comment

Progress and challenges in analyzing rodent

energy expenditure
Whole-body energy expenditure is the summed metabolic activities of tissues and, to remove the influence
of body size, ratios of energy expenditure to body mass are often applied but can generate spurious differences.
In 2011, a group of experts proposed adoption of ANCOVA for the analysis of metabolic rate but, seven years
later, analyses based on ratios remain the most frequent. We discuss some of the barriers to adopting better
analytical procedures.

Rodrigo Fernández-Verdejo, Eric Ravussin, John R. Speakman and Jose E. Galgani

I
n humans, average body mass index has a The extent to which these molecules are
increased over the last few decades1. This 2.0 1.5 kcal d–1 s.d. retained in a living organism is a function
between groups (kcal d–1)

is mostly explained by a shift towards the 1.0 kcal d–1 s.d. of the energy flux12. Such flux is dynamic,
Detectable difference

right in the distribution of individuals with
larger body mass, while body mass of leaner
individuals has remained unaltered2. Such
heterogeneous propensity to weight-gain
prompts the search for factors that alter the
energy balance.
Technologies to measure energy
expenditure in humans and rodents have
made remarkable progress3,4. Additionally,
genetic manipulation of rodent models
allows discernment of the influence
of genetics on metabolic rate with an
unprecedented sophistication. However,
this has not been mirrored by improved
methodologies to analyze metabolic
rates. Metabolic rate is often divided by
body mass to “normalize” data for size
differences between animals, which is
mathematically wrong.
Knowledge of this problem, and the
approach to overcome it, dates back to
the 1950s5,6. However, compared with
calculating ratios, manual calculation of the
analysis of covariance (ANCOVA) was time
consuming. In the 1980s, computational
analysis eased calculations and increased
interest in the appropriate analyses in
the field of human metabolism7 and in
ecological studies of small mammals8.
ANCOVA thus became the common
approach to adjust for body size. In contrast,
studies of small laboratory rodents persisted
in using ratios. This erroneous practice has
continued despite papers pointing out the
potential for spurious results9,10.
To reach a consensus on how mouse
energy metabolism should be analyzed, a
large group of experts in mouse and human
metabolism discussed conceptual, analytical
and practical aspects involved in measuring
energy metabolism in mice11. They proposed
ANCOVA as a way to conduct, analyze and
interpret metabolic studies. Here, we assess
Nature Methods | www.nature.com/naturemethods
1.5
1.0
0.5
0 of chronic positive energy balance. Accurate,
0 6 16 32
Mice per group (n)
b for understanding body weight control.
ANCOVA, ratios and no normalization (n = 3)
ANCOVA and no normalization (n = 3)
ANCOVA and ratios (n = 2)
ANCOVA only (n = 6)
No normalization (n = 10)
Ratios and no
normalization (n = 8)
Ratios only (n = 45)
Total = 77 articles
Fig. 1 | Analytical aspects in rodent metabolic
studies. a, Detectable difference in energy
expenditure according to the number of mice
per group. The mean energy expenditure of
the control group was assumed as 10 kcal d–1,
with standard deviations (s.d) for control and
experimental groups of 0.5, 1.0 and 1.5 kcal d–1.
SAS 9.2 (SAS Institute) was used for calculations,
using two-sided Student’s t test with α = 0.05
and β = 0.20. b, Frequency of the method used to
compare energy expenditure in mice among the
studies retrieved.
how successful the application of ANCOVA
has been in mouse metabolism studies.
Energy balance and weight gain
Body weight represents the combined mass
of water, protein, fat and carbohydrate.
0.5 kcal d–1 s.d. with a constant exchange between the
macromolecules entering the body through
feeding, and the molecules leaving the body
through oxidation. A mismatch of this flux
(in versus out) modifies body weight and
composition, leading to obesity in the case
48 64
precise and continuous determination of
energy expenditure and intake is thus crucial
In the best scenario, we should isolate the
primary event driving energy imbalance,
that is, higher energy intake or lower energy
expenditure, before changes in body mass
occur. Living organisms are, however, almost
