Phaesants: Apparently ‘poor’ cognitive performance (slow reversal speed suggesting low behavioural flexibility) correlates with fitness benefits in at least some circumstances, implying a novel mechanism by which continued exaggeration of cognitive abilities may be constrained

The quick are the dead: pheasants that are slow to reverse a learned association survive for longer in the wild. Joah R. Madden, Ellis J. G. Langley, Mark A. Whiteside, Christine E. Beardsworth, Jayden O. van Horik. Philosophical Transactions of the Royal Society B: Biological Sciences, September 26 2018, Volume 373, issue 1756. http://rstb.royalsocietypublishing.org/content/373/1756/20170297

Abstract: Cognitive abilities probably evolve through natural selection if they provide individuals with fitness benefits. A growing number of studies demonstrate a positive relationship between performance in psychometric tasks and (proxy) measures of fitness. We assayed the performance of 154 common pheasant (Phasianus colchicus) chicks on tests of acquisition and reversal learning, using a different set of chicks and different set of cue types (spatial location and colour) in each of two years and then followed their fates after release into the wild. Across all birds, individuals that were slow to reverse previously learned associations were more likely to survive to four months old. For heavy birds, individuals that rapidly acquired an association had improved survival to four months, whereas for light birds, slow acquirers were more likely to be alive. Slow reversers also exhibited less exploratory behaviour in assays when five weeks old. Fast acquirers visited more artificial feeders after release. In contrast to most other studies, we showed that apparently ‘poor’ cognitive performance (slow reversal speed suggesting low behavioural flexibility) correlates with fitness benefits in at least some circumstances. This correlation suggests a novel mechanism by which continued exaggeration of cognitive abilities may be constrained.

---
1. Introduction

One powerful approach to understand how natural selection may act on cognition is to measure the performance of individuals in a particular cognitive domain, and then explore how their performance correlates with a (proxy) fitness measure [1,2]. This is achieved by deploying explicit psychometric tasks targeting specific, defined cognitive processes [3,4]. Because fitness itself is hard to measure [5], researchers tend to use proxies that are presumed to correspond to reproductive success and/or survival. This correlational approach has predominantly revealed a positive relationship between an individual's performance in the psychometric task and a (proxy) measure of their fitness. Ants Lasius niger that exhibited faster route learning had greater colony-level foraging success [6]. Male African striped mice (Rhabdomys pumilio) that escaped quickly from mazes also had increased probability of surviving to the breeding season [7]. Male guppies (Poecilia reticulata) that learned mazes quickly were preferred by females [8]. Male bitterling (Rhodeus ocellatus) (practising a ‘sneaker’ strategy) that exhibited better maze learning subsequently had higher reproductive success [9]. Male song sparrows (Melospiza melodia) that demonstrated better control in a detour-reaching task had a larger song repertoire [10]. Male starlings (Sturnus vulgaris) with better spatial learning exhibited longer song bouts [11]. One study of female Australian magpies (Cracticus tibicen dorsalis) reported a link between their reproductive success and a general factor summarizing their performance in a battery of four tasks [12]. By contrast, a study of spotted bowerbirds (Ptilonorhynchus maculatus) found no relationship between a male's mating success and his performance in a battery of six tasks, either individually or when his performance was summarized by a single component [13]. Only one study has reported a negative relationship: male song sparrows that were fast at spatial learning also had smaller song repertoires [14]. This implies that natural selection generally leads to more exaggerated cognitive performance and associated abilities.

Interpretation of these previous studies is complicated by three factors. First, in all cases except one [12], a single assay has been used for each cognitive process being investigated. Reliance on a single assay risks a misattribution of the mechanisms driving individual performance. For example, learning to discriminate between two colour cues may indicate the specific ability or inherent motivation to prefer one colour over another [3], rather than the more general ability to learn associatively. A more robust method would be to use two (or more) tests that assay the same putative cognitive mechanism but differ in format or cue uses and hence triangulate on the outcome (Volter et al. [15]). We considered two ubiquitous cognitive processes and tested each using two different test variants. Associative learning involves learning to associate a stimulus with a reward and may be tested using a binary discrimination. Reversal learning may be measured by the speed at which such a previously learned association can be reversed. Reversal learning is considered to indicate an individual's ability to exert executive, inhibitory control and thus be behaviourally flexible ([16,17] corvids (Gymnorhinus cyanocephalus, Nucifraga columbiana, Aphelocoma californica); but see also [18] humans). The processes have been linked to specific behaviours and fitness consequences. Associative learning performance determines adult foraging strategies ([19] sparrow, Passer domesticus) and rapid learning speeds enhance individual's foraging or reproductive success ([20,21] grasshopper, Schistocerca americana; wasp, Biosteres arisanus). Flexibility permits rapid switching between different optimal decisions in changeable environments [22] so that more behaviourally flexible individuals have improved invasion success ([23] Birds) or a better ability to track fluctuating social groups ([24] Primates). The two processes (associative learning and reversal learning) may be closely related to one another. In several other species, speeds of associative learning and reversal learning are negatively related ([25–27] myna, chickadee, scrub-jay (Aphelocoma coerulescens), red junglefowl (Gallus gallus)). However, this negative relationship is not inevitable ([13,28] bowerbird, robin (Petroica longipes)) and indeed may be positive [29] or moderated by another factor e.g. testosterone [25].

