The prevalence of overweight and obese individuals is a significant global problem(1). Obesity develops due to an imbalance between energy intake and energy expenditure. Energy balance represents a conglomerate of traits, each influenced by numerous variables such as behaviour, diet, environment, social structures and genetics. Complex interactions among these variables (for example, gene–environment, gene–diet, gene–gene) contribute not only to individual differences in the development of obesity, but also in the response to interventions and treatments (Fig. 1).
Weight maintenance is complex, involving both central and peripheral inputs that are simultaneously affected by highly interactive genetic architecture (involving 100s or even 1000s of genes) and multiple environmental stimuli. Here, we have attempted to depict only a fraction of the components that contribute to obesity, specifically focusing on the novel phenotypes described in the body of the text (exercise, diet and the gut microbiome) and the potential interactions among them.
Mouse models have historically played valuable roles in understanding the genetics of traits related to energy balance and obesity in many mammalian species including companion animals, agricultural species and man. Initially, mouse models were successful in identifying individual genes that had large effects and that were inherited in a simple Mendelian pattern(Reference Bennett, Farber and Orozco2). Mice with naturally occurring mutations, as well as those with mutations that were genetically engineered or chemically induced, have played a crucial role in furthering our understanding of how specific genes may contribute to disease aetiology. In many instances (for example, the ob/ob and db/db mice, and the melanocortin-4 receptor (MC4R) knock-out mouse)(Reference Campfield, Smith and Guisez3–Reference Ste Marie, Miura and Marsh5), these mutations have identified key proteins involved in obesity and the regulation of energy balance. Yet, investigation into the role of these target proteins has revealed that they do not work in isolation and that the alteration or absence of a single protein cannot explain common obesity-related phenotypes in human populations(Reference Considine, Considine and Williams6). Thus, it is generally accepted that obesity and energy balance are highly complex and polygenic traits, regulated by numerous genes that have an impact on many physiological systems, and that interact not only with each other but also with many environmental stimuli.
Recent advancements in genetics research have employed the use of genome-wide association studies (GWAS) to determine the genetic underpinnings of energy balance and obesity in human populations. Several SNP have been implicated in obesity, yet many have failed to replicate across studies(Reference Hindorff, Sethupathy and Junkins7–Reference Andreasen, Mogensen and Borch-Johnsen9), and confirmed findings account for only a small percentage of the phenotypic variance across the population tested(Reference Bogardus10). To date, GWAS have consistently identified two SNP significantly associated with obesity, one in an intron of the FTO gene and one downstream from the MC4R gene. These two SNP have relatively small effects, accounting for less than 2 % of the variance in adult BMI(Reference Frayling, Timpson and Weedon11, Reference Loos, Lindgren and Li12). In conjunction with the relatively low level of explanatory power with regard to percentage phenotypic variance, GWAS have additional limitations. For example, hundreds of thousands to million of SNP may be evaluated in a single sample, reducing the statistical power to identify significant associations. Moreover, heterogeneity requires extremely large sample sizes to reveal moderate to small genetic effects, limiting the discovery of contributions of rare alleles(Reference Cohen, Kiss and Pertsemlidis13–Reference Cheverud, Lawson and Fawcett15).
In mice, early studies on the genetic architecture of complex traits employed linkage analysis based on simple F2 crosses of phenotypically divergent inbred strains to identify obesity loci, but the mapping resolution of such crosses proved prohibitive for identifying target genes(Reference Bennett, Farber and Orozco2). Currently, in light of the ability to evaluate human populations directly using GWAS, mouse models have evolved to better address the complexity of the genetic architecture of obesity and related traits. Novel approaches include more complex population structures of advanced intercross lines (AIL) derived from inbred lines and heterogeneous outbred stocks, as well as association analysis based on haplotype–phenotype relationships across large numbers of inbred strains. Moreover, a unique and novel mouse population, the Collaborative Cross (CC), was designed specifically to model the genetic diversity of the human population(Reference Churchill, Airey and Allayee16, Reference Threadgill, Hunter and Williams17). Modern mouse models offer specific advantages over GWAS in that it is possible to completely control environmental variables, population structure and trait variants, lending more power to detect statistically significant associations(Reference Cheverud, Lawson and Fawcett15). Furthermore, studies using mouse models enable the interrogation of obesity-related phenotypes that are not easily amenable to large-scale analysis in humans, including, for example, levels of voluntary exercise.
