Sex Differences in Genetic and Environmental Influences on Percent Body Fatness and Physical Activity

Abstract

Background:

Knowing the extent to which genetic and environmental factors influence percent body fatness (%Fat) and physical activity (PA) would be beneficial, since both are tightly correlated with future health outcomes. Thus, the purpose was to evaluate sex differences in genetic and environmental influences on %Fat and physical activity behavior in male and female adolescent twins.

Methods:

Subjects were adolescent (age range 8.3 – 16.6 yr) twins. %Fat (n = 518 twins) was assessed by bioelectrical impedance analysis (BIA) and PA (n = 296 twins) was measured using 3-Day PA Recall. Each activity was converted to total MET-minutes. Univariate twin models were used to examine sex differences in genetic and environmental factors influencing %Fat and PA.

Results:

%Fat was influenced by genetic effects in both boys and girls (88% and 90%, respectively), with slightly higher heritability estimates for girls. PA was influenced solely by environmental effects for both sexes with higher shared environmental influences in boys (66%) and higher nonshared effects in girls (67%).

Conclusions:

When developing interventions to increase PA in adolescents, it is important to consider the environment in which it takes place as it is the primary contributor to PA levels.

INTRODUCTION

Physical activity (PA) and body size are related to weight gain in adulthood and influenced by both genetics and environment^1–3. However, the relative contribution of genetics to PA shows a wide (0 – 83%) and inconsistent range^1–3. One reason for this wide range in heritability estimates is that studies often only consider one particular aspect of PA behavior, such as leisure-time PA⁴, sports participation⁵, or physical fitness⁶. It is also possible that geographic and cultural factors may also slightly influence variation⁷. Fewer studies have examined the contribution of genetics to PA across a wide range of activities, especially during childhood and adolescence⁸. Using methods that limit PA domains discount valuable information regarding other PA accrued throughout the day. To adequately examine the influence of genetics and the environment on PA, investigators must consider capturing more than one particular PA domain.

Twin research provides investigators a unique opportunity to study the magnitude of genetic and environmental influences on PA participation. Monozygotic (MZ) twins are assumed to be genetically identical, while dizygotic (DZ) twins are assumed to share on average half of their genetic material^9,10. Thus, any MZ intra-pair differences are due solely to environmental factors and measurement error, while any intra-pair differences in DZ twins are also a function of genetic influences. The use of twin data allows for the examination of the extent to which genetic and environmental factors influence a variety of behaviors and characteristics, such as PA.

Although there is some research examining genetic and environmental influences on PA, findings have been inconsistent. Using the University of Washington Twin Registry, Duncan et al. examined 1,389 same-sex twins (M age = 30 yr) and found that shared (i.e., factors that co-twins share, making them more similar) and nonshared (i.e., factors unique to co-twins that contribute to their behavioral differences) environmental effects (18% and 71%, respectively) influenced variation in PA levels more than genetic factors (11%)¹¹. Of particular importance was a finding that when using the cut-point of 150 min/week to categorize twins as exercisers or non-exercisers, the best fitting model included shared and nonshared environmental influences with the nonshared environment contributing 72% of the variance in phenotype, which also included any measurement error¹¹. However, other research has shown different results. Carlsson et al. studied 13,362 twin pairs (age range=14–46 yr) and showed that genetic factors may have played a more important role than environment in PA⁴, accounting for approximately 57% of the influence on PA in males and approximately 50% in females⁴. Furthermore, De Moor et al. examined 3, 525 adolescent twins (13–18 yr) and found contrasting results between boys and girls for the genetic contribution (42% vs. 36%, respectively), shared environmental factors (41% vs. 52%, respectively), and nonshared environmental factors (14% vs. 12%, respectively)¹². These contrasting findings result in differing opinions for improving PA levels. It is not entirely clear why findings regarding the heritability of PA are inconsistent; however, discrepancies may be due to differences in study design, analytical methodology, operational definitions of PA and exercise¹³, and age differences in the samples examined. Thus, the present study aimed to determine the extent to which genetic and environmental factors influence PA in children and adolescents.

Knowing the extent to which genetic and environmental factors influence percent body fatness would be beneficial, since PA (a modifiable behavior) and percent body fatness may be correlated. In a review of twenty-four studies, most of the articles showed a protective effect of PA on adiposity; however, others demonstrated a null association¹⁴. Previous researchers have also found that adolescents who participate in exercise and competitive sports were more likely to exercise in adulthood. This finding demonstrates a need to target this age group, as PA behavior can track into adulthood¹⁵. Also, examining potential sex differences should be considered as they may partially account for PA heritability differences among adolescents³ which could influence body size and fatness. Thus, determining the influence of genetic, shared environmental, and nonshared environmental effects on both body size and PA levels will enable researchers to implement more effective PA interventions in a school-based setting. The purpose of this study was to evaluate sex differences in genetic and environmental influences on body composition and PA behavior in male and female adolescent twins.

