- Top of page
- Research Methods and Procedures
Objective: To investigate the role of genetic admixture in explaining phenotypic variation in obesity-related traits in a sample of African-American women (n = 145) and to determine significant associations between obesity traits and admixture genetic markers.
Research Methods and Procedures: Associations between genetic admixture and BMI, resting metabolic rate, fat mass, fat-free mass, and bone mineral density were tested using linear regression considering the estimation of admixture by 1) a maximum-likelihood approach (MLA) and 2) a Bayesian analysis.
Results: Both the conservative MLA and the Bayesian approach support an association between African genetic admixture and BMI. Evidence for the associations of African genetic admixture with fat mass and fat-free mass was supported by the Bayesian analysis; the MLA supported an association with bone mineral density. When the individual ancestry informative markers that were used to estimate admixture were tested for associations with BMI, significant associations were identified in chromosomes 1, 11, and 12.
Discussion: These results provide evidence supporting the application of admixture mapping methods to the identification of genes that result in higher levels of obesity among African-American women. Further research is needed to replicate and further explore these findings.
- Top of page
- Research Methods and Procedures
Although health disparities among races and ethnic groups have been declared a priority for public health initiatives, the extent to which genetic and environmental factors account for ethnic and racial differences in complex traits such as obesity, cardiovascular disease, and diabetes continues to be enigmatic. Efforts to identify genetic variations accounting for group differences in such complex polygenic phenotypes continue to challenge researchers.
Most of the ethnic groups in the United States have resulted mainly from the intermixing of European, African, and Native-American populations during the colonization and continued habitation of the New World. Genetic variants or alleles from these previously isolated parental populations were brought together in new combinations establishing the gene pools of the various contemporary European-, African-, Hispanic-, and Native-American resident populations. Consequently, individuals of these populations, who inherit variants that predispose them either to disease-related traits or to a greater sensitivity to environmental exposure, will have a greater chance of acquiring the disease.
Epidemiological evidence has supported differences among individuals of various backgrounds in different obesity-related traits, including BMI and resting metabolic rate (RMR),1 particularly when African-American women are compared with European-American women. For example, obesity is more prevalent in African-American women than their white counterparts, even after controlling for socioeconomic status (SES) (1). In addition, there are reports of lower levels of energy expenditure in African-American women (2, 3, 4, 5). Higher bone mineral density (BMD) in African-American individuals when compared with European Americans has also been reported (6, 7). Because these differences might result, in part, from the inheritance of disease-predisposing alleles from parental populations, the identification of such alleles may be a valuable tool for exploring the genetic component contributing to these health disparities. The estimation of the degree of parental admixture on members of genetically admixed populations has been described in the literature as an approach to identifying genetic influences of this sort (8, 9, 10). This approach has been used by Williams et al. (11) to suggest a genetic susceptibility for both diabetes and obesity by the negative association between European admixture and prevalence for type 2 diabetes and obesity in Pima Indians.
This study capitalizes on the estimation of the degree of individual genetic admixture to identify genetic influences in obesity-related traits using a specially selected panel of ancestry informative markers (AIMs). Associations between individual estimates of genetic admixture with measures of body composition, bone density, and RMR are tested in a sample of African-American women. Additionally, these AIMs are tested for an association with these phenotypes.
- Top of page
- Research Methods and Procedures
Table 2 demonstrates descriptive statistics for the site-specific admixture estimate, age, and body composition variables. No significant differences were observed in levels of African admixture among the three sites. However, age significantly differed among the three sites (F = 103.6, p < 0.0001). When the mean values of the obesity-related phenotypes at each site were compared, significant differences were observed for BMI (F = 29.03, p < 0.001), FM (F = 34.09, p < 0.001), FFM (F = 9.41, p < 0.001), and BMD (F = 4.46, p < 0.013), as qualified by a one-way ANOVA. After adjusting by FM and FFM, levels of RMR were not significantly different among the three sites (F = 0.363, p < 0.696). Even after adjusting these variables by age, the results remained the same: significant differences across study sites in the body composition variables and no significant differences in RMR.
