The locomotor performance achieved in a challenging situation depends not only on physiological limitations, such as the aerobic exercise capacity, but also on behavioral characteristics, such as adequate coping with stress. The stress response is mediated largely by the hypothalamic–pituitary–adrenal (HPA) axis, through modulated release of glucocorticoids. We used a unique experimental evolution model system to test the hypothesis that the evolution of an increased aerobic exercise performance can be facilitated by modification of the glucocorticoid-related stress-coping mechanisms. Bank voles (Myodes glareolus) from ‘aerobic’ (A) lines, selected for 22 generations for high maximum swim-induced rate of oxygen consumption (V̇O2,swim), achieved a 64% higher V̇O2,swim than those from unselected, control lines. The temporal pattern of exercise during the swimming trial also evolved, and the A-line voles achieved V̇O2,swim later in the course of the trial, which indicates a modification of the stress response characteristics. Both V̇O2,swim and the average metabolic rate measured during the trial tended to increase with baseline corticosterone level, and decreased with the post-exercise corticosterone level. Thus, increased baseline corticosterone level promotes high metabolic performance, but a high corticosterone response to swimming acts as an inhibitor rather than stimulator of intense activity. However, neither of the corticosterone traits differed between the A-selected and control lines. Thus, the experiment did not provide evidence that evolution of increased aerobic performance is facilitated by the modification of glucocorticoid levels. The results, however, do not exclude the possibility that other aspects of the HPA axis function evolved in response to the selection.
In several circumstances, such as predator–prey encounters, inter-male contests or demanding migratory events, an animal's life and Darwinian fitness depend on its locomotor performance. However, as we know well from the experience of both amateur and Olympic athletes, actual achievements depend not only on biomechanical or physiological limitations, such as the aerobic exercise capacity, but also on adequate coping with the stress posed by the challenging situation (e.g. Burton, 1988; Chamberlain and Hale, 2007). The same appears to apply to other animals (e.g. Baugh et al., 2013; Carere et al., 2010; Henderson et al., 2017; Koolhaas et al., 2010; Vitousek et al., 2014). Sensitivity to stress is often considered to be an element of an animal's personality or ‘behavioral syndrome’, determining its performance in particular situations (Bensky et al., 2017; Huyghe et al., 2009; Kern et al., 2016; Pryke et al., 2007; Vindas et al., 2017; Yewers et al., 2017), which in turn affects its fitness (Biro and Stamps, 2010; Réale et al., 2010). Thus, selection for increased performance under a particular challenging condition should lead to improved stress coping in this specific situation. In line with the general idea that the evolution of complex physiological adaptations is triggered by the evolution of behavior (Garland and Carter, 1994; Rhodes and Kawecki, 2009), it can be even hypothesized that the evolution of increased exercise performance is facilitated by improved mechanisms for coping with stress associated with the challenge. Surprisingly, although the importance of stress coping is well recognized in professional sports medicine and psychology, and the behavior–physiology nexus has become part of the evolutionary physiology paradigm, to the best of our knowledge such a ‘performance–stress coping nexus’ hypothesis has not yet been explicitly tested within the framework of evolutionary physiology. Here, we applied an experimental evolution model and asked whether stress coping changed in lines of a non-laboratory rodent selected for the ability to achieve high aerobic exercise metabolism.
line of bank voles selected for high V̇O2,swim (‘aerobic’)
body mass averaged from the two values obtained before the swimming trial and at baseline blood sampling
- HPA axis
non-selected line of bank voles (‘control’)
baseline corticosterone level (obtained from undisturbed animals)
ratio of corticosterone increase during the V̇O2,swim measurement (Cswim–Cbase, following log-transformation of raw Cort values)
swim-induced corticosterone levels (obtained directly after the V̇O2,swim measurement)
rate of oxygen consumption
average rate of oxygen consumption during the swimming trial
maximum 1 min rate of oxygen consumption achieved in a swimming trial
In vertebrates, the response to stressors is mediated largely by the hypothalamic–pituitary–adrenal (HPA) axis, through the modulated release of glucocorticoid hormones (McEwen and Wingfield, 2003; Romero et al., 2009). These hormones affect both behavior and many physiological parameters including oxygen and metabolic substrate distribution (Wingfield and Ramenofsky, 1999), but their effect depends on both the hormone concentration and the respective receptor density (Duclos et al., 2007; Jones et al., 1998; Kolber et al., 2008; Reichardt et al., 2000; Ridder et al., 2005; Sapolsky et al., 2000; Schilling et al., 2013). A strong glucocorticoid response results in the suppression of metabolism-supporting traits and exploratory behavior (Breuner et al., 1998; Dallman et al., 1993; Wingfield and Ramenofsky, 1999). This mechanism supports hiding from and waiting out the stressor, especially when it is impossible to escape the threat (Landys et al., 2006). However, moderate increases in glucocorticoid level are typically associated with the promotion of performance traits, immune function, resource use and allocation, and metabolism in general (Sapolsky et al., 2000; Wingfield and Ramenofsky, 1999). This is reflected in increased glucocorticoid levels at metabolically demanding life history stages (de Bruijn and Romero, 2018; Kenagy and Place, 2000; Monaghan and Spencer, 2014; Romero, 2002) or following intense physical activity (Coleman et al., 1998; Duclos and Tabarin, 2016; Hill et al., 2008; Viru, 1992). Therefore, at least at moderate levels, glucocorticoids can also be viewed as metabolism-stimulating hormones (Jimeno et al., 2017, 2018a,b; McEwen and Wingfield, 2003).
