A genome-wide linkage scan for body mass index on Framingham Heart Study families

Background Genome-wide scan data from a community-based sample was used to identify the genetic factors that affect body mass index (BMI). BMI was defined as weight (kg) over the square of height (m), where weight and height were obtained from the first measurement available between the ages of 40 and 50 years. Results Significant familial correlations were observed in mother:father (spouse) relative pairs and in all relative pairs examined except parent:daughter pairs. Single-point sib-pair regression analysis provided nominal evidence for linkage (p < 0.05) of loci to BMI at 23 markers. Multi-point sib-pair regression analysis provided nominal evidence for linkage to BMI at 42 loci on 12 chromosomes. Empirical p-values showed results consistent with the multi-point results; all but three of the loci identified by multi-point analysis were also significant. Conclusion The largest regions of nominally significant linkage were found on chromosomes 2, 3, and 11. The most significant evidence for linkage was obtained with markers D2S1788, D2S1356, D2S1352, D3S1744, and D11S912 from multi-point sib-pair single-trait regression analysis. Our results are in agreement with some of the recently published reports on BMI using various data sets including the Framingham Heart Study data.


Background
Data from the ongoing NHLBI-supported Framingham Heart Study (FHS) of factors that contribute to cardiovascular disease (CVD) was made available to the Genetic Analysis Workshop 13 (GAW13). The FHS has collected physical exam and lifestyle information from a community-based sample that has identified major CVD risk factors, such as high blood pressure, cholesterol, smoking, and obesity http://www.nhlbi.nih.gov/about/framing ham/timeline.htm. Obesity is a major risk factor for morbidity and mortality from many chronic diseases including CVD [1]. It is known that obesity-related traits such as BMI are influenced by genetic factors with heritability estimates of 40-90% [2]. Most studies have provided evidence for a polygenic etiology of common forms of obesity and BMI [2]. To identify the obesity and BMI loci, genome-wide linkage analyses have been performed in several populations including Europeans [3], Mexican Americans [4][5][6], Pima Indians [7,8], and African Americans [5,9]. In these studies, possible linkage (3 > LOD > 2) of obesity and obesity-related traits such as BMI and diabetes to markers on chromosomes 2, 3,6,7,11,13,17, and 20 were reported.
Recently two groups reported results of genome-wide linkage scans for BMI. Wu et al. found strong evidence for the presence of a quantitative trait locus (QTL) for BMI at 3q27 (marker D3S2427, LOD = 3.40, p = 0.03) in a metaanalysis of studies performed on individuals with Caucasian-, African-, and Mexican-American ethnic ancestry [5]. At least two other studies support linkage of BMI to this region. In a genome scan involving 507 nuclear families, Kissebah et al. reported the presence of a QTL locus on 3q27, which was linked to six traits including BMI [10]. Vionnet et al. also reported significant linkage with obesity-related diabetes to the same region of chromosome 3q [11]. Recent reports for linkage of BMI to other chromosomes includes a study by Deng et al. that identified a major QTL for BMI on 2q14 near the marker D2S347 (LOD = 4.04) in a Caucasian-American sample ascertained through a low-BMI proband [12]. One recent study on the FHS families reported linkage of BMI to chromosome 6q (marker D6S1009, LOD = 4.64) [13]. This study also reported some evidence of linkage to chromosome 11q14 markers, namely D11S4464 and D11S912 [13].
In the present study, we performed a genome scan using BMI and subjects from Framingham Cohorts 1 and 2. We chose to perform sib-pair analysis because the sib-pair method is an efficient way of screening for linked loci and is robust to generational differences in a trait such as BMI. Furthermore, we chose to limit the BMI measurements to a single time-point (ages 40-50) in an attempt to minimize the positive correlation of age on BMI.

