Selective breeding for high voluntary exercise in mice increases maximal (V̇_O2,max) but not basal metabolic rate

Metabolic rate, i.e. the rate at which organisms acquire and expend energy, is a fundamental aspect of an animal's physiology, forming a link from the first principles of physics and chemistry to the biology of individual organisms (Brown et al., 2004; Lovegrove, 2019). For vertebrates, two key boundaries of animal energetics are the maximal rate of oxygen consumption attained during exercise (V̇_O2,max) and the resting (RMR) or basal (BMR) metabolic rate (usually also measured as O₂ consumption) (Hulbert and Else, 2004). In general, these extremes set the upper and lower bounds for energy expenditure (although exceptions exist, e.g. cold-induced summit metabolism in small mammals can exceed V̇_O2,max; Chappell and Hammond, 2004; Andrew et al., 2019), and some (e.g. Biro et al., 2018) have argued that the difference between these bounds (i.e. aerobic scope) may constrain variability in the expression of some behaviors. A recent meta-analysis found that various measures of whole-animal metabolic rate (including resting and maximal metabolic rates) were not strongly related to aspects of movement behavior (e.g. activity in familiar environments, exploration of novel environments, dispersal) at the level of individual variation (Wu and Seebacher, 2022). From a macroevolutionary perspective, Boratyński (2020) compared 52 species of mammals and found that home range size (corrected for body size) was positively correlated with V̇_O2,max but negatively correlated with BMR, thus suggesting that ‘aerobic scope plays a prominent role in constraining home ranges’ (Boratyński, 2020, p. 468; see also Albuquerque et al., 2015). A potential link between metabolic rate and behavior underlies the aerobic capacity model for the evolution of vertebrate endothermy, which posits that directional selection favored high levels of sustained aerobic physical activity, which required an increase in V̇_O2,max and, because of unspecified linkages with V̇_O2,max, also increased BMR (Bennett and Ruben, 1979; Taigen, 1983; Hayes and Garland, 1995).

On first principles, maximal and resting rates of O₂ consumption should be positively correlated, given that the two metabolic states share many organ systems (e.g. cardiovascular), cellular properties (e.g. mitochondrial density) and metabolic pathways (Bennett and Ruben, 1979). First principles do not, however, support a 1:1 correlation between V̇_O2,max and BMR (as pointed out by Taigen, 1983), given that different tissues account for the bulk of O₂ consumption at rest versus during activity. During sustained activity, skeletal muscle is responsible for the majority of O₂ consumption, but has a relatively low metabolic rate when an animal is at rest (Weibel et al., 2004). At rest, O₂ is mainly consumed by visceral organs and the brain (Konarzewski and Diamond, 1995; Książek et al., 2004; Konarzewski and Książek, 2013), whose collective metabolic rates are relatively low during sustained, aerobically supported activity. Nevertheless, the brain and some visceral organs (e.g. heart, liver) may remain quite active during periods of activity and have been consistently correlated with V̇_O2,max and/or BMR at the level of individual variation (e.g. Garland, 1984; Konarzewski and Diamond, 1995; Chappell et al., 1999; Książek et al., 2004; Rezende et al., 2006b; Gębczyński and Konarzewski, 2009; Konarzewski and Książek, 2013), which should result in some degree of positive correlation between the two metabolic states.

Alternatively, a mechanistic link between V̇_O2,max and BMR may stem from cellular properties. Mitochondria consume O₂ and produce ATP via a series of protein complexes embedded in their inner membrane. This inner membrane can be ‘leaky’, decoupling O₂ consumption and ATP production. This leakiness is a major contributor to BMR (Else and Hulbert, 1987; Else et al., 2004). Therefore, higher mitochondrial densities could provide the capacity for higher rates of O₂ consumption, but at the cost of a higher resting rate of O₂ consumption (Else and Hulbert, 1981; Hulbert and Else, 1989; Hulbert et al., 2006), although the general application of this as a unifying explanation is debated (Konarzewski and Książek, 2013).