It can be hypothesized that through modulation of metabolic substrate availability and motivation to undertake physical effort in response to a challenge, the HPA axis sets the upper limit for the metabolic rate an organism achieves under particular circumstances. It is commonly assumed that the baseline level of glucocorticoids is associated with an organism's metabolic rate (McEwen and Wingfield, 2003; Romero et al., 2009), although intra- and inter-specific studies provide mixed results (Buehler et al., 2012; Francis et al., 2018; Haase et al., 2016; Jimeno et al., 2017). Depletion of glucocorticoid production drastically reduces the running endurance and spontaneous activity of rats, an effect reversible by supplementation with exogenous glucocorticoids (Devenport et al., 1993; Moberg and Clark, 1976). In humans, artificially elevated glucocorticoid level may also improve endurance at moderate-intensity exercise (Tacey et al., 2017), but it does not enhance maximal exercise performance (Baume et al., 2008; Soetens et al., 1995). Moreover, as mentioned before, a high dose of glucocorticoids may even impede the athlete's performance (Burton, 1988; Chamberlain and Hale, 2007). Thus, if metabolic performance in a given task is indeed limited by glucocorticoids, it is likely that their release is fine-tuned to optimize the strength of the response to the situation. Therefore, two opposing scenarios for the HPA axis adaptation can be proposed: boosting locomotor activity through an increased glucocorticoid response, and reduction of the activity-inhibiting effect of high glucocorticoid levels through weakening the stress response. Both possibilities are interesting with respect to the factors determining an organism's physical performance.
The experimental evolution approach offers a powerful tool to experimentally test hypotheses regarding responses to selection for well-defined traits (Swallow et al., 2009), and the problem of adaptation in the HPA axis has already been approached in several studies on rats or mice selected for locomotor-related traits. The baseline concentration of corticosterone (Cort), the primary glucocorticoid hormone of small rodents, is elevated in mice selected for high voluntary running activity (Malisch et al., 2006), particularly at night (Malisch et al., 2008), but the effect is to a large part explained by the amount of spontaneous activity before blood sampling (Girard and Garland, 2002). The Cort level achieved in response to restraint stress was not affected by the selection, but because of the difference in baseline level, the relative scope of the response decreased in the selected lines (Malisch et al., 2006). Conversely, rats selected for high or low running endurance did not differ in baseline or restraint stress-induced Cort levels (Waters et al., 2008, 2010). However, neither of these studies tested whether and how the HPA axis activity affects selection trial performance under the experimental evolution scenarios.
Here, we tested the hypotheses regarding the evolutionary aspects of correlation between metabolic performance and glucocorticoid response to the challenge using a unique model system: bank voles (Myodes glareolus) from lines selected for high swim-induced aerobic metabolism (A-lines), measured as the maximum swim-induced rate of oxygen consumption in the swimming trial (V̇O2,swim). In generations 19–22, voles from the A-lines achieved over 60% higher V̇O2,swim than those from unselected, control (C-) lines (Fig. S1; Jaromin et al., 2016; Jaromin et al., 2018). Importantly, the swimming trial does not force the animals to achieve the maximum metabolic rate per se; the animals can either swim vigorously or ‘hang’ in the water with the tip of their nose above the surface. Not surprisingly, in the C-lines, V̇O2,swim was 20% lower than the maximum forced-exercise aerobic metabolic rate (V̇O2,max), but the two traits were nearly equal in the A-lines. Thus, both the aerobic exercise capacity per se increased, and behavior evolved toward more pro-active in the A-lines: faced with the swim-test challenge, A-line voles were ready to work up to their physiological performance limit (Jaromin et al., 2016, 2018, 2019). Voles from the A-lines also spent more time on active swimming than those from the C-lines (Jaromin et al., 2018). The pro-active personality of the voles was also evident in a standard open-field test (performed in generation 15: Maiti et al., 2018). One of the mechanisms underlying the difference could be a change in neuronal transmission in the circuits responsible for motivation to undertake physical activity, similar to that observed in mice selected for high wheel-running activity (Keeney et al., 2012; Rhodes and Garland, 2003; Rhodes and Kawecki, 2009). This possibility was tested in a series of pharmacological experiments, which indeed showed an altered characteristic in the noradrenaline (norepinephrine) but not in the dopamine, serotonin or endocannabinoid signaling pathways (Jaromin et al., 2016, 2018, 2019). Here, we addressed another (not mutually exclusive) possibility, that the A-line voles evolved an altered HPA axis activity, and that the change is reflected in the blood concentration of Cort, at both its baseline and post-swimming level.
