During the last 30 years, specific study designs (e.g. twin, family, and adoption studies) have been used to calculate the total genetic influence on body weight (for a recent review, see 24). More recently, genome-wide association studies (GWAS) on BMI, waist circumference (WC), and FM have been undertaken with the goal to identify human genes that biologically cause overweight. While familial correlation in BMI is high in monozygotic twins, so far GWAS on obesity at the population level explained a minor proportion of the variation in adult BMI only.

Up to now, a total number of 19 rare monogenetic defects associated with severe obesity are impressive manifestations of disturbed control of body weight ²⁵ . These rare cases are in favor of the idea that specific genes influence EI and/or EE and thus EB and body weight. However, these findings do not explain population-wide obesity. Although specific variants in individual genes (e.g. the FM- and obesity-associated gene, FTO) are considered suitable candidates to explain the individual variability in (i) the predisposition to become obese or (ii) individual responses to weight loss strategies in obese patients, the proposed genetic basis of obesity is still uncertain ²⁴ .

Using specific study designs (e.g. twin, family, and adoption studies), heritability ¹ estimates (which are considered synonymous with genes) of BMI, FM, and visceral adipose tissue (VAT) have been calculated in numerous studies, which gave evidence for a biological influence on body weight ^{26– 41} . The familial correlations in BMI were between 0.20 and 0.23 in parent–offspring pairs and 0.20 and 0.34 in dizygotic twins and reached 0.58 to 0.88 in monozygotic twins. In general, additive genetic factors explaining the proportion of variation in BMI varied between 0.31 and 0.85. Molecular mechanisms of heritability may not be limited to DNA sequence differences, since epigenetic factors also contribute to the phenotype. In fact, analyzing DNA-methylation profiles in pairs of monozygotic and dizygotic twins may be due to epigenomic differences in the zygotes adding to heritability estimates ⁴² .

When different age groups of twin pairs were compared, the proportion of BMI explained by genetic and epigenetic factors increased until late adolescence with no or only minor effects of the shared environment ⁴⁰ . By contrast, shared environmental factors related to education and/or culture seemed to have a stronger influence during mid-puberty. Furthermore, in pooled cohorts of a total of 140,379 complete twin pairs from different regions of the world, the heritability estimates of BMI decreased from 0.77 in young adults to 0.59 in adults aged 70 to 80 years, which was independent of the obesity prevalence in the populations studied ⁴¹ . However, heritability estimates cannot explain steep increases in the prevalence of obesity, which are due to non-biological and thus environmental changes as the driving factors. Furthermore, heritability does not take into account the complexity of the genotype–phenotype relationship. In addition, additive and non-additive genetic effects cannot be addressed separately. Finally, incomplete adjustments for co-variates like growth spurts during puberty or regional diversities in the environments affect heritability estimates.

In addition to cross-sectional data from population studies, differences in the response to overfeeding studied in pairs of monozygotic twins showed that inter-pair variances in gains of weight, FM, and VAT were found to be three to six times higher than the respective intra-pair variance ^{26, 27} . This was taken as evidence for a “genotype-overfeeding interaction” that determines weight and fat gain as well as fat distribution. Vice versa with negative EB (due to a controlled exercise program ³¹ ), the intra-pair variances in changes in weight, FM, and VAT were also lower than the inter-pair variances, suggesting a “genotype-underfeeding interaction” as well. However, these data have to be compared with considerable intra-individual variances in changes of body weight, which have not been taken into account in the studies cited ²⁴ .

The high heritability estimates were not supported by the results of recent GWAS on BMI, FM, or VAT in greater populations ²⁴ . Up to now, 115 genetic loci have been identified where sequence variation was statistically associated with the BMI, explaining 2 to 3% of the variation in adult BMI only ⁴³ . In addition, longitudinally, no significant associations were found between any lead single nucleotide polymorphisms (SNPs) and weight changes ⁴⁴ . Obesity thus has a polygenic architecture, with small effects of each associated gene ⁵ . The polygenic basis of adiposity may provide small sensitivities to environmental influences. More recent data suggested that obesogenic environments accentuate the genetic risk of obesity ⁴⁵ : using BMI as an outcome and a 10-variant genetic risk score in a socially deprived population, researchers found that genetic risk was associated with 3.8 kg extra weight in a normal subject. These data were compared with 2.9 kg extra weight in the least-deprived group ⁴⁵ . This finding may be taken as evidence for a moderate gene–obesogenic environment interaction.