In the present review, we summarise recent advances in the development and use of mouse models designed to understand the genetic architecture of obesity and its component traits. We highlight innovative and novel approaches to understanding the complexity of obesity, such as the CC, which will provide unprecedented resolution for mapping complex traits. In addition, we describe recent advancements in applying mouse models to begin to dissect genetic predisposition to previously understudied traits such as voluntary exercise, and the interactive relationship between host genetics and composition of the gut microbiome.
Mouse models for understanding the genetic architecture of obesity and related traits
Inbred mouse strains
Phenotypic differences in body weight and adiposity across multiple inbred mouse strains reveal how genetic background influences phenotypic expression(Reference Bachmanov, Reed and Beauchamp18, Reference Reed, Bachmanov and Tordoff19). Exploiting this phenotypic variability, interbreeding strains with high and low values for a specific phenotype, creates segregating populations allowing for the detection of genome regions associated with the phenotype of interest, or quantitative trait loci (QTL). To date, hundreds of QTL for body weight and adiposity have been described(Reference Wuschke, Dahm and Schmidt20), yet little is known about the actual genes and the responsible genetic variation that may contribute to the expression of obesity-related phenotypes(Reference Wuschke, Dahm and Schmidt20). Many of these QTL were identified using F2 intercross or backcross populations. QTL mapping of F2 intercross and backcross populations is generally low in power and generalisibility since identified QTL typically account for such small proportions of the phenotypic variation and are background strain dependent(Reference Stylianou, Korstanje and Li21). Furthermore, few F2 intercross QTL studies have used obese or obesity-prone mouse strains such as New Zealand obese (NZO) or KK, biasing the results toward more common inbred strains(Reference Wuschke, Dahm and Schmidt20). Recombinant inbred lines offer increased mapping resolution due to increased recombination resulting from generations of inbreeding(Reference Peirce, Lu and Gu22–Reference Williams, Gu and Qi25). However, many traits may not be mapped to any region of the genome because the two founder strains of the recombinant inbred lines share a large proportion of genetic variation. For example, 48·6 % of genomic regions in BXD strains will not be mapped due to shared variation(Reference Kirby, Kang and Wade26).
Recent advances in computational analysis such as haplotype association mapping (HAM) enable QTL mapping by simultaneously evaluating dozens of inbred strains, providing more statistical power for identifying significant QTL and incorporating more of the variation within the mouse genome. HAM overcomes the resolution issue of traditional F2 and backcrosses(Reference Kirby, Kang and Wade26) by enabling mapping of QTL to much smaller confidence intervals, potentially harbouring fewer candidate genes. HAM is not without limitations, however. For example, the power to detect genetic contributions of small effect is low, making it more difficult to identify QTL associated with complex traits such as obesity(Reference Kirby, Kang and Wade26–Reference Payseur and Place28). In addition, population structure must be taken into account when performing the analysis to eliminate false-positive associations(Reference Kirby, Kang and Wade26–Reference Payseur and Place28).
The Hybrid Mouse Diversity Panel (HMDP) has been proposed as a solution to challenges with population structure in association studies(Reference Bennett, Farber and Orozco2). Comprised of 100 inbred strains and seventy-one recombinant inbred lines, the HMDP is specifically designed to map traits with small variance contributions and to increase power to detect loci with small effects(Reference Bennett, Farber and Orozco2). The combination of recombinant inbred and classical inbred strains will disrupt long spans of linkage disequilibrium within each of the populations, thereby increasing mapping resolution. Preliminary mapping studies using the HMDP have replicated existing QTL identified in classical inbred crosses, yet with greater mapping resolution(Reference Bennett, Farber and Orozco2).
Selectively bred mouse lines
Selective breeding experiments, which have been used for decades to gain an understanding of the genetic underpinnings of growth, fatness and other traits related to obesity and energy balance, can capture extreme phenotypic variation beyond the range of the founding population(Reference Bunger, Laidlaw and Bulfield29). For example, high growth selection lines have been established with mean body weights of up to 77 g, far exceeding the body weights within the base populations from which these lines were selected, as well as body weight in at least 400 inbred mouse strains(Reference Bunger, Laidlaw and Bulfield29, Reference Svenson, Von Smith and Magnani30).
