Evidence for Environment‐Specific Pace‐of‐Life Syndromes

TL;DR

The study shows that how behavior relates to life history in mice depends on food quality, with different patterns emerging in high versus low resource environments.

Contribution

The study provides empirical evidence that pace-of-life syndromes are context-dependent, particularly influenced by resource quality.

Findings

Under low food quality, more exploratory mice showed a faster pace-of-life, aligning with a risk–mortality trade-off.

In high food quality, more active mice delayed reproduction, following a slower pace-of-life linked to asset protection.

Pace-of-life syndromes are context-dependent, with environmental factors like food quality strongly influencing behavioral and life-history correlations.

Abstract

The pace‐of‐life syndrome (POLS) hypothesis posits that consistent individual differences in behaviour are integrated with physiology and life‐history traits such that behaviour mediates how individuals resolve life‐history trade‐offs. For instance, individuals exhibiting higher exploration tendencies may accelerate reproduction by gaining access to resources more quickly, but this same behaviour could reduce survival through increased risks of predation and competition. While empirical support for POLS remains mixed, recent theory emphasises the role of environmental context in resolving some inconsistencies. Resource quality, in particular, may strongly mediate context‐dependent effects, yet its functional role has received little empirical attention. To address this, we monitored the complete life‐histories of 344 female house mice (
Mus musculus domesticus
) across four semi‐natural enclosures running in parallel, provisioned with either high or standard‐quality food. We first assessed how resource quality influenced life‐history traits and then repeatedly measured behaviour to investigate the among‐individual correlations between behaviour and life‐history within each food quality treatment. Two axes captured most of the variation in life‐history in both food quality treatments, with the primary axis reflecting a fast–slow continuum. The relationship between behaviour and life‐history was context‐dependent at the among‐individual level: under a lower quality treatment, more exploratory females exhibited a faster pace‐of‐life, consistent with a risk–mortality trade‐off. By contrast, in higher quality food conditions, individuals that covered more distance in an open‐field, that is, more active stress‐copers, delayed reproduction and followed a slower pace‐of‐life, suggesting a POLS that incorporates aspects of asset protection. Our results indicate that pace‐of‐life syndromes are context‐dependent, emerging most clearly when behavioural variation interacts with environmental factors that affect some aspect of fitness. More broadly, we provide evidence that POLS vary profoundly in different ecological conditions, highlighting the importance of considering environmental context when testing fundamental links between behaviour and life‐history.

The correlation between behaviour and life‐history depends on environmental conditions. We show this is true when considering the quality of the food in the environment.

Figures

Sample sizes and levels of analysis in this study.

Among‐individual correlations and their 95% credible intervals between three behavioural traits and two life‐history principal components (PC1 and PC2), estimated separately for the standard (SQ) and high‐quality food treatment (HQ). PC1 is an axis describing the pace‐of‐life from fast to slow, and PC2 describes the reproductive decisions of individuals contrasting large litters with smaller litters and the number of successful litters produced (see Section 3.2). The estimates are presented on the correlation scale (r). A star (*) indicates a correlation where the 95% credible interval does not overlap with zero.

Scatterplots showing the relationship between individual behaviour and pace‐of‐life across food‐quality treatments: (a) distance covered in the open‐field (log‐transformed Gaussian errors); (b) number of exploration trips (Poisson errors). Each point represents one individual measurement from the raw dataset. Linear trends (as drawn from the multivariate models: Solid lines) and 95% CI are shown separately for each food‐quality treatment, displayed in separate panels (high‐quality enclosure versus standard‐quality enclosure). Significance is indicated above each panel (* = 95% CI does not contain 0; ns = CI does contain 0).

Among‐individual variance in behaviour of female mice under different food quality treatments. Left: Variance in the total distance covered in the open‐field; Right: Variance in the number of exploration trips. Bars show posterior mean estimates of among‐individual variance, and error bars indicate 95% credible intervals.

Tables

TABLE 1. All the models we used in our analyses.

Model	Response(s)	Fixed and random‐effects structure
Model 1: Repeatability (Gaussian)	Log (distance covered in the OF a )	Trial + (1 | enclosure ID) + (1|female ID)
Model 2: Repeatability (Gaussian)	Sqrt (time in centre of the OF)	Trial + (1 | enclosure ID) + (1|female ID)
Model 3: Repeatability (Poisson)	Number of trips	Trial + (1 | enclosure ID) + (1|female ID)
Model 4: Life‐history*behaviour (Gaussian + Gaussian)	PC1 + Log (distance covered in the OF)	1 + (1| enclosure ID) + (1| female ID, by = treatment)
Model 5: Life‐history*behaviour (Gaussian + Gaussian)	PC2 + Log (distance covered in the OF)	1 + (1| enclosure ID) + (1| female ID, by = treatment)
Model 6: Life‐history*behaviour b (Gaussian + Gaussian)	PC1 + Sqrt (time in centre of the OF)	1 + (1| enclosure ID) + (female ID, by = treatment)
Model 7: Life‐history*behaviour b (Gaussian + Gaussian)	PC2 + Sqrt (time in centre of the OF)	1 + (1| enclosure ID) + (female ID, by = treatment)
Model 8: Life‐history*behaviour b (Gaussian + Poisson)	PC1 + number of trips in the NE c	1 + (1| enclosure ID) + (female ID, by = treatment)
Model 9: Life‐history*behaviour b (Gaussian + Poisson)	PC2 + number of trips in the NE	1 + (1| enclosure ID) + (female ID, by = treatment)

TABLE 2. Loadings of life‐history traits on the first two principal components (PC1 and PC2) under high‐quality and standard‐quality food treatments.