never in neutral energy balance, because
organisms acquire and dissipate energy at
variable rates. Therefore, distinguishing
normal hour-to-hour energy disparity from
the imbalance that will lead to obesity is
challenging. Thus, energy balance is often
compared in animals of different body
mass and/or composition. Nevertheless,
as body mass and composition influence
energy metabolism, these variables must
be controlled to determine the influence of
other variables (for example, genotype).
Comparison of energy expenditure
in mice
Several studies have assessed the effect of
factors, such as diets and genes, on energy
expenditure of animals that differ in body
mass and/or composition. Such differences
may alter metabolic rate. Therefore, a highly
debated issue is how to normalize energy
expenditure for differences in body size.
An extensively used normalization method
is dividing metabolic rate by body mass.
However, this approach is inappropriate
because the regression line between body
mass and metabolic rate does not have a
zero intercept, and is therefore not linearly
comment
proportional13,14. Scaling body mass to a
power was intended to solve this problem.
However, the power exponent commonly
utilized (that is, 0.75)15 derives from
between-species comparisons that may not
apply for within-species comparisons. To
account for differences in body composition,
the metabolic-rate-to-lean-mass ratio is
calculated. This approach does not solve
the problem, because the regression line
between lean mass and metabolic rate also
has no zero intercept.
Generalized linear modeling (GLM)
and ANCOVA are appropriate approaches
for normalizing metabolic rate to body
mass or composition. These approaches
allow combination of discrete (for example,
genotype) and continuous (for example,
body mass) traits as determinants of energy
expenditure. Lean and fat masses can also be
included to account for body composition.
Considering tissues and organs expend
energy at different rates, ANCOVA could
also include the masses of individual
organs and tissues as covariates16,17. Thus,
a predictive model of metabolic rate can
be generated from control mice based on
the size of, among others, the liver, brain or
skeletal muscle.
Alternatively, metabolic rate can be
compared before differences in body mass
and composition are evident. In this case,
metabolic rate can be simply reported
as kcal d–1. However, the ability to detect
‘statistically significant’ differences depends
on the reliability of the instrument, the
statistical approach and the sample size.
With the sample size often included in
studies (∼6 per group), differences in
body mass or composition may remain
undetectable, which may influence the
results by including or not including a
normalization of the data. True differences
in energy expenditure may also be
undetectable, and the detectable differences
in energy expenditure may be a function
of the sample size (Fig. 1a). With six mice
per group and a standard deviation of
0.5 kcal d–1, differences in energy
expenditure of less than 10% cannot be
detected. On the one hand, such lack of
power may be considered an advantage as
it reduces the risk of making a type I error
and only genes with large effect size will be
identified. On the other hand, genes with
small, but perhaps significant, long-term
influence may be wrongly discarded
(an increased risk of a type II error).
Notably, ANCOVA may be unnecessary
and could generate spurious results when
mice display no overlap of body weight
between groups18,19. See ref. 11 for further
details on the appropriate analysis of
metabolic rates.
The concepts aforementioned also apply
for energy intake. ANCOVA should thus
be used for comparison of energy intake
when body mass or composition differs
between groups. Importantly, since energy
expenditure and intake are required to
understand weight gain, reporting them in
the same units is essential; using different
units blurs the significance of results.
How energy expenditure is analyzed
In 2011, Tschöp et al.11 found that only
2% of studies used ANCOVA or GLM to
correct for body mass or composition, and
recommended more widespread use. Have
these recommendations been followed since
then? We set an arbitrary search criterion
to include a wide range of publications.
