We examined the etiology of two disordered eating characteristics.
We examined the etiology of two disordered eating characteristics.
Participants included 1,470 female adolescent and young adult twins and their female nontwin siblings. Phenotypic factor analyses of a seven-item eating pathology screening tool yielded two factors: weight and shape concerns and behaviors (WSCB) and binge eating (BE). Univariate and bivariate extended twin analyses (including cotwins and nontwin siblings) were used to estimate the magnitude of genetic and environmental influences on these characteristics.
Analyses indicated that individual differences in WSCB and BE could be explained by additive genetic influences (a2 = 0.43 (95% CI: 0.33–0.52) and 0.49 (95% CI: 0.36–0.58), respectively), with the remaining variance due to nonshared environmental influences. The genetic correlation between WSCB and BE was estimated at 0.64; the nonshared environmental correlation was estimated at 0.27.
These results corroborate previous findings on genetic and environmental influences on disordered eating characteristics and suggest that findings can be extended to nontwin populations. © 2010 by Wiley Periodicals, Inc. Int J Eat Disord 2010; 43:751–761
Investigations of genetic and environmental influences on disordered eating characteristics have primarily relied on the classical twin design (e.g., Refs.1–3). Disordered eating characteristics that are often present in anorexia and bulimia nervosa are binge eating and weight and shape concerns. Twin studies of these characteristics have shown that heritability point estimates ranged from 17% to 82% for binge eating,1, 4–7 62%–64% for shape concerns,2, 3 and 66% for weight concerns,3 with the remaining phenotypic variance due to nonshared environmental effects (i.e., factors that account for differences among individuals in the same family). Notably, in the study by Wade et al.,2 shared environment (i.e., environmental influences that account for similarities among members of the same family) and nonshared environment influenced weight concerns in adult women in roughly equal proportions. In a study investigating multiple age ranges, shared and nonshared environment predominantly influenced a combined measure of weight and shape concerns in 10- to 12-year-old girls, whereas additive genetic and nonshared environment solely influenced this same trait in 13- to 41-year-old adolescent girls and women.8
Although these studies suggest that genetic factors influence these important characteristics, there has been little attention to understanding how common genetic and common environmental factors may influence these frequently co-occurring traits. A few studies have utilized multivariate analyses to investigate common etiological factors underlying binge eating and obesity4 and binge eating and purging behaviors,9 and one study10 examined genetic and environmental influences on bulimia nervosa by including individual binge eating and weight/shape concerns items in a common-factor model. To date, no study has directly investigated the potential for common etiological factors underlying binge eating and weight/shape concerns.
This study aimed at extending previous literature in two ways. First, we examined the genetic and environmental risk factor overlap between binge eating and weight/shape concerns. While prior studies have examined these characteristics individually, the use of a bivariate model is important to understand whether these two traits, which are frequently present in individuals with anorexia and bulimia nervosa, share genetic and environmental risk factors. Second, we included nontwin siblings to test whether there were twin-specific shared environmental influences on disordered eating characteristics, as this has not been previously examined in the eating pathology literature. Although earlier work has shown that there are no differences between twins and nontwin samples in rates of psychiatric disorders,11 recent studies of substance use behaviors that include twins and nontwin siblings suggest that twins share more similar environments than other siblings in the family,12 and substance use often co-occurs with eating pathology.13, 14 The presence of such twin-specific shared environments could reflect more shared peer groups among twins than nontwin siblings, twin cooperation, and/or use of a co-twin instead of a nontwin sibling as a model for eating attitudes and behaviors. The use of nontwin siblings not only extends the generalizability of these findings to nontwin populations,15, 16 but also increases power.17
