A Recurring Problem With the Analysis of Energy Expenditure in Genetic Models Expressing Lean and Obese Phenotypes

The fact remains that an urgent need exists for the development of new therapies to control body weight and improve insulin action, and modulation of energy expenditure provides one potential therapeutic target. The availability of spontaneous and targeted mutations in mice is an important resource for identifying novel genes involved in energy homeostasis. Hardly a month goes by without publication of data demonstrating novel mechanisms for regulating adipose mass or further characterizing pathways of previously identified genes involved in regulating satiety or metabolism (Table 1). Often these mutant genes are reported to affect obesity/leanness and diabetes phenotypes as a consequence of their effects on substrate oxidation and energy expenditure. However, few of these have ultimately led to the identification of the novel underlying thermogenic mechanism.

The problem facing the field is that, with the exception of UCP1/BAT thermogenesis, there is little evidence for an alternative thermogenic mechanism. A case has been made for UCP1/BAT as a thermogenic mechanism for diet-induced thermogenesis at thermoneutrality (29,30). However, mice in research colonies and in the wild and humans before World War II did not spend most of their time at thermoneutrality, and the evidence is clear that at even slightly reduced ambient temperatures (24°C) UCP1/BAT is not required for diet-induced thermogenesis (31,,–34). Even severely obese Lep^ob/Lep^ob mice, which exhibit clear deficits in BAT thermogenesis in response to cold exposure, exhibit a robust increase in oxygen consumption in response to hyperphagia (15,35). While it is clear that mutations and environmental conditions that affect UCP1/BAT will impact diet-induced obesity, these conditions can affect fuel supply for BAT thermogenesis, for example, acyl-CoA dehydrogenases knockouts (36), or can inactivate transcription factors (37,38) or receptors (39,40) associated with the cold-mediated induction of thermogenesis BAT. Although these mutations primarily affect the efficiency of a thermogenic mechanism devoted to body temperature regulation, the efficiency of UCP1/BAT thermogenesis indirectly affects body weight regulation as less efficient systems for maintaining core body temperature require more calories.

That thermogenic systems for dissipation of calories have not been found does not mean that ion/substrate cycles could not serve this function (41,,,–45). Presumably some targeted mutations with adiposity phenotypes are based upon UCP1-independent alternative thermogenic mechanisms. However, inspection of the phenotypic data for energy expenditure in the studies cited in Table 1 readily leads one to the disappointing conclusion that it is probable that none of them have a bona fides defect in energy expenditure.

What is the basis for this provocative statement? It will be obvious to anyone that has used indirect calorimetry for comparing lean and obese mice that total 24-h energy expenditure (TEE) increases as a function of body weight. Studies comparing 24-h energy expenditure in humans with a wide range of body weight also demonstrate an increase in 24-h energy expenditure (46). To compensate for the often dramatic differences of body mass in lean and obese models, and perhaps to estimate energy expenditure at a cellular level, investigators have resorted to normalizing energy expenditure reported as oxygen consumption (Vo₂) to body weight. This approach assumes that lean and adipose tissues contribute equally to Vo₂; however, while adipose tissue is not metabolically inert, it contributes relatively little to the total energy expenditure of an organism compared with lean mass. Furthermore, this distortion in estimates of energy expenditure will increase in proportion to the accumulation of lipid in the adipocyte as obesity increases.

The theoretical and experimental bases for the effect of mass on energy expenditure are complex and have been covered in depth by many physiologists for over a century. A particularly lucid account can be found in “Energy Metabolism in Animals and Man” by Kenneth Blaxter (NY, Cambridge University Press, 1989). During the past 7 years, in excess of 10 papers published in high profile journals devoted to the phenotypic characterization of new targeted gene mutations affecting body composition have calculated whole-body energy expenditure by dividing the volume of oxygen consumed by the total body weight (Table 1). This use of body weight to normalize energy expenditure data between groups of mice with often marked differences in body weight makes the assumption that all tissues have an equivalent metabolic demand. A variation in estimating energy expenditure to account for differences in the mass of mutant versus wild-type mice divides oxygen consumption by body weight to the 3/4 power. It has long fascinated biologists that body mass and energy expenditure among a range of species vary as a function of the 3/4 power (47,48). However, it is assumed that species that conform to this relationship will have normal body composition, that is, neither obesity nor insulin resistance. Therefore, including a highly variable fat mass in the body weight when the lean mass is minimally variable will similarly distort estimates of energy expenditure.