Empirical studies have tested for a positive correlation between V̇_O2,max and BMR at several levels. For example, an allometric comparison indicated that both V̇_O2,max and BMR are on average ∼6 times higher in mammals as compared with lizards, with the ratio of V̇_O2,max/BMR or SMR being ∼9 for both lineages (see table 1 of Garland and Albuquerque, 2017). Among species within lineages, a comprehensive comparative analysis of 176 vertebrate species (including fish, amphibians, reptiles, birds and mammals) found a positive correlation between residual V̇_O2,max and BMR or SMR (Nespolo et al., 2017). Among individuals within species, Pough and Andrews (1984) found no correlation between residual exercise V̇_O2 and either standard or resting rates of O₂ consumption in the lizard Chalcides ocellatus (see also Garland, 1984), whereas Chappell and Bachman (1995) reported a significant positive correlation between residual V̇_O2,max and BMR in the wild rodent Spermophilus beldingi. With respect to quantitative genetics, Dohm et al. (2001) reported a positive additive genetic covariance between residual V̇_O2,max and BMR in an outbred strain of laboratory house mice, but only in a reduced model containing additive and environmental variance. Hence, the authors advocated that their results be interpreted with a degree of caution, given potential bias resulting from imposed modelling constraints. Similarly, Sadowska et al. (2005) found that the additive genetic covariance between V̇_O2,max and BMR in bank voles was also positive, and was significant across several models (i.e. potentially more robust). Overall, interspecies comparisons generally report a positive correlation between maximal and resting rates of O₂ consumption, whereas comparisons within species are less consistent (Auer et al., 2017; Nespolo et al., 2017).

One fairly direct way to test for correlations among physical activity behavior, whole-animal metabolic rates and lower-level traits is through replicated selection experiments, which allow the study of evolution in real time and in response to well-defined and reproducible selective regimes (Swallow and Garland, 2005; Swallow et al., 2009; Storz et al., 2015). Several selection experiments have tested for a positive relationship between V̇_O2,max, BMR and lower-level traits. For example, Książek et al. (2004) selectively bred two lines of laboratory mice for high versus low mass-corrected BMR. Starting at generation 7, the between-line difference in mass-corrected BMR increased, and after 19 generations of divergent selection, the between-line difference in BMR was 8.9 ml O₂ h⁻¹, equivalent to ∼2.3 phenotypic standard deviations. This increase in BMR (+18%) was accompanied by a larger small intestine, liver, kidneys and heart in the high-BMR mice. However, the low-BMR mice had significantly higher (+4%) V̇_O2,max (elicited by forced swimming) than those from the high BMR group, contradicting the idea that V̇_O2,max and BMR are positively related (Książek et al., 2004). Similarly, Gębczyński and Konarzewski (2009) found that 10 generations of selection for high body mass-corrected V̇_O2,max (elicited by forced swimming) in laboratory mice resulted in a 12% increase in V̇_O2,max, but no change in BMR. Additionally, V̇_O2,max was positively correlated with higher masses of gastrocnemius muscles and heart, but not other visceral organs (intestine, stomach, liver and kidneys). Using a colony of wild-derived bank voles (Myodes glareolus), Sadowska et al. (2015) conducted a multiway artificial selection experiment meant to mimic an adaptive radiation, with four lines each bred for high aerobic metabolism during forced swimming, predatory behavior on crickets, or the ability to maintain body mass on a low-quality plant diet. After 11 generations, mass-corrected V̇_O2,max and BMR were both significantly higher in the four swimming-selected lines as compared with four non-selected control lines, although the magnitude of these increases differed greatly (+49% in V̇_O2,max, +7.3% in BMR). Finally, Wone et al. (2015) bred four replicate lines of laboratory house mice for high mass-independent V̇_O2,max during forced treadmill exercise, four antagonistically selected lines for high V̇_O2,max and low BMR, and maintained four non-selected controls. After eight generations, V̇_O2,max significantly increased (+11%) in lines bred for high V̇_O2,max, while BMR had not significantly changed (+2.5%). In the antagonistically selected lines, V̇_O2,max increased (+5.3%) while BMR decreased (−4.2%, not statistically significant), which, while it does not falsify the notion that V̇_O2,max and BMR may be linked, provides support for the independent evolution of the metabolic traits.

