The mode of inheritance as revealed by complex segregation analysis (CSA) can provide evidence whether there is a major gene effect for Apo A1 concentrations, which is important for subsequent gene localization [11]. Even if some major susceptibility genes are already identified, CSA can help shed light on whether other genes exist [12]. So far studies of CSA on Apo A1 concentrations have been conducted mainly in different samples such as from hospital-based patients or community-based adults, with conflicting findings reported [7,8,13-18]. First, earlier CSA in families of probands who either had coronary heart diseases or underwent cardiac catheterization tended to support the existence of major gene effect for Apo A1 concentrations [15,16]. However, a recent study with a large sample in relatives of probands who underwent cardiac angiography failed to find evidence for the major gene effect [8]. Second, two studies in families of adult probands identified from the community found evidence for major gene effect for Apo A1 concentrations [7,17]. But in the HERITAGE Family study, Both Apo A1 at baseline (sedentary state) and its response the training, the major gene was not inferred due to the ambiguous transmission of the major effect from parents to offspring after they followed a supervised exercise training program for 20 weeks [18].

Proportion of Apo A1 variation in this sample explained by gender, age, age², BMI, smoking and drinking habits was only 7.3%. The correlation between residual Apo A1 concentrations among family members after adjustment for nonlipid covariates is displayed in Table ​Table2.2. The mother-father correlation coefficient was relatively high as compared with other familial correlations, indicating a strong environment effect and a weak genetic effect on Apo A1 levels. It is possible that there is a sex-specific influence in the variation of Apo A1 levels in the study subjects. The mother-daughter (0.18) and sister-sister (0.23) correlations were higher than father-son (0.11), brother-brother (0.00) or cross-sex correlations for the variation in Apo A1 levels.

Commingling analysis showed that a 4-component rather than a single, two or three-component distribution was the best-fit model for Apo A1 variance. The component means, variances, and proportions for the 4-component distribution model were (-0.741, 0.036, 1.142, 2.124), (0.167, 0.593, 4.813, 0.066), and (18.6%, 73.8%, 5.2%, 2.4%), respectively. The χ² for comparing the 4-component with 3-component distribution was 14.21 (df = 3, p = 0.003), while that for comparing the 4-component with 5-component distribution was 4.31 (df = 3, p = 0.230). The finding of multiple distributions is compatible with a major gene hypothesis; however, commingling may also arise through other causes. Thus, segregation analysis was used to determine whether these major effects segregated in families according to Mendelian expectations.

The results of complex segregation analysis of Apo A1 are presented in Table ​Table3.3. The sporadic model, familial correlations model, and Mendelian model were rejected due to a P value for each model's χ² of less than 0.001. The environmental model is the best-fit one. Under this model, factor A is pervasively present in the population (frequency = 0.97) and the mother-father correlation coefficient was 0.30. The mother-offspring correlation was higher than the father-offspring correlation (0.26 vs. 0.15). If the four familial correlations in the general model were reduced to two (ρMF = 0.28 ρMO = ρFO = ρSS = 0.18), the fit was not significantly worse (-2lnL = 6332.82, χ² = 5.50, df = 2, P = 0.064). The estimated heritability was 23.9%.

To rule out the possibility that our rejection of the major gene effect was due to numerical failure, sensitivity analysis of likelihood values was conducted for the Mendelian model. The results are plotted in Figure ​Figure1.1. Only three initial values can reach the global maximum, and the likelihood curves were irregular. The model with parameters estimated under this global maximum could still be rejected as it was compared with the general model.

Genetic heterogeneity was tested by plotting -2ln(LE/LM), in which LE refers to the likelihood of the environmental model and LM to that of the Mendelian model (Figure ​(Figure2).2). One outlier (i.e., a value of 25.52 for -2ln(LE/LM)) was found favoring the Mendelian model. A repeated segregation analysis after excluding this outlier demonstrated that the environmental model was still the best-fit model.

Two main features of this study are worth noting. First, there is considerable homogeneity in this study population. Most of the subjects live in the same community, hence their social and living environments tended to be more similar than those in different communities. Second, the results are particularly relevant for a population at low risk for atherosclerosis, since the probands were randomly selected from adolescents in the community.

Our results were consistent with those of previous familial correlation and commingling studies [19], which demonstrated that more than one component is needed to explain the distribution of lipids. Previous estimates for the heritability of Apo A1 were 0.2 to 0.3 [20,21]. The high mother-father correlation found in this study indicated that there might be assortative mating and/or common household factors affecting Apo A1 concentrations. We found there were 6.04% parents who knew their hyperlipidemia disease, and less than 1% had history of coronary heart disease. There were only 1.65% parents taking regular hypolipidemic drugs. We think the effects of lipid-lowering drugs were minimal for parent's lipid profiles. We found that the correlation between mother and offspring over Apo A1 was greater than the correlation between father and offspring. Similar results were reported in HDL cholesterol concentrations [22]. Similarity in patterns of lifestyle and physical activity among family members may explain the effects of assortative mating and common household effects [23].

