Distinct Penetrance of Obesity-Associated Susceptibility Alleles in the Hungarian General and Roma Populations

Abstract

Aims

The aim of our study was to explore differences in genetic predisposition to obesity between the Hungarian general and Roma populations.

Methods

A total of 1,152 samples from the Hungarian Roma population and 1,743 samples from the Hungarian general population were genotyped for 20 single nucleotide polymorphisms (SNPs) associated with the risk of obesity. Two types of multilocus genetic risk scores were constructed to estimate the combined effect of selected SNPs.

Results

Risk allele frequencies differed significantly between the two populations for 11 SNPs, with no enrichment in any of the two study groups. Variants (rs1558902, rs1121980, rs9939609, and rs9941349) in the fat mass and obesity-associated (FTO) gene exhibited strong but ethnicity-independent association with obesity. Genetic risk scores showed stronger associations with obesity in the Roma population compared with the Hungarian general population; however, without significant gene-population interaction.

Conclusion

Differences in obesity prevalence between the Hungarian general and Hungarian Roma populations could not be explained by their distinct genetic susceptibility, rather by ethnicity-related environmental and behavioral factors. Nonetheless, particular gene-environment interactions might contribute to the distinct penetrance of the obesity-associated genetic factors in populations of different ethnic backgrounds.

Introduction

Obesity, one of the strongest cardiovascular risk factors, is a serious public health challenge for the 21st century. Recently, a health examination survey revealed 26.2% and 30.4% prevalence of obesity in the Hungarian adult male and female populations, respectively [1]. These values ranked Hungary as the number one “most obese member state” in the European Union [2], which clearly indicates the need of collecting more information and conducting studies about the potential factors influencing obesity on the Hungarian population.

The population of Hungary is reasonably diverse with many ethnic minorities, among them the Roma (Gypsy) who account for approximately 7% of the population [3]. Hungarians originate from the eastern side of the Ural Mountains, whereas the Roma people are from the Indian subcontinent. Genetic ancestry studies have provided sufficient evidence that the Roma's ancestors migrated to Europe about 850 years ago, with a severe founder event and a very little genetic admixture occurring with populations encountered on the way. After settled in the Balkans, this initial population fragmented and dispersed throughout Europe resulting in the mixture of their genomes with those of various European ethnic groups, including Hungarians. Despite the fact that today's Hungarian Gypsies have genetic variants typical of India and Europe, the two populations fundamentally differ from each other in their genetic architectures [4, 5]. Recent genetic epidemiological studies also indicated significant genetic differences between the Hungarian general (HG) and Hungarian Roma (HR) populations [6, 7, 8, 9].

Generally, the Roma minority is considered to possess unfavorable health and live in a less healthy environment (colonies) than the majority of the population, not only in Hungary but also in other European countries [10]. The low educational status as well as high unemployment and poverty rates may further contribute to ill health among the Roma [11]. Because of the generally high prevalence of adverse risk factors such as obesity, metabolic syndrome, hypertension, smoking and alcohol consumption, this population is at an increased risk of diabetes, cardiovascular diseases and, consequently, early mortality [12, 13, 14]. Studies investigating anthropometric features in Roma populations in Slovakia, Spain, and Serbia have suggested that the prevalence of obesity and BMI are higher among the Roma than in the general population [15, 16, 17].

Environmental factors play a crucial role in the development of obesity; however, there is no doubt that it is also influenced by genetics [18]. In the past few years, genome-wide association studies (GWAS) identified dozens of polymorphisms in genes or gene regulatory regions involved in energy homeostasis, appetite regulation as well as lipid and carbohydrate metabolism that have cumulative effects on body weight [19]. However, most of these studies were performed in populations of European descent. The results of genetic replication studies investigating the relationship between SNPs identified by GWAS and obesity-related traits were not always consistent. The reasons could be that these studies were conducted in populations with different environmental characteristics and/or were performed on samples of relatively small size. Furthermore, there is a considerable genetic heterogeneity across populations in terms of allele frequency, linkage disequilibrium (LD), and haplotype structure that can arise as a result of multiple factors such as genetic drift, mutations, and natural selection. If a true functional SNP is in strong LD with the lead SNP in the population in which the GWAS was performed, but not in a different population, then the lead SNP will not be associated with the phenotype in this second population. Thus, interethnic differences in LD block patterns limit the generalizability of GWAS association signals across populations [20]. Consequently, differences within and among human communities in terms of susceptibility to obesity could be explained by the different genetic background of a population and/or its interaction with various environmental exposures [21]. Moreover, it has not been clarified how much these SNPs contribute to obesity risk and related quantitative factors if combined and whether they can be considered as possible predictors of obesity, which might have implications for early prevention and intervention.