None of the aforementioned selection experiments directly tested the specific scenario proposed by the aerobic capacity model (Bennett and Ruben, 1979), which has been more broadly interpreted as suggesting a fundamental link between physical activity behavior, V̇_O2,max and BMR that is of general applicability to vertebrates (Taigen, 1983; Hayes and Garland, 1995). We addressed this scenario with a well-established mouse model in which four replicate High Runner (HR) lines bred for high voluntary wheel-running behavior were compared with four non-selected control lines (Swallow et al., 1998a; Garland, 2003; Wallace and Garland, 2016). In a sample of females from generations 88 and 96, we measured maximal and basal rates of O₂ consumption, and recorded the mass of the kidneys, spleen, liver, brain, heart (ventricles) and lungs. We hypothesized that: (1) selection for high voluntary exercise would have resulted in an increased V̇_O2,max for HR mice, (2) that HR mice would also have an increased BMR, and (3) that the mass of some organs (e.g. brain, heart, liver), which are quite active during aerobic exercise as well as under basal conditions, would be increased in HR mice. Although several previous studies have reported elevated V̇_O2,max in the HR lines (e.g. Rezende et al., 2005, 2006a,b; Kolb et al., 2010; Dlugosz et al., 2013b), only one previous study reported BMR (of aged individuals), finding no statistical difference between the HR and control lines (Kane et al., 2008).

For logistical reasons, we sampled from two generations of an ongoing selection experiment for high voluntary wheel-running behavior (Swallow et al., 1998a; Garland, 2003; Careau et al., 2013; Wallace and Garland, 2016): 50 females from generation 88 and 22 from generation 96. Only females were used because: (1) the number of mice per day that could be tested for V̇_O2,max and BMR was limited, such that a smaller sample size or lengthier testing period was required; and (2) females can be housed four per cage as adults, whereas males often need to be individually housed to prevent fighting. The delay between generations was approximately 2 years and was primarily the result of COVID-19-related restrictions on personnel and research.

Individuals were measured for V̇_O2,max, BMR and organ mass, but were not exposed to wheels at any time; thus, they represent ‘baseline’ or untrained conditions. Additionally, we collected wheel-running data from siblings that were part of the routine selective breeding procedures. For each generation of the selection experiment, mice are housed four per cage by sex from weaning (21 days of age) until ∼6–8 weeks of age, when they are housed individually with access to an activity wheel (1.12 m circumference). Over 6 days, wheel revolutions are recorded in 1 min intervals. For the four replicate HR lines, the highest-running male and female from each family are chosen as breeders for the next generation, with no sibling pairings allowed. For the four replicate control lines, breeders are chosen without regard to wheel running (Swallow et al., 1998a; Careau et al., 2013). Animals were maintained in accordance with NIH guidelines, and all procedures were approved by the IACUC of University of California, Riverside, which is accredited by AAALAC.

The original base population of mice used to start the selection experiment included individuals with hindlimb muscles that were ∼50% smaller than normal-muscle individuals (Garland et al., 2002; Houle-Leroy et al., 2003). This ‘mini-muscle’ phenotype is caused by a single nucleotide polymorphism that acts as a Mendelian recessive and was present at a frequency of ∼7% in the base population (Kelly et al., 2013). The phenotype was only ever observed in one control line and in two HR lines. The phenotype eventually disappeared from the control line, became fixed in HR line 3, and remains polymorphic in HR line 6 (Hiramatsu et al., 2017; Cadney et al., 2021; Castro et al., 2022). Of the 72 mice used here, all 14 in HR line 3 had the mini-muscle phenotype (as expected) and 5 of the 22 mice in HR line 6 had the mini-muscle phenotype.

