New insights into understanding obesity: from measures to mechanisms

Introduction

So far, understanding the role of obesity in determining broader health and mortality has largely been a story of indirect measurements (eg, body mass index) and complex inferences.1 Although associations are reliable,2 3 applied causal estimates compelling,4 5 and (new and established) intervention effects apparently helpful,6 7 most of these relationships are the product of a large collection of possible pathways to effect. To dissect the nature of obesity as a disease and as a risk factor for comorbidities and mortality, there is clearly a need to understand mechanism, but also to navigate available measures that relate to factors both shaping and shaped by the underlying biology of obesity. Taken in this way, the study of mechanism does not refer exclusively to the causes of obesity, but rather to the biological events captured by the measures we have available to characterise adiposity. In this context, the phenotypes (eg, body mass index) or disease states (eg, obesity) typically used to identify excess adiposity can be considered as indirect markers revealing underlying biological features that both feed into observed variation in adiposity and are the result of it.

This open approach to understanding mechanisms is in line with the more holistic definition of obesity, as presented in the Obesity Society 2018 position statement (“a multi-causal chronic disease recognised across the lifespan resulting from long term positive energy balance with development of excess adiposity”8) and is key to the interpretation of a growing collection of available data that should enable a re-evaluation of the phenotypic definition of obesity.9 Based on a series of examples, this review therefore focuses on the molecular characterisation of obesity as an approach to better understanding mechanism also articulating the concern that increasingly granular intermediate measurement does not automatically result in the clarification of causal pathways.

Epidemiology of obesity

The World Obesity Atlas report for 2023 suggests that more than four billion people could have overweight or obesity (body mass index ≥25) by 2035.10 This count reflects an increase from 38% of the world's population in 2020 to more than 50% (excluding children aged <5 years) and an increase in the prevalence of obesity (body mass index ≥30) alone from 14% to 24%. Based on global statistics for adults (aged ≥20 years) in 2020, rates of obesity are higher in women (18%) than in men (14%).10 In terms of worldwide distribution, the prevalence of obesity typically increases with increasing income (as defined by the World Bank income levels), with 7% of the population in low income countries having obesity compared with 25% in high income countries.11

The largest rise in the prevalence of obesity to 2035 is predicted to occur in children and adolescents, with expectations that global prevalence will rise from 10% to 20% in boys and from 8% to 18% in girls.10 Based on data from 1975 to 2016, rising trends in body mass index in children and adolescents seem to have plateaued in many high income countries, but have accelerated in parts of Asia.12 This trend is set to continue, with the prevalence of obesity in young people in low income countries predicted to increase dramatically by 2035.10 These global trends are important because overweight and obesity are potent risk factors for other diseases, making it a primary target for intervention.13 Similarly, clinically used continuous measures of healthy weight, such as body mass index, have outstanding associations with health outcomes.14–17 For example, 3.6% of all new cancer cases in adults in 2012 have been attributed to high body mass index, which increases to 5.3% when considering only those countries with high human development indices.18

Sources and selection criteria

For this specialist review, evidence was pooled by an expert author group focused on allocated subsections that explore the complexity of available measures increasingly used to examine associations between obesity and health. This task was divided between authors, with each area including a selected set of references, which were known to the authors, and also searched for with relevant search tools (PubMed and Google Scholar). Evidence collected was for the period 1999-2023 (with an emphasis on recent publications) and content was reviewed by a clinical contributor (DJP) and the chair of the UK Office of Life Science Obesity Mission (NS). Lastly, independent readers guided the presentation of the review so as to provide evidence of the complexity in interpreting obesity related measures and to carefully assess the appropriate definition of obesity and the likely broad phenotypic footprint of variation in adiposity.

Using molecular phenotypes to measure mechanism

Emerging approaches to deep molecular phenotyping can theoretically provide insights into the mechanisms underlying obesity while simultaneously assessing the molecular changes that might be driven by obesity. Considering the molecular traits that form the basis of complex phenotypes, updated biochemical techniques and informatic developments are now delivering automated high throughput arrays capable of efficiently gathering huge numbers of measures over multiple domains.19 20 The resultant datasets are transforming our ability to describe physiological states and are providing detailed readouts of cellular activity behind exposures of interest.21 When implemented across different study designs, this molecular fingerprinting can help dissect the causes and consequences of variation in adiposity.