The present study was designed to explore differences in genetic predisposition to obesity between the HG and HR populations by investigating the distribution of 20 recently identified obesity susceptibility loci as well as their association, individually and in combination, with the prevalence of obesity.

Subjects and Methods

Study Sample of the Hungarian General Population

In 2006, a cross-sectional study was conducted based on the General Practitioners' Morbidity Sentinel Stations Program to define the prevalence of metabolic syndrome in the Hungarian population [22]. The source population of the study included all individuals aged 20–69 years who were registered by the 59 participating general practitioners (GPs) of eight Hungarian counties. The study population was randomly selected proportional to the size of the practices to represent the Hungarian adult population based on geographic, age, and sex distributions. GPs performed physical examinations (weight, height, blood pressure measurements) and collected blood samples for laboratory investigations and DNA isolation. Information on the sociodemographic status, family history, and lifestyle were also collected using self-administered questionnaires. In the framework of this study, 1,783 DNA samples were acquired.

Study Sample of the Hungarian Roma Population

The study population was enrolled from counties in Northeast Hungary, the area where the Roma are most prevalent and the majority of segregated Roma colonies are located. The participants were interviewed by Roma field workers in the framework of two surveys conducted recently [23, 24]. The ethnicity of the participants was assessed by self-declaration. 92 segregated colonies with more than 100 inhabitants were considered in the surveys, of which 65 were randomly selected using the GPs' validated household lists. Afterwards, 25 households in each colony were randomly chosen, and one person from each household was invited to the GPs. As a part of the health examination survey, physical examinations (weight, height, blood pressure measurements) were performed, and whole blood samples were taken for routine laboratory tests and DNA extraction. Data on sociodemographic factors, lifestyle, and self-assessed health status were obtained using interviewer-assisted questionnaires. A total of 1,170 DNA samples from 20- to 69-year-old Roma individuals were used for genotyping.

Genotyping

Genomic DNA was extracted from EDTA-coagulated whole blood samples using the MagNa Pure LC DNA Isolation Kit - Large Volume (Roche Diagnostics, Mannheim, Germany) according to the manufacturer's instructions. Genotyping was performed in the Mutation Analysis Core Facility (MAF) of the Karolinska University Hospital (Stockholm, Sweden) using the Sequenom MassARRAY platform (Sequenom Inc., San Diego, CA, USA) with iPLEX Gold chemistry [25]. Validation, concordance analysis and quality control were conducted by the MAF according to their protocols, resulting in a successful genotyping outcome for 2,895 (1,743 Hungarian and 1,152 Roma) DNA samples.

SNP Selection and Computation of Genetic Risk Score

A systematic literature review was undertaken in the PubMed database with an emphasis on GWAS to identify SNPs that were found to play a significant role in the development of obesity. Given that most GWAS have described common obesity-associated gene loci in European-descent populations, multi-ancestral meta-analyses were also examined to discover gene polymorphisms showing association with obesity in populations with different ethnic background. SNP selection was based on HapMap data for a European ancestry population sample (CEU) with a minor-allele frequency > 5% (www.hapmap.org) and previously published results showing significant associations between these polymorphisms and obesity-related traits. The top 20 SNPs, identified by the magnitude of the described association (odds ratio or beta value), were selected for further genotyping (table 1).

Table 1.