Interestingly, voles from the A-lines also evolved an increased basal metabolic rate (Sadowska et al., 2015). If Cort indeed acts predominantly as a metabolic hormone (Jimeno et al., 2017; Jimeno et al., 2018a,b; McEwen and Wingfield, 2003), we can expect increased base Cort level in the A-lines, and a positive correlation between both the base and post-swim Cort level and V̇O2,swim. If, however, a high Cort level reflects a distress status, its post-swim level should be lower in the A- than in the C-lines (as a result of indirect selection for improved stress coping), and it should be negatively correlated with V̇O2,swim.
MATERIALS AND METHODS
Animals and the selection experiment
This work was performed on bank voles, Myodes (Clethrionomys) glareolus (Schreber 1780), from generations 22 and 23 of an ongoing artificial selection experiment maintained at the Jagiellonian University (Poland). The rationale, history and protocols of the experiment have been presented in our earlier papers (Chrząs´cik et al., 2014; Sadowska et al., 2008, 2015). Briefly, in the selected, ‘aerobic’ (A-) lines, the selection criterion was the maximum 1 min rate of oxygen consumption (V̇O2,swim), achieved during 18 min of swimming at 38°C. The high water temperature eliminates the costs of thermoregulation. Thus, unlike in a similar experiment performed on laboratory mice (Gebczyński and Konarzewski, 2009, 2011), the V̇O2,swim in our experiment measures just the exercise-related performance. The V̇O2,swim values used as selection criteria were mass adjusted (residuals from ANCOVA including other covariates and cofactors). Four replicate lines for both selected and unselected control (C-) lines were maintained (to allow valid tests of the effects of selection; Henderson, 1997), with 15–20 reproducing families in each of the eight lines (which avoids excessive inbreeding). After 11 generations, the selection resulted in an approximately 50% increase in V̇O2,swim in the A-lines (Konczal et al., 2016; Sadowska et al., 2015), and the difference increased to about 60% in generations 19–22 (Fig. S1; Jaromin et al., 2016, 2018, 2019).
The animals were maintained in standard mouse cages (mostly opaque, polypropylene) with sawdust bedding, at constant temperature (20±1°C) and photoperiod (16 h:8 h light:dark; light phase starting at 02:00 h). At the age of 17 days, the animals are weaned, marked temporarily by fur clipping and kept in family groups until 30–35 days. At about 34 days, all individuals were marked permanently with mouse ear tags (model 10005-1; National Band and Tag, Newport, KY, USA; mass 0.18 g) and later maintained in same-sex groups of three individuals in model 1264C cages (Tecniplast, Bugugiatte, Italy) or up to four individuals in the larger model 1290D cages (Tecniplast). Cages were changed every 5–14 days, depending on the number of animals in the cage, and cage size and cleanliness. Water and food (a standard rodent chow: 24% protein, 3% fat, 4% fiber; Labofeed H, Kcynia, Poland) was provided ad libitum. Every day, all cages were visually inspected for the presence of food and water or dead animals. The colony was under the supervision of a qualified veterinary surgeon. During any kind of measurements, if symptoms of poor condition were observed in an animal (problems with breathing or moving, injury, etc.), it was removed from the experiment. Depending on the judgment of the observer or animal care personnel, the animal was either allowed to recover or was euthanized. Depending on the circumstances, one of three methods of euthanasia was used: exposure to a rising concentration of CO2, cervical dislocation or isoflurane inhalation (AErrane, Baxter; applied using the open-drop technique).
The selection V̇O2,swim test was performed at the age of 75–85 days, between 08:00 h and 18:30 h. On the day of measurement, home cages were transferred from the housing room to the laboratory. Animals were removed from their cages shortly before the start of measurement, weighed and placed in a temporary container while the respirometric system was being prepared. To measure V̇O2,swim, we used simultaneously two systems with respirometric chambers (15 cm diameter 3 l glass jars) partly filled with water. The water temperature was set to 38°C to ensure that the increase of metabolism was solely due to locomotor activity and not to thermoregulatory demand. After recording the initial baseline for 30 s, the chamber was opened and the animal gently placed on the water surface and allowed to swim freely. The test lasted for up to 18 min, unless an individual began to drown or oxygen consumption rapidly decreased. After removing the animal from the chamber, the chamber was closed, and after sufficient washout time (about 2 min) a 30 s final baseline was recorded. After the swimming trial, the animals were wiped with paper towels, returned to their home cages, which had been fitted with fresh bedding, and placed under a heating lamp to support fur drying. The animals were subsequently returned to the colony.
All the breeding, selection and experimental procedures were approved by the 1st Local Ethical Committee in Krakow, Poland (decision no. 170/2014).