Study participants included 1,398 female twins (710 monozygotic (MZ), 437 dizygotic (DZ), 251 opposite-sex (OS)) and 72 of their female nontwin siblings, for a total of 1,470 individuals. Participants included 53 trios (i.e., twin-twin-sibling; 38 MZ, 15 DZ), 507 twin-twin pairs (310 MZ, 197 DZ), 19 twin-sibling pairs (1 MZ, 1 DZ, 17 female OS twins with a female sibling), and 259 single-responding twin individuals (13 MZ, 12 DZ, 234 female siblings of OS pairs). All subjects were participants in the Colorado Center for Antisocial Drug Dependence (CADD; PIs Crowley & Hewitt), which is an ongoing, longitudinal twin and family study of substance use, antisocial behavior, and comorbid psychopathology. The CADD twin sample was drawn from the Colorado Community Twin Study (CTS) and Longitudinal Twin Study (LTS), conducted at the Institute for Behavioral Genetics at the University of Colorado.18 For CTS families, twins were recruited by birth records obtained through the Colorado Department of Health, Division of Vital Statistics or through primary and secondary schools in Colorado beginning in 1983. Twins born from 1968 onwards were invited to participate in the CTS and willing participants born between 1980 and 1990 were included in the CADD study. The LTS population is a somewhat more restricted sample, as families were recruited only from the Colorado Department of Health, Division of Vital Statistics, only from the greater Denver metropolitan area, and only from families whose twins had healthy birth weights and gestational ages born between 1984 and 1990. The CADD sample was comparable with the larger sample of twin families from which it was drawn.18
Participants were included in this study based on the following criteria: (1) all individuals were female; (2) individuals must have been 16 years old or older (in order to minimize the effects of pubertal development on genetic and environmental influences on eating pathology19–22); (3) twins and nontwin siblings who were part of the same family must have been tested within two years of each other; and (4) when there were multiple assessments (the study is an ongoing longitudinal study), the first assessment which met the age 16 or older criterion was utilized. Males were initially examined in the data set, but due to the difference in factor structure and low item endorsement rates for all of the items, they were not included in this study. The average age difference between twins at time of testing was 0.03 ± 0.18 years, the average age difference at time of testing between twin 1 and her sibling was 2.62 ± 1.42 years, and the average age difference at time of testing between twin 2 and her sibling was 2.53 ± 1.40 years.
The CADD protocol was approved by the University of Colorado's Institutional Review Board. All participants gave informed consent (if 18 years old or older) or assent (if 17 years old or younger), and parents also provided informed consent for participants under age 18. All subjects were paid for participation.
Participants answered seven questions about eating attitudes and behaviors23, 24 (see Table 1). These questions asked participants if they had ever endorsed or engaged in these eating attitudes and behaviors in their lifetimes. Individuals with incomplete data were included in basic descriptive statistics (see Table 2). For the binary factor analyses and all subsequent analyses, complete data were required (i.e., answering yes or no to each of the seven items of the eating pathology screening tool). However, 99% (n = 1,389) of twins and 100% (n = 72) of nontwin siblings had complete data on the seven items. Thus, there was minimal missing data.
|Items||Item Endorsement||Factor 1||Factor 2|
|Do you feel fat even though others say you look thin?||0.46||0.77||0.03|
|Do you think about staying thin almost all the time?||0.31||0.87||−0.10|
|Do you ever make yourself throw up after eating?||0.03||0.62||0.06|
|Do you ever eat less than usual for several days when you are upset?||0.28||0.50||0.02|
|Do you ever eat in secret?||0.10||0.19||0.47|
|Is it sometimes hard to stop eating?||0.24||0.14||0.66|
|Do you ever eat more than usual for several days when you are upset?||0.19||−0.12||0.96|
|Age (mean ± SD); range||18.28 ± 1.63 years; 16.02–25.91||18.08 ± 1.56 years; 16.02–26.05||18.81 ± 1.66 years; 16.00–25.01||19.43 ± 2.15 years; 16.11–24.94||18.37 ± 1.68 years; 16.00–26.05|
|% of participants endorsing WSCB items|
|% of participants endorsing BE items|
Item-level phenotypic factor analyses were conducted using Mplus version 4.125 to investigate the dimensional structure underlying the seven-item eating pathology screening tool. Because the items were dichotomous (i.e., yes/no), we used binary factor analysis, which takes into account differences in endorsement rates for each of the items. Promax rotation was used to allow a correlation between the factors. We used three sources for determining the number of factors to be extracted: Kaiser's Rule (i.e., eigenvalues greater than one), the scree plot, and interpretation of the factor loadings. Item endorsements ranged from 0.03 (self-induced vomiting) to 0.46 (feeling fat although others say you look thin) (Table 1).