As an example, we provide data from our studies analyzing energy expenditure in genetic mutants that exhibit obesity owing to increased fat mass (Fig. 1,A). Obesity associated with mutations in either the leptin or melanocortin-4 receptor (Mc4r) genes is associated with a 30–40% increase in TEE expressed as kilojoules per mouse per day (Fig. 1,B) (49,50). Normalization of TEE by body weight (which includes substantial increases in the amount of adipose in the obese models, Fig. 1 A) appears to indicate lower TEE per unit of body weight with extreme obesity, at least in females. Two-way ANOVA indicates significant effects of genotype (P < 0.0001) and sex (P < 0.0001), with the effect of genotype dependent on sex (P < 0.05). TEE per gram of body weight was significantly lower in female Lep^ob/Lep^ob mice compared with wild type (P < 0.0001).

When the same data are normalized using lean mass, a very different outcome is observed. TEE per gram of lean mass is significantly (and markedly) increased with obesity (effect of genotype, P < 0.0001, P < 0.002 between all genotypes). Sex also significantly affected TEE per gram of lean mass, with males having a lower value irrespective of genotype (P < 0.0001); however, sex had no affect on the impact of genotype. Note that the quite large differences of fat mass observed between Mc4r^−/− and Lep^ob/Lep^ob mice (∼10 g, or 15–20% of total body weight) do not correlate with significant changes in TEE (Fig. 1 B). This suggests that the additional fat mass is not associated with any appreciable increase of metabolic activity.

One interpretation of this data is that severe obesity is supported by increased energy expenditure. As both genetic models of obesity may be hyperphagic relative to lean controls, the oxidation of any additional calories could contribute to increased energy expenditure. Note that while these genetic mutants obviously accumulate more mass over several weeks to a few months, these recordings were taken over a time frame of a few days where minimal or no significant change in body mass would occur. In other words, the state of energy balance (intake less expenditure) is likely near neutrality, and any additional calories ingested are likely being oxidized or used to support the maintenance of the additional mass (i.e., oxygenation and turnover of protein and lipid in the extra tissue). Similar results were observed whether the analysis used total energy expenditure or resting energy expenditure, assessed using data obtained during period of minimal activity.

The rationale for dividing Vo₂ by body weight to calculate energy expenditure is also sometimes applied to food intake. That is, food intake is expressed as gram or calorie per gram of body weight. Of course, when applied universally the outcome makes no sense, since it would result in the conclusion that leptin-deficient Lep^ob/Lep^ob mice are hypophagic. In some instances authors will present one parameter adjusted for body weight and one expressed per animal (or per lean mass). Quite often, the presentation of data appears to reflect a desire to align the data with assumptions made a priori, as opposed to an unbiased analysis of the contribution of food intake or energy metabolism to the phenotype being investigated.

To illustrate the problem we draw attention to two recent examples, one a report by Plum et al. in Cell Metabolism (51), and the other by Hofmann et al. in the Journal of Clinical Investigation (52). Plum et al. performed a technological tour-de-force by deleting the phosphatidylinositol-3-phosphate (PIP3) kinase Pten, specifically neurons expressing the long-form of the leptin-receptor (ObRb). The deletion of Pten in ObRb neurons (Pten^ΔObRB) resulted in the over activation of the phosphatidylinositol-3 (PI3) kinase pathway, previously reported to be involved in the anorectic actions of leptin (53). As predicted, these mice exhibit a lean phenotype, with a marked reduction in perigonadal fat pad weight and a 10–15% (2- to 3-g) reduction in body weight. The lean phenotype was largely attributed to energy expenditure, based upon increased Vo₂ expressed as milliliters per kilogram per hour; however, this increase in energy expenditure would disappear if expressed as Vo₂ in milliliters per mouse per hour or milliliters per gram of lean mass per hour. Additionally, the authors confound the problem by presenting food intake accumulated over 3 weeks by adjusting to body weight (grams per grams per day). Based on the body weight data, it appears likely that there would be mild reduction in total food intake in Pten^ΔObRB mice compared with controls, which could contribute to the lean phenotype. More surprisingly, in the same figure, food intake during a 6-h period after an overnight fast was expressed as gram per animal and was not different. No rationale is provided for the ad hoc adjustment of food intake for body weight. Their conclusion is that Pten ablation in leptin-sensitive neurons reduces adiposity by increasing energy expenditure.