Rapidly changing analytical opportunities are being presented through developing epidemiological resources in this area. In a time like no other, analyses can cross examine studies comprising hundreds of thousands or even millions of participants, all with individual participant level data available. This progress in conducting population health studies is promising but also generates complexity. The analysis of indirect measures in study frameworks with excellent analytical precision can generate conditions that not only confuse inference (often through a multitude of observed associations) but also present potential hazards in applied analyses.22 This situation therefore generates the question as to how we should take a context of reliable associations between health and obesity and combine it with current data and approaches to better understand excess adiposity? Clearly we should try to avoid a natural tendency to assume that detailed omic measures used to characterise variables or risk factors are themselves direct measurements of pathways to affect. However, we should be able to embrace multiple, detailed, domain and study specific evidence to capture the mechanisms associated with obesity. In the following sections, we explore this using a limited set of example measures and experimental designs to show how it is possible to bring together a more comprehensive view of the biological mechanisms contributing to obesity and its consequences (figure 1).

Request permission

Genetic contributions to obesity: anchoring inference

There is consistent evidence that inter-individual variation in body mass index and the risk of developing obesity and overweight have a substantive genetic component, with heritability estimates for body mass index typically in the range of 40-70%.23 Although heritability estimates are expected to vary between studies because of natural variations in the balance between environmental and genetic contributions, differences in study design have also been shown to contribute to this substantial heterogeneity in reported estimates.24 There is also a suggestion that the genetic contribution to body mass index varies with age, being relatively greater during childhood.24 Within these genetic underpinnings, a range of effects exist that span rare, highly penetrant mutations with large genetic effects through to common genetic changes that have smaller phenotypic effects and complex patterns of penetrance. This reflects a broad genetic architecture that is driven by the variable nature of these genetic effects, the range of adiposity measures and the relative proximity of these measures to underlying genotypic effects.25

Genetic contributions to variation in adiposity give insight into factors likely to be important in understanding mechanisms captured by these traits. There is also the theoretical insight gained from the primacy of genotype over naturally downstream processes and factors (the central dogma), which can enable insight that, unlike other observational measures, puts more emphasis on the mechanisms contributing to, rather than being associated with, obesity. As an illustration and starting with rare and relatively simple genetic variation, variants within the MC4R gene region were identified in early family based linkage studies23 26 and these variants have some of the largest observed associations with obesity. Functional MC4R mutations disrupt the leptin-melanocortin pathway, which is involved in the regulation of appetite. Recent population based work undertaken in thousands of life course study participants has shown that individuals carrying MC4R loss of function mutations have substantial differences in body mass index, weight, fat mass, and lean mass from as early as age five years.27 Furthermore, although still considered rare in isolation, collectively these loss of function mutations in MC4R might occur at a considerably higher frequency than previously estimated.27–31 This information is not only useful for estimating the number of individuals that might have these mutations and for reducing the stigma around obesity through greater understanding, but also for describing specific biological mechanisms that are likely involved in obesity more widely. For example, we now know that loss of function mutations in any one of several genes acting in the leptin-melanocortin pathway (eg, POMC, LEPTIN, LEPR) can cause hyperphagia induced and severe early onset obesity,32 leading to increased rates of morbidity and mortality in carriers, including in children.33

At the other end of the genetic architecture spectrum, studies focusing on identifying common genetic variation have combined population scale health investigations with common variant data for the whole genome, using genome-wide association studies (GWAS). Advanced by technological developments in the early 2000s, the ability to examine associations between millions of genetic variants capturing most of the common changes in the human genome facilitated a shift from targeted gene studies to hypothesis free scans of the entire genome.34 The resultant GWAS are now common in genetic epidemiology and the costs sufficiently low to enable very large studies to be conducted.

