in two contests, two subjects participated in three contests and five subjects participated in only one contest. This came from the need to have data from enough contests for our analyses while having as few contests as possible for ethical reasons. We considered a subject to have a higher social rank if her opponent avoided her three consecutive times (sensu Garnham et al. 2019) and contests were gently terminated immediately after this occurred. We only tested each pair once, as, in red junglefowl, individual rank is generally stable over time (Collias 1943; Collias and Collias 1996). Furthermore, during the time period over which data was collected, the subjects’ social group did not undergo any changes that could have influenced individual rank (e.g., removal, or addition, of birds).
To measure how inhibitory control linked to foraging efficiency, subjects that had completed the detour test (n = 26) took part in a foraging test in a room (284 × 298 × 305 cm; L x W x H) in which they were presented with nine glass petri dishes with lids (84.5 Ø mm) equally spaced 80 cm from each other. Each petri dish had three accessible mealworms on the top of, and three inaccessible mealworms below, its lid. Therefore, we created a foraging situation in which remaining at a dish trying to access inaccessible food, rather than leaving it to find accessible food elsewhere, led to lower foraging efficiency. Due to potential individual foraging speed differences, and wanting to limit how long subjects spent in testing, the test had three possible end points: 1) As subjects could potentially visit all patches in two minutes (confirmed by pilot tests), if a subject spent more than two minutes in the test, the test ended when she had visited three patches, as this was enough to get sufficient data on her foraging efficiency. 2) If a subject had visited more than three patches, but less than seven patches, by two minutes, the test ended at two minutes. 3) If a subject had visited seven patches before two minutes, the test ended at the seventh patch, as we needed to leave at least a couple of patches un-foraged for the simulated predator attack (described below). To assess a subject’s foraging efficiency, we recorded the number of patches she visited, divided by the time she spent foraging (i.e., eating, pecking at food, scratching for food, or walking directly to a food patch). A higher score implied higher foraging efficiency. Subjects always ate all the accessible worms at a patch before moving to another patch, thus no subjects obtained a higher foraging efficiency by leaving accessible worms unconsumed. To test if persistence in this test could affect foraging efficiency, and if individuals showed similar levels of inhibitory control in both the detour and foraging tests (sensu van Horik et al. 2018), we recorded persistence in this test as how many times a subject pecked towards the inaccessible food for all patches (‘Persistence FT’, FT referred to foraging test, a higher score implied higher persistence). As all subjects ate available worms when they came across them and tried to reach the inaccessible worms, we feel that all subjects had similar levels of food motivation during this test.
Simulated predator attack test
Immediately after the foraging test, to explore whether inhibitory control was linked to risk taking under perceived threat of predation, subjects (n = 26) were exposed to a simulated predator attack test (sensu Favati et al. 2016). We released a black model outline of a hawk (62 × 31 cm; W x L) down a zip-line placed diagonally across the room in which the foraging test had taken place (from 287—114 cm above ground level). The model imitated the flying silhouette of a hawk of the Accipiter family. Hawks and other raptors are natural predators of Gallus species (Evans et al. 1993; Collias and Collias 1967). Thus, this outline was predicted to trigger anti-predator behaviour in our subjects (e.g., Evans et al. 1993; Wilson and Evans 2009). We scored the intensity of each subject’s response to the model, which could indicate how threatening they perceived the simulated attack as being, thus we termed this measure ‘Threat perception’ (0–5, 5 suggests that the subject found the predator model very threatening, Table 1). As a subject may remain vigilant for longer, and, therefore, take longer to return to foraging after being exposed to a predator stimulus, if they perceive it as a greater threat, we also measured latency to resume foraging and termed this measure ‘Time spent vigilant’. ‘Threat perception’ reflected a subjects initial response to the simulated predator, whereas ‘Time spent vigilant’ reflected a more delayed response. In this test, an a subject that perceived a greater level of threat (i.e., had higher ‘Threat perception’) can be considered more risk taking than one that perceived a lower level of threat, but took the same amount of time to return to foraging (i.e., same ‘Time spent vigilant’). Therefore, we made a combined ‘Risk taking’ measure (‘Time spent vigilant’/’Threat perception’), which we used in our analyses, in which a lower value indicated higher risk-taking. All but three of the subjects responded to the predator stimulus by running and/or alarm calling (typical responses to predators in fowl, Evans et al. 1993; Wilson and Evans 2009). Thus, overall, our study subjects appeared to perceive the predator stimulus as a threat.