Cort response to swimming trial
The experiment was performed on 24 A-line and 24 C-line voles from generation 22, and 29 A-line and 28 C-line voles from generation 23 (105 animals in total). At the age of approximately 30 days, 6–8 animals of either sex were sampled from each of the eight replicate lines (one animal per family), and assigned to eight approximately balanced measurement blocks. The animals were maintained under the same conditions as those from the main colony, except that they were kept individually in individually ventilated cages (AERO Mouse IVC Green Line: Techniplasty). Unlike rats or mice, bank voles are solitary in nature (Bujalska, 1990), and thus in their case social isolation is unlikely to elicit stress-related disorders. The individuals were not subjected to the selection tests.
To reduce non-specific stress resulting from human activity in the vicinity of a cage, as well as to habituate animals to handling procedures, the voles were handled once per day for 17–20 consecutive days (generation 22) or 14 consecutive days (generation 23) preceding the swimming trial and until the final blood sampling. In mice, 14 days of daily handling was sufficient for habituating the animals to human presence (Balcombe et al., 2004). However, for logistical reasons, the testing procedures in generation 22 had to be postponed by a few days, and the handling procedure was prolonged accordingly to maintain the habituation status. During handling, the animals were removed from their cages by the neck scruff and briefly immobilized by hand. The entire procedure, from taking a cage from the shelf, through capturing and handling the animal to returning it to the cage lasted about 40 s, and human activity in the vicinity of the cages (including preparatory activity and handling of all animals from an experimental block) lasted for a minimum of 15 min. Handling was performed between 08:00 h and 18:00 h, and the exact timing, degree and method (animal held in the hand or on a flat surface) of hand immobilization was diversified among the handling days to habituate the animals to an array of daily human encounters rather than an event of predictable timing and scenario. During handling, two animals were recognized as diabetic (see Bartelik et al., 2013) and were excluded from the experiment.
The swimming trials were performed at the age of 72–85 days, during the same periods and with the same systems as the regular V̇O2,swim measurements, but not on the same days. The measurements were performed between 08:00 h and 12:30 h. The animals were transferred from the housing room to the laboratory in two batches at about 08:00 h and 10:00 h and kept in their home cages in the laboratory. An animal was removed from its cage shortly before measurements started, weighed and placed in a temporary container for up to 5 min while the respirometric system was prepared. Then, the animal was placed at the water surface inside a respirometric chamber, where it was allowed to swim freely. Unlike in the standard protocol used in the selection procedure, which lasted up to 18 min unless the individual began to drown or oxygen consumption rapidly decreased, the measurement on the experimental group was 10 min, to ensure that most animals completed the trial and spent the same time in the water. As the initial period of up to 1 min after closing the chamber is not suitable for calculating the rate of metabolism, the trials provided 9 min of useful record. In four cases, the trials had to be interrupted because the animals started to drown, and in one case the respirometric system was disturbed during the measurement. These animals were excluded from further analyses.
Retro-orbital blood samples were taken immediately following the V̇O2,swim measurements (within 0.5–2.5 min, i.e. 10.5–12.5 min from the start of the trial) using 70 µl heparinized capillary tubes (Medlab Products, Raszyn, Poland). Among a number of blood sampling procedures applicable to small rodents (Joslin, 2009; Kim et al., 2018), only retro-orbital sampling allows sufficient sample volume from bank voles to be obtained. No anesthesia was applied prior to blood sampling, as it could impair the ability to obtain samples of sufficient volume through reduction of blood flow. Moreover, application of anesthesia would prolong the blood sampling procedure, which would compromise measurement of Cort levels (Kim et al., 2018). After blood sampling, the animals were dried with paper towels and returned to home cages fitted with fresh bedding.
A second blood sample, representing the basal, non-stimulated state, was taken 7 days after the swimming trial (rather than before), because bleeding the animals could result in decreased aerobic performance for several days. In contrast, after a single short stressor, Cort level recovers to baseline within hours (De Kloet, 2017), and therefore it can be assumed that after 7 days it is not affected by the stress associated with the swimming trial. Moreover, this sequence allowed us to avoid futile blood sampling from individuals that had not completed the swimming trial. Prior to blood sampling, the animals were transferred from the animal housing room to a surgery room. The time between the first disturbance (removing the cage from a rack) and the completion of blood sampling was always under 3 min (38–174 s, including the duration of transfer between the rooms). Body mass was measured after the blood sampling. The samples were taken between 08:00 h and 12:30 h, approximately at the same time as the V̇O2,swim trial for each individual.
The capillary tubes containing blood samples were stored on ice and centrifuged for 15 min at 14,000 g. Plasma samples were separated into Eppendorf tubes, frozen at −20°C, and transferred to the Leibniz Institute for Zoo and Wildlife Research (Berlin), where Cort concentration (ng ml−1) was measured according to the method described in Dehnhard et al. (2003). Samples from seven individuals were lost during processing, and these individuals were excluded from further analyses because of incomplete datasets.