We searched in PubMed for original
articles published or available online from
July 2012 until June 2018, using the key:
(“energy expenditure” OR “energy balance”
OR “heat production” OR “metabolic
rate” OR “oxygen consumption”) AND
(“2012” [Date - Publication]: “3000” [Date
- Publication]) AND (“mouse” OR “mice”).
We also searched papers that cited the guide
by Tschöp et al11. The inclusion criteria
were: comparison of genetically altered
mice versus control mice (mouse models
injected with anti-sense oligonucleotides or
adenoviruses were also included); mice fed
ad libitum or pair-fed designs; mice awake
and without movement restrictions at the
time of measurement; analysis at room
temperature or at thermoneutrality; mice
fed with chow, high-fat or western diet; and
journal impact factor ≥ 15.0 according to
the Journal Citation Reports 2016.
We identified 77 articles, of which
14 (18%) cited Tschöp et al. Eighteen percent
of articles used ANCOVA (n = 14), of which
almost two-thirds included body mass as the
covariate (Fig. 1b and Supplementary Table).
Only five articles conducted ANCOVA
with lean mass as covariate (Supplementary
Table). As reported11, normalization using
ratios with body mass or lean mass were the
most frequent methods (> 50%) (Fig. 1b).
Some studies analyzed data using more than
one approach. For instance, three studies
analyzed data without normalization,
with ANCOVA and with ratios (Fig. 1b).
This multiple-analysis strategy appears
unnecessary and may lead to confusion if
the approaches generate discordant results.
Presenting individual data for the regression
between energy expenditure and body
mass (or lean mass), and comparing groups
with ANCOVA, appears to be the most
appropriate analysis.
Thousands of dollars are spent on genetic
mouse models, and energy metabolism
is measured with expensive equipment.
Unfortunately, data is often analyzed in a
way that negates all the previous investment
in time, effort and money. Although the
search conducted by ourselves and by
Tschöp et al.11 use different search criteria
and are not directly comparable, our data
suggest that while ANCOVA is used more
often than before, its implementation still
needs to be increased.
Why does this issue still exist?
We notice potential barriers for adopting
appropriate methods. First, reviewers
are unfamiliar with the best practice.
Second, since most studies use the wrong
method, new studies simply repeat this
wrong method. This is particularly true
if researchers are unfamiliar with the
correct analysis, which may be perceived
as complex and hard to interpret. But with
the appropriate statistical packages (such as
STATA; https://www.stata.com), conducting
ANCOVA is as easy as calculating ratios.
Mina et al.20 recently proposed a free web-
based tool (CalR; https://CalRapp.org/) that
provides appropriate analyses of indirect
calorimetry data in various experimental
designs. Third, using ANCOVA often
reveals no effect of a given manipulation
(for example, genotype) making publication
more difficult, particularly if it does
not conform to a previous consensus.
Researchers could thus feel pressured to
continue with a particular method, even
if it is wrong. Fourth, some commercial
equipment automatically expresses
metabolic rate per kg of body mass,
suggesting this is an appropriate analysis.
Finally, some authors cite the guide for
analysis by Tschöp et al.11, but still use ratios.
This incongruence may lead some to assume
that using ratios was recommended by
Tschöp et al.
Concluding remarks
Mouse models provide an invaluable
opportunity to compare metabolic
phenotypes to better understand weight-
related diseases. These models are
accompanied by major progress in the
measurement of energy metabolism using
indirect calorimeter cages. Unfortunately,
such advances have not been associated with
appropriate data analysis, potentially leading
to biased interpretations.
Editorial boards and peer-reviewers
have a major responsibility in demanding
appropriate analyses. Including specific
guidelines for metabolic data in the
‘Information for authors’ may promote
better analyses. A straightforward strategy
is to require energy expenditure plotted
against body mass (or lean mass). This will
force consideration of mass effects, which
Nature Methods | www.nature.com/naturemethods