The mode of inheritance of Apo A1 concentrations in this study was best explained by a model of mixed environmental effect and familial correlation. No major gene effect was found. This is similar to the finding of a recent study [18] but contrary to the findings from two studies in families of adult probands identified from the community [7,17]. Several reasons may account for the lack of major gene effect in this study. First, it may reflect the complexity of Apo A1 metabolism. Although Apo A1 is the direct product of a single gene, it is distributed across the full range of HDL particles and other lipoproteins as well. Therefore, Apo A1 concentrations are likely to be influenced by many genetic as well as environmental factors. Hence, this renders the detection of any singular effect of a major gene difficult. Intriguingly, CSA of HDL cholesterol, one phenotype strongly related to Apo A1, also revealed that the environmental model was the best-fit model [24]. Second, there might still be residual confounding on Apo A1 levels from covariates that were not controlled for in this study, such as physical activity and hormone replacement. Third, there were no extreme Apo A1 levels found in this population. Thus, for variation of Apo A1 within the normal range, there might be no major gene effect of considerable magnitude. Finally, the different study population characteristics were one possible reason for incongruent results on the mode of inheritance studies of HDL cholesterol and Apo A1 levels, and this study was presented as subjects with normal range of Apo A1 levels. Also, it is important to consider the inverse relationship between triglyceride and HDL and Apo A1 levels. The rationale for not including triglyceride or HDL in the model was as follows. The genomic profiles controlling Apo A1 and triglyceride or HDL might be overlapped. Adjusting triglyceride or HDL will eliminate the genetic proportions of controlling HDL and triglyceride, and a part of Apo A1 levels. It was contradictory to our primary hypothesis to explore the genetic components of Apo A1 levels. Therefore, the genetic results may be confounded by triglyceride or HDL components, and did not separate the background of specific traits.

Another possible reason for our failure to detect a major gene effect for Apo A1 is the young age of our subjects. Age plays an important role for genetic control on phenotype expression, and environmental factors add complication in lipid traits. A recent study on the effect of quantitative trait loci for lipid phenotypes in the rat indicated that genetic components become important factors to the control of phenotype expression as age increases [25]. Among human, Apo A1 was also reported to have intergenerational differences in heritability [26]. However, lack of a major gene in the study families does not exclude the possibility that there are major genes that will influence Apo A1 levels in late adulthood. For example, there have been reports suggesting pleiotropy affecting lipid-related traits and obesity. Substantial evidence for quantitative trait loci with pleiotropic effects influencing BMI and HDL were found in the Framingham Heart Study [40]. We conducted analyses with BMI as a covariate. If pleiotropy exists affecting body weight and lipid-related traits, adjusting for BMI would likely hide the genetic effect on ApoA1. This could explain why our study did not identify a major gene contribution. Finally, we have collected several items about food intakes among the family members, such as vegetarian, meats, and rice amounts. The high spouse correlation coefficient was not associated with above variables. It is the limitation of our study to explain the possible common household effects on Apo A1 levels.

Segregation analysis is typically very sensitive to ascertainment, and false assumptions on ascertainment could invalidate the estimates obtained through segregation analysis (11). Highly selected samples obtained through a phenotype-based ascertainment scheme may lead to biased estimates. Therefore, the ascertainment scheme is very important on segregation analysis. Although we did not ascertain probands on ApoA1 trait, the likelihoods corrected for ascertainment on probands' ApoA1 levels provided efficient estimation. The results of likelihood without correction of proband ascertainment were similar to corrected likelihood results.

Complex segregation analysis was used to ask whether an inheritable trait is controlled by a single major gene plus residual polygenes, if so, gene mapping tasks would be warranted. If not, gene mapping studies would be lack of power because traditional statistical programs and sparse genetic markers would not provide sufficient tools to tackle a trait without major gene effects. However, this situation has changed in the past few years. Now we know a complex trait is less likely to be influenced by a single major gene. Advanced statistical programs, dense genetic markers plus other new technologies has allowed us to investigate a complex trait where genes with small to moderate effects. In the post-genomics era, segregation analysis is considered as an intermediate tool to help investigators to plan further sophisticated genomic studies. Although the method did not give information of exact DNA locus to find the genes, it can provide heritability estimates and parameters for further parametric linkage analysis. So segregation analysis should be useful and not obsolete. Further genotype-based methods such as linkage and association will be planned for elucidating genetic effects and locations.