Pace-of-life syndrome: linking personality, metabolism and colour ornamentation in male guppies

Within populations, there commonly exists consistent among-individual differences in behaviour. As a potential mechanism maintaining this variation, the pace-of-life-syndrome (POLS) hypothesis posits that consistent differences in the behaviour of individuals (i.e. ‘ personality ’ ) are intercorrelated and have coevolved with consistent individual differences in metabolism and life-history traits. Here, using adult male guppies, Poecilia reticulata , we tested under laboratory conditions (1) whether behavioural and metabolic traits vary consistently among individuals over time ( (cid:1) 7 days) and (2) the POLS prediction that repeatable behavioural traits should be intercorrelated with each other and with repeatable metabolic rate at the among-individual level. Furthermore, based on indicator models of sexual selection, we expected that sexually selected male colour ornamentation would predict individual personality and metabolic rate. We repeatedly assayed three behavioural traits (sociability, boldness, exploration) and three metabolic traits (resting metabolic rate, maximal metabolic rate, aerobic scope) in a set of males that varied in body size and colour ornamentation. All behavioural trait measures were repeatable, consistent with individual personality. Of the three metabolic traits, only resting metabolic rate (RMR) was repeatable. Behavioural trait measures were signi ﬁ cantly intercorrelated and thus integrated in a behavioural syndrome. More ‘ proactive ’ males were bolder, more exploratory and more sociable than ‘ reactive ’ ones. However, contrary to the POLS hypothesis, RMR was not signi ﬁ cantly correlated with any of the behavioural trait measures, suggesting that consistent individual differences in RMR do not drive or support differences in personality, or vice versa. Male colour ornamentation did not covary with behaviour or RMR and thus does not appear to be a reliable predictor of either behavioural or metabolic phenotypes. We therefore did not ﬁ nd any compelling support for the POLS hypothesis, suggesting that individual differences in metabolism do not underlie the evolution and maintenance of among- individual behavioural variation in our study population.

One proposed explanatory functional hypothesis of particular interest here is the extended pace-of-life syndrome (POLS) hypothesis (sensu R eale, Garant, et al., 2010). This hypothesis posits that variation in repeatable behaviour among individuals within populations should covary or integrate with repeatable amongindividual differences in physiology and life-history traits along a slow-to-fast pace-of-life continuum that arises ostensibly due to trade-offs in investments in current versus future reproduction (Auer et al., 2018;Biro & Stamps, 2008Careau et al., 2008;Mathot et al., 2019;Mathot & Frankenhuis, 2018;R eale, Garant, et al., 2010;Royaut e et al., 2018;Stamps, 2007;Wolf et al., 2007). Moreover, the POLS hypothesis proposes that selection should favour particular locally adaptive suites (syndromes) of correlated behavioural, physiological and life-history traits among individuals. For example, according to this hypothesis, 'fast', 'proactive' or 'risk-prone' (i.e. more exploratory, more aggressive, bolder, less sociable) individuals are expected to have higher maintenance metabolic rates and faster life histories (e.g. higher growth rate, earlier maturity, higher rate of reproduction, shorter life span) than 'slow', 'reactive' or 'risk-averse' (i.e. less exploratory, less aggressive, timid, more sociable) individuals within and across populations (cf. R eale et al., 2007). In support of the POLS hypothesis, there is empirical evidence for a general positive association, and coevolution, between maintenance metabolic rate and pace-of-life history (Auer et al., 2018) and between maintenance metabolism and behaviour (Biro & Stamps, 2008Careau & Garland, 2012;Mathot et al., 2019;Niemel€ a & Dingemanse, 2018) across diverse taxa. However, this body of evidence is highly heterogeneous with respect to effect sizes and support for the POLS hypothesis across individual studies (reviewed in Biro & Stamps, 2010;Careau et al., 2008Careau et al., , 2014Careau & Garland, 2012;H€ am€ al€ ainen et al., 2021;Mathot et al., 2019;Niemel€ a & Dingemanse, 2018;R eale, Garant, et al., 2010;Royaut e et al., 2018).
In our current study, we focus particularly on the POLS's hypothesized general relationship between repeatable behavioural traits and repeatable metabolic rate at the among-individual, within-population level. Metabolic rate reflects the energetic costs of living, is thought to set the pace of life and is functionally linked to behaviour (Auer et al., 2018;Biro & Stamps, 2010;Careau et al., 2008;Careau & Garland, 2012;Mathot et al., 2019;Metcalfe et al., 2016). Similar to behavioural traits, metabolic rates have been shown to vary consistently among individuals and are repeatable and heritable in general (Biro & Stamps, 2010;Careau et al., 2008;Holtmann et al., 2017;Metcalfe et al., 2016;Nespolo & Franco, 2007;White et al., 2013). Consequently, metabolic rates are subject to selection and can evolve (reviewed in Auer et al., 2018;Killen et al., 2016). According to the POLS hypothesis, among-individual variation in maintenance metabolism (i.e. minimum energy required to maintain the homeostatic mechanisms necessary for life) can be expected to be associated with among-individual variation in behaviours that contribute net energy (e.g. boldness, foraging, food defence) and/or cost net energy (e.g. locomotion, foraging, aggression, courtship) to the individual (Biro & Stamps, 2010;Careau et al., 2008;Mathot et al., 2019;Metcalfe et al., 2016). However, recent meta-analyses have found either little to no correlation (Royaut e et al., 2018) or statistically significant weak or modest correlations (Mathot et al., 2019;Niemel€ a & Dingemanse, 2018) between behaviour and maintenance metabolic rate within and among taxa, suggesting that the direction and strength of this relationship are highly variable and equivocal in general.
Trends between behaviour and metabolism appear equivocal and correlations weak or near zero because these two traits are labile and the relationship between them depends on several mediating factors, such as ambient environmental conditions, sex and behaviour type (Auer et al., 2020;Biro & Stamps, 2010;H€ am€ al€ ainen et al., 2018Killen et al., 2016;Mathot et al., 2019;Mitchell & Houslay, 2021;Niemel€ a & Dingemanse, 2018;Royaut e et al., 2018;Tarka et al., 2018). Interestingly, another potential mediator of the behaviouremetabolism relationship not yet well explored is sexual selection (H€ am€ al€ ainen et al., 2018). The action of this mediator might be expected given that the sexes differ in their reproductive investments and strategies (Andersson, 1994;H€ am€ al€ ainen et al., 2018). For example, males in numerous species vary in their expression of costly sexually selected ornamental traits and courtship behaviour (e.g. Andersson, 1994;Houde, 1997), and females can sexually prefer males that express certain personalities (e.g. high levels of boldness and/or exploration) over other males (e.g. Godin & Dugatkin, 1996;Scherer et al., 2020;Schuett et al., 2010).
Here, using laboratory-born adult male guppies, Poecilia reticulata, originating from a feral population in Australia, we tested under standardized laboratory conditions (1) whether behavioural and metabolic traits vary consistently among individuals over time (i.e. are repeatable) and (2) the POLS prediction that repeatable behavioural traits should be intercorrelated with each other and with repeatable metabolic rate, thus constituting a behavioural syndrome and behaviouremetabolism syndrome, respectively, at the among-individual within-population level. In addition, based on the POLS hypothesis, indicator models of sexual selection (Andersson, 1994) and prior studies on the personality of male guppies (e.g. Godin & Dugatkin, 1996;Irving & Brown, 2013), (3) we hypothesized that sexually selected male colour ornamentation that potentially indicates male quality would predict individual personality and metabolic rate. More specifically, we might expect that more colour-ornamented males would be bolder, more exploratory, less sociable and exhibit higher resting metabolic rates than their less colour-ornamented counterparts. We focused on adult male guppies as experimental subjects because they exhibit determinate growth at sexual maturation (Houde, 1997;Magurran, 2005;, which controls for any potential trade-off between metabolic energy allocation to somatic growth versus behavioural activities and its effect on the behaviouremetabolism relationship (Biro & Stamps, 2010), and they allow for the testing of the potential mediating role of costly male sexual ornaments (i.e. polymorphic body coloration; Endler, 1995;Houde, 1997) on the relationship between behaviour and metabolic rate.

