With this study, we wanted to investigate whether the experimental design, particularly the stress it induces, would influence the consistency of learning performances in cognitive tasks and their associations with personality traits. We therefore designed both low- and high-stress versions for each test. Across all tests, we observed a significantly higher temperature increase during the tests that have been designed to be more stressful compared to the low-stress tests. Moreover, the difference in temperature increases with the duration of the tests in a high-stress test while it remains constant in a low-stress situation (suppl. mat. S3). In addition to that, some behaviors also indicated that the high-stress situations were really experienced as more stressful. In the mazes, the latency to enter decreased by the 2nd trial in the low-stress situation, indicating a quick habituation. On the contrary, in the high-stress condition, the latency to enter stays steady across all five trials. In the PS tasks, the latency to approach the setup seems higher in the low-stress condition, potentially indicating a higher motivation to get access back to their home cage in the HSPS. However, the different approach latencies could also indicate a difference in perception and/ or attention rather than indicating a difference in stress. All together, the physiological and behavioral results show that the tests that were designed more stressful were also perceived as more stressful by the mice.
In addition to the measure of the stressfulness of the design, infrared thermography could be used as a measure of individual sensibility to stress. This method has been proposed to be a new non-invasive measure of stress in a couple of studies35. In mice, for example, differences in temperature increase have been linked with some personality and behavioral traits34,37. In this study, the temperature increase was consistent over quite a substantial time period (ca. 6 months) and across contexts, revealing consistent inter-individual differences in this physiological trait.
We did not find consistency between learning performances in low- and high-stress conditions although in general, wild mice exhibit consistency in the problem-solving tasks we used38 and we found problem-solving to be consistent when only considering the high-stress situations. Furthermore, we found that the personality-cognition correlations differ between the high and low stress conditions. This shows that the design-induced stress can have an impact on both trait consistency and cognitive syndromes.
Stress and cognition: absence of consistency
In the meta-analysis from Cauchoix et al.4, cognition was generally found to be repeatable across time and contexts. Despite the high variability in consistency between species and cognitive tasks, problem-solving and spatial learning were always found to be repeatable in adult rodents (including Mus musculus) except some species of the genus Sciurus4,9,39. While problem-solving was also repeatable in Mus musculus (even of the same population as used in this study) and in Apodemus agrarius using the same problem-solving test setups as in our low-stress situation38,40, here, we could not find consistency across the high- and low stress situation.
Stress is already known to influence behavior in such a way that it impacts the repeatability of personality traits. For example, acclimation time has been shown to influence the repeatability of activity in guppies41. The methodology is then expected to also influence results in cognitive experiments. Different spatial learning testing set-ups have already been shown to induce different stress levels22 and influence memory retention and working memory42. Nevertheless, the influence of stress varies from one context to the other: while strong or chronic stress is expected to impair cognitive performances, mild fear could have the opposite effect21,43. Additionally, perceived stress will be different from one individual to the other, which will modulate the effect of stress on cognitive performance by displacing or changing the shape of the relations stress-cognition. For example, it has been suggested that the inverted U-shape of the cognitive abilities could be moved along the stress axis following the personality of an individual. This means that the “optimal stress” (i.e., the stress level at which learning performances are the highest) could vary from one individual to the other20. In such a situation, stress-sensitive individuals would be better learners in a low-stress situation but the worse learners in a more stressful task, and vice versa, explaining the lack of consistency found in this study. A couple of studies tend to support this statement20,31,44 while some other studies do not33,45,46, however, the lack of control for the stress level of the task in the latter makes it difficult to interpret the generality and importance of this phenomenon.
Some factors other than stress are known to influence learning performance and might consequently decrease consistency if not accounted for in experimental designs. For example, previous experience of a task has been shown to increase performance in a similar task47,48. However, this has been controlled for in this study by splitting our set of animals into 2 groups which started either with high- or low-stress tests and our results show that there is no influence of the sequence of the test on the different learning measures (see Suppl mat S5). Motivation is another important factor that can influence performance in a cognitive test47. We tested the effect of motivation in the LSPS tasks by measuring the latency to eat the first mealworm. Despite a trend of the latency to eat the mealworm to influence the latency to solve the PS, the 2 other learning measures were not influenced by motivation (see Suppl mat S5) in accordance with previous results on wild mice38. This result suggests that, at least for the LSPS task, motivation did not influence learning performances.
Another factor explaining the lack of consistency is that some tasks that appear very similar to us could in fact rely on different mechanisms4. For example, an experiment of Troisi et al.49 showed that learning was not consistent in a spatial and a reversal learning task between 2 different spatial scales. They hypothesize that this could be due to differences in the cognitive mechanisms implied at different spatial scales (allo- vs egocentric navigation and cue use)49. This could contribute to the lack of consistency we observe between the high- and the low-stress maze, as one of the differences between them is their size. However, in their study, for both the spatial learning and the reversal learning tasks, the smaller scale test took place in the individual’s cage, while the larger scale tests happened in another room. We showed that such a difference in design leads to a different perceived stress for the animals, both, behaviorally and physiologically, while the stress level was not controlled for in the study of Troisi et al.49. Furthermore, a scale difference may only explain a lack of consistency in the mazes while all the problem-solving setups were of approximately the same size but still lacking in consistency. Nevertheless, this example shows that experimental setups and their intended effects on animals need to be designed very rigorously and several potential confounding factors need to be considered.