Experiments evaluating the genetic underpinnings of obesity-related traits using selection lines have identified several QTL for early growth, obesity, fatness and exercise behaviour (wheel running)(Reference Rocha, Eisen and Van Vleck31–Reference Allan, Eisen and Pomp33). A large QTL analysis of obesity-related traits in 993 mice from an F2 population derived from a cross between M16i, a mouse strain selected for high growth and body weight at 3–6 weeks of age and inbred, and L6, selected for low growth at 6 weeks of age, revealed many significant QTL for growth and fatness(Reference Rocha, Eisen and Van Vleck31, Reference Rocha, Eisen and Van Vleck32). In addition, several epistatic effects (gene × gene interactions) were identified(Reference Ankra-Badu, Pomp and Shriner34). Evaluating epistasis in QTL analyses aids in the discovery of loci that may have small main effects but large effects when combined with interacting loci(Reference Ankra-Badu, Pomp and Shriner34). Similarly, extensive research examining adiposity and body composition traits has been performed using Large (LG/J) and Small (SM/J) mice, inbred lines selected for large and small body size at age 60 d(Reference Kramer, Vaughn and Pletscher35–Reference Hrbek, de Brito and Wang37). Several QTL for adiposity and body weight were identified, each of only small effect. Furthermore, pleiotropic QTL for adiposity and tail length demonstrated interaction in the genetic control of obesity and body size(Reference Kramer, Vaughn and Pletscher35, Reference Cheverud, Vaughn and Pletscher36).
Although evidence from mice, other model species such as Caenorhabditis elegans (Reference Ashrafi, Chang and Watts38) and humans point to the contribution of literally hundreds of genes to the genetic architecture of body weight, a recent study in dogs paints a distinctly different picture(Reference Boyko, Quignon and Li39). Boyko et al. (Reference Boyko, Quignon and Li39) found that only a few genes of large effect collectively explain most variability in canine body size. This unusual genetic architecture was attributed to several unique aspects of selection in dogs. For example, the authors speculated that many of the modern breeds of dogs were created during the Victorian era where novelty was a focus of selection and breeders favoured the preservation of discrete mutations(Reference Boyko, Quignon and Li39). Repeated selection for rare or novel traits in dogs contributed to genetic variation, thus increasing the power to identify causal genes.
Advanced intercross lines (AIL)
AIL have been used to refine the mapping of genetic loci associated with obesity- and energy balance-related traits, as well as to identify new genetic variation. AIL offer increased mapping resolution of QTL through increased recombination resulting from random intercrosses over multiple generations of breeding(Reference Darvasi and Soller40). Initial AIL experiments using the LG/J by SM/J cross with continued interbreeding through generations F9 and F10 resulted in five times the mapping resolution of the original F2 population(Reference Fawcett, Jarvis and Roseman41). This F10 AIL identified ninety significant single trait QTL for phenotypes related to obesity such as increased size, adipose deposition and body weight, many of which were multiple narrow QTL with small effect size resolved from original large single peak QTL in F2/F3 populations. Such studies only further highlight the complexity of the genetic architecture of obesity, and the potentially enormous number of genes involved.
Additional AIL resources for fine-mapping complex traits such as obesity include heterogeneous outbred stocks of mice. Outbred stocks are derived from multiple founder inbred strains and numerous random matings, resulting in a population with closely spaced recombinants ( < 2 cM) and fine mapping resolution ( < 1 cM) for QTL of small effect(Reference Valdar, Solberg and Gauguier42, Reference Mott, Talbot and Turri43). GWAS performed in outbred stocks identified hundreds of QTL with effect sizes less than 2 %. Many of these QTL replicated existing QTL identified in crosses between inbred strains that had been included in the creation of the heterogeneous stock(Reference Valdar, Solberg and Gauguier42). In some instances, large-effect QTL were identified with mapping intervals encompassing three or fewer genes, uncovering significant targets for causal genes(Reference Valdar, Solberg and Gauguier42).
The mapping approaches discussed above have provided valuable information as to the general genetic architecture of obesity and related traits. Numerous genetic loci have been identified, emphasising the complex and polygenic nature of the genetic architecture of obesity, but specific target genes have yet to be determined. Although these approaches have been informative, several shortcomings have been recognised. For example, the lack of genetic diversity among traditional inbred laboratory mouse strains has left large regions of the genome unmapped, and has afforded little power to detect small effects and identifying QTL that are largely strain dependent(Reference Stylianou, Korstanje and Li21). AIL and outbred stocks have been established to increase mapping resolution and increase power to identify loci of small effect. However, the lack of genetic diversity still exists, and mapping power remains too low for identification of the underlying causative genetic variation(Reference Valdar, Solberg and Gauguier42, Reference Mott, Talbot and Turri43). To address these shortcomings, a unique and innovative mouse model designed to mimic human genetic diversity, the CC, has been created.
