Reduced Nonexercise Activity Attenuates Negative Energy Balance in Mice Engaged in Voluntary Exercise

Abstract

Exercise alone is often ineffective for treating obesity despite the associated increase in metabolic requirements. Decreased nonexercise physical activity has been implicated in this resistance to weight loss, but the mechanisms responsible are unclear. We quantified the metabolic cost of nonexercise activity, or “off-wheel” activity (OWA), and voluntary wheel running (VWR) and examined whether changes in OWA during VWR altered energy balance in chow-fed C57BL/6J mice (n = 12). Energy expenditure (EE), energy intake, and behavior (VWR and OWA) were continuously monitored for 4 days with locked running wheels followed by 9 days with unlocked running wheels. Unlocking the running wheels increased EE as a function of VWR distance. The metabolic cost of exercise (kcal/m traveled) decreased with increasing VWR speed. Unlocking the wheel led to a negative energy balance but also decreased OWA, which was predicted to mitigate the expected change in energy balance by ∼45%. A novel behavioral circuit involved repeated bouts of VWR, and roaming was discovered and represented novel predictors of VWR behavior. The integrated analysis described here reveals that the weight loss effects of voluntary exercise can be countered by a reduction in nonexercise activity.

Introduction

Physical exercise alone generally fails to result in meaningful weight loss in obese populations (1,2). Similarly, numerous reports suggest that the same is true regarding the effectiveness of voluntary wheel running (VWR) for inducing weight loss in mice (3–5). The ineffectiveness of voluntary exercise for weight loss is, at least in part, due to compensatory feedback mechanisms that counterbalance increased energy expenditure (EE) and preserve energy balance. Chronic compensation to exercise (>2 weeks) includes increased food intake (6,7), reduced resting metabolic rate (8–10), and decreased nonexercise physical activity (11). Acute compensation to exercise (<2 weeks) in mice can be independent of increased food intake (12), suggesting that a decline in nonexercise physical activity may help offset exercise-induced EE in accordance with a “constrained” model of EE (13). Indeed, mice engaged in VWR have decreased nonexercise physical activity, termed off-wheel activity (OWA) (14–16). The metabolic impact of decreased OWA with VWR has not been previously reported.

There are limited data regarding the contributions of VWR and OWA to EE in mice. This is because VWR and OWA are voluntary and typically are performed at low to moderate intensity, thus requiring sensitive instrumentation to detect an increase in EE (17). Only one report to date (18) has demonstrated an acute (within minutes) effect of VWR on EE and was performed in outbred deer mice. O’Neal et al. (14) were able to detect a contribution of VWR to the daily energy budget in male C57 mice during a short-duration (3 days) study and observed a diminishing contribution of VWR to EE during a longer-duration (21 days) study. Surprisingly, however, the increase in EE from VWR observed by O’Neal et al. (14) during their chronic study was not positively correlated with VWR distance. Brown et al. (19) observed a distance-dependent increase in daily Vo₂ in female C57BL/6N mice, but a caveat to this study is that female mice run much farther than male mice (20). Although a few studies have directly measured EE during VWR, no studies to date have been able to detect the independent contribution of OWA to EE (21); thus, potential interactions between VWR and OWA and their impact on EE and energy balance have not been addressed.

There is significant variability in voluntary physical activity patterns in mice, even within an inbred strain (14). Skeletal muscle (SkM) mitochondria are the primary source of ATP during low- to moderate-intensity exercise, and exercise training has been shown to improve SkM mitochondrial efficiency (22). Recent work (23) has revealed that muscle mitochondrial efficiency also acutely increases as a function of metabolic demand, suggesting that this variable may be a novel factor explaining differences in VWR behavior between mice.

The purpose of the studies described here was to combine indirect calorimetry, context-specific behavioral mapping, and mitochondrial energetics in C57BL/6J mice to 1) quantify the metabolic contributions of VWR and OWA to EE, 2) examine whether reduced OWA is a mechanism to preserve energy balance in response to VWR, and 3) identify behavioral and metabolic factors that explain variance in VWR between mice.