It has been stated already that it is impossible to directly assess a set point in humans ⁴⁶ . Even in controlled experiments (e.g. using overfeeding or underfeeding protocols), EI and EE cannot be tightly controlled, owing to compensations on both sides of the EB and the components of EB being dynamically inter-related. This issue became obvious in a controlled underfeeding and overfeeding protocol in healthy normal-weight subjects ¹³ . Comparing the differences between (i) the EB data calculated from the difference between EI and EE and (ii) EB data based on changes in accurate measurements of FM and FFM (assuming their constants for energy equivalents) as an alternative approach resulted in considerable discrepancies between the two measures of EB of about 800 to 1,000 kcal/day ⁴⁷ . These discrepancies cannot be explained by the limited precision of the methods used to assess either EI or EE or body composition only ⁴⁷ .

Although EI is considered as the major driver of individual weight gain and the population-wide obesity issue ²³ , it cannot be measured with confidence ⁴⁸ . There are large errors involved in the methods used under everyday living conditions to assess EI at the individual as well as the population level ^{48, 49} ; thus, true variations and between-group differences cannot be differentiated from measurement errors. Today, the measurement of 24-hour EE (by DLW) together with direct assessments of FM and FFM are considered the most precise and valid way to investigate EI during periods of more than 3 to 4 weeks ^{48, 50, 51} . However, this approach does not provide detailed data on food and nutrient intakes. Hence, the great dilemma of nutrition research is obvious.

It is obvious that limitations in both our present concepts of biological control and methods used to assess the individual components of EB limit the direct assessment of a biological control of body weight in humans.

Globally, about 20 to 25% of adult populations are presently obese ^{52, 53} . When compared to the previous generation, there was a more than twofold increase in the prevalence of obesity in affluent societies. This is indirect evidence for the idea that a tight control of body weight is unlikely to exist in humans living in an obesogenic environment. Alternatively, the obesity epidemic is presumably due to environmental, societal, and economic “drivers” rather than the proposed biological determinants of body weight ^{20, 23} .

A tight biological control of body weight is also questioned by repeated measurements of body weight, which show a considerable intra-individual variance in its spontaneous day-to-day changes ^{13, 24} . Even during controlled underfeeding and overfeeding, there are high intra-individual day-to-day variances in weight loss and weight gain which resemble inter-individual variances in weight changes ^{13, 24, 54} . However, when compared with weight loss, gaining body weight is slow, suggesting again that body weight is, to a certain degree (almost not perfectly), defended. It has been proposed that this defense may constrain weight changes ⁵⁵ .

As far as weight-reduced obese patients are concerned, only 20% maintain at least 10% weight loss over a period of 1 year, suggesting that in free-living trials weight loss maintenance is difficult to hold ⁵⁶ . This is also true in lifestyle intervention trials, such as the Diabetes Prevention Program ⁵⁷ and the Look AHEAD Study ⁵⁸ . In the "Biggest Loser Competition", an extreme weight loss was observed with 58 kg at the end of the competition ⁵⁹ . However, the regain was 41 kg after 6 years of follow up ⁵⁹ . This regain was taken as evidence for the set point paradigm. In addition, long-term adaptation and a stable body weight after weight loss following bariatric surgery has been proposed to reflect a permanent re-setting of the body weight set point, restoring “normal” leptin signaling (or its downstream signals) in the hypothalamus ^{60, 61} . Accordingly, after bariatric surgery, the proposed re-programming of the body weight defense mechanisms at a lower body weight was not associated with increased hunger feelings or reduced EE ⁶¹ . By contrast, animal data suggest no increase in hypothalamic leptin sensitivity after weight loss due to bariatric surgery, questioning the idea of re-programming a set point ⁶⁰ .

