Consistent individual differences give rise to ‘ caching syndromes ’ in a food-storing passerine

Caching food for later retrieval is vital for many animals' survival, but little is known about how this behaviour varies among individuals in the wild. If individuals consistently differ in where, when and how they store food, it could indicate that caching is subject to selection. In this study, we experimentally quanti ﬁ ed the repeatability and relationship between different aspects of caching behaviour over two winter seasons in a wild population of toutouwai, Petroica longipes . Individuals were repeatable both within and between years in the number of cache sites they created, the distance they travelled to cache and food item handling time prior to caching. All three of these caching behavioural measures were positively correlated with one another, suggesting that toutouwai exhibit a caching syndrome analogous to a behavioural syndrome, with individuals ranging between ‘ clump caching ’ and ‘ scatter caching ’ . The number of food items that birds ate prior to caching and latency to begin cache retrieval were also consistently correlated, indicating a second ‘ fast - slow ’ syndrome linking cache retrieval to satiation level. In both syndromes most individuals were normally distributed between the behavioural extremes, indicating that birds tended to partake in intermediate caching rather than clustering at either end of the continuum. We posit that this distribution may be the result of stabilizing selection that balances the costs and bene ﬁ ts of each extreme. © 2024 The Authors. Published by Elsevier Ltd on behalf of The Association for the Study of Animal Behaviour. This is an open access article under the CC BY license (http://creativecommons.org/licenses/ by/4.0/).

