|
|
| |||||||||||||||||||||||||||||||
Background The extent to which serum lipid levels are affected by genetic and environmental factors remains a point of controversy. We examined both genetic and environmental influences on serum lipid levels in twins reared either together or apart who participated in the Swedish Adoption/Twin Study of Aging.
Methods We studied 302 pairs of twins (mean age, 65.6 years; range, 52 to 86); 146 pairs had been reared apart. We simultaneously compared the twins on the basis of both zygosity and rearing status, which allowed joint estimation of genetic and environmental influences on serum lipid levels. Genetic influence was expressed in terms of heritability, the proportion of the population variation attributable to genetic variation (a value of 1.0 indicates that all of the population variation is attributable to genetic variation). The serum lipids and apolipoproteins measured included total cholesterol, high-density lipoprotein cholesterol, apolipoproteins A-I and B, and triglycerides.
Results Structural-equation analyses revealed substantial heritability for the serum levels of each lipid measured, ranging from 0.28 to 0.78. Comparisons of the twins reared together with those reared apart suggested that the environment of rearing had a substantial impact on the level of total cholesterol (accounting for 0.15 to 0.36 of the total variance). Sharing the same environment appeared to affect the other lipid measures much less, however, than did genetic factors and unique environmental factors not shared by twins. Comparisons of younger with older twins suggested that heritability for apolipoprotein B and triglyceride levels decreased with age.
Conclusions The effect of genetic factors on the serum levels of some but not all lipids appears to decrease with age. Early rearing environment appears to remain an important factor in relation to levels of total cholesterol later in life, but it has less effect on other serum lipids and apolipoproteins in the elderly.
Although these studies of specific defects have a profound implication for persons with monogenic familial syndromes, they explain only a minor portion of the population variation in serum lipid levels and rates of coronary heart disease. Some research suggests that monogenic disorders are present in 20 to 30 percent of survivors of myocardial infarction who are less than 50 years old, and perhaps in only 7 percent of older survivors4,8,9. Since monogenic familial syndromes apparently account for a relatively small percentage of cases of coronary heart disease, it is relevant to identify other genetic factors. Twin and family studies have addressed this issue through quantitative genetic approaches. The importance of the influence of genetic factors on a trait can be expressed as heritability, defined as the proportion of population variation attributable to genetic variation10.
A common criticism of twin and family studies is that they may overestimate heritability by confounding genetic and environmental factors, in that relatives living together share environments as well as genes. A more ideal design is to combine a twin study with an adoption study, by comparing identical and fraternal twins reared together with twins reared apart. The Swedish Adoption/Twin Study of Aging (SATSA) has incorporated such a design11. The principal aim of that study is to examine individual differences in aging, providing a unique opportunity to examine age differences in heritability for serum lipids. Although some epidemiologic studies have reported on lipid levels in the elderly,12,13,14,15 most genetic studies have focused on younger people. This report presents new information on genetic and environmental influences on lipid levels in an older population. The examination of the elderly is timely, given current discussion of the role of elevated cholesterol levels as a continuing risk factor for coronary heart disease in this age group13,14.
Methods
Study Sample
The SATSA sample of twins separated early in life was identified through the Swedish Twin Registry, which includes nearly 25,000 pairs of twins of the same sex born in Sweden between 1886 and 195816. The SATSA subregistry was compiled in 1984 by contacting pairs of twins identified in the registry as having been reared apart, along with matched pairs of twins who had been reared together. The SATSA sample has been described in detail elsewhere17.
The data on serum lipids reported here were recorded in the subgroup of 302 pairs of twins who were given physical examinations during testing between 1986 and 198811. Informed consent was obtained from all subjects. When the subjects were grouped according to zygosity (confirmed by the identification of serologic markers11) and rearing status (twins reared together vs. twins reared apart), the sample consisted of 46 pairs of monozygotic twins reared apart, 67 monozygotic pairs reared together, 100 dizygotic pairs reared apart, and 89 dizygotic pairs reared together.
Serum Lipids and Apolipoproteins
The subjects were requested to fast for 12 hours before testing. During an interview accompanying physical examination, they were asked about all medications used in the past month. Of the 604 older adults in this study, none reported taking drugs prescribed specifically to lower lipid levels.