Variable	PC1 (high‐quality food)	PC2 (high‐quality food)	PC1 (standard‐quality food)	PC2 (standard‐quality food)
Litter size	+0.054	−0.441	+0.036	+0.281
Number of litters	+0.041	+0.773	+0.153	−0.731
Lifetime growth (grams)	−0.464	−0.177	−0.540	+0.272
Survival in months	+0.660	+0.196	+0.676	−0.027
Onset of reproduction	+0.588	−0.373	+0.477	+0.558

Full text

Introduction

1

Across species, animals exhibit substantial variation in growth, survival, reproduction and lifespan, that is in life‐history traits (Healy et al. 2014, 2019; Roff 1992; Stearns 1992). Classical life‐history theory explains much of this variation through trade‐offs in energy acquisition and allocation, such as how resources are divided among growth, reproduction and maintenance (Stearns 1989; van Noordwijk and de Jong 1986). A major, though not exclusive, axis of variation is the pace‐of‐life: some species and populations grow rapidly, reproduce early and die young, whereas others grow more slowly, reproduce later and live longer (Bielby et al. 2007; Healy et al. 2014, 2019; Salguero‐Gómez et al. 2016).

Individuals within populations also differ in their life‐histories and in behavioural tendencies, that is, have distinct personalities (Dingemanse and Réale 2005, 2013) in terms of risk‐taking, exploration, sociability or boldness. However, the mechanisms maintaining such behavioural variation remain unclear (Laskowski et al. 2022). Consequently, the pace‐of‐life syndrome (POLS) hypothesis has been proposed to explain consistent individual differences in both behaviour and life‐history traits (Dall et al. 2004; Haave‐Audet et al. 2022; Réale et al. 2010). Specifically, the POLS proposes that suites of phenotypic (life‐history, physiological and behavioural) traits covary in a fast‐slow axis due to differences in energy allocation between individuals that invest in a current versus a future reproduction (Dammhahn et al. 2018). Under the POLS, “fast” individuals are expected to combine life‐history traits such as rapid growth, short lifespans, early and frequent reproduction and high offspring numbers with behavioural traits such as high boldness and exploration. Conversely, “slow” individuals should combine slower life‐history traits such as delayed reproduction, lower reproduction rates and longer lifespans with cautious, risk‐averse behaviours coupled with lower metabolic levels (Dammhahn et al. 2018; Réale et al. 2010). The behavioural traits in these correlations have been suggested to covary with life‐history trade‐offs through correlative selection or the effects of genetic and/or hormonal pleiotropy (Réale et al. 2010; Stamps 2007; Wolf et al. 2007).

In general, the predicted correlations between behaviour and life‐history have been hard to detect (Dammhahn et al. 2018; Montiglio et al. 2018; Royauté et al. 2018). Because of that, mostly theoretical, work has suggested that behaviour–life‐history correlations remain elusive unless one distinguishes variation in resource acquisition from variation in resource allocation (Dammhahn et al. 2018; Laskowski et al. 2021; Montiglio et al. 2018; Royauté et al. 2018). Recent meta‐analyses support this view, either by showing no negative among‐individual phenotypic correlations of life‐history traits or that positive among‐individual correlations arise when acquisition varies most, and vice versa (Chang et al. 2024; Haave‐Audet et al. 2022; Moiron et al. 2020). This acquisition–allocation trade‐off is well‐established in classic life‐history theory (Y‐model): individuals that acquire more resources have more energy available to invest across competing traits and when trait expression depends more on total acquisition than on allocation phenotypic correlations between traits are expected to be positive (van Noordwijk and de Jong 1986). More recently, this framework has been incorporated to the POLS literature to explain the mixed empirical support for behaviour–life‐history correlations at the among‐individual level (Laskowski et al. 2021). Specifically, the predicted correlation between behaviour and life‐history may depend on environmental variation, which can create differences in how individuals acquire resources (Haave‐Audet et al. 2022; Laskowski et al. 2021; Smallegange and Guenther 2024). Empirical studies have explored how such environmental heterogeneity can mask the true structure of pace‐of‐life, while also providing evidence for rapid behavioural adjustments under novel or rapidly changing conditions.

Reviews (Haave‐Audet et al. 2022; Hämäläinen et al. 2021; Laskowski et al. 2021; Montiglio et al. 2018) and limited evidence indeed suggest that factors related to environmental variation, such as predator abundance (Cote et al. 2013; Dhellemmes et al. 2021; Lapiedra et al. 2018), food‐quality (Prabh et al. 2023), temperature (Debecker and Stoks 2019) and resource availability (Bergeron et al. 2013; Haines et al. 2020) can all mediate the covariance between behaviour and life‐history traits. For example, activity in an open‐field assay is positively correlated with risk‐taking and growth but negatively with survival in shark subpopulations with lower predation (Dhellemmes et al. 2021); female exploration and survival decreases in the presence of predators compared to predator‐free environments in lizards (Lapiedra et al. 2018); dispersal is personality‐dependent but only in the absence of predators in fish (Cote et al. 2013); and, at the population‐level, mice feeding on a lower food‐quality are more risk‐prone but reproduce more slowly than mice having access to richer (i.e., in calories) food (Prabh et al. 2023). Three important considerations have emerged from these: first, support for core predictions of the POLS, even when accounting for aspects of variation in some environmental factors (e.g., weather or food availability), is equivocal (Bergeron et al. 2013; Haines et al. 2020), underscoring the need for more studies at the among‐individual level; second, most of the aforementioned studies are partly constrained by the inherent difficulties of obtaining repeated behavioural and life‐history measurements in the wild, therefore usually focus on a section of individuals' lifetime (Dhellemmes et al. 2021; Lapiedra et al. 2018); or at population‐level behavioural differences (Prabh et al. 2023); third, the role of food quality—not just the amount of available resources—has received comparatively little attention, even though resource quality is expected to shape both life‐history and behavioural traits by influencing the energy individuals can acquire and allocate.