Thus, observational studies on greater populations living in affluent societies and also clinical data on obese patients do not provide consistent evidence for a biological control of body weight.

Following the framework of the “dual intervention model” ^{1, 20} , biological control of body weight may be overridden by strong environmental and economic drivers (i.e. uncompensated factors) of overweight. Thus, studying overweight subjects in affluent societies is unlikely to address biological control (if it exists). Alternatively, body weight homeostasis should be investigated in lean subjects undergoing weight gain, weight loss, and weight maintenance.

Investigating crude phenotypes like BMI and WC (i.e. phenotypes which have been most frequently used in studies on heritability estimates as well as in GWAS) is spurious and cannot provide any deep insights. This is because BMI and WC are merely surrogate measures of nutritional status. BMI is calculated from weight and height squared and has no biological meaning. Both BMI and WC have practical value in daily clinical practice. By contrast, they are weak outcomes in research on body weight homeostasis ^{62– 64} . Furthermore, there are ethnic differences in the associations between these crude measures and FM, FFM, or VAT which lead us to question their value as a phenotype to be used in multicenter studies on subjects with different ethnic backgrounds ⁶⁵ . It is worthwhile to remember that the concept of BMI dates back to a period of underdeveloped scientific methodologies and simplistic theories ^{63, 64} . By contrast, at the time of modern biomedical science, still keeping to BMI and/or WC is unacceptable.

Alternatively, addressing a suitable phenotype, a framework is needed to assess the structures of the body ^{66, 67} . To do so, we should start with a simple question: what do we really want to know? In the case of obesity, we are interested in excess FM, which can be easily measured by, for example, bioelectrical impedance analysis (BIA) in population studies and/or by either densitometry (as assessed by air displacement plethysmography [ADP]) or dual energy absorptiometry (DXA) in clinical studies. Obesity-related cardio-metabolic risks are characterized by hyperinsulinemia (i.e. basal plasma insulin levels >7–10 μU/mL), hypertriglyceridemia, elevated biomarkers of inflammation, and high blood pressure. Alternatively, an estimate of liver fat (as measured qualitatively using ultrasound [US] or MRS, Magnetic Resonance Spectroscopy or by biochemical estimates, e.g. liver enzymes and fetuin A) or VAT (as measured by magnetic resonance imaging [MRI]) can be used as a risk estimate. As far as malnutrition is concerned, this is characterized by recent weight loss (in the case of wasting diseases) and/or a low muscle mass (i.e. sarcopenia, which is characterized by either DXA or MRI or multifrequency BIA validated against those two reference methods). Fluid overload is again measured with confidence using multifrequency BIA or D ₂O dilution. Finally, in the case of risk of osteoporosis, bone mineral density plus skeletal muscle mass are assessed by DXA measurements.

Presently, it is unclear whether body weight control is about control of (i) the static masses of the body, including masses of individual organs and tissues (which add up to body weight), and/or (ii) the association between FM and FFM and their concerted changes when body weight changes. In any case, focusing on body mass or masses of organs and tissues alone does not take into account different levels and systems of control ²⁴ . Thus, to go on a more systemic approach to body weight homeostasis, we need to take into account different levels and systems of control ⁶⁷ .

Up to now, studies on body weight control in humans have addressed the structural level only (level 1 in Figure 1). According to a more advanced model ⁶⁷ , control is about relationships within and between structures, their related functions and systemic outcomes, and thus not about body weight or its individual components only. The masses of organs and tissues and their inter-relationships (level 1) have to be addressed in the contexts of neurohumoral control (level 2) together with metabolic (e.g. EE, i.e. level 3) and systemic outcomes (e.g. heart rate, blood pressure, respiration, excretion, and body temperature, i.e. level 4) ( Figure 1).