The Collaborative Cross, a next-generation mouse model
The CC is a large panel of recombinant inbred mouse lines derived from a genetically diverse set of eight founder strains created to advance discovery of the genetic underpinnings of complex traits, such as energy balance and body weight regulation (Fig. 2) (Reference Churchill, Airey and Allayee16, Reference Threadgill, Hunter and Williams17). The eight founder strains consist of five standard laboratory strains (C57BL A/J, C57BL/6 J, 129S1/SvImJ, NOD/LtJ and NZO/H1LtJ) and three wild-derived strains (WSB/EiJ, CAST/EiJ and PWK/PhJ), representing each of the subspecies of Mus musculus and capturing an unprecedented high degree of genetic diversity(Reference Aylor, Valdar and Mathes44, Reference Roberts, Pardo-Manuel de Villena and Wang45). Not only were the eight founder strains chosen specifically for their genetic diversity, but also to include genetic variants specifically linked to known complex disorders such as type 1 and 2 diabetes, obesity and insulin insensitivity (NOD/LtJ and NZO/H1LtJ)(Reference Morahan, Balmer and Monley46). This genetic diversity affords the CC increased mapping precision for complex traits, a distinct advantage over AIL and existing recombinant inbred lines(Reference Churchill, Airey and Allayee16, Reference Threadgill, Hunter and Williams17). It is expected that the mapping resolution of the fully inbred lines will be approximately 1 Mb, more precise than the finest resolution presented thus far(Reference Broman23, Reference Valdar, Solberg and Gauguier42, Reference Valdar, Flint and Mott47). Furthermore, the CC lines will be completely sequenced, and sequences for each line will be archived and available for shared use. Contrarily, each new individual in a specific cross or AIL is genetically unique and thus must be individually genotyped or sequenced, making experiments in these populations costly and labour intensive(Reference Churchill, Airey and Allayee16, Reference Threadgill, Hunter and Williams17, Reference Morahan, Balmer and Monley46). The genetic variability and eternal reproducibility of the CC makes it a superior resource for systems genetics research. This large reference panel of recombinant inbred strains will enable the evaluation of complex gene–environment and gene–gene interactions, providing a platform for the investigation into the genetic architecture underlying complex interactions among diverse phenotypes and molecular networks(Reference Churchill, Airey and Allayee16).
The Collaborative Cross (CC) is a large panel of recombinant inbred mouse lines designed to model human genetic diversity. Each individual line of the CC (one theoretical example depicted here) will represent a genetic mosaic of the eight founder strains. Derived from the crossbreeding of five classic inbred lines (C57BL A/J, C57BL/6 J, 129S1/SvImJ, NOD/ltJ and NZO/H1LtJ) and three wild-derived mouse lines (WSB/EiJ, CAST/EiJ and PWK/PhJ), the CC captures more than 90 % of the genetic diversity across the mouse genome. The genetic and phenotypic diversity of the eight founder strains captures genetic variants specifically linked to body size and known complex disorders such as type 1 and 2 diabetes, obesity and insulin insensitivity. Some images in this figure are courtesy of Fernando Pardo Manuel de Villena.
Initial analysis of genetic architecture confirmed that generations F5–F12 of the incipient CC lines (i.e. during the inbreeding process) encompassed the total genetic diversity of the eight founder strains with alleles from all eight strains dispersed among the genomes of each incipient CC line(Reference Aylor, Valdar and Mathes44). Furthermore, phenotypic analysis of the eight founder strains and 176 incipient CC lines revealed extreme diversity in obesity- and metabolism-related traits. For example, body fat percentages for the founder strains varied from 2 to 32 % for the A/J and NZO/H1LtJ strains, respectively(Reference Mathes, Aylor and Miller48). Likewise, body weights for the founder strains ranged from 13 g for WSB/EiJ to 48 g for NZO/H1LtJ. Variation in adiposity and body weight was mirrored in the incipient CC lines. Body fat percentages ranged from 1 to 32 %(Reference Mathes, Aylor and Miller48). Body weights for the incipient CC lines were slightly less varied, spanning from 15 to 35 g. However, several of the traits measured in the incipient CC lines surpassed the phenotypic variation observed in the founder strains. For instance, average wheel-running distances on days 11 and 12 of wheel access ranged from 0 to 13 km in the founders and from 0 to 19 km in the incipient CC lines(Reference Mathes, Aylor and Miller48). Moreover, body weight change in response to wheel running varied from a 6 g loss to a 3 g gain in the founder strains and from a 7 g loss to an 8 g gain in the incipient CC lines(Reference Mathes, Aylor and Miller48). These findings demonstrate that the genetic diversity highlighted in the eight founder strains was captured in the incipient CC lines resulting in equivalent or increased phenotypic diversity for complex obesity-related traits(Reference Mathes, Aylor and Miller48).