Although the mice used within this study never had access to a running wheel, we did have the wheel-running data from their siblings, which were part of the overall selection experiment. Briefly, mice are housed with access to an exercise wheel (1.12 m circumference) for 6 days. During this period, we recorded the wheel revolutions in each 1 min interval over a period of 23 h. We then computed the total number of revolutions (i.e. daily running distance), the number of 1 min intervals that had at least one revolution (i.e. minutes of wheel activity), the mean revolutions per minute (i.e. average running speed), and the maximum revolutions per minute (i.e. maximum running speed) (Koteja and Garland, 2001; Hiramatsu and Garland, 2018). A measure of wheel freeness was also included as a covariate in analyses of wheel running (e.g. Girard et al., 2007; Kolb et al., 2010).

V̇_O2,max was measured in an enclosed wheel metabolic chamber (effective volume 900 ml; ∼15 cm diameter), as described in Dlugosz et al. (2013a). Briefly, an upstream mass flow controller set incurrent air flow to ∼2000 ml min⁻¹. Excurrent air was subsampled at ∼150 ml min⁻¹, scrubbed of H₂O and CO₂ by Drierite and soda lime, respectively, and directed to an O₂ sensor. Data from an O₂ analyzer (Applied Electrochemistry Inc., S-3A) were recorded in 1 s intervals on a computer equipped with a National Instruments A–D converter and LabHelper software (M. A. Chappell, Warthog Systems, https://warthog.ucr.edu/).

Each mouse was tested twice, with a day of rest between trials. The repeatability of V̇_O2 achieved in forced running trials was determined by performing a paired Student's t-test using the raw V̇_O2 values, which tests for differences in the average values from one day to the next. In addition, after regressing the V̇_O2 of each trial on its corresponding body mass and age, we performed a paired Student's t-test of the residuals. Pearson's correlation and the associated t- and P-values for these tests are reported below. Mice averaged 58 days of age (range 50–66 days) at the start of testing were randomized with regard to time of day and testing order, and all tests were performed at 22–25°C during the photophase. Each test was less than 10 min and consisted of: (1) 1 min reference reading of incurrent air at the start; (2) ∼1 min adjustment period after the mouse was placed in the chamber; (3) testing period wherein (a) the wheel was manually propelled by one of two researchers across all measurements, (b) the initial speed was used to elicit a walking pace from the mouse, and (c) the researcher accelerated the wheel (by hand) approximately every 30 s until V̇_O2 did not increase for ∼3 min or the mouse could not continue running; (4) ∼1 min recovery period before removal from chamber; and (5) 1 min reference reading. The same protocol was applied to all mice in this study, and so any measurement error should be comparable across individuals. Values reported here are similar to those previously reported for these mice when using a treadmill (Rezende et al., 2005; Kolb et al., 2010) or the wheel apparatus (e.g. Claghorn et al., 2017; Cadney et al., 2023).

After every trial, each mouse was given an objective score of exhaustion, indicated by the number of seconds after the trial before the mouse began walking again. These data were analyzed on a scale of 1 to 5, where an exhaustion of 1 indicated that 1 s had elapsed and an exhaustion of 5 indicated that 5 or more seconds had elapsed. Additionally, each mouse was given a subjective score of overall cooperativity, indicated by whether the mouse consistently ran with the direction of wheel rotation, or sometimes ran in the opposite direction. This was also on a scale of 1 to 5, where a cooperativity score of 1 indicated the mouse did not run when prompted, and a cooperativity score of 5 indicated the mouse attempted to run even at the highest speeds. In cases of uncooperative mice (e.g. cooperativity scores of 1 or 2) or technical difficulties, a third trial was conducted (N=16) and used to replace the poor trial.

BMR was determined by measuring O₂ consumption at rest in postabsorptive mice at ∼32°C (within their thermal neutral zone: Lacy and Lynch, 1979). The setup for recording BMR was similar to that for V̇_O2,max, except incurrent air flow was ∼500 ml min⁻¹, mice were in plastic respiration chambers (10 cm×7.5 cm×7.5 cm), and excurrent air was subsampled at ∼100 ml min⁻¹.