Methods
Subjects for this analysis were participants in the ongoing NHLBI-supported Framingham Heart Study (FHS), which included 5209 subjects in Cohort 1 who were recruited in 1948 and 5124 children of the original subjects and spouses of these children in Cohort 2 who were recruited in 1971. Selected data on these subjects were made available to GAW13. The available data included family histories on all FHS participants, phenotypic information including height and weight data, as well as results of genetic analysis of 401 polymorphic markers on chromosomes 1 to 22 in the largest families.
The data investigated for phenotypic assessment and familial correlation in this analysis were obtained from Cohort 1 and Cohort 2 and included the BMI calculated from the height and weight information provided by the FHS. The first recorded height and weight measurement between the ages of 40 and 50 years for each subject was used for the analysis. BMI was calculated by dividing weight in kilograms by the square of height in meters. The average BMI at ages 40 to 50 for individuals in Cohorts 1 and 2, 25.70 kg/m 2 and 26.54 kg/m 2 , respectively, differed significantly (p < 0.01, one-tailed t-test). We tested for significance of sex, cohort, and calendar year on BMI values across the entire data set of 2252 individuals. Sex was significant in all models while cohort was significant when included in models without calendar year. Calendar year was never significant (results not shown). Therefore sex and cohort were used as covariates in all linkage analyses. Our assumption was that sibs would be highly correlated for age at BMI and calendar year of BMI. We tested this assumption; this will be discussed later (see Discussion). Individuals without height and weight measurements between ages 40 to 50 years were removed from all analyses.
GAW13 phenotypic and genotypic data were imported into a Microsoft Access database and exported in appropriate files for familial correlation and linkage analysis in S.A.G.E. (Statistical Analysis for Genetic Epidemiology) 4.2 [14]. Familial correlations and the asymptotic standard errors were estimated using FCOR from S.A.G.E. 4.2. We used GENIBD from S.A.G.E 4.2 to generate identityby-descent (IBD) sharing distributions of the GAW13 data. We applied single-marker as well as multiple-marker IBD analyses to generate IBD distributions for each pair of relatives at each marker. SIBPAL from S.A.G.E 4.2 was used to perform single regression of the mean corrected trait cross-product using IBD information with one marker being evaluated at a time. Furthermore, empirical p-values for all the markers were calculated using the SIB-PAL program from S.A.G.E. 4.2. The covariates sex and cohort were included in the regression models.

Results
The total number of individuals in the total FHS sample supplied to GAW13 is 4692, where 1702 have genotype information and 2252 have weight and height information. The total number of participants with BMI ≥ 30 (i.e., obese) is 248. Two hundred and forty-five individuals among the 1702 that were genotyped lacked data for BMI between the specified age range and were removed from all analyses. Eight individuals had BMI values that were 4 standard deviations above the mean (mean = 25.98). These eight outliers, all female with mean BMI of 47.5, were removed from all analyses.
Familial correlation analysis was done using S.A.G.E. 4.2 (FCOR). Table 1 provides the correlation coefficient and 95% confidence intervals for various relative pairs (sexspecific and non-sex-specific). The most significant familial correlation is seen among brother:brother pairs. A genome-wide linkage analysis was performed using the GENIBD and SIBPAL programs of S.A.G.E 4.2. Through single-point linkage analysis, nominal evidence (p < 0.05) for linkage to BMI was found at 23 markers on 12 chromosomes (Table 2). Evidence for linkage to BMI at a pvalue of less than 0.01 was found with markers D3S1764, D3S3053, D16S516, and D19S246 ( Table 2). The strong-est evidence for linkage using single-point analysis was found at D19S246 (p = 0.000051) on chromosome 19q.
Multi-point linkage analysis identified 42 markers on 12 chromosomes with p-values < 0.05. Empirical p-values for all these markers were calculated and are consistent with multi-point values (Table 3). Through multi-point analy-  To further evaluate the linkage results, -Log 10 (p-value) was plotted for all markers for chromosomes 2, 3, and 11 that showed the strongest evidence for linkage ( Figures  1a,1b,1c). -Log 10 (p-value) was also plotted for all markers on chromosome 6 for comparison with the findings of Atwood et al. [13] (Figure 1d).