Mapping experiments in the incipient CC lines as a proxy for the power of the CC resource to uncover the genetic architecture of complex traits revealed significant novel QTL for wheel-running traits and replicated previously described QTL for body weight-, body fat- and metabolism-related traits (Aylor et al. (Reference Aylor, Valdar and Mathes44); WF Mathes, DL Aylor, DR Miller, GA Churchill, EJ Chesler, F Pardo-Manuel de Villena, DW Threadgill and D Pomp, in press). A significant body weight QTL was identified on proximal chromosome 4, Bwq14, which replicated existing QTL for body weight identified using the KK strain, a strain phenotypically similar to NZO(Reference Suto49, Reference Suto, Matsuura and Imamura50). Expression QTL analysis for Bwq14 revealed twenty-one genes with significant local expression QTL. Using allele effect estimates and existing SNP data sources for the eight founder strains, the genes underlying Bwq14 could be narrowed to a single positional candidate, aspartate-β-hydroxylase, Asph (Reference Aylor, Valdar and Mathes44).
The genetic and phenotypic diversity that has manifested in the incipient CC lines provides strong evidence that the CC, once fully inbred, will be a powerful tool for complex trait analysis. Early success in mapping QTL for body weight in the incipient CC lines confirms that the CC will aid in the discovery of the genetic variables that predispose individuals to obesity and related traits. The advantages of the CC over existing resources are clear. The CC will offer increased genetic and phenotypic diversity that will translate into the discovery of QTL within regions that were previously ‘blind’ to investigation. The elaborate mating scheme used to create the CC will eliminate strain-specific bias and reduce linkage disequilibrium, allowing for mapping resolution precise enough to identify specific genes that underlie complex traits. Moreover, the CC will be an invaluable tool for the examination of gene–environment interactions that contribute to energy balance- and obesity-related phenotypes. Taken together, the CC will inform human health by elucidating biological networks involved in the development of obesity as well as providing predictive models from which one can determine the variables that may contribute to the aetiology of obesity-related diseases on an individual basis.
Novel phenotypes: exercise
As described above, genetic architecture unquestionably plays a major role in weight dysregulation and traits related to obesity. However, behaviours such as food consumption and exercise are also crucial in maintaining a balance between energy intake and expenditure, and ultimately constancy of body weight. In addition to weight loss and reductions in adiposity, physical activity has also been associated with reductions in TAG and LDL levels, increased HDL and enhanced insulin sensitivity(51, 52). Despite the growing body of evidence of positive health benefits, especially with regard to weight reduction(Reference Church and Blair53), there remains considerable variation in the amount that children and adults voluntarily exercise(51, Reference Duke, Huhman and Heitzler54). While a voluntary behaviour, the predisposition to engage in exercise is heritable(Reference Lightfoot, Turner and Daves55), and thus, the genes that underpin physical activity traits are integral to our understanding of the genetic architecture of obesity. As such, multiple mapping studies have been directed at understanding the genetic architecture of physical activity-related traits and how they may have an impact on body weight and obesity. Moreover, as changes resulting from exercise remain considerably variable within populations, especially with regard to weight and adiposity, studies are now aiming to uncover the genetic architecture regulating the change in weight and adiposity in response to exercise (i.e. Mitchell et al. (Reference Mitchell, Church and Rankinen56)).