Mice were separated into two groups, and tests began at either 12:00 h or 16:00 h. Mice averaged 68 days of age (range 57–80 days) at the start of testing. Food was removed 4 h prior to testing, which is adequate for obtaining a postabsorptive state in mice (Jensen et al., 2013). Mice were tested over a period of 4 h, wherein excurrent air was subsampled for 45 min, then incurrent air was subsampled for 15 min. Two mice were measured at a time using separate channels (Oxzilla, Sable Systems International). Four mice were tested each day, over a period of 13 days. Analysis of BMR was the same as for V̇_O2,max, except data were recorded in 2 s intervals and the lowest 5 min continuous average was used. For each of the lowest values, we verified that the trace was stable, thus indicating that animals were at rest.

Mice were euthanized by decapitation without anesthesia (average age 76 days, range 70–81 days) and blood samples were immediately collected from the trunk via heparinized micro-hematocrit tubes, then spun in a micro-hematocrit centrifuge for 5 min. Approximately four samples were collected for each mouse and readings were averaged. The whole brain, heart ventricles, kidneys, liver, lungs, spleen and triceps surae muscle group were each collected and weighed.

All mice were weighed at weaning, before V̇_O2,max, before and after BMR trials (average used), and before dissection. Body composition was measured by non-invasive quantitative magnetic resonance (EchoMRI-100, Echo Medical Systems LLC, Houston, TX, USA), which independently calculated fat and lean mass, after the first V̇_O2,max trial, after the BMR trial and before dissection.

Statistical analyses were performed using SAS Proc Mixed v15 (SAS Institute, Cary, NC, USA). HR and control lines were compared by a mixed model, using the restricted maximum likelihood (REML) method, with line type and mini-muscle status as fixed effects. Replicate line (4 HR and 4 control) was nested within line type as a random effect using the containment method for d.f., such that the d.f. for line type were always 1 and 6. We tested the significance of the random effect of the replicate lines using the COVTEST option. This yields the estimates, standard errors and statistical significance of any covariance parameters, which were restricted to non-negative covariance estimates. The effect of replicate line was never statistically significant for any measured trait (Tables 2 and 4). For wheel-running traits, separate variances were allowed for HR and control lines, as previous studies have established a greater variability in wheel-running behavior among HR lines (Garland et al., 2011). We also checked for any interactions between body mass and line type for all measured traits (i.e. heterogeneity of slopes). None were statistically significant, and thus were not included in the final model (i.e. slopes for body mass were assumed to be homogeneous). Generation was included as a random effect in preliminary analyses, but was never statistically significant, and thus was removed from final analyses. Additionally, some values were removed because of known problems (e.g. loss of tissue during dissection, equipment malfunction) prior to analysis. Outliers were removed when the standardized residual was greater than ∼3 standard deviations and/or the difference from the next value was greater than ∼1 standard deviation. Least squares means (LSMs) and associated standard errors are presented to compare HR with control lines and mini-muscle versus normal mice.

Correlations between V̇_O2,max, BMR and relevant organ masses were calculated in two ways: (1) for individual mice, using the standardized residuals for each trait, derived from each of the SAS Proc MIXED analyses with line type and mini-muscle status as fixed effects, and line nested within line type as a random effect; (2) for line means, using the LSMs derived from SAS Proc MIXED analyses with ‘line’ (N=9, separating mini- and normal-muscle mice in Line HR6) as a fixed effect. Finally, a multiple regression analysis was performed (list-wise deletion of data; P to enter=0.05) to test for combined predictors of V̇_O2,max.

To avoid any training effect on the comparison of V̇_O2,max, BMR or organ mass, focal mice did not receive wheel access. However, for their female siblings, HR mice ran ∼3-fold more revolutions per day than control mice (Fig. 1A; Table S1). This increase in wheel-running behavior was caused primarily by a significant increase in average running speed (+142%: Fig. 1C; Table S1), accompanied by a non-significant increase in running duration (+27%: Fig. 1B; Table S1). Additionally, maximum running speed was significantly higher in HR mice (+94%: Fig. 1D; Table S1).