Exploratory factor analyses suggested that there were one or two factors underlying the data (eigenvalues: Factor 1 = 3.46, Factor 2 = 1.09). Although the three- and four-factor solutions were also considered, these solutions yielded eigenvalues less than 1.00 (Factor 3 = 0.74, Factor 4 = 0.54). Moreover, only one item loaded on the third factor (eating more than usual when you are upset). Given that factors having only a single item loading are poorly defined, confirmatory factor analyses only compared the one- and two-factor solutions. Model-fits were assessed using chi-square (χ2) values and the root mean square error of approximation (RMSEA), where lower RMSEA values indicated better fit.26 Confirmatory factor analyses indicated that the one-factor solution provided a poorer fit to the data (χ2 = 135.17, df = 12, p < 0.01, RMSEA = 0.08) than the two-factor solution (χ2 = 26.30, df = 11, p = 0.01, RMSEA = 0.03). Four items loaded highest on Factor 1 (feeling fat though others say you are thin, thinking about staying thin, self-induced vomiting, and eating less than usual when upset) and three items loaded highest on Factor 2 (eating in secret, finding it hard to stop eating, and eating more than usual when upset) (Table 1). Items loading on the first factor reflect key attitudes and behaviors that are seen in both anorexia and bulimia nervosa, whereas items loading on the second factor are commonly seen in individuals with the binge-purge subtype of anorexia and bulimia nervosa and binge eating disorder. The correlation between the two factors was 0.60.
Given these results, a two-factor solution was retained and used for subsequent analyses. Factor 1 was termed weight and shape concerns and behaviors (WSCB) and Factor 2 was termed binge eating (BE) (see Table 1). Ordinal factor scores were derived by simply summing (unit weights) the items loading highest on the respective factors. Thus, scores for WSCB ranged from 0 to 4 and for BE ranged from 0 to 3. Because of the low number of participants who endorsed all four WSCB items (n = 23), individuals who endorsed either three or four of these items were combined into one category (labeled 3+). Higher scores indicate more severe levels of disordered eating. In the current study, the internal consistency of WSCB and BE was modest (Chronbach's α = 0.55 for WSCB and 0.57 for BE). Previous studies23, 24 have shown that these items are reliable (Chronbach's α ≥ 0.69)24 and show acceptable sensitivity and specificity between women with anorexia or bulimia nervosa and controls.23, 24
All opposite-sex twins were necessarily assigned DZ twin status. For 98% of the same-sex twins for whom genetic data were available (n = 1,125), zygosity was determined by genetic analysis of at least 11 highly informative short tandem repeat polymorphisms.18 Co-twins concordant for all genetic markers were classified as MZ and co-twins discordant for any genetic markers were classified as DZ. For the 2% of same-sex twins for whom genetic data were unavailable (n = 22), zygosity assignment was determined using a modified version of the Nichols and Bilbro questionnaire27: a nine-item assessment of physical characteristics, such as eye color, hair color and curliness, height, weight, how often the twins were mistaken for their cotwin by others, and opinions of their own zygosity. Rules for assigning zygosity based on the Nichols and Bilbro questionnaire have been shown to classify MZ and DZ twins with greater than 90% accuracy.27
Because the distributions of overall scores on WSCB and BE were positively skewed and reflected ordinal versus continuous measures, we chose to use a threshold approach to analyze the data. Multiple threshold models were fit to the raw ordinal data for the disordered eating characteristics in the statistical package Mx,28 which allowed for missing data. Threshold models assume a normal, continuous underlying liability distribution, where particular thresholds on the liability distribution give rise to the observed ordinal scores. Thus, for both WSCB and BE, three thresholds were used that reflected a change in the ordinal score from 0 to 1 (i.e., threshold 1), 1 to 2 (i.e., threshold 2), and 2 to 3/3+ (i.e., threshold 3).
