Introduction

Metabolism and behavior are frontline responses to environmental challenges, and their variations among species, or among populations of the same species, are expected to reflect adaptations to local environmental conditions (Réale et al. 2010; Piersma and van Gils 2011; Goulet et al. 2017). Until recently, variations in metabolism and behavior have been foremost the topic of separated fields of research (i.e., eco-physiology vs. behavioral ecology). The relevance of such separate approaches is currently challenged by a growing number of theoretical studies suggesting that metabolic and behavioral traits are likely to covary and emphasizing that the mechanisms generating such covariation remain however little investigated (Careau et al. 2008; Biro and Stamps 2010; Réale et al. 2010; Careau and Garland 2012; Mathot and Dingemanse 2015; Sih et al. 2015; Holtmann et al. 2017). At least two biological mechanisms can account for covariations between metabolism and behavior among populations and individuals of the same species.

Firstly, there is increasing evidence that natural selection can generate convergence in phenotype among distinct populations of the same species exposed to a common selective force. This can occur through recurrent recruitments of the same genetic changes (Christin et al. 2010; Hague et al. 2017), through recurrent phenotypic changes fostered by phenotypic plasticity (Dennis et al. 2011) or through a combination of those two processes (Conover and Schultz 1995; Oke et al. 2016). Because variations in metabolism or behavior can be shaped by environmental and genetic factors (e.g., behavior: Dingemanse et al. 2002; Bell et al. 2009; Bize et al. 2012; metabolism: Nespolo and Franco 2007; McKechnie 2008; Nilsson et al. 2009), it predicts that metabolism and behavior can exhibit phenotypic correlations among populations when responding (independently from each other) to co-varying selective forces. This prediction can be illustrated through the following simple hypothetical example. Because ambient temperature decreases with an increase in elevation, in endotherms, selection along an altitudinal gradient may favor individuals with greater metabolic rate and capacity to produce heat either through genetic variation or developmental plasticity (Gillooly et al. 2001; Lovegrove 2003; see also Goulet et al. 2017). Furthermore, because food resources may often become rarer and patchier with an increase in elevation, selection might also favor animals with greater exploratory activity at higher elevation (e.g., Kramer and Weary 1991). This hypothetical and assuredly simplistic example would predict the occurrence of a strong positive phenotypic covariation between metabolism and behaviors among populations of endotherms sampled at different elevations. Of note, when considering phenotypic plasticity only, similar arguments can be applied further over distinct ontogenetic periods (growth vs. adulthood) and time periods (e.g., winter vs. summer, breeding vs. non-breeding) rather than spatially distinct populations, asking for careful considerations of both spatial and time co-varying selective forces when studying phenotypic covariations between metabolism and behavior (e.g., for an example of the effect of breeding on trait covariation, see Lantová et al. 2011).

Secondly, because by definition behavioral traits are energy-demanding, selection could also lead to the phenotypic correlation of behavior and metabolism among individuals sampled within the same population (Biro and Stamps 2010; Réale et al. 2010). Accordingly, Careau and collaborators (Careau et al. 2008) have recently suggested two different models for their correlation, using basal metabolic rate (BMR) as a benchmark of metabolism. In birds and mammals, BMR is the lowest measure of metabolic rate of an adult individual that is at rest during its normal period of inactivity, post-absorptivity, and non-reproductivity, and within its thermoneutral zone (McNab 1997). Thus, BMR is often viewed as the minimum energy cost of living (White and Seymour 2004; Speakman et al. 2004). The “performance model” points out that BMR is determined by the size of organs responsible for acquiring and processing food. Because active or aggressive individuals need high energy throughput, they would be expected to have larger-than-average organs, and thus higher-than-average BMR which in turn should lead to a positive phenotypic correlation between behavior and metabolism among individuals from the same population. Alternatively, the “allocation model” points out that, because resources are often limited in nature, the amount of energy allocated to behavioral traits can come at the expense of the amount of energy devoted to body maintenance, which in turn should lead to a negative correlation between BMR and behavioral traits among individuals from the same population. A recent review by Mathot and Dingemanse (2015) shows that empirical results are predominantly supporting the performance model.