HR mice did not significantly differ from control mice for whole-body, lean or lean-adjusted fat mass when measured at V̇_O2,max, BMR or dissection, although HR mice did have consistently less lean-adjusted fat mass at each measurement (Tables 1–4). Mini-muscle mice were significantly smaller when measured at V̇_O2,max, BMR and dissection, as a result of decreased lean mass, but also had increased lean-adjusted fat mass at V̇_O2,max, BMR and dissection (Tables 1–4).

The V̇_O2 achieved in forced running trials was repeatable (r=0.821, P<0.0001), although the second of the V̇_O2 trials was consistently higher than the first (t=2.554, P=0.013). Additionally, after regressing each value on its respective body mass, residual V̇_O2 was also repeatable (r=0.751, P<0.0001).

The mass-corrected V̇_O2,max (higher of the two V̇_O2 values) of HR mice was ∼13.6% higher than that of control mice (Fig. 2A, Tables 1 and 2). However, mass-corrected BMR did not significantly differ between HR and control mice, although HR mice had +6.5% higher mass-corrected BMR than control mice (Fig. 2C, Tables 1 and 2). Mini-muscle status did not significantly affect either V̇_O2,max or BMR (Fig. 2A,C, Tables 1 and 2). However, when lean mass was used as a covariate (see above), V̇_O2,max was ∼4.6% higher in mini-muscle mice (Fig. 2B, Tables 1 and 2). Mini-muscle mice were also significantly more exhausted after V̇_O2,max (Tables 1 and 2).

HR mice did not significantly differ from control mice for any mass-adjusted organ mass, nor did they differ in hematocrit; however, they tended to have smaller lungs than control mice (Fig. S1; Tables 3 and 4). Mini-muscle mice had ∼50% less hindlimb muscle mass, as expected (Garland et al., 2002; Houle-Leroy et al., 2003); they also had significantly larger livers and lungs, and tended to have larger kidneys and spleens (Fig. S1; Tables 3 and 4).

V̇_O2,max and BMR were not significantly correlated at the level of residual (individual) variation or for line means (with HR line 6 split into normal and mini-muscle individuals, i.e. based on nine values) (Table 5). However, residual V̇_O2,max and ventricle mass were significantly positively correlated among all individuals and for line means (Table 5). BMR and ventricle mass were significantly positively correlated among line means, but not among individuals (Table 5). Mass-corrected V̇_O2,max and BMR were not significantly correlated with any other lower-level trait (Table 5). In a forward regression analysis (listwise deletion of data; P to enter=0.05), only residual ventricle mass entered (N=65). Correlations among organ masses are presented in Table S2.

Several alternative, though not necessarily mutually exclusive, hypotheses have been proposed to explain the often-observed positive relationship between V̇_O2,max and BMR, at the level of proximate and/or ultimate causation. The aerobic capacity model (Bennett and Ruben, 1979) suggests that, with respect to ultimate causation, selection for high levels of sustained aerobic activity would require an increase in V̇_O2,max, and that an increase in BMR would also occur as a result of hypothetical mechanistic linkages (proximate causation). Although originally proposed in the context of the evolution of avian and mammalian endothermy, this model might also apply more generally. As outlined in the Introduction, empirical studies have provided mixed support for this model at several levels. However, a direct test of the primary assertions of the aerobic capacity model has not previously been conducted.

