Introduction

At the moment, dietary guidelines are based on nontoxic daily allowance for the general population, and only cluster the population into a few subgroups, such as children or pregnant women, to be given specific or additional dietary recommendations. With increasing knowledge of gene-diet interactions for macronutrients and micronutrients, it will be possible to tailor dietary recommendations for specific subgroups based on their genetic makeup. Today, we already use diets tailored to lessen severe diseases; for example, a phenylalanine-free diet is used for phenylketonuria, which is caused by mutations in the phenylalanine hydroxylase gene leading to higher concentration of phenylalanine in the blood, neurotoxicity, and subsequently mental retardation [1].

The possibility of using nutrigenetics to optimize health and prevent disease also applies to the dietary management of obesity. Obesity prevalence has increased epidemically during the past decades, and in 2008, half a billion adults could be categorized as obese (body mass index [BMI] ≥30 kg/m2) [2]. There are several possible contributors to the obesity epidemic, but the main causes are an increase in the availability of energy-dense foods and an increase in sedentary activities, whereas everyday physical activity has declined [3]. Despite these changes in the common obesity-promoting environment, a large proportion of individuals remain lean. This suggests that nonenvironmental factors contribute substantially to the individual obesity susceptibility.

Several theories have been put forward to explain our current obesity epidemic [4••]. One of the suggested models is the dual intervention point model that determines our susceptibility by a lower and higher body weight set point between which we will move determined on environmental pressure, whereas the upper limit is dependent on our genetic profile [4••]. This hypothesis is in line with the drifty genotype—a reform of the thrifty genotype—that entails that the genetic variation predisposing to obesity has at one point been an evolutionary advantage but without the predator pressure on the upper limit for humans allowing for genetic drift for the upper weight limits [5].

Heritability studies have shown that monozygotic twin pairs are more concordant than dizygotic twin pairs for obesity traits and adoptees are more similar to their biological parents than their adopted parents for obesity degree [6, 7]. These studies and several other studies of obesity heritability supports a significant heritable component estimated to be relatively high by twin studies and relatively low by family studies, resulting in a general estimate of genetic variation accounting for 40–70 % of the variance in obesity [4••]. Thus, the current environment explains the high worldwide prevalence of obesity, but genetic variation explains the interindividual differences in obesity susceptibility in the obesogenic environment. Equally, twin studies of underfeeding and overfeeding have shown that the interpair variance is much larger than the intrapair variance, indicating that weight loss and weight regain are both highly influenced by genetic variation [8].

Genetics has moved from linkage studies in large families, assuming cosegregation with the disease, and years spent on fine mapping relevant candidate genes, to candidate gene studies, examining one gene at the time, often only the coding regions, for genetic variation, and on to the hypothesis-free genome-wide association studies (GWAS) testing millions of single nucleotide polymorphisms (SNPs) at the same time. The use of GWAS does not require a link between biological function and the gene, but merely tests SNPs identified by the sequencing of the human genome sequence and construction of the human HapMap [9–11], for frequency among cases and controls. Despite these advances in methods, very few genetic loci or SNPs have been consistently associated with obesity [12, 13]. This lack of accounting for the estimated heritability by genetic variation in the DNA sequence often is referred to as the “missing heritability.” The missing heritability could be due to private relatively rare mutations, copy number variants (CNVs), epigenetic factors, which affect gene transcription without changing the DNA sequence [14], interactions between genes, or interactions between gene and environment [15].

Genetics of Obesity

The three main forms of obesity attributed to genetics are: syndromic obesity, nonsyndromic or monogenic obesity, and common or polygenic obesity. Syndromic obesity is the combination of extreme obesity, physical dysmorphology, and intellectual disabilities and includes syndromes such as Prader-Willi (15q11.2), Bardet-Biedl (BBS2-15), and Alstrom (2p13.1) [16].