Obesity predisposing SNPs selected for the study

SNP	Gene		Described association		Reference
	name	abbreviation	odds ratio of obesity	mean difference in BMI (β)
rs1137101	leptin receptor	LEPR	1.13	–	Paracchini et al. [26]
rs2815752	neuronal growth regulator	NEGR1	–	0.13	Speliotes et al. [27]
rs2867125	transmembrane protein 18	TMEM18	–	0.31	Speliotes et al. [27]
rs6548238			–	0.26	Willer et al. [28]
rs1801282	peroxisome proliferator-activated receptor gamma	PPARG	–	0.06	Galbete et al. [29]
rs2241766	adiponectin, C1Q and	ADIPOQ	1.39	–	Wu et al. [30]
rs1501299	collagen domain containing		0.89	–	Lu et al. [31]
rs10938397	glucosamine-6-phosphate deaminase 2	GNPDA2	–	0.18	Speliotes et al. [27]
rs16139	neuropeptide Y	NPY	–	0.58	Yeung et al. [32]
rs925946	brain-derived neurotrophic	BDNF	1.11	–	Thorleifsson et al. [33]
rs6265	factor		1.12	–	Thorleifsson et al. [33]
rs660339	uncoupling protein 2	UCP2	–	0.39	Brondani et al. [34]
rs659366	(mitochondrial, proton carrier)		1.06		Brondani et al. [34]
rs6499640	fat mass and obesity-	FTO	1.16	–	Thorleifsson et al. [33]
rs1558902	associated protein		–	0.39	Speliotes et al. [27]
rs1121980			–	0.31	Vimaleswaran et al. [35]
rs9939609			–	0.33	Willer et al. [28]
rs9941349			–	0.56	Cronin et al. [36]
rs17782313	melanocortin 4 receptor	MC4R	–	0.2	Willer et al. [28]
rs12970134			1.12	–	Thorleifsson et al. [30]

Two types of multilocus genetic risk score (GRS) were constructed for each individual in the study populations. GRS1 incorporated all independent SNPs, while GRS2 included only SNPs associated with obesity with a p < 0.05 in the present study. Only one representative SNP was selected from a LD block if there were absolute LDs with r² ≥ 0.2. GRS was defined by a simple count method (i.e., by summation of the number of risk alleles) assuming additive genetic model in which each risk allele contributes equally to the risk for obesity, and summed over all the SNPs in the set.

Statistical Analysis

Data were analyzed using PLINK v1.07 (http://pngu.mgh.harvard.edu/purcell/plink) [37], Stata v12.0 (StataCorp LP, College Station, TX, USA) and CaTs [38] software. Departure from Hardy-Weinberg equilibrium (HWE) as well as differences in obesity prevalence and the allele frequencies of individual SNPs were calculated with Pearson's chi-square test. Distributions of risk allele frequencies and genetic risk scores in the two study populations were compared by binomial distribution and Kolmogorov-Smirnov tests, respectively. To estimate the risk of obesity (18.5 ≤ BMI < 25 kg/m² vs. BMI ≥ 30 kg/m²) for each SNP, we calculated the odds ratio (OR) and 95% confidence intervals (95% CI) using multivariate logistic regression. To ensure the highest statistical power, data of the two study groups were analyzed jointly. In the regression analyses, the additive model was used and adjusted for covariates including age, gender and ethnicity. The possible interethnic differences in the association patterns were assessed by introducing a multiplicative interaction term in the model assuming additive allelic effect and dichotomus ethnicity. The threshold for statistical significance was 0.05. The power to detect association was calculated using the minor allele frequency of each SNP in our case-control samples and the effect size calculated in the allelic association analysis.

In a further analysis, we evaluated the association of genetic risk scores with obesity risk (18.5 ≤ BMI < 25 kg/m² vs. BMI ≥ 30 kg/m²) using multivariate logistic regression models. Interactions between genetic effects and ethnicity were analyzed by introducing the corresponding interaction term into the models as well.

Results

Demographic Characteristics and Distribution of Obesity Prevalence in the Study Populations

Differences in demographic characteristics, as well as in the distribution of obesity prevalence were observed between the HG and HR population samples (table 2, fig. 1). The female proportion was higher in the HR (59.6%) than in the HG (53.1%) population, although women were the majority in both study samples in all age groups, except the 20- to 29-year-old HG group. The age distribution of the HR sample was slightly shifted towards the younger age groups compared to the HG sample. The average ages in the HG and HR populations were 46 and 41.4 years, respectively.