Research Design and Methods

Animal Housing and Study Design

Vanderbilt University Animal Care and Use Committee approved all experiments. Twelve-week-old male C57BL/6J mice (n = 12) were purchased from The Jackson Laboratory and were housed in the Vanderbilt University Mouse Metabolic Phenotyping Center. Mice were singly housed starting at 15 weeks of age. Indirect calorimetry studies were performed from 19 to 21 weeks of age. At 23 weeks of age, after a 4-h fast and under isoflurane inhalation, the red gastrocnemius muscle was excised for mitochondrial energetics assays.

Indirect Calorimetry and Voluntary Physical Activity Behavior

Mice were singly housed in a Promethion Metabolic Analyzer (Sable Systems, North Las Vegas, NV) at ambient temperature (20–22°C). Cages contained Pure-O’Cel Bedding (The Andersons Inc., Maumee, OH), and mice had ad libitum access to 5L0D chow diet (LabDiet, St. Louis, MO) and water throughout the study. Running wheels were present for the entire time period, but were locked for the first 4 days. O₂ and CO₂ were continuously monitored for assessment of EE. Mouse cage behavior, including roaming (XYZ beam breaks), food and water intake, and VWR were monitored continuously with second-to-second resolution. Contextual behavioral mapping was achieved by integrating XYZ beam break data with mass measurement modules for food intake, water intake, and body weight. Additional technical information is provided in the Supplementary Data.

Behavior mapping analysis for each animal was performed for the last dark phase (12 continuous hours) of each condition (locked or unlocked wheel) using ExpeData Software (Sable Systems). This analysis yielded a complete sequential list of behaviors. For each behavior, the duration and frequency were measured. From the sequential list of behaviors, time budgets were generated as percentages. Time budget data were analyzed with two-dimensional hierarchical clustering (x-axis = animal, y-axis = behavior) and visualized as heat maps representing individual behavior budgets for individual mice relative to the group mean. Behavioral transitions were determined for the last continuous 24 h of data for each condition, and probability matrices were visualized as a Markov chain.

SkM Mitochondrial O₂ Consumption, ATP Production, and Oxidative Efficiency

Mitochondrial ATP production, O₂ consumption, and oxidative phosphorylation efficiency (ATP/O) were measured in permeabilized myofiber bundles from red gastrocnemius muscle as previously described (23).

Calculations

The contributions of OWA or VWR to total EE (EE_total) were determined from time-matched behavior and metabolic data collected every 5 min over the last 72 h (435 data points/mouse) of each period (i.e., wheel locked or unlocked). Individual data points were binned based on OWA or VWR activity speed. The EE at rest (i.e., activity = 0) was determined then subtracted from EE during OWA or VWR. The remaining EE was multiplied by the percentage of time spent active at a given speed to determine the absolute contribution of OWA to EE (EE_OWA) or the absolute contribution of VWR to EE (EE_VWR). To determine EE_OWA in mice with an unlocked running wheel, frequency distributions of OWA speed were generated as described above. Notably, however, the fraction of EE above rest in mice with a locked wheel was used instead of that measured when the wheel was unlocked. This was done because VWR typically occurred in concert with OWA, thereby convoluting the specific contribution of OWA to EE. This calculation relies on the assumption that the energetic cost of OWA at a given speed does not change when the wheel was unlocked. The metabolic cost of exercise (MCE) of VWR was calculated by dividing the total number of kilocalories expended at a given binned speed by the total distance traveled at that speed. The average energy cost of VWR across all speeds was used to calculate the MCE in kilocalories per meter for each individual mouse.

Energy intake (in kcal) = food intake (g) * 2.91 (kcal/g). The 2.91 kcal/g value represents the metabolizable energy obtained from the diet based on manufacturer data. Daily energy balance = energy intake (kcal/day) − EE (kcal/day). ΔOWA, ΔEE, Δenergy intake, and Δenergy balance were each calculated as (unlocked wheel − locked wheel).

Statistics

A paired Student t test was used to compare data when the wheel was locked versus unlocked. One-way ANOVA with Tukey post hoc test was used to examine rate-dependent metabolic responses. Simple linear regression analysis was used to generate correlation coefficients (r) or coefficients of determination (r²), as appropriate. An F test was used to determine the statistical significance of linear regressions. Unless otherwise indicated, all data are presented as the mean ± SD. Statistical significance was defined as P < 0.05.