Table 2. Descriptive statistics for subjects studied at each of three sites (data shown as means ± SD)
| || || || || ||Body composition|
|Site||N*||Age||Admixture estimate||RMR||BMI (kg/m2)||FM||FFM||BMD|
|Alabama||46||33.08 ± 5.53||82.72 ± 14.06||1317.72 ± 191.07||25.52 ± 3.41||26.81 ± 8.50||42.90 ± 4.19||1.20 ± 0.08|
|Maryland||38||56.32 ± 5.97||80.92 ± 12.90||1546.26 ± 213.65||34.71 ± 4.24||44.29 ± 9.19||48.11 ± 6.29||1.24 ± 0.09|
|New York||61||34.57 ± 10.80||82.79 ± 16.86||1446.61 ± 220.39||29.86 ± 6.47||33.23 ± 12.44||53.11 ± 19.14||1.24 ± 0.07|
|All||145||39.83 ± 12.86||82.28 ± 14.96||1432.23 ± 226.18||29.78 ± 6.16||34.32 ± 12.41||48.62 ± 13.69||1.23 ± 0.08|
R2 and significance values resulting from the association between the obesity-related phenotypes and AFADM are reported in Table 3. R2 values indicated in the table refer to the contribution of AFADM on the dependent variable after removing the contribution of covariates (i.e., “semi-partial” R2 values). Significant R2 values were obtained for BMI and BMD, whereas values for FM, FFM, and RMR (adjusted by FM and FFM) were of borderline significance.
Table 3. Results of the association between AFADM and the obesity-related phenotypes
|Dependent variable||Slope||Lower CI||Upper CI||R2*||p value|
Table 4 shows the independent associations of genetic markers with BMI. Four different markers were significantly associated with BMI at the 0.05 probability level. Results from the AFADM-adjusted models are also shown in Table 4. The results of the adjusted models remained significant even when the adjustment for the association was performed using a value of admixture that included all markers for the individual admixture estimate (data not shown).
Table 4. Results of the associations of the independent markers and BMI using MLA
| || || ||p value|
Results of the association between admixture and the phenotypic measures using BA are demonstrated in Table 5. The posterior means and 95% credible intervals from BA are asymptotically equivalent (with large sample size and noninformative prior distributions) to maximum-likelihood estimates and 95% confidence intervals. A 95% credible interval that does not overlap 0 is, thus, asymptotically equivalent to a p value <0.05. The estimate of the mean of population African admixture was 83% (95% credible interval: 81 to 86), and there was no evidence that income was related to admixture or that it differed among sites. The Bayesian analysis supported associations of BMI, FM, and FFM with African admixture with 95% credible intervals that did not overlap zero. The results from the test for linkage in the model with genes underlying ethnic differences in BMI supported only two loci (APOAI and GNB3) with evidence for linkage significant at p < 0.05. Because the test at locus APOAI uses additional information from the adjacent locus DRD2TAQD, which is only 4 cM away, it is not surprising that both DRD2 and APOA1 show associations in the same direction.
Table 5. Results from the Bayesian model
| ||BMI (kg/m2)||BMD||FM||FFM||RMR*|
| ||Posterior mean||2.5%||97.5%||Posterior mean||2.5%||97.5%||Posterior mean||2.5%||97.5%||Posterior mean||2.5%||97.5%||Posterior mean||2.5%||97.5%|
- Top of page
- Research Methods and Procedures
Both the conservative MLA and the BA support an association between AFADM and BMI. Evidence for the associations of AFADM with FM and FFM is stronger with the BA, which makes fuller use of the data, than with the classical analysis; however, for BMD, the proximity of overlap to 0 in the BA makes it equivalent to a p value just >0.05. Nevertheless, the results of this investigation support the relevance of using AIMs in studying the genetics of complex traits in admixed populations and suggest that the differences in the prevalence of obesity-related phenotypes among African-American and European-American women could be partly attributable to genetic factors. These findings also provide insight into the role of genes accounting for racial differences in complex biomedical traits, suggesting a genetic contribution to the well-documented higher levels of BMI in African-American vs. European-American individuals.
After establishing a significant association between AFADM and BMI, it was intuitively reasonable to expect similarly significant results with measures of FM and FFM in the MLA. Only close-to-significant results for both phenotypes were shown by MLA, most likely because of the small sample size, whereas the BA supported the expected associations with both FM and FFM. Using BA, we noted a significant association between AFADM and percentage of body fat, which was not supported by MLA (data not shown).