The analyses were performed separately for two datasets: (1) data from all individuals in generation 22 and 23 in which V̇O2,swim was measured as part of the regular selection program, and (2) data from 46 A-line and 45 C-line animals (91 animals total) for which both V̇O2,swim and Cort data were obtained. In the latter, four aspects of aerobic metabolism were measured during the swimming trial: (i) an array of nine average rates of oxygen consumption in each subsequent 1 min interval of the test (V̇O2; ml min−1), (ii) the maximum 1 min rate of oxygen consumption (V̇O2,swim, the highest 1 min running average; ml min−1), (iii) the average rate of oxygen consumption during the swimming trial (V̇O2,avg; ml min−1) and (iv) the time of achieving V̇O2,swim (time in seconds from the start of the trial). Two Cort traits were measured directly: post-swimming Cort level (Cswim) and baseline Cort level (Cbase); a third one, Cort proportional increase, was calculated as the ratio of Cswim and Cbase (Cratio). Because of a strong right-skewness of the Cort trait distribution, clearly visible on diagnostic graphs (histograms of residuals, residual versus quantile plots, and residuals versus predicted values plots), all three were log-transformed prior to statistical analyses.
The analyses were performed with cross-nested mixed ANCOVA models, using SAS v.9.4 (SAS Institute, Inc., Cary, NC, USA) Mixed procedure (with REML method of estimation and variance components restricted to positive values). As several complex models were fitted, in addition to the description presented here, we provide a commented SAS code used for the analyses (Lipowska et al., 2019; Dryad digital repository: https://doi.org/10.5061/dryad.n17dg25). All the models included selection direction (A- versus C-lines), sex and generation as the main fixed factors, random effect of replicate line (nested within selection direction), and two covariates: time of performing the trial and body mass (unless body mass was the subject of analysis). Additionally, models for the metabolic traits included a fixed cofactor of respirometric system and generation×respirometric system interaction. The hierarchical structure of the statistical model (replicate lines nested in selection direction) is required to allow a proper distinction of the effects of selection from random genetic effects, such as genetic drift (Henderson, 1997). This basic model structure was further expanded to accommodate additional factors adequate for specific analyses, as indicated in the next two paragraphs.
One type of model was applied to V̇O2,swim measured as part of a standard selection protocol in generations 22 and 23, and body mass measured immediately before the selection swimming trial. The measurements were performed on all available individuals from A-lines, and a sample of individuals from C-lines (typically one male and one female from a family). Therefore, to properly handle non-independence of observations obtained on individuals from the same full-sib families, the models included random factor of family (nested within replicate line and generation). The models also included the litter number of a particular family as a cofactor (the first, second and third or further litter) and litter size as a covariate.
The other type of model was applied to Cort data and associated metabolic traits. The models did not include the family effect because each individual on which the tests were performed represented a separate family, but included random factor of experimental block (nested within generation). Six versions of the models were analyzed. (1) A model for average body mass (BMavg; mean of two values, recorded before the swim trial and after baseline Cort blood sampling). The analyses were performed for BMavg rather than separate values because body mass was highly repeatable (Pearson correlation: 0.97), and the mean mass is a better representation of animal size during the period of the experiment than a single record. For the same reason, BMavg was also used as a covariate in the following models. (2) Models for three Cort traits (Cbase, Cswim and Cratio), which included the respective disturbance duration (time to completing blood sampling) as an additional covariate or covariates (for Cratio, both the post-swim and base-level sampling duration were included). (3) A model for analyzing partial correlation between two Cort levels, with Cswim as a dependent variable and Cbase and the disturbance duration as additional covariates. (4) A model for analyzing the temporal pattern of average 1 min V̇O2 during the swimming trial, with consecutive minute number (excluding the first minute of recording) as an additional, repeated measures (within-individual) fixed factor. (5) Models for time of achieving V̇O2,swim and two metabolic traits: V̇O2,swim and V̇O2,avg. (6) Models for analyzing partial correlation between the metabolic and Cort traits, with the metabolic traits as dependent variables and Cort traits as predictors: either Cbase and Cswim entered simultaneously, or Cratio.
The initial models included interactions between the main fixed categorical factors (selection, sex and generation), between the main fixed factors and main covariates (body mass, Cort values), and all the respective random interactions with the replicate line. The models were then step-wise reduced by removing non-significant interactions (P>0.05). However, the interaction between selection direction and sex was a priori considered biologically meaningful and retained in all models, together with the respective random line×sex interaction, irrespective of their significance. Models from 4 and 6 above had too many predictors to include all possible two-way interactions. Their initial versions were therefore based on the final models from 5 (from which non-significant interactions were removed). Models from 6 were created by adding the Cort traits and interactions between the Cort traits and the main fixed factors (which were then stepwise reduced, as in the other models) to final models from 5. The model from 4 included all factors and interactions present in the final model of V̇O2,avg from 5, additional fixed factor of consecutive minute of test (repeated for individual), and its one-way interactions with the main fixed categorical factors. Models from 4 were initially fitted with either equal or unequal residual covariance (compound symmetry or Toeplitz covariance structure), and for final analyses we chose the better one based on a lower Akaike information criterion (AIC).