Here, we studied 344 female house mice under semi‐natural conditions in four replicate populations that differed only in food quality (see Methods). First, we quantified each individual's pace‐of‐life (POL) as a composite of life‐history traits, including litter size, number of litters, growth rate and survival. We then tested whether the POL was associated with consistent behavioural differences within each food‐quality treatment (high versus standard‐quality food). We predicted that high‐quality food could reduce the phenotypic expression of trade‐offs as the net energy available per gram is higher than in standard‐quality food, potentially weakening behavioural–life‐history correlations. Conversely, standard‐quality food may intensify the detection of these trade‐offs, leading to stronger correlations and a greater adaptive value of behaviour.

Methods

2

Animals and Enclosures

2.1

House mice are ecological opportunists that thrive and reproduce in variable environments (Bronson 1979; Ganem 2012) and show consistent behavioural differences across time and contexts (Krebs et al. 2019; Küçüktaş and Guenther 2022). We focused on females that lived in semi‐natural enclosures (19.6 m^2^; Figure S1) designed to mimic their natural habitat, such as fields or barns (König and Lindholm 2012; Laurie 1946), and monitored their life‐history traits from birth to death (birth to month 1 = lactation period; month 1 to month 9 = potentially reproductively active period; see Methods Section 2.2) during monthly censuses.

Founding mice of these enclosures originated from 16 pairs out of wild populations of the Cologne/Bonn region of Germany (CB N = 18 original breeding pairs) (50°45′ N–51° N, 6°45′ E–7° E) and were distributed to the four replicate enclosures with a starting generation of 20 males and 20 females per enclosure. During monthly censuses, we caught all individuals within an enclosure (all RFID‐chipped), measured their weight and took a tissue sample to assign parentage and assess reproductive events (i.e., litters, litter size, lifetime fitness) via microsatellite analysis. Young pups heavier than 10 g received an RFID PIT tag (Planet ID, 1.4 × 9 mm). As such, the whole population was known, allowing us to infer survival because we repeated this process monthly. The experiment ran for approximately 3 years. Importantly, whenever a new generation reached at least 140 individuals per enclosure old/heavy enough to receive a PIT tag, we removed the older generation to prevent negative effects of social crowding. Usually, this happened every 6–10 months per enclosure and population: under standard‐quality food the enclosures exceeded the threshold on 8% of censuses while under high‐quality food on 8.2%, showing that the removal procedure was consistent across food quality treatments.

We quantified the life histories of 344 females living in four semi‐natural enclosures. These females were those that produced at least one litter with at least one pup that survived long enough to be PIT‐tagged and genotyped (≥ 10 g at a census). Consequently, our dataset excludes females for which no offspring survived to weaning, reducing bias in our analyses. These females could represent slower strategies with delayed or infrequent reproduction, but equally could reflect faster strategies associated with high early reproductive investment followed by offspring loss. In addition, it is well established that most female mice become pregnant within the first couple of months of their lives even though many of them may never be observed to raise pups. Authors generally interpret this as females losing foetuses early during pregnancy (Christian and Lemunyan 1958).

Two of the four enclosures (1 and 2) always received standard‐quality food (Altromin 1324, SQ), while the other two (3 and 4) always received high‐quality food (Altromin 1414, HQ). The standard‐quality food provided 3227 kcal/kg metabolizable energy (24% protein, 11% fat, 65% carbohydrates), whereas the high‐quality food provided 3680 kcal/kg (28% protein, 22% fat, 50% carbohydrates). Thus, the main difference between the two food treatments was the amount of energy available per gram of food. Importantly, previous research has indicated the importance of the food quality for reproduction and behaviour in mice (Bronson 1979; Lopez‐Hervas et al. 2024), revealing that animals under a standard‐quality food need to eat more throughout 24 h even in a non‐reproductive state (Prabh et al. 2023). Within each enclosure, food and water were equally distributed across nine stations and replenished daily to be ad libitum, to mimic natural house mice populations (in barns or fields) where food is abundant (König and Lindholm 2012; Laurie 1946). Each enclosure had natural daylight and temperature fluctuations, with underfloor heating keeping a minimum of 10°C since mice were not able to dig burrows to escape cold temperatures. Enclosures were covered by a roof, inaccessible to predators and designed to resemble the natural conditions under which house mice would establish a colony, like a barn.

For all analyses, we pooled observations across enclosures within each food‐quality treatment (i.e., enclosures 1 and 2 as one group, and enclosures 3 and 4 as the other; Figure 1) to increase sample size and statistical power. This approach aligns with our primary goal of testing how the covariance between behaviour and life‐history varies with food quality and is justified because the only systematic difference between the enclosures is food quality. Importantly, to account for among‐enclosure variability, we included semi‐natural enclosure ID (i.e., 1, 2, 3 or 4) as a random effect in all models (see Statistical Analyses).

Sample sizes and levels of analysis in this study.

Life‐History Traits

2.2

All life‐history traits were analysed until death from natural causes (high‐quality treatment: mean = 5.5 and range = 1–9 months; standard‐quality treatment: mean = 6.3 and range = 1–12 months). While females possibly do not exceed 14 months of age in barns (König and Lindholm 2012), the mean life expectancy in feral populations (e.g., Skokholm) is ~3–4 months (Bellamy et al. 1973); and tooth‐wear, which indicates relative age across mammals, doubles after ~8–10 months on wild caught mice (Morgan and Bellamy 1975). Crucially, in a previous paper that used three generations of house mice, only ~7% (19/286) of females reproduced after 7 months of age (Prabh et al. 2023).

