Twin and adoption studies suggest that family environment has little, if any, influence on body mass index (BMI) in adulthood. We investigated the hypothesis that the differences in the years of birth between siblings influence their similarity in BMI at comparable ages, which would give evidence for a possibly modifiable influence of the environment shared by family members.
Swedish full-brother pairs (N=261 712) born between 1951 and 1983 were measured for BMI in conscription examination at 16–26 years (median: 18.2 years) of age and were divided into quartiles by the difference between their birth-years (< 2.25 years, 2.25–3.33 years, 3.34–5.08 years and >5.08 years). Furthermore, 1961 dizygotic twin brother pairs from the same population representing brothers born at the same time were included. In addition, the log BMI of the younger brother was modeled as a linear function of the log BMI of the older brother. Subsequently, the significance of the interaction between birth-year difference and the BMI of the older brother was tested.
Intraclass correlation for BMI in dizygotic twin pairs was higher (0.431, 95% confidence interval (CI) 0.394–0.466) than the correlation for full-brothers in the first quartile of birth-year difference (0.376, CI 0.342–0.408). Among full-brothers, the BMI correlation decreased from 0.378 (CI 0.372–0.385) in the first quartile to 0.338 (CI 0.331–0.345) in the last quartile. The regression analysis showed a statistically significant decrease in correlation with increasing birth-year difference (P<0.001).
The influence on BMI in young men of the environment shared by dizygotic twin brothers is greater than between non-twin full-brothers, indicating important influences of concomitant exposure to the same early life environment before and/or after birth. Among non-twin siblings there is a slight possibly modifiable influence as evidenced by declining correlations by increasing distance in years of birth.
The prevalence of obesity has increased at an alarming rate globally in the past decades, and in the United States, more than a third of all men and women are now obese.1, 2 In addition, in the youngest age groups, obesity has become one of the most serious public health challenges of the twenty-first century.3 As obesity in early life tracks into adulthood,4 preventive efforts targeted at children and adolescents is a cost-effective strategy to combat obesity for all ages.
The family setting is often considered a target in public health strategies for obesity prevention. However, the evidence for the importance of childhood family environment for the development of obesity is not consistent. Studying similarity of body mass index (BMI) between family members can give evidence on the role of family environment in the development of obesity.5 The classical twin design, which compares monozygotic (MZ) and dizygotic (DZ) twin pairs, is the prevailing approach to disentangling contributions of shared environment from genetic influences.6 Although the influence of shared environment on BMI has been found in studies of twin children,7 the effect is typically not found among adolescents and adults.8 The classical twin design may, however, not be an ideal approach to study shared environmental effects as it is unable to differentiate additive genetic effects from combined effects of nonadditive genetic effects and common environmental effects.9 The study of phenotypic resemblance in adoptive siblings reared together is a methodologically rigorous approach to test for shared environmental influences. Several such studies have indicated that the shared family environment influences BMI as long as the adoptees live with their foster parents, but thereafter the influence vanishes.10, 11, 12 Nevertheless, adoption studies have limitations attributable to the study design (relatively small sample sizes and age differences at the time of assessment) and adoptions not taking place immediately after birth.
The aim of this study is to investigate the part of shared environmental influences on BMI in young adults that changes as time goes by within families. This is done by investigating whether the BMI resemblance within pairs of brothers measured at similar ages depends on the year-of-birth interval between them. The siblings born shortly after each another may share a more similar environment than those with larger intervals between the births. This is because the rearing environment in families may vary over time, which could lead to different childhood experiences for two brothers within the same family. The inclusion of DZ twins allows assessment of the influence of concomitant exposure to the prenatal and/or early postnatal environment when compared with non-twin full-siblings, as the DZ twins, by sharing the same amount of genetic material as full-siblings, represent siblings with no time interval between births. Our hypothesis is that the highest BMI correlation is found for DZ twins and that the correlations gradually decline with longer intervals between births, indicating modifiable influences on BMI of the shared environment.
Materials and methods
The data comprise height and weight measurements of all Swedish men born between 1951 and 1983 (N=1 736 369) who underwent military conscription examination at the age of 16–26 years (median: 18.2 years, s.d.: 2.97). Height was measured using a wall-mounted stadiometer and weight using an analog or digital scale. BMI was calculated as weight (kg)/height2 (m2). Extreme BMI values (BMI <15 and >50 kg m−2) were excluded to reduce the risk of misclassification because of measurement errors or data entry errors. During the years covered by this study, conscription examination was compulsory by law for all young men with Swedish citizenship. Only men with severe diseases or disabilities were exempted based on a certificate issued by a physician with information on diagnosis.
Information on sibling status was acquired by linking the conscript data with the Swedish Multi-Generation Register and the Swedish Twin Register using the Swedish personal identification number (ID). The ID numbers of the biological parents were included in the record of their offspring, and on this background 259 751 pairs of full-brothers and 1961 DZ twin pairs were identified among the conscripts. Data on zygosity were obtained from the Swedish Twin Register and from the Swedish Young Male Twins Study13, 14 where zygosity was obtained using well-validated questions on physical similarity.
The full-brother pairs were divided into quartiles on the basis of the interval between births: quartile 1, <2.25 years; quartile 2, 2.25–3.33 years; quartile 3, 3.34–5.08 years; and quartile 4, >5.08 years. Intraclass correlation coefficients and 95% confidence intervals (95% CIs) were calculated on BMI for DZ twin pairs and within each quartile of full-brother pairs using the SPSS statistical software, version 19 (IBM Corporation, Armonk, NY, USA). If a smaller BMI correlation is found among brother pairs with a greater interval between the birth-years, this would suggest that siblings do share environmental factors that may change over time in their influence on BMI. As the genetic correlation is 0.5 for both full-brothers and DZ twin pairs, a difference between these two groups in the intrapair BMI correlations could suggest an influence from the intrauterine or early postnatal environmental factors shared by the twins but not by the non-twin siblings. In addition, we performed a linear regression analysis with the log-transformed and birth-year-specific BMI z-score of the younger brother as a function of the log-transformed and birth-year-specific BMI z-score of the older brother using the proc reg procedure in the SAS statistical software, version 9.2 (SAS Institute Inc., Cary, NC, USA). An interaction term between the logarithm-transformed BMI of the older brother and the difference in birth-year was included into the model. As we used standardized values of BMI for both brothers, the regression parameter can be interpreted as the correlation coefficient. A statistically significant negative interaction would therefore indicate that the intrapair resemblance in BMI decreases with increasing birth-year intervals. Hence, the regression analysis gives a formal statistical test of the study’s hypothesis. In addition, we performed a corresponding regression analysis with the BMI of the younger brother as a predictor of the BMI of the older brother to see if the results were sensitive to the ordering of brother pairs. It should be noted that as the DZ twins were born on the same date, the ordering was random in this group. Subsequently, we performed the same regression analyses without DZ twins to see if different results would be obtained. Furthermore, we used the regression parameter of the older brother on the younger brother to extrapolate the full-brother correlation to a hypothetical birth-year difference of zero. If this extrapolated correlation is higher than that of DZ twin brothers, it is evidence of uniquely shared environmental conditions among DZ twins relative to full-brothers.
To test the validity of the findings, we did a series of sensitivity analyses. First, the quartile correlations were calculated using birth-year-specific z-scores of BMI. The birth-year-specific mean was subtracted from each individual raw score and divided by the birth-year-specific s.d. This sensitivity analysis is relevant as the environment surrounding the families changes over calendar time, and these secular changes in the environment could also result in a more diverse rearing environment for brothers with larger birth-year differences. By standardizing the BMI distributions, the secular changes in the environment that have brought about changes to the BMI distribution are adjusted for. The remaining difference in BMI values between brothers who may belong to different birth-years should then reflect the degree of similarity in family upbringing rather than the secular environmental changes. In other words, the birth-year specific z-score standardization allows us to look isolated at the influence from shared environmental influences on brother-pair correlations. The BMI values were logarithm transformed before calculating z-scores in both the quartile and the regression analysis. The logarithm transformation was used to reduce a slight positive skewness of the BMI distribution. However, the log transformation influences the results as the procedure brings BMI observations that are far apart closer together. Therefore, we investigated possible changes to the interaction parameter in the regression analysis when using z-scores that were not log transformed. Furthermore, we investigated whether the same results, still using BMI z-scores, could be obtained when excluding all brother pairs with an age difference at conscription (measurement) >1 year (9.8% of the population), as this parameter could influence the findings if it varies with distance in birth-years. In addition, we simulated non-paternity. If there is a higher occurrence of non-paternity among full-brothers with larger birth-year intervals, this could generate a decrease in BMI resemblance. A previous study suggests a rate between 2 and 5% in Western populations.15 To determine whether non-paternity could explain a decrease in BMI resemblance with larger birth-year intervals, we replaced 5% of the full-brother pairs in the first quartile of birth-year difference with maternal half-brother pairs from the same underlying study population. The correlation was subsequently compared with the correlation in the last quartile where the largest birth-year difference and the lowest correlation (see Results) were found. With a misclassification rate of <4%, studies have confirmed that questionnaires are a valid method of zygosity determination.16 Nevertheless, to investigate the possible significance of this bias, we performed a sensitivity analysis in which the 4% most similar DZ twin pairs, in terms of BMI, were removed.
Table 1 shows summary statistics for the groups of DZ twins and full-brothers stratified by quartiles of birth-year difference. The BMI mean and s.d. for full-brothers and DZ twins were 21.75 (2.89) kg m−2 and 21.05 (2.37) kg m−2, respectively.
Figure 1 (left panel) shows intraclass correlations of BMI for DZ twins and full-brothers stratified by quartiles of birth-year difference. Among full-brothers, the BMI correlation decreased with increasing difference in birth-year from 0.378 (CI 0.372–0.385) in the first to 0.338 (CI 0.332–0.345) in the last quartile. It should however be noted that the CI’s were overlapping except for the first and last quartiles. The BMI correlation among DZ twin pairs (0.431, CI 0.394–0.466) was higher than the correlation in all pairs of full-brothers (0.359, CI 0.356–0.363) (not shown in Figure 1) and each of the quartile of full-brothers. The same analyses were run using Pearson’s correlations and were found to give virtually the same results as the intraclass correlation coefficient. The same overall tendency was found when using logarithm-transformed BMI z-scores to calculate the correlations (Figure 1, right panel). The correlations showed a slightly attenuated decline across quartiles compared with the raw BMI values. This suggests that secular changes in the environment surrounding the families are partly responsible for the decrease in raw BMI correlations.
The results from the regression analysis and the corresponding sensitivity analyses are presented in Table 2. A small but statistically significant interaction was found between the effect of the BMI of the older brother and the intrapair difference in birth-year (P<0.001) on the BMI of the younger brother (model 1). The regression parameter was 0.378 (CI 0.372–0.385) corresponding to the hypothetical BMI correlation within brother pairs when the birth-year difference is zero. This correlation is significantly lower than the DZ twin correlation. The value of the interaction parameter was −0.0034 (CI −0.0047 to −0.0021), which is the estimated average decrease in correlation with each incremental year of birth-year difference. We performed the corresponding analysis with BMI of the younger brother as the predictor of the BMI of the older brother (model 2) with similar results. The results were similar when using BMI scores that were not log transformed before z-score standardization (models 3 and 4). Excluding DZ twins from the model did not significantly change the regression or interaction parameters.
The same results were obtained in both the quartile and the regression analysis when excluding brother pairs with an age difference of >1 year at measurement. Finally, the simulation of non-paternity showed a marginal decrease in correlation among full-brothers in the first quartile of birth-year difference from 0.373 to 0.371 (CI 0.365–0.377) when 5% of the full-brother population was replaced with maternal half-sibs. However, there was still no overlap in CI’s between the first and last quartile, implying that non-paternity alone is unlikely to explain the findings. The removal of 4% of the most similar DZ pairs gave a moderate decrease in BMI correlation from 0.430 to 0.402 (CI 0.363–0.439).
We found a statistically significant deviation from normality of the residual values in the regression models. Hence, as a sensitivity analysis we calculated robust standard errors of the parameters in model 1 using robust ‘sandwich’ estimation in STATA (version 9.2) statistical software. The CI’s for the regression coefficient widened slightly from 0.372–0.385 to 0.371–0.386. Similarly, the CI for the interaction parameter widened marginally (from (−0.0047–0.0021) to (−0.0050–−0.0018)). The interaction parameter was still statistically significant (P<0001) when using robust standard errors.
As genetic correlations are similar for DZ twins and for non-twin full-brother pairs, DZ twins can be conceived of as full-brother pairs with no time interval between births (although it should be noted that only the twins concomitantly share the intrauterine environment). Hence, the gradual decline in the correlations from DZ twin pairs through the successive quartiles of full-brothers shows a systematic relationship between length of birth interval and the degree of similarity in BMI within brother pairs. This relationship was formally tested in the regression model in which the interaction between birth-year difference and the BMI of the older brother on the BMI of the younger brother is negative and statistically significant. It should, however, be noted that the interaction parameter is relatively small, corresponding to ∼1% decrease (for example, in model 1 the value of the interaction is 0.0034, which corresponds to a 0.0034/0.378=0.9% decrease in BMI correlation for each 1-year increase in birth-year difference) in BMI correlation for each 1-year increase in birth-year difference. The results were similar, although not identical, when the ordering of brother pairs was reversed; that is, when BMI of the younger brother was used as predictor for the BMI of the older brother.
The results indicate that non-twin full-brothers with longer intervals between births have grown up under slightly different family conditions influencing BMI variation at conscription age. In other words, brothers with a shorter interval between births appear to share environmental factors to a greater extent than those with a longer birth-year interval. It is a potentially important finding that the difference in correlations between DZ twin brothers and non-twin brothers is relatively large compared with the more moderate decline in correlations across quartiles of non-twin brothers. This could imply that the concomitantly shared intrauterine environment is important. This notion is strengthened by the fact that the predicted correlation in the regression model among full-brothers with a birth-year difference of zero (regression parameter in Table 2, model 1) was significantly lower than the correlation for DZ twins. It is, however, also possible that DZ twins to a greater extent than ordinary full-brothers share environmental factors during childhood for reasons not caught by a measure of birth-year difference. This finding could also be attributed to other critical periods in the later development where the environment changes rapidly and the individuals are vulnerable to its exposure in terms of development in adipose tissue.
Both secular changes in the environment surrounding the families, as well as independent changes in the family environment, could result in a more diverse rearing environment for brothers with longer birth intervals. However, the results using birth-year-specific log-transformed BMI z-scores took the birth-year-related changes in BMI into account by standardizing the distributions for each birth-year. Hence, the similar results obtained from the analysis using BMI z-scores indicate that secular environmental changes only account for a slight part of the difference between the correlations. The findings show that environmental factors shared by family members and possibly modifiable over time have an influence on BMI.
The main strength of this study is the population-based nature of the material with large number of full-brother pairs identifiable through linkage of the Swedish conscript database with the Swedish Multi-Generation Register, allowing detection of even very small differences in correlations. Another important strength is that brothers were measured at almost the same age. If there were larger age differences between brothers at measurement, this would attenuate the correlations because partly different sets of genes seem to affect BMI at different ages.17
The current approach also has limitations. First, there are various risks of misclassification: non-paternity may confound the analyses by being more likely the larger the difference in year of birth. However, our simulation of non-paternity in the first quartile showed that the attenuation of the BMI correlation was too small to explain the difference between the quartiles. Hence, non-paternity is unlikely to be entirely responsible for the findings. Although we consider the decline in the between-brother correlations as due to a decline in a common environmental influence, we admit that there may be other reasons beyond non-paternity that we have not been able to address in this study. One possibility could be changes over time in transmitted epigenetic effects on BMI because of changes in the parental environment influencing their germ cells, although this in principle may be considered a contributor to shared environment among the offspring.
Another issue is that a small fraction of the DZ twin pairs may in fact be MZ twin pairs, which would inflate the correlation. The sensitivity analysis in which the 4% most similar DZ twins were removed (assuming them to be MZ twins) showed a moderate decrease in correlation from 0.430 to 0.402. This questions the notion of an unexpectedly large difference in correlation between DZ twins and full-brothers. The reduced correlation in the sensitivity analysis implies that the difference between DZ twins and full-brothers with a short interval between births approaches the general decline in correlation with larger birth-year intervals among full-brothers. A second issue is that although there is considerable variation in the length of birth intervals in the data set, the difference in correlations between groups of shorter and longer intervals between births is unlikely to reflect the full range of influence from the shared environment; brother pairs with a long interval between births have nevertheless in most cases still lived in the same household during childhood. A third limitation is that we only have measurement in late adolescence/young adulthood, whereas siblings possibly have a more similar lifestyle at younger ages; young adult brothers may have more unique lifestyles and are thereby less influenced by the rearing environment. This could dilute the effect of shared environment (as compared with the influence in childhood); had brother pairs been measured in childhood or early adolescence, the difference in correlation between quartiles could have been larger. Another possible limitation is that the log transformation influences the results as the procedure tends to bring BMI observations that are far apart closer together. However, the sensitivity analysis showed little changes to the decrease in correlation with larger birth-year intervals.
Finally, it should be noted that the BMI mean and variance for DZ twins is lower than for full-brothers. The possible causes and consequences of this difference have previously been reported and thoroughly discussed.18 Most importantly, it may imply that twins are not representative of the general population. The environment shared by full siblings may differ from the constellation of environmental factors shared by twins because of the concomitant sharing of the intrauterine environment. Hence, the observed difference in correlation between DZ twins and non-twin full-brothers may imply that possibly modifiable intrauterine exposures contribute to BMI variation and that these influences last into adulthood. Nevertheless, it should be noted that when excluding the 4% most similar DZ twins, the correlation decreased from 0.430 to 0.402. If in fact the excluded twin pairs are misclassified MZ twins, it indicates that the difference in correlation between DZ twins and first-quartile full-brothers is more in line with the decline in correlations observed across the quartiles of full-brothers.
Our study shows that the influence of the family environment on BMI variation lasts at least into young adulthood, as demonstrated by the difference between DZ twin brothers and non-twin full-brothers and by the decline in the correlations in BMI between full-brothers by the increasing distance in years of birth. These environmental influences suggest existence of targets in the family environment for prevention of obesity during childhood and adolescence.