We performed genome scan linkage analyses for BMI measured at four different time points (over 16 years) and the mean of BMI over the four time points in a select sample from the Framingham Heart Study. The strongest evidence of linkage was observed on the short arm of chromosome 16 near centromere (16p11.2-12) at time point 1 (LOD = 3.0). Although this LOD score does not quite reach the genome-wide significant linkage level (3.3), we observed the same maximum peak with all other time points and the MEAN, with the lowest LOD score at time point 4 (LOD = 1.9). Using the same Framingham Heart Study data, Atwood and colleagues detected two linkage regions on chromosome 6 and 11 [3]. The maximum LOD score was 4.6 in the chromosome 6q23-25 region at time point 1. We observed a suggestive linkage with a LOD score of 2.9 at time point 1 in the same region. The obvious difference between our analyses and that of Atwood et al. is in sample selection. Atwood et al. used all available data at time point 1, while we selected a subset from the total sample, which has complete data at all four time points. By comparing the age (mean: 40.4 vs. 40.8) and BMI (mean: 25.5 vs. 26.0) distributions between Atwood et al.'s and our samples, there seems to be no obvious difference between the two samples. Therefore, the difference in the magnitude of LOD scores at 6q23-25 could be due to sample size difference. However, it is not clear why evidence for linkage at 16p11.2-12 was not identified in Atwood et al.'s analysis.
It is intriguing that the highest LOD score at 16p11.2-12 was observed at time point 1, while the lowest was at time point 4, and that three other suggestive linkages were all observed at time point 1. Since all subjects have data on all time points, the only difference between each time point is age and BMI values. Time point 1 captured a younger stage of the same population as compared with later time points. The observations that the maximum LODs decreased over time, and that other suggestive linkages were observed only at time point 1 suggest that these QTLs may have greater influence on BMI during the early years of life. Such variations of genetic influence with age may be further supported by a few available heritability studies of BMI. In a study of the National Heart, Lung, and Blood Institute male veteran twins, Fabsitz et al. estimated a heritability of 82% at age 20 and observed a somewhat reduced heritability after age 48 (72–78%) [4]. They also suggested that different sets of genes may be active at different age periods. Similarly, in the longitudinal Quebec Family Study [5], maximal heritability (2 × rsibling) at time 1 was higher (44%) than 12 years later (36%). These authors also suggested that some familial factors affect BMI consistently overtime and some additional familial factors affect BMI at different times. If such age-related heritability variations are true, and they are determined by different genes, it is conceivable that certain genes can only be identified during certain periods of life, for example suggestive linkage at 6q23-25 was only observed at time point 1. The consistent location of the maximum LOD score for BMI over time on 16p11.2-12 in our study indicates that this putative QTL has a relatively lasting effect on BMI, while such effect may be veiled by environmental factors over time. The chromosome 16 locus may be a true susceptibility locus for BMI, but this is the first report of this locus so an independent confirmation would increase its validity.
Using longitudinal observations of the same trait, BMI, we derived MEAN as a longitudinal phenotype that represents an overall status of BMI across 16 years. However, MEAN did not improve LOD score at either the previously reported 6q23-25 location or at our best peak on chromosome 16. One explanation could be that the effects of these QTLs were reduced at later years of life so that the composite measure, MEAN, from all age periods averaged out the effect of these QTLs. Although in this study, this longitudinal design did not provide more power to identify QTLs for BMI than a single cross-sectional data at time point 1, without longitudinal data, we would not be able to observe variations of genetic effect overtime. With such knowledge, in order to increase power to identify QTLs for BMI, we may design future linkage studies by ascertaining families with young adults. However, keep in mind that the effects of different QTLs and environmental factors may be different during a lifetime. When we restrict sample selection, we may gain power to identify certain genes but may not be able to identify others.
In summary, BMI varies with age, and different genes may determine variations in the population at different age periods. For example, genes that influence childhood obesity may be different from genes that influence adulthood obesity. Nevertheless, when we analyzed cross-sectional BMI data at different time points and the mean of all time points, we observed a consistent linkage with BMI at 16p11.2-12 across all time points, and three other suggestive linkages, including the previously reported 6q23-25. In addition, we observed variation in LOD score over time with the highest at time point 1 and the lowest at time point 4 (16 years later). These results indicate that there may be a QTL on chromosome 16 that contributes to BMI and this locus, and maybe others, is more likely to affect BMI during early adulthood.