Of note is the wide age range (16–26 years) in the participants. Correcting for age using standard regression procedures, as is done with continuous data, is not appropriate for ordinal data. For ordinal data, age-dependent thresholds were used for calculating all correlations and for biometrical model-fitting. Taking into account age-dependent thresholds allows the endorsement rates on the ordinal WSCB and BE scales to differ depending on an individual's age. Separate thresholds were estimated for six different age groups: 16, 17, 18, 19, 20, and 21- to 26-year-olds. Participants who were 21- to 26-years-old were combined into one group because of the low number of individuals at each of those particular ages.
Descriptive statistics were analyzed using SPSS version 12.0.29 For these analyses, all female subjects (i.e., MZ twins, DZ twins, OS female twins, and nontwin siblings) were examined to investigate the underlying structure of the seven-item eating pathology screening tool.
Polychoric correlations (within-trait and cross-trait sibling correlations) for MZ twins (rMZ), DZ twins (rDZ), and siblings (rSib) were computed in Mx28 to provide initial information regarding genetic and environmental influences on these traits. Polychoric correlations were utilized instead of Pearson correlations because data were ordinal rather than continuous. Interpretations of polychoric correlations [and 95% confidence intervals (CIs)] are identical to that of correlations obtained for continuous measures.
With respect to twin correlations, additive genetic influences are present if the rMZ is greater than the rDZ, and nonshared environmental influences are present if the rMZ is less than one. By comparing the rDZ to half of the rMZ, we can determine if shared environmental influences or dominance effects are present. If rDZ is less than half of the rMZ, there is evidence for dominance effects. If rDZ is equal to half of the rMZ, shared environmental effects are absent. If rDZ is greater than half of the rMZ, there is evidence for shared environmental effects.
The rSib is compared with the rDZ to determine whether any twin-specific shared environmental influences (i.e., environmental influences that are shared only among twins or same-age siblings) are present. If the rSib is lower than the rDZ, there is evidence for twin-specific shared environment; if the two correlations are equal, any shared environmental influences present in the data are due to environments shared by all siblings in a family (i.e., there is no twin-specific shared environment).
Correlations between WSCB and BE were also examined in the three groups, including cross-sibling, cross-trait correlations. These demonstrate the relative influence of genetic and environmental effects to the correlation (orcovariance) between the two traits. This rationale is identical to the univariate analysis described above, except that the etiology of the covariance between traits rather than the variance of the traits is examined, and nonshared environmental influences on the correlation are present if the MZ cross-trait correlation is less than the phenotypic correlation.
Univariate model-fitting analyses examined the magnitude of additive genetic (A, sum of the accumulation of multiple genes impacting a trait), shared environmental (C, environmental influences that account for similarities among members of the same family), and nonshared environmental (E, environmental influences that account for differences among members of the same family) influences on WSCB and BE. Notably, nonshared environmental influences also include measurement error. Because we included nontwin siblings in the analyses, we were also able to examine the presence of twin-specific shared environmental influences (T, environmental influences that are shared only among twins or same-age siblings).
A Cholesky decomposition (or lower diagonal factorization) was used for bivariate twin analyses30 in order to assess the magnitude of A, C, T, and E. A path diagram of the full bivariate model between WSCB and BE is shown in Figure 1. This diagram includes the latent variables that give rise to the additive genetic effects common to WSCB and BE (A1) and unique to BE (A2), as well as the shared environmental effects common to WSCB and BE (C1) and unique to BE (C2). T1 describes the twin-specific shared environmental effects common to WSCB and BE, whereas T2 describes these effects which are unique to BE; E1 describes the nonshared environmental effects that are common to the two traits, whereas E2 describes nonshared environmental effects that are unique to BE. Further, the diagram includes the path coefficients that determine the additive genetic (aij), shared environmental (cij), twin-specific shared environmental (tij), and nonshared environmental (eij) variances and covariances for the two traits.