Here, we used an ongoing artificial selection experiment wherein mice were bred for high voluntary exercise behavior during days 5 and 6 of a 6 day exposure to wheels (Swallow et al., 1998a; Garland, 2003; Wallace and Garland, 2016). After 10 generations of selection, mice from the HR lines ran, on average, ∼75% more revolutions per day than those from the control lines, and had ∼7% higher body mass-corrected V̇_O2,max (Swallow et al., 1998a,b), though BMR was not measured. In later generations, HR mice reached a selection limit at which they ran approximately 3-fold more than control mice on a daily basis (Careau et al., 2013), which has remained true across tens of generations (e.g. Singleton and Garland, 2019; Cadney et al., 2021; McNamara et al., 2022; present study). HR mice also had higher activity in home cages when housed individually without wheels (Malisch et al., 2009; Copes et al., 2015), and higher food consumption, both with and without wheels (Copes et al., 2015; see also Rezende et al., 2009), as compared with control mice. Several additional studies have reported V̇_O2,max, and most have verified higher values for HR lines (Rezende et al., 2005; 2006a,b; Kolb et al., 2010; Dlugosz et al., 2013b). However, the only study of BMR found no significant effect of selection, did not measure V̇_O2,max or any organ masses, and used mice that were far older (∼22.5 months) than the normative wheel-testing age (∼2 months) for the selection experiment (Kane et al., 2008). Thus, more information was needed to determine whether the HR mouse model supports the aerobic capacity model of vertebrate energetics. In the present study, HR mice had significantly higher V̇_O2,max (+13.6%; Tables 1 and 2, Fig. 2A), but did not have significantly higher BMR (+6.5%; Tables 1 and 2, Fig. 2C). Additionally, V̇_O2,max and BMR were not correlated at any level (e.g. among individuals or replicate lines; Table 5). Finally, aside from the positive correlation between ventricle mass and V̇_O2,max among individuals (consistent with a previous study: Rezende et al., 2006b) (but not among replicate lines; Table 5), and between ventricle mass and BMR among replicate lines (but not among individuals; Table 5), V̇_O2,max and BMR were not correlated with the mass of any other organs. Thus, the two metabolic states do not appear to be mechanistically linked through the lower-level traits measured here. Overall, our results offer limited support for the aerobic capacity model, consistent with the three rodent selection experiments that targeted V̇_O2,max and/or BMR (see Introduction).

Beyond the aerobic capacity model, our results, and the HR selection experiment as a whole, may offer insights into other hypotheses regarding links between V̇_O2,max, BMR and other traits (Hayes and Garland, 1995; Hillman et al., 2013; Careau et al., 2015; Auer et al., 2017). For example, under the assimilation capacity model (Koteja, 2000), selection favors high-intensity parental care, especially the feeding of juveniles, which requires higher daily energy expenditure (e.g. due to foraging; see also Farmer, 2000), and thus an increased rate of energy processing. Here, BMR increases as a correlated response to the increased capacity of the alimentary tract, as these organs are a primary contributor to BMR (Konarzewski and Diamond, 1995; Książek et al., 2004; Konarzewski and Książek, 2013). HR mice in the present study did not have statistically larger internal organs (e.g. liver, kidney) (or BMR) and have not been shown to have a larger alimentary tract (Kelly et al., 2017), although mini-muscle mice (a subset of the HR mice) have higher stomach dry mass and longer small intestines (Kelly et al., 2017). Additionally, Koteja (2000) proposed that high daily energy expenditure was driven by high-intensity parental care, which has not been found to differ in an important way between HR and control lines (Girard et al., 2002; Keeney, 2011).

In closing, we note that the HR mouse selection experiment is relevant to the ‘behavior evolves first’ model (e.g. Blomberg et al., 2003; Huey et al., 2003; Rhodes and Kawecki, 2009). Specifically, our results demonstrate that selection for a behavioral trait can result in very large changes (in our case, an approximately 3-fold increase in daily running distance; Table S1; Fig. 1), without large modifications to underlying morphological or physiological traits (here, only a 13.6% increase in V̇_O2,max (Tables 1 and 2, Fig. 2A), a 6.5% increase in BMR (Tables 1 and 2, Fig. 2C), and no statistically detectable changes in organ mass or hematocrit (Tables 3 and 4; Fig. S1).

We thank members of the Garland lab for their assistance. We thank M. A. Chappell for providing training and technical support. We also thank L. Graham for providing technical support. We thank members of the vivarium staff for their assistance in animal care. Finally, we thank P. Koteja for very thorough comments on an earlier version of the manuscript.