The fat mass and obesity related locus FTO was the first reliable common variant association with body mass index. The implicated specific single nucleotide variants (called single nucleotide polymorphisms) reliably associated with adiposity were initially detected in one of the first GWAS studies undertaken in thousands of participants for type 2 diabetes,35 but later confirmed for primary associations with body mass index.36 Implicated single nucleotide polymorphisms have been associated with energy intake37 and diet,38 both suggesting that behavioural mechanisms contribute to adiposity. However, work has also suggested that variants near the FTO gene region might alter the function of nearby genes (IRX3 and IRX5), which in turn influence how fat cells (adipocytes) are involved in thermogenesis (the burning of stored energy to produce heat). In this case, apparently functional FTO alleles (which affect the expression of local genes IRX3/5 during early adipogenesis) appear to repress mitochondrial thermogenesis in adipocyte precursor cells. This results in a shift from heat producing adipocytes to energy storing adipocytes, with a reduction in thermogenesis, as well as an increase in lipid storage.39 Other model work has suggested that altered thermogenic capacity and resistance to adiposity induced by a high fat diet might explain the genetic associations.40 Of relevance for this review, these insights highlight how common genetic variant signals are potentially complex in interpretation but have advanced possible pathways relating to variation in levels of adiposity.

Sample sizes of GWAS have grown substantially and the international GIANT (Genetic Investigation of Anthropometric Traits) consortium (https://portals.broadinstitute.org/collaboration/giant/index.php/GIANT_consortium) periodically deliver updated meta-GWAS for a range of measures of human body size and shape. Expanding on an early GWAS meta-analysis and the discovery of six single nucleotide polymorphisms with association signals detected in a study of 32 000 individuals,41 the largest and most recent meta-GWAS for body mass index identified more than 900 independent single nucleotide polymorphisms associated with body mass index in more than 650 000 individuals.42 Although limited in penetrance and individual effect sizes, these collections of genetic contributions not only confirm the validity of the common disease, common variant hypothesis,43 but also reflect the multidimensional nature of obesity. These signals include over-representation of loci involved in complex, multiple pathway complexes, such as the central nervous system,44 among a back collection of other candidates and poorly explained associations. Importantly, these findings highlight the likely heterogeneous nature of mechanisms contributing to obesity and a need to consider the breadth of possible downstream pathologies related to this complex disease and a need for more dynamic approaches to the interpretation of detected effects.

So far, genomics initiatives have been limited by their focus on European populations and a reliance on simplified measures of adiposity variation (ie, mainly in analyses of obesity as a state and body mass index as a continuum), although some work does exist employing measures of regional adiposity.45–47 These relatively simple studies, however, have led to numerous follow-up techniques and analyses looking to explain signals of association and learn from this initiative.23 48 49 Furthermore, the use of aggregate scores summarising common variant carriage and trained to the results from analytically powerful studies of body mass index (as done for other health outcomes) have been used in efforts to track associations with anthropometric measures throughout life50 and risk profiles in later life.51

Not covered in detail here, the related disciplines of transcriptomics and epigenetics also offer further potential insight into how variation in DNA sequence translates into differences in phenotype and risk of disease, albeit with the added complication of also reflecting response to outward stimuli. Transcriptome-wide association studies have evolved, originally with microarrays but now increasingly performed with RNA sequencing data. So far, studies have uncovered new genes associated with body mass index as well as confirming DNA sequence based findings.52 Epigenetic changes are genetic modifications of genes that change their activity without changing the DNA sequence, the most studied (at least at scale) being DNA methylation. Epigenetic variability is influenced by genetics (ie, DNA sequence variation) as well as the environment and studies support a role for epigenetics in the development of obesity but also provide evidence for epigenetic changes in response to the disease.53 54 Likely of clinical interest is the potential for epigenetics to contribute to phenotypic (and risk) correlations across generations as well as the potential therapeutic opportunities afforded by its reversible nature.55 56

Intermediate molecular phenotypes and mechanistic insight

The measurement of biological "intermediates"—including proteomics, metabolomics, and the microbiome—at scale can improve understanding of events coincident with obesity as well as providing insight into routes to subsequent pathology. Protein and metabolite datasets derived from serum and plasma are the largest and most numerous intermediate phenotype or omic datasets currently available and, when implemented across different study designs, association signals can help delineate the causes and consequences of variation in adiposity.