General Methodological Approach
We followed contemporary recommended guidelines for rigorous studies of animal personality and behavioural syndromes, namely, repeated measures of multiple traits on the same set of subjects, standardized protocols, controlling/accounting for potential confounding factors, and using univariate and multivariate modelling frameworks as appropriate White et al., 2020). Importantly, we used state-of-the-art univariate and multivariate linear mixedeffects models Houslay & Wilson, 2017;Nakagawa & Schielzeth, 2010;Stoffel et al., 2017;Westneat et al., 2015) to decompose the phenotypic (co)variances in labile traits, namely behaviour and metabolism, into their component among-individual and residual within-individual (co) variances. With the latter component (co)variances, we estimated, respectively, trait repeatability and among-individual and residual within-individual correlations between multiple behavioural traits, multiple metabolic traits and multiple elements of a sexual selected trait (body coloration) in the same set of male subjects.

Subjects and Husbandry
Under licence from the Northern Territory government, freeranging adult guppies were collected in 2014 from a feral population in a freshwater drainage ditch connected to an estuarine creek, known to contain fish predators (Irving & Brown, 2013), that empties into Fannie Bay, Darwin, Northern Territory, Australia. The fish were placed in aerated, water-filled plastic bags and airfreighted to us in Sydney by a commercial company (Aquagreen, Howard Springs, Australia). Upon arrival, they were housed in large mixed-sex stock aquaria (180-litre) filled with aged, aerated and filtered tap water (range 26.0e27.0 C) and containing plastic replica aquatic plants for enrichment and shelter. The aquaria were exposed to overhead full-spectrum fluorescent lighting on a 12:12 h light:dark cycle. The fish were fed ad libitum daily with commercial flake food (Nutrafin Max, R.C. Hagen Ltd, Castleford, U.K.). On days when they were tested in behavioural or metabolism trials (described below), subjects were fed after completion of trials.
Test subjects (N ¼ 33) were first-generation laboratory-born adult males selected haphazardly (i.e. without known bias) with a large dip-net from stock aquaria containing over 200 individuals and held individually in clear plastic mesh containers (10 Â 10 and 20 cm high) placed in a large communal water bath under the same conditions as the stock aquaria. We numbered each container as a means of identifying individuals. The spatial position of every container within the water bath was periodically shuffled to minimize potential spatial effects and to expose all fish to a range of conspecific visual and chemical cues.

Behavioural Assays
We first quantified sequentially three behavioural traits (sociability (1 measure); boldness (2 different measures); exploration (2 different measures)) in each of the test males, one trait per day on three successive days. Each of the three traits was similarly assayed a second time 7 days later, in the same fixed order as in the first set of assays, to estimate their repeatability. We pseudorandomized the time of day each fish was tested to control for any potential time-of-day effect on behaviour. Following common practice in studies of behavioural syndromes (reviewed in Bell, 2013), we chose to assay the three behavioural traits for each test fish in the same fixed sequential order rather than a random one to minimize and standardize across all subjects any potential carryover effect of an individual's experience in a given test on its behaviour in a subsequent test. No test order effect was observed in a previous experimental study of behavioural syndromes in the guppy (Smith & Blumstein, 2010).
Only one of us (J.-G.J.G.) observed and quantified the behaviour of the fish to avoid interobserver measurement errors. Between repeated assays, individual test fish were returned to their respective home container in the communal water bath.

Sociability
Sociability is an individual's propensity to socially associate with others and to form social groups (Mathot et al., 2019;R eale et al., 2007). To assay this behavioural trait, we presented individual test males with a mixed-sex group of conspecifics (¼stimulus shoal) constrained in a small compartment (16 Â 30 cm) at one end of a rectangular test aquarium (88 Â 30 and 15 cm high; Fig. 1a) constructed of white acrylic plastic. The aquarium contained a white sand substratum and was filled with aged, aerated tap water to a depth of 8 cm. We added 1 litre of water from our guppy home holding tank to the water in the test aquarium to saturate it with collective guppy chemical cues, which presumably rendered the environment chemically familiar to the test fish and minimized any stress associated with its transfer into the test aquarium.
Guppies typically associate socially with each other within mixed-sex shoals in the wild (Croft, Arrowsmith, et al., 2003;Magurran, 2005). The stimulus shoal comprised two adult males and two adult females, which were chosen haphazardly from stock aquaria. All test subjects were unfamiliar with the stimulus shoal fish prior to being experimentally assayed for their sociability. We used the same stimulus shoal in all sociability test trials for stimulus consistency and to minimize the total number of fish used in the assay across all trials.
The shoal compartment (Fig. 1a) was separated from the remainder of the test aquarium by a clear perforated plastic screen, allowing both visual and chemical cues to be transmitted between the shoal and test male. A 10 cm wide social association zone directly in front of the stimulus shoal compartment was delineated with markers placed on the substratum. The width of this zone corresponds approximately to four to five male guppy body lengths, a commonly observed interindividual distance among shoaling fishes in nature (Pitcher & Parrish, 1993) including the guppy (Croft, Arrowsmith, et al., 2003). The entire apparatus was placed inside a blind to minimize external disturbances and uniformly illuminated overhead with full-spectrum fluorescent tubes.
Prior to the onset of a trial, an opaque screen was lowered in front of the stimulus shoal compartment and a test male was placed in a perforated clear plastic cylinder (7 cm diameter) at the opposite end of the test aquarium from the stimulus shoal. After a 10 min acclimatization period, we slowly raised the opaque screen in front of the stimulus shoal compartment with a pulley system and then raised the cylinder with a separate pulley system, thereby allowing the test male to swim freely in the test aquarium. We filmed (at 1080p, 25 frames/s) the test male using an overhead Logitech HD webcam (model C615) coupled to a computer. We quantified the test male's sociability (i.e. shoaling tendency) as the cumulative time it spent within a 10 cm wide zone adjacent to the stimulus shoal (Fig. 1a) over a 5 min test period.
Because individual test males were exposed to the same stimulus shoal in repeated tests 1 and 2, it is possible that a test male's behavioural response to the stimulus shoal might have been confounded by the identity of the individual fish within the shoal. However, this is unlikely for at least four reasons. (1) The initial exposure to the stimulus shoal (test 1) was only for 5 min, limiting the time available to learn the identity of the individual members of the shoal. (2) Individual acquisition of social familiarity/recognition of conspecifics within a shoaling context requires multiple prior repeated exposures to them over at least 10 consecutive days in the guppy (Griffiths & Magurran, 1997), a time frame that exceeds the 7-day period between the repeated sociability test assays in our current study. (3) The mean (±SE) time spent associating with the stimulus shoal did not differ between repeated test 1 (200.9 ± 11.7 s) and test 2 (191.2 ± 13.1 s) across all test male guppies (paired t test: t 32 ¼ 0.932, P ¼ 0.359) in the current study, suggesting that there was no apparent carryover effect of the behaviour of focal test males towards the stimulus shoal from the first to the second test. (4) Supporting the latter finding, our multivariate models did not reveal any effect of repeat test number on any of our behavioural measures, including sociability (see Results below).