Monogenic obesity often is caused by mutations in genes in the leptin-melanocortin signaling pathway or genes expressed in hypothalamic nuclei [17]. Among these monogenic forms of obesity, the most common is due to mutations that reduce the activity or expression of the melanocortin 4 receptor (MC4R) [17], with a frequency of 2–6 mutated alleles per 100 persons with obesity [18–20]. The mutations in the MC4R have been shown to be directly coupled with the molecular function of the receptor in vitro as there is an allele dose effect on ad libitum energy intake [18]. Despite a clear association with obesity, most probably caused by haploinsufficiency, it has been observed that the penetrance of the obesity is dependent on age [20] and for some heterozygous carriers of functional mutations there is still an influence of environmental factors [21]. Other less frequent monogenic forms of obesity include mutations in the leptin (LEP), leptin receptor (LEPR), proopiomelanocortin (POMC), prohormone convertase (subtilisin/kexin-type) 1 (PCSK1), single-minded homologue 1 (SIM1), brain-derived neurotrophic factor (BDNF), neurotrophic tyrosine kinase receptor type 2 (NTRK2), and SH2B adaptor protein 1 (SH2B1) [16].

Treatment opportunities for monogenic obesity, apart from the treatment of leptin-deficient children with recombinant leptin [22], are limited, and only a few studies have examined the response of individuals with monogenic obesity to diet with/without physical activity. Weight loss either with diet alone or with the combination of diet and physical activity was similar for obese individuals with POMC or MC4R and for individuals without mutations in these genes [21, 23, 24]. However, children who had MC4R mutations were less successful in maintaining their weight loss compared with children who did not have MC4R mutations [24]. In addition to dietary management of monogenic obesity, bariatric surgery is a treatment option. Specifically for monogenic obesity in which uncontrolled hyperphagia often is a significant phenotypic manifestation, it is critical to determine the response of mutation carriers to treatment as it is expected that approximately a 5 % fraction of morbidly obese individuals undergoing bariatric surgery are carriers of MC4R mutations. One case study has suggested that complete MC4R deficiency might influence long-term weight loss after adjustable gastric banding [25], whereas MC4R haploinsufficiency, despite large effects on obesity, does not limit weight loss after Roux-en-Y gastric bypass (RYBG), compared with individuals who do not have MC4R mutations [26].

Besides the monogenic obesity cases that were discovered on the basis of mouse models of obesity and the exploration of the leptin-melanocortin pathway [27], it has been hypothesized that since obesity is a common condition, the genetic variation might also be common variants (polygenic obesity). As such polygenic obesity is assumed to be associated with common variants that have small additive or nonadditive effect sizes without a Mendelian pattern of inheritance. Based on the thrifty/drifty gene hypothesis, it is not assumed that all obese individuals have the same combination of SNPs, but rather that several genetic changes contributed to evolutionary advances [4••].

Genome-wide linkage and candidate gene studies have identified 52 genomic regions associated with obesity, with a minimum of two replications [12]. The two approaches have been largely abandoned after obesity-related SNPs were identified by GWAS. These GWAS have led to the identification of novel obesity-associated genes, such as FTO [28, 29], which would not necessarily have been identified based on its biological function but also has identified loci, such as the MC4R, that have been previously associated with obesity [30]. The FTO and MC4R SNPs increase BMI with 0.4 kg/m2/allele and 0.25 kg/m2/allele, respectively, and except from TMEM18 [31, 32], they are the two genes that confer the highest increase in BMI [33]. In total, 32 loci have been identified from current GWAS that are robustly associated with obesity, ranging in effect size from 0.05-0.40 BMI units per risk-allele for KCTD15 and FTO, respectively [13, 34, 35].

Dietary Intake and Genetic Predisposition

A change in taste preference or food intake to energy-dense palatable foods, such as foods rich in fat or sugar, could be one of the mechanisms by which common SNPs increase susceptibility to obesity. In fact, not only do some forms of dietary intake show a moderate heritability but it also has been shown that a significant proportion of the heritability is shared with BMI [36, 37], suggesting that taste preferences or food preferences might be genetically driven and impact the development of obesity.