METHODS

Subjects:

Subjects were drawn from a population-based twin registry (Michigan State University Twin Registry¹⁶) focused on understanding risk factors for health behaviors and internalizing and externalizing psychiatric disorders across development. The study was approved by the Michigan State University Institutional Review Board. All subjects were informed of potential study risks and benefits before verbal assent (subjects) and written informed consent (parents) were obtained. Subjects included 518 adolescent twins for the percent body fat (%Fat) analyses (n = 206 males, 110 monozygotic [MZ] and 96 dizygotic [DZ]; and n = 312 females, 162 MZ, 150 DZ). Although height and weight were measured on all study participants, percent body fat was not obtained for a subset of twins who were assessed in their homes (instead of the laboratory). Thus the final sample size for %Fat analyses included 503 adolescent twins (n = 196 males, 100 MZ, 96 DZ; and n = 307 females, 161 MZ, 146 DZ). A smaller sample of 296 adolescent twins was used for the PA analyses (n = 94 males, 52 MZ, 42 DZ; n = 202 females, 92 MZ, 110 DZ). Exclusion criteria included twins using oral contraceptives, antibiotics, psychotropic or other medications, or individuals with known clinical illnesses or a history of health problems that could affect hormone levels (e.g., diabetes). Twin zygosity was determined through five standard, within-pair physical similarity questions that have been shown to be over 95% accurate¹⁷.

Procedures:

Twin pairs were requested to come to the laboratory together for their study visit. Anthropometric measurements included height (m) and body weight (kg, in light clothing and no shoes) with a stadiometer and physician’s scale, respectively. Body mass index was calculated as kg^.m^-2.

Instruments:

Percent body fatness was measured according to standard procedures with bioelectrical impedance analysis (BIA: RJL Quantum II analyzer; Clinton Township, MI). Subjects were instructed to refrain from eating or drinking anything (including water) for four hours prior to assessment to reduce the impact of hydration status on the percent body fatness measure. The equation of Houtkooper¹⁸, which was developed on 10–19 year old boys and girls was used to determine fat free mass (kg). Percent body fatness was then calculated by subtracting fat free mass from body weight, then dividing by body weight, and multiplying the result by 100.

After anthropometric data were obtained, the study coordinators took each twin into separate rooms for questionnaire completion. Physical activity was assessed using the 3-Day Physical Activity Recall (3D-PAR)¹⁹. Each child was asked about his/her previous day’s activities for three consecutive days in 30-minute blocks from 7:00 am to 12:00 am (midnight). Whenever possible, the children were asked about 2 weekdays and 1 weekend day. Metabolic equivalent (MET) values were then assigned to each activity based on the Compendium of Physical Activities²⁰. MET values for each day were summed to find total MET-minutes over the three day period. Also, blocks spent in moderate (3.0–5.99 METs) to vigorous (>6.0 METs) PA (MVPA) were then combined and summed. This was done to determine the number of children meeting current pediatric PA recommendations (≥60 minutes of MVPA per day)²¹. Given that it is a daily recommendation, the children needed to report at least two 30 minute blocks of MVPA on all three days to be considered meeting recommendations.

Data Analysis:

To prepare the data because of the relatively large age-range in our sample and the association between age, PA, and percent body fat, twin age was regressed out of the PA and percent body fat variables prior to analyses²². Tests for normality of variance were conducted between members of twin pairs within same-sexed zygosity groups.

Descriptive statistics included Mann-Whitney U-Tests because the data were non-normally distributed. The student’s t-test was used to compare PA between zygosity groups and chi-square tests were used to determine whether twins met the current PA recommendations²¹.

Twin Correlations: Intraclass twin correlations (e.g., Twin 1’s percent body fatness with Twin 2’s percent body fatness) were calculated to provide initial indications of genetic and environmental influences on percent body fatness and PA separately by sex. These correlations were used to provide indications of genetic and environmental factors on each phenotype individually across boys and girls. Greater MZ than DZ twin correlations suggests that genes influence the trait or phenotypic association in question. By contrast, similar MZ and DZ twin correlations suggest that shared environmental factors are important. Finally, nonshared environmental influences (which include measurement error as well as child-specific environmental influences) are implicated if the MZ twin correlations are less than 1.00. Correlations were also conducted to determine the relationship between PA and percent body fatness.