Discussion
The results of the familial correlation analysis on this data set provides evidence that significant familial correlations exist for BMI (Table 1) with all sibling pair and parent:son pair types tested, but that both genetic and environmental factors influence this correlation. The rank order of the point-estimate of familial correlation by relative pair type is brother:brother > brother:sister ~ sibling m other:father > paternal:son ~ sister:sister. Brother:brother and brother:sister relative pairs are the only sex-specific relative pair type that have familial correlations for BMI with greater point-estimates than the correlation between mothers and fathers (i.e., spouses), however, brother:brother familial correlation is substantially greater than any other familial correlation. This may be due to the effect of same-sex environmental factors on BMI. When sibling pairs are evaluated for familial correlation without respect to sex, the point estimate of familial correlation for BMI is slightly greater than that of mother:father familial correlation.
In the present study, nominal evidence of linkage to BMI has been found with closely spaced markers on chromosomes 2, 3, and 11 (Table 3, Figure 1a,1b,1c). The results of multi-point analysis (Table 3) are consistent with the single-point analysis results (Table 2) on markers with pvalues <0.01. The empirical p-values are also consistent with the p-values obtained from multi-point analysis ( Table 3).
Because of the significant difference in BMI between Cohorts 1 and 2, which is consistent with secular upward trends in obesity over the last several decades, we tested the significance of calendar year of BMI in the 40 to 50 age range with respect to BMI values. Calendar year was not found to be an important factor and we expected age at BMI to be highly correlated among the sibs in our analysis with calendar year. In order to test our assumption that age at BMI in the narrow 40 to 50 age range did not play an important role, we repeated the analysis for the most significant chromosomes (2, 3, and 11) with age at BMI as a covariate. We found that including age at BMI as a covariate did not affect the linkage results (data not shown).  6 38.33 47.97 55.51 64.29 73.61 73.61 86.82 90.82 99.41 103.16 114.42 125.18 132.58 145.08 147.4 149.89 152 6 112.4 119.1 124.2 134.6 138 152.6 161 176.5 181.9 188.3 201.1 209.4 215.8  1.325.5422.3 26.25 36.65 44.81 57.92 61.52 70.61 78.64  At least three other studies have suggested linkage of BMI to marker D3S2427 and/or other markers within that region of chromosome 3. Marker D3S2427, located on 3q27, was found to be strongly linked to BMI in the study by Wu et al. [5]. In the study by Kissebah et al., another marker on 3q27 at the map location of 190 cM, which is very close to the map location of D3S2427 at 188.29 cM, was strongly linked to BMI [10]. Another study reported significant linkage to the same region of chromosome 3 with diabetes or glucose intolerance in individuals below the age of 45 [11]. Since obesity and diabetes are often correlated with each other, some of the same loci may increase susceptibility to both diabetes and obesity. Our results also suggest possible linkage of BMI to markers D6S503 and D6S1027 on chromosome 6 (184.51-187.23 cM) ( Table 3, Figure 1d). Duggirala et al. reported linkage of certain obesity-related traits such as insulin levels and leptin concentrations to chromosome 6q markers at 135-161 cM in nondiabetics from Mexican-American families [16]. This group also reported a LOD score of greater than 2.0 for linkage of these traits to marker D6S503, which is in agreement with our findings. More recently Atwood et al. reported very strong evidence for linkage of BMI to marker D6S1009 at 137.7 cM on chro-mosome 6q in FHS families [13]. The linkage region in our multi-point trace does not include the marker reported by Atwood et al., though the single-point linkage results for chromosome 6 ( Table 2) include markers flanking D6S1009. The differences between our results and those of Atwood et al. could possibly reflect the methodological differences between our two studies. Atwood et al. performed their analysis using six different measurements of BMI taken over a span of 28 years from 1971 to 1988 [13]. We chose to perform our analysis using a single measurement between the ages of 40 and 50 for all subjects. In addition, Atwood et al. used a variance-components method for linkage analysis, which was different from the method used by our group.

Conclusions
We report evidence for nominally significant linkage of BMI to three regions on chromosomes 2, 3, and 11 by performing multi-point sib-pair single-trait regression analysis and calculating empirical p-values for all markers using the Framingham Heart Study data.