One of the genes that have been investigated for associations with energy intake is the MC4R that in addition to predispose to monogenic obesity, has several SNPs in the proximity of the locus that are associated with common obesity [30, 38]. The MC4R rs2229616 is associated with an increased total carbohydrate intake (57 g/day) [39], and the MC4R rs17782313 is associated with increased intake of protein (4.4 g protein/day) in women but not with carbohydrate intake [40], whereas a study of 756 healthy adult twin pairs did not identify significant associations between MC4R SNPs and macronutrient intake [41].

Additionally, several studies have investigated whether the strongest GWAS hit for obesity is associated with food preference and energy intake. FTO rs9930609 has been associated with increased fat intake [42] as well as total energy intake (independently of body weight) in children [43]. Likewise in adults, FTO rs9939609 has been associated with an increased intake of fat [44] and a higher intake of dietary protein and energy from fat and lower carbohydrate intake [45]. Similarly, the FTO rs8050136, which is in complete linkage disequilibrium (LD) with rs9939609, has been associated with a higher habitual dietary intake of protein [46], and FTO rs1421085, which is in LD with rs9939609 (r2 = 0.97), with a higher energy percent from fat (0.5E%/allele) [47•]. On the contrary, several studies have not identified any association between FTO SNPs and macronutrient intake [41, 48–50]. The fact that some studies do not identify associations between FTO and macronutrient intake could reflect a requirement for large samples sizes due to small effect sizes. This is highlighted by a recent genome-wide association meta-analysis that investigated associations between SNPs and macronutrient intake in 37,537 individuals, with replication in two additional cohorts of 7,724 and 33,533 individuals, which confirmed the association of FTO rs1421085 (in high LD with rs9939609, r2 > 0.97) with a 0.1 % higher protein intake [51•].

In addition to the FTO and MC4R, a few other obesity-associated genes have been associated with dietary macronutrient intake. SH2B1 rs7498665 is associated with increased intake of fat (1.1 g/d/G-allele) and saturated fatty acids (SFA, 0.6 g/d/G-allele) [52]. BDNF rs6265 is associated with an increased energy intake of 100 kcal/day/allele and TNNI3K rs1514176 was associated with a lower percentage intake of dietary protein of 0.3E%/allele [47•]. Additionally, there are some indication that rs35874116 in one of the sweet taste receptors TAS1R2 is associated with less carbohydrate intake, including sugars and fibers [53].

The studies of food preference and dietary intake suggest that previously identified obesity-associated GWAS hits could account for some of the shared heritability between obesity and dietary intake but also indicate that additional genetic variation related to taste perception could contribute to this shared heritability.

Genetic Variation and Dietary Management

In addition to food preference, several physiological mechanisms are involved in determining energy balance, including appetite, absorption, metabolism, and energy expenditure. Previous candidate gene studies have identified several genes involved in energy expenditure, appetite control, and lipid metabolism to influence the response to weight loss, with and without interaction with diet or lifestyle [54–56]. Of these candidate genes, the PPARG Pro12Ala (rs1801282) has been thoroughly investigated [57, 58]. In the Diabetes Prevention Program (DPP) that includes 2,234 individuals with overweight or obesity and impaired glucose tolerance (IGT) rs1801282 was associated with weight loss after 6 months (−0.63 kg/allele) and after 2 years (−0.93 kg/allele) [59], and in 1,465 individuals on a Mediterranean diet, PPARG rs1801282 interacted with dietary fat in determining obesity and weight loss [60]. Similarly to the PPARG, the apolipoprotein A-II (APOA2) rs5082 interacts with diet to modify weight loss success, which has been shown consistently in 6 independent studies including more than 10,000 individuals [61, 62].

In the post-GWAS era, gene-diet interaction analyses have mostly focused on selecting the SNPs that are associated with obesity in GWAS to explore these in a single SNP fashion or by a modified candidate gene approach, including SNPs identified from GWAS and/or additional SNPs from relevant biological candidate genes, trying to select for gene-diet interactions with a moderate effect size.

