The gain slope, gain mean and overall mean, were shown to be significantly heritable, with the phenotypes based on mean values exhibiting much larger heritability estimates than the gain slope. The most apparent gain from considering the nonlinear relationship in the definition of the BMI phenotype can be seen in the comparison of LOD scores for gain mean and overall mean. Namely, it appears that linkage analysis based on the gain mean phenotype provided us with possible chromosomal locations influencing an individual's tendency to be heavier-set, while the analysis using the overall mean phenotype (including both gain and decline phases) did not produce strong linkage evidence at these potential locations. The gain mean phenotype provided two regions with two-point LOD scores greater than 3 with no such regions for the overall mean phenotype. The proximity of the two elevated LODs for the adjacent markers on chromosome 9 provided additional evidence that this location is worthy of future study. Moreover, the use of the gain mean phenotype detected chromosomal locations that have already been implicated in previous studies using more direct measures of the obesity phenotype.
The Québec Family Study [8], a genome-wide scan, found nine QTLs affecting abdominal subcutaneous fat, two of which were on chromosomes 4 and 9. The region on chromosome 9 reported in Table 2 is the same region (D 9S 257 at 92 cM) reported in the Québec Family Study to influence abdominal subcutaneous fat.
A study of Pima Indians [3], who have a high prevalence of both type 2 diabetes and obesity, performed genome scans for loci linked to type 2 diabetes and obesity. Variance components linkage analyses were conducted on sibships. Phenotypic information came from the participants of their original longitudinal study, measuring the age at onset of type 2 diabetes. The mean age at onset of diabetes among affected offspring was 34 years (SD = 10.6) and the mean age at last examination of nondiabetic offspring was 35.5 years (SD = 11.1). The maximum BMI observed in the study period was used as an individual's BMI phenotype, and the largest associated LOD score was on chromosome 11 in the same region as that found in the current study (157xh6 at 131 cM).
Only weak linkage signals were observed in our study for the gain slope. No chromosomal locations linked to a slope phenotype were reported in the Human Obesity Map [1]. However, it is interesting to note that the signals observed in the current study were in the same general region on chromosome 4 and did increase in size in the multipoint analysis. This region is approximately 20 cM, calculated using the Marshfield Map [9], from the location reported for chromosome 4 in the Québec Family Study [7] (D 4S 2417), which has been suggested to contain a potential candidate gene.
The only linkage results for obesity reviewed in the Human Obesity Map [1] that corresponded to the regions found in the current analysis were those from the Québec Family Study [8] and the site on chromosome 11 found by Hanson et al. [3], who studied a relatively young sample that would, presumably, have not yet entered their decline phase. A measure such as abdominal subcutaneous fat used in the Québec Family Study is a more accurate measure of obesity. However, such a direct and invasive measure is only available in studies by design. Accounting for the nonlinearity of BMI by using the gain mean of BMI, when longitudinal data is available, seems to be a practical and simple alternative. The gain mean is not plagued by the fluctuations of a single cross-sectional measurement yet it is still easily calculated from the height and weight data that is often collected in studies with an alternative focus. The gain mean may also increase the power to detect linkage to obesity by removing the decline phase of individuals, which appears to be introducing competing sources of variation.
In cross-sectional study designs one might incorporate the results of this study by choosing to restrict BMI phenotypes on the basis of age; selecting the measurement for the BMI phenotype to be at ages less than the lower confidence limit for the average age at maximum BMI. The current study observed a large standard deviation for the age at maximum BMI. This may not reflect the true population standard deviation, because an individual's age at maximum BMI had to occur at a particular point in time (i.e., study visit). Additionally, the data available on some individuals did not consist of data in both the gain and decline phases. For example, if data were available for an individual only in their gain phase then their age at maximum BMI may have been underestimated. However, when choosing a BMI phenotype in a cross-sectional study, with the intent to omit decline phase measurements, the possibility of a large standard deviation in the age at maximum BMI should be considered.
It is unknown how prevalent the use of a slope phenotype is for studying obesity-related traits, because there has been little discussion of these phenotypes in the published literature and no slope phenotypes exhibiting linkage were reported in the Human Obesity Map [1]. The gain slope may provide a means to distinguish genetic components controlling the rate of weight gain for an individual by omitting the decline phase in the definition of this phenotype. The analysis using the gain slope phenotype did suggest a potential region for future study, although the evidence, as measured by the LOD score, was weak. Inclusion of the decline phase in the definition of this phenotype might have led some, who have attempted to use slope phenotypes in past studies, to overlook this potential region.