Many birds engage in the storage of food in caches for later retrieval, which allows them to survive times of shortage and manage their energy budget (McNamara et al., 1990;Pravosudov, 2006).As a vital survival behaviour, caching is thought to be subject to selection (Smith & Reichman, 1984), but few studies have quantified individual differences in food storage in the wild.Of particular interest is whether individuals repeatably vary in the spatiotemporal aspects of food storage, which could suggest a heritable component to caching (Dall et al., 2012;Wauters et al., 1995).Additionally, if individuals vary consistently across different aspects of the caching process, it could suggest the existence of a caching-related behavioural syndrome resulting from different choices made during caching (Bell, 2007;Dochtermann & Jenkins, 2007;Zwolak & Sih, 2020).Addressing the nature of individual differences in food storage necessitates the development of methods for quantifying and conceptualizing how individuals vary in their caching behaviour.
Investigations into the individual consistency of caching are complicated by the logistical difficulties involved in quantifying the behaviour in sufficient detail to reveal repeatable individual differences.Directly observing caching in the field is difficult in species where caching occurs over large spatiotemporal scales, or in species that are wary towards human observers (Ekman et al., 1996;Morand-Ferron et al., 2016).As a result, only a limited portion of caching behaviour is quantifiable in most study systems.For instance, one study found individual differences in the rate at which mountain chickadees, Poecile gambeli, carried away seeds from artificial feeders, presumably to cache them (Croston et al., 2016).In a mammalian caching species, Merriam's kangaroo rat, Dipodomys merriami, individuals varied between scatter hoarding seeds throughout their territories and larder hoarding seeds in their dens (Dochtermann & Jenkins, 2007).In Canada jays, Perisoreus canadensis, individual birds were given artificial sources of food to cache, with the finding that individuals varied their caching behaviour depending on the location of previously made caches (Waite & Reeve, 1992a).To determine whether such behavioural variation is truly reflective of individual preferences, it is important not only to measure behaviour across the entire caching process (e.g.Brodin, 1994), but also to test whether individuals' behaviour is temporally repeatable (Carter et al., 2013).A recent study on mountain chickadees found that caching rate was significantly repeatable across years (Sonnenberg et al., 2022).Additionally, Class et al. (2021) found that the number and mass of cached prey items was significantly repeatable in Eurasian pygmy owls, Glaucidium passerinum.These two studies represent the only attempts that have been made to assess multiyear repeatability in any aspect of food-storing behaviour.
Caching is a multifaceted behaviour that presents individuals with numerous decisions: from the initial choice to cache food, to the selection of a cache site, to the timing of retrieval (Smith & Reichman, 1984).Individuals can approach the various decisions involved in caching in different ways, raising the question of whether there is a single, ideal way of caching in a species, or whether multiple strategies may exist (Wolf et al., 2007).If individuals vary consistently in how they make decisions throughout the caching process, it could indicate the existence of a 'caching syndrome' analogous to a behavioural syndrome.Behavioural syndromes describe suites of behaviours that consistently correlate with one another across time and/or context, resulting in a continuum between behavioural types (Sih et al., 2012).In the context of caching, correlative relationships between independent components of the caching process (e.g. the number of sites used and the distance that cached items are moved relative to where they were acquired) may indicate the presence of multiple 'caching types.'These behavioural phenotypes need not be bimodally distributed: the defining characteristic of a syndrome is the correlative relationship between multiple independent behaviours (Sih & Bell, 2008).If a caching syndrome does exist, it could indicate that different individuals have different ways of addressing the challenges associated with caching, resulting in a reservoir of behavioural diversity in the population that can buffer against changes in the selective pressures acting on caching birds.
The North Island robin, Petroica longipes, henceforth referred to by their M aori name toutouwai, provides an ideal system for establishing individual repeatability in caching and investigating the existence of a caching-specific behavioural syndrome.The toutouwai, a small passerine endemic to New Zealand, engages in short-term caching behaviour: invertebrate prey items (or their dismembered parts) are stored in cache sites throughout the bird's small territory, typically in tree axles, knots, or in the fissures of branches (Powlesland, 1980).Retrieval of these caches occurs within several days, but often within a single hour (Burns, 2009).Having evolved without mammalian predators, toutouwai are naïve and will approach humans, and readily cache provided food items such as mealworms (Alexander et al., 2005).Previous research has taken advantage of this naivety to examine aspects of cache pilferage, numerical cognition and spatial memory (Armstrong et al., 2012;Shaw et al., 2019;Van Horik & Burns, 2007).Toutouwai are thus an ideal model system for empirically investigating repeatable individual differences in caching behaviour at a fine scale.
In this study, we quantified individual differences in toutouwai caching, with a focus on determining repeatability over short and long timescales and assessing the relationship between different components of the behaviour.We experimentally provided wild toutouwai with mealworms, standardized the overall caching effort across individuals by allowing each bird to cache a set number of food items on their winter territories, and then waited until retrieval commenced.We measured the number of food items eaten prior to caching, the number of sites food items were distributed between, the distance the bird travelled to cache each item, the length of time spent deciding where to store, and the length of time before retrieval of these sites began.We ran multiple sessions per bird and examined the temporal repeatability of these caching behaviours both within each year and across a 2-year period.We then examined whether the patterns in the birds' decision making were consistent with the existence of one or more caching syndromes.

Study Subjects and Site
The study subjects were wild, free-living toutouwai residing in Zealandia Ecosanctuary in Wellington, New Zealand, which is a 225 ha wildlife sanctuary surrounded by fencing designed to exclude introduced mammalian predators.The toutouwai inhabit a mosaic of regenerating native forest and introduced pine forest, with breeding pairs in the study site defending small territories of less than 1 ha (Shaw et al., 2019).Since 2014, birds residing within our 25 ha study area have been given individually identifiable combinations of coloured leg bands and have been closely monitored (Shaw et al., 2015(Shaw et al., , 2019)).As a result, each bird's territory was known, and birds could be tested in central locations where conflicts with neighbours could be avoided.All subjects were already habituated to humans and accustomed to approaching experimenters for mealworm food rewards (Shaw et al., 2015(Shaw et al., , 2019)).

Ethical Note
The experiment was approved by the Victoria University of Wellington Animal Ethics Committee (ID: 0000028395) and carried out under permit from the Department of Conservation (Authorization number: 56493-FAU).The study subjects consisted of 69 wild birds of which 37 were male and 32 were female.The birds were members of a long-term study population monitored since 2014.As free-living animals tested in their own established territories in the wild, the birds were not forced to take part in the experiment and could cease participation at any time.