Table 2.

Demographic characteristics of the study populations

	Hungarian general population							Hungarian Roma population
	male		female		total		female / male ratio	male		female		total		female / male ratio
	n	%	n	%	n	%		n	%	n	%	n	%
Number of subjects	818	100	925	100	1,743	100	1.13	465	100	687	100	1,152	100	1.48
Age groups
20–29 years	147	17.9	137	14.8	284	16.2	0.93	106	22.8	138	20.1	244	21.1	1.3
30–39 years	155	18.9	160	17.3	315	18.0	1.03	111	23.8	174	25.3	285	24.7	1.57
40–49 years	167	20.4	202	21.8	369	21.1	1.21	119	25.5	191	27.8	310	26.9	1.61
50–59 years	203	24.8	236	25.5	439	25.1	1.16	91	19.5	140	20.3	231	20.1	1.54
60–69 years	146	17.8	190	20.5	336	19.2	1.3	38	8.2	44	6.4	82	7.1	1.16

Fig. 1.

Age-specific prevalence (%; 95% CI) of obesity in the Hungarian general and Roma populations among males (A) and females (B). Statistically significant differences are indicated by asterisks (*p < 0.05).

The prevalence of obesity in both men and women were higher in the HG than in the HR population with significant difference only between HG and HR males. Obesity was less frequently observed among men than women in both study groups. The age- and sex-specific prevalence of obesity followed a similar pattern in younger age groups in both samples; however, in age groups older than 40–49 years, the proportion of HR people with obesity, especially of males, was strongly reduced.

Frequency and Impact of SNPs in the Study Populations

All SNPs were in Hardy-Weinberg equilibrium in both study groups. Eleven obesity-predisposing polymorphisms showed significant difference in the risk allele frequencies between the two study groups (table 3). Six of them had significantly higher risk allele frequency in the HR population. Nonetheless, when comparing the risk allele distribution of the entire SNP panel between the two study populations, none of them were significantly enriched by obesity-associated genetic markers.

Table 3.

Allele frequencies of individual SNPs in the Hungarian general and Roma populations

Gene/SNP	Risk allele	Risk allele frequency
		HG	HR	Pfor difference
LEPR
rs1137101	G	0.46	0.44	0.267
NEGR1
rs2815752	A	0.66	0.73	<0.001
TMEM18
rs2867125	C	0.81	0.86	<0.001
rs6548238	C	0.81	0.89	<0.001
PPARG
rs1801282	C	0.13	0.05	<0.001
ADIPOQ
rs2241766	G	0.11	0.11	0.916
rs1501299	T	0.29	0.28	0.32
GNPDA2
rs10938397	G	0.46	0.41	0.001
NPY
rs16139	C	0.04	0.02	<0.001
BDNF
rs925946	T	0.26	0.37	<0.001
rs6265	C	0.8	0.89	<0.001
UCP2
rs660339	G	0.59	0.63	0.001
rs659366	C	0.63	0.65	0.146
FTO
rs6499640	A	0.59	0.51	<0.001
rs1558902	A	0.45	0.43	0.037
rs1121980	A	0.47	0.46	0.571
rs9939609	A	0.44	0.43	0.373
rs9941349	T	0.45	0.43	0.246
MC4R
rs17782313	C	0.22	0.22	0.629
rs12970134	A	0.24	0.22	0.063

HG = Hungarian general population; HR = Hungarian Roma population.