In the case of RMR, a small sample size might also have explained the lack of association between AFADM and RMR using the MLA. In the BA, the 95% posterior interval for the effect of admixture overlaps zero, but with a very wide range (up to 10 SDs), which might suggest that not enough information was available to fit a complex model with seven covariates. The association between AFADM and RMR was somewhat expected, based on evidence from epidemiological studies, wherein lower RMR levels have been found in African-American compared with other North-American, genetically admixed groups (3, 4, 5, 13), and yet the slope of the relationship between RMR and body composition between African-American and African populations did not differ (21). Interestingly, on preliminary data analysis, a significant association between RMR and AFADM was observed in the New York sample, but not in the Maryland or Alabama samples (data not shown). This observation might provide insight into reported differences in energy levels according to geographic location (22). Future investigations relating the role of genetic admixture to RMR in large samples of admixed populations are warranted for enhancing understanding of this phenomenon.
Interpretations regarding the extent to which AFADM explains variation in each of the studied phenotypes deserves discussion. The R2 values presented in Table 3 represent biased estimates of the association of admixture with these phenotypes. In part, this is because of the lack of information that the value of AFADM provides to the model. AFADM ranged in the samples from 35 to 100; therefore, no information is available at lower levels of AFADM. Also, the R2 estimates depend on how much AFADM varies among individuals, and the MLA tends to underestimate the true relationship by not allowing for uncertainty in the individual admixture estimates. Finally, the sample was selected at some sites for extremes of BMI that can inflate the estimate of proportion of variance. Therefore, the percent variance explained is not necessarily a meaningful measure of strength of association in relation to observed ethnic difference. Nevertheless, the results of the study support the hypothesis that AFADM contributes to obesity traits in this population, characterizing the role of genetics in observed racial/ethnic differences in these phenotypes. Future research using representative samples will be needed to produce unbiased estimates of percent variance.
The genotype/phenotype associations described in the results support previously reported findings for obesity-related traits. The significant association in chromosome 1 for BMI occurred in a region close to the LMNA gene, which has been associated with BMI in other populations (23, 24) and close to the locus at which Vionnet et al. (25) reported suggestive linkage to type 2 diabetes. Variants of the dopamine receptor in chromosome 11 (DRD2TAQD) have also been associated with weight and height (26) and BMI (27), and the GNB3 allele has been associated with BMI in a number of populations (28, 29). It is important to reiterate that the association tests in this study were performed using those AIMs, namely markers that have demonstrated differences in parental frequencies, which were available to the investigators. Although some of these AIMs may be nonfunctional polymorphisms, the use of these markers is appropriate because their association with the phenotype allows the identification of chromosomal regions influencing the trait. Given the limited amount of markers in the panel tested in this study, only limited regions of the genome were scanned. There are other regions across the genome influencing obesity-related traits that were not tested in this study because the regions were not defined in our battery of markers.
In the MLA, several markers showed significant association with BMI, even after adjustment for admixture. The strongest association of a marker with BMI was with FY-Null, which is almost perfectly informative for ancestry because the null allele is almost fixed in West-African populations and very rare in non-Africans. A limitation of the MLA is that it relies on “plugging in” the maximum-likelihood estimates of admixture into a regression model that does not allow for the uncertainty in these measurements. Thus, associations of the trait with marker loci may persist after adjustment for admixture because of residual confounding. In the BA, this uncertainty is correctly accounted for by using a score test that averages over the posterior distribution of admixture. However, the fact that other associations with obesity-related traits have been reported with regard to this region does not rule out the possible involvement of the chromosomal region in explaining ethnic and racial differences in obesity-related traits and deserves further research.
Several other limiting aspects of the study deserve further discussion. Although this investigation reports interesting results, it is limited by a relatively small sample size and the limited number of AIMs. The study did not measure any cultural or environmental component that might influence the phenotypes, and the proxy measure of SES, although not significant in any of the model, was fairly crude and might not necessarily reflect the SES of the subjects at the time of participation in the study. The strength of environmental effects on obesity is apparent from the large differences in mean BMI among the three study sites, which were not explained by differences in admixture. Without more detailed measurements of social, economic, and behavioral factors related to obesity, we cannot rule out confounding by environmental factors as a possible explanation for the observed relationship with admixture. Also, the design and size of the study did not allow for the consideration of models testing for possible interactions that might have contributed to phenotypic variability.
This study supports the use of admixture mapping as a tool to identify genetic influences in obesity traits, as previously proposed by Williams et al. (11). Further research is needed with this approach, including the identification of more markers, the consideration of other admixed populations, such as Afro-Caribbeans, the inclusion of environmental measures, and the use of larger samples.