Results

Daily light- and dark-phase VWR for the entire experimental period is shown in Fig. 1A, and the average from days 11 to 13 (shown in Fig. 1B) was used for subsequent analyses. There was no significant relationship between VWR distances during the light phase and the dark phase (Fig. 1C). The EE_total for the entire experimental period is shown in Fig. 1D. The EE_total increased during both light and dark phases when the running wheel was unlocked (Fig. 1E). The change in dark-phase EE_total (ΔEE_total) closely correlated with VWR distance (Fig. 1F).

Figure 1.

VWR elicits a distance-dependent increase in EE. VWR was measured during the light (open boxes and circles) and dark (closed boxes and squares) phases over the entire experimental period (A) and averaged over the final 3 days (B). %CV = coefficient of variation. No relationship was observed between light-phase and dark-phase VWR (C). EE_total was measured during the light phases (open boxes) and dark phases (closed boxes) over the entire experimental period (D). EE_total increased in both the light (circles) and dark (squares) phases of mice with access to a running wheel (E). The change in EE_total (ΔEE_total) after unlocking the wheel was tightly correlated with VWR distance (F). N = 12 mice; *P < 0.05 between bracketed groups using the paired Student t test. Regression lines are shown for relationships with P < 0.05 with the F test.

The relative amount of time mice spent at rest or performing VWR at different speeds is shown in Fig. 2A. EE_total normalized to time (kcal/h) was significantly elevated when VWR was performed at any speed relative to inactivity (Fig. 2B). The EE_total linearly increased as a function of VWR speed (r² = 0.96, P < 0.0001; slope = 3.0 ± 0.4 kcal/km [mean ± SEM]). The EE_VWR at a given speed is shown in Fig. 2C. Summing these individual contributions, we concluded that VWR accounted for 26.6 ± 7.7% of EE_total. MCE (kcal/m) (Fig. 2D) decreased as VWR speed increased and tightly fit a one-phase exponential decay curve (r² > 0.98 for all 12 mice). MCE at <5 m/min was positively correlated with VWR distance (r² = 0.47, P < 0.05), but MCE at higher speeds of VWR was not significantly correlated. These data reinforce the observation made in Fig. 1 that VWR is a direct contributor to EE and implicate MCE at low VWR speeds as a predictive indicator of VWR behavior.

Figure 2.

Acute changes in EE and metabolic efficiency as a function of VWR speed. Real-time data from each mouse (435 data points/mouse) over the final three dark phases of the experimental period were separated based on running speed over each 5-min interval. The fraction of time mice spent inactive (VWR = 0) or running at different VWR speeds was measured (A). EE_total of mice during inactivity or at various VWR speeds was determined (B). The fraction of EE_total attributed to VWR at a given speed (active EE − inactive EE) was multiplied by the time spent at that speed to determine the contribution of VWR to EE_total (EE_VWR) (C). MCE (kcal/m traveled) was determined by dividing EE_VWR by the distance traveled at each speed (D). N = 12 mice; * or #P < 0.05 between bracketed groups. hr, hour.

We hypothesized that SkM oxidative phosphorylation efficiency would be positively related to physical activity, thus providing a link between efficient aerobic metabolism and the motivation to exercise. Rates of ATP production (JATP) and O₂ consumption (JO₂) by permeabilized muscle fibers, as well as the resulting ATP/O values measured as a function of ADP concentration, are provided in the Supplementary Data. Surprisingly, plotting ATP/O at each ADP concentration studied (5, 20, 75, and 200 μmol/L) against VWR distance, ΔEE_total, or MCE did not reveal any significant relationships (r² < 0.1 for all ATP/O values tested). These data suggest that SkM mitochondrial oxidative efficiency is not associated with VWR performed over a relatively brief period of time.