For the single gene/SNP studies, it has been most consistently shown that the effect of FTO on obesity can be attenuated by physical activity [63]. In addition to an interaction with physical activity, it also has been shown that high intakes of saturated fatty acids (SFA) interact with FTO rs9930609 to accentuate the effect on obesity [64, 65], and likewise to interact with energy-adjusted fat intake and carbohydrate intake to determine obesity [45]. In the latter study, it was also shown that low leisure time activity accentuated the association of FTO rs9930609 with obesity. Additionally, it has been shown that FTO rs9939609 is associated with better weight loss on a Mediterranean diet over 3 years [66] and a better weight maintenance after weight loss over 40 weeks [67]. However, several studies have not identified any gene-diet interaction effect on weight management of the FTO SNPs [50, 68–74]. Collectively, the association between FTO SNPs and obesity can be accentuated by an increased preference for energy-dense foods and interaction between FTO and these nutrients, and attenuated by physical activity.

One interesting notion, which could be related to the non-consistency of findings in the FTO-diet interaction studies, is that FTO rs9939609 has been reported to be associated with underreporting [45, 70]. This underreporting could potentially reduce the probability of identifying interactions between FTO and energy-dense foods as this association might reflect a specific underreporting of energy dense and likely “unhealthy” food items [75].

Together with the FTO and MC4R, the transcription factor 7-like 2 (TCF7L2), which is the gene that has been most consistently associated with type 2 diabetes and confers the highest risk [76], has been investigated for the effect on weight management. TCF7L2 HapA (rs7903146 and rs10885406) interacts with dietary protein to attenuate weight change [77], and TCF7L2 rs7903146 is associated with a smaller 10-week weight loss (2.8 kg) on a high fat compared to a low fat diet [78]. In addition to these GWAS hits, there are additional single SNP studies that show gene-diet interaction in regulating weight, including TFAP2B [79] and GIPR [80].

In the two EU-funded Pan-European projects Nutrient-gene interaction in obesity (NUGENOB) and Diet and Genes (DIOGENES), we have explored gene-diet interactions in relation to weight loss and weight regain [81, 82]. In NUGENOB, we examined 42 SNPs in 26 genes for an interaction with fat content of a 10-week hypocaloric diet on weight loss in 648 participants but found no interactions [82]. In DIOGENES, we used a replication strategy to explore associations of the same SNPs in a large observational study and in a smaller but well-controlled dietary intervention study with detailed phenotypic characterization. In the observational study that included 5,507 cases and controls, one significant diet-SNP interaction was identified among 123 SNPs in 15 hypothalamic genes [83]. The interaction between Neuromedin B (NMB) rs7180849 and glycemic index (GI) increase weight gain 25 g/year per allele per GI unit. In the intervention study, we selected 69 nutrient-sensitive candidate genes and included a total of 651 tagSNP selected to cover the genetic variance of the included genes/loci (±5 kb) with LD (r2) = 0.7–0.8 and a minor allele frequency above 5 %. We examined the interaction between these 651 tagSNPs and either dietary protein or GI on 6-month maintenance of weight, waist, or fat mass after a minimum of 8 % weight loss obtained in 8 weeks on a low-calorie diet (LCD) in 742 individuals [81]. After correction for multiple testing, none of the SNP-diet interaction effects were significant. The inclusion of both an observational and an intervention study to explore gene-diet interactions in DIOGENES was done to replicate initial findings from the observational study in the intervention study on weight regain; however, NMB rs7180849 did not interact to increase weight regain over 6 months in the intervention study [81].

In the MONICA/KORA study, 7 GWAS SNPs (TMEM18, NEGR1, MTCH2, FTO, MC4R, SH2B1 and KCTD15) were investigated in 12,462 individuals, but for none of the SNPs a significant modification by carbohydrate and fat dietary intake was observed on BMI [84]. In a study of 401 German children and adolescents with obesity, it was observed that out of 10 GWAS SNPs associated with childhood obesity, serologically defined colon cancer antigen 8 (SDCCAG8) rs10926984, rs12145833 and rs2783963 were associated with reduced weight loss during the 1-year lifestyle intervention program [85]. However, attempts to replicate the findings in the above-mentioned NUGENOB study with 648 adults on a 10-week hypocaloric diet failed [85]. Similarly there was a lack of replication of interactions for 38 genes and dietary intake between two populations of 1,173 African Americans (1,086 SNPs) and 1,165 Caucasians (897 SNPs) [86].

Combining all 32 GWAS hits for obesity by allele loading showed that having a high number of obesity-associated alleles increased BMI more the higher the intake of sugar-sweetened beverages, while no such association was found with artificially-sweetened beverages [87••]. Recently, a GWAS for dietary fat intake among 598 adolescents identified that the opioid receptor μ 1 (OPRM1) rs2281617 was associated with a lower fat intake and lower BMI and with a higher amygdala volume, indicating that the OPRM1 can influence reward behaviors through regulation of amygdala volume [88]. This study suggests that not only hypothalamic genes are involved in the regulation of energy intake, but that gene-diet interactions with genes in the mesolimbic system could be relevant to investigate in relation to dietary management in obesity.

The above-mentioned gene-diet interaction studies, although several of them are designed to investigate gene-diet interactions in weight management, are retrospective. Studies that assign dietary management based on genetic profile are scarce. One published study uses the genetic profile to adjust dietary and physical activity advice based on genotypes for 24 SNPs in 19 genes (from a DTC-genetic profiling company) [89]. The study found that genetic profile-based advice increased weight loss success, compared to following standard dietary advice (73 % vs. 32 %). Likewise, reports of a similar study with 5 SNPs in 4 genes increased weight loss success over 6 weeks [90]. However, both of the studies were small, including approximately 50 individuals in the genetic profile group, the participants were not all obese and given that one used 24 SNPs and the other 5 SNPs, it would seem that there is no consistency between them. In addition, all SNPs from GWAS or candidate gene studies with suggestive evidence of an interaction with diet on weight changes were not included [89, 90].

Conclusions and Perspectives

Despite the fact that the currently published papers on genetic predisposition to obesity only explain very little of the estimated heritability and that only suggestive gene-diet interactions have been identified, several DTC genetic profiling companies offer predictive tests and advice on diet and physical activity for optimal health based solely on the genetic profile [91]. Currently, we cannot predict whether an individual will be obese based on the information from GWAS hits nor does this genetic information add to predictions based on asking simple questions, such as parent or childhood obesity [92]. The only exceptions are monogenic obesity cases where we can predict obesity with a high sensitivity and specificity from rare variants. Likewise, we cannot predict the response of an individual to a particular dietary intervention based on the genetic profile or use the genetic profile of an individual to give personalized nutrition advice on how to lose weight or remain weight stable.

Before we can achieve the ultimate goal of providing dietary recommendations for optimal health to subgroups based on nutrigenetics, we need to be able to show consistent associations between genetic variation and response to dietary components. Presently, the studies are fairly small, even for observational studies, and there is lack of replication between studies. Additionally, current studies are focused on macronutrients [51•] and only a few on the quality of the macronutrients [83] or a specific food product [87••], but the focus might as well be on dietary patterns, micronutrients, or the source and quality of the nutrient. To dissect these contributions to dietary management in obesity, more sensitive tools need to be applied to record dietary intake, other measurements of obesity than the crude use of BMI might be necessary and integration of omics data will be essential [93••]. Likewise, the genetic contribution might be viewed as more complex than just the common disease–common variant approach, used in genetic linkage studies, hypothesis-based biological candidate studies and hypothesis-free genome-wide association studies, which have not offered an explanation of the missing heritability. Current strategies include identification and investigation of mutations with lower frequency or CNVs by sequencing either of the entire genome or of the exome [94, 95•, 96]. Additionally, epigenetic changes [14] and possibly telomere length [97] need to be considered in relation to genetic predisposition to obesity.

In conclusion, genetic predisposition in dietary management is still in its infancy, but further exploration of the genetic architecture and the integration with other omics data will allow for dietary recommendation to additional subgroups defined by genetic profiles to optimize health and prevent disease.