All the subjects used in this study had been regularly handled since soon after hatching, were accustomed to human presence, and were habituated to being alone in testing arenas prior to testing. When in their home pen, all subjects had access to enrichment (perches, shelters, and sawdust for dustbathing) and ad libitum access to commercial poultry food and water, thus they were never starved prior to testing. All training and testing sessions lasted max 15 min, and we returned subjects to their home pen for at least one hour rest between sessions. Subjects were also given at least one hour rest between contests. Regarding our simulated predator test, we accept that the model predator acted as a stressor to our subjects. However, the stress they experienced was short lived, and all bar two subjects returned to foraging within two minutes after the simulated attack (the two that took longer to return to foraging did so within five minutes). Thus, we do not think the experience of the simulated predator attack had any long-term detrimental effects on our subjects. Concerning our social rank tests, previous pilot work in our group has found that using mirrors does not produce relevant behavioural data. Further, as chicken have a higher flicker fusion frequency than humans (Lisney et al 2011), using video playback would have been unlikely to produce relevant behavioural data as our subjects would most likely have perceived this as not normal behaviour. Consequently, contests with live individuals were necessary to investigate current social rank. We arranged contest pairs, and only had one contest per pair, to make the overall number of contests as low as possible. We were always ready to break up aggressive interactions if they occurred, and veterinary care was readily available if needed. No contests resulted in any serious fighting and no subjects obtained any injuries.
All analyses were conducted in R studio (v. 3.5.2). We used non-parametric statistical analyses, as our data did not meet the assumptions of parametric statistics. Results of statistical analysis with a p-value < 0.05 were considered significant.
For the first stage of our analyses, we investigated if how subjects were housed, or the replicate they were housed in, as chicks, biased our results. As this investigation was not a focus of this study, how it was conducted is described in the supplementary information. If we found an effect of how subjects had been housed, or the replicate they were housed in, as chicks, on any of our measures, for these measures we performed separate analyses on data from subjects housed in different ways, or in different replicates, as chicks.
To explore the temporal consistency of ‘Impulsive action’ and ‘Persistence DT', between November 2019 and February 2020, and relationships between ‘Impulsive action’, ‘Persistence DT’ and ‘Persistence FT’, measured in November 2019, we used Spearman’s rank correlations. Spearman’s rank correlations were also used to investigate whether ‘Impulsive action’, ‘Persistence DT’ or ‘Persistence FT’, measured in November 2019, linked to ‘Foraging efficiency’, or ‘Risk taking’. To examine whether ‘Impulsive action’, ‘Persistence DT’ or ‘Persistence FT’, measured in November 2019, predicted social rank measured in same-sex pairwise contests, we used paired Wilcoxon signed-rank tests (winners vs. losers). While higher ‘Impulsive action’, ‘Persistence DT’, and ‘Persistence FT’ all reflected lower inhibitory control, we analysed each of these behaviours separately as each may be affected by different aspects of inhibitory control.
As previous experience could affect performance in cognitive tests (e.g., Boogert et al. 2018; Dochtermann 2010), we ran all our analyses both with, and without, the three subjects that did not experience the detour test as chicks. Results are reported from the analyses with all subjects, and any qualitative differences in results that came from removing the subjects that did not experience the detour test as chicks, are given for each result below. We also ran the analyses with, and without outliers. First, we only removed extreme outliers (i.e., points more than three times the interquartile range from upper and lower quartiles, Hodge and Austin 2004; Levend Asikoglu 2017), then we removed both extreme and mild outliers (defined as points between 1.5 and three times the interquartile range from upper or lower quartiles, Hodge and Austin 2004; Levend Asikoglu 2017). However, neither of these methods of outlier removal qualitatively affected our results, thus we only report results including all data.