In the present study, our aim was to test for phenotypic covariation in resting metabolic rate (RMR) and exploration activity (EA) at both the among- and within-population levels in adult common voles (Microtus arvalis) captured from different populations at low (< 520 m a.s.l.) and high elevations (> 1400 m a.s.l.). EA is important for collecting information on food abundance and predation risk, and therefore, EA is expected to be both under strong environment-specific natural selection and plastic to be able to respond to changes in the environment. Furthermore, as highlighted above, elevation provides a natural gradient that encompasses several ecologically relevant factors (e.g., temperature, humidity, predation risk, vegetation, nutrients). We defined two main objectives.

Our first objective was to describe the variation and covariation in RMR and EA with changes in elevation. In this study, we did not measure variation of ecological factors in relation to elevation nor their effects on RMR and EA; we had no a priori prediction on how RMR and EA should vary with elevation. As discussed before, for example, we can expect voles from higher elevation to have higher RMR if ambient temperature decreases with elevation and higher RMR facilitates the maintenance of a high body temperature in the cold (Hayes 1989). Alternatively, we can expect voles from higher elevation to have lower RMR if food becomes less abundant at higher elevation and a lower metabolism helps sparing energy (Selman et al. 2001). Alternative predictions can also be made for EA. For instance, voles from higher elevation may have higher (or lower) EA if food becomes less abundant at higher elevation and high EA helps them to discover new food patches (or lower EA helps them sparing energy) (e.g., Sears et al. 2009). The sign of the covariation between RMR and EA would depend on the specific effect of elevation of both variables.

Our second objective was to test for covariation between RMR and EA at two different scales: among populations sampled at different elevations (i.e., among-population level) and among individuals sampled from the same population (i.e., within-population level). If RMR and EA showed similar responses along the elevation gradient, populations from the same elevation levels should show phenotypic convergence (Dennis et al. 2011). The covariation between RMR and EA along the elevation gradient would exist only at the among-population level if it is due to independent co-varying selection pressures or phenotypic plasticity. However, if RMR and EA are genetically correlated (i.e., integrated, following Careau et al. 2008), then the covariation should also exist at the within-population level. The sign of the phenotypic correlation at the within-population level would indicate support for either the performance or the allocation model (Careau et al. 2008). Note that a phenotypic correlation at the within-population level may also be driven by correlated plasticity if for instance the two traits respond similarly to seasonal cues. Here, it is important to point that our phenotypic correlation is based on single, and not repeated, measures of RMR and EA per individual. Phenotypic correlations based on single measures need to be interpreted with care since they can be influenced by both among- and within-individual covariation, and thus, their value may misrepresent the true correlative value expected at the individual level when among- and within-individual variations are not identical (Niemelä and Dingemanse 2018).

Materials and methods

General methods

We used Longworth traps to capture 36 voles in four high-elevation populations (1428 to 1695 m a.s.l.) and 43 voles in five low-elevation populations (439 to 513 m a.s.l.) in cantons Vaud and Valais, Switzerland (Table 1). Live trapping of voles from high-elevation populations took place in mid-October 2010 and of voles from low-elevation populations in early-November 2010. Within 24 h after their capture in the field, voles were brought to the University of Lausanne (397 m a.s.l.), topically treated with the antiparasitic Ivomec© (Merial), and housed individually in polycarbonate cages (42.5 × 26.6 × 18.5 cm) in an animal facility room with a 14-h light:10-h dark cycle and a constant temperature of 22 ± 1 °C. Cages contained sawdust, hay, and a flower pot as cover. Water and food pellets were available ad libitum, and apples and endives were offered three times a week. Voles were acclimatized to the laboratory conditions during 30.8 ± 0.9 days (mean ± SE) before the measurement of their behavior.

