Effects of Heritability, Shared Environment, and Nonshared Intrauterine Conditions on Child and Adolescent BMI




Heritability studies of BMI, based upon twin samples, have identified genetic and shared environmental components of BMI, but have been largely silent about the nonshared environmental factors. Intrauterine factors have been identified as having significant long-term effects on BMI and may be a critical source of nonshared environmental influence. Extant studies based on samples of either unrelated individuals or twins cannot separate the effects of genetics, shared environments, and nonshared intrauterine conditions because the one lacks variation in the degree of relatedness and the other has insufficient variation in intrauterine conditions. This study improves upon these prior studies by using a large, sibling-based sample to examine heritability, shared environmental, and nonshared intrauterine influences on BMI during two age periods in childhood (6–8 years; 12–14 years). The primary interest was in determining the effects of the intrauterine environment on BMI as a component of the nonshared environment and in determining whether there were sex-specific differences in heritability and/or in the intrauterine factors. These were estimated using regression-based techniques introduced by DeFries and Fulker. Heritability of BMI was estimated to be 0.20–0.28 at 6–8 years and 0.46–0.61 at 12–14 years. Differences in heritability were found at 12–14 years between same-sex as compared to mixed-sex pairs. The shared environmental effect was significant at 6–8 years but insignificant at 12–14 years. Differences in birth weight were significant in all groups at 6–8 years suggesting long-term effects of the nonshared intrauterine environment; at 12–14 years, birth weight was no longer significant for girls.


There has been a dramatic increase in childhood and adolescent obesity in the past 20 years (1). Because of the rapidity of this increase, most of the explanatory focus has been on changes in the environment, rather than genetic changes. This, however, does not diminish the large body of evidence that finds variation in BMI largely due to heritable genetic differences (2). Heritability is the proportion of phenotypic differences among individuals that can be attributed to genetic differences (3). The current thinking is that although genetic changes per se are not the “cause” of this rise, it is probably a result of a genetic susceptibility that is manifested in an obesogenic environment (4). Because obesity is known to be a complex disease influenced by multiple genetic and environmental factors, a thorough understanding of these determinants requires studies that can tease apart the genetic from the environmental influences on obesity (5).

Sex differences in genetic influences on BMI have been found in children, but results have been mixed and may be related to the child's developmental stage. A few studies indicate heritability to be greater in females (6), others find it to be higher in males (7), while others report no difference (8). Twin studies that include opposite-sex dizygotic pairs find a lower correlation between BMI for opposite-sex dizygotic pairs than same-sex pairs, suggesting sex differences in heritability and/or environmental influences (9).

Estimates of heritability for BMI are primarily based on twin studies, although there is also evidence from general sibling studies (10). Twin studies suggest that genetic factors explained 35–90% of the variance in BMI during childhood and adolescence (11). Heritability estimates of BMI based on general sibling pairs have been reported in the range of 25–85% (10,12). In general, the estimates of heritability are higher in twin samples than in general sibling samples because of both increased shared genetic material, as well as a more common shared environment.

Common environmental influences reflect the proportion of the variance in BMI attributable to shared lifestyles between children growing up in the same family such as eating habits and lack of exercise. Twin studies report estimates of common environmental influences for children aged 4 and 5 years of 25% for a pooled sample of boys and girls (13) and gender-specific estimates of 44% for boys and 12% for girls (14). Estimates for children at age 9 years range from 10% for twin samples (15) to 25% for a sample of twins plus same-age adoptees raised in the same household (16). In one longitudinal study, estimates of common environmental influences fell from 15 to 25% for twins during early adolescence (11–14 years of age) to insignificance by age 17 years (9).

Nonshared environmental factors are thought to be etiologically important (17), accounting for ∼20% of the variation in some studies (18,19). These factors are for the most part unknown (20), though recent studies have suggested that intrauterine factors may be a critical source of nonshared environmental influence (21,22). This may be particularly true for BMI as recent studies have demonstrated that growth in utero, maternal glucocorticosteroid production, maternal nutrition, and maternal behaviors (e.g., smoking (4)) may have significant and long-term effects on a person's weight over their lifetime (23). Moreover, this environment may have different effects by sex. Birth weight is a well-recognized summary marker of intrauterine processes (24), as well as an independent risk factor for later obesity (4). It has been difficult to examine these intrauterine factors separately from the genetic and shared environmental factors as most obesity and early life studies have been done using samples of unrelated individuals (25). But heritability studies limited to twins cannot examine nonshared effects of the intrauterine environment because twins share similar intrauterine environments. Only heritability studies using nontwin siblings can address these issues.