Initial explorations into the genetic architecture of exercise-related traits have been conducted in second-generation intercross (F2), backcross, and advanced intercross (G4) mouse populations. Using F2 crosses between low active (wheel running) C3H/HeJ and high active C57L/J have revealed multiple QTL for running duration, distance and speed(Reference Lightfoot, Turner and Pomp57, Reference Leamy, Pomp and Lightfoot58). Moreover, analysis of body weight changes in response to exercise in this population identified additional QTL for energy balance(Reference Leamy, Pomp and Lightfoot59) and loci that were pleiotropic for both exercise- and body weight-related traits(Reference Leamy, Pomp and Lightfoot60). QTL for weight change in response to exercise did not co-localise with QTL for running traits or previously described QTL for body weight.
Nehrenberg et al. (Reference Nehrenberg, Wang and Hannon61) utilised a backcross between mice selectively bred for high voluntary wheel running (high running (HR) line) and the inbred strain C57BL6/J to identify novel exercise QTL as well as loci that were sex-specific. The HR line originated from a long-term replicated artificial selection experiment for high voluntary wheel-running behaviour on days 5 and 6 of a 6 d wheel exposure (for a review, see Swallow et al. (Reference Swallow, Hayes, Koteja, Garland and Rose62)). By generation 16, and continuing through generation 50 and beyond, the HR lines (four replicates) had diverged from the control lines (C lines; four replicates) with an approximate 2·5- to 3·0-fold increase in total revolutions per d. As demonstrated by the use of the HR line in Nehrenberg et al. (Reference Nehrenberg, Wang and Hannon61), selective breeding has not only facilitated the uncovering of the genetic architecture of weight-related phenotypes, but has aided in the search for genetic loci associated with exercise phenotypes as well.
Using a different HR replicate (out of the possible four), Kelly et al. (Reference Kelly, Nehrenberg and Pierce63) created an advanced intercross line population (G4) and used it to identify forty-one significant and twenty suggestive QTL for running traits (distance, duration and speed) and body mass. Although several of these loci are close in proximity to physical activity QTL identified in F2 and backcross populations, they do not overlap, suggesting that they may be independent loci. However, the strains used to create each of these populations differ in their phenotypic expression for physical activity- and body weight-related traits, making a direct comparison between the two populations difficult. Each of the QTL identified in the AIL explained only a small percentage of the phenotypic variations observed in this population, with the largest effect accounting for only 6·6 %(Reference Kelly, Nehrenberg and Hua64). In addition to daily running traits, Kelly et al. (Reference Kelly, Nehrenberg and Pierce63) identified seven significant and five suggestive QTL representing the slope and intercept of a linear regression across all 6 d of running, some representing a combination of the daily traits. The QTL associated with the trajectory of running across multiple days of wheel exposure were the first of their kind and may provide valuable information with regard to the continuation of exercise programmes.
An additional strategy recently employed by Lightfoot et al. (Reference Lightfoot, Leamy and Pomp65) utilises HAM in thirty-eight inbred mouse strains. In addition to discovering new and refining previously mapped QTL for wheel running, the use of HAM across a diverse set of genetic backgrounds allowed the expansion of genomic coverage and the ability to more confidently generalise the results. In the study by Lightfoot et al. (Reference Lightfoot, Leamy and Pomp65), 448 mice from thirty-eight inbred strains were phenotyped and genotyped for physical activity-related traits such as running distance, speed and time. Results identified twelve significant QTL for running distance and speed, some with sex-specific effects. QTL identified in this experiment did not replicate existing QTL; however, existing QTL for physical activity traits are based solely on crosses using C3H/HeJ, C57L/J and HR lines(Reference Lightfoot, Leamy and Pomp65). The three strongest QTL identified in this HAM study had relatively narrow confidence intervals, spanning genomic regions with few genes. In fact, only eight predicted genes and one annotated gene fell within the confidence intervals of these peaks(Reference Lightfoot, Leamy and Pomp65).
Collectively, the findings discussed above have just begun to uncover the genetic architecture of physical activity and how it may relate to obesity and its related traits. Further investigation is warranted to understand the complex relationship among these traits and the genetic underpinnings of the said relationship.