Table 1 Identity of common vole populations along with elevation of the study sites, coordinates, number of animals measured (N), and means ± SE body mass before the metabolic measurements, resting metabolic rate (RMR) divided by body mass, and exploratory behavior (EB) measured as the number of lines crossed during 3 min of exploration in an open-field test. Populations were divided in low- (< 513) and high (> 1428 m)-elevation categories in the statistical analyses
Full size table

Behavioral measurements

We measured EA using an open-field test (OFT) made of a 40 × 40 × 30-cm squared Plexiglass™ box with its outside walls and floor covered with white paper and its ceiling with a dark sheet to stabilize the lightning conditions in the box. The testing room was adjacent to the animal facility room, and each vole was accustomed to handling by being carried under its flower pot (i.e., cover) for 1 min 2 days prior to the test. Animals were denied food for the hour preceding the test to standardize their nutritional status, and in turn their motivation to explore their novel environment. Voles were tested between 08:20 and 16:40. At the beginning of the test, each vole was placed under its pot in one corner of the open field box, the door of the pot being closed, and it was given 3 min to settle down. Then, the door of the flower pot was opened, and three additional minutes were waited before removing the pot and starting the behavioral recording, even if the vole left the pot. Eighteen of the 79 voles tested left their pot before it was removed (nine voles from each elevation groups). The behavior of each vole was video-recorded during 3 min with a digital camera fixed on top of the open field arena. We cleaned the open field box with 70% ethanol before each behavioral test. To obtain an estimate of EA, we divided the floor of the open field arena into 36 squares of 44.4 cm2, and for each individual, we counted the number of lines crossed during the 3-min-trial. Videos were analyzed by two undergraduate students blindly with respect to elevation; the identity of the analyzer did not affect the results (effects of observer identity on EA measurements: ANOVA: F1,77 = 0.26, P = 0.61).

Metabolic measurements

At the end of the OFT, voles were transferred back to the animal room facility and individually housed in polycarbonate cages (36.5 × 20.7 × 14.0 cm) which can be sealed with a lid for metabolic measurement purpose performed 1 day after their behavioral measurement. We measured resting metabolic rate via indirect calorimetry (O2 consumption [\( {V}_{O_2} \)] and C02 production [\( {V}_{{\mathrm{CO}}_2} \)]) using a SM-MARS-4 open flow system allowing the measurements of 3 animals in parallel (Sable Systems International, Las Vegas, USA) as previously described in Lehto Hürlimann et al. (2014). The food was removed 1 h prior to the measurements that were performed between 08:00 and 16:00. The common vole forages and feeds throughout the day in regular episodes spaced by ca. 150 min (Gerkema et al. 1993), and thus, our measurements spanned both active and inactive periods. The cages were placed individually on an activity detector (Sable Systems MAD-1) in a versatile environmental test chamber (MLR-350H; Sanyo, Japan). Measurements were conducted in the dark at a relative humidity of 50% and constant temperature of 30 ± 1 °C, which is within the thermoneutral zone for this species (Devevey et al. 2008). The air was pumped out of each cage with MFS-5 pumps with 1 L/min rate to the Multiplexer (MUX) where the air sample (500 mL/min) from only one cage at a time was pumped with subsampler (SS4) to the water vapor (RH-300)-, CO2 (CA-10)-, and O2-analyzers (FC-10) in this order. The raw data was analyzed using ExpeData software (Sable Systems International, Las Vegas, USA). One measuring cycle consisted of 2-min baseline recording (air from the climate chamber) in the beginning and at the end of the cycle to control for the baseline drift, and each cage being measured for 2-min period for 5×. Sample was taken every second and an average of 45 last samples of every 2-min period was used as single reading. In total, three measuring cycles (2-min measuring periods per cage for 15×) were recorded. The oxygen consumption (mL O2/h) was calculated according to the equation

$$ \mathrm{V}{\mathrm{O}}_2=\frac{\mathrm{F}\mathrm{R}\times \left({\mathrm{F}}_{\mathrm{i}}{\mathrm{O}}_2-{\mathrm{F}}_{\mathrm{e}}{\mathrm{O}}_2\right)-{\mathrm{F}}_{\mathrm{i}}{\mathrm{O}}_2\times \left({\mathrm{F}}_{\mathrm{e}}{\mathrm{CO}}_2-{\mathrm{F}}_{\mathrm{i}}{\mathrm{CO}}_2\right)}{1-{\mathrm{F}}_{\mathrm{i}}{\mathrm{O}}_2} $$

