A consensus guide to preclinical indirect calorimetry experiments

Abstract

Understanding the complex factors influencing mammalian metabolism and body weight homeostasis is a long-standing challenge requiring knowledge of energy intake, absorption and expenditure. Using measurements of respiratory gas exchange, indirect calorimetry can provide non-invasive estimates of whole-body energy expenditure. However, inconsistent measurement units and flawed data normalization methods have slowed progress in this field. This guide aims to establish consensus standards to unify indirect calorimetry experiments and their analysis for more consistent, meaningful and reproducible results. By establishing community-driven standards, we hope to facilitate data comparison across research datasets. This advance will allow the creation of an in-depth, machine-readable data repository built on shared standards. This overdue initiative stands to markedly improve the accuracy and depth of efforts to interrogate mammalian metabolism. Data sharing according to established best practices will also accelerate the translation of basic findings into clinical applications for metabolic diseases afflicting global populations.

Fig. 1:. Standard visualizations of body composition in diet-induced obesity.

Littermate male C57BL/6J mice were divided into two equal groups at week 2 to receive either a HFD (green) or LFD (yellow). a, Weekly group averages for body weight (g) of mice on the LFD or HFD. The dashed box represents stable weight change in both groups. The slopes of weight change per week are indicated. Body composition was determined at the conclusion of the study. b, Plot of body composition showing the absolute amount of fat, lean and total mass (g). Other body mass (for example, bone and skin) that was not detected by qNMR is not shown. c, A plot showing percentage body composition, which might be misinterpreted to indicate a lower absolute amount of lean mass in HFD-fed mice. The tick and warning icons in b and c indicate appropriate and inappropriate analyses, respectively. Statistical comparisons by unpaired t-test. ***P < 0.001, n.s., not significant, n = 16 mice per group. Data are shown as mean ± s.e.m.

Fig. 2:. Visualizing indirect calorimetry results.

In an ideal experimental scenario, indirect calorimetry would be performed before body weights had diverged. In this example, body weights of littermate male C57BL/6J mice are significantly different between groups (HFD, green; LFD, yellow). a–j, Plots of energy expenditure (kcal h⁻¹), energy intake (kcal h⁻¹), energy balance (kcal) and statistical analysis over 48 h (grey bars indicate dark photoperiods). Ten weeks after randomization to HFD or LFD, the mice were placed in an indirect calorimeter. Data versus time are plotted as the group mean ± s.e.m. for each diet, as represented by the line and ribbon, respectively. Data are plotted every 5 min with a 6-point rolling-window (30-min) smoothing function applied to the visualizations, but not to values for statistical analysis. a, Energy expenditure versus time. b, Boxplots representing the 25th percentile, median and 75th percentile of the daily energy expenditure per mouse. Points represent each mouse. c, Cumulative energy expenditure. d, Energy intake (that is, food mass × energy content of food) versus time. e, Boxplots representing the daily mean energy intake per mouse. f, Cumulative energy intake. g, Energy balance (metabolizable energy intake minus energy expenditure); average values for dark, light and combined photoperiods are shown. For statistical analysis, ANCOVA was used. h, Cumulative energy balance versus time. i,j, Statistical analysis by ANCOVA for energy expenditure and energy intake, with body mass as a covariate. In both cases, the groups are statistically different. A mass × group interaction effect was not significant. **P< 0.01; ***P < 0.001.

Fig. 3:. Quality control and visualization guide.

a,b, Quality control in indirect calorimetry experiments of littermate male C57BL/J mice on a HFD (green) or LFD (yellow) (grey bars indicate dark photoperiods). a, The mass change (Δ mass) is determined from measurements taken before and after the indirect calorimetry experiment. Calculated daily energy balance (kcal day⁻¹) versus Δ mass for each animal over 48 h. The three cages with malfunctioning feeders are excluded from the analysis (open circles). Linear regression is predicted to cross the origin (dashed line) with slope of 4 kcal g⁻¹, corresponding to weight change from lean mass (pink), or slope of 9 kcal g⁻¹, corresponding to weight change from fat mass (purple). b, Cumulative food intake (kcal) for each mouse, overlayed with group means. The values for the three excluded mice are shown in black. c–l, Visualization guide. Plots of the RER, the ratio of VCO₂ to VO₂ and an indicator of substrate oxidation (1.0 corresponds to more carbohydrate oxidation, 0.7 corresponds to more fatty acid oxidation). Mice on a HFD have more fatty acid oxidation and a lower RER than do mice on a LFD. c,d, The RER for one mouse on a LFD (c) or one mouse on a HFD (d). Group mean values of RER versus time for 48 h (e, g, i, k) or 12 h (f, h, j, l) (corresponding to the dashed box). Plotting choices include the visualization of ribbons or error bars to represent the s.e.m. and raw group mean values or a 6-point rolling average (30 min smoothing). e,f, Lines represent group means ± s.e.m. with ribbons and smoothing. This is our recommended visualization (indicated with the tick icons). g,h, Lines represent group means ± s.e.m. with error bars and no smoothing. i,j, Lines represent group means ± s.e.m. with error bars and 30-min smoothing. k,l, Points and lines represent group means ± s.e.m. with ribbons and smoothing.

Fig. 4:. Correct and incorrect visualization of metabolic rates.

In a second cohort of littermate male C57BL/6J mice on a LFD (yellow) or HFD (green), plots of the average VO₂ versus time; boxplots with the median, 25th percentile and 75th percentile over 48 h; and regression plots of the VO₂ versus total body mass. a–c, The standard visualization of metabolic rate (VO₂ in ml h⁻¹), with no normalization applied (tick icons indicate that this is our recommended visualization). d–f, An incorrect and misleading visualization achieved by dividing VO₂ by total body weight (VO₂ in ml kg⁻¹ h⁻¹). With this erroneous manipulation (indicated with cross icons), mice with the largest mass now appear to have the lowest metabolic rate. Dividing by total body weight assumes that the groups of mice have homogeneous body compositions; this assumption is violated in this example, because mice on a HFD have greater fat mass than do mice on a LFD.

Fig. 5:. Statistical approach for indirect calorimetry and choosing the appropriate covariate.

a,b, Statistical tests. For analysis of mass-dependent variables (for example, energy expenditure or VO₂), we first perform a statistical test to determine whether the slopes are parallel, by testing for an interaction effect between the covariate (total mass) and the group variable. In example 1 (a), the non-parallel lines produce a statistically significant interaction effect and dictate that ANCOVA cannot be used in this analysis. The diverging lines make the interpretation of the experiment more complex; the statistical differences between the groups will depend on the specific mass range at which the comparison is made. For statistical analysis, we will use the general linear model (GLM) when an interaction effect is significant. In example 2 (b), we first run a test for the interaction effect and find that it does not reach statistical significance. This result allows us to proceed with ANCOVA. c–f, Comparison of energy expenditure (kcal h⁻¹) as the dependent variable versus different mass covariates as the independent variable. c, Energy expenditure versus total mass (g). d, Energy expenditure versus lean mass. e, Energy expenditure versus fat mass. f, There is a positive correlation (dashed line) between the two independent variables, fat mass versus lean mass. This correlation complicates the use of lean and fat mass as multiple independent variables for regression.