Host genetic control over composition of the gut microbiome
The primary cause of obesity is an imbalance between energy intake and expenditure, with traits related to energy balance being exceptionally complex, simultaneously affected by genetic architecture and interacting environmental factors. Nonetheless, as we have described, mouse models have been, and continue to be, fruitful in the elucidation of the underlying genetic architecture of obesity. A vast majority of studies, only a fraction of which are discussed above, attempting to identify genetic susceptibility of obesity have focused on the penultimate outcome trait (i.e. weight and/or adiposity), with fewer studies examining traits associated with the behavioural and physiological mechanisms contributing to obesity (for example, food choice, intake and assimilation efficiency; RMR; exercise propensity). As described above, literature is now emerging that attempts to uncover the genetic architecture of behaviours that negatively affect energy balance, specifically exercise. Here we present an additional new paradigm for the study of obesity, highlighting a recent example(Reference Benson, Kelly and Legge66), namely the analysis of host genes that influence composition of the gut microbiome, a climax population of microbial species that enter into intimate metabolic and immune interactions with host gastrointestinal tissues and potentially affect many nutritionally relevant traits and diseases.
Recent literature examining the role of the gut microbial ecology may now enlighten the view that obesity simply results from an excess of energy intake v. energy expenditure (for a review, see Turnbaugh & Gordon(Reference Turnbaugh and Gordon67)). This emerging hypothesis predicts that lean and obese individuals possess unique gut microbial populations that may in turn affect energy extraction and subsequent deposition of fat stores from food sources. Thus, the impact of the gut microbial community is most relevant for the proportion of intake that can be metabolised. In other words, even when energy intake and expenditure are held constant, variability in energy metabolism and fat deposition will vary depending on the composition of the gut microbes.
In mammals, the colonisation and establishment of micro-organisms in the gastrointestinal tract begin at birth and stabilise around the time of weaning(Reference Tannock68). This dense microbial population, up to 100 trillion organisms, is collectively known as the gut microbiota(Reference Turnbaugh and Gordon67). The gut microbiota is populated by thousands of bacterial species belonging to a small number of phyla, and varies dramatically among communities(Reference Kuczynski, Costello and Nemergut69) and individuals within a single population(Reference Ley, Peterson and Gordon70). Contributing to the individuality of the gut microbiota are differences in the relative ratios of dominant phyla and variation in genera and species found in an individual host(Reference Ley, Peterson and Gordon70).
The composition of the gut microbiota is undoubtedly shaped by environmental factors, highly resistant to change once established, and relevant to our understanding of human diseases such as obesity, CHD(Reference Fava, Lovegrove and Gitau71), diabetes(Reference Wen, Ley and Volchkov72) and digestive maladies(Reference Frank, St Amand and Feldman73). However, few studies have directly tested, in the absence of confounding genetic diversity (in human populations) and strong environmental effects, the hypothesis that host genetic architecture may also play a role in the gut microbiota composition(Reference Turnbaugh and Gordon67, Reference Deloris Alexander, Orcutt and Henry74, Reference Khachatryan, Ktsoyan and Manukyan75).
In an attempt to control the strong environmental effects exerted over the gut microbiota, several studies have tested the impact of the host genotype on the gut microbiota utilising comparisons of genetically distinct mouse inbred strains. Friswell et al. (Reference Friswell, Gika and Stratford76) performed uterine transplantations in order to eliminate any non-genetic maternal effects, and found that mice with different genotypes that were born together had similar microbiota. In another comparison, Loh et al. (Reference Loh, Brodziak and Blaut77) used germ-free inbred strains (C57BL/10, C3H, BALB/c), inoculated individuals from a single mouse donor, and compared the strains over a 13-week period. Between 4 and 8 weeks post-inoculation, cluster analyses revealed that C3H and BALB/c mice were similar and distinct from C57BL/10. But at 8 weeks post-inoculation, environmental effects (housing in different cages) appeared to play a greater role relative to host genetic control on intestinal microbiota composition.
An alternative approach to the comparison of inbred strains is a candidate gene approach. This approach, primarily driven by transgenic mouse studies, aims to identify single genes that strongly influence the gut microbiota. Thus far, identified candidate genes have been primarily related to immune system function(Reference Vijay-Kumar, Aitken and Carvalho78) and metabolism(Reference Jones, Begley and Hill79). For example, the obese phenotype (ob/ob) of leptin-deficient mice, compared with lean mice, has been shown to be associated with a 50 % reduction in the abundance of Bacteroidetes while proportionally increasing the gut population of Firmicutes(Reference Ley, Backhed and Turnbaugh80). Subsequently, Turnbaugh et al. (Reference Turnbaugh, Ley and Mahowald81) demonstrated that this shift in relative abundance of Bacteroidetes and Firmicutes, among obese individuals, resulted in greater capacity for energy extraction from the diet.