For each female we measured: (1) the onset of reproduction as the month a female conceived successfully for the first time (high‐quality treatment: mean = 3.5 and range = 1–8 months; standard‐quality treatment: mean = 4 and range = 1–8 months). This was inferred from the microsatellite analysis (see below) of each female's first litter: because pups are chipped at ~50 days of age, we back‐calculated the approximate conception date by subtracting 2 months from the chipping date and assigned as “onset of reproduction” the census occasion closest to that month; (2) number of litters, determined via microsatellite parentage assignments, with pups sharing the same mother and chipping date considered a single litter; (3) mean litter size (litters with ≥ 1 pup); (4) growth rate, calculated as the difference between first and last recorded mass divided by the number of intervals between measurements; (5) survival, quantified as the number of monthly censuses a female was recorded alive (e.g., a female found for 8 censuses before death = 8).

Parentage was determined with seventeen microsatellite markers using the procedure adapted from Linnenbrink et al. (2013). In short, DNA from the ear clips was amplified using Multiplex PCR kits (QIAGEN), samples were run on an ABI 3730 Sequencer (Applied Biosystems). GeneMarker (V2.6.4) was used to call alleles and Colony [COLONY | Zoological Society of London] to assign the parentages based on the maximum likelihood of each potential parent pair.

Behavioural Traits

2.3

We assessed aspects of risk‐taking using one voluntary exploration test (novel environment) and one forced exploration test (open‐field). In total, we collected 162 novel‐environment and 176 open‐field observations. Because sampling effort was uneven across individuals, each behavioural variable was analysed in its own model, allowing us to account for individual‐level variation and estimate repeatability for each trait. Second trapping sessions were conducted 4 weeks after the first, an interval consistent with previous studies in house mice and other rodents (Erixon et al. 2024, 2025; Lopez‐Hervas et al. 2024; Prabh et al. 2023), spanning a substantial portion of individuals' lifespans. This design allowed us to capture both within‐ and among‐individual behavioural variation over time.

For the novel environment, 81 individuals were measured once for “trips” while 40 were measured twice. The number of trips was defined as the number of exploration bouts in a novel environment. Recording of this voluntary exploration behaviour was directly conducted within semi‐natural enclosures by inserting the mice caught in live‐traps into a transparent Macrolon cage (Techniplast) (40.5 × 28.0 × 50.0 cm) close to the point of capture of each mouse. The cage contained three objects unknown to the mice. The number of exploration trips (NE trips) was recorded under red light for 10 min (Panasonic Full HD camera 10 MP HC‐V 180).

For the open‐field, females were released in the middle of a 60 × 60 cm arena for 5 min and the total distance covered and the percentage of time spent 10 cm away from the wall was measured using the software VideoMotion2 (TSE). 64 individuals were measured once and 56 twice for the variables “distance covered” and “time in the centre” Trapping started earliest at sunset (i.e., at different times depending on the season) and lasted about 4 h. Crucially, the open‐field has been validated as ecologically credible in our populations (Krebs et al. 2019) and dates almost a century (Hall and Ballachey 1932) for analyses related to rodent behaviour: the distance covered reflects individual exploratory tendencies and active versus passive stress‐coping behaviour, while the time spent close to the walls (as opposed to the centre) is a measure of risk‐taking and is associated with stress levels (Gould et al. 2009; Krebs et al. 2019; Lopez‐Hervas et al. 2024). We aimed to catch and test N = 30 animals (1st round) and at least N = 20 animals for a repeated trial (2nd round) per enclosure, which usually lasted three to four trapping nights in a row. Traps were checked every few minutes and animals that were not immediately used for testing were released.

Statistical Analyses

2.4

R 4.2.1 (2022‐06‐23, R Foundation for Statistical Computing, Vienna, Austria; https://www.R‐project.org/) was used in all analyses.

We first run two standard unrotated Principal Component Analyses (PCA; prcomp function of the stats package) on all life‐history traits per female (growth rate, the onset of reproduction, the number of litters, mean litter size and survival), separately for each food‐quality treatment to account for treatment‐specific life‐history variation (Prabh et al. 2023). Each PCA reduced the dimensionality of correlated life‐history traits into orthogonal axes, summarising key population‐level biological patterns. This approach is common (Stott et al. 2024; de Van Walle et al. 2023) and is also used when exploring the correlations among different behaviours (Erixon et al. 2024). All variables were standardised (mean = 0, SD = 1) prior to analysis to ensure comparability. After fitting the PCAs, we extracted the major individual PC metrics that summarised variance across individuals (Table 2, Figure S2).

As a first step towards testing pace‐of‐life syndrome (POLS) theory—which assumes that individuals differ consistently in behavioural tendencies—we assessed the repeatability R of each behavioural trait (distance in the open‐field, time in the centre of the open‐field, number of trips in a novel environment) fitting three Bayesian mixed effects models in the brms package (Bürkner 2017). Each model included one behavioural measure as the response, trial number as a fixed effect and semi‐natural enclosure ID and female ID as random effects. The distance covered in the open‐field was log‐transformed and the time spent at the centre of it was square‐root transformed, both to achieve approximately Gaussian residuals. The number of exploration trips were modelled using a Poisson error distribution. All variance components were obtained from the posterior draws and the distribution of R was summarised using the posterior median and 95% credible intervals. To ensure that including individuals with only a single behavioural measurement did not bias the repeatability estimates, we conducted a sensitivity analysis restricted to females with at least two observations for each trait. Repeatability estimates, 95% credible intervals and posterior distributions were virtually unchanged, confirming that single measurements do not introduce measurement noise.