In all analyses, Satterthwaite's approximation was used to calculate the effective degrees of freedom (d.f.) for t-tests or the denominator d.f. for F-tests (i.e. the d.f. was computed from a combination of the d.f. of respective random grouping effects and residual term, weighted by variance contribution of the terms; SAS Institute Inc., 2011). Thus, the d.f. could take any real value between d.f. of the random factor and d.f. of the residual term. In the analyses of metabolic rates, the interaction between selection direction and body mass was significant (or nearly significant for the smaller dataset), and the final model had to retain heterogeneous slopes. Therefore, the effect of selection was tested for the average (22 g), minimal (16 g) and maximal (32 g) body mass relevant for both selection groups (using ‘at’ option in SAS ‘lsmeans’ statement), but as it was highly significant in the entire range of body masses, only results for average body mass are reported. Significance of the random effects was tested with the likelihood ratio test, based on results from models with the same structure as described above, but with variance components not restricted to positive values (‘nobound’ option in SAS Mixed procedure).
In analyses of Cswim from 2 and 3 above, one individual was recognized as an outlier (studentized residual <−3.5) and excluded from these analyses. The residuals of this individual did not stand out in analyses of other traits, so it was retained in the database.
Tables with descriptive statistics, as well as the raw data, are available from the Dryad digital repository (Lipowska et al., 2019; https://doi.org/10.5061/dryad.n17dg25), and tables with extended statistical results are presented in the supplementary information (Tables S1–S4). We provide results limited to significance of the main factors of interest and least squares means with 95% confidence intervals (LSM±CI), computed for the approximate mean value of covariates, for all analyses.
Body mass of 1434 animals that completed the regular selection protocol in generation 22 and 23 ranged from 13.7 to 36.9 g (mean±s.d.: 23.6±3.8 g), and did not differ significantly between the A- and C-lines (P=0.19). Animals from the A-line had significantly higher V̇O2,swim than those from the C-line in the entire range of relevant body mass (LSM±CI at body mass of 22 g, A-line: 5.49±0.19 ml O2 min−1; C-line: 3.36±0.20 ml O2 min−1; P<0.0001). Complete results of these analyses are presented in the supplementary information (Table S1, Fig. S1B).
For the 91 animals used in the Cort analyses, body mass (BMavg) ranged from 15.4 to 34.0 g (mean±s.d.: 22.3±3.7 g) and tended to decrease throughout the day (F1,82.6=2.85, P=0.095). Males were heavier than females (F1,79.3=61.6, P<0.0001), but the body mass did not differ significantly between selection directions or generations (P≥0.2; Table 1; Table S3).
The baseline and swim-induced levels of Cort varied among individuals by an order of magnitude (Cbase: from 6 to 96 ng ml−1; Cswim: from 40 to 326 ng ml−1; Fig. 1). In all individuals, Cswim was higher than Cbase, and the ratio between the two (Cratio) ranged from 1.7 to 28.2. None of these Cort measures were affected by body mass (P≥0.7; Fig. 1) or time of day (P≥0.2). Cbase significantly decreased and Cratio increased with duration of baseline blood sampling (Cbase: P=0.026; Cratio: P=0.027), whereas duration of post-swimming blood sampling did not affect Cswim or Cratio (P≥0.6; Fig. S2). After correction for the above effects, females had higher Cswim and tended to have higher Cbase than males (Cswim: P=0.001; Cbase: P=0.074), whereas Cratio did not differ between sexes (P=0.5), but the effects of selection, selection×sex interaction or generation were not significant for any of the Cort measures (P≥0.4; Fig. 2, Table 1). Cswim was not significantly correlated with Cbase (F1,77.1=0.43, P=0.5), and introduction of Cbase to the model concerning Cswim did not affect significance of other factors in the model.
The average 1 min V̇O2 increased with body mass (P=0.011) and decreased throughout the day (P=0.024; Table 2; Table S4). The slope of the relationship with body mass was steeper in the A- than in C-lines (P=0.014), and the values were significantly higher in the A- than in the C-lines over the entire relevant range of body mass (18–32 g; P≤0.001; Fig. 3A). The effect of consecutive minute of measurement was complicated by a significant interaction with selection direction (P<0.0001): in the A-lines the average 1 min V̇O2 kept rising in consecutive minutes (F8,167.8=8.32, P<0.0001), whereas in the C-lines there was no significant variation in metabolic rate among the minutes of a trial (F8,168.3=1.67, P=0.11; Fig. 3A). The effects of sex, selection×sex interaction, generation, respirometric system and generation×respirometric system interaction were not significant (P≥0.2).
The V̇O2,swim was achieved later in A- than in C-lines (A: 496±87 s; C: 328±87 s; P=0.015; Table 3, Fig. 3B; Table S4). The effects of body mass, time of day, sex or selection×sex interaction were not significant (P≥0.04). The simple effects of generation and gas analyzer system were not significant (P≥0.2), but a generation×respirometric system interaction was observed (P=0.002): in generation 22 (but not 23), animals tested in one of the systems achieved V̇O2,swim faster than in the other (F1,74.9= 9.28; P=0.003). Introduction of the Cort measures into the model did not affect these conclusions, and did not reveal any influence of Cort traits on time of achieving V̇O2,swim (P≥0.9).