Following that “multi-level–multi-systemic” model, body weight control (if it exists) is likely to happen between, but not at, a single mass and level (i.e. control is about operating within and between levels). Then changes in body weight follow the control of the associations within levels (e.g. between FM and FFM or between individual organs and tissues) as well as between different levels (e.g. between structures and metabolic functions). Thus, strictly speaking, the issue of body weight control is about associations within and between different levels and systems which add up to maintain a stable body weight.

Consequently, phenotypes worthwhile of study are related to body mass–body function relationships rather than to body weight itself ⁶⁷ ( Figure 1). The concept of functional body composition ^{67, 68} refers to, for example, the association between FFM (and its anatomical and physical characteristics) and REE in the context of neurohumoral control (e.g. thyroid state and SNS activity) and related systemic outcomes (e.g. heart rate and body temperature). Similarly, the association between FM (or its distribution) and plasma leptin levels has to be seen in the contexts of insulin resistance, O2 consumption and CO ₂ production, as a measure of lipid oxidation and respiration. To go on with that idea, the structure–function associations and their changes have to be studied separately in different situations, e.g. before, during, and after weight change as well as during weight maintenance. As far as weight loss is concerned, it has been shown recently that control systems involved in metabolic adaptation differ between weight loss and weight maintenance ⁶⁹ .

Adaptive thermogenesis (AT) refers to changes in EE which are independent from changes in FFM and FFM composition ^{69– 73} . Since there is a considerable inter-individual variance, AT may provide a suitable phenotype to investigate in future studies. AT is asymmetric (i.e. during weight loss it adds to energy sparing, whereas no energy dissipation is observed with overfeeding ^{70, 71} ). AT is also observed after massive weight loss following bariatric surgery in severely obese patients ⁷² . During early starvation, AT is related to hepatic glycogen depletion and the fall in insulin secretion ^{13, 69} . Thus, AT is considered a metabolic adaptation to meet glucose oxidation in the brain (i.e. the brain's energy metabolism requires 80–100 g glucose per day ⁶⁹ ). AT is associated with systemic outcomes, i.e. decreases in body temperature, heart rate, and glomerular filtration rate ¹³ . These data suggest that AT is part of the concerted physiological response to weight loss, with the fall in insulin secretion as its major characteristic ⁶⁹ . In the long term, AT adds to weight loss maintenance ^{69, 73– 75} where AT is related to the low plasma leptin levels sparing triglycerides stored in adipose tissue where a low FM limits biological functions like reproduction ⁶⁹ . Obviously, the meaning of AT varies with the phase of caloric restriction and weight loss.

AT has been related to the set point and settling point paradigms. Recently, three models for AT have been proposed ⁷⁴ . First, a “mechanical model” related to the settling of body weight; second, a “threshold model”, where AT is related to the decrease in FM below a minimum (i.e. a set point); and, third, a so-called “spring-loading model” with an effect on the dynamics of weight loss. No model fully explained AT. However, during weight maintenance, decreases in REE were consistent with the threshold model and thus a low set point related to ES in adipose tissue and plasma leptin levels.

As far as metabolic adaptation during early weight loss is concerned, another threshold (or a low set point) related to the depletion of hepatic glycogen stores has been proposed ⁶⁹ . This finding is in line with the idea that energy allocation to the brain controls EE (i.e. the brain is assumed to have a hierarchical position in whole-body energy metabolism ⁷⁵ ). The brain has a high metabolic rate ^{14, 76} and is the only organ which does not lose weight with weight loss ¹³ . When compared with muscle-specific metabolic rates, the brain and skeletal muscle differ by a factor of 18 ⁷⁶ . Since the brain demands a constant energy budget, it has “pole position” in a competitive situation of whole-body energy allocation and thus control of whole-body EE ( Figure 2).