For both the maze and PS, correlations between learning measures and personality were different in high- and low-stress conditions. Learning performances in the HSM were linked with individual sensitivity to stress in the OF, while we found no personality-learning correlations with the LSM. On the contrary, learning in the HSPS tasks wasn’t linked with any personality measure but learning in the LSPS task was linked with the behavior and the temperature increase in the NE.
Correlations between different biological traits can result from shared underlying trade-offs10. For example, an active and bold individual is expected to explore a task faster, find cues easily and react to them more quickly than a shy individual. However, in this study, different correlations between personality and cognition are found following the stress level of the task, suggesting that these relations might be more complex than imagined17. As suggested for the lack of consistency, this could also be due to different underlying mechanisms. In an experiment on great tits (Parus major), Titulaer et al (2012) found personality-learning correlations only in the most difficult reversal learning task. They hypothesize that easy and hard tasks might be differently perceived for example by the attention that is paid to the cues. These context differences could then reveal mechanisms that underlie cognition16. Specifically, it is known that different neurobiological mechanisms are involved in low- and high-stress cognitive tests19. It has been shown in a spatial learning task that the memory-related protein ERK2 was activated during learning in the amygdala, a brain region linked to fear, in a high-stress situation, but not in a low-stress one50. Similarly, stress-related hormones such as glucocorticoids can bind to their respective receptors in the brain and affect cognitive performances51. In rats, corticosterone blockade in a stressful situation decreased learning performances in a spatial learning task, while corticosterone enhancement in a low-stress situation increased them52. This could easily explain both the lack of consistency found between the HSM and the LSM and that learning performances in a high-stress situation such as the HSM are correlated with stress-sensitivity. On the other hand, exploration is often linked with cognitive performances8, and especially aspects of problem-solving33,53. This explains that learning in a low-stress condition such as the LSPS task was correlated with personality in the NE. The involvement of these different physiological reactions following the stress context indicates that the cognitive performances are strongly influenced by the design of the experiment, while personality remains relatively consistent, hence, leading to different personality-learning associations. In addition, correlations between cognitive measures appeared stronger in high-stress conditions compared to the low stress conditions. It is suggesting that some of the mechanisms involved in the stressful situation are common to multiple learning tasks, while the different cognitive tasks in the low-stress situation are more independent or more strongly influenced by external factors.
Additionally, the lack of consistency between the two stress conditions suggests that cognitive traits may be particularly flexible. On an evolutionary point of view, the flexibility of the underlying mechanisms of cognition can be adaptive, as the optimal cognitive phenotype can be different in different stress contexts. Indeed, multiple trade-offs are involved in the determination of cognitive abilities and the optimal outcome can be different following the situation. The first one is the trade-off costs-benefits of cognition4. Advantages provided by learning are sometimes counteracted by direct (time and energy demand) and indirect (correlations with other traits such as personality) costs54. Secondly, the speed-accuracy trade-off describes the compromise between the precision/accuracy during a task and the time necessary to perform it55. The stakes of low- and high-stress situation are intrinsically different, as the individual’s life is supposedly more at risk in a stressful situation. This can influence the energy allocation and the importance given to speed and/or accuracy in a cognitive task55,56. Acting distinctively on cognitive performances in low- and high-stress situation would allow for a better fine-tuning of these traits.
However, we also find some differences between learning tasks. Cognitive syndromes are only present in the low-stress condition in the PS tasks, but we only find personality-learning correlations in the high-stress condition in the maze. In addition, while learning performances in the mazes are drastically different between the 2 stress conditions, it is much less strong in the PS tasks. Indeed, even if consistency could not be found between the learning measures from LSPS and HSPS, they still tended to be positively correlated. In the same way as consistency in cognition4, our results suggest that the effect of stress on cognitive performances and its strength are highly dependent on the cognitive task. Different mechanisms are probably involved in the different learning tasks and, therefore, they are differentially mediated by the stress level of the test. More cognitive abilities should be tested in order to have a broader view on the influence of stress on cognitive performances.
Evolution of cognition
For the two studied cognitive tasks, learning performances have been shown to be not consistent across different stress contexts and the correlations between the measured learning and personality traits were different. Controlling for these conditions would be very important to unravel when cognitive syndromes may affect fitness and what are their underlying mechanisms are. There are 3 important conditions for a trait to be subject to evolution2,57. The first one, the presence of inter-individual variation in a population, seems to be generally verified for cognition4. Secondly, the cognitive traits should be heritable. Even if not completely clear, multiple clues tend to indicate that at least some of these traits are transmitted to the next generation2,58. Finally, the inter-individual differences should lead to differences in fitness. A causality between cognition and fitness is hard to determine, but several correlations have been found2,4,54 suggesting that such a relationship exists. In addition to that, some studies found differences in cognition between populations or related species with different ecology that could be adaptive38,59. These results imply that cognitive traits probably under selection pressure. It also means that if cognition is correlated with some other traits, they won’t evolve in isolation8,57. Then, evolutionary changes in one of these traits would lead to concurrent changes in correlated traits even if these are not under direct selection60,61. The differences found in learning performances and their correlations with personality between stressful and non-stressful conditions shows the importance of taking stress into account when designing a cognitive experiment, as it could drastically bias and induce misleading conclusions on the underlying mechanisms and therefore also on the evolution of cognition.