Boldness
Boldness is an individual's willingness to be exposed to a perceived risk (Mathot et al., 2019;R eale et al., 2007), such as a potential threat of predation (Dugatkin & Godin, 1992a). We assayed individual boldness using a test apparatus consisting of a rectangular white acrylic plastic aquarium (88 Â 30 and 15 cm high; Fig. 1b). The test aquarium was filled to a depth of 8 cm with aged, aerated tap water and uniformly illuminated overhead with full-spectrum fluorescent light tubes. As for the above sociability test, we added 1 litre of water from our guppy home holding tank to the water in the test aquarium to saturate it with collective guppy chemical cues. At one end of the aquarium, a predator compartment (16 Â 30 cm) was separated from the open zone (72 Â 30 cm) by a sealed clear glass partition. This compartment contained a live juvenile Australian bass, Macquaria novemaculeata, a native predator of the Australian guppy. For each trial, we haphazardly selected a bass from a large stock tank containing approximately 30 unmarked individuals (ca. 10e12 cm total length). The open zone was demarcated at 5 cm intervals using black marker lines on the uniformly white bottom of the aquarium to facilitate the recording of the test fish's position. Prior to the onset of a trial, an opaque removable screen was lowered in front of the predator compartment and a test male guppy was transferred from the holding tank to a clear perforated plastic cylinder (7 cm diameter) at the other end of the test aquarium, next to a submerged plastic plant on the  bottom, which could serve as a potential refuge (Fig. 1b). The fish were left undisturbed to acclimatize for 10 min. The entire apparatus was placed inside a blind to minimize external disturbances and uniformly illuminated overhead with full-spectrum fluorescent tubes.
Following the acclimatization period, we began a 5 min trial by remotely raising the opaque screen in front of the predator compartment with a pulley system, allowing the test fish to view the fish predator at a distance for 2 min. We then slowly raised the cylinder with a separate pulley system, thereby allowing the test male to swim freely in the test aquarium and approach (¼ 'inspect') the predator. The long body axis of the bass predator typically faced the guppy broadside during the trials (as depicted in Fig. 1b). Guppies commonly and cautiously approach potential fish predators in the wild (Dugatkin & Godin, 1992b;Magurran & Seghers, 1994). We video filmed the test male using an overhead Logitech HD webcam (model C615) coupled to a computer. We concurrently quantified two measures of boldness, namely (1) the number of times the test fish inspected or approached the predator (¼ inspections) and (2) the nearest distance of the test fish to the predator compartment at the end of each inspection visit. From the latter data, we calculated the test fish's median inspection distance (¼ approach distance) to the predator.
Predator inspection or approach behaviour is a commonly used measure of individual risk-taking or boldness behaviour in animals in general and in fishes in particular (e.g. Dugatkin & Godin, 1992a, b;Smith & Blumstein, 2010). During a predator inspection visit, a prey individual cautiously approaches a putative predator at a distance in a series of brief directed approaches alternating with short stationary pauses, during which the inspector is visually fixated on the predator (Dugatkin & Godin, 1992a). An inspection visit is deemed to have ended when the inspector stops its directed forward movements and moves away from the putative predator. Because approaching predators is inherently dangerous (Dugatkin, 1992;Dugatkin & Godin, 1992a;Milinski et al., 1997) and the risk of being captured by a predator is greater the closer the prey is to the predator at the instance of an attack (Lima & Dill, 1990;Milinski et al., 1997), we therefore assumed here that the degree of boldness (risk taking) of a fish to be positively related to its frequency of predator inspections and inversely related to its median approach distance to the predator during the assay.

Exploration
Exploratory behaviour is commonly defined as an individual's movement in a novel environment and/or response to novel objects or resources (Mathot et al., 2019;R eale et al., 2007). Male guppies are more mobile than females and they actively move within and between pools (potentially novel habitats) in streams and rivers in nature (Croft, Albanese, et al., 2003).
We assayed males individually for their exploratory behaviour in an open field novel environment, consisting of a white acrylic plastic tank with angled walls (64.5 Â 64.5 Â 12.5 cm high at the bottom, 69 Â 69 and 12.5 cm high at the top; Fig. 1c). The experimental tank was filled with aged, aerated tap water to a depth of 4 cm. This novel environment contained two circular beige sandstone blocks (4 cm height, 5.5 cm diameter) as visual landmarks on the bottom. The two blocks were placed in fixed positions along one or the other diagonal of the tank depending on the repeated test. Following Jones and Godin (2010), the diagonal positioning of the blocks was alternated between the two repeated tests on each fish, with the order of presentation randomized, to mitigate potential habituation to the environment. The entire apparatus was placed inside a blind to minimize external disturbances and uniformly illuminated overhead with full-spectrum fluorescent tubes.
For a given trial, a male was first transferred from his holding aquarium compartment to a perforated clear plastic cylinder located in the start corner of the experimental tank ( Fig. 1c) and left undisturbed to acclimatize for 10 min. Following this period, the cylinder was gently raised remotely, allowing the test fish to swim freely in the novel environment ('explore') for 5 min. We filmed the test male using an overhead webcam as above. The video files were converted to .avi files and later analysed using the open-source tracking software 'CTrax' v.0.1 (CalTech, Pasadena, CA, U.S.A.) and the associated manual correction package 'FixErrors GUI' in MatLab (Mathworks, Natick, MA, U.S.A.). We superimposed a digital grid of 36 equal-sized squares (10 Â 10 cm) over the water surface of the entire tank (see Appendix, Fig. A1). CTrax produced consistent tracks and recorded the position (xey coordinates) of the test male every 0.12 s (i.e. at a rate of 5 scans/s) and enumerated the number of grid squares entered at least once by the test fish (Appendix, Fig. A1). From the tracking data, we quantified two measures of exploration behaviour: (1) the proportion of the total area of the novel environment explored by the test fish (¼ proportion explored), expressed as the number of different grid squares it entered at least once divided by 36 (the total number of available squares) and (2) total distance (to the nearest 1.0 cm) the test fish travelled in the environment (¼ travel distance) during the 5 min assay.

Metabolic Rate Assays
The aerobic metabolic rate of individual animals is typically estimated from their integrated oxygen uptake rates measured using respirometry under standardized conditions (e.g. Careau et al., 2008Careau et al., , 2014Chabot et al., 2016;Norin & Clark, 2016). After allowing individual male guppies 11e23 days to recuperate in their respective home containers following their use in the above behavioural assays, we then assayed their individual whole-animal metabolism (cf. Hayes, 2001) twice, with 22e27 h elapsed between the repeated measures, to estimate their repeatability. We pseudorandomized the time of day the fish were tested to control for any potential time-of-day effect on metabolic rate. Given the minimal number of repeated measures (k ¼ 2) of individual traits in the current study and for logistical reasons (limited availability of the respirometers), we chose a shorter interval (1 day) between repeated measures of metabolism than for repeated measures of behaviour (7-day interval) to minimize any spontaneous changes in the fish's internal state that could affect its energetic needs and thus metabolic activity between repeated measurements.
Following Seebacher et al. (2013), we quantified in fixed order on the same day whole-animal maintenance (resting) and maximal activity-induced oxygen uptake rates for each of our test males using small respirometers (Appendix, Fig. A2) and a fibre-optic oxygen-measuring system (Witrox 4, Loligo Systems, Viborg, Denmark) according to the manufacturer's instructions. These respiration rates are proxies for individual resting metabolic rate (RMR) and maximal metabolic rate (MMR or VO 2max ) during strenuous exercise over short periods, respectively (Auer et al., 2017;Careau et al., 2014). Here, RMR is defined as the lowest metabolic rate of a postabsorptive, nonreproductive, undisturbed and resting fish acclimated to a given ambient temperature, which includes the costs of low-level spontaneous movements and maintenance of body posture and equilibrium (Careau et al., 2014;Careau & Garland, 2012;Chabot et al., 2016). The maximal active metabolism (MMR) of animals is the upper bound of their aerobic capacity and thus limits their performance Norin & Clark, 2016). For each test male, we always measured RMR before MMR.
The oxygen content of the water inside the respirometers was adjusted for ambient water temperature and local barometric pressure. The measured oxygen uptake rates of individual fish were adjusted for their wet body mass by including body mass (in grams) as a fixed-effect variable in our statistical models. We cleaned the respirometers with fresh tap water after use and left them to air dry overnight before being reused. At the time of their testing, individual fish had been previously fasted for 19e23 h and were thus likely to have completely evacuated food from their gastrointestinal tract and to be in a postabsorptive state (cf. He & Wurtsbaugh, 1993).
To ensure reliability of measurements, we calibrated daily the O 2 sensor spots inside our respirometers using water that was either fully saturated (100%) with or completely depleted (0%) of dissolved oxygen. Dissolved oxygen saturation and anoxia were achieved by bubbling compressed air into the calibration water bath or by adding sodium sulphite to the water bath, respectively. Only one of us (J.-G.J.G.) quantified the metabolic rates of the fish to ensure consistency of execution of the protocols and to avoid interobserver measurement errors.