Studies of the human microbiome

A compelling intermediate measure, proposed as a cause of obesity and as a link between obesity and adverse health outcomes, is the gut microbiome.86 Found predominantly within the large intestine, the gut microbiome is a complex system of microorganisms (ie, bacteria, fungi, and viruses) that help digestion, provide protection against harmful pathogens, and create essential metabolites that humans would otherwise be unable to produce.87 Measurable, labile, and intuitively linked to weight regulation processes (eg, digestion, gut physiology, and metabolism), the microbiome is promising as a biological intermediate and potential mechanism. The number of studies researching the role of the gut microbiome in human health has increased substantially, with some suggesting causality in relationships, whereby changes in microbiota distributions may lead to better or worse health. For example, in a small study of species level microbiome variation and dietary interventions (n=12 participants with obesity and n=2 controls), evidence suggested that those living with obesity had fewer Bacteroidetes and more Firmicutes before intervention, that over time the relative abundance of Bacteroidetes and Firmicutes changed irrespective of diet type, and that increased abundance of Bacteroidetes was associated with percentage loss of body weight.88

While promising, current associations have been largely observational, based on small samples and often inconsistent.89–91 Most evidence comes from model or small scale human studies that have the limitations of classic epidemiological approaches (ie, confounding, bias, and reverse causality).92 93 Furthermore, many human studies have focused on European individuals and, therefore, the generalisability of findings can be questionable, especially when drawing inference for populations with substantially different environmental, phenotypic, and genetic backgrounds. To illustrate the challenges in this research area, about 20 studies have already looked at the contribution of common genetic variation in humans (as host) to measures of the gut microbiome, but limited overlap has been found in the genetic signals highlighted so far. Reliable results are currently limited in real terms to MCM6/LCT and the ABO locus,94 95 probably because of the limited sample sizes (ranging from 93 to 18 340 individuals), population differences, and variation in study protocols, including in microbiome measurement and variable derivation, sequencing platforms, bioinformatic processing, variable transformations, and analysis pipelines.

Despite challenges, efforts have been made to build inference around specific associations between the microbiome and adiposity. Specifically, applying genetic causal inference analysis96–98 to assessing the role of the amount of Bifidobacterium on traits related to adiposity suggested that lower amounts of Bifidobacterium could be associated with waist circumference, weight, and body mass index.94 Given the current landscape of GWAS of the gut microbiome and the overall lack of understanding of the mechanisms linking host genetic variation and the microbiome, however, these relationships could also be explained by alternative mechanisms. For example, microbiome related genetic variation can influence host behaviour (eg, milk consumption), which in turn alters the microbiome profile. This effect could be a problem in the case of hypothetically coincident effects of genotype on both exposure (microbiome) and outcome (body mass index). Apparently causal estimates of this type could also be explained by genetic variation related to the microbiome having a primary effect on adiposity, thus reversing the direction of inference and suggesting a potentially more parsimonious explanation—that traits related to adiposity and their precursors might be driving variation in the microbiome. More evidence in this area is needed.

Value of detailed measurement in the context of therapeutic interventions

As we enter a new era in the management of obesity,99 an opportunity exists to learn about mechanisms associated with obesity by complementing the existing epidemiological and experimental data with that examining the effect of treatment. Therapeutic interventions developed from basic scientific discovery can be used to confirm causal pathways and also to generate more knowledge about the disease.100 Current treatments include multicomponent lifestyle interventions, as well as bariatric surgery or drug treatment, or both (table 1).101–103 The recent emergence of incretin related treatments has demonstrated a promising opportunity for substantial weight reduction in individuals with overweight and obesity.104 105 Although clinically distinct, the effect of these treatments is multifactorial and overlapping, each engendering a host of changes that are both physiological and behavioural, but which target common routes to altering adiposity.