Benson et al. (Reference Benson, Kelly and Legge66) have most directly tested the assumption of host genetic control of the gut microbiome using a sophisticated murine advanced intercross line model, in which the genetic background was evaluated and environmental factors controlled. This large G4 population (n 645) was originally derived from an intercross between C57BL/6J and an ICR-derived outbred line (HR), and designed to uncover the genetic architecture of voluntary exercise while providing a high level of mapping resolution (as discussed above, see the Novel phenotypes: exercise section). DNA was extracted from faecal samples, and using quantitative pyrosequencing, sixty-four conserved taxonomic groups, which varied across animals in the population, were defined. While a proportion of the variation in the abundance of taxonomic groups could be explained by environmental variables such as litter and cohort, Benson et al. (Reference Benson, Kelly and Legge66) found a substantial contribution from the host genetic architecture. Using 530 fully informative SNP markers, QTL analyses revealed eighteen loci that show significant or suggestive genome-wide linkage with relative abundances of specific microbial taxa (Fig. 3). Some loci controlled individual microbial species, some controlled groups of related taxa, and some had pleiotropic effects on groups of distantly related organisms. These findings clearly establish host genetics as a factor in determining composition of the gut microbiome.
A summary of host murine quantitative trait loci (QTL) controlling gut microbiome composition as described in Benson et al. (Reference Benson, Kelly and Legge66). ■, SNP (n 530) used for QTL mapping, with corresponding chromosomal positions. QTL confidence intervals are shaded in colours corresponding to major phyla: 
, Actinobacteria; 
, Proteobacteria; 
, Firmicutes; 
, Bacteriodetes.
Benson et al. (Reference Benson, Kelly and Legge66) have for the first time presented direct evidence for environmental and host genetic loci that control variability in the abundances of different taxa in the mouse gut microbiome, and have therefore hypothesised that the gut microbiota composition as a whole can be understood as a complex, polygenic trait. Additionally, the genetic control identified by Benson et al. (Reference Benson, Kelly and Legge66) appears to encompass, for example, host genetic factors such as those influencing mucosal immunity. Consequently, host genetic loci that affect composition of the gut microbiome are likely to partially contribute to an individual's overall predisposition to obesity and other nutritionally relevant traits and diseases. How changes in nutrition may influence this host–microbiome relationship, and thus make an impact on weight regulation, remains an interesting yet untested question.
Summary: the genetic architecture of obesity
The many mapping strategies presented within the present paper have outlined significant progress in uncovering the genetic architecture of obesity and related traits. Initial studies examining obesity and related traits used simple F2 crosses of inbred strains to parse out regions of the genome associated with these traits. The need for finer mapping resolution and greater capacity to detect small effects drove the development of AIL, outbred crosses, and diverse populations such as the HMDP to investigate the genetic architecture of complex traits. From these studies it became clear that obesity is a complex trait with genetic effects originating from literally hundreds of small-effect QTL across the genome. In addition, it was recognised that many environmental factors, and interactions between genes and the environment, contribute to the development of obesity and its related traits. The CC is a next-generation mouse model designed to maximise genetic and phenotypic diversity, refine mapping resolution, and enable the detection of small effects. Early evaluation of the CC demonstrates that it has the potential to be a remarkable tool for complex trait analysis and could potentiate significant discoveries within the fields of genetics and systems biology. The completion of the CC in 2012 may afford unprecedented mapping precision to aid in the search for causal genes for obesity and its related traits. It is important to note that novel phenotypes such as physical activity and wheel running in mice can contribute to the development of obesity. Likewise, genetic sequences of host and gut microbiota may influence energy utilisation contributing to obesity and overweight. Understanding the complex interplay among these traits will provide insight into the intricate network of phenotypic, genotypic and environmental variables that predispose individuals to obesity and its related traits.
Acknowledgements
Sections of the present review were supported by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) (grants no. RC1DK087346 and no. DK076050). Some phenotypes were collected using the Animal Metabolism Phenotyping core facility at the University of North Carolina's Nutrition Obesity Research Center (funded by NIDDK grant no. DK056350). W. F. M. was supported by National Institutes of Health grant no. K12 HD01441. S. A. K. was supported through a National Institute of Mental Health-funded (no. 5T32-MH-075854-04) Interdisciplinary Obesity Training (IDOT) programme and from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) (grants no. RC1DK087346 and no. DK076050).
W. F. M., S. A. K. and D. P. all contributed to the writing of the present review.
There are no conflicts of interest.