Dark-phase OWA was decreased in mice with an unlocked running wheel (Fig. 3A and B), but this was not related to VWR distance (Fig. 3C) or time spent performing VWR (r² = 0.10, P = 0.31). Energy intake was not different in the light or dark phases when the running wheel was unlocked (Fig. 3D and E) or related to VWR (Fig. 3F). Daily energy balance became negative when the running wheel was unlocked (Fig. 3G and H), but changes in energy balance were not significantly related to dark-phase VWR distance (Fig. 3I).

Figure 3.

Energy balance is a determinant of the compensatory reduction in physical OWA in mice performing voluntary exercise. OWA was measured over the entire experimental period during light (open boxes) and dark (closed boxes) phases (A). Average dark-phase OWA decreased in mice with an unlocked running wheel (B) but was not correlated with VWR distance (C). Energy intake was measured over the entire experimental period (D) and was not different when the running wheel was unlocked (E) or correlated with VWR distance (F). Daily energy balance (energy intake − EE) was measured over the entire experimental period (G). Average daily energy balance became negative in mice performing VWR (H) but was not correlated with VWR distance (I). The change in OWA when the running wheel was unlocked (ΔOWA) was not correlated to ΔEE (J) or Δenergy intake (ΔEI) (K) but was positively related to Δenergy balance (ΔEB) (L). N = 12 mice. *P < 0.05 between bracketed groups using the paired Student t test. Regression lines are shown for relationships with P < 0.05 with the F test.

To explore how unlocking the running wheel might lead to decreased OWA, we compared the decline in OWA (ΔOWA) to changes in key metabolic variables. ΔOWA was not related to ΔEE_total (Fig. 3J) or Δenergy intake (Fig. 3K) but was positively associated with Δenergy balance (Fig. 3L). These data suggest that reduced OWA may partially offset negative energy balances in response to the increased EE of VWR.

Frequency distributions of OWA speed (Fig. 4A), EE_total as a function of OWA rate (Fig. 4B), and EE_OWA (Fig. 4C) binned in 0.1 m/min increments were calculated. EE_OWA was not related to total OWA distance (Fig. 4D). EE_OWA measured at speeds <0.3 m/min also failed to correlate with OWA distance (Fig. 4E). EE_OWA at speeds >0.3 m/min was positively correlated with OWA distance (Fig. 4F). We next examined food intake and water intake as they related to OWA and EE. The percentage of time spent eating was greater at low OWA (Fig. 4G), whereas time spent drinking was greater at high OWA (Fig. 4H). The rates of food and water intake were both greater, as expected, at low OWA compared with high OWA (Fig. 4I and J). Food intake corresponded to an ∼30% increase in EE_total during low OWA but only an ∼3% increase during high OWA (Fig. 4K). Similar effects were observed for water intake (Fig. 4L). These data demonstrate for the first time that OWA is an independent contributor to EE_total in mice. Since the contribution of OWA to EE_total was observed only at OWA speeds >0.3 m/min, it is reasonable to conclude, as others have (21), that OWA at low speeds is not a direct contributor to EE.

Figure 4.

OWA is an independent contributor to EE in mice. Time spent performing OWA at various speeds (A), EE_total when performing OWA (B), and EE_OWA (C) were determined from 435 data points collected during three consecutive dark phases of data (days 2–4). OWA distance was plotted against total EE_OWA (D), EE_OWA with OWA <0.3 m/min (E), and EE_OWA with OWA >0.3 m/min (F). Time spent eating (G) was greater when OWA was low (<0.3 m/min) compared with high (>0.3 m/min), whereas the opposite was true for time spent drinking (H). Average intake rates of food (I) and water (J) were both greater during low OWA compared with high OWA. EE_total during food intake (K) and water intake (L) were both increased at both high and low rates of OWA, but the effect on EE was much larger at low rates of OWA. In panel B, *P < 0.05 compared with OWA = 0, #P < 0.05 compared with OWA <0.1 m/min, †P < 0.05 compared with OWA <0.3 m/min using ordinary one-way ANOVA with Tukey post hoc test; in all other panels, *P < 0.05 between bracketed group using paired Student t test. hr, hour.