Resting metabolism
For resting oxygen uptake measurements, we used four identical cylindrical glass respirometers (19 mm internal diameter, 100 mm long, 27 ml volume; Appendix, Fig. A2) equipped with a fibre-optic sensor spot (Loligo Systems) affixed internally at their midpoint and coupled to an oxygen meter via a fibre-optic cable. The respirometers were submerged in an opaque plastic container filled with aged and aerated tap water under temperature control. On any given day, fish were concurrently tested in sets of three, one fish in each of three respirometers. The fourth respirometer did not contain a fish and served as a blank control for any microbial respiration in the water. A small vertical piece of opaque Plexiglas was placed between adjacent respirometers so that the fish would not be able to view each other during measurements. We gently placed a fish inside each respirometer and allowed them to rest undisturbed for 2 h with O 2 -saturated water (26.2 ± 0.07 C, mean ± SE) pumped through the respirometers by peristaltic pumps (model i150, iPumps, Tewkesbury, U.K.). This initial 2 h resting period is sufficient time for a small fish to recover from handling stress and for its O 2 consumption rate to stabilize in this respirometer (Seebacher et al., 2016). We then remotely turned off the pumps and automatically recorded (at intervals of 1 s) the linear decrease in dissolved oxygen owing to respiration over a 10 min period or until a steady rate of oxygen decrease was established. We determined oxygen uptake rates by the fish from the slope of the regression line-of-best-fit of the decrease in oxygen content inside each respirometer over time using PRISM 6.0 (GraphPad Software Inc., La Jolla, CA, U.S.A.; see Appendix, Fig. A2). We subtracted the slope of any temporal decline in the oxygen content of the water in the blank control respirometer from the slope of the decrease in oxygen content of the water in each of the respirometers containing a fish. The resultant adjusted slope estimates the RMR of the fish.

Maximal active metabolism
Following the measurements of their RMR, the fish were gently removed from their respective cylindrical respirometer and placed individually into a closed-system, circular glass respirometer (124 ml volume; Appendix, Fig. A2) equipped internally with a fibre-optic sensor spot (same as above) and filled completely with O 2 -saturated aged water (26.4 ± 0.04 C) to quantify their activityinduced maximal rate of oxygen uptake rate (MMR), as described in Seebacher et al. (2013). After a 10 min acclimatization period, the fish was induced to swim against a circular water current inside the respirometer. Water current velocity was gradually increased until the fish swam steadily at its maximum sustained swimming speed while maintaining its position in the water column and facing upstream. Oxygen uptake was automatically measured every second during 10 min of steady swimming. We estimated MMR from the slope of the decrease in oxygen content (Appendix, Fig. A2). As for the above measurements of RMR, we ran a blank control (empty respirometer) to account for any microbial respiration and to accordingly adjust the maximal oxygen uptake rate of each fish within the set of three fish tested on a given day.
For both RMR and MMR, whole-animal rates of oxygen uptake (in mmol O 2 /min; Hayes, 2001) by test fish were determined as the adjusted slope of the decrease in dissolved oxygen content in the water multiplied by the volume of the respirometer. We statistically controlled for interindividual variation in body size by including the body mass (in grams) of individual fish as a fixedeffect variable in all our models.

Aerobic scope
We calculated metabolic aerobic scope (AS) as the absolute difference between an individual's whole-animal MMR and RMR oxygen uptake rates (Norin & Clark, 2016). AS represents a fish's maximal metabolic energy use after its maintenance energetic needs have been met (Norin & Clark, 2016).

Measuring Fish Body Mass, Length and Colour Ornamentation
We measured, weighed and anaesthetized/photographed individual subjects only once, immediately following their use in the experiment, to minimize stress and to avoid its potential carryover effects on behaviour and metabolism between repeated test assays.
Each test male was wet-weighed to the nearest 1.0 mg on a Sartorius balance and then lightly anaesthetized in a solution of clove oil (40 mg/litre), placed on a piece of white Plexiglas alongside a metric ruler and its left side photographed with a digital camera under standardized lighting. Following photography, the fish was transferred to a container containing aerated aquarium water to recuperate before being returned to a stock aquarium. Using the open-source software ImageJ (http://imagej.nih.gov/ij/), we measured the standard body length of each male to the nearest 0.1 mm and scored its body colour ornamentation from its digital photograph as follows. The standard body length and wet body mass of test male guppies averaged (±SD) 19.8 ± 1.6 mm and 151.3 ± 38.6 mg, respectively.
Using the area tracing tool in ImageJ, we measured separately the areas of orange-yellow (hereafter orange), black and 'fuzzy black' pigmentation and iridescent structural colours (silver, blue, green, purple) on the left side of each test male's body, including the caudal fin (see example guppy photograph in Fig. 1). To control for interindividual variation in body size, each separate area of coloration was divided by the body area (including caudal fin area) of the fish and thus represented separate univariate colour elements. We then calculated the relative area of total body coloration ((orange þ black þ fuzzy black þ iridescence areas)/total body area) for each male as their individual composite colour ornamentation score. Female guppies most likely assess and respond to overall or composite male colour ornamentation when choosing mates (Brooks & Endler, 2001a, b;Endler & Houde, 1995;Houde, 1997). Only one of us (J.-G.J.G.) photographed, weighed and measured the fish and analysed all photographs to avoid interobserver measurement errors. Body length, mass and coloration measurements were taken blind of the subjects' behavioural and metabolic trait measures.
All of the aforementioned colour classes are heritable (Brooks & Postma, 2011) and are sexually selected to varying degree (depending on the population) through female mate choice in both the Trinidadian (reviewed in Houde, 1997) and Australian guppy (e.g. Brooks & Endler, 2001a, b). Moreover, the degree of expression of at least orange coloration in male guppies is condition dependent (Houde, 1997;Magurran, 2005) and thus a potential honest indicator of male quality (cf. Andersson, 1994;Houde, 1997). There is recent meta-analytic evidence (White, 2020) for a significant positive correlation between iridescent colours and male quality (as measured by body condition and immune function) in animals in general.

Data Analyses
All data were analysed in R v.3.3.3 (R Core Team, 2017) and all statistical tests were two tailed.

Trait repeatability
Following Nakagawa and Schielzeth (2010) and Stoffel et al. (2017), we estimated the repeatability (R) and associated 95% credible interval (CI) of each behavioural and metabolic trait measure separately by fitting univariate linear mixed-effects (LMM) models (Bates et al., 2015) to our repeated-measures data using the 'rptR' package. The five behavioural trait measures (inspections, approach distance (log-transformed), sociability, travel distance and proportion explored (logit transformed; cf. Warton & Hui, 2011)), were modelled assuming Gaussian error distributions. The three proxy measures of metabolism (RMR, MMR, AS) were similarly modelled assuming Gaussian error structures. All models included individual test male identity (ID) as a random effect and individual body mass and repeat test number as fixed effects. Repeat test number as a fixed effect controls for mean differences in behaviour or metabolic rate between the first and second tests. Individual body mass was controlled for because metabolic rate generally scales allometrically with body mass (Auer et al., 2018;Careau et al., 2008;Killen et al., 2010Killen et al., , 2016 and body mass can influence behaviour (e.g. Brown & Braithwaite, 2004;Darby & McGhee, 2019;Dowling & Godin, 2002) in animals. Specifically, we used the 'rptGaussian()' core function designed for fitting Gaussian data (Stoffel et al., 2017). The 'rpt' core functions employ parametric bootstrapping for estimating 95% credible intervals (Nakagawa & Schielzeth, 2010;Stoffel et al., 2017); we used 1000 bootstraps in our models. We confirmed by visual inspections of histograms and QeQ plots of the residual errors and using the ShapiroeWilks test that the model residuals were normally distributed for all behavioural and metabolic trait measures (Sha-piroeWilks test: 0.979 < W < 0.996; 0.330 < P < 0.999), with the exception of the model residuals for proportion explored, which were quasi-normally distributed. Mixed-effects models are generally very robust to violations of the assumptions about the distribution of residual effects (Schielzeth et al., 2020).
Repeatability was computed as the intraclass correlation or R ¼ V I /(V I þ V R ), where V I is the among-individual variance and V R is the residual within-individual variance Nakagawa & Schielzeth, 2010). R thus represents the proportion of the total observed phenotypic variation in a given trait that can be attributable to among-individual variation (V I ). The residual within-individual variance (V R ) represents statistical measurement error and general environmental variation over time . V R is also of biological relevance because it includes average within-individual plasticity towards any stimulus that is statistically unaccounted for (Westneat et al., 2015) and can potentially influence correlations between traits Mitchell & Houslay, 2021). The V I and V R variance components underlying each computed R estimate are shown in the Appendix (Table A1). We calculated the precision of each observed R estimate, measured as the width (CIW) of its associated 95% credible interval (CIW ¼ upper CI e lower CI; Wolak et al., 2012). We approximated the statistical power of each of our models to identify nonzero values of repeatability (R), and the expected precision of each of our computed R estimates, by visual extrapolation of the simulation results of Dingemanse and Dochtermann (2013, Table 1), respectively, for the case of k ¼ 2 repeated measures and N ¼ 33 individual subjects. Although the latter simulation results are overall robust, individual approximated estimates of CIW and statistical power (see Results) might not necessarily be robust. We note that our repeatability estimates (R) represent adjusted repeatabilities (sensu Nakagawa & Schielzeth, 2010;Stoffel et al., 2017), because they were adjusted for individual body mass and repeat test number as fixed effects in the aforementioned models.