Then, we quantified the among‐individual covariance between each behavioural trait with the two primary life‐history axes (PC1 and PC2) using six multivariate Bayesian mixed‐effects models in brms, depending on the error distribution of behavioural and PC data (Table 1; models 4–9): (1) Gaussian distance in the open‐field (log‐transformed) and Gaussian PC1; (2) Gaussian distance in the open‐field (log‐transformed) and Gaussian PC2; (3) Gaussian time in the centre of the open‐field (square‐root transformed) and Gaussian PC1; (4) Gaussian time in the centre of the open‐field (square‐root transformed) and Gaussian PC2; (5) Poisson number of trips in a novel environment and Gaussian PC1; (6) Poisson number of trips in a novel environment and Gaussian PC2. For PC1 and PC2, because they are only expressed once per female and thus have no within‐individual variation, we fixed their residual variance to one following other studies (Houslay and Wilson 2017; Moiron et al. 2020). Behavioural traits, which were repeatedly measured, retained free residual variances with weakly informative priors. In all models, each behavioural variable was further standardised prior to fitting to facilitate convergence and improve numerical stability. Semi‐natural enclosure ID and individual female ID were fitted as random effects in all six models to account for among‐enclosure and among‐individual variance. Trial number (in behavioural measurements) was omitted because the repeatability models (see above) indicated that it explained negligible variation in behaviour. Importantly, and to test our main prediction that pace‐of‐life syndromes will vary as a function of food quality treatment, we allowed the among‐individual variances and covariance components to vary within each food quality treatment using gr (ID, by = treatment).

Validity of Bayesian models was assessed visually (caterpillar plots of the posterior distribution), statistically using the Gelman‐Rubin diagnostic (Vats and Knudson 2021), and by calculating the variance inflation factor (no predictors were highly correlated). The models were run for at least 100,000 iterations (depending on convergence issues) for 2 chains and had a burn‐in of 1000.

Results

3

For the high‐quality treatment, the PCA yielded eigenvalues of 1.74, 1.29, 0.93, 0.78, and 0.25 for the five axes, corresponding to 34.8%, 25.9%, 18.6%, 15.7%, and 5.0% of the total variance explained, respectively. In the standard‐quality treatment, eigenvalues were 1.86, 1.20, 1.05, 0.63, and 0.26, explaining 37.2%, 24.0%, 21.0%, 12.7%, and 5.2% of the variance. Variable loadings were comparable across treatments (Table 2) and allowed us to compare individuals along a common fast–slow life‐history axis (PC1) and simultaneously confirm treatment‐specific differences in subsequent components (Prabh et al. 2023). Importantly, for our analyses, we focus on the first two axes per treatment, which explained most of the variance.

PCA in the High‐Quality Food Treatment

3.1

According to the PCA, the first principal component (PC1, 34.8% variance explained) represented a trade‐off between survival, the onset of reproduction and growth rate (Table 2, Figure S2a): individuals scoring positively on PC1 delayed first reproduction, survived longer and grew more slowly, whereas individuals scoring negatively grew faster, initiated reproduction earlier but died younger. Thus, PC1 reflected a trade‐off between growth and early reproduction versus maintenance and longevity. The second component (PC2, 25.9% of variance explained) primarily captured reproductive patterns, contrasting individuals producing a higher number of successful litters with those producing fewer, larger litters. In the high‐quality treatment, PC2 thus reflected a distinct life‐history trade‐off: females investing in larger litters emphasised current reproduction whereas those producing more litters across their lifespan emphasised future reproduction.

PCA in the Standard‐Quality Food Treatment

3.2

A comparable, though not identical, pattern emerged for mice in the standard‐quality treatment (Table 2, Figure S2b). PC1 (37.2% of variance) captured the primary life‐history axis, with longer‐lived, later‐reproducing individuals scoring positively and faster‐growing individuals scoring negatively, reflecting the same pace‐of‐life trade‐offs as in the high‐quality treatment. In contrast, PC2 (24.1% of variance) reflected the trade‐off between the onset of reproduction and a number of successful litters. This indicates that under lower‐quality resources the timing of first reproduction is more tightly linked to lifetime reproductive output than under high‐quality conditions: females that reproduced earlier produced more litters, suggesting that fitness differences are more pronounced when resources are more limiting.

Repeatability Analyses

3.3

For all Bayesian models, we interpret a result as biologically meaningful (i.e., similar to statistically significant) if the 95% credible intervals did not overlap 0 (Table S1).

All behaviours were significantly repeatable across individuals and were therefore retained for downstream analyses, although the magnitude of repeatability varied among traits. The distance moved in the open‐field exhibited moderate repeatability (R = 0.52; 95% CI: 0.29–0.69), whereas the time spent in the centre of the open‐field showed the lowest estimate (R = 0.14; 95% CI: 0.001–0.37). The number of exploration trips in a novel environment was highly repeatable (R = 0.84; 95% CI: 0.77–0.90).

Pace‐of‐Life Syndromes

3.4

All results are summarised in the Supporting Information.

We found food quality‐specific among‐individual correlations between behaviour and life‐history (Figure 2).

Among‐individual correlations and their 95% credible intervals between three behavioural traits and two life‐history principal components (PC1 and PC2), estimated separately for the standard (SQ) and high‐quality food treatment (HQ). PC1 is an axis describing the pace‐of‐life from fast to slow, and PC2 describes the reproductive decisions of individuals contrasting large litters with smaller litters and the number of successful litters produced (see Section 3.2). The estimates are presented on the correlation scale (r). A star () indicates a correlation where the 95% credible interval does not overlap with zero.*

In high‐quality enclosures, the distance moved in the open‐field was strongly positively correlated with the pace‐of‐life (PC1: r = 0.61, 95% CI: 0.14–0.96; Figure 3a), so that more active females in the open‐field tended to postpone reproduction relative to less active ones. Under the standard‐quality diet, however, no credible association between open‐field activity and PC1 emerged. Importantly, the among‐individual variance was similar under high and standard‐quality food (0.77 > 0.61, respectively; Figure 4).