Experimental Procedure
The experiment took place during two consecutive austral winters (2020e2021).In 2020, we tested 51 (N males ¼ 28, N females ¼ 23) toutouwai between 0900 and 1530.Each bird was given one session per day on 3 consecutive testing days.Occasionally, factors such as adverse weather meant that a bird could not be tested on consecutive days; in such cases the next session was given on the next possible day.Each session occurred at approximately the same time of day (±3 h) for each bird.Prior to the start of a session, we attracted the target bird to a central location within its territory, where we placed a feeding stage in an open site with no overhanging vegetation <1 m above it.The feeding stage consisted of a rectangular wooden block (13.5 Â 9 cm and 2.5 cm high).The location of the feeding stage was shifted 5 m in a random direction between each session to determine whether birds would reuse sites across days regardless of starting position.The session commenced only if the target bird was alone, with no additional conspecifics in the immediate vicinity.
A session consisted of two phases: a caching phase and a retrieval phase.To begin the caching phase, the experimenter placed one freshly killed mealworm (Tenebrio molitor larvae) in the centre of the feeding stage before stepping back approximately 2 m to an observation position and allowing the bird to approach.As the position of the experimenter potentially affected where birds chose to cache, it was kept constant within each session.If the bird consumed the mealworm, the experimenter immediately approached and replaced it.The sequential presentation of single mealworms continued until the bird began to cache.Once caching began, the experimenter video recorded the bird's movements using an iPhone 5 and noted the location of the bird's chosen cache site.Immediately after the bird stored a mealworm and left the cache site, the next mealworm was placed on the stage.The caching phase continued until the bird cached five mealworms in total.On the rare occasions (40 instances, or 4.9% of the total number of mealworms cached) that we were unable to determine the precise location of a bird's cache site, this was noted but did not count towards the requisite five caches; more mealworms were provided as appropriate.Upon the fifth stored mealworm, no further mealworms were provided, ending the caching phase.Video examples of toutouwai caching are available in the Supplementary material.
The retrieval phase began immediately following the termination of the caching phase and consisted of a 30 min period during which we recorded any cache retrieval by the bird.During this time the experimenter remained as still as possible, moving only to ensure the bird remained in sight, and avoided approaching active cache sites.This length of time has previously been used to assess cache retrieval in toutouwai (Burns & Van Horik, 2007).If the bird retrieved a mealworm, the time and site retrieved from were recorded.At the end of 30 min, we measured the distance and compass bearing of each site from the feeding location, as well as the height of each cache site above ground level.After measurements were taken, the bird was weighed by allowing it to hop onto an electronic scale to retrieve an additional mealworm, a method employed by past toutouwai studies (Shaw et al., 2015(Shaw et al., , 2019)).Body condition could then be calculated by weight divided by tarsus length, a metric previously used to assess body condition in toutouwai (Shaw, 2017).
To examine whether birds demonstrated long-term consistency in their caching behaviour, we repeated the study procedure in AprileJuly 2021 on 44 (N males ¼ 23, N females ¼ 21) toutouwai, of which 26 (N males ¼ 14, N females ¼ 12) had also participated in the previous year.The data collection was largely identical to 2020 procedures.However, due to time constraints in the winter testing period, our caching observations were carried out in tandem with an experiment measuring inhibitory control (McCallum & Shaw, 2023).Thus, rather than taking the mealworm directly from the feeding stage, birds detoured around a clear or opaque plastic barrier to acquire it.In addition, the feeding stage location remained constant across all three sessions, rather than shifting 5 m between sessions.As each bird received 20 detour-reaching trials per day, it was possible for more or fewer than five caches to be made during a detour-reaching session.In sessions where a bird finished a detour-reaching session before caching five mealworms, the bird was provided with mealworms sequentially until five were cached in total.In sessions during which a bird made more than five caches during the detour-reaching task, only the first five instances of caching were used in analyses.Thus, for the analyses each bird had 15 trials across three sessions in both 2020 and 2021.