Relationships between the metabolic traits
It is notable that there was considerable phenotypic variation in the three proxy metabolic trait measures of whole-animal metabolism (i.e. RMR, MMR, AS) in our test subjects. We characterized this variation as phenotypic correlations between RMR, MMR and AS. To do this, we carried out separate bivariate regressions using linear mixed-effects models ('lm()' function in the 'lme4' package; Bates et al., 2015), with the body mass of individual males and repeat test number included as fixed-effect covariables in the models. The residuals of all three models were normally distributed (ShapiroeWilks test: 0.980 < W < 0.983, 0.349 < P < 0.516).

Trait (co)variation
For any two labile and repeatable traits in general, the amongindividual and within-individual correlations (covariances) jointly contribute to the phenotypic correlation between the traits Dingemanse et al., 2012). To ultimately test for among-individual correlations between the behavioural trait measures and between behaviouremetabolism ecolour ornamentation measures in male guppies, we carried out the following procedural steps.
First, to test for a behavioural syndrome, we concurrently estimated the among-individual and within-individual (co)variances for each of the five repeatable behavioural trait measures, as determined by the above univariate 'rptR' models (see Results), using a multivariate modelling framework recommended for studies of behavioural syndromes and the POLS Houslay & Wilson, 2017). With these (co)variances, respectively, we then estimated among-individual (r I ) and residual withinindividual (r R ) correlations between the behavioural trait measures to test for the presence of a behavioural syndrome. Covariation among behaviours that arises from significant among-individual differences is taken as evidence for the presence of a behavioural syndrome (e.g. Houslay & Wilson, 2017;Sih et al., 2004).
This multivariate analysis was carried out using the 'MCMCglmm()' function in the R package 'MCMCglmm' (Hadfield, 2010), following Houslay and Wilson (2017) and Kniel and Godin (2019). The 'MCMCglmm()' function fits generalized linear mixedeffects models (GLMM) in a Bayesian framework using Markov chain Monte Carlo (MCMC) techniques. Our multivariate model included the five repeatable behavioural trait measures as response variables, the body mass of individual test subjects and repeat test number as fixed effects, male identity (ID) as a random effect, and family Gaussian with identity link. We mean-centred and scaled to units of 1 standard deviation the response variables, mean-centred the repeat test number (fixed effect) and mean-centred and scaled to standard deviation units male body mass (fixed effect). Mean centering and scaling place each behavioural response variable on the same scale, which facilitates model fitting and interpretation of the results. We defined a parameter-expanded flat prior centred around 0 and with a large variance (of 625) that should be uninformative for our model. During the fitting process, the model was iterated 800 000 times with a burnin of 60 000 and thinning interval of 200. We confirmed model convergence by running four separate chains and applying the GelmaneRubin and HeidelbergereWelch diagnostic tests ('gelman.diag()' and 'heidel.diag()' functions, respectively, in the R package 'coda').
Second, we could not carry out a similar multivariate analysis to test for the presence of among-individual correlations between the proxy measures of metabolism (i.e. RMR, MMR and AS), because only one of these measures (namely, RMR) was significantly repeatable (see Results below).
Third, to test for behaviouremetabolismecolour ornamentation covariation, we used a separate, more inclusive multivariate 'MCMCglmm' model to estimate among-individual and residual within-individual correlations between each of the behavioural trait measures, the repeatable resting metabolic rate (RMR) and the composite colour ornamentation scores of individual male guppies. Because male body colour elements were measured only once per individual (see above subsection), there was no residual withinindividual covariance in the coloration scores of individuals and therefore residual within-individual correlations involving coloration must be 0. However, because variances have to be positive, we fixed the within-individual variance in body coloration to a very small positive number (i.e. 0.0001) in all our multivariate models that included coloration as a response variable (cf. Houslay & Wilson, 2017).
Lastly, given that the direction and strength of female mating preferences for orange, black, fuzzy black and iridescent male colours vary within and between populations of the guppy (Brooks & Endler, 2001a, b;Endler & Houde, 1995;Houde & Endler, 1990), it is biologically meaningful to also test separately whether each of these four colour elements is correlated with any of the repeatable male behavioural traits and/or RMR. We did this with a separate 'MCMCglmm' model, which included these four distinct colour elements among the response variables.
We illustrated the correlational relationships between paired traits as matrices of scatterplots using the 'scatterplot()' function in the R package 'car' (Fox & Weisberg, 2019). Each scatterplot plots the raw data and shows the among-individual correlation (r I ) between the paired traits with its corresponding Bayesian 95% credible intervals (CIs) obtained from the relevant multivariate mixedeffects model (described above). In addition, as recommended by Dingemanse and Wright (2020) and Houslay and Wilson (2017), we also report more fully the results of our models as separate matrix tables in the Appendix, each showing among-individual mean variances (V I ), among-individual between-trait mean covariances (COV I ) and among-individual correlations between paired traits (r I ) with their corresponding 95% CI. The 95% CI can function as an informative indicator of statistical significance and uncertainty (Nakagawa & Schielzeth, 2010). As is common practice (Houslay & Wilson, 2017), we interpreted any Bayesian 95% CI that did not span zero to indicate statistical significance in the classical (frequentist) sense. Conversely, if the 95% CI associated with an amongindividual correlation (r I ) between traits crossed zero, then we deemed this correlation coefficient to be statistically nonsignificant.

Ethical Note
Our study received prior approval from the respective institutional animal care/ethics committee at the University of Sydney (protocol number 2014/747) and Carleton University (file number 100937), and thus adheres to the laws of both Australia and Canada and the ASAB/ABS Guidelines for the use of animals in research.
Our experimental fish were held in relatively large glass aquaria, provided with aerated, aged and filtered water at stable temperatures, and submerged and floating plastic aquatic plants for enrichment and shelter, and fed ad libitum daily with commercial flake foods. These aquaria were visually checked daily for any abnormal behaviour and fungal and bacterial infections in the fish (none were detected during the course of our study). We made every effort to minimize any discomfort or stress in our fish. Access to our aquatic laboratory was restricted to the investigators and, as such, external disturbances to the fish were minimized. The behavioural essays (10 min acclimatization þ 5 min assay) and metabolism essays (2 h acclimatization þ 10e15 min assay) were kept relatively brief, while allowing for adequate acclimatization to the testing apparatuses and the recording of the behavioural and metabolic rate variables under study. In the boldness assay, individual test guppies were provided with a submerged aquatic plant for refuge and were visually exposed for only 5 min to a live fish predator (as stimulus) located behind a sealed glass partition, which did not allow for attacks or actual predation. The guppies were free to move away from the predator. Following the behavioural and metabolism essays, individual male guppies were lightly anaesthetized for digital photography, a procedure that typically lasted 60e90 s, after which they rapidly recuperated (<5 min) in well-aerated water. At the end of the study, all experimental fish were healthy and were returned to the breeding stock aquaria described above.

Trait Repeatability
Our univariate linear mixed-effects models revealed statistically significant adjusted repeatability estimates (R) for all proxy measures of the three behavioural traits (i.e. sociability, boldness, exploration), but only for one of the three metabolic traits, namely resting metabolic rate (RMR) ( Table 1). The R estimates for MMR and AS were relatively small and not significantly different from zero, partly owing to low statistical power to identify nonzero values of repeatability for these two measures of metabolism (Table 1). Approximated power level for all the other trait measures was high (>0.800; Table 1) and can thus be considered acceptable (cf. . The mean repeatability estimate (R ¼ 0.622) and precision (CIW ¼ 0.413) for the behavioural trait measures were considerably greater than the corresponding mean values for the metabolic traits (R ¼ 0.343, CIW ¼ 0.547; Table 1).