EE_OWA when OWA was >0.3 m/min decreased when running wheels were unlocked (Fig. 5A). This decrease in EE_OWA was linearly related to the decrease in OWA distance (Fig. 5B), confirming that it represents a change in the direct contribution of OWA to EE_total. We next compared the measured EE_total in mice with an unlocked running wheel to the predicted EE_total if OWA were not reduced. The formula for this calculation was as follows: predicted EE_total = measured EE_total + (EE_OWA locked − EE_OWA unlocked). This calculation predicts that, if OWA was not reduced, EE_total would be reduced ∼5% (Fig. 5C) and energy balance would be ∼45% less negative (Fig. 5D). These data are the first to demonstrate a direct impact of reduced nonexercise physical activity on exercise metabolism.

Figure 5.

Reduced OWA mitigates negative energy balance in mice performing VWR exercise. A: EE_OWA when OWA >0.3 m/min decreased in mice with an unlocked running wheel. B: ΔEE_OWA when OWA >0.3 m/min was positively related to the change in OWA distance when the wheel was unlocked. If OWA was not decreased, it is predicted that EE_total would be ∼5% greater (C) and energy balance (EB) would be ∼45% more negative (D) than measured. N = 12 mice. *P < 0.05 between groups using paired Student t test. hr, hour.

Since OWA impacts EE_total and energy balance, we sought to define which OWA behaviors are altered by VWR. A behavioral analysis approach that quantifies the relative amount of time a mouse spends performing various activities was used. When the wheel was unlocked, mice spent 17.71 ± 5.4% (mean ± SD) of their time on the wheel. Time spent running was positively correlated with VWR distance (r² = 0.86; P < 0.0001) and negatively correlated with time spent roaming (r² = 0.92; P < 0.0001). Notably, however, the change in time spent roaming (unlocked − locked) was not correlated with the change in OWA distance (r² = 0.11, P = 0.29), suggesting more intricate remodeling of OWA in response to VWR. To first examine the changes in OWA elicited by unlocking the wheel, independent of wheel-running behavior, the time spent running was removed as a variable. As a group, mice spent more time performing short roaming and less time performing long roaming (Fig. 6A) when the wheel was unlocked. The relative change in behavioral frequency for each mouse (unlocked/locked) was visualized as a two-dimensional hierarchical cluster (Fig. 6B), revealing a consistent increase in time performing short roaming. Mice performed long roaming at a slower speed when the wheel was unlocked with no change in short-roaming speed (Fig. 6C). The total distance traveled during long roaming was reduced ∼65% when the running wheel was unlocked (Fig. 6D). Despite a decrease in long-roaming distance, VWR distance was correlated with time performing short roaming, but not long roaming (Fig. 6E). Plotting the percentage change in short roaming in response to an unlocked wheel against VWR distance revealed an even stronger relationship (Fig. 6F). These findings demonstrate that the impact of VWR on OWA is intricate and multifaceted.

Figure 6.

Reduced duration of episodic inactivity after wheel access predicts VWR distance. A: Behavior frequency was determined for 12 consecutive hours during the last dark phase with the wheel locked or unlocked. B: The response of individual mice to an unlocked wheel (unlocked/locked) was analyzed using two-dimensional hierarchical clustering (mouse, x-axis; behavior, y-axis) and visualized as a heat map. C: The average speed of long or short roaming was measured in mice with a locked or unlocked wheel. D: The total distance traveled during short or long roaming was determined in mice with a locked or unlocked wheel. E: Time spent short roaming (open circles) when the wheel was unlocked was positively associated with VWR. F: The response in short roaming to the wheel being unlocked revealed an even stronger positive association with VWR. N = 12 mice. *P < 0.05 between groups using paired Student t test.