Scatterplots showing the relationship between individual behaviour and pace‐of‐life across food‐quality treatments: (a) distance covered in the open‐field (log‐transformed Gaussian errors); (b) number of exploration trips (Poisson errors). Each point represents one individual measurement from the raw dataset. Linear trends (as drawn from the multivariate models: Solid lines) and 95% CI are shown separately for each food‐quality treatment, displayed in separate panels (high‐quality enclosure versus standard‐quality enclosure). Significance is indicated above each panel ( = 95% CI does not contain 0; ns = CI does contain 0).*

Among‐individual variance in behaviour of female mice under different food quality treatments. Left: Variance in the total distance covered in the open‐field; Right: Variance in the number of exploration trips. Bars show posterior mean estimates of among‐individual variance, and error bars indicate 95% credible intervals.

Conversely, the number of exploration trips showed a qualitatively different pattern. Under standard‐quality food, more exploratory individuals exhibited a faster pace‐of‐life (r = −0.61, 95% CI: −0.97 to −0.03; Figure 3a), an effect absent under high‐quality food (Figure 3b). Again, the among‐individual variances did not differ between treatments (Figure 4).

All other among‐individual correlations involving time spent in the centre of the open‐field, and between behaviour and the second life‐history axis (PC2, representing reproductive decisions) were weak as their 95% CIs overlapped zero. Importantly, the among‐individual variance in the time spent at the centre of the open‐field was again equivalent across food‐quality treatments (Figure 4).

Discussion

4

Here, we investigated how behaviour and life‐history covary among individuals in two different food‐quality environments (i.e., in the amount of energetic resource provision for individuals), testing principal assumptions of the POLS hypothesis (Dammhahn et al. 2018; Montiglio et al. 2018; Réale et al. 2010). Our approach is timely because environmental variation might ultimately influence the strength and direction of behaviour–life‐history correlations at the among‐individual level (Haave‐Audet et al. 2022; Hämäläinen et al. 2021; Laskowski et al. 2021; Polverino et al. 2018). Indeed, we detected and report food quality‐specific POLS.

We found that the correlation between an individual's pace‐of‐life and behaviour depended on food quality. Under a standard‐quality food regime, females that were more explorative exhibited a faster pace‐of‐life—they reproduced earlier, grew faster, but survived less time. In contrast, under high‐quality food, which is comparatively of significantly higher nutritional quality (Prabh et al. 2023), this relationship was absent, while more active stress‐copers showed instead a slower pace‐of‐life. Our findings highlight that behaviour–life‐history correlations are not static, but instead depend on environmental context. When resources are abundant or environmental variation is overlooked, the expected correlations can be reversed, weakened, or obscured, potentially explaining why pace‐of‐life syndromes are often difficult to detect (Royauté et al. 2018; Smallegange and Guenther 2024).

Life‐History Variation

4.1

Our results summarising the full suite of life‐history traits aligned with core predictions of life‐history theory (McNamara and Houston 1996; Stearns 1989, 1992): we detected an axis (PC1) characterised by an early onset of reproduction and fast growth that was traded off against survival; and a reproductive trade‐off between offspring quantity, quality and possibly, ultimately fitness (PC2). As such, although our dataset included only females with at least one surviving offspring, the observed patterns are consistent with classical life‐history predictions, supporting the validity and generality of our results.

We interpret PC1 as a fast–slow continuum, the prominence of which here (explained > 30% of variation in life‐history in both food quality treatments) aligns with studies demonstrating positive correlations between growth and maturation (Guenther 2018), the onset of reproduction and other life‐history traits (Fay et al. 2016), as well as with comparative analyses across animal (Healy et al. 2014, 2019; de Van Walle et al. 2023) and plant species (Salguero‐Gómez et al. 2016). As predicted by the POLS (Réale et al. 2010), we found here that females that reproduced faster died earlier, which in our case occurred even in the absence of predation pressure. This suggests that the increased mortality might be well‐embedded (in our system or more generally) within life‐history if: (1) the onset of reproduction is under (negative) correlational selection with longevity; (2) investing in early reproduction trades off with investment into maintenance, leading to cellular and organismal damage accumulation “disposable soma hypothesis” (Kirkwood 1977), as some studies have found (Hammers et al. 2013; Hayward et al. 2014; Nussey et al. 2006); (3) the high physical, physiological and metabolic activity needed for accessing and processing food resources to feedback a fast reproduction may elevate oxidative stress, accelerating senescence (Soulsbury and Halsey 2018); (4) the increasing resource demands of individuals that are both still growing and reproductively active may heighten competition with conspecifics, which fast individuals may not be able to successfully cope with over time.

A divergence in life‐history traits among food quality treatments was observed for PC2, which quantified reproductive decisions/patterns. Under high‐quality food, females with an earlier reproductive onset produced more but smaller litters, whereas those reproducing later produced fewer, larger litters. This pattern reflects a trade‐off between current and future reproduction: investing in smaller litters allows resources to be conserved for future reproductive events and vice versa. In the standard‐quality food, this trade‐off was much weaker: PC2 primarily captured the link between a fast onset of reproduction and the number of litters produced during the lifetime. This contrast highlights that environmental variation can mask phenotypic trade‐offs when it disproportionately affects one trait over another (Stearns 1989). Our results mean that high‐quality food boosts energy intake, allowing females to invest in both reproduction and growth simultaneously. Consequently, the trade‐off of current versus future reproduction is alleviated at the phenotypic level since all females have enough resources to invest heavily into current and future reproduction. In standard‐quality food, however, energy allocation and acquisition constraints probably emerge and limit reproductive success (van Noordwijk and de Jong 1986) even though food per se is not limited.