Second, previous studies have not attempted to explicitly test how performance in abstract cognitive tasks relates to specific behaviours upon which selection may act. For example, it is not clear how improved inhibitory control as revealed by performance in a detour task may relate to song-learning processes [10], or how the ability to navigate a maze manifests in improved mating success [9]. One possibility is that cognitive performances and natural behaviours are linked by an overarching personality, such that an individual's behaviour in one context (a cognitive task) is linked to their behaviour in another context [30,31]. Alternatively, a cognitive ability has an immediate link to a natural behaviour, independent of personality. For example, performance in maze learning may correspond to the methods by which an individual learns to navigate their environment and recall feeding and refuge locations. By explicitly testing how cognitive abilities relate to broader personality assays, or more specific behaviours likely to relate to fitness outcomes, we can better understand how selection may act on these abilities.

Finally, studies have either had to test wild individuals for whom prior experience, social ranking and/or age is unknown, or they have relied on laboratory systems where the putative fitness consequences are hard to relate to the natural world. Administering controlled psychometric tests to wild animals, in which a large, random and reasonably complete sample of individuals participate over a large number of repeated presentations is problematic [1–4]. One solution is to capture animals from the wild and take them into captivity where they can be tested before release back into the wild. This approach encounters two problems. First, capture may not be random [32], so that the sample tested is not representative of the wild population. Second, individuals may have undergone different prior experiences that could lead to biases or preferences (e.g. for a particular colour) developed in other contexts that skew their performance in tests [3]. Such problems may be overcome by testing captive-reared individuals where prior experiences can be controlled and participation ensured. However, captive animals are not subject to natural selective pressures because predators are excluded and resources are provided in excess, and hence robust and relevant fitness measures are difficult to collect. This may explain why previous studies have used proxy measures of fitness.

We made use of a unique study system, the common pheasant (Phasianus colchicus) (hereafter pheasants). In the UK, these birds can be reared in captivity from hatching and subsequently released into the wild (for hunting). This ensures that individuals all experience identical developmental trajectories and prior experiences, all can be tested under controlled conditions and, critically, after release can be subject to natural selective pressures in the wild, where their fates can be monitored. We reared pheasant chicks from hatching to 10 weeks under controlled conditions in 2014 and 2015, and during this time we could subject them to psychometric tests of acquisition and reversal learning [33]. We used two sets of tests of particular processes, specifically the acquisition and reversal of associations between cues and rewards, using two different task paradigms (one discriminating colours and the other discriminating spatial positions on the test apparatus), with one task paradigm used in each year, to improve our confidence that it was the cognitive process that we were measuring rather than simply response to one particular set of cues. Critically, we then released birds into the wild and followed their fates, using survival as an unambiguous indicator of their fitness. Pheasant survival may be affected by year [34,35], sex ([36,37], but see also [38]), mass ([39], but see also [40]) and interactions between them (e.g. [41]). Therefore, we considered these in conjunction with performances in the cognitive tests. Pheasant mortality is typically high, especially in early life when birds are first independent, due to both terrestrial and avian predators [41,42]. This mortality occurs when pheasants disperse from their open-topped release pens and hence encounter novel predators and move away from artificial food provision. Pheasants that leave such safe release sites and fail to learn new foraging locations or refuges from predators are likely to be highly susceptible. We asked whether survival was predicted by a pheasant's early life performance in psychometric tests of learning and reversal, controlling for other non-cognitive factors such as sex and mass. Given the ambiguous relationship previously reported between an individual's speed of acquisition and reversal [13,25–29], we tested how performances in these two tasks were related to each other in pheasants. We then explored two mechanisms by which any such relationships between cognition and fitness may be mediated by their movement and exploration. As pheasants moved further away from their point of release (in a protected and provisioned pen—see below), they would encounter higher densities of predators and lower densities of artificial food supplies, and hence face an increased risk of predation or starvation. First, we tested how an individual's exploratory behaviour in a series of assays under controlled conditions when five weeks old correlated with their cognitive ability. Second, we tested how early life cognitive performance related to adult ranging behaviour after release.
The quick are the dead: pheasants that are slow to reverse a learned association survive for longer in the wild. Joah R. Madden, Ellis J. G. Langley, Mark A. Whiteside, Christine E. Beardsworth, Jayden O. van Horik. Philosophical Transactions of the Royal Society B: Biological Sciences, September 26 2018, Volume 373, issue 1756. http://rstb.royalsocietypublishing.org/content/373/1756/20170297

Abstract: Cognitive abilities probably evolve through natural selection if they provide individuals with fitness benefits. A growing number of studies demonstrate a positive relationship between performance in psychometric tasks and (proxy) measures of fitness. We assayed the performance of 154 common pheasant (Phasianus colchicus) chicks on tests of acquisition and reversal learning, using a different set of chicks and different set of cue types (spatial location and colour) in each of two years and then followed their fates after release into the wild. Across all birds, individuals that were slow to reverse previously learned associations were more likely to survive to four months old. For heavy birds, individuals that rapidly acquired an association had improved survival to four months, whereas for light birds, slow acquirers were more likely to be alive. Slow reversers also exhibited less exploratory behaviour in assays when five weeks old. Fast acquirers visited more artificial feeders after release. In contrast to most other studies, we showed that apparently ‘poor’ cognitive performance (slow reversal speed suggesting low behavioural flexibility) correlates with fitness benefits in at least some circumstances. This correlation suggests a novel mechanism by which continued exaggeration of cognitive abilities may be constrained.

---
1. Introduction

One powerful approach to understand how natural selection may act on cognition is to measure the performance of individuals in a particular cognitive domain, and then explore how their performance correlates with a (proxy) fitness measure [1,2]. This is achieved by deploying explicit psychometric tasks targeting specific, defined cognitive processes [3,4]. Because fitness itself is hard to measure [5], researchers tend to use proxies that are presumed to correspond to reproductive success and/or survival. This correlational approach has predominantly revealed a positive relationship between an individual's performance in the psychometric task and a (proxy) measure of their fitness. Ants Lasius niger that exhibited faster route learning had greater colony-level foraging success [6]. Male African striped mice (Rhabdomys pumilio) that escaped quickly from mazes also had increased probability of surviving to the breeding season [7]. Male guppies (Poecilia reticulata) that learned mazes quickly were preferred by females [8]. Male bitterling (Rhodeus ocellatus) (practising a ‘sneaker’ strategy) that exhibited better maze learning subsequently had higher reproductive success [9]. Male song sparrows (Melospiza melodia) that demonstrated better control in a detour-reaching task had a larger song repertoire [10]. Male starlings (Sturnus vulgaris) with better spatial learning exhibited longer song bouts [11]. One study of female Australian magpies (Cracticus tibicen dorsalis) reported a link between their reproductive success and a general factor summarizing their performance in a battery of four tasks [12]. By contrast, a study of spotted bowerbirds (Ptilonorhynchus maculatus) found no relationship between a male's mating success and his performance in a battery of six tasks, either individually or when his performance was summarized by a single component [13]. Only one study has reported a negative relationship: male song sparrows that were fast at spatial learning also had smaller song repertoires [14]. This implies that natural selection generally leads to more exaggerated cognitive performance and associated abilities.

Interpretation of these previous studies is complicated by three factors. First, in all cases except one [12], a single assay has been used for each cognitive process being investigated. Reliance on a single assay risks a misattribution of the mechanisms driving individual performance. For example, learning to discriminate between two colour cues may indicate the specific ability or inherent motivation to prefer one colour over another [3], rather than the more general ability to learn associatively. A more robust method would be to use two (or more) tests that assay the same putative cognitive mechanism but differ in format or cue uses and hence triangulate on the outcome (Volter et al. [15]). We considered two ubiquitous cognitive processes and tested each using two different test variants. Associative learning involves learning to associate a stimulus with a reward and may be tested using a binary discrimination. Reversal learning may be measured by the speed at which such a previously learned association can be reversed. Reversal learning is considered to indicate an individual's ability to exert executive, inhibitory control and thus be behaviourally flexible ([16,17] corvids (Gymnorhinus cyanocephalus, Nucifraga columbiana, Aphelocoma californica); but see also [18] humans). The processes have been linked to specific behaviours and fitness consequences. Associative learning performance determines adult foraging strategies ([19] sparrow, Passer domesticus) and rapid learning speeds enhance individual's foraging or reproductive success ([20,21] grasshopper, Schistocerca americana; wasp, Biosteres arisanus). Flexibility permits rapid switching between different optimal decisions in changeable environments [22] so that more behaviourally flexible individuals have improved invasion success ([23] Birds) or a better ability to track fluctuating social groups ([24] Primates). The two processes (associative learning and reversal learning) may be closely related to one another. In several other species, speeds of associative learning and reversal learning are negatively related ([25–27] myna, chickadee, scrub-jay (Aphelocoma coerulescens), red junglefowl (Gallus gallus)). However, this negative relationship is not inevitable ([13,28] bowerbird, robin (Petroica longipes)) and indeed may be positive [29] or moderated by another factor e.g. testosterone [25].

Second, previous studies have not attempted to explicitly test how performance in abstract cognitive tasks relates to specific behaviours upon which selection may act. For example, it is not clear how improved inhibitory control as revealed by performance in a detour task may relate to song-learning processes [10], or how the ability to navigate a maze manifests in improved mating success [9]. One possibility is that cognitive performances and natural behaviours are linked by an overarching personality, such that an individual's behaviour in one context (a cognitive task) is linked to their behaviour in another context [30,31]. Alternatively, a cognitive ability has an immediate link to a natural behaviour, independent of personality. For example, performance in maze learning may correspond to the methods by which an individual learns to navigate their environment and recall feeding and refuge locations. By explicitly testing how cognitive abilities relate to broader personality assays, or more specific behaviours likely to relate to fitness outcomes, we can better understand how selection may act on these abilities.

Finally, studies have either had to test wild individuals for whom prior experience, social ranking and/or age is unknown, or they have relied on laboratory systems where the putative fitness consequences are hard to relate to the natural world. Administering controlled psychometric tests to wild animals, in which a large, random and reasonably complete sample of individuals participate over a large number of repeated presentations is problematic [1–4]. One solution is to capture animals from the wild and take them into captivity where they can be tested before release back into the wild. This approach encounters two problems. First, capture may not be random [32], so that the sample tested is not representative of the wild population. Second, individuals may have undergone different prior experiences that could lead to biases or preferences (e.g. for a particular colour) developed in other contexts that skew their performance in tests [3]. Such problems may be overcome by testing captive-reared individuals where prior experiences can be controlled and participation ensured. However, captive animals are not subject to natural selective pressures because predators are excluded and resources are provided in excess, and hence robust and relevant fitness measures are difficult to collect. This may explain why previous studies have used proxy measures of fitness.

We made use of a unique study system, the common pheasant (Phasianus colchicus) (hereafter pheasants). In the UK, these birds can be reared in captivity from hatching and subsequently released into the wild (for hunting). This ensures that individuals all experience identical developmental trajectories and prior experiences, all can be tested under controlled conditions and, critically, after release can be subject to natural selective pressures in the wild, where their fates can be monitored. We reared pheasant chicks from hatching to 10 weeks under controlled conditions in 2014 and 2015, and during this time we could subject them to psychometric tests of acquisition and reversal learning [33]. We used two sets of tests of particular processes, specifically the acquisition and reversal of associations between cues and rewards, using two different task paradigms (one discriminating colours and the other discriminating spatial positions on the test apparatus), with one task paradigm used in each year, to improve our confidence that it was the cognitive process that we were measuring rather than simply response to one particular set of cues. Critically, we then released birds into the wild and followed their fates, using survival as an unambiguous indicator of their fitness. Pheasant survival may be affected by year [34,35], sex ([36,37], but see also [38]), mass ([39], but see also [40]) and interactions between them (e.g. [41]). Therefore, we considered these in conjunction with performances in the cognitive tests. Pheasant mortality is typically high, especially in early life when birds are first independent, due to both terrestrial and avian predators [41,42]. This mortality occurs when pheasants disperse from their open-topped release pens and hence encounter novel predators and move away from artificial food provision. Pheasants that leave such safe release sites and fail to learn new foraging locations or refuges from predators are likely to be highly susceptible. We asked whether survival was predicted by a pheasant's early life performance in psychometric tests of learning and reversal, controlling for other non-cognitive factors such as sex and mass. Given the ambiguous relationship previously reported between an individual's speed of acquisition and reversal [13,25–29], we tested how performances in these two tasks were related to each other in pheasants. We then explored two mechanisms by which any such relationships between cognition and fitness may be mediated by their movement and exploration. As pheasants moved further away from their point of release (in a protected and provisioned pen—see below), they would encounter higher densities of predators and lower densities of artificial food supplies, and hence face an increased risk of predation or starvation. First, we tested how an individual's exploratory behaviour in a series of assays under controlled conditions when five weeks old correlated with their cognitive ability. Second, we tested how early life cognitive performance related to adult ranging behaviour after release.