We further investigated the relationship between short roaming and VWR using behavioral transition matrices. With this approach, we tested the hypothesis that short roaming and VWR constitute a repeated behavioral circuit that predicts VWR distance. Behavioral transitions (data provided in Supplementary Data) were visualized as a Markov chain (Fig. 7). Consistent with our hypothesis, short roaming was the most common behavior preceding VWR, occurring 39% of the time, and VWR was the most common behavior that preceded short roaming, occurring 86% of the time. Bouts of short roaming adjacent to (before or after) VWR bouts were performed at a higher speed than other short-roaming bouts (0.91 ± 0.14 vs. 0.62 ± 0.15 cm/s, P < 0.0001). The frequency with which short roaming preceded VWR was a positive predictor of VWR distance (r² = 0.54, P < 0.01). These findings led us to hypothesize that the number of repeated running bouts would predict VWR distance. Surprisingly, the total number of repeated running bouts did not predict VWR distance (r² = 0.23, P = 0.12). The average VWR bout duration was tightly correlated with VWR distance (r² = 0.59, P < 0.01). Long, repeated VWR bouts (>100 s) positively predicted VWR distance (r² = 0.72, P < 0.001), whereas no relationship was observed between VWR distance and the number of short repeated running bouts (<100 s) (r² < 0.1). These data demonstrate functional interactions between VWR and OWA that more comprehensively describe voluntary exercise behavior.

Figure 7.

Duration and rapid successive bouts predict VWR distance. Mouse behavior transitions were continuously monitored for the last 24 h with the wheel unlocked. The coupling of behaviors in sequence was mapped using a Markov chain plot that revealed “short roaming → VWR” and “VWR → short roaming” as prominent behavioral transitions. Lines represent the group mean for N = 12 mice.

Discussion

In this report, we characterize metabolic and behavioral remodeling that occurs in mice engaged in VWR. We have quantified the direct contributions of VWR and OWA to EE and found that decreased OWA can preserve energy balance in response to VWR. Remodeling of OWA is complex and includes both gross reductions in OWA distance and time spent roaming, but also changes in the duration and context of roaming behavior. These intricate and assorted modifications in cage behavior suggest an important interplay between VWR and OWA in the regulation of energy balance and provide a framework for assessing mouse metabolism in the context of both exercise and nonexercise behavior.

VWR contributing to EE_total (Fig. 1D–F) is consistent with previous work in outbred mice (18,24), as well as a recent study in C57BL/6 mice (14), but is at odds with another report (21). We refine our understanding of the relationship between VWR and EE_total by showing a strong positive linear correlation between dark-phase EE_total and VWR distance (Fig. 1F). One reason a relationship between VWR and EE_total has been difficult to detect may relate to the interval of metabolic gas sampling. For example, Chappell et al. (18) used a fast sampling interval (90 s) and demonstrated a speed-dependent contribution of VWR to EE_total. In the current study, a 5-min sampling interval was used, and we observed a similar relationship between VWR and EE_total. O’Neal et al. (14) used 13-min intervals and observed an increase in daily EE_total when mice had an unlocked wheel, but the rise in EE_total was not correlated with VWR distance. Finally, Virtue et al. (21) used 18-min intervals and failed to detect a contribution of VWR to EE_total in mice. Thus, sampling rate is a critical factor in the ability to resolve the contribution of VWR to EE_total.

We were also able to detect the acute (5 min) contribution of VWR to EE_total (Fig. 2A). From these data, we found that MCE decreased as a function of VWR speed (Fig. 2D) and was positively related to VWR distance. This observation suggests that metabolic efficiency may be a predictor of voluntary physical activity behavior and warrants further investigation. SkM mitochondrial oxidative efficiency was not related to VWR measured over this relatively brief time period (Supplementary Data), suggesting that differences in MCE may be primarily due to biomechanic or cardiovascular differences as opposed to variations in oxidative metabolism per se.

As far as we are aware, this is the first study to report an independent contribution of OWA to EE_total. From these data, we also provide a potential explanation for why previous reports have been unable to do so. For example, Virtue et al. (21) found a linear relationship between EE_total and low speeds of OWA. Examining only low OWA rates to determine EE_OWA, they found no significant relationship between EE_OWA and OWA speed and thus concluded that the independent contribution of OWA to EE_total was essentially zero. Our data show the same poor relationship between EE_OWA at low OWA speeds (<0.3 m/min) and total OWA distance (Fig. 4E), likely due to the concurrent intake of food and/or water (Fig. 4K and L). However, we demonstrate that EE_OWA at >0.3 m/min is positively correlated with OWA distance (Fig. 4F) where food/water intake have a minimal effect on EE. These data collectively suggest that food and water intake explain, at least in part, why previous reports have been unable to detect an independent contribution of OWA to EE_total.