Univariate Twin Models:

Univariate twin models were fit to raw data using the maximum likelihood option in Mx²³ to directly examine additive genetic (A: the effect of individual genes summed over loci that acts to increase twin similarity relative to the amount of genes shared), shared environmental (C: the part of the environment common to siblings that makes them similar to each other), and nonshared environmental (E: environmental factors, and measurement error, that make co-twins within a pair different from each other) influences on percent body fatness and PA, separately (see Figure 1). The raw data option in Mx considers missing data to be missing at random and can lead to less biased estimates than pairwise or listwise deletion²³. When fitting models to raw data, variances, covariances, and means are first freely estimated to obtain a baseline index of fit (minus twice the log-likelihood; −2lnL). The −2lnL under this unrestricted baseline (saturated) model is then subtracted from the −2lnL under more restrictive biometric models. This result is a likelihood-ratio chi-square test of goodness of fit for the model (χ²), which is then used to calculate the Akaike’s information criterion (AIC)²⁴, AIC=χ²-2df, the traditional fit index of behavioral genetics research. The AIC measures model fit relative to parsimony and lower AIC values indicate the preferred model. In addition to the AIC, the χ² difference test (∆χ²) can also be used as a measure of model fit. The ∆χ² examines changes in chi-square values and degrees of freedom between nested models. If the ∆χ² is nonsignificant, then the most parsimonious (i.e., reduced) model is preferred.

Figure 1.

The Basic Twin Model: Model fitting for univariate individual differences model for body composition and physical activity. A = additive genetic variance, C = shared environmental variance, E = nonshared environmental variance. Paths a, c and e are the effects of A, C and E on the quantitative trait of interest (e.g., percent body fatness). The genetic correlation (r_G) is 1.0 for monozygotic (MZ) twins and 0.5 for dyzygotic (DZ) twins. The shared environmental correlation (r_C) is assumed to be 1.0 for MZ and DZ twins reared together.

Initially, a fully unconstrained (i.e., sex-differences) ACE model was fit to the data. The ACE unconstrained model allows variances, covariances, and means to be estimated freely by minimizing the −2lnL of data. Next, a fully constrained ACE model was examined to directly test the presence of sex differences in effects. If the unconstrained model provides the best fit, then sex differences in estimates are present. By contrast, if the constrained model is the best-fitting model, sex similarities in parameter estimates are indicated. Submodels that differentially constrained A, C, and E were then fit to the data to determine which was significant across sex.

RESULTS

The data were normally distributed within sex. Results from tests for normality of variance indicated that the variances were not significantly different between members of twin pairs within same-sexed zygosity groups (all p values greater than .47). Subjects’ physical characteristics and BMI percentiles²⁵ can be seen in Table 1. In our sample, the boys’ average was 21.0 ± 7.0% fat and the girls’ average was 26.5 ± 7.5% fat. There were no significant differences in BMI or %Fat by zygosity in either sex. When examining the PA data, total number of MVPA blocks averaged across three days was 12.7 ± 11.0 blocks for boys and 9.3 ± 7.3 blocks for girls, with no differences by zygosity in either sex. Percent of twins who met the current pediatric PA recommendations of ≥60 minutes of MVPA per day across three days was 37.2% in boys and 29.7% in girls, again with no difference by zygosity in either sex.

Table 1.

Descriptive statistics of the subjects with PA data and %Fat data. Mean ± Standard Deviation.

	Physical Activity (n = 296 twins)		Percent Body Fat (n = 518 twins)
	Boys (n=94 twins)	Girls (n=202 twins)	Boys (n=196–206 twins)	Girls (n=307–312 twins)
Age (yrs)	12.3 ± 1.4	12.5 ± 1.4	12.8 ± 1.6	12.6 ± 1.5
Height (cm)	153.2 ± 11.8	154.3 ± 10.4	156.0 ± 12.8	154.5 ± 10.0
Weight (kg)	46.7 ± 15.0	49.9 ± 13.8	49.7 ± 15.8	49.2 ± 13.1
BMI (kg·m−2)	19.6 ± 3.9	20.8 ± 4.5	20.0 ± 4.1	20.5 ± 4.3
BMI Percentile (%)	50–75	75–85	50–75	50–75
Fatness (%)	21.0 ± 7.9	26.5 ± 7.5	20.8 ± 7.7	25.9 ± 7.5
Physical Activity	201.3 ± 53.1	184.2 ± 30.3	--	--

Note. There were no zygosity differences within gender. Sample size varies for percent body fat due to missing data. BMI percentile was determined according to the Centers for Disease Control and Prevention website.²⁵

Twin Analyses:

Twin intraclass correlations for %Fat and PA are presented in Table 2. MZ twin correlations were larger than DZ twin correlations for %Fat. By contrast, shared and nonshared environmental factors appeared to predominate for PA for boys, as MZ/DZ twin differences in correlations were non-significant, and the MZ correlation was less than 1.00. Unexpectedly, the DZ twin correlation for PA was larger than the MZ twin correlation in both boys and girls. However, the confidence intervals for these correlation estimates largely overlap, thus it is not clear whether these estimates would differ with a larger sample. Overall, the phenotypic correlation between percent body fat and PA was −0.14 (p = 0.02). When separated by sex, the correlations were r = −0.02 (p = 0.89) for boys and r = −0.22 (p = 0.003) for girls.

Table 2.

Twin intraclass correlations for %Fat and PA (total MET-minutes) in girls and boys.

Variables	Girls				Boys
	MZ (n = 40–81 pