Physical activity is directly associated with total energy expenditure without evidence of constraint or compensation

Significance

Our paper should be of interest to broad readership in that it provides insight into the relationship between physical activity (PA) and total energy expenditure (TEE) across a wide range of PA levels including that habitually performed by ultraendurance runners. Our study provides evidence for a positive linear relationship between PA and TEE independent of fat-free mass across a wide range of PA levels. PA was inversely related to sedentary behavior, but there was no relationship between PA, TEE, or resting metabolic rate and adjusted biomarkers of immune, reproductive, or thyroid function. Taken together, our observations are inconsistent with constraint or compensation in total energy expenditure. Instead, our findings suggest that PA directly adds to TEE in humans.

Abstract

The prevailing linear model of physical activity (PA) and total energy expenditure (TEE) has been challenged by models that predict an upper limit of TEE linked to a compensatory reduction elsewhere in the energy budget in response to increased PA. We determined the equation of best fit between PA and TEE and explored relationships between PA and behavioral and physiological compensation. Using linear and nonlinear modeling, we observed a positive linear relationship between PA and TEE either without or after adjustment for fat-free mass (R²= 0.3492, TEE = 0.00685*PA + 7.124: R²=0.3667, TEE_ADJ(FFM) = 0.00511*PA + 8.598). Higher PA was associated with lower sedentary time (R²= 0.7207, %SPA= −0.0211*X + 91.261). There was no association between PA, TEE, or resting metabolic rate and adjusted biomarkers of immune, reproductive, or thyroid function after Bonferroni correction. The findings of this observational study do not support the constrained/compensated model but affirm the conventional additive relationship between PA and TEE across a broad range of PA levels.

The health benefits of increasing energy expenditure with physical activity are well documented (1), but how physical activity influences the energy budget remains controversial (2–4). The prevailing additive model predicts a linear relationship in which increasing levels of PA lead to linear increases in energy expenditure which adds directly to TEE (5). In contrast, the constrained model posits that while low-to-moderate levels of PA lead to increases in TEE, higher levels of PA do not result in further increases in TEE (i.e., the maximal TEE is constrained, reflecting the existence of an upper limit). This also implies energetic trade-offs elsewhere in the energy budget (5). This constrained model may also be regarded as driving compensatory reductions in other components of expenditure, but compensation may also occur in the absence of a constraint (compensation model) leading to a nonlinear relation between PA and TEE even before a hypothetical constrained limit on TEE has been reached (4).

Sustained maximal energy expenditure is often expressed as the quotient of total energy expenditure (TEE) and resting metabolic rate (RMR) and is referred to as metabolic scope (MS). Several mechanisms have been suggested to impose constraints on the maximal MS including an alimentary limit (6–8), a limit due to heat dissipation capacity (9), and limits imposed by the metabolic capacity of different body tissues (10, 11). Petersen et al. examined metabolic ceilings of a variety of species and observed a typical limit between 1.5 and 5.0 times basal metabolic rate (BMR) among endotherms including humans (12). Pontzer et al. (13) observed an asymptotic relationship between PA and TEE which reached an asymptote at an MS of ~1.68× RMR. Thurber et al. (6) reported MS of ~2.5 to 3× RMR based on pooled data from extreme endurance events. Higher MS were proposed to be limited by alimentary energy supply resulting in increased reliance on body fuel reserves and body mass loss (6). Some elite endurance athletes, including professional cyclists (14–16) and collegiate (17) and elite open water swimmers (18) have been reported to achieve MS as high as 3 to 6× RMR for days to weeks with little or no body mass loss. However, these observations were limited to a small number of individuals. These athletes may have been uniquely able to meet their energy requirements through a carefully planned and implemented fueling strategy (19) allowing them to circumvent alimentary limits, and/or experienced increased convective heat loss associated with their chosen mode of PA thereby alleviating any heat dissipation constraint (9). Nevertheless, these data suggest that the relationship between PA and TEE is not clearly defined, particularly among weight-stable individuals habitually performing large volumes of PA (14, 17, 20, 21).

The purpose of the present study was to examine the relationship between habitual PA and TEE across a wide range of habitual PA levels in weight-stable adults to test between the additive, constrained, and compensatory models. The additive model predicted the relationship between PA and TEE would be linear, while constrained and compensatory models both predicted an asymptotic curve. The position of the asymptote would be approximately 2.5 to 3× BMR if it was due to constraint, but lower if only compensation was involved. We further sought to directly establish mechanisms of compensation by determining whether discretionary behavioral or physiologic processes were downregulated at high levels of physical activity by examining the relationships between PA and sedentary behavior, RMR, and select biomarkers of immune, reproductive, and metabolic function (3, 22–24).

Results

Sample Characteristics.

We enrolled 75 adults (37 female) aged 18 to 63 y from the west central region of Virginia (latitude 37° N) who represented a spectrum of PA levels based on their self-reported weekly walking or running (0.0 to 128.7 km/wk). Three participants were aged 61 to 63 y, and while energy expenditure has been shown to decline at 63.0 y (95% CI: 60.1,65.9) (25), their inclusion did not affect study outcomes. The sample reflected greater diversity (71.9% white) than observed in the region (Montgomery County is 86.1% White) (26). More than 80% of participants reported living in Montgomery County or in an adjacent County or City which on average is > 600 m (251 to 1,315 m) above sea level (27). Sample characteristics by sex are shown in Table 1.

Table 1.

Participant characteristics

	Mean (SD)
	Female	Male	All
N	37	38	75
Age (years)	35 (10)	38 (12)	37 (11)
Body Mass (kg)	59.6 (6.39)	74.4 (10.03)	67.1 (11.22)
Height (cm)	164.3 (5.4)	180.0 (7.1)	172.3 (10.1)
Body Mass Index (BMI)(kg/m2)	22.1 (1.96)	22.9 (2.51)	22.5 (2.28)
Fat Free Mass (FFM)(kg)	44.4 (4.61)	61.4 (7.25)	53.0 (10.5)
Fat Mass (FM)(kg)	15.5 (4.94)	13.4 (4.93)	14.4 (5.01)
% Fat Mass (%FM)	26.7 (6.46)	18.5 (5.16)	22.5 (7.13)
PA(VA CPM)(physical activity)(vector magnitude counts/min)*	773 (222.5)	838 (242.6)	806 (233.4)
PA (physical activity)(km/week)	49.5 (33.68)	52.4 (30.89)	51.0 (32.11)
Maximal Oxygen Consumption (VO2 max)(mL/kg/min)	46.0 (8.94)	54.6 (9.28)	50.3 (10.03)
Resting Metabolic Rate (RMR)(kcal)	6.03 (0.636)	7.52 (0.886)	6.79 (1.073)
Total Energy Expenditure (TEE)(MJ/day)†	11.29 (2.250)	14.15 (2.478)	12.76 (2.761)
Metabolic Scope (MS)(TEE/RMR)†	1.87 (0.270)	1.89 (0.330)	1.88 (0.300)
% Time Sedentary (%SPA)	74.6 (6.17)	74.0 (5.45)	74.3 (5.79)

^*N = 74 (female = 37) due to lost damaged accelerometer, n = 1 excluded from analysis that included VA CPM as a variable.

^†N = 74 (female = 36) due to implausible TEE values n = 1 excluded from analysis that includes TEE as a variable or values calculated from TEE (i.e., MS and AEE).

There was no difference in age, objective PA using accelerometry [vector magnitude counts per minute, PA(VM CPM)], or subjective PA as self-reported miles per week (converted to km) running or walking [PA(km/week)] between males and females. Nor was there a difference between sexes in metabolic scope (MS) (TEE/RMR) or percent of waking hours spent in sedentary physical activity (%SPA, < 100 vector magnitude counts per minute). We achieved wide and similarly distributed spectra of PA in both males and females (Fig. 1).

Fig. 1.

Physical activity distribution by sex (t tests). (A) N = 74, Mean physical activity measured in vector magnitude counts per minute using accelerometry [(PA(VM CPM)] did not differ by sex (P = 0.2339) and was similarly distributed. (B) N = 75, Mean physical activity in self-reported usual km per week running or walking [PA (km/week)] did not differ by sex (P = 0.6953) and was similarly distributed.

The majority of participants were enrolled in the spring and winter months (SI Appendix, Fig. S1A). There was no difference in mean PA(VM CPM) of the participants across seasons by category (SI Appendix, Fig. S1B), nor was there a relationship between environmental temperature and reported weekly distance running or walking (SI Appendix, Fig. S2A) or PA (VM CPM)(SI Appendix, Fig. S2B).

Repeated height, weight, and densitometry measures were obtained on day 1 and day 15 of doubly labeled water (DLW) measurements. To establish that participants were in energy balance, energy stores were calculated based on fat mass and lean mass from densitometry (28). Comparing these time points %CV was < 1.0 for these measures. Repeated measures of RMR were also performed to document reproducibility. RMR was highly reproducible (%CV 4.3, SI Appendix, Table S1).

Calculation of total energy expenditure using assumed and estimated respiratory quotients.

The calculation of TEE from DLW may be confounded by using the assumed respiratory quotient (RQ) if there are deviations in macronutrient intake from the typical diet (29, 30). As such, we calculated TEE using rCO2 from DLW analysis and the assumed RQ of 0.85 versus RQ estimated from habitual dietary intake analysis. Using the equation described by Elia, (30) we calculated estimates of RQ from food quotients (FQ) using three dietary recalls [TEE(recall)], 4 d of food intake records [TEE(FIR)] as well as the mean from the combined dietary assessment measures [TEE(FQ)] for each participant. We then computed TEE using the Weir equation (31), combining rCO₂ from DLW analysis, and the four estimates of RQ as described. There were no differences in TEE calculated using the assumed RQ (0.85) vs using each of the three estimates of RQ from dietary assessments (SI Appendix, Fig. S3).

Recent Physical Activity and RMR.

We calculated the total time in moderately vigorous PA (MVPA) bouts 12 to 24 h prior to measuring RMR at 15 d to determine whether allowing exercise up to 12 h prior to RMR was associated with RMR (Fig. 2). There was no correlation between minutes spent in bouts of MVPA 12 to 24 h prior to RMR and RMR whether unadjusted (Fig. 2A) or adjusted for FFM (Fig. 2B). Furthermore, there was no association between minutes spent in exercise bouts 12 to 24 h prior to RMR and the residuals of actual vs predicted RMR using either the Pontzer et al. (25) equation (RMR= −0.954 + 0.707 * Ln (FFM) + 0.019 * Ln (FM)) (Fig. 2C) or the Mifflin, et al. (32) equation (RMR = 413 + 19.7 * FFM) (Fig. 2D).

Fig. 2.

Time spent in moderately vigorous PA 12 to 24 h before RMR. Minutes in MVPA 12 to 24 h before RMR testing did not correlate with RMR or explain variability in actual vs. predicted RMR. (A) Simple linear regression, N = 71, RMR regressed on minutes in MVPA bouts, P = 0.8219 (B) RMR adjusted for fat-free mass RMR_Adj(FFM) regressed on minutes in MVPA bouts, P = 0.5553. (C) The residuals of actual vs predicted RMR using the Pontzer et al. equation P = 0.6121 (D) and using the Mifflin equation, P = 0.7309 were not associated with the time in MVPA in the 12 to 24 h preceding the RMR.

Physical activity and energy expenditure.

The relationship between PA(VM CPM) and total energy expenditure (TEE) (Fig. 3A), unadjusted and adjusted for fat-free mass [TEE_Adj(FFM)] (Fig. 3B), RMR [TEE_Adj(RMR)] (Fig. 3C) or using MS as the dependent variable (Fig. 3D) was best characterized by a linear model (SI Appendix, Tables S2 and S3). Similarly, we modeled PA(VM CPM) on RMR unadjusted (Fig. 3A) and adjusted for FFM (Fig. 3B), and there was no significant relationship.

Fig. 3.

The relationship between PA and TEE (N = 73) and RMR (N = 74) using simple linear regression. (A) There was a positive linear relationship between PA(VM CPM) and but not RMR (P = 0.1335). (B) The relationships between PA(VM CPM) and TEE and RMR (P = 0.7233) were unchanged after adjusting for FFM. (C) There was a positive and linear relationship between PA(VM CPM) and TEE adjusted for RMR. (D) There was a positive and linear relationship between PA(VM CPM) and metabolic scope (TEE/RMR). ***P < 0.0001.

Physical activity energy expenditure and sedentary behavior.

High levels of activity energy expenditure (AEE = 0.9*TEE – RMR) may be compensated for by increasing the amount of time spent in sedentary behavior (i.e., behavioral compensation) (24, 33). In contrast, we observed a strong positive linear relationship between PA(VM CPM) and AEE unadjusted or adjusted for FFM (Fig. 4 A and B). There was a strong negative relationship between PA(VM CPM) and % time in sedentary PA (%SPA, <100 VM CPM by accelerometry) (Fig. 5A) as well as between PA (km/week) and % time in SPA (%SPA) (Fig. 5B). Individuals with PA activity levels spent a lower percent of their waking hours in sedentary PA. At the same time, PA was positively associated with percent time spent in low (R² = 0.224446, P = <0.0001) and moderate PA (R²= 0.173139, P = 0.0002).

Fig. 4.

The relationship between PA and AEE (N = 73), simple linear regression. (A) Positive and linear relationship between PA and AEE. Y Intercept (0.2607, 95% CI: −0.9346) was not different from 0. (B) The relationship between PA and AEE was not altered by adjusting AEE for FFM. Y Intercept (0.8698, 95% CI: −0.2915,2.0310) was not different from 0. ***P <0.0001.

Fig. 5.

The relationship between PA and % of accelerometry wear time spent sedentary, simple linear regression, N = 74. (A) There was a strong negative correlation between PA(VM CPM) and %SPA. (B) There was a significant negative correlation between PA (miles/week) and %SPA. ***P < 0.0001.

Circulating biomarkers of inflammation, stress, and reproductive and metabolic function.

High levels of physical activity may induce physiological compensation in the form of reduced energy expended in other physiologic domains such as inflammatory and stress responses (22, 23, 34) and reproduction (35–37). If this were the case, lower circulating inflammatory or reproductive biomarkers would be predicted at higher levels of energy expenditure.

We performed prediction screening for an array of serum, plasma, and whole blood biomarkers that have been reported to relate to PA (SI Appendix, Table S6) (22, 23, 34–38). We adjusted these biomarkers for the most predictive baseline characteristics (anthropometric measures, sex, and age) or season. PA(VM CPM) was positively correlated with absolute serum c-reactive protein (CRP) (SI Appendix, Fig. S4A) and CRP adjusted for season (SI Appendix, Fig. S4B). These relationships were no longer significant after applying a Bonferroni correction for multiple comparisons (α = 0.002) (SI Appendix, Table S7). Among the unadjusted and adjusted biomarkers, there were no significant relationships identified with TEE_Adj(FFM) (SI Appendix, Table S8) and similarly no correlation between adjusted circulating markers and RMR_Adj(FFM) or AEE_Adj(FFM) (SI Appendix, Tables S9 and S10).

Ambient temperature and TEE.

While there was a positive linear relationship between environmental temperature and PA(VM CPM), there was no significant relationship between environmental temperature and MS or TEE adjusted for FFM (Fig. 5 A and B. However, adjusting TEE for both FFM and PA(VM CPM) resulted in a negative correlation, but adjusting MS for PA(VM CPM) did not (SI Appendix, Fig. S5 C and D).

Discussion

The major findings of our study were severalfold. First, we observed a positive linear relationship between PA and both TEE and MS. These relationships remained significant after adjusting TEE (or MS) for FFM or RMR. This relationship is consistent only with the additive model and does not support the constraint or compensatory models. Second, there was no significant relationship between PA and RMR regardless of whether adjusted for FFM or not, suggesting no compensatory change in RMR as PA increased. Moreover, there was a positive linear relationship between PA and AEE which is also inconsistent with both constraint and compensatory models. Together, these additional data provide further support for the additive model by indicating no compensatory changes in the resting expenditure as level of activity increased. Third, we observed that PA was positively associated with %SPA. This was opposite of the prediction of the constrained and compensatory models. Finally, we observed that higher levels of PA may be associated with lower adjusted absolute monocyte counts and cortisol and higher c-reactive protein but not with any other inflammatory, reproductive, or metabolic biomarkers that we examined. In addition, there was no relationship between any of the circulating biomarkers and adjusted TEE or its components. This suggests that these changes while significant do not contribute in any meaningful way to the overall energy budget. Taken together, these observations support the additive model of impacts of PA on TEE and do not support the constrained or compensatory models.

The results of previous studies focused on athletes engaged in endurance and ultraendurance events have demonstrated that high levels of TEE (MS > 5.0) can be achieved and sustained for ≥ 10 d (6, 16, 20, 39). However, a compilation of previous studies involving extreme athletic events demonstrated that MS at this level or higher was associated with body fat loss (20). In addition, the results of clinical (6, 20) and population (40) studies suggested a limit to TEE of ~ 2.5× RMR (6) with little or no change in body weight (20). Our observation of MS of < 2.5 in our sample of weight-stable individuals is consistent with these observations. It could be argued then that the reason we did not observe any constraint in the relation between TEE and PA was because the level of activity was insufficient to push individuals up to the level where constraints apply. It is important to note that the most active individuals in our sample were running more than 72 km per week (0.0 to 128.7 km/day, AEE = 0.51 to 6.1 MJ/day) or more than 10 km per day and had VM CPM above 1,000. If we extrapolate the relationship between PA and activity out to an MS of 2.5, that would require about twice these levels of activity. Hence, while it is theoretically possible that the constrained model is still correct, the predicted levels of activity necessary to achieve the constraint lie well above the activity levels of the overwhelming majority of individuals—including those generally considered to be extremely active. It is worthy of note that the first paper in which constraint in relation to PA was introduced (3) and tested using the same types of measure utilized here (DLW and objectively measured PA by accelerometry), the constraint was suggested to become apparent at a VM of around 200, which is a lower level of activity than the most sedentary individuals in our cohort.

Physical activity may result in energy compensation via behavioral or metabolic mechanisms (3, 13). We observed a positive linear relationship between PA and AEE and between both PA (and AEE) and %SPA. In contrast, the results of a meta-analysis by Franssen et al. (24) indicated that athletes spent more time sedentary than less active individuals. The reasons for this discrepancy are not clear but may be related to methodological differences between studies; the majority of studies included in the Franssen et al. (24) meta-analysis did not objectively measure PA or sedentary behavior. Neither non-exercise activity thermogenesis nor nonexercise PA (NEPA) was measured in the present study. However, we observed that percent time spent in light (100 to 1951 VM CPM) and moderate (1952 -5724 VM CPM) physical activity was positively correlated with PA suggesting no compensatory reduction in NEPA.

Klasson et al. (22) reported that higher levels of accelerometry-derived PA were associated with lower inflammatory markers (total and differential white blood cell count, fibrinogen, c-reactive protein) and markers of thyroid function (thyroid-stimulating hormone and free T4). Klasson et al. (22) also hypothesized that these changes were due to suppressed energy expenditure of other physiological systems resulting from increased physical activity and, in turn, contributed to compensation. Although we also observed downregulation of some unadjusted biomarkers in relation to PA, we found no relationship between these biomarkers and either TEE or RMR.

There are several notable strengths of our study. Our sample was relatively large given the careful phenotyping using rigorous state-of-the-art methods including DLW, indirect calorimetry, accelerometry, and dual energy x-ray absorptiometry (DXA). Multiple approaches were used to assess RQ/FQ. Our participants reported being body weight stable for the prior 6 mo and subsequently maintained their body weight throughout the DLW dosing period. The stability of body composition and RMR were also verified during the DLW dosing period. Importantly, our findings should be considered in the context of a state of energy balance (4, 6, 20, 41, 42).

There are also some limitations of our study. First, it was cross-sectional in nature, as such, we cannot conclude causality in our observations. Second, there may have been compensation that occurred in other aspects of the energy budget not considered or directly measured in the present study. For example, we did not consider circadian fluctuations in RMR; the actual RMR may have been lower resulting in an underestimation of AEE (3, 43–46). Furthermore, the thermic effect of food was not measured in the present study but rather assumed to comprise 10% of TEE. However, the thermic effect of food has been reported to be higher or similar, not lower, in endurance-trained athletes compared with controls (46–49). As such, change in TEF seems an unlikely candidate for a trade-off with increased AEE. Finally, there could be error in our measurement of PA despite the use of a validated accelerometer (2, 50, 51).

In summary, our findings were inconsistent with the constrained or compensatory models of PA effects on TEE but rather support an additive model, whereby increasing energy expenditure spent on activity is linearly added to a fixed resting expenditure that is independent of activity levels.

Materials and Methods

Study Design.

Our general approach was to use linear and nonlinear modeling to determine the nature of the relationship between habitual physical activity (PA) and total energy expenditure (TEE). To that end we implemented a cross-sectional study and measured TEE across a wide spectrum of PA level in healthy weight-stable adults.

Screening and Enrollment.

Following Virginia Tech institutional review board approval (#21-567) in July 2021, employing a stratified approach investigators recruited participants ensuring a balanced distribution of sex and PA levels from sedentary (no routine formal exercise) to ultra-endurance-trained individuals. Recruitment and enrollment spanned from October 2021 through April 2023.

We conducted initial screening using an online web-based survey software (Qualtrics (Provo, UT) and QuestionPro (Austin, TX) and conducted virtual screening and consenting visits using a privacy and security compliant institutional Zoom for MacOS 5.8.0 (San Jose, CA) account. Written and verbal informed consent was obtained from all participants prior to their enrollment, in accordance with institutional and ethical guidelines. We reviewed the participant’s health history to exclude health conditions, medications, or other criteria incompatible with the study (Table 2). We collected additional subjective PA data including reported typical number of run or walked miles in a week. Finally, we emailed participants a standard health history and the Godin Leisure Exercise Questionnaire (52) to be completed and returned at the first in person visit.

Table 2.

Inclusion and exclusion criteria for the study

Inclusion criteria

Aged 18 to 65 y

Stable body weight over last year (± 5 pounds)

Sedentary, recreationally active, or competitive endurance or ultra-endurance-trained athlete

Exclusion Criteria

Cardiovascular, pulmonary, metabolic, kidney, or malignant neoplastic disease

Pregnancy or plans to become pregnant during the study period

Recent injury or condition or change that resulted in change of physical activity in the last 12 mo.

Medications that may impact the outcomes of the study including but not limited to beta-blockers or blood glucose–lowering agents.

Smoking

*Serum LDL cholesterol > 160 mg/dL

Serum triglycerides > 500 mg/dL

*Blood pressure ≥ 140 mmHg systolic or 90 mmHg diastolic

BMI≥ 30 kg/m2

^*Otherwise healthy and highly active individuals who met moderate risk ACSM & PAR-Q + criteria (e.g., resting systolic blood pressure ≥140 and ≤159 or diastolic blood pressure ≥90 and ≤99; or LDL cholesterol >160 and <200) were required to obtain clearance by their primary care physician or licensed clinician to participate in the exercise capacity testing.

Sample.

We recruited 75 healthy male and female participants aged 19 to 63 across a spectrum of physical activity levels ranging from sedentary-to-recreationally active to ultraendurance athletes. The athletes had participated in at least one ultraendurance running event (any event longer than a marathon length of 26.1 miles (42 km) (53) in the past year. We enrolled participants stratified according to their self-reported typical weekly mileage from running or walking. We sought to achieve an equal distribution of running/walking volumes across 10-mile (16 km) increments while maintaining a balance between males and females across the spectrum of physical activity. Table 2 shows inclusion and exclusion criteria.

Procedures.

General timeline.

We obtained anthropometric measures, body composition, resting metabolic rate and urine, serum, and whole blood analyses on day 1. Over the course of 2 wk, we measured total energy expenditure, physical activity, and obtained dietary histories. We repeated anthropometric measures, body composition, and resting metabolic rate and measured $\dot{V}$ O₂ max on day 15.

Anthropometric measures and body composition.

We measured body mass (to nearest 0.1 kg) and height (to nearest 0.5 cm) (Scale-Tronix 5002, Welch Allyn, Inc, Skaneateles Falls, NY) in triplicate on days 1 and 15, with participants in laboratory-provided preweighed light clothes. We measured body composition (fat mass (FM), fat-free mass (FFM), and % FM) on days 1 and 15 using dual-energy x-ray absorptiometry (DXA) (General Electric, Lunar Digital Prodigy Advance, Madison, WI) and analyzed with enCORE software (version 15, GE Healthcare, Madison, WI).

Resting metabolic rate.

We measured resting metabolic rate using indirect calorimetry (Parvo Medics, TrueOne 2400 Metabolic Measurement System, OUSW 4.3.4; Murray, UT) as described previously (54, 55). We fit participants with a ventilated canopy in the supine position in a dimly lit and temperature-controlled room (22 to 24 C) following a 12-h fast and at least 12 h after the last exercise training session in runners to not interfere with their habitual physiological state. The last 30 min of a 45-min period was used for analysis. We performed measurements of RMR on two occasions, separated by 14 d during which body mass was stable. We used the mean of these measurements for our analysis.

Maximal oxygen consumption.

Maximal oxygen consumption was measured during graded treadmill exercise to volitional exhaustion using open-circuit indirect calorimetry (TrueMax 2400, Parvo Medics, Salt Lake City, Ut). Standard criteria for valid achievement of maximal oxygen consumption were met (56).

Accelerometry.

To objectively assess physical activity, we used a waist mounted ActiGraph triaxial accelerometer (wGT3X+, ActiLife v6.13.4 software, Pensacola, FL).

Triaxial accelerometry detects and records accelerations in three axes: axis 1 = vertical, axis 2 = horizontal, axis 3 = perpendicular. The ActiGraph wGT3X also detects ambient light (lux). Analysis of the postprocessed data produces step counts from vertical axis acquired data. The orientation and timing of removal of the device can also be determined by the inclinometer in postprocessing. Each epoch is interpreted as device off 0), Standing 1), lying horizontal 2), and sitting 3).

We set the sample rate to 30 Hz. On day 15 we uploaded the data to a computer with ActiLife software installed creating AGD(wGT3X-BT) files with epoch lengths of 10 s, data from all 3-axes, step counts, lux, and inclinometer data. We scored the data using the following algorithms: Freedson combination (57) for energy expenditure, Freedson Adult (51, 57, 58) for METs and vector magnitude counts per minute cut points (sedentary: 0 to 99; light: 100 to 1,951; moderate: 1,952 to 5,724; vigorous: 5,725 to 9,498; very vigorous: 9,499 and above). We used only the standard filter (51).

We employed global date and time filters beginning and ending with the predetermined start and stop times and entered wear time manually using the participant wear time diary. We resolved any ambiguities with the participant and reconciled any remaining gaps in the diary by comparison with the ActiLife software’s wear time algorithm. The daily report was examined and only days that contained at least 600 min (10 h) of wear time were considered as valid days and included in the analysis in keeping with standard practices (58).