A primary goal of the experiments performed was to define the impact of decreased OWA on energy balance in mice performing VWR. In these experiments, an unlocked wheel decreased energy balance (Fig. 3G and H) independent of VWR distance (Fig. 3I) and changes in energy intake (Fig. 3F). A lack of change in energy intake has been reported in mice that engage in VWR for brief periods of time (14). The negative energy balance observed with an unlocked wheel was associated with a decrease in OWA (Fig. 3L) but not VWR distance (Fig. 3I), EE_total (Fig. 3J), or energy intake (Fig. 3K). These findings collectively suggest that energy balance may be a determinant of OWA in mice performing VWR.

We observed a reduction in EE_OWA at high OWA speeds (i.e., >0.3 m/min) when the running wheel was unlocked compared with when it was locked (Fig. 5A). This change in EE_OWA was linearly related to the change in OWA distance resulting from the wheel being unlocked (Fig. 5B), reflecting a decrease in the direct contribution of OWA to EE_total. The decrease in EE_OWA when the wheel was unlocked is estimated to reduce EE_total by ∼5% and consequently mitigate the drop in daily energy balance by ∼45% (Fig. 5C and D). This difference in energy balance, specifically due to reduced OWA, translates to a decrease in weight loss of 0.4 g of fat per week. These data are consistent with a “constrained” model of EE_total where reduced nonexercise activity serves to maintain a ceiling for EE (13,25). Human studies examining whether nonexercise physical activity decreases with exercise are somewhat inconsistent (reviewed in the study by Melanson [26]), so it is possible that our approach can be coupled with mouse genetic engineering to better understand these variable responses.

VWR behavior is highly variable between individual mice, even within an inbred strain (14). OWA behavior of the C57BL/6J mice studied here dramatically changed when they were allowed to run on a wheel. For example, mice spent less time roaming when given access to a wheel. Specifically, long roaming was performed less often (Fig. 6A) and at a slower speed (Fig. 6C) in mice with an unlocked wheel. These changes resulted in an ∼65% decrease in distance traveled during long roaming (Fig. 6D). Meijer and Robbers (27) recently showed that wild mice run significant distances on a wheel placed in their natural environment for no clear reason, so it is interesting to consider whether this observation could be explained by VWR and long roaming being functionally similar behaviors associated with foraging.

Mice performing VWR engaged in a distinct behavioral circuit consisting of repeated bouts of short roaming and VWR (Fig. 7). Engaging in this circuit was found to be a positive predictor of VWR distance, suggesting it may be a tractable mechanism for better understanding voluntary exercise behavior. One possible explanation for the existence of this circuit may be to avoid overexertion incurred by longer bouts. These findings provide valuable insight into our understanding of why mice, and potentially humans, have such broad variability in their willingness to engage in voluntary exercise and present new behaviors to interrogate.

The findings described here should be considered with the following caveats. The ability to assess activity-induced changes in EE was limited by the temporal resolution of gas analysis (5-min intervals). These studies examined a relatively acute time frame for VWR, and it is unknown whether these behavioral changes persist over longer periods of time. Chow-fed male mice were studied, so it is unknown whether obese mice in positive energy balance would respond similarly. Finally, it is unknown whether female mice have the same behavioral and metabolic responses to VWR.

The use of rodent models to study voluntary metabolic and behavioral responses to VWR has been constrained by an inability to detect the contributions of VWR and OWA to EE, distinguish between context-specific behaviors, and account for individual variability in VWR performance within a group of animals. By addressing these gaps in our understanding of voluntary exercise metabolism and behavior, our findings (summarized in Fig. 8) provide the most comprehensive assessment to date of the metabolic and behavioral response to voluntary exercise and provide a framework for deconstructing the complex regulation of energy balance in mice.

Figure 8.

Reduced nonexercise physical activity preserves energy balance in response to exercise. Top: In the absence of exercise, energy intake is matched with EE and the resulting maintenance of body mass over time. Middle: Voluntary exercise increases EE that would be predicted to be additive for EE and thereby induces a pronounced negative energy balance. Bottom: The change in energy balance with voluntary exercise is less than predicted because of a reduction in nonexercise physical activity.