where FR = flow rate (mL/h), FiO2 = fractional concentration of O2 in incurrent air (baseline), FeO2 = fractional concentration of O2 in excurrent air, FiCO2 = fractional concentration of CO2 in incurrent air, FeCO2 = fractional concentration of CO2 in excurrent air. All values are corrected to standard temperature and pressure (STP) and corrected for water vapor pressure. As a measurement for RMR, we used the average of two lowest, consecutive readings when the voles were inactive (measured by the activity detector). If those criteria were not met, or the measurement failed because of technical reasons, the vole was measured again in 1 to 3 weeks and discarded if also the second trial did not meet the requirements. Measurements of four individuals (two from low elevation and two from high elevation) were discarded from the final analyses leading to final sample sizes of 34 and 41 voles from high- and low-elevation populations, respectively.

Statistical analyses

Firstly, we used univariate mixed models in “lme4” R package (Bates et al. 2013) to investigate which factors best explained phenotypic variation in RMR (mL O2/h) and EA (number of lines crossed during the OFT). In the models, we entered elevation (2 levels: low vs. high), sex (2 levels: male vs. female), body mass (continuous trait; log-transformed), number of acclimatization days in the laboratory (continuous trait), and time of the day at the start of the measure (continuous trait) as fixed factors and population identity as a random effect. The explanatory variables RMR and EA were, respectively, log-transformed and square root–transformed to normalize the distribution of the residuals. P values of type III F test for mixed models, with denominator degrees of freedom calculated using Satterthwaite’s approximation, were computed using the function anova in “lmerTest” R package (Kuznetsova et al. 2013).

Secondly, we used a bivariate mixed model using “MCMCglmm” R package (Hadfield 2010) to investigate covariation between RMR and EA at the among- and within-population levels. To this end, we entered log-transformed RMR and square root–transformed EA as response variables with Gaussian distributions. To avoid over-parametrizing our bivariate model, we only entered as fixed effects the factors identified as significant in the univariate mixed models described above. We entered population identity as a random effect on both traits. Given the structure of the data, only one observation per individual on each trait, the residual variance in both traits and the residual covariance are interpreted as the phenotypic (co)variance between RMR and EA within populations. The response and explanatory variables were scaled before analysis (mean centered on 0 and SD reduced to 1) to facilitate the interpretation of the results (Schielzeth 2010). We estimated the covariance between RMR and EA generated by elevation using two complementary approaches. First, the covariance between BMR and EA due to their covariation with elevation can be estimated as Cov(RMR, EA) = a × b × Var(elevation), where a and b are the regression coefficients of altitude on BMR and EA, respectively. Elevation was fitted as a 2-level factor and thus had a variance of 0.25. Secondly, we inspected the changes in the variance and covariance matrix caused by the inclusion or exclusion of elevation in the bivariate model. We used the following priors for the residual (V = diag(2), nu = 0.002) and random effect matrices (V = diag(2) × 0.002, nu = 1.002, alpha.mu = rep(0.2), alpha.V = diag(2)). To compute the posterior distribution, the model was run over 200,000 iterations, with a burn-in of 30,000 and a thinning interval of 100, to obtain an effective sample size between 1930 and 2185 with an autocorrelation level between retained iterations lower than 0.05. Parameter convergence and appropriate mixing of the chain were assessed visually for each parameter. We also tested for an overall phenotypic correlation between RMR and EA within the 9 different populations by performing a meta-correlation using the “meta” R package (Schwarzer 2007) on the 9 correlation coefficients computed within each population.

Results

Males were heavier than female voles (F1,76 = 22.09, P < 0.001), and there was no difference in body masses between high- and low-elevation voles (F1,76 = 0.007, P = 0.93) (mean ± SE body mass in grams for males vs. females in high-elevation populations, 26.1 ± 1.4 vs. 20.7 ± 1.2; males vs. females in low-elevation populations, 25.9 ± 0.9 vs. 20.5 ± 0.8).

Voles from high-elevation populations had significantly higher RMR than voles from low-elevation populations (F1,69 = 16.23, P < 0.001; Fig. 1a). RMR increased with log-transformed body mass (estimate ± SE = 0.834 ± 0.174, F1,69 = 22.63, P < 0.001) (Fig. 2) and decreased with the number of acclimatization days in the laboratory (estimate ± SE = − 0.014 ± 0.006, F1,69 = 5.60, P = 0.021). Sex and time of the day when measured did not explained significant variation in RMR (sex: F1,69 = 0.17, P = 0.69; time of the day: F1,69 = 0.08, P = 0.77).