Doubly labeled water isotope elimination.

We assessed total energy expenditure using the doubly labeled water technique.

Doubly labeled water (DLW) is water that is enriched with stable isotopes of hydrogen (²H) and (¹⁸ O) oxygen. These isotopes occur naturally in all water sources at low levels of enrichment and can safely be ingested by humans. By precisely measuring the baseline enrichment of ²H and ¹⁸ O in body fluids, most often urine, we can detect their disappearance following ingestion over time with serial urine sampling and analysis.

The difference between the elimination of the two isotopes in urine yields exhaled O₂ as CO₂ since O₂ is eliminated both in the form of water in body fluids and exhaled CO₂ whereas H is lost only as water. When the ratio of CO₂ production (rCO₂) to O₂ utilization (rO₂)—known as the respiratory quotient (RQ)- are known, total energy expenditure can be calculated using the Weir equation (31).

$$\begin{array}{c}TEE (MJ/d)={\mathit{rCO}}_2\left[1.106+(3.94/RQ)\right]4.184/103.\hfill \end{array}$$ [1]

DLW mixing and dosing procedures.

DLW was mixed to a ratio of 2 to 1 of PPM of ¹⁸O to ²H in the final dilution. We determined the working enrichment for dosing purposes from the certificate of enrichment (¹⁸O) (Sigma Aldrich, CAS # 14314-42-2, St Louis, MO) or the label (² D) (Sigma Aldrich, CAS# 7789-20-0). We filtered all isotope enriched water using a 0.45 μm filter and a sterile 50-cc syringe and placed the mixture in sterile autoclavable glass bottles and sterilized the batches of DLW at 121 °C and retained samples of each batch for analysis.

We based individual dosing on day 1 body mass in g using the following equation in which enrichment refers to ¹⁸O isotope = 120 ppm and background enrichment $\approx$ 100,000 ppm

$$\begin{array}{c}Dose,g=(body mass,g x desired excess enrichment)/dose enrichment.\hfill \end{array}$$ [2]

As a range of dosing is acceptable, there is room for a degree of imprecision in calculating the putative dose, however in measuring the actual ingested dose a high degree of precision is required. We employed a multistage approach to obtain serial measurements.

The rate of isotope disappearance was determined using the plateau method as described by Speakman (59). Using this sample approach, we obtained a 3-h urine sample post-DLW ingestion, then daily first morning urine samples for 2 wk aiming for consistent daily timing. Urine specimens and times were kept in a spread sheet and time from DLW dosing was calculated for each sample.

Urine was aliquoted into 4 2 mL cryotubes for each daily collection and held at −80° C and shipped on dry ice to the University of Aberdeen Doubly Labeled water laboratory for analysis using the ABB Los Gatos liquid water triple isotopic water analyzer (TIWA) (San Jose, CA) by an experienced team. CO₂ production was calculated using the following (60) equation in which N refers to the total body water, ko and kd refer to the rates of oxygen and hydrogen isotope loss, respectively.

$$\begin{array}{c}{\mathit{rCO}}_2(L/d)=[N\times ((0.45859x{k}_o)(0.47498 x k_d))]\times 22.26.\hfill \end{array}$$ [3]

Dietary assessment.

Using the Weir equation (31), RQ is required to compute TEE. This ratio of oxygen consumed to carbon dioxide produced can be measured using indirect calorimetry, estimated from the proportions of dietary macronutrients (fats, carbohydrates, and proteins ± alcohol)—known as food quotient (FQ)–or assumed based on the typical dietary composition of a population. For a typical Western diet, RQ is presumed to be between 0.84 and 0.86. The use of FQ may be a strategy to mitigate error compared to the use of an assumed RQ, especially if there is significant dietary diversity or departure from a standard American diet of ~ 55% carbohydrate intake (61). While estimating energy intake by dietary history is error-prone, dietary macronutrient composition can be accurately estimated with careful dietary intake histories (62).

We employed a combination of three 24-h dietary recalls (recalls) using a multiple pass method and four-day food intake records (FIRs) furnished by the participants (63). We input the data into Nutrition Data System for Research software (2021 version, Nutrition Coordinating Center, University of Minnesota) as soon as practical (typically within 72 h) and created averaged reports for recalls and for FIRs accounting for macronutrient content (absolute and % of dietary intake) and energy intake. We calculated FQ using recalls and FIR separately and as a combined average using the following equation described by Elia et al. (30) and implemented by Hall et al. (61) and others.

$$\begin{array}{c}FQ=(1.0\times CHO(kcal)+(0.7\times Fat(kcal))+(0.835\times Pro(kcal))\hfill \\ +(0.67\times ETOH(kcal).\hfill \end{array}$$ [4]

We computed TEE in kcal using an assumed RQ of 0.85 typically reflective of an American diet and using FQ at individual levels in the Weir equation (31).