Blood samples were frozen at -70 °C, transported in dry ice, and thawed and analyzed when received by the laboratory. Serum cholesterol and triglycerides were measured with an enzymatic calorimetric assay (Boehringer-Mannheim automated analysis for Hitachi systems 717, Diagnostica, Mannheim, Germany). High-density lipoprotein (HDL) cholesterol was measured after lipoproteins containing apolipoprotein B were precipitated with phosphotungstate-magnesium chloride18. Apolipoproteins A-I and B were measured with commercial radioimmunoassay kits (RIA 100, Pharmacia Diagnostics, Uppsala, Sweden). The intraassay coefficients of variation for cholesterol, HDL, triglycerides, apolipoprotein A-I, and apolipoprotein B were 1.1, 2.5, 2.7, 2.3, and 2.2 percent, respectively; the respective interassay coefficients of variation were 1.1, 3.7, 3.3, 4.8, and 4.9 percent. Because the distributions of values for triglycerides and the two apolipoproteins were skewed, these values were logarithmically transformed before statistical analysis.
Lipid Analyses
Analyses of the lipid measurements included descriptive statistics and genetic analyses. The effects of age and sex on lipid levels were explored by multiple regression19 predicting lipid levels from sex, age (in years), and an interaction term for sex and age. Intraclass correlations and model-fitting analyses were used to evaluate genetic and environmental influences. Since estimates of heritability in twins of the same sex may be biased unless the effects of age and sex on lipid levels are controlled for,20 multiple regression was used to assess the linear effects of age and sex on lipid levels; the residuals (differences between the observed and the predicted scores) were then used in subsequent comparisons of intraclass correlations. Similarly, because the levels of lipids and lipoproteins can be affected by a variety of medications,21,22 the linear effects of estrogens, beta-blockers, thiazide diuretics, and loop diuretics were removed in a prior multiple regression analysis.
Intraclass Correlations
Our genetic analyses were based on quantitative genetic theory, which defines a phenotype as the sum of the effects of both a genotype and an environment10. Our goal was to separate the phenotypic variance that we observed into its genetic and environmental components by comparing the similarity between the twins of each pair (measured by intraclass correlations) according to their zygosity and type of rearing. For example, monozygotic twins share identical genotypes, so any differences between them are theoretically due to their environments. Dizygotic twins, in contrast, are no more alike genetically than siblings, sharing on average 50 percent of their segregating genes. The extent to which monozygotic twins are more alike than dizygotic twins should therefore reflect genetic influences. In the classic twin method, the difference between intraclass correlations for monozygotic twins and those for dizygotic twins is doubled to estimate heritability. The remaining population variation can then be attributed to environmental factors.
Studying twins reared apart is a powerful method for evaluating not only genetic influences but also different types of environmental effects. The influence of a shared rearing environment is estimated by comparing the intraclass correlations for twins reared together with those for twins reared apart: if twins reared together are more similar than twins reared apart, this finding demonstrates the importance of a shared rearing environment. However, even twins reared apart may share similar, or "correlated," environments, perhaps because of selective placement (placing adopted children in homes that resemble those of their biologic parents). Other factors producing a correlated environment, for both twins reared together and apart, may include prenatal influence, post-rearing contact, or similarities in aspects of adult lifestyle such as dietary habits. The effects of correlated environment become apparent when the similarity of a pair of twins cannot be explained by either zygosity or the type of rearing. Unlike traditional studies of twins reared together, the study of twins reared apart allows separate quantification of the effects of a shared rearing environment and those of a correlated environment.
In addition to shared rearing and correlated environments, which lead twins to resemble each other, other environmental effects occur throughout life that are not shared by twins. After the effects of zygosity and shared environments have been accounted for, dissimilarity between twins can be attributed to such unique, or nonshared, environmental effects.
Model-Fitting Analyses
Estimates of genetic and environmental effects based on comparisons of intraclass correlations have low power and large standard errors and do not simultaneously use all available information. Model-fitting approaches are more powerful and permit analysis of groups of twins simultaneously, making assumptions explicit and testing the relative fit of different models. In our study, variances and covariances among twins were subjected to structural-equation modeling with the LISREL 7 program23 to estimate the genetic and environmental components of variance. The use of structural-equation models has become standard in twin research,24 and the application of these techniques to SATSA has been described previously25,26,27. The models, which are based on the expectations of quantitative genetics about factors contributing to similarities and differences among twins, are explained further in the Appendix.
Age and Sex Differences in Genetic Factors
To evaluate age differences, the SATSA sample was divided into two age groups at the median age. This dichotomy, chosen to maximize sample size and statistical power for hypothesis testing, separated subjects 65.4 years of age or older in January 1987 (the midpoint of the testing period) from the other subjects. Intraclass correlations and the results of model-fitting analyses were then compared according to age group and sex. Age (in years) was also included as a covariate within models, to obtain estimates for genetic and environmental factors that are independent of linear effects of age within cohorts (described by Neale and Martin28).
To examine further the linearity of age effects without arbitrary categorization of the sample into cohorts, hierarchical multiple regression analyses were also performed in which the lipid levels of one twin were predicted from the levels of the other twin, as well as by terms relating to zygosity and age. The results of such analyses indicate whether the similarity of twins is affected by zygosity (i.e., whether there is evidence of significant genetic effects) and whether genetic effects vary linearly as a function of age. Hierarchical multiple regression analyses, which have been used extensively in twin studies,29,30 complement the model-fitting analyses and are described in further detail in the Appendix.
Results
Descriptive (Nongenetic) Analyses
In the study sample as a whole, cholesterol levels ranged from 139.0 to 413.0 mg per deciliter (3.6 to 10.7 mmol per liter), with an overall mean of 262.7 (6.8 mmol per liter). The mean HDL level was 57.0 mg per deciliter (1.5 mmol per liter) (range, 19.3 to 104.2 mg per deciliter [0.5 to 2.7 mmol per liter]); the apolipoprotein A-I level, 1.4 g per liter (range, 0.8 to 2.2); the apolipoprotein B level, 1.1 g per liter (range, 0.6 to 2.2); and the triglyceride level, 144.0 mg per deciliter (1.62 mmol per liter) (range, 40.7 to 476.1 mg per deciliter [0.5 to 5.4 mmol per liter]). Table 1 shows the mean lipid levels according to zygosity, rearing status, and sex; the values for total cholesterol, HDL, and apolipoprotein A-I suggest that differences in these concentrations were sex-related.
|
|
Intraclass correlations for subgroups of the study sample are shown in Table 3. In general, correlations for the monozygotic twins were higher than those for the dizygotic twins, indicating genetic influence. The importance of the influence of the rearing environment on levels of total cholesterol and apolipoprotein B was suggested by consistently higher correlations for the twins reared together than for those reared apart.
|
Comparisons of correlations obtained with and without control for the use of estrogen and hypotensive drugs revealed virtually no differences. Although therapy with estrogen and hypotensive drugs may greatly affect serum lipid levels, these agents had little influence on the similarity between twins in serum levels of lipids and apolipoproteins.
Model-Fitting Analyses
Table 4 summarizes the estimates for the components of variance that were obtained with LISREL modeling, which suggested that there were age differences in heritability. Heritability in the younger group ranged from 0.63 for total cholesterol to 0.78 for apolipoprotein B; in the older group, it was lowest for triglycerides and total cholesterol -- 0.28 and 0.32, respectively -- and about 0.5 for apolipoprotein B, apolipoprotein A-I, and HDL. The most striking difference between the age groups occurred in triglyceride levels. For total cholesterol, HDL, and apolipoprotein A-I, models could be fitted that constrained equal estimates across the two age groups. Constrained models could not be fitted for apolipoprotein B and triglycerides, suggesting significant age differences in heritability for these variables.
|
Differences in the environmental components of variance were also evident. For total cholesterol, a shared rearing environment appeared to be more important among the older twins than the younger twins, and more important among women than men. For apolipoproteins A-I and B, a shared rearing environment was important among the older twins but not among the younger twins. A correlated environment, which may be due to selective placement, prenatal influences, or post-rearing contact, was found to be important for only one variable -- HDL in the younger twins. However, it only approached significance; a model without correlated environment is shown in Table 4.
Hierarchical Multiple Regression
Hierarchical multiple regression analyses to predict a lipid value in one twin from the value in the other twin and interaction terms for zygosity and age (see the Appendix) indicated that heritability varied linearly with age for total cholesterol (F = 3.92, 281 df; P = 0.045) and apolipoprotein B (F = 9.60, 287 df; P = 0.002), but not for triglycerides, HDL, or apolipoprotein A-I. The significant linear effect of age on heritability for cholesterol was not found in model-fitting analyses that examined the sample as two age groups.
Discussion
The mean levels of total cholesterol in this sample of elderly twins were relatively high according to general clinical guidelines and in comparison to levels in American subjects2. However, they are consistent with normative data from Sweden and other Northern European countries31. A substantial genetic influence was observed for each lipid across sex and age groups, in keeping with conventional twin and family studies. Estimates of heritability in the younger twins, whose ages most closely approximated those of subjects studied by other investigators, were similar to estimates reported for total cholesterol and HDL in other twin studies (Table 5). In this age group, heritability was highest for apolipoprotein B (0.78), a result similar to that found in a sample of Japanese twins 50 to 74 years old46. The role of genetics in population variability of serum triglyceride levels has been less clear than its role in the variability of serum cholesterol levels. Our estimate of 0.72 for heritability in the younger twins suggests a strong role for genetic influence on variability of triglyceride levels in this population. Reports to date on apolipoprotein A-I have also been conflicting. Some studies47,48 have found no heritability for apolipoprotein A-I, whereas others39,46,49 have reported heritability in the range of 0.42 to 0.66; the latter value is similar to that of 0.69 for our younger twins.
|
Age and Sex Differences in Genetic Factors
A unique feature of this study is the age of the twins; the mean age at testing was 65.6 years. Dramatic differences were found between the younger and older groups in heritability for apolipoprotein B and triglycerides. This apparent decrease with age in the importance of genetic factors was not observed for HDL and apolipoprotein A-I, for which models could be fitted that constrained the effects in the two age groups to be equal. Evaluation of the estimates for cholesterol, as well as the hierarchical multiple regression analyses, suggests that there are age differences in heritability, although model-fitting analyses do not support this conclusion. These contradictory findings may reflect a lack of statistical power to find age differences when a sample is divided into two age groups. The effect of age on heritability for serum lipids and apolipoproteins has not been examined before, although in two early studies, Osborne et al.50 found no effect of age on the similarity of total cholesterol levels in twins 18 to 55 years old and McDonough et al.51 found that differences between twins in cholesterol levels increased with age.
The apparent age differences in heritability for apolipoprotein B and triglycerides may reflect an increase in the importance of accumulated experiences unique to individuals. A decrease in the importance of genetic factors with age may also reflect a survivor effect, in that persons predisposed to genetically mediated lipid disorders may die at earlier ages, and are thus differentially represented in the two age groups. Alternatively, these results may reflect cohort differences.
Our results indicate that there are no sex differences in the relative importance of genetic factors for any of the lipids studied except apolipoprotein B. As discussed below, genetic effects, as well as the effects of shared rearing, emerged as a more important influence on apolipoprotein B levels in women than in men.
The Role of a Shared Environment
The variable of shared environment in our study differs from that used in conventional twin and family studies,34,36 because studying twins reared apart allowed us to focus on early rearing experiences as a cause of variation. In studies of twins or other family members reared together, the effects of a shared rearing environment cannot be reliably differentiated from those of other factors, such as prenatal and post-rearing effects. The inclusion of twins reared apart allowed us to distinguish these effects as our variables of shared rearing environment and correlated environment. The results of model fitting indicated little evidence of an effect of correlated environment, in contrast to a shared rearing environment, which had a significant effect on cholesterol levels in each sex and age group. A shared rearing environment also significantly influenced levels of apolipoprotein A-I and B in the older twins and apolipoprotein B levels in women.
Our findings of an influential role for a shared rearing environment emphasize the importance of early rearing experiences in determining variation in total cholesterol levels much later in life. Whitfield and Martin36 discussed sex differences in the effect of a shared environment on serum cholesterol levels and suggested that such differences may reflect a greater retention of family dietary or cooking habits by women than by men. The difference between the age groups in the effect of a shared rearing environment on apolipoprotein B levels may reflect a similar cohort difference in the retention of family dietary or lifestyle habits as well.
Our results suggest an apparent reduction with age in the influence of genetic factors on some but not all serum lipids. The early rearing environment appears to remain important in influencing total cholesterol levels later in life but seems less important in influencing other serum lipid and apolipoprotein levels in our elderly sample.
Supported by grants from the National Institute on Aging (AG-04563), the MacArthur Foundation Research Network on Successful Aging, the Swedish Medical Research Council (09533), and the Swedish Lung and Heart Foundation.
Source Information
From the Program in Biobehavioral Health (D.A.H., G.E.M.) and the Center for Developmental and Health Genetics (D.A.H., N.L.P., G.E.M.), College of Health and Human Development, Pennsylvania State University, University Park; the Division of Cardiovascular Medicine, Department of Medicine, Karolinska Hospital, Stockholm, Sweden (U.F.); the Department of Epidemiology, Institute for Environmental Medicine, Karolinska Institute, Stockholm, Sweden (N.L.P.); and the Department of Clinical Chemistry, Umea University Hospital, Umea, Sweden (G.D.).
Address reprint requests to Dr. Heller at the Program in Biobehavioral Health, 210 E. Health and Human Development, Pennsylvania State University, University Park, PA 16802.
References
Model-Fitting Analyses
The models in this study are based on the following equations, which describe similarity between twins in terms of the contributions to variance of additive genetic effects (heritability), shared rearing environment, and correlated environment (nonshared environmental effects contribute only to dissimilarity between twins and are therefore not included in the equations): (1) covariance of monozygotic twins reared apart = heritability + correlated environment; (2) covariance of monozygotic twins reared together = heritability + shared environment + correlated environment; (3) covariance of dizygotic twins reared apart = 1/2 heritability + correlated environment (because dizygotic twins share on average only 50 percent of their segregating genes); and (4) covariance of dizygotic twins reared together = 1/2 heritability + shared environment + correlated environment.
A maximum-likelihood procedure with LISREL 7 was used to fit the expected covariances for twins, based on the above equations, to the observed covariances23. The application of model-fitting analyses to data on twins has been discussed fully by Jinks and Fulker52 and Boomsma et al24. Model-fitting procedures in LISREL 7 yield parameter estimates, standard errors, and chi-square values, which indicate how well the data fit the model.
For each lipid, a series of models was tested; all variables (heritability, shared environment, correlated environment, and nonshared environment) were included in the initial model but were excluded individually or collectively in subsequent models. A significant increase in the chi-square value after a variable has been excluded indicates that the reduced model fits the data less well than the full model. The model selected is the most parsimonious -- i.e., it contains the fewest variables but still fits the data well. The parameter estimates provided by LISREL 7 correspond to partial regression coefficients and, when squared, represent components of variance. The squared coefficients, then, are used to calculate the proportions of phenotypic variance due to genetic and environmental factors.
For these genetic analyses, cohort and sex differences in genetic and environmental factors were evaluated by comparing models in which parameter estimates were constrained to be equal across subjects in an age or sex group with models in which the estimates were allowed to differ. The relative fits of these constrained and unconstrained models were then evaluated by chi-square criteria.
Hierarchical Multiple Regression
As an adjunct to the model-fitting analyses that compared results across broadly defined age groups, hierarchical multiple regression analyses were also performed to evaluate the linear effects of age on similarity of twins and heritability. These analyses assess whether similarity between twins differs as a function of other variables40. The basic model may be summarized as follows:
Y = B1X1 (twin's score) + B2X2 (age) + B3X3 (zygosity) + B4X1X2 (twin's score x age) + B5X1X3 (twin's score x zygosity) + B6X2X3 (age x zygosity) + B7X1X2X3 (twin's score x age x zygosity) + constant,
where Y denotes one twin's predicted score (e.g., lipid level), X1 the other twin's score, X2 the covariate age, and X3 zygosity coded as a dichotomous variable (e.g., 0 = a dizygotic twin and 1 = a monozygotic twin). B1 through B7 are the partial regression coefficients corresponding to the main effects and the two-way and three-way interaction terms.
The main effects and interaction effects are tested sequentially for significance. A significant effect for B1X1 indicates significant similarity between twins; a significant effect for B2X2 indicates the simple effect of age on lipid levels. A significant interaction effect for B4X1X2 indicates that similarity between twins varies as a function of age, but does not distinguish between the genetic and environmental components of similarity.
The two-way interaction B5X1X3, if significant, indicates that similarity between twins varies as a function of zygosity (i.e., there are significant genetic effects). The three-way interaction B7X1X2X3, if significant, indicates that the effect of zygosity on similarity varies as a function of age (i.e., heritability varies linearly with age).
Model-fitting analyses and hierarchical multiple regression analyses should be regarded as complementary methods. For example, a linear effect of age on heritability may be detected by hierarchical multiple regression, but this analysis does not provide the discrete estimates of heritability that can be obtained and compared by model fitting across age groups. Conversely, model-fitting analyses of defined age groups cannot assess the linearity of age effects on heritability as thoroughly as hierarchical multiple regression analyses.
| |||||||||||||||||||||||||||||||
This article has been cited by other articles:
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | TERMS OF USE | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2009 Massachusetts Medical Society. All rights reserved. |
Previous