Behavioural Measures
We took five behavioural measures that we considered representative of the entirety of the caching process.While all five of these behaviours were components of the overall caching process, they were specifically chosen to avoid measures that were autocorrelated (e.g.we did not include height, as greater heights in the context of our measures were by definition inherently correlated with longer distances).Only characteristics of mealworm caches made by the bird during the caching session were analysed; wildcaught food items cached during the retrieval phase were not included in analyses, even if these were retrieved within 30 min.In the very rare instances (six times, or 0.4% of caches) where a bird recached a previously stored mealworm to a different site during the caching phase, we used the original site's measurements in our analyses.

Number of mealworms eaten
In 2021 the number of mealworms eaten during the caching phase was taken as a measure of an individual's propensity to engage in caching versus direct consumption of food.Typically, birds consumed several mealworms before the onset of caching, after which all subsequent mealworms were cached, but occasionally mealworms were eaten between instances of caching, and these were included in the total.The number of mealworms eaten during the caching session was not included in analyses in 2021, as the detour-reaching task trials varied greatly in duration, meaning that birds did not receive mealworms at a steady rate as they did in 2020.

Number of sites
The number of sites that a bird created within a session to store the five mealworms was taken as a measure of propensity to scatter or clump cached food items.Items were considered to be in the same site if they were all stored in the same topographical feature (e.g. in the same crevice, crack or branch angle).

Distance travelled to cache
Distance (m) was measured as a straight line between the feeding stage and the cache site, using a tape measurer for distances <5 m and a Bushnell rangefinder for distances >5 m.We used distance as a measure of effort expended in finding a cache site as well as wariness: birds that travelled further were considered choosier in selecting a cache site, and possibly more cautious of potential pilferers, based on previous research linking the presence of conspecific pilferers with the creation of more distant cache sites (Burns & Steer, 2006;Van Horik & Burns, 2007).

Handling time
We used BORIS behavioural coding software (Friard & Gamba, 2016) to code the handling time (s) for each mealworm from the videos taken during the sessions.Handling time was defined as the interval between a bird's first contact with a mealworm and its final placement in a cache site.We excluded flight time from this calculation to avoid confounding distance and handling time, as birds that cached further away were assumed to fly further to do so.Thus, handling time primarily consists of time that the bird spent perching while holding the mealworm, alongside any small hops made by the bird along branches/the ground.Birds that took longer to decide on a cache site were considered more selective.During the second year of the study (2021) the experimenter provided narration during filming to verbally confirm deposition of the food item into the cache site, as videos from the first year were sometimes difficult to decipher due to poor resolution.

Retrieval latency
The number of minutes before first retrieval was used as a measure of the bird's choice to deplete its caches versus continuing to expend energy foraging for novel food items.Birds that did not retrieve within 30 min were assigned a retrieval latency of 31.Due to time constraints retrieval latency was not measured in 2021.

Statistical Analyses
All analyses were conducted using R version 4.1.3(R Core Team, 2022) with the use of the packages lme4 (Bates et al., 2015) and rptR (Stoffel et al., 2017).