Sedentary behavior and physical activity level also impact the control of appetite, satiety, and body weight. Becoming sedentary does not downregulate EI, i.e. at low physical activity there is a weak coupling between EE and EI ^{16, 77, 78} . By contrast, an increase in physical activity improves satiety signaling ^{16, 77} and also increases activity EE (AEE). Then both REE and AEE drive EI ⁷⁷ . In fact, following the spectrum from low to high physical activity, there are “regulated” and “non-regulated” areas of EI ⁷⁷ . The association between physical activity and EI is J-shaped ¹⁶ . However, in clinical practice, physical activity (or even exercise) does not add much to the treatment of obese patients ^{79, 80} . This paradox is explained by compensations in the individual components of EE. EE increases with low and moderate physical activity but plateaus at high activities to maintain EE within the target range (that is, EE is constrained with respect to physical activity ^{81, 82} ). In fact, with a high level of exercise, the compensation is explained by a reduced REE, which resembles a decrease in basal biological functions, e.g. reproduction.

The finding of a tight control of EE gives rise to speculation that body weight control (if it exists) is about control of EE ¹⁴ . This idea provides an alternative paradigm, which is a putative set point or dual intervention point model of EE. The latter framework is characterized by a lower boundary (or lower set point of EE) given by the lowest metabolic rate needed for survival, whereas the upper set point of EE is explained by maximum mitochondrial capacity of cells of the body. If this EE–set point paradigm holds true, this would put recent concepts of body weight homeostasis into perspective. Then, any change in body weight follows the variance of EI at a tightly controlled EE, i.e. body weight itself is not controlled but results from the balance between a tight control of EE at loosely controlled EI.

During the last 25 years, most research on body weight control has been adipocentric. However, since FM accounts for only 10 to 40% of body weight, any control of FM can represent only a similarly sized part of the body weight control. In addition, FM (as far as it can be assessed today) is constant throughout a day, and it starts to decline within 72 hours in response to starvation ¹³ . Thus, FM by itself (as far as it’s mass can be measured by todays methods) has no association with daily meal frequency and short-term calorie restriction ¹ . However, since decreases in plasma leptin concentrations observed with caloric restriction precede and exceed the changes in FM ¹³ , the adipocentric view of body weight control refers to both the size and the secretory activity of FM.

It is only recently that, besides FM, FFM came into the center of research on body weight control ^{16, 77, 78, 84, 85} . Obviously, there is need for a broader view on the control of EB and body weight taking into account biological inputs from sensory systems (i.e. taste and olfactory signals), the gut and feedback related to FFM (controlling protein and glycogen content) and FM (controlling fatness), and the role of hedonism and rewards in the contexts of environmental and behavioral pressures ⁷⁶ . Comparing different inputs, tonic afferences from FFM (signaling energy demands and metabolic requirements) and FM (signaling ES) have to be differentiated from episodic or dynamic feedback from the gut (signaling nutrient availability and meal and macronutrient intake by neural and enteroendocrine signals) ¹⁶ .

EI is regulated by interactions between homeostatic and non-homeostatic mechanisms; it is thus influenced by experiences, learning, and culture ⁸⁶ . This helps to explain why individual factors (like leptin and insulin) may have limited effects on EI. Using functional MRI (fMRI) after intra-nasal insulin application, selective insulin resistance in the prefrontal cortex (responsible for cognitive control and decision making) and in the hypothalamus was characterized as being associated with reduced inhibition of EI, food craving, and thus overeating in obese patients ⁸⁷ . In addition to the “cognitive brain”, hedonic and incentive signals related to brain reward systems of the “emotional brain” (related to the mesoaccumbal dopamine system) may further add to overeating (i.e. eating is pleasurable and rewarding). In line with this idea, two set points, a homoeostatic and a hedonic set point, have been proposed, with obesity affecting the balance between the two and one inducing shift in the other ⁸⁸ . Going on with that idea, “metabolic obesity” (with a genetically determined set point) and “hedonic obesity” (due to hedonic overeating overriding the homoeostatic set point) has been defined, and the two types of obesity may serve as a future stratification in the treatment of obese patients ⁸⁹ . However, the quantitative effects of non-homeostatic influences on the set point model are unknown.