The FTO gene variants, namely rs1558902, rs1121980, rs9939609 and rs9941349, showed a strong association with obesity in the joint analysis, as indicated by the odds ratios of 1.34 (95% CI 1.17–1.54; p < 0.001), 1.35 (95% CI 1.18–1.56; p < 0.001), 1.35 (95% CI 1.17–1.55; p < 0.001) and 1.41 (95% CI 1.22–1.62; p < 0.001). Further association signals could be replicated for SNPs in the GNPDA2 (rs10938397), NPY (rs16139), FTO (rs6499640), and MC4R (rs17782313 and rs12970134) genes. The sample size of the joint analysis permitted relatively high power (86–100%) for these SNPs to detect association when allowing a false-positive rate of 0.05. However, the sample size showed rather low power for the other SNPs (5–64%) suggesting that they have almost no or a rather weak contribution to the development of obesity in the study populations. In addition, significant interaction between genetic variants and ethnicity could be observed only for rs1801282 in the PPARG gene, indicating virtually the absence of ethnicity-related genetic predisposition to obesity in our study sample (table 4).

Table 4.

Associations of individual SNPs with obesity in the study populations (joint analysis) and interactions between individual SNPs and ethnicity^*

Gene/SNP	Risk allele	ORa (95% CI)	p	Power (%)b (p = 0.05)	Expected sample sizec		Interaction analysisd
					case	control	ORa (95% CI)	p
LEPR
rs1137101	G	1.01 (0.87–1.16)	0.941	5	>100.000	>100.000	0.88 (0.66–1.75)	0.391
NEGRI
rs2815752	A	1.13 (0.97–1.31)	0.119	64	1,313	1,313	0.95 (0.69–1.31)	0.772
TMEM18
rs2867125	C	1.08 (0.89–1.30)	0.416	23	4,826	4,826	1.09 (0.74–1.63)	0.641
rs6548238	C	1.15 (0.95–1.39)	0.159	58	1,447	1,447	1.33 (0.87–2.04)	0.188
PPARG
rs1801282	C	1.08 (0.85–1.37)	0.519	16	7,595	7,595	0.57 (0.33–0.98)	<0.05
ADIPOQ
rs2241766	G	1.08 (0.86–1.34)	0.639	17	6,905	6,905	1.11 (0.71–1.74)	0.635
rs1501299	T	1.13 (0.97–1.32)	0.129	63	1,370	1,370	0.91 (0.66–1.24)	0.556
GNPDA2
rs10938397	G	1.17 (1.01–1.34)	<0.05	87	749	749	1.21 (0.91–1.61)	0.19
NPY
rs16139	C	0.56 (0.38–0.81)	<0.01	100	239	239	0.88 (0.37–2.11)	0.778
BDNF
rs925946	T	1.13 (0.97–1.32)	0.114	64	1,332	1,332	1.03 (0.76–1.39)	0.859
rs6265	C	1.16 (0.95–1.39)	0.139	64	1,311	1,311	0.85 (0.56–1.28)	0.434
UCP2
rs660339	G	1.09 (0.95–1.26)	0.235	41	2,398	2,398	1.21 (0.9–1.61)	0.206
rs659366	C	1.09 (0.94–1.26)	0.245	40	2,454	2,454	1.12 (0.84–1.49)	0.054
FTO
rs6499640	A	1.12 (0.98–1.29)	0.103	62	1,389	1,389	1.07 (0.81–1.42)	0.631
rs1558902	A	1.34 (1.17–1.54)	<0.001	100	241	241	1.21 (0.91–1.61)	0.184
rs1121980	A	1.35 (1.18–1.56)	<0.001	100	232	232	1.21 (0.91–1.61)	0.184
rs9939609	A	1.35 (1.17–1.55)	<0.001	100	229	229	1.19 (0.89–1.58)	0.224
rs9941349	T	1.41 (1.22–1.62)	<0.001	100	181	181	1.24 (0.93–1.65)	0.142
MC4R
rs17782313	C	1.19 (1.01–1.42)	<0.05	86	782	782	0.9 (0.64–1.23)	0.563
rs12970134	A	1.19 (1.04–1.41)	<0.05	87	754	754	0.87 (0.62–1.22)	0.431

^aIncrease in the odds of being obese (BMI ≥ 30 kg/m²) versus being normal weight (18.5 ≤ BMI < 25 kg/m²) for each additional risk allele.

^bCaTs software [38] were used for the power calculation under additive genetic model and the prevalence of obesity was assumed 28% according to the literature [1].