Statistical Analysis.

We performed all analyses not otherwise described using JMP 16.2.0 (570548) statistical software, SAS Institute Incorporated (North CA). We performed repeated measures ANOVA with post hoc Tukey analysis and created all figures unless otherwise indicated in GraphPad Prism 10 for macOS Version 10.2.0 (335), GraphPad Software, LLC (Boston, MA, www.graphpad.com).

Repeatability of Variables.

For repeated anthropometric measures and resting metabolic rate, we calculated individual means, SD, and coefficients of variation (%CV) to determine the repeatability of measures by timepoint or method at an individual level and averaged to determine overall repeatability.

Outliers and Data Exclusion.

For variables obtained from accelerometry (%SPA and VM CPM), we considered data valid if there were 8 valid days present. A day was considered valid if there was ≥ 10 h of wear time (58). Three individuals did not meet these wear time criteria. Nonetheless, we compared analyses with and without these individuals and found no meaningful differences in outcomes. One individual returned with a broken accelerometer resulting in the loss of data and was therefore eliminated from any analysis that included PA(VM CPM). In the analysis of FQ vs RQ in the calculation of TEE, there was 1 individual for whom 24-h food recalls were missing. We analyzed the data imputing recall data that were calculated using the mean differences between recalls and FIR variables.

We performed outlier diagnostics using Cook’s D and identified no highly influential outlier using this method. We eliminated the TEE data for one subject from the final analysis as the results of the DLW analysis were considered physiologically implausible. As a result, N = 73 for analysis of TEE and AEE though exclusion of this datum did not alter any of the conclusions.

Linear Regression.

We screened potential predictor variables using JMP software and performed simultaneous forward and backward regression with candidate predictor variables (FFM, FM, % FM, BMI, Height, Age, and season) for TEE, RMR, AEE, %SPA, and serum, plasma, and whole blood biomarkers. Statistically significant (P < 0.05) or clinically meaningful variables that increased R² were initially included in the model. Multicollinearity was determined using the variance inflation factor (VIF) for each variable and predictors were eliminated as required to eliminate collinearity. The predictors in models with the lowest Akaike Information Criterion (AIC, differences < 2 were considered equivalent) were used to adjust the outcome variables (64). We regressed PA (VM CPM) as an objective measure on TEE, RMR, and AEE and on the array of biomarkers using linear and nonlinear polynomial modeling. We established a critical value of P for multiple comparisons of circulating biomarkers using Bonferroni correction (α/number of comparisons). To eliminate the effect of significant baseline characteristics, we adjusted outcome variables by summing the variable mean to the residuals. We presented TEE, AEE, and RMR as an absolute value and adjusted for FFM. We also presented TEE adjusted for RMR.

To account for potential variability in RMR due to differences in exercise in the 12 to 24 h prior to RMR assessment, we calculated the total time in moderately vigorous PA (MVPA) bouts 12 to 24 h prior to measuring RMR at the 15 d measurement point to determine whether exercise up to 12 h prior to RMR correlated RMR values. We regressed mean time in MVPA bouts defined by Freedson et al. on RMR and RMR adjusted for FFM. Next we determined expected RMR for each participant using the well-established Mifflin et al. equation for RMR and the Pontzer equation which we previously published as the equation that best fit our data (54). Using linear regression we determined the residuals between expected and actual RMR then regressed PA(MVPA) bouts 12 to 24 h prior to measuring RMR on those residuals.

Repeated Measures ANOVA.

To determine the effect of dietary macronutrient composition on the calculation of TEE, we obtained macronutrient intake to calculate FQ using 24 h Recalls (3 d) and FIRs (4 d). We then calculated TEE using the Weir equation with rCO₂ from isotope elimination and either an assumed RQ of 0.85, TEE(RQ) or FQ TEE(recall), TEE(FIR), and TEE(FQ). We compared all means using repeated measures ANOVA. As there were no family-wise differences, post hoc analysis was not performed. We performed the main outcome analysis using TEE (FQ).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Partial restrictions apply to the anonymized daily energy expenditure data, as it is available in the International Atomic Energy Agency International Doubly Labelled Water Database via application. To request data access, please see: https://www.iaea.org/resources/hhc/nutrition/databases/double-labelled-water-dlw/how-to-request (65).