For both univariate and bivariate model-fitting analyses, multiple submodels of the full ACTE model were fit to the data that constrained the magnitude of one or more of these influences to zero (i.e., the submodels include the ACE, ATE, AE, TE, and E models). Model-fitting analyses were conducted using the statistical package Mx.28 Standard chi-square (χ2) difference tests28 were used to compare the fit of nested submodels to the full ACTE model. When comparing nested models, p-values less than 0.05 for the χ2 difference test indicate that the nested submodel provides a poorer fit to the data than the full model. Akaike's Information Criteria (AIC31) was used for comparing non-nested models. Lower AIC values indicate better-fitting and more parsimonious models.
Table2 describes the demographic characteristics of the sample and endorsement rates for WSCB and BE. The mean age of all participants in the sample was 18.35 ± 1.68 years (range, 16.00–26.05). The average number of days between assessing participants within the same family was 27.3 days, with 97.2% tested within one year of each other. The percentage of females who endorsed at least one WSCB item was 59.9%; the percentage who endorsed at least one BE item was 34.8%.
Sibling polychoric correlations are presented in Table3. Of particular note is the rSib column. When we tested whether there were differences between the nontwin sibling correlations for each of the three zygosity groups (i.e., MZ, DZ, and OS), the different nontwin sibling correlations could be constrained to be equal. Therefore, the rSib column represents the nontwin sibling correlation utilizing all nontwin sibling pairings irrespective of the twin zygosity status.
|Factor||rMZ (n = 348 pairs)||rDZ (n = 212 pairs)||rSib (n = 125 pairs)|
|WSCB||0.42 (0.31–0.52)||0.28 (0.12–0.41)||0.13 (−0.13–0.32)|
|BE||0.47 (0.34–0.60)||0.29 (0.09–0.46)||0.22 (−0.07–0.42)|
|WSCB-BE||0.28 (0.19–0.37)||0.16 (0.00–0.27)||0.17 (−0.03–0.32)|
Comparisons of the sibling polychoric correlations show that the rMZ is larger than the rDZ for both WSCB and BE, which suggests that there is evidence for genetic influences on both traits. There is also evidence for twin-specific shared environmental effects since the rSib for WSCB and BE is lower than the respective rDZ. Cross-sibling, cross-trait correlations (i.e., WSCB-BE) again suggest that there are genetic influences contributing to the correlation between these traits, as the rMZ is larger than the rDZ. However, the cross-trait correlations between WSCB and BE for DZ twin pairs (rDZ) and nontwin siblings (rSib) are almost identical, indicating no twin-specific shared environmental influence on the covariation between WSCB and BE. Note that there are relatively wide 95% CIs for all of the correlations. Confidence intervals for the nontwin sibling correlations for WSCB, BE, and WSCB-BE include zero, indicating that we have limited power to differentiate between shared environmental influences common to all siblings in a family (C), and shared environmental influences specific to same-age siblings (T). Moreover, z-tests of equality suggest that the rDZ and the rSib are not significantly different for WSCB, BE, and WSCB-BE (all ps > 0.05); thus, twin-specific shared environmental influences are likely not significant.
Table4 shows the univariate model-fitting results for the full ACTE model and various submodels. Model comparisons indicated that the CE and E models were the only models that could be clearly rejected by χ2 difference tests for both WSCB and BE. Further, with the exception of the CE model, shared environmental influences (c2) were estimated near zero in all other models, indicating little evidence for environmental influences shared by nontwin siblings. Thus, any shared environmental influences that are present appear to be specific to twins or same-age siblings (t2). The univariate ATE, AE, and TE models all provided adequate fit to the data; however, the AE model had the lowest AIC value for both WSCB (780.40) and BE (−206.36). According to this criterion, the best-fitting and most parsimonious model is the AE model for both WSCB and BE. Under an AE model, additive genetic influences accounted for 43% (95% CI: 0.33–0.52) of the overall phenotypic variance for WSCB and 48% (95% CI: 0.36–0.58) of the overall phenotypic variance for BE. Nonshared environmental influences accounted for 57% (95% CI: 0.48–0.68) and 52% (95% CI: 0.42–0.64) of the overall phenotypic variance for WSCB and BE, respectively.
|WSC||ACTE||0.29 (0.00–0.51)||0.00 (0.00–0.25)||0.13 (0.00–0.37)||0.58 (0.49–0.69)||3703.31||783.31||—||—||—|
|AE||0.43 (0.33–0.52)||—||—||0.57 (0.48–0.68)||3704.40||780.40||1.09||2||0.06|
|BE||ACTE||0.39 (0.00–0.58)||0.02 (0.00–0.33)||0.07 (0.00–0.37)||0.52 (0.42–0.65)||2727.39||−202.62||—||—||—|
|AE||0.48 (0.36–0.58)||—||—||0.52 (0.42–0.64)||2727.63||−206.38||0.24||2||0.89|
Parameter estimates and 95% CIs from the full ACTE bivariate model (Model 1 in Table5) are presented in Figure2. Path coefficients that could be dropped from the model without yielding a significant decrement in fit are represented by dotted arrows, whereas those that could not be dropped from the model are represented by solid arrows. Consistent with the univariate findings, the shared environmental effects common to nontwin siblings (c2) were estimated at zero and could therefore be dropped from the model. Thus, C was dropped from all subsequent submodels, resulting in ATE, AE, TE, and E only submodels (Table5). Despite identifying the AE model as the best-fitting model in the univariate analyses (according to AIC), the ATE and TE models still provided adequate fit to the data, as determined by other fit statistics. The use of a bivariate analysis increases power to distinguish between multiple models. When comparing the nested AE (Model 3) and TE (Model 4) models to the ATE (Model 2) model in the bivariate analysis, only the AE model provided an acceptable fit to the data (Δχ2 = 2.98, p = 0.39). Bivariate results corroborate those of the best-fitting univariate twin models, indicating that additive genetic influences account for 42% and 49% of the overall phenotypic variance in WSCB and BE, respectively. The remaining variance is due to nonshared environmental influences. Formulas for how to calculate these estimates based on the path coefficients in Figure1 are presented in Table6.
|Model Number||Model||Model-Fit||Comparative Fit|
|6*||AE, drop a21||6308.76||2930||448.76||38.97||1||<0.001|
|7*||AE, drop a22||6286.66||2930||426.66||16.86||1||<0.001|
|8*||AE, drop e21||6285.19||2930||425.19||15.39||1||<0.001|
|9*||AE, drop a21 e21||6432.02||2931||570.02||162.23||2||<0.001|
|Additive genetic influences|
|a2BE||a212 + a222||0.49|
|WSCB-BE genetic covariance||a11 × a21||0.29|
|WSCB-BE genetic correlation, ra||a11 × a21 / √(a2WSCB × a2BE)||0.64|
|Nonshared environmental influences|
|e2BE||e212 + e222||0.52|
|WSCB-BE nonshared environmental covariance||e11 × e21||0.15|
|WSCB-BE nonshared environmental correlation, re||e11 × e21 / √(e2WSCB × e2BE)||0.27|
|% of phenotypic correlation due to additive genetic influences||(a11 × a21)/(a11 × a21) + (e11 × e21)||0.65|
|% of phenotypic correlation due to nonshared environmental influences||(e11 × e21)/(a11 × a21) + (e11 × e21)||0.35|
Although the AE model (Model 3) provided the best-fit of the first five models in Table5, we wanted to test whether the additive genetic and/or nonshared environmental covariances could be dropped from the model. Thus, we compared the fit of Models 6 through 9 with the fit of Model 3. Models 6 through 9 investigated whether specific path coefficients within the AE model are necessary or can be dropped. Results indicated that all four of the submodels provided a significantly worse fit to the data; both the additive genetic and nonshared environmental covariances could not be dropped from the model (Figure3). These results suggest that there are additive genetic and nonshared environmental influences contributing to the covariance between WSCB and BE. Although the univariate analyses suggested that twin-specific shared environmental effects may contribute to the variance of WSCB and BE, bivariate analyses suggest that these effects are not substantial and do not contribute to the covariation between the two disordered eating characteristics. The genetic correlation between the two disordered eating characteristics was estimated at 0.64 and the nonshared environmental correlation was 0.27. Also, 65% of the expected phenotypic correlation between WSCB and BE was due to common genetic influences, with the remainder of the correlation due to nonshared environmental influences. Although model fits and estimates of genetic and environmental correlations are invariant to the order of the variables in the analysis, for completeness we reversed the order of the dependent variables to examine the residual additive genetic and nonshared environmental effects on WSCB with BE entered first. The parameter estimates and 95% CIs for the genetic and environmental factor loadings were very similar across the two models. Table 6 shows how the genetic and environmental covariations and the covariance components are derived from the parameter estimates shown in Figure3.
This study aimed to examine the magnitude of genetic and environmental influences on weight/shape concerns and behaviors (WSCB) and binge eating (BE), and their covariance, in adolescent and young adult female twins and their female nontwin siblings. Item-level phenotypic factors derived from a seven-item eating pathology screening tool produced two factors: WSCB and BE. In the univariate analyses, the best-fitting models indicated that only additive genetic and nonshared environmental effects influenced WSCB and BE. Similarly, an AE model was the best-fitting model for the bivariate analysis (a2 = 0.42 for WSCB; 0.49 for BE). The genetic correlation between WSCB and BE was 0.64 and the nonshared environmental correlation was 0.27. Also, 65% of the overall phenotypic variance was due to common genetic influences, with the remainder due to nonshared environmental influences. Thus, our findings suggest that there are additive genetic and nonshared environmental factors common to both phenotypes, as well as influences that contribute independently to each phenotype, perhaps via different genetic and environmental mechanisms.
Our univariate biometrical model-fitting results corroborate prior studies in adolescents and adults in suggesting moderate additive genetic and nonshared environmental influences on weight/shape concerns and binge eating.1–5, 7, 8 One exception is the study published by Wade et al.,2 which found no additive genetic influences on weight concerns. Perhaps the age of the two samples impacted these results, as our sample was younger (mean age of 18.37) than the sample from Wade et al.,2 in which women were required to be between 30 and 45 years old to be included in the study. Moreover, bivariate results suggest that only additive genetic and nonshared environmental factors influenced the covariation between WSCB and BE.
An important assumption in the classical twin design is the equal environments assumption. The equal environments assumption assumes that the shared environmental influences contributing to the resemblance of twins for a given trait or phenotype under study are equivalent (i.e., are of the same kind and magnitude) for MZ and DZ twins. If this is violated, one cannot attribute the greater similarity of MZ twins (who are genetically identical) over DZ twins (who share, on average, only half their alleles identical-by-descent) to genetic effects alone. Thus, violations of this assumption will lead to an overestimation of the importance of genetic effects in the classical twin design. Given that twins in more frequent contact with each other are likely to experience more shared environmental experiences than those in infrequent contact, serious violations of the equal environments assumption can be examined by testing whether MZ twins who are in more frequent contact are more similar phenotypically than MZ twins in less frequent contact. Because MZ twins are genetically identical, any differences cannot be attributable to genetic effects but must be environmental in nature. Therefore, we tested whether the MZ correlations for WSCB and BE were significantly different in twin pairs who had frequent contact (i.e., living together or having daily contact; n = 275 for WSCB and n = 276 for BE) and those who had less frequent contact (i.e., less than daily contact; n = 34 for WSCB and BE). The correlation between MZ twin pairs who were in frequent contact was not significantly different from MZ twin pairs in less frequent contact (WSCB: 0.40 vs. 0.27, respectively, p > 0.05; BE: 0.39 vs. 0.42, respectively, p > 0.05). Although no significant differences were detected, we should point out that the number of MZ twin pairs with infrequent contact was low, but expected given the age of our sample; thus, we do not have sufficient data in this sample to conduct a strong test of the equal environments assumption.