^cMinimum sample size that permits to detect significant (p<0.05) association with the power of 80%.

^dBased on model assuming additive allelic effect and dichotomus ethnicity in a multiplicative interaction term.

^*Regression models were adjusted for age, gender and ethnicity.

Obesity according to Genetic Risk Score

A set of 14 independent variants was used to calculate GRS1, while GRS2 was constructed by using four independent SNPs that were associated significantly with obesity in the target samples. SNPs were considered as independent if the LD between them had r² < 0.2. Individual GRS1 and GRS2 ranged from 5 to 20 and from 0 to 6 among our study subjects, respectively, with higher scores indicating a higher genetic predisposition to obesity. Individuals with less than 7 or more than 19 GRS1 were aggregated because of the small number of observations in those groups. GRSs followed normal distributions which were not significantly different (GRS1 p = 0.3; GRS2 p = 0.872) between the two study groups (fig. 2).

Fig. 2.

Distribution of the GRS1 combining 14 SNPs (A) and GRS2 combining 4 SNPs (B) in the Hungarian general (grey) and Hungarian Roma (black) populations.

Significant positive linear relationships were observed between both GRSs and the studied phenotype (fig. 3). GRS2 displayed higher predictive value than GRS1, as indicated by higher odds ratios. Each additional GRS1 and GRS2 unit, corresponding to one risk allele, was significantly associated with a 3% (95% CI 1.01–1.05; p = 0.003) and 19% (95% CI 1.1–1.28; p < 0.001) increase in risk of obesity, respectively, if the two populations were analyzed together. Although significant interaction between the genetic risk and ethnicity could not be observed, both GRSs exhibited different association patterns between the HG and HR populations (table 5). Stratification by ethnicity revealed that a one-unit increment in GRS1 and GRS2 in the HR individuals was significantly associated with a 12% (95% CI 1.05–1.18; p = 0.001) and 27% (95% CI 1.12–1.45; p < 0.001) higher risk of obesity, respectively. The correlation between GRSs and obesity was less pronounced in the HG study sample.

Fig. 3.

Association of GRS1 combining 14 SNPs (A) and GRS2 combining 4 SNPs (B) with obesity in the study populations (joint analyses). Histograms (Y axis on right, grey bars) represent the percentage of individuals in each risk score category. Odds ratios and mean BMI (Y axis on left) are plotted. Error bars represent 95% confidence intervals.

Table 5.

Associations of the GRS1 combining 14 SNPs and GRS2 combining 4 SNPs with risk of obesity in the Hungarian general and Hungarian Roma populations

	Hungarian general population		Hungarian Roma population
	ORa (95% CI)	Pfor trendb	ORa (95% CI)	Pfor trendb
GRS1	1.11 (1.04–1.17)	<0.001	1.12 (1.05–1.18)	0.001
GRS2	1.14 (1.02–1.27)	0.018	1.27 (1.12–1.45)	<0.001

^aIncrease in the odds of being obese (BMI ≥ 30 kg/m²) versus being normal weight (18.5 ≤ BMI < 25 kg/m²) for each additional genetic risk score.

^bP_{for trend} values were adjusted for age and gender.

Discussion

There is general agreement that in a genetically isolated population with low environmental variability, such as the Roma, the founder effect and genetic drift can cause an increase in the frequency of particular alleles [39]. In this study, we reported significantly higher risk allele frequencies for SNPs in NEGR1 (rs2815752), TMEM18 (rs2867125 and rs6548238), BDNF (rs925946 and rs6265), and UCP (rs660339) genes but significantly lower risk allele frequencies for variants in GNPDA2 (rs10938397), NPY (rs16139), PPARG (rs1801282), and FTO (rs6499640 and rs1558902) genes in HR compared with the HG population. By comparing the distribution of investigated risk allele variants, increased genetic predisposition to obesity in the Roma could not be confirmed in this study.