Fig. 1
figure 1

Mean ± SE resting metabolic rate (RMR) (a) and exploration activity (EA) (b) of common voles from low- (< 513 m) and high (> 1428 m)-elevation populations

Full size image
Fig. 2
figure 2

Relationship between log-transformed resting metabolic rate (RMR) and log-transformed body mass in common voles issued from low- (< 513 m; open circles) and high (> 1428 m; closed circles)-elevation populations. The linear regression line (solid line) is presented with its 95% confidence interval (dashed lines)

Full size image

Individual variations in EA was best explained by elevation and number of acclimatization days in the laboratory. Individuals from high elevation were more explorative than those from low elevation (F1,73 = 7.75, P = 0.007; Fig. 1b). The level of exploration decreased with the number of days in the laboratory from their capture in the field to the behavioral test (estimate ± SE: − 0.18 ± 0.09; F1,73 = 4.19, P = 0.044). Effects of sex, body mass, and time of the day when measured did not explained significant variation in EA (sex: F1,73 = 1.58, P = 0.21; body mass: F1,73 = 0.07, P = 0.79; time of the day: F1,73 = 0.10, P = 0.76).

The use of a bivariate mixed model on scale values of RMR and EA values as response variables showed that elevation had a significant effect of similar magnitude on RMR (mean [95% credible interval]: − 1.08 [− 1.58; − 0.56]) and EA (− 1.28 [− 2.15; − 0.36]) (Table 2A), which has for consequence to generate phenotypic correlation in RMR and EA across populations of voles sampled at low versus high elevation of 0.28 [0.064; 0.658] (Fig. 3). Inspection of the variance components of models with vs. without elevation in the explanatory variables (Table 2A vs. 2B) provided similar findings. Indeed, when elevation was included in the explanatory variables, there was almost no variance left to be explained at the among-population level in both RMR (0.010 [1.02 × 10−9; 0.046]) and EA (0.017 [5.68 × 10−10, 0.079]) and no covariance between them (− 0.001 [− 0.027; 0.023]) (Table 2A). In contrast, when elevation was removed from the explanatory variables, it was possible to detect at the population level some variance in both RMR (0.085 [1.52 × 10−10; 0.342]) and EA (0.087 [3.26 × 10−8; 0.383]) and a positive (though non-significant) covariance between them (0.051 [− 0.068, 0.299]) (Table 2B). Finally, although both models showed significant phenotypic variance within populations (estimated as residual variance in the model) in RMR (model with vs. without elevation: 0.639 [0.445; 0.853] vs. 0.736 [0.494; 0.994]) and EA (0.952 [0.655, 1.305] vs. 1.021 [0.709, 1.384]), it provides no evidence of phenotypic covariation (0.005 [− 0.162; 0.195] vs. 0.077 [− 0.142; 0.292]) (Table 2). Examination of the phenotypic correlation coefficients between RMR and EA computed within each of the nine populations showed no consistent pattern. Correlation coefficients ranged from strongly negative to positive values (Fig. 4), which led to a non-significant overall phenotypic correlation of r = 0.14 [− 0.14; 0.40] (mean [95% CI]; raw data are shown on Fig. 1S in the Supplementary Material).

Table 2 Estimates of fixed effects and variance components for resting metabolic rate (RMR) and exploration activity (EA) in wild-captured common voles, obtained from bivariate mixed models. Panel (A) reports the estimates of fixed effects and variance components with elevation included in the explanatory variables, and panel (B) reports the variance components after removing elevation from the explanatory variables. Variation in response and explanatory variables were scaled before analysis to facilitate the interpretation of the estimates (Schielzeth 2010). Given the structure of the data, the residual variance and covariance are interpreted as the phenotypic (co)variance within populations. The table gives the mean posterior distribution and its 95% credible interval (CI)
Full size table
Fig. 3
figure 3