The Hofmann et al. article is another molecular genetic tour-de-force in which adipocyte LDL receptor–related protein-1 was specifically inactivated in white adipose tissue to assess its role in diet-induced obesity. However, the authors use the same problematic experimental approach to calculate energy expenditure from indirect calorimetry, food intake, and, in addition, loss of fat mass during fasting, and they come to a similar conclusion, namely that reduced adiposity is caused by elevated energy expenditure from enhanced muscle thermogenesis.

In both studies if the energy expenditure was calculated by normalizing to lean body mass or total energy expenditure per mouse, then no significant differences would have been observed. It is important to emphasize that these and other studies in Table 1 represent only a few of many articles published in a broad spectrum of journals that publish such research.

This issue with the analysis of indirect calorimetry data was raised in a letter in Science in 1997 by Himms-Hagen (5). This letter was a response to the publication of the first studies on the effectiveness of recombinant leptin to correct the obese phenotype of Lep^ob/Lep^ob mice (4). The authors of that study claimed that leptin increased energy expenditure in these mice when calculating energy expenditure by dividing oxygen consumption by body weight. Himms-Hagen pointed out that if energy expenditure is estimated from oxygen consumption per mouse, it leads to the conclusion that leptin stimulates fat oxidation and reduces food intake in the Lep^ob/Lep^ob mouse, but it does not stimulate energy expenditure. The serious negative consequence of the conclusion that Lep^ob/Lep^ob mice have a slow metabolism led to years of fruitless and expensive clinical studies on obese humans seeking to find obese individuals with a slow metabolism. None were found, and the blame for these fruitless studies was placed on the Lep^ob/Lep^ob mouse, claiming that its obesity is not similar to human obesity (54). However, a review of energy expenditure data in the Lep^ob/Lep^ob, Mc4r^−/−, and wild-type mice from our laboratories illustrates how Lep^ob/Lep^ob have a slow metabolism when oxygen consumption data are expressed per gram of body weight (Fig. 1) but not when oxygen consumption data are expressed per gram of lean mass or per mouse. Thus, the problem is not with the mouse model but with the inability to appreciate that a unit of fat mass has a lower level of oxygen consumption than a unit of lean mass.

The letter by Himms-Hagen should have significantly reduced the publication of studies that employed similar methods; however, the extensive list of high-profile publications since 1997 in which changes in adiposity have been erroneously attributed to modulations in energy expenditure indicates that the problem continues (Table 1). Consequently, these conclusions as to how various metabolic pathways contribute to obesity phenotypes through modulation of energy expenditure impair progress in this area. The effect of genetic manipulation on the basic adiposity phenotype is not in dispute, rather the manner in which indirect calorimetry has been used to provide a physiological mechanism based upon energy expenditure is the problem.

This point is further illustrated using a simple thought experiment showing that subtle differences in energy expenditure can account for marked differences in the accumulation of adipose over time. A difference of 0.5 g/week in the rate of accumulation of fat between a mutant and control mouse would, if constant, result in a 10-g difference in body weight over 5 months. These estimates are not unreasonable, as studies examining obesity phenotype usually involve mice in the 4- to 6-month age range (if not older); a difference of 10 g represents a marked change for the B6 mouse, which weighs 25–30 g when maintained on standard chow, increasing to 45–50 g when fed a high-fat diet. However, this number represents a difference of 18.8 kJ per week, or <2.9 kJ/day. Assuming a range of energy expenditure of 50–60 kJ/day and identical caloric intake, then this would mean a difference in daily energy expenditure of ∼5%, which is at the limits of the discriminatory capacity of indirect calorimetry.

Of course, this crude estimation does not take into account the actual efficiency of the conversion of nutrients to triglyceride stores in adipose. It also does not take into account differences in the balancing of nutrient composition with substrate oxidation, indicated by the respiratory exchange ratio, which may also impact energy homeostasis. However, it does highlight the small changes in energy balance that can explain relatively large differences in the accumulation of adipose mass over time. Similar reasoning has been applied to the normally exquisite regulation of energy balance affecting adiposity over time in humans (55).