In this study, we are interested in examining the effects of the intrauterine environment on BMI as a component of the nonshare environment in a large US-based general sibling cohort. In addition, we are interested in determining whether there are sex-specific differences in heritability and/or in intrauterine factors to further clarify possible processes associated with BMI during two childhood periods: 6–8 years and 12–14 years.

Methods and Procedures


This study used data from the US-based National Longitudinal Survey of Youth's Child–Mother file. It was based on the National Longitudinal Survey of Youth 1979, which began in 1979 as a household probability sample with all 14- to 21-year-old youths in selected households. This survey includes a representative cross-section of the civilian population living in the United States in December 1978, and an oversample of African Americans and Hispanics. Beginning in 1986, data were collected biennially on biological children of these women. As of 2006, the latest round of survey data available, 9,294 children had been born to these women (26). The design of the child component has resulted in a sample with multiple kinship ties, including cousins, half siblings, full siblings, and monozygotic twins. Rodgers, Johnson, and Bard have created relationship codes for the children born through 2000 using an algorithm that assigns genetic relatedness to 98.8% of child-pairs (27). This procedure has been validated using child heights.

The current study was based on two samples of children: an early school age-group born between 1978 and 2000 (n = 7,466) who were ages 6–8 years and an early adolescent group, born between 1972 and 1994 (n = 7,126) who were 12–14 years of age. The early school age study sample included 5,453 children (73% of the eligible sample) with 4,915 unique relationship pairs (1,214 boys, 1,189 girls, and 2,512 mixed sex). The adolescent study sample included 4,994 children (70% of the eligible sample) with 4,506 unique relationship pairs (1,156 boys, 1,088 girls, and 2,262 mixed sex). There were 2,845 children in both samples.


Anthropomorphic measures. Mothers of the children were given the option to report their children's heights and weights or have the interviewer complete these measurements. Approximately 74% of all the heights were measured; 69% of the weights were measured. Measures were taken with clothes, but without shoes. Interviewers were trained to conduct these measurements by the National Opinion Research Center at the University of Chicago. BMI-for-age percentile (BMIpct) and BMI-for-age z-scores (BMIz) were calculated using the CDC (Centers for Disease Control and Prevention) SAS program (CDC, Atlanta, GA) designed for this purpose. An advantage of this program is that it includes cut-points for biologically implausible values for heights and weights by age and sex, providing a reliable mechanism for data cleaning (28). The study sample includes only children with biologically plausible values and BMIz scores between −3 and 3.

Dependent measure. The BMIz scores were adjusted for gender, year of birth, an indicator for whether height was measured, and an indicator for whether weight was measured using linear regression. Based on separate regressions for school-age and adolescent children, the residualized BMIz was used as the dependent measure in the analyses. This approach was adopted by Jacobson and Rowe (12).

Sibling pairs. Sibling groups were formed based on the sex of the child, with girl–girl, boy–boy, and girl–boy (mixed) pairings.

Genetic relatedness for additive genetic influences. This study assigns codes for additive genetic influences as follows: monozygotic twins (R = 1), full siblings (R = 0.5), half siblings (R = 0.25), and sibling pairs who could not be classified with confidence as either full or half siblings (R = 0.375). We could not reject the hypothesis that the correlation between residualized BMIz scores of unclassified sibling pairs was equal to the correlation for half siblings, but rejected the hypothesis that the correlation equaled that for full siblings. Therefore, the unclassified sibling pairs were reassigned the relatedness code for half siblings (R = 0.25).Genetic relatedness for nonadditive genetic influences. This study assigns codes representing nonadditive genetic variance for dominance as follows: monozygotic twins (R2 = 1), full siblings (R2 = 0.25), and half siblings (R2 = 0). Unlike monozygotic twins, full siblings share almost no nonadditive effects due to epistasis (29).Measures of the intrauterine environment. Birth weight, recorded in ounces, was reported by mothers during the first interview following the birth of the child. Approximately 9% of the sample had missing data on birth weight. Missing observations were imputed based on gender, maternal race/ethnicity, education, and age. Mothers also reported on smoking during pregnancy, prepregnancy weight, and height. Height and weight reports were combined to create a variable for prepregnancy BMI.

Additional controls. Indicators for whether height and weight were measured or mother-reported and year of birth.


DeFries and Fulker (30) introduced a multiple regression model for analyzing kinship data that accounts for both additive genetic and common environmental influences. A number of authors have extended the basic DF framework to account for nonshared environmental influences (17,31,32), which in this study were measures of the intrauterine environment. This model is referred to as the ACE model and can be expressed as


where T1, is the trait score for the first member of the kinship pair, T2, is the trait score for the second member of the pair, R is the coefficient of genetic relatedness (R = 1 for monozygotic twins, R = 0.5 for full siblings, etc.), (NS1NS2) is the transpose of a vector of the pair-difference between elements of the nonshared environment, and ε is the error term. E1) = c2 is an estimate of the common environmental influences within the population, E3) = h2 is an estimate of trait heritability within the population, and the vector E(β) is an estimate of a vector of the effect of differences in the nonshared environment.DF analysis was initially developed to study twin data with two levels of genetic relatedness. Its underlying assumptions include trivial assortative mating, equal shared environmental influences, and additivity. Although the equal shared environment is plausible with twins, it is somewhat less plausible when applied to siblings born at different points in time. Studies that apply DF analysis to samples with multiple levels of genetic relatedness interpret the estimate of E1) = c2 as the average common environmental influence across all levels of genetic relatedness (33). A number of authors have extended DF analysis to fit a model with nonadditive genetic, dominance genetic, and nonshared environmental influences (34,35), referred to as the ADE model, which is expressed as


where the notation is the same as above except R2 is a measure of genetic relatedness for nonadditive genetic influences that takes a value of 1 for monozygotic twins, 0.25 for full siblings, and 0 for half siblings. Although some studies use multiple sibling groups defined by gender to implement a model with concurrent nonadditive and shared environmental influences to study BMI in children (example Jacobson and Rowe (12)), most studies compare goodness of fit between the two models and report results from the model that fits better. We adopt this latter strategy and use the Akaike information criterion to determine the preferred model. We allow for sex differences in the effects of heritability, common environmental influences, and nonshared intrauterine environmental influences by interacting the relevant variables with indicators for whether the pair is girl–girl, boy–boy, or mixed sex. F-tests are then used to determine whether there are statistically significant sex differences in the effects.

DF analysis requires double entry of all related pairs, so that sibling pair (1,2) and pair (2,1) are both entered in the data set. The regression errors for these two observations violate the assumption of independence. Therefore, it is necessary to use Generalized least-squares methods, such as sandwich estimators, to control for heteroscedasticity and obtain unbiased estimates of standard errors for the regression coefficients (36). All analyses were conducted using Stata 9.1 (StataCorp, College Station, TX) for models with and without controls for sibling differences in nonshared environment.


Table 1 reports the sample characteristics by individual child at each time period. In the 6- to 8-year sample, 48% were non-Hispanic white, 30% African Americans, and 22% Hispanic. Mean BMIpct was 55.9. In the 12- to 14-year sample, 43% were non-Hispanic white, 35% African Americans, and 22% Hispanic, and the mean BMIpct of the sample was 65.01. At both ages, maternal education was about 13 years.

Table 1.  Individual child sample characteristics
inline image

Table 2 reports the mean absolute differences in BMIpct, BMIz, and birth weight for each age-group by sex-specific sibling relationship. As relatedness increased, i.e., more shared genetic material, the mean absolute differences in BMIpct and BMIz decreased in the boys and mixed groups at both ages; statistically significant differences in BMIpct and BMIz were seen in these groups at the older age. This was not true for girls, as there were no statistical differences between the full and half sibs in either BMIpct or BMIz at either age. Statistical differences in birth weights were found between full and half sibs in each of the sex-determined groups, with greater variability in the half sibs at the younger age; in the older group, statistically significant differences were found in the girls and mixed groups, but not the boys. The absolute differences in residualized BMIpct and residualized BMIz are not reported, but were very similar to the absolute differences in the unadjusted variables.

Table 2.  Sample size and absolute differences in variable means by relationship and gender types for unique pairs of siblings
inline image

Table 3 reports the correlations by relationship code for BMIpct, BMIz, and birth weight at each age. For boys and the opposite sex group, the correlations increased with genetic relatedness at both ages. For girls, the results were mixed. At the younger age, the correlations for BMIpct and BMIz were lower in the full siblings than half siblings. Correlations on birth weight at both ages, and BMIpct and BMIz at the older age were higher in full siblings than half siblings. The correlations between residualized BMIpct and residualized BMIz are not reported, but were comparable to the correlations between the unadjusted variables.

Table 3.  Correlations between key variables by relationship and sex
inline image

Based on the Aikake Information Criterion, the ACE model fit better than the ADE model for both early school-age and adolescent children. We report only the results for the ACE models in what follows. Controls for difference in indicators for measured weight were not significant in any of the specifications, nor were differences in maternal prepregnancy BMI and smoking during pregnancy. The reported results are based on specifications that exclude these variables. Finally, there was no evidence of sex differences in the effect of common environmental influences, and the results are reported for a specification without gender-specific interactions for this coefficient.

The coefficients from the ACE models are reported in Table 4 for both age-groups. Model 1 excludes controls for sibling differences in nonshared environment; and Model 2 includes the difference in birth weight, the only nonshared environmental factor that was significant. Shared environmental factors were significant in both models at the younger age (C2 = 0.08), but were insignificant in both models at the older age. At the younger age, the heritability estimate (h2) for the mixed group and girls only was about 0.20 in both models 1 and 2. For the boys, it was about 0.28 in both models. However, there were no statistically significant differences in estimates of h2 between any of the three gender-specific groups at the younger age. The coefficients on birth weight were significant for all gender-specific groups at the younger age, but the differences between them were not statistically significant. The heritability estimates were higher at the older age than at the younger age, approximately doubling in magnitude for the boys and mixed-sex groups, and almost tripling for girls. There were significant differences in the estimates of h2 between boys and the mixed-sex groups in both models at the older age, but not between girls or the other two gender-specific groups. The coefficient on birth weight differences was significant for boys and the mixed group but not for girls at the older age. There were statistically significant differences in the estimated birth-weight effects for the boys and girls groups, but not between the mixed-sex group, and either the boys- or girls-only groups.

Table 4.  Regression results with residualized BMIz as dependent variable, by age and sex group (robust 95% CI in parentheses)
inline image


Heritability estimates of BMI were between 0.20 and 0.28 at the younger age and 0.46–0.61 at the older age. The point estimates were higher for boys-only pairs than the mixed-sex or girls-only groups. The estimates were not affected by inclusion of birth-weight differences in the regression. These heritability estimates are lower than what has been found from twin studies, but there is concern that heritability in twin studies may be inflated because of the increased difficulty in separating direct from indirect genetic effects that arise from parental responses to genetically driven child behaviors. Indirect genetic effects reflect the influence of the shared environment on heritability and thus will be more correlated for twins than nontwin siblings. Butte, et al. have suggested using general sibling samples to minimize the bias in heritability estimates attributable to shared indirect genetic effects present in twin studies (20). The home environment observed over time is likely to have a stable, common component, but at any point in time may reflect specific family events, e.g., job loss experienced by a parent, divorce, parental health problem, etc. Twins experience more commonality in indirect genetic effects than full siblings because they share both the stable and idiosyncratic parts of the home environment, whereas full siblings who are born at different points in time share only the common component. Our estimates, while smaller than most reported in twin studies, still suggest a substantial component of variation in BMI due to heritable genetic differences.

Sex differences in heritability estimates were found between the mixed-sex group and the boys-only group at age 12–14 years. This may reflect different sex-specific growth patterns related to the maturational processes in puberty (37,38). The magnitude of the heritability estimates as well as the correlation coefficients suggest a stronger heritability effect in boys than girls, but the estimates were not significantly different in the regression analyzes. The results of this study also confirm other findings that suggest heritability increases with age. An effect of common environmental influences was found only for the younger group. There was no evidence of sex differences in the effect of common environmental influences at either age.

The differences in birth weight were statistically significant for all sex-specific pairs measured at school age, indicating that sibling differences in BMI have their beginnings in utero. By adolescence, differences in birth weight were not significant for girl-only pairs, but were significant for mixed and boy-only pairs. Interactions of birth-weight differences with relationship codes were not significant, suggesting that this nonshared influence does not have a genetic component. In contrast to birth weight, sibling differences in maternal prepregnancy BMI and smoking were not associated with sibling differences in BMI. However, further analysis showed that for most mothers, these behaviors were similar across pregnancies. The lack of within-mother variation in these variables may contribute to their lack of significance in the regression.

The study has several limitations. Fathers may influence the obesogenic environment so that half siblings have less commonality in the environment than full siblings. Although children are measured at similar ages, full siblings born in different years may be raised in different social environments. Other limitations are that the study sample represented only 70% of the eligible sample and that some of the heights and weights were mother-reported instead of measured. However, we do have indicators for mother reports that allowed us to control statistically differences in reporting method through use of residualized BMIz as the dependent measure. Finally, all birth weights were mother-reported with variable time lags between actual birth and birth-weight reports.

In conclusion, the study investigated sex differences in the role of genetics, shared environment, and nonshared intrauterine environment on BMI among children and adolescents. There was some evidence of sex differences in heritability and the effect of birth weight in adolescence. The birth-weight effect was not explained by genetic interactions, indicating a possible important environmental influence; the intrauterine environment difference between boys and girls also suggests that there may be a sex-specific effect operating. Further study is needed to explore these sex-specific differences.


This work was supported with a RO1 grant from NIH and NINR.


The authors declared no conflict of interest.