Repeatability of measures and sex/age/body condition effects
We measured the repeatability of each of the above behavioural measures across the three sessions within a year using the R package rptR, which provides repeatability estimates using mixedeffects models with individual identity as a random effect (Stoffel et al., 2017).The repeatability estimates for the number of mealworms eaten during a session, as well as the number of sites created, were calculated with a generalized linear mixed model (GLMM) using a Poisson error distribution.Repeatability of distance travelled during each instance of caching (m), handling time of each cached item (s), and retrieval latency (min) for each session were calculated using linear mixed models (LMMs).We also quantified interyear repeatability between 2020 and 2021 in the number of cache sites used per session, the distance travelled to cache each food item and the handling time of each food item prior to caching for the 26 birds that participated in both years; in these models, study year was included alongside individual identity as a random effect.
To examine whether sex, age, body condition or time of day influenced caching behaviour, we constructed LMMs and GLMMs for each of the five behavioural measures.Distance travelled to cache each item (m), handling time of each item (s) and retrieval latency (min) were modelled using LMMs, while mealworms eaten and number of cache sites per session were modelled using a GLMM with a Poisson error distribution.In each model, the behavioural measure was the response variable, with age, sex, body condition and time of day included as fixed factors and individual bird as a random factor.Year (2020 or 2021) was included as an additional fixed factor for the three measures that were recorded in both years.

Caching syndromes
We constructed a Spearman rank correlation matrix to test the correlations between each of the five caching behavioural measures taken in 2020 and between each of the three behavioural measures taken in 2021.As the raw data for these measures consisted of repeated measures that could not be used in a single principal component analysis (PCA) and given the significant within-year individual repeatability of all caching measures (see Results for details), a single mean value of each behavioural measure across all sessions within a year was calculated for each individual.The 2020 data included all five caching measures, while the 2021 data omitted the two variables that were not measured in that year: mealworms eaten and retrieval latency (min).To avoid the increased possibility of type II errors, we did not apply post hoc tests to the P values (Armstrong, 2014).We then used a PCA with an unrotated factor solution to summarize associations between these variables within each year.
PCA has previously been used to quantify behavioural syndromes: when a suite of behaviours loads onto a single principal component (PC), that component can be interpreted as representing the variance due to a behavioural syndrome (Dingemanse et al., 2007;Found & St. Clair, 2016;H€ ojesj€ o et al., 2011).As the caching measures we quantified describe different aspects of the same behaviour, we interpreted significant loadings of these variables onto the same PC as capturing variation from an underlying 'caching syndrome'.Following the Kaiser -Guttman criterion, we retained only those PCs with eigenvalues greater than 1.We then interpreted the presence/absence of a caching syndrome by examining the PC loadings of each behavioural measure.We considered a factor loading of >0.5 to be relevant (Budaev, 2010).By extracting individual birds' scores for relevant PCs, we obtained a 'caching score' for each bird: a measure of where along the axis of the PC the individual lay (Found & St. Clair, 2016).Within each year we then tested for possible sex, age or body condition effects on the extracted scores of each PC using a linear model (LM) with PC score as the response variable.

RESULTS
On average, birds consumed 6.8 ± 0.2 (mean ± SE) mealworms during a caching session, mealworms were handled for 14.9 ± 0.2 s and cached 5.2 ± 0.1 m from the feeding location.Birds used a mean of 2.5 ± 0.1 sites per session and frequently reused sites across days.There was a mean latency of 15.0 ± 0.7 min prior to first retrieval from a cache site.Retrieval did not occur within 30 min in 15 of 153 sessions (9.8%).All caching measures were significantly repeatable across sessions within both years of the study (Table 1).In the first year of the study, number of mealworms eaten, number of sites used, distance travelled to cache and handling time had low to moderate repeatability, while retrieval latency exhibited moderate repeatability across sessions.Across-session repeatability was similar during the second year of the study, with number of sites, distance travelled to cache and handling time all demonstrating low to moderate levels of repeatability.There was significant low to moderate repeatability across years in distance travelled, number of sites used and handling time (Table 1).Taken together these results suggest that individual birds remained consistent in their behaviour in both the short term (across sessions) and long term (between years).
There was no evidence that sex or body condition influenced any caching measure, while age only influenced retrieval latency: younger birds tended to delay retrieval longer than older birds (LMM retrieval latency coefficient estimate (CE) ¼ À0.803, 95% confidence intervals (CI) ¼ À1.438 to À0.168).The time of day that sessions took place influenced the number of mealworms eaten during the caching phase prior to creating five caches: birds ate more mealworms during sessions taking place later in the day (GLMM number of mealworms eaten CE ¼ 0.099, 95% CI ¼ 0.007 to 0.192).Handling time was significantly lower in 2021 than 2020 (LMM handling time CE ¼ À1.697, 95% CI ¼ À3.015 to À0.358).This latter finding is most likely explained by the increased quality of video recording in the second year of the study, which allowed for more precise measurement during video coding.For full model results see Appendix Table A1.
Tables 2 and 3 show the correlations between caching measures in both years of the study.There were significant positive correlations between number of sites, distance travelled (m) to cache at each site and handling time (s) in both years of the study.There was also a significant positive correlation between the number of mealworms eaten and retrieval latency (min) during the one year when these two variables were measured; neither of these variables correlated significantly with any others.Table 4 shows the results of the PCA for all five cache measures taken in 2020, as well as results for the smaller three-variable PCA run on the data collected in 2021.PCA results from the first year revealed that number of sites, distance travelled to cache (m) and handling time (s) loaded onto the same PC (all loadings >0.5), PC1, which accounted for 39.5% of variation.PC1 appears to describe a behavioural axis governing cache site selection: birds that created more cache sites were more likely to travel further away to cache and spent more time handling food items before placing them ('scatter caching').Conversely, birds that created fewer cache sites tended to travel less far to cache in them and to spend less time handling items before caching ('clump caching').The three variables that loaded strongly onto PC1 in 2020 were once again examined in 2021, with similar results to the first year: all three variables loaded strongly onto PC1, which explained 63.6% of variance, indicating that the scattering -clumping syndrome was present in both years.In 2020, the number of mealworms eaten and retrieval latency loaded strongly onto PC2, which accounted for an additional 30.7% of variance.PC2 describes behavioural variation whereby birds that took longer to initiate caching were more likely to delay cache retrieval for longer ('slow' caching), while birds that began caching quickly were more likely to retrieve sooner ('fast' caching).
In all cases, extracted PC scores formed a continuum between the extremes of each correlative relationship, with most birds falling close to zero (Fig. 1).However, extracted individual PC1 scores during the first year of the study were not normally distributed (Shapiro -Wilk test: W ¼ 0.930, P ¼ 0.005), with a significant negative skew from outlying low-scoring (scattering) birds (Fig. 1a).This may be the result of the differences in test protocol between years: the shift of the feeding stage between each session in 2020 may have inflated the degree to which birds scattered their caches.PC2 scores in the first year of the study (Shapiro -Wilk test: W ¼ 0.984, P ¼ 0.717) and PC1 scores in the second year of the study (Shapiro -Wilk test: W ¼ 0.985, P ¼ 0.843) were normally distributed (Fig. 1b and c).Extracted individual PC1 scores were not related to sex, age or body condition in 2020, but there was a significant sex difference in 2021, with males engaging in more clumping than females (LM 2021 PC1 score CE ¼ 1.080, 95% CI ¼ 0.143 to 2.017).Meanwhile, PC2 scores (calculated for 2020 only) were independent of sex and body condition but were influenced by age cohort (LM PC2 score CE ¼ À0.122, 95% CI ¼ À0.242 to À0.003).Older birds tended to score lower, eating fewer mealworms prior to caching and retrieving sooner than younger birds.(See Appendix Table A2 for full model results.)