Randomised controlled trials published so far have effectively characterised the effect of these treatments on several distinct and well defined patient outcomes focusing on treatment efficacy, usually in terms of the primary outcome of weight loss, as well as the potential for (short term) adverse side effects. However, there remains the obvious question as to whether these effective interventions are similar at other levels? Randomised controlled trials are not generally designed to capture wider treatment effects, such as concomitant changes in diet, drug treatment use, and lifestyle, nor do they enable a molecular cross examination of response to treatment. These characteristics might be important determinants of the effectiveness of treatments and will provide evidence of the common effects across treatment modalities. By measuring detailed molecular phenotypes over the course of an intervention, better characterisation of the intervention effects and biological associates of variation in adiposity will be possible. Challenges related to small sample sizes, limited length of follow-up and the predominantly cross sectional design of nested studies remain, but the presence of specific and randomised interventions with simultaneous detailed characterisation present opportunities to investigate the mechanisms of—and responses to—obesity.77 106

Emerging studies

The ongoing development of substantial resources in epidemiology, in particular those designed to prioritise diversity and representativeness, such as Our Future Health in the UK (https://ourfuturehealth.org.uk/) and All of Us in the US (https://allofus.nih.gov/), when combined with detailed molecular characterisation of the effect of interventions will allow researchers of obesity to look at more nuanced questions about the disease and its implications for the health of populations. One of the priorities must be to improve our understanding of the heterogeneity of overweight and obesity as a disease and how different presentations of the disease might affect patient outcomes (eg, incidence of comorbidities) and their response to the range of therapeutic interventions currently available.107 Further analysis of the mechanisms associated with obesity in this context has the potential to identify clinically relevant obesity subtypes and work has already begun on this topic.107 Studies that have used clinical biomarker data or genetic data, or both, have provided evidence of heterogeneity in obesity,108 including approaches that exploit the apparent variation in risk of cardiometabolic comorbidities (eg, type 2 diabetes) in individuals with a similar body mass index.109 110 Going forward, the integration of common omic measures from different study designs and phenotypic measures (eg, imaging based measures of body fat distribution), will inevitably contribute to the important question of disease heterogeneity.

Conclusions

In this review, we have invited a reconsideration of how to approach studies exploring the mechanisms of obesity. Focusing on adiposity as well as obesity, we have updated areas of potential mechanistic insight and described the reality of those insights with respect to both cause and consequence. Although assuming that mechanism implies causation might be natural, we argue that mechanisms (or our best measures of them) can be associated with obesity without being causal. Despite the growing collection of detailed intermediate measures in human datasets, it is therefore not immediately possible to assert that observed variation in molecular phenotypes are clear targets for next step obesity intervention. However, through a combination of study designs, differential exposures, and evidence integration, a growing view of mechanism and downstream effect is emerging. The use of multi-omics approaches in analysing observational, genetic, and intervention studies is revealing a broad physiological response to increased adiposity which agrees with the extensive collection of health outcomes likely causally related to excess adiposity and presents many potential targets that might enhance the growing number of treatments for obesity. Although this review has focused on the molecular characterisation of obesity as an approach to understanding mechanism, with the increase in the global incidence of obesity over the past four decades12 inextricably linked to the corresponding spread of an obesogenic environment,111 improved understanding of the environmental and societal components of the disease will also be necessary to tackle the obesity epidemic.112

Effective integration of different experimental designs, where an array of measures of health outcomes and intermediate phenotypes are available, will be necessary if we are to further enrich an open interpretation of obesity associated risk. This approach will be important for more effective management of the health burden related to obesity by better use of existing treatments and could also lead to other less obvious targets. The effective use of a growing portfolio of obesity treatments for patients can only be done with information about the common and unique on-and-off target effects of these interventions. Whether cause or consequence, the ability to characterise these effects and associations will be enhanced by the types of study described here and will inform clinical decision making. This improved understanding should also help to effectively reduce the stigma surrounding obesity and overweight and, where relevant, simplify approaches to the management of obesity at a population scale—something that might need simplicity in delivery (against a backdrop of major problems such as social inequality) and that could also correspond with casual messages from the most effective interventions.

Obesity is a complex trait in the truest sense. The heterogeneity behind this trait at an individual level and the breadth of associations with it encourage an abandoning of simplified searches for mechanisms of obesity. In place of this, we would encourage a more fluid consideration of mechanisms associated with obesity if we are to make best use of the tools available now and in the future.