POLS and Environmental Variation

4.2

In a previous study, we showed that food quality influences the average behavioural expression, with individuals receiving lower‐quality food exhibiting a more active stress‐coping phenotype (Prabh et al. 2023). Consistent with these findings, we show here that the structure of among‐individual covariation between behaviour and life‐history also depends on diet quality. Specifically, we show a context‐dependency using a resource‐quality manipulation experiment. Our findings join a growing number of empirical studies suggesting that associations between behaviour and life‐history traits (i.e., pace‐of‐life syndromes) are not universal but strongly context‐dependent. For instance, in mosquitofish, individuals from a slow‐growing population are on average bolder and more active—despite identical laboratory conditions (Polverino et al. 2018). In lemon sharks, individuals from low‐risk, low‐predation populations (i.e., of “higher quality”, as our population receiving high‐quality food) that showed higher exploratory behaviour also achieved higher growth rates, a fast life‐history outcome (Dhellemmes et al. 2021).

The distance covered in the open‐field, a measure of stress‐coping behaviour in rodents (Gould et al. 2009; Lopez‐Hervas et al. 2024) was correlated with a slow pace‐of‐life (PC1) under a high but not a standard‐quality food treatment. In environments that provide a highly nutritious food source, all individuals can satisfy their basic energetic requirements. As such, females that are less active stress‐copers channel their resources directly into reproduction—perhaps because they are less willing to take risks in respect to competition or finding resources (e.g., shelters/territories). In contrast, more active stress‐copers forego immediate reproduction and a fast growth in favour of survival—a behavioural response that aligns with the asset protection principle (Monaghan 2007; Moran et al. 2021). This might be because they experience the negative effects of mating competition more strongly. Females receiving high‐quality food are, on average, more passive stress‐copers than females receiving standard‐quality food, yet they still develop a faster pace‐of‐life compared to females on a lower‐quality diet (Prabh et al. 2023), indicating evolution towards an asset protection behavioural phenotype. Similarly, within high‐quality food, individuals that are more passive stress‐copers reproduce faster, reflecting a pace‐of‐life syndrome (POLS) linking behaviour and life‐history with asset protection at the among‐individual level. Importantly, though, females that reproduce later do not achieve higher fitness as quantified by PC2 because fast reproducers produce many small litters—a dynamic consistent with classic trade‐offs between offspring quantity and quality (Charnov and Ernest 2006; Lack 1947; Roff 1992). In other words, in high‐quality food environments, there was no evidence for positive life‐history correlations that would justify among‐individual variation in resource acquisition (Laskowski et al. 2021; van Noordwijk and de Jong 1986).

A contrasting pattern of covariance between exploration and individual pace‐of‐life emerged in the (comparatively lower quality) standard‐quality food. There, fast‐paced females were more exploratory but traded their fast growth and reproduction for reduced survival, aligning with previous reports (Adriaenssens and Johnsson 2011; Réale et al. 2010; Stamps 2007; Wolf et al. 2007) and core predictions of the original POLS formulation (Réale et al. 2010). Individuals under standard‐quality food may need to invest extra time and effort in foraging to obtain sufficient energy before they can reproduce. Consequently, those that make more trips to the food source might be the ones capable of meeting the energetic demands of rapid reproduction. Alternatively, the elevated exploration tendencies, combined with their potentially higher aggressiveness that form a common syndrome across species (Erixon et al. 2024; Thys et al. 2017), may confer an advantage under nutritionally limited, low‐quality conditions as their early reproductive onset allows them to contribute more to growing populations (Wright et al. 2019). In other words, exploration becomes the behaviour most tightly linked to life‐history variation only when energy is limiting, highlighting that the functional relevance of a behavioural trait for fitness can depend on ecological context (Laskowski et al. 2021).

Comparatively, a recent meta‐analysis (Moran et al. 2021) indicated that poorer nutritional circumstances promote elevated risk‐taking and much evidence supports the view that individuals born in worse conditions are generally more risk‐prone (Lopez‐Hervas et al. 2025). Similar to our results, fast‐exploring rabbits (which are risk‐prone during the subadult stage) survive less well (Rödel et al. 2015) and activity (which is possibly positively correlated with risk‐taking) in squirrels is negatively correlated with long‐term survival (Boon et al. 2008). At the among‐population level, personality‐mediated life‐history trade‐offs are context‐dependent, from insects (Debecker and Stoks 2019) to sharks (Dhellemmes et al. 2021) and mammals (Prabh et al. 2023). A growth‐mortality trade‐off mediated through risk‐taking is linked to predation in sharks (Dhellemmes et al. 2021) and a similar trade‐off among mouse populations is mediated through activity differences linked to food quality (Prabh et al. 2023). Also, differences in the pace‐of‐life among two damselfly populations raised under different temperatures (Ischnura elegans) have been found; and larvae that grew faster were more active and explorative within populations (Debecker and Stoks 2019). Our study extends these findings indicating that the core predictions of the POLS are detectable (at the among‐individual level) only within populations experiencing some constraints. Here, such constraints are not linked to food availability but rather food quality. Overall, the variation in ecological conditions among our populations, driven by resource quality, clearly shaped the among‐individual payoffs of personality variation, thereby modulating whether activity/boldness aligns with fast or slow life histories.