In response to a reviewer's concern, we also investigated the extent to which individual differences in body mass index (BMI) may have contributed to the genetic correlation between the two disordered eating characteristics. We examined this relationship in three ways. First, we examined the phenotypic correlations between BMI (calculated as weight [in kilograms]/height [in meters] squared) and WSCB and then BMI and BE. Although the correlations were statistically significant (p-values < 0.001) due to the large number of participants, the point estimates were low (0.10 and 0.17, respectively). Second, we computed the partial phenotypic correlation (controlling for BMI) between WSCB and BE (partial r = 0.31) and found that BMI accounted for only 10% of the phenotypic relationship (zero-order r = 0.35). Third, we tested for significant differences in the MZ and DZ correlations between BMI and WSCB and then BMI and BE. Using z-tests of equality, there were no significant differences between the MZ and DZ correlations for BMI and WSCB or BMI and BE (all p-values > 0.05). Taken together, these data suggest that although BMI makes a modest contribution to the phenotypic correlation between WSCB and BE, it does not appear to influence the genetic correlation between the two disordered eating characteristics. These findings support prior research showing that genetic and environmental influences on disordered eating were not solely accounted for by BMI.19, 32
With respect to the twin-specific shared environmental influences, the full univariate models for both disordered eating characteristics suggested that these influences contribute to individual differences in WSCB and BE. However, effect sizes were relatively modest, estimated at 13% and 7%, for WSCB and BE, respectively. The best-fitting AE model (based on AIC) for both traits, as well as the bivariate model (which is more powerful than the univariate model), suggest that twin-specific shared environment is not an important contributor to these disordered eating characteristics. These results suggest that findings from twin studies on disordered eating characteristics can be generalized to nontwin populations. Still, the number of nontwin siblings in our sample was small, limiting our power to detect these influences.
There were important strengths to our study. First, the use of a late adolescent and young adult sample allowed us to avoid the possible influence of pubertal effects on individual differences in disordered eating characteristics. Second, this was the first study to directly examine whether common genetic and environmental influences on weight/shape concerns and binge eating exist. Last, this was the first study to include nontwin siblings in order to assess whether shared environmental effects were unique to twins (twin-specific shared environmental influences).
A number of limitations should also be noted. First, in our analyses, we required all siblings in a family to be tested within two years of each other (97% within one year) to minimize time-of-measurement influences (e.g., twins and siblings were not assessed on the same day, and there was a trend for twins and their nontwin siblings to be tested further apart in time than cotwins). Second, the use of ordinal data reduced the power of our analyses, despite relatively large sample sizes. Consequently, 95% CIs for the nontwin sibling polychoric correlations included zero, and relatively wide 95% CIs were observed for the parameter estimates in biometrical model-fitting. Third, the internal consistency reliabilities for our ordinal measures were modest (Cronbach's α = 0.55 for WSCB and 0.57 for BE). However, confirmatory factor analysis indicated that two factors best described the data. Fourth, given that some of our participants were as young as 16 years old, it is likely that at least a portion of the participants had not fully passed the age of risk for onset of disordered eating symptoms. Last, we did not examine males in our study, as the number of males who endorsed any of the seven items was much lower than in females in our sample. In addition, it is possible that these items do not adequately measure disordered eating characteristics in the same way across genders. Exploratory factor analyses in males (not shown) indicated only a single factor (versus the two-factor solution observed for females) underlying the data. For this reason, we did not include males in this study.
In summary, similar to other studies, WSCB and BE were moderately heritable (a2 = 0.43–0.49), with a genetic correlation of 0.64 and a nonshared environmental correlation of 0.27. Although the number of nontwin siblings used in the analyses was relatively small, these results provide the first evidence suggesting that findings from twin studies of disordered eating characteristics can be generalized to the general nontwin population.
The authors acknowledge Laura E. Sobik for her earlier work examining the seven-item eating pathology screening tool in male and female twins as part of her doctoral dissertation. In addition, they would like to thank Kelly L. Klump for an earlier review of this manuscript, Carol A. Beresford and Thomas P. Beresford for their thoughts on the eating pathology screening tool, and the families and staff who have made this project a success. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Visit: http://www.ce-credit.com for additional information. There may be a delay in the posting of the article, so continue to check back and look for the section on Eating Disorders. Additional information about the program is available at www.aedweb.org