DISCUSSION
Toutouwai displayed significant repeatability and individual variation in their caching behaviour.We found moderate repeatability in several of our caching measures both within and between years, indicating that birds were temporally consistent in the decisions they made throughout the caching process.Our findings support the results of previous studies that have quantified repeatable individual variation in food-storing behaviour.Class et al. (2021) found levels of repeatability similar to ours in the contents of Eurasian pygmy owl cache sites.Likewise, a laboratory experiment on captive Merriam's kangaroo rats and Ord's kangaroo rats, Dipodomys ordii, found repeatability in the spatial distribution of cached seeds (Jenkins, 2011).These findings indicate that repeatable individual variation in caching behaviour could be ubiquitous across food-storing taxa.Unlike the above studies, however, we measured variation not just in the contents and distribution of cache sites, but in the behaviour of the caching individual.Many previous studies have quantified food-storing behaviour via direct observation in the wild (e.g.Grubb & Woodrey, 1990;Haftorn, 1954;Haftorn, 1956aHaftorn, , 1956bHaftorn, , 1956c;;Petit et al., 1989;Waite & Reeve, 1992a, 1992b); here we add to this literature with the first measures of repeatability across the entire caching process in the wild, from food acquisition to retrieval.Our approach of providing a standardized number of mealworms allowed us to avoid confounds associated with prey encounter rate and variable prey size that occur in purely observational studies (e.g.Brodin, 1994Brodin, , 2005)).Our findings suggest that variation in caching behaviour cannot be wholly ascribed to changes in short-term environmental factors such as weather or time of day.Instead, they suggest that individual differences may be due to heritable genetic variation that may be subject to selection, although further research is needed to confirm a link with fitness proxies in toutouwai (Dall et al., 2012).One factor potentially influencing caching that we cannot explicitly rule out in our study is the possibility that territory structure influenced cache site availability.However, given the similarity in most birds' territories (a mixture of regenerating native forest and introduced pines) we believe it unlikely that this could solely explain the observed variation.Our finding that caching measures were repeatable in the first study year, even when the feeding stage was shifted across sessions, also suggests that territory structure is not a major confound.
The consistent correlation of the distance travelled to cache, cache site number and handling time across individuals indicated the existence of a caching syndrome in which individuals fell between the two extremes of 'scatter caching' (creating many sites, travelling longer distances, with longer handling times) and 'clump caching' (creating few sites, travelling short distances, with shorter handling times).The correlation of these independent components of food storage resulted in a behavioural axis analogous to a behavioural syndrome (Sih et al., 2004).Individuals were not bimodally distributed into discrete scatter-caching and clumpcaching categories: instead, an intermediate mode of caching prevailed.Such a normally distributed caching syndrome may suggest that food storing in toutouwai is characterized by stabilizing selection that prevents scatter-cachers and clump-cachers from diverging (Han & Brooks, 2013).
The predominance of an intermediate strategy may indicate the deleterious nature of committing to either behavioural extreme.A bird that engages in a pure clumping strategy may take on much greater risk of pilferage, as the loss of one site results in the loss of a much larger proportion of their total hoarded food amount (Clarke & Kramer, 1994).Conversely, a purely scatter-caching bird may avoid a high cost of pilferage at the cost of spending more time and energy in the creation and retrieval of many distant caches (Waite & Reeve, 1995).Thus, we hypothesize that the ideal caching strategy for a toutouwai appears to involve engaging in enough scatter caching to lessen the impact of pilferage, but not so much that the benefits of caching are negated by the time and effort expended in retrieval.We did not explicitly quantify pilfering in this study: we tested birds with no conspecifics present and the experimenter's presence may have discouraged heterospecific pilferage during the retrieval sessions, but previous research has confirmed that toutouwai caches are pilfered by both conspecific and heterospecific birds (Steer, 2006;Van Horik & Burns, 2007).Linking instances of pilferage with variation in caching is a natural next step for future research to take.