The values of V̇O2,swim increased with body mass (P=0.001) and decreased throughout the day (P=0.011; Table 3, Fig. 4A). The slope of the relationship with body mass was steeper in A- than in C-lines (P=0.080). V̇O2,swim adjusted for mean body mass (22 g) was 62% higher in A- than in C-lines (LSM±CI, A-line: 5.34±0.44 ml min−1; C-line: 3.30±0.44 ml min−1; P=0.0001; Table 3, Fig. 3C; Table S2). The effects of sex, selection×sex interaction, generation, respirometric system and generation×respirometric system interaction were not significant (P≥0.2; Table 3). These results were not markedly affected by the introduction of Cort measures into the model (Table 3). However, V̇O2,swim, corrected for the above effects, tended to decrease with increasing Cswim (P=0.057) and increase with increasing Cbase (P=0.097), and it decreased significantly with increasing Cratio (P=0.020; Table 3, Fig. 5A–C).
The results for the average rate of oxygen consumption from the whole swimming trial (V̇O2,avg) were similar to those for V̇O2,swim (Figs 3D and 4B, Table 3; Table S2), but the relationship with Cort traits was clearer. V̇O2,avg decreased significantly with increasing Cswim (P=0.024) and tended to increase with increasing Cbase (P=0.054), and decreased highly significantly with increasing Cratio (P=0.006; Table 3, Fig. 5D–F).
Our results confirmed that artificial selection for high aerobic exercise metabolism resulted in both an immense increase in the maximum metabolic rate achieved during swimming (V̇O2,swim) and a shift toward a more pro-active response to the trial. In generations 22 and 23, animals from the A-lines achieved 64% higher V̇O2,swim than those from the C-lines when measured with a standard, 18 min selection protocol. V̇O2,swim achieved in the shortened trial (10 min) associated with measuring the Cort response was only about 2.5% lower than that achieved in the standard trial, and the magnitude of the difference between A- and C-lines (62%) was similar, too (Fig. 3). We have already reported that during the tests, animals from the A-lines spend more time on active swimming than those from the C-lines (Jaromin et al., 2018). Here, we show a difference in the timing of achieving V̇O2,swim: the C-line voles typically achieved V̇O2,swim shortly after being placed in the water, and did not tend to have a clear pattern of metabolic rate changes throughout the measurement, whereas the A-line voles tended to gradually increase their metabolic rate until reaching V̇O2,swim toward the end of the measurement (Fig. 3A,B). Both the increased proportion of time spent on active swimming and the steady growth of metabolic rate during the trial indicate that selection modified the swim–stress response characteristics.
Interestingly, the effect of selection on the behavioral response to swimming in bank voles from our selection experiment was opposite to that in mice selected for high voluntary wheel-running activity: during a swimming trial, mice from the high-runner lines spent less time on active swimming than those from the control lines (Malisch et al., 2009). It can be hypothesized that the mechanisms motivating mice to perform intense voluntary exercise in running wheels differ from those motivating voles to swim vigorously when placed in the water. It also cannot be expected that a change in Cort level observed in response to one stressor is the same as changes in response to all other stressors, particularly if selection has also affected personality traits (as in voles from our selection experiment; Maiti et al., 2018). Therefore, to learn whether a modification in HPA axis activity facilitated the evolution of high aerobic exercise performance in bank voles, we analyzed changes in Cort level in response to the challenge faced by the voles in the selection scheme.
The animal's physical activity during a swimming trial can be affected by both its personality and the stress it experiences (Castro et al., 2012; de Kloet and Molendijk, 2016; Stepanichev et al., 2018; Veenema et al., 2003a,b). Both factors are associated with glucocorticoid-mediated HPA axis signaling (Careau et al., 2008; Castro et al., 2012; Charmandari et al., 2005; Koolhaas et al., 1997), suggesting it is involved in the mechanism underlying the divergence in behavior during the swimming trial. Exposure to such a challenge results in an increase in plasma glucocorticoid level, which can be caused by stress associated with the test procedure (Sapolsky et al., 2000) and/or be an element of the system promoting intense physical activity (Viru, 1992; Wingfield and Ramenofsky, 1999). To investigate this subject, we measured levels of Cort, the primary glucocorticoid hormone of small rodents, after the swimming trial and under baseline conditions. Both values varied up to an order of magnitude among individuals, similarly to what was observed in brown lemmings (Fauteux et al., 2017) or in California mouse (Dlugosz et al., 2012). This finding is also in line with the pulsatile character of Cort release observed in rats, in which plasma Cort levels fluctuated by one or more orders of magnitude in the course of less than an hour (Windle et al., 1998). Besides the inter- and intra-individual variation of Cort level, Cort can also vary greatly among species (Romero et al., 2008), making it difficult to draw comparisons between studies performed on other animal models. Moreover, the analytical method has a profound effect on the glucocorticoid values measured (Gatti et al., 2009); thus, comparison between absolute values measured with different methods is not very informative. However, both baseline and swim-induced Cort values of bank voles reported in this study fit within the range of Cort values reported for mice that varied in the amount of exercise carried out prior to blood sampling (Coleman et al., 1998; Girard and Garland, 2002), and the ratio of the swim-induced to baseline Cort was similar to or higher than that observed after forced swim tests on other rodent species (Anisman et al., 2001; Sinton et al., 2000; Taymans et al., 1997). We also found that female voles have higher Cort than males, a difference also observed in other rodent species (Harpaz et al., 2013; Seale et al., 2004; Taymans et al., 1997).