Trait (Co)Variation
The estimated among-individual correlations (r I ) between all possible paired combinations of behavioural trait measures, except two (sociabilityeproportion environment explored, approach distance to predatoredistance travelled), were statistically significantly different from zero (Fig. 2, Appendix, Table A2). The latter two nonsignificant correlations (r I ¼ 0.417 and À0.485) were Univariate 'rptR' model estimates of repeatability (R), adjusted for individual body mass and repeat test number, and associated 95% credible intervals (CIs) for behavioural and metabolic trait measures that were each assayed twice for individual test male guppies. CIW is the width of the observed 95% confidence interval (¼ upper CI e lower CI) and taken as a measure of the precision of computed estimates of R; lower CIW values represent greater precision. Boldfaced P values denote repeatability estimates (R) that are significantly different from zero. Also shown are approximate values of the statistical power of our univariate models to identify nonzero values of repeatability that were visually extrapolated from the simulation results of Dingemanse and Dochtermann (2013), and approximate expected values of CIW extrapolated from the simulation results of Wolak et al. (2012), for N ¼ 33 subjects and k ¼ 2 repeated measures in our study. Each scatterplot also shows the coefficient of the among-individual correlation (r I ) between paired traits, with its corresponding 95% credible interval (CI, in brackets), the regression line and data-concentration ellipses (inner and outer ellipses represent the 50% and 95% confidence levels, respectively). The r I correlation coefficients and credible intervals were obtained from a multivariate generalized linear mixed-effects model fitted to the data that included the five repeatable behavioural trait measures as response variables, male body mass and repeat test number as fixed effects, and individual male identity (ID) as a random effect. Scatterplots with similarly coloured ellipses illustrate relationships between paired trait measures, one or both of which fall within the same behavioural functional category (red ¼ boldness; blue ¼ exploration). More details are provided in the Appendix, Table A2.
nevertheless relatively strong and their respective 95% credible intervals only marginally overlapped zero ( Fig. 2a and f, respectively). This observed suite of intercorrelated behavioural trait measures represents a sociabilityeboldnesseexploration behavioural syndrome, which is characterized by bolder males being more exploratory and more sociable than shyer ones, and more sociable individuals being more exploratory than less sociable ones. By definition, more exploratory males visited a greater proportion of the novel environment and swam longer distances than less exploratory ones (Fig. 2j), and bolder males inspected the predator more frequently and more closely than more timid individuals (Fig. 2g), both at the among-individual and within-individual correlational levels (Appendix , Table A3) for exploration and boldness, respectively. As expected from the structure of the observed behavioural syndrome (Fig. 2, Appendix, Table A2), the estimated among-individual correlations between paired behavioural trait measures were all substantially greater than the corresponding estimated residual within-individual correlations, most of which were not significantly different from zero but nevertheless in the same direction (sign) as the corresponding among-individual correlations (Appendix, Table A3). As an aside observation, the strong and significant among-individual correlations (r I ) between paired behavioural trait measures that were estimated by our first and simplest multivariate model (Fig. 2, Appendix, Table A2) were weaker when the same data were fitted to our most inclusive multivariate model (Appendix , Table A4), which additionally included RMR and colour ornamentation as response variables. This was driven entirely by the reduction in statistical power that resulted from estimating more parameters in the latter model and highlights the limits of our models and modest sample size and minimal replicate number (see also related section of the Discussion).
As noted above, because MMR and AS were not significantly repeatable (Table 1), we could not test for among-individual correlations between these three metabolic trait measures within our multivariate GLMM framework. Nevertheless, there was considerable phenotypic variation in the whole-organism resting (RMR) and maximal active (MMR) metabolic rates and aerobic scope (AS) in the test subjects (Appendix, Fig. A3). Unsurprisingly, MMR values (mean ± SE ¼ 0.09858 ± 0.00341 mmol O 2 /min) were approximately four-fold greater than RMR values (0.02409 ± 0.00201 mmol O 2 /min) on average. With individual body mass and repeat test number controlled for statistically, our mixed-effect linear regression models revealed a nonsignificant positive association between RMR and MMR (estimate ± SE ¼ 0.2120 ± 0.1915, t 65 ¼ 1.107, P ¼ 0.272; Appendix, Fig. A3a). RMR and AS were significantly inversely related (estimate ¼ À0.7880 ± 0.1915, t 65 ¼ À4.115, P < 0.001; Appendix, Fig. A3b), such that males with a low RMR tended to have a greater metabolic scope for aerobic activity than males with higher RMR. Conversely, AS was significantly positively correlated with MMR (estimate ¼ 0.9085 ± 0.0826, t 65 ¼ 10.999, P < 0.0001; Appendix, Fig. A3c).
Measures of boldness, exploratory activity and sociability tended to be weakly to modestly positively associated with resting metabolic rate, RMR (Fig. 3). Most importantly, however, our multivariate model revealed no significant among-individual correlation (r I ), residual within-individual correlation (r R ) or covariation (COV I ) between any of the paired combinations of behavioural trait measures and maintenance metabolism (RMR) (Fig. 3, Appendix, Table A4). As such, we found no compelling evidence for a behaviouremetabolism syndrome. Interestingly, across the behavioural and RMR trait measures, all but one of the estimated among-individual correlations between paired trait measures were greater in magnitude than the corresponding residual withinindividual correlations (Appendix , Table A3, A5). This suggests that among-individual covariance between the aforementioned traits contributed relatively more than residual within-individual covariation did to the between-trait phenotypic correlations observed in our study population of guppies, conditional on the magnitude of the repeatability of each trait.
Likewise, our multivariate analysis showed no significant among-individual correlation (r I ), nor residual within-individual correlations (r R ), between any of the possible paired combination of measures for the behavioural and metabolic (RMR) traits and male body colour ornamentation score (Fig. 3, Appendix, Table A5). Similarly, none of the univariate colour elements (i.e. orange, black, fuzzy black and iridescence) was significantly correlated with any of the five behavioural trait measures nor with RMR, either at the among-individual or within-individual level (Appendix, Table A6). The latter two findings suggest that body colour ornamentation did not predict individual male personality or maintenance metabolism in our current study.
Lastly, the posterior mean intercept values for the fixed-effect covariables (male body mass, repeat test number) entered in our multivariate GLMM models were all very small (most approximated 0) and statistically nonsignificant and therefore did not systematically influence any of the behavioural trait measures (body mass: 0.178 < P < 0.885; repeat test number: 0.327 < P < 0.999). Similarly, male body coloration was not affected by body mass (0.189 < P < 0.803). However, body mass significantly influenced resting metabolic rate (0.011 < P < 0.009), as expected, but repeat test number did not (0.382 < P < 0.403).

DISCUSSION
We found repeatable among-individual variation in several ecologically important behavioural traits (sociability, boldness, exploration) and resting metabolic rate (RMR) in individual male guppies originating from a feral population in Australia. Proxy measures of the behavioural traits were significantly intercorrelated and thus integrated into a proactiveereactive behavioural syndrome. Most importantly, the behavioural trait measures did not significantly covary with RMR among individuals, contrary to the central prediction of the POLS hypothesis. In addition, neither multivariate nor univariate body colour elements in individual males predicted their personality or maintenance metabolism, as would be expected based on indicator models of sexual selection.

Trait Repeatability
Controlling for individual body mass, repeat test number and ambient temperature, all proxy measures of behaviour and one proxy measure of metabolism (namely RMR) varied significantly among individual male guppies and were thus repeatable over time in our current study, consistent with the existence of personality and metabolism axes, respectively. This result corroborates previous findings that these same behavioural traits (sociability, boldness, exploration/activity; Bell et al., 2009;Holtmann et al., 2017;Wolak et al., 2012) and metabolic traits (Biro & Stamps, 2010;Careau et al., 2008;Holtmann et al., 2017;Metcalfe et al., 2016;Nespolo & Franco, 2007;White et al., 2013) are commonly, but not universally, repeatable in animals. For the guppy in particular, previous studies have also reported significant repeatability estimates for one or more of the above behavioural traits (e.g. Ariyomo & Watt, 2013;Biro et al., 2016;Harris et al., 2010;Houslay et al., 2018;Irving & Brown, 2013;Kniel & Godin, 2019;Trompf & Brown, 2014;White et al., 2016), but not clearly so for metabolic traits as the limited evidence for their repeatability is mixed (e.g. Biro et al., 2016;Chappell & Odell, 2004;Santostefano et al., 2019;White et al., 2016).  . Behaviouremetabolismecolour ornamentation relationships. Matrix of scatterplots of the raw data for each possible pair of (1) behavioural trait measures and resting metabolic rate (RMR) and (2) behavioural trait measures and colour ornamentation exhibited by test male guppies. The among-individual correlation coefficients (r I ) and corresponding 95% credible intervals (CIs, in brackets) were obtained from a multivariate generalized linear mixed-effects model fitted to the data that included the five repeatable behavioural trait measures, RMR and colour ornamentation scores as response variables, male body mass and repeat test number as fixed effects, and individual male identity (ID) as a random effect. Scatterplots with similarly coloured ellipses illustrate relationships between paired trait measures, one or both of which fall within the same behavioural functional category (red ¼ boldness; blue ¼ exploration; green ¼ sociability). The remainder of the caption is as shown for Fig. 2. More details are provided in the Appendix ,  Table A4.
The adjusted repeatability estimates obtained for guppy behavioural traits (R ¼ 0.488e0.800) in our current study were relatively high compared with reported mean repeatability estimates for behaviours in the literature (R ¼ 0.289e0.480;Bell et al., 2009;Dochtermann et al., 2015;Garamszegi et al., 2013;Holtmann et al., 2017;Wolak et al., 2012), whereas our adjusted repeatability estimates for metabolic traits (R ¼ 0.257e0.503) were similar to or lower than mean repeatability estimates reported for mean metabolic rates (R ¼ 0.451e0.458;Holtmann et al., 2017;White et al., 2013), in animals generally. In theory, the repeatability of a phenotypic trait reflects the upper bound of its heritability (Boake, 1989; but see Dochtermann et al., 2015;Dohm, 2002). Repeatability is linked to quantitative genetics theory through the potential contribution of additive genetic variation to among-individual trait variation (Dochtermann et al., 2015). Repeatability estimates for phenotypic traits can thus provide insights into their heritability, potential responsiveness to selection and evolvability Dingemanse et al., 2012;Dochtermann et al., 2015). The significant repeatability estimates for behavioural traits and maintenance metabolic rate (RMR) observed in our current study therefore suggest that these traits have underlying additive genetic variation and can potentially respond to selection and evolve in our guppy study population. Indeed, there is direct and indirect evidence for additive genetic variation underlying among-individual variation in personality  and mass-independent maintenance metabolic rate (Auer et al., 2018), and thus for their heritability, in the Trinidadian guppy.

Trait Measure (Co)Variation
Because individual male guppies were consistent in both their behaviour and maintenance metabolism in the current study, significant intercorrelations (covariation) between them at the among-individual level are possible (cf. Garamszegi et al., 2012).
First, with regards to personality, our measures of sociability, boldness and exploration were indeed strongly intercorrelated at the among-individual level and integrated in a proactiveereactive behavioural syndrome, as assumed by the POLS hypothesis (cf. R eale, Garant, et al., 2010). Proactive or risk-prone male guppies were bolder, more exploratory and more social compared to reactive or risk-averse males who were shyer, less exploratory and less social. Similarly, Irving and Brown (2013) previously observed a sociabilityeboldnesseactivity syndrome in male guppies from seemingly the same Australian guppy population as ours. Santostefano et al. (2019) and Smith and Blumstein (2010) reported positive correlations between boldness and exploration/activity in male Australian and Trinidadian guppies, respectively, but other studies did not find any evidence for such a behavioural syndrome in male Trinidadian guppies of different provenance (Diaz Pauli et al., 2015;Houslay et al., 2018;White et al., 2016). Evidence for behavioural syndromes in female guppies is also mixed (e.g. Diaz Pauli et al., 2015;Houslay et al., 2018;Irving & Brown, 2013;Kniel & Godin, 2019;Trompf & Brown, 2014;White et al., 2016). So, unsurprisingly, the existence and structure of behavioural syndromes in the guppy vary between the sexes and populations, as they do in other species (e.g. Dingemanse et al., 2007;H€ am€ al€ ainen et al., 2018Tarka et al., 2018), most likely owing to differences in selection on the sexes and to differences in selection regimes across environments/populations (e.g. Endler, 1995;Magurran, 2005), as well as to potential differences between studies in trait measures, how correlations between traits were estimated and experimental protocols.
The sociabilityeboldnesseexploration/activity syndrome in male guppies observed in our study population (current study; Irving & Brown, 2013) suggests that these three component behavioural traits are strongly linked phenotypically and might have coevolved through correlational selection if they also covary genetically (cf. Lande & Arnold, 1983), resulting in the emergence and maintenance of different personality types along a proactiveereactive behavioural continuum in the population. However, being strongly linked together in a syndrome, the plasticity of these three behavioural traits and their independent responses to selection would potentially be constrained (e.g. Sih et al., 2004). The relatively weak residual within-individual correlations in the behavioural traits comprising our observed behavioural syndrome suggests limited or constrained plasticity in these traits, at least under our experimental conditions (cf. Westneat et al., 2015). Nevertheless, it is plausible that this particular behavioural syndrome is adaptive and stable in our guppy study population, given that sociability, boldness and exploration/activity are associated with fitness-related benefits and costs (Barber & Dingemanse et al., 2010;Dingemanse et al., 2004;Dugatkin & Godin, 1992a;Krause & Ruxton, 2002;Moiron et al., 2020;Smith & Blumstein, 2008) and commonly positively associated with growth, fecundity and other life-history traits in general (Biro & Stamps, 2008Stamps, 2007; but see White et al., 2016 for an exception). The coexistence of different personality types in our study population might therefore reflect different fitness benefitecost trade-offs faced by individuals expressing different linked levels of sociability, boldness and exploration in the wild. For example, proactive or riskprone male guppies may be more likely to disperse and invade novel habitats (Cote et al., 2010;Deacon & Magurran, 2016;Myles-Gonzalez et al., 2015) and encounter and acquire more resources, such as food and mates (e.g. Dyer et al., 2009;Trompf & Brown, 2014;Godin & Dugatkin, 1996;Herdegen-Radwan, 2019), but at the cost of higher risks of mortality from predation and parasitism (Barber & Dingemanse, 2020;Dugatkin, 1992;Hulth en et al., 2017;Smith & Blumstein, 2008), compared with reactive or risk-averse individuals who may access fewer resources but at lower mortality risks.
Second, regarding the metabolism of male guppies in the current study, we could not test for among-individual correlations (i.e. for the presence of a metabolic syndrome) between RMR, MMR and AS using a multivariate modelling framework because only maintenance metabolism (RMR) was found to be significantly repeatable. Our relatively modest sample size (N ¼ 33 subjects) and minimum repeat test number (k ¼ 2) likely contributed to low power to detect repeatability estimates significantly different from zero for MMR and AS. Nevertheless, we did observe a strong negative phenotypic correlation between RMR and AS, suggesting that male guppies exhibiting a low RMR tended to have a greater absolute scope for metabolic aerobic activity than males with higher RMR, as previously shown for some other fish species (Metcalfe et al., 2016).