are obscured in histograms. A check-list
verifying that energy expenditure was
analyzed with ANCOVA and plotted
correctly should be mandatory. Analysis
of energy intake, along with the detailed
reporting of the composition of diets,
also deserves consideration. These efforts
should increase the consistency, rigor and
reproducibility required for comprehensive
understanding of the role that biological and
environmental traits have on obesity. ❐ 4. Lighton, J. R. B. Eur. J. Clin. Nutr. 71, 301–305 (2017).
Rodrigo Fernández-Verdejo1, Eric Ravussin2,
John R. Speakman3,4 and Jose E. Galgani1,5*
1
Departamento de Ciencias de la Salud, Carrera
de Nutrición y Dietética, Facultad de Medicina,
Pontificia Universidad Católica de Chile, Santiago,
Chile. 2Pennington Biomedical Research Center,
Baton Rouge, LA, USA. 3Institute of Biological
and Environmental Sciences, Zoology Building,
University of Aberdeen, Aberdeen, UK. 4Institute
of Genetics and Developmental Biology, Chinese
Academy of Sciences, Beijing, China. 5Departamento
de Nutrición, Diabetes y Metabolismo, Facultad de
Nature Methods | www.nature.com/naturemethods
Medicina, Pontificia Universidad Católica de Chile,
Santiago, Chile.
*e-mail: jgalgani@uc.cl
Published: xx xx xxxx
https://doi.org/10.1038/s41592-019-0513-9
References
1. NCD-RisC. Lancet 387, 1377–1396 (2016).
2. Wardle, J. & Boniface, D. Int. J. Obes. 32, 527–32 (2008).
3. Lam, Y. Y. & Ravussin, E. Mol. Metab. 5, 1057–1071 (2016).
5. Tanner, J. M. J. Appl. Physiol. 2, 1–15 (1949).
First paper calling for attention regarding the fallacy of
expressing physiological variables per unit of body size.
6. Cochran, W. G. Biometrics 13, 261–281 (1957).
7. Ravussin, E., Lillioja, S., Anderson, T. E., Christin, L. &
Bogardus, C. J. Clin. Invest. 78, 1568–1578 (1986).
8. Packard, G. C. & Boardman, T. J. Physiol. Zool. 61, 1–9 (1988).
9. Arch, J. R. S., Hislop, D., Wang, S. J. Y. & Speakman, J. R.
Int. J. Obes. 30, 1322–1331 (2006).
10. Butler, A. A. & Kozak, L. P. Diabetes 59, 323–9 (2010).
11. Tschöp, M. H. et al. Nat. Methods 9, 57–63 (2011).
12. Hall, K. D. et al. Lancet 378, 826–37 (2011).
13. Heymsfield, S. B. et al. Am. J. Physiol. Endocrinol. Metab. 282,
132–138 (2002).
14. Ravussin, E. & Bogardus, C. Am. J. Clin. Nutr. 49, 968–975 (1989).
Remarkable analysis of the main determinants of energy
expenditure in humans. This paper also elaborates the need
comment
for proper analysis when comparing individuals of different
body size.
15. Kleiber, M. The Fire of Life: An introduction to animal energetics
(Wiley, 1961).
16. Mitchell, S. E. et al. Oncotarget 8, 17453–17474 (2017).
17. Javed, F. et al. Am. J. Clin. Nutr. 91, 907–12 (2010).
18. Speakman, J. R. Front. Physiol. 4, 34 (2013).
19. Tracy, C. & Sugar, J. Physiol. Zool. 62, 993–997 (1989).
20. Mina, A. I. et al. Cell Metab. 28, 656–666 (2018).
Acknowledgements
R.F.V. was supported by Fondecyt #11180361 (Conicyt,
Chile). J.E.G was supported by Fondecyt #1170117
(Conicyt, Chile).
Author contributions
J.E.G., E.R. and J.R.S. conceived the work. R.F.V. and J.E.G.
designed, acquired and analyzed the data, and drafted the
manuscript. All authors interpreted the data, substantively
reviewed the manuscript, approved the submitted
version and agreed to be personally accountable for their
contributions.
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information is available for this paper at
https://doi.org/10.1038/s41592-019-0513-9.