Covariation between resting metabolic rate (RMR) and exploration activity (EA) of common voles issued from low- (< 513 m) and high (> 1428 m)-elevation populations. The identity of each population is written next to its corresponding mean ± SE population value; information on each population is reported in Table 1

Full size image
Fig. 4
figure 4

Results of a meta-correlation of the within-population phenotypic correlations between resting metabolic rate (RMR) and exploration activity (EA) of common voles issued from five low (< 513 m)-elevation populations (BML, CSL, LSL, ROL, YVL) and four high (> 1428 m)-elevation populations (MOH, JOH, COH, LAH). Correlation coefficients (COR) are reported together with their 95% confidence intervals (95% CI) and sample sizes (N)

Full size image

Discussion

Our study shows that voles from high-elevation populations had higher resting metabolic rate (RMR) and higher exploration activity (EA) compared with their counterparts from low-elevation populations. Furthermore, the bivariate analyses between RMR and EA highlight that distinct populations exposed to a common selective force, here an elevation gradient, lead to a phenotypic correlation between RMR and EA at the among-population level. However, we found no evidence of a phenotypic correlation between RMR and EA among individuals from a same population, and thus no evidence of their phenotypic integration. This last result needs nonetheless to be interpreted with care since our phenotypic correlation relies on single measures per individual of RMR and EA, and this approach does not allow to tease apart the contribution of among- and within-individual variations at generating correlation between labile traits (Niemelä and Dingemanse 2018; see also the discussion below). Overall, our results suggest that the environmental correlation between RMR and EA is due to phenotypic plasticity and/or adaptation to co-varying selection pressures.

Altitudinal variation in RMR and EA

The clear differences in RMR and EA observed between voles from low- and high-elevation populations are opening two major questions. Firstly, are those differences fostered by genetic effects and/or phenotypic plasticity, and secondly what is/are the natural selective force(s) driving those differences. With our design, we cannot say if the observed differences in RMR and EA were driven by independent genetic adaptations to elevation due to co-varying selection pressures and/or by phenotypic adaptations of similar genotypes exposed to low- and high-elevation environmental conditions (i.e., phenotypic plasticity). Our results suggest nonetheless that both processes might be important. Pilot results on a small subsample of voles from this study that were measured twice during their time in captivity showed that RMR and EA were significantly repeatable (r = 0.58 and n = 23 individuals for RMR, and r = 0.61 and n = 17 for EA, see Supplementary Material for details). Because repeatability establishes the upper limit for heritability (Lessells and Boag 1987; Falconer and Mackay 1996; Ronning et al. 2005; but see Dohm 2002), our results highlight plausible additive genetic effects on variations in RMR and EA in the common vole. Our findings are also pointing toward plastic responses in RMR and EA. Both traits were decreasing with the time spent in the laboratory, which indicates that voles were acclimatizing to their new laboratory conditions. When brought from the field to the laboratory, obvious major changes in their environment were higher ambient temperatures (22° in the laboratory vs. 8.9 °C or 2.5 °C for the elevations 455 m or 1974 m, respectively, based on mean temperatures of October–November for locations Pully and Le Moléson in years 1981–2010; Federal office for meteorology and climatology MeteoSwiss, www.meteoswiss.ch), ad libitum access to food and water, no predators, and smaller living space. Common garden experiments where animals from different elevations are housed, for instance, at different ambient temperatures are now required to address the importance of genetics and adaptive plasticity in shaping the RMR and EA of adult common voles (for an example of such approach, see Tsuchiya et al. 2012).