Covariation between personality and metabolism
Given that sociability, boldness and exploratory activity in vertebrates are generally associated with extrinsic mortality risk and energetic gains and costs (Biro & Stamps, 2008Mathot et al., 2019;Smith & Blumstein, 2008;Stamps, 2007), it can be logically assumed therefore that these behavioural traits in male guppies might be linked to their individual metabolism in some way (cf. Biro & Stamps, 2010).
Contrary to the central prediction of the POLS hypothesis (cf. R eale, Garant, et al., 2010), behaviour and maintenance metabolism were not significantly correlated among individual male guppies, after controlling for individual body mass, repeat test number and ambient temperature, in our current study. This negative finding corroborates the results of two other recent studies with guppies. White et al. (2016) also did not find any evidence for correlations between repeatable behavioural traits (exploratory activity, boldness) and metabolism (RMR, AS) at the among-individual level in female Trinidadian guppies. Likewise, Biro et al. (2016) did not observe among-individual covariation between stress-induced peak metabolism and behavioural traits (courtship, activity, foraging), but did report a significant negative correlation between peak metabolic rate and courtship display rate in male guppies originating from an Australian population different from ours. So, overall, there is no compelling evidence to date for a behaviouremetabolism syndrome in either male or female guppies, suggesting that individual differences in personality are not supported or driven by individual differences in RMR and that personality and maintenance metabolism can therefore potentially evolve independently of each other if there is no underlying genetic covariation.
Considering our findings more broadly, recent meta-analyses have revealed that the relationship between behaviour and metabolism at the among-individual level in animals in general is either significant, but only modestly so (Mathot et al., 2019;Niemel€ a & Dingemanse, 2018), or weak to nil (Royaut e et al., 2018). This within-and among-species heterogeneity in the behaviouremetabolism relationship is mainly owing to the lability of these two functional traits and the dependence of this relationship on several mediating factors. These mediators include context, ambient environmental conditions, study environment (laboratory versus wild), resource availability, sex, behaviour type, thermal type (ectothermy versus endothermy), trophic level, phylogeny, the energy management and life-history strategies of individual animals, experimental protocols (e.g. whether or not traits of individual subjects are measured repeatedly) and whether a particular behaviour contributes or costs net energy to individuals (reviewed in Auer et al., 2020;Biro & Stamps, 2010;H€ am€ al€ ainen et al., 2018Killen et al., 2016;Mathot et al., 2019;Mitchell & Houslay, 2021;Niemel€ a & Dingemanse, 2018;Royaut e et al., 2018;Tarka et al., 2018). So, within this broader context, the lack of evidence for a behaviouremetabolism syndrome in the guppy to date (current study; Biro et al., 2016;White et al., 2016) is not unusual.
There are at least two plausible alternative explanations for this negative result of our study. First, our feral study population of guppies is descended from ancestral guppies introduced to Australia from multiple endemic populations in either the western part of Trinidad or Guyana, South America about a century (ca. 200e300 generations) ago (Lindholm et al., 2005). It is very likely that these ancestral invaders experienced different selection regimes over time in their novel Australian habitats compared to their ancestral source habitats. This might plausibly have led to the breakup of locally adapted genetic architecture of ancestral polygenic traits over time in descendent Australian populations. Alternatively, but not exclusively, the ancestral guppies introduced to novel environments in Australia originated from different endemic populations in Trinidad and Guyana (Lindholm et al., 2005). There is genetic evidence for historical gene flow between the descendent, genetically differentiated guppy populations in Australia (Lindholm et al., 2005). Such admixture (mixing of genes from different populations) is common in species invading novel habitats and it can also disrupt the genetic architecture of polygenic traits of ancestral populations (e.g. Grant & Grant, 2008;Verhoeven et al., 2011). If either or both of the latter phenomena apply to our study population of Australian guppies, then we could expect reduced among-individual phenotypic and genetic correlations between polygenic traits such as behaviour, physiology and body coloration.
Second, our current study has a modest sample size (N ¼ 33 subjects) and minimum replicate measures (k ¼ 2) per trait, and our test guppies experienced a relatively homogeneous and benign laboratory environment compared to natural habitats. Given that multivariate linear mixed-effects models in particular are 'data hungry' (Hadfield, 2010), a larger sample size and/or larger repeat test number would have statistically increased the likelihood of observing significant among-individual correlations between behavioural and metabolic traits (cf. Wolak et al., 2012) and increased the robustness of our findings. Regarding the benign laboratory environment limitation, it is also plausible that the energy requirements for the expression of sociability, boldness and exploration by individual male guppies in our current study were minimal and not strongly linked to net energy gains or losses. The fish were being regularly fed at libitum daily in a controlled and stable laboratory environment that lacked natural ecological stressors such as predators, low dissolved oxygen levels and extreme variation in ambient temperatures, for example. If true, this might also explain our negative finding for amongindividual covariance between behaviour and metabolism. So, if male personality and maintenance metabolism do covary in the wild in our study population, then this association may have become uncoupled in guppies that were born and raised in a relatively stable and benign laboratory environment and tested under standardized laboratory conditions.

Covariation between personality, metabolism and colour ornamentation
In male guppies, at least orange body coloration is known to be condition dependent and a reliable indicator of male quality/ viability (reviewed in Houde, 1997;Magurran, 2005), as expected from indicator models of sexual selection (Andersson, 1994), and can be correlated with individual personality traits, such as boldness (Godin & Dugatkin, 1996; but see Ariyomo & Watt, 2013), that females prefer in potential mates. Contrary to our a priori hypothesis, neither the composite colour ornamentation of individual male guppies nor any of the univariate colour elements were significantly correlated with any of their behavioural traits or RMR at either the among-or within-individual levels in our current study. Similarly, metabolism is not correlated with body coloration (orange, black, fuzzy black) in Australian male guppies from a different Australian population than ours (Santostefano et al., 2019). These findings collectively suggest that body colour ornamentation in male guppies does not consistently predict their personality or metabolism, and that the underlying physiology associated with body coloration does not constrain the expression of personality and whole-organism metabolism in male guppies. Clearly more research is needed to comprehensively understand the relationships between body coloration, behaviour and metabolism in model vertebrate study systems such as the guppy.

Conclusion
We have demonstrated here that ecologically important behavioural traits in male guppies were repeatable and intercorrelated at the among-individual level and were thus integrated into a sociabilityeboldnesseexploration syndrome. However, personality and maintenance metabolism were not significantly correlated at the among-individual level, contrary to the POLS hypothesis. Male body colour ornamentation did not predict either personality or metabolism. Therefore, our current study does not offer any compelling evidence in support of the POLS hypothesis. Within the limitations of our study, we cautiously conclude that individual differences in maintenance metabolism do not drive or maintain differences in personality among males in our study population of the Australian guppy. As such, our findings contribute to further understanding the potential role of among-individual differences in physiology in the evolution and maintenance of multiple behavioural types within populations in general.

Data Availability
Data in support of our results are available from Dryad (Godin et al., 2022).

Declaration of Interest
None.

Table A1
Estimates of the among-individual (V I ) and residual within-individual (V R ) variances for each of the behavioural and metabolic trait variables obtained from univariate linear mixed-effects models ('lmer()' function) in the R package 'lme4' RMR: resting metabolic rate; MMR: maximal active metabolic rate; AS: metabolic aerobic scope. The body mass of individual male guppies (N ¼ 33) and repeat test number were included as fixed effects variables and male identity (ID) as a random effect in the models. Standard deviations of the variance estimates are shown in brackets. We used the 'lme4' package ('lmer()' function) to obtain estimates of the V I and V R because the 'rpt.Gaussian()' function in the R package 'rptR', which we used to estimate trait repeatabilities (R), does not directly output variance estimates. The 'rpt.Gaussian()' function calls 'lmer()' (LMM method) as its underlying mixed-model fitting engine (Stoffel et al., 2017).  Figure A1. Example of the movement track of test male guppy 13 (test 1) in the experimental novel environment during a 5 min trial produced by CTrax. The grid squares entered at least once by the test male are automatically identified by CTrax using an inverted green triangle, and the grid squares that have not been entered by the fish are identified by a red X. The dark circular objects inside the environment are sandstone blocks. Each univariate measure of colour is the relative area (proportion) of that colour class on the male's body. Correlation coefficients whose 95% CI overlap 0 are not statistically significant. Example regression plots of oxygen consumption rate during rest (resting metabolism, lower plot) and maximal sustained swimming (maximal active metabolism, upper plot) of a test male guppy. Each data point is an automated measurement of the dissolved oxygen concentration in the water of the respirometer containing the fish (see Methods for details). The slopes of the linear regression lines depicted represent the instantaneous rate of oxygen consumption. The difference between these two slopes represents the fish's metabolic scope for aerobic activity (AS). 0.06 0.08 0.1 0.12 0.14 0.16 MMR Figure A3. Bivariate regression plots of (a) resting metabolic rate (RMR) versus maximal active metabolism (MMR), (b) RMR versus metabolic aerobic scope (AS) and (c) MMR versus AS, obtained from linear mixed-effect models that included the body mass of individual male guppies and repeat test number as fixed effects. The metabolic rates shown on the axes are whole-animal rates of oxygen uptake (mmol O 2 /min).