In addition, we observed that among‐individual variance in activity in the open‐field was higher under high food quality, consistent with the idea that abundant resources can amplify individual differences by relaxing energetic constraints. Under standard‐quality food, in contrast, energetic limitations compress individual variation in some traits while selectively enhance variation in others that affect resource acquisition (here exploration). This further supports the view that ecological context shapes not only mean trait expression (Dhellemmes et al. 2021; Prabh et al. 2023) but also patterns of individual differentiation and integration across the phenotype–life history trait space. Importantly, exploration and stress‐coping did not consistently influence all life‐history dimensions, as indicated by the non‐significant correlation with reproductive output (PC2); and the time spent at the centre of the open‐field was decoupled from individual life‐history. This means that not all behaviours influence all life‐history dimensions (Dhellemmes et al. 2021) or that exploration might, for example, primarily relate to variation in onset of reproduction (PC1), or the independent component of PC2 (after accounting for PC1) may not covary with exploration.

Our Findings in the Context of Current POLS Theory

4.3

Two core ideas explain the functional importance of behaviour in mediating individual variation in life‐history: that behaviour is linked to differential allocation of resources among life‐history traits (Réale et al. 2010; Wolf et al. 2007) or that behaviour mediates differential resource acquisition (Haave‐Audet et al. 2022; Laskowski et al. 2021; Moiron et al. 2020). Since the latter predicts a positive correlation between life‐history traits—e.g., reproduction and survival (Haave‐Audet et al. 2022; Laskowski et al. 2021; Moiron et al. 2020)—, as indicated in classical life‐history theory (van Noordwijk and de Jong 1986), we report evidence for a trade‐off of current versus future reproduction under standard‐quality but not under higher quality food conditions. Under such lower conditions, true differences in quality among individuals might emerge because behaviour directly affects the primary factor limiting fitness, namely energy acquisition for investment into reproductive output (Haave‐Audet et al. 2022; Laskowski et al. 2021). Indeed, standard‐quality populations produce, on average, fewer and smaller litters than those receiving high‐quality food (Prabh et al. 2023). This shows that food quality directly affects the absolute amount of energy individuals can acquire, which in turn constrains how they allocate resources to life‐history traits. Consequently, resource quality alters the phenotypic expression of life‐history trade‐offs, reducing the strength of observable correlations among traits and thereby making underlying allocation conflicts less apparent in an environment‐specific manner, commonly referred to as the “big house, big car effect” (van Noordwijk and de Jong 1986).

A recent meta‐analysis (Moiron et al. 2020)—but see (Smith and Blumstein 2008)—has concluded that risk‐taking is not linked to decreased survival. Here, however, we found that more exploratory females experienced a mortality cost. It might be that ad libitum food and the absence of mortality due to predation in our enclosures have affected this axis of life‐history—but note that house mice are found in areas where food is abundant and practically sit on food (Bronson 1979; König and Lindholm 2012; Laurie 1946). Alternatively, current findings about survival may in some cases be confounded by the unaccounted fluctuating environmental variation that the studied populations face. In any case, Moiron et al. (2020) reported that behaviour explained only some proportion of variance in survival, an indication that unaccounted factors might shift life‐history–behaviour correlations. Similarly, Haave‐Audet et al. (2022) found a positive, but statistically non‐significant, correlation between survival and reproduction.

Conclusion

5

In summary, we tested the POLS hypothesis by studying the full spectrum of individual life‐history traits in replicated populations living in under different food quality conditions. For the first time, we demonstrate that differences in food quality create different covariance structures between life‐history and behaviour. Our results align with some original suggestions of the POLS (Réale et al. 2010) but only when environmental heterogeneity is modelled (Dammhahn et al. 2018; Haave‐Audet et al. 2022; Hämäläinen et al. 2021; Jablonszky et al. 2018; Laskowski et al. 2021; Royauté et al. 2018); and especially only in relatively worse conditions. Thus, we offer a possible explanation for the equivocal support of the POLS hypothesis in the literature (Dammhahn et al. 2018; Laskowski et al. 2021, 2022; Montiglio et al. 2018; Royauté et al. 2018). Our results demonstrate that the environment modulates trait covariation rather than enforcing a fixed POLS pattern. Mechanistically, changes in the environment should alter the underlying structure of variance and covariance among traits because individuals differ in both their baseline trait values and their plastic responses to environmental conditions. As such, we predict that in lower quality environments POLS that integrate risk‐taking and exploration will emerge because acquisition differences heavily influence fitness. In higher quality environments, though, where acquisition differences are relatively unimportant due to the elevated quality, POLS integrating aspects of stress‐coping will emerge, allowing individuals to minimise the negative effects of social crowding. More generally, how behaviour mediates life‐history traits as a function environmental heterogeneity should be interpreted relative to the specific limiting factors that most strongly influence fitness.

Author Contributions

Fragkiskos Darmis: conceptualization (equal), data curation (equal), formal analysis (equal), investigation (equal), methodology (equal), software (equal), validation (equal), visualization (equal), writing – original draft (lead), writing – review and editing (equal). Anja Guenther: conceptualization (equal), funding acquisition (lead), investigation (equal), project administration (lead), supervision (lead), writing – review and editing (equal).

Ethics Statement

The animals were handled and experimental procedures carried out according ASAB/ABS guidelines. Keeping of animals was approved by the “Veterinäramt Plön” under permit: 1401‐144/PLÖ‐004697. Experimental procedures were approved by the “Ministerium für Landwirtschaft, ländliche Räume, Europa und Verbraucherschutz, Referat IX 55 Tierschutz” under licence: V244‐31223/(2019(62‐5/19)). No animal was euthanized for this study.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: Supporting Information.