The SNPs with the most robust effect on obesity in our target populations were found in the intronic region of the FTO gene, which is mainly involved in the regulation of energy homeostasis and body composition. However, despite intensive research its exact pathomechanism is still not completely understood [40]. The impacts of FTO gene polymorphisms found in this study were consistent with previous reports not only in European-descent populations [28, 35, 36] but also in the Slovakian Roma [41], indicating that the effect of these susceptibility loci for obesity persists across populations with various ancestral origins. Nonetheless, common variants in the FTO gene did not show an obvious association with obesity-related phenotypes in Spanish Roma populations [42].

In the analysis of the combined effect of the selected variants, GRS1 and GRS2 yielded similar results; both of them displayed significant positive correlation with obesity. The inclusion of only a subset of SNPs that were found to have significant effects in this study has allowed for creating a genetic risk score profile (GRS2) with increased predictive validity, as reflected by the association measures. The analysis of GRSs stratified by ethnic groups demonstrated differences between the two study populations. Whereas both GRSs revealed a stronger relationship with obesity risk in the HR population, these associations were weakened in the HG sample, indicating difference in the penetrance of obesity-related variants in our study populations. This finding is an interesting observation because significant interactions of individual SNPs or GRSs with ethnicity could not be established if the two populations were analyzed jointly.

Despite the lack of marked heterogeneity of genetic risk to develop obesity in the study populations, excess body weight, surprisingly, was less frequently observed among Roma than in the HG population, which is consistent with a previous report [43]. This observation may be explained in part by the Roma's generally deprived socioeconomic status [44], which imposes limits the transition from their traditional lifestyle to a westernized one. Although their nutritional characteristics (consumption of fruits and vegetables and type of fat used for cooking) were reported to be generally unhealthier than that of HG population [43], their overall energy intake does not appear to reach the degree with which the HG population can be characterized. Regardless of dietary intake, it is important to consider that other ethnic-specific lifestyle components (alcohol consumption, physical activity levels, or smoking habits) could also act as modifiers of the effect of a given genetic variant on body weight [18]. Our findings suggest that the high environmental variability associated with more favorable socioeconomic conditions and the cultural habits resulting in abundant energy intake of the Hungarian general population highlight the importance of environmental and lifestyle factors in the attenuation of penetrance and expression of obesity-predisposing genetic variants. These facts may be possible explanations for the phenotypic differences observed between the two populations.

Several limitations of the present investigation need to be considered. An important point for discussion is that the HR study population is not fully representative of the overall Roma population in Hungary. Due to the sampling design, those Roma who have assimilated within the Hungarian general population or declared themselves as Hungarians were not included in the HR population survey. Moreover, it is also necessary to note that the representative sample of the Hungarian general population included some people who are Roma because the data collection of HG sample did not involve the ascertainment of ethnicity. Similarly, the presence of participants with mixed Roma/non-Roma ancestry could not be excluded in the recruitment of both study populations. Their inclusion may result in a slight underestimation of the differences between the populations. Furthermore, the lack of information both on environmental factors, including physical activity and dietary habits, and gene-environment interactions is a limitation of our work. Lastly, we acknowledge that the impact of SNPs selected from meta-analyses has not been confirmed by GWAS and the possible absence of their association signals may decrease the predictive value of GRS computed in our study. On the other hand, a series of GWAS-established loci were not included in our analysis but could also contribute to the phenotypic differences between the two populations.

In conclusion, this study is the first to investigate the joint effect of obesity-related gene variants among Roma living in segregated colonies and to compare them with data of the majority of the population. Our findings suggest that the differences in obesity prevalence between the HG and HR populations could be primarily explained by ethnicity-related environmental and behavioral factors, and not by genetics. Nevertheless, particular gene-environment interactions might contribute to the distinct penetrance of the obesity-associated genetic factors in populations of different ethnic backgrounds.

Ethics Approval

All procedures performed in studies involving human participants were in accordance with the ethical standards of the Ethical Committee of the University of Debrecen (reference No. 2462-2006), the Ethical Committee of the Hungarian Scientific Council on Health (reference No. 8907-O/2011-EKU, 285/PI/11) and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

This article does not contain any studies with animals performed by any of the authors.

Informed consent was obtained from all individual participants included in the study.

Disclosure Statement

The authors declare that they have no conflict of interest.