Our scatter-clumping caching syndrome differs from true behavioural syndromes in that it describes variation in independent (i.e.nonautocorrelated) measures within the context of a single behaviour, rather than consisting of measures taken from separate assays of behaviour (Sih et al., 2004).Our findings do not rule out the possibility that other 'classical' behavioural syndromes such as boldness or exploration may influence a bird's caching (Sih, 2011;Thompson & Morand-Ferron, 2019;Zwolak & Sih, 2020).Toutouwai are known to aggressively defend their caches from pilferers (Burns, 2009;Steer, 2006), raising the possibility that boldness or aggression could be linked to caching: birds may be more likely to risk storing more items in clumped caches if they are better able to defend these (Dally et al., 2006;Vander Wall et al., 2005).As male toutouwai are known to be dominant to females (Burns & Steer, 2006), this may account for the tendency of males to engage in clump caching in the second year of the study, although it is unclear why this effect was not observed in the first year.However, research on kangaroo rats found that propensity to scatter caches was positively associated with boldness, potentially because of bolder individuals' greater tolerance of predation risk when making numerous caches (Dochtermann & Jenkins, 2007;Steele et al., 2016).The pressures driving individual variation in caching animals may thus vary greatly across taxa, making generalization across species difficult.
There was an additional repeatable positive correlation between the number of items eaten prior to caching and the onset of retrieval behaviour, which pointed to early retrieval in less satiated birds.This relationship, along with the short latency to first retrieval (<30 min for nearly all birds), backs up previous research on the closely related kakaruwai, Petroica australis, in which caching serves more as a method for balancing the birds' daily energy budget, and less as a stockpiling technique for seasonal food shortages (Barnett, 2001;Barnett & Emura, 2014;Pravosudov & Grubb, 1997).Toutouwai caching may primarily serve as a shortterm insurance against food stress for a bird whose primary food source, large invertebrates such as tree w et a (Hemideina spp.), is often patchily distributed (Kelly, 2006).Our finding that birds ate more food items prior to caching in sessions that took place later in the day reinforces previous research showing that as night approaches, birds should switch to consuming food directly rather than caching it, in anticipation of the energetic demands of surviving the night (Pravosudov & Grubb, 1997).The repeatability of this relationship across days suggests that there may be inherent individual differences in the level of satiation that 'trigger' caching behaviour, making this 'fast -slow' continuum a second syndrome apparently unrelated to the scattering -clumping syndrome detailed above.The tendency of younger birds towards what we term 'fast' caching (eating more and delaying retrieval, as opposed to 'slow' caching birds that did the opposite) may be the result of first-year birds not yet developing adult-level caching behaviour, which previous research has shown takes more than 12 weeks in toutouwai (Clark & Shaw, 2018).As in the scatter-clumping syndrome, individuals were normally distributed between pure 'fast' and 'slow' caching, indicating a similar balance between either end of the behavioural axis.
Our findings constitute a uniquely detailed quantification of individual variation in caching behaviour.However, much remains to be discovered about the causes and consequences of this variation.Investigating the links between individual variation in caching behaviour and pilferage could help establish whether the latter is a selective pressure driving evolution of the former.The extent to which the individual differences reported here are heritable also remains an important question to address.Likewise, comparisons of individual caching type with proxies of fitness would allow for the identification of a possible role for stabilizing selection in maintaining an ideal caching type.Finally, toutouwai are known to vary consistently in their spatial memory capabilities (Shaw et al., 2019).As spatial memory is essential to cache retrieval in many avian food-storing species (Bednekoff et al., 1997;Bennett, 1993;Clayton & Krebs, 1994), establishing a link between individual differences in spatial cognition and individual variation in food storage behaviour is an exciting avenue for further research in toutouwai.

Figure 1 .
Figure 1.Histograms showing the distribution of individual 'caching scores' in PC1 (scattering -clumping syndrome) in (a) 2020 and (b) 2021.Negative PC1 scores indicate greater 'scattering' propensity, while positive scores indicate greater 'clumping' propensity.(c) PC2 (fast -slow syndrome) from 2020.Negative PC2 scores indicate 'slow' birds that ate more mealworms before caching and waited longer to retrieve, while positive scores indicate 'fast' birds that ate fewer mealworms before caching and retrieved sooner.

Table 1
Repeatability estimates of caching variables Mean R is given with SE and the P value.N/A: not applicable.T. I. F. V amos, R. C. Shaw / Animal Behaviour 211 (2024) 43e51

Table 2
Spearman correlation matrix of the relationships between mean caching measures in 2020