The number of studies where energy expenditure has been expressed as Vo₂ divided by total body weight appears not to have been diminished by Himms-Hagen's admonitions. In most cases articles identifying a specific thermogenic mechanism associated with the phenotypes have not followed these initial reports. In other cases, modest changes in Ucp1 mRNA expression especially in white adipose tissues have been interpreted to implicate alterations in UCP1 as the thermogenic mechanism underlying the differences in energy expenditure (51,56). However, whether a 50% change in Ucp1 expression affects overall adiposity is questionable given that complete deficiency of UCP1 does not affect the obese state or energy expenditure at an ambient temperature of 23°C (34). Cold tolerance has also been used by many studies as an ad hoc indicator of reduced metabolism; however, even this interpretation is complicated, since homozygous Lep and Ucp1 mutant mice can, given the right conditions or, in the case of UCP1-deficient mice on an F1-hybrid background, adapt to cold (57).

We have approached this discussion of the problems arising from calculations of energy expenditure from Vo₂ data based upon an oversimplified obesity model in which the only morphological variable is body fat content. Mutations in pleiotropic signaling molecules may not only affect body fat content, but also change the overall size and body composition and affect behavioral phenotypes changing physical activity and food intake. Two recent studies on the targeted inactivations of the FTO and IKK[epsilon] genes have the potential of having a high impact on studies of human obesity, especially FTO, given the association of FTO alleles with human obesity. These studies also propose that differences in diet-induced obesity are due to differences in energy expenditure (58,59); however, this interpretation is uncertain because of significant differences in both lean and fat mass in the mutant versus wild-type mice. Each of these subphenotypes will have an effect on calculations of energy expenditure. Therefore, data provided by systems for assessing energy expenditure require evaluation with respect to each of the morphological and the behavioral variables. If a change in adiposity is determined by mutations to enzymes of fat metabolism or to preferential fat oxidation, then the respiratory quotient can provide additional useful information. In addition, an independent evaluation of energy expenditure can readily be acquired by determination of metabolic efficiency, that is, by measuring changes in adiposity and lean mass relative to caloric intake over the course of several weeks or months.

Finally, regression analysis of energy expenditure against body composition data, which is the method of choice in determining energy expenditure in humans (46), could be used. However, this approach is impractical for mouse studies characterizing the effects of targeted mutations on obesity and energy expenditure. In most laboratories, relatively small numbers (6,,,–10) of age- and sex-matched mutant and wild-type mice are available for indirect calorimetry studies. However, if variation in lean mass or body size was present as revealed by duel-energy X-ray absorptiometry or nuclear magnetic resonance, then analysis by regression analysis would be justified. Alternatively, National Institutes of Health (NIH)-funded investigators in institutions without the technology for indirect calorimetry or analysis or body composition by nuclear magnetic resonance can send animals to the NIH-funded Mouse Metabolic Phenotyping Centers (http://www.mmpc.org/).

We have attempted to refocus attention on a significant but easily corrected problem encountered often in the literature in studies determining energy expenditure in animals with variations in adiposity, and first clearly enunciated by Jean Himms-Hagen 12 years ago (5). Establishing physiologically meaningful phenotypes of energy balance is a challenging endeavor influenced by variations in physical activity, circadian rhythms, subtle differences of substrate utilization, and diet composition. Careful phenotyping can provide unique insights into energy balance at the whole animal level that cannot be obtained through molecular or cell culture experiments. However, major problems lie ahead as our ability to generate new knockout models has outstripped our capacity for a thorough investigation of phenotype. The technical feasibility of rapidly generating knockout gene models has led to an international consortium to knockout every gene in the mouse (http://www.genome.gov/17515708). However, we are not the first to voice concern about whether the experience, resources, and financing are available first to design physiologically meaningful gene knockouts of energy balance and second to create and support the laboratories with the experience to phenotype the mice for defects in energy homeostasis (60).

A.A.B. is supported by DK073189 and the American Diabetes Association (7-08-RA16). L.P.K. is supported by HD008431. A.A.B. and L.P.K. have received support from CNRU Center Grant 1P30 DK072476 entitled “Nutritional Programming: Environmental and Molecular Interactions” sponsored by the National Institute of Diabetes and Digestive and Kidney Diseases. A.A.B. has received funding support from Biomeasure, Inc., IPSEN. No other potential conflicts of interest relevant to this article were reported.

The authors thank Dr. Karima Begriche for help with preparing the manuscript.