Both the maximum and average V̇O2 achieved during swimming were negatively correlated with the Cort response to swim stress, measured as either its post-swim level (Cswim) or the factorial increase with respect to the baseline level (Cratio; Table 3, Fig. 5). The result implies that Cort exceeded the level at which it stimulates the animal's metabolism and motivation to undertake physical effort in response to a challenge, and instead represents the level of stress perceived by the animal, blunting its physical performance. However, although the difference in behavior between voles from the A- and C-lines indicates a shift in perception of stress or motivation to perform activity, it is not reflected in a lower level of Cswim or Cratio (Fig. 2, Table 1). It could be argued that the lack of a significant difference results from a huge inter–individual variation of Cort level, which masks the hypothetical differences between the lines. However, the large number of individuals provided enough statistical power to detect the differences between sexes and phenotypic (individual level) correlation between Cort level and metabolic rate. Thus, the lack of the effect of selection on Cort values seems to have a strong empirical basis.
The swim-induced rate of metabolism tended to increase with baseline Cort level, although the relationship was nearly significant only for the average V̇O2 recorded during the trial. This observation is in concert with the concept of metabolism-stimulation function of Cort at low or moderate levels, but the support is statistically weak. However, although basal metabolic rate is higher in the A- than in the C-lines (Sadowska et al., 2015), the selection did not affect baseline Cort level, whereas mice selected for high running activity have higher baseline Cort but similar basal metabolic rate to those from non-selected lines (Kane et al., 2008). It is therefore unlikely that in these animals Cort plays a major role in determining the metabolism of resting animals.
In conclusion, the results provide only weak support for the idea that at low concentrations, Cort acts predominantly as a metabolic hormone (Jimeno et al., 2017; Jimeno et al., 2018a,b; McEwen and Wingfield, 2003), but support its involvement in signaling distress status. However, the results do not provide evidence that alteration of the HPA axis activity towards reducing the adverse effects of stress induced by a locomotor challenge is a crucial factor in the evolution of high aerobic exercise capacity. It cannot be stated, however, that evolution of this performance trait did not affect the stress response system as a whole. The selection could have affected perception of the Cort-mediated signal by the target tissues through modulation of glucocorticoid receptor expression (Veenema et al., 2004). Under- or over-expression of glucocorticoid receptors markedly affects sensitivity to Cort and the behavioral reaction to stress in mice (Kolber et al., 2008; Reichardt et al., 2000; Ridder et al., 2005). In humans, polymorphism in sensitivity to the glucocorticoid signal is also correlated with aspects of energy metabolism, such as BMI, insulin release and plasma cholesterol (van Rossum and Lamberts, 2004). Moreover, modulation of Cort sensitivity can be tissue specific, given the weak correlation between the receptor expression in different tissues of same individual, as indicated by studies on house sparrows (Lattin et al., 2015). As the selection experiment on bank voles is continued, the ‘performance–stress coping nexus’ hypothesis and putative regulatory mechanisms underlying the evolution of high V̇O2,swim can be addressed in further studies.
We thank Wolfgang Goymann for help in designing the project, organizing corticosterone analyses, and comments on the manuscript. We are also very grateful to the many technicians and students who helped with animal maintenance and sampling, especially B. Bober-Sowa, K. Baliga-Klimczyk and J. Wyszkowska, and the two anonymous reviewers for their constructive criticism.
Conceptualization: U.B., P.K.; Methodology: M.M.L., E.T.S., P.K.; Software: P.K.; Formal analysis: M.M.L.; Investigation: M.M.L.; Resources: E.T.S.; Data curation: M.M.L., E.T.S.; Writing - original draft: M.M.L.; Writing - review & editing: M.M.L., E.T.S., U.B., P.K.; Visualization: M.M.L.; Supervision: P.K.; Project administration: E.T.S.; Funding acquisition: P.K.
The study was supported by the National Science Centre in Poland (Narodowe Centrum Nauki, grant 2014/13/B/NZ8/04683 to P.K.) and the Jagiellonian University (DS/WBINOZ/INOS/757).
Data are available from the Dryad digital repository (Lipowska et al., 2019): https://doi.org/10.5061/dryad.n17dg25
The authors declare no competing or financial interests.