Many ecologically relevant factors vary along the elevation (just as from the field to the laboratory) and can affect RMR and EA. One of the best studied factors is the ambient temperature, with rodents issued from higher elevations, but also from cooler environments, showing higher BMR or RMR (Hayes 1989; Lovegrove 2003; Russell and Chappell 2007; but see Hammond et al. 1999; Rezende et al. 2004). Accordingly, we found that common voles from high-elevation populations had higher RMR than those from low-elevation populations, and that RMR declined when acclimatized to laboratory conditions (22 °C). Those findings are consistent with the idea that the altitudinal gradient can shape RMR through greater need for endotherms to produce heat at higher elevations/cooler temperatures (Rodríguez-Serrano and Bozinovic 2009). Interestingly, because greater energetic for heat production can lead to greater food intake, as observed in rodents exposed to cold or acclimatized to high elevations (e.g., Selman et al. 2008; Hammond et al. 1999), the same ecological factors might also favor greater activity and exploration activity in the quest for food (Sears et al. 2006, 2009). Although we do not have information about food abundance and predation risk in our study populations, they are also potential factors to vary along an elevation gradient. In higher elevations, food resources might be scarcer, which could lead to higher levels of EA (Kramer and Weary 1991). As exploration increases the susceptibility to predators, between population differences in predation risk could also select for different optimal levels of EA (Herczeg et al. 2009; Bergeron et al. 2013).

Phenotypic correlation among- and within-populations between RMR and EA

Because natural selection is often impacting more than one trait at a time, one first important consequence is that phenotypic convergence of alternative, isolated traits can lead to their phenotypic correlation among populations. A second important consequence of selection for trait combinations is that this might also lead to the evolution of a functional integration among traits at the level of the individual, and ultimately to the existence of genetic correlations between these traits at the level of the population (Duckworth and Badyaev 2007; Versteegh et al. 2012). In agreement with the idea that phenotypic convergence of alternative traits in response to natural selection can lead to their phenotypic correlation among populations, the strong positive effects of elevation on RMA and EA led to positive correlation among them of 0.28 [0.064; 0.658]. This result was further supported by changes in the covariance matrix of the bivariate models where a positive phenotypic correlation, though non-significant, between RMR and EA was only becoming apparent when elevation was removed from the models. The lack of a significant covariance at the population level in the second approach, despite strong effects of elevation on both traits, is most likely explained by low statistical power coming from the fact that only nine populations from two elevations were sampled. Our results in voles are in line with recent findings in western stutter-trilling cricket (Gryllus integer) suggesting that correlations between physiological, behavioral, and life history traits could have occurred due to environmental factors rather than due to genetic constraints (Niemelä et al. 2013; see also Krams et al. 2017).

It has also been suggested that selection on energy acquisition or allocation can lead to either positive or negative correlations between RMR and EA at the individual level (Careau et al. 2008; Mathot and Dingemanse 2015). Results from the bivariate models and the meta-correlation provided, however, no evidence of consistent phenotypic correlations within-populations between RMR and EA: the more explorative individuals from a population had neither a higher RMR nor a lower RMR than the less explorative ones. Our results are nonetheless based on single measures per individual, and work using repeated measures of RMR and EA from the same individuals (Boulton et al. 2015), repeated measures over multiple generations from the same families (i.e., pedigree approach; Careau et al. 2011), or experimental selection on RMR or EA (Vaanholt et al. 2007; Gebczynski and Konarzewski 2009; Careau et al. 2010) is required to adequately test integration between RMR and EA at the individual level. Such studies remain rare and are currently providing contrasting results. For instance, the use of long-term pedigree approach in wild-derived deer mice (Peromyscus maniculatus) showed a positive genetic correlation, but without a phenotypic correlation, between RMR and EA (Careau et al. 2011). Artificial selection experiments on laboratory mice showed that selection for high BMR led to higher activity (Gebczynski and Konarzewski 2009) whereas selection for higher activity led to lower RMR (Vaanholt et al. 2007). In contrast, dog breeds selected for high aggressiveness have higher energetic needs (Careau et al. 2010).

Conclusion

Links between physiological, behavioral, and life-history traits are gaining increasing interest among evolutionary biologist as they are central for understanding evolutionary potential and limitations of populations and species (Careau et al. 2008; Biro and Stamps 2010; Ketterson et al. 2009; Réale et al. 2010; Careau and Garland 2012; Mathot and Dingemanse 2015; Sih et al. 2015). An important step is to understand the relative importance of genetics and environment factors in shaping complex (multivariate) phenotypes (Swallow and Garland 2005). In our study, we could show, by comparing among- and within-population correlations, that RMR and EA can be linked by independent responses to co-varying selection environmental forces. Future studies on correlations between complex phenotypes should take into consideration the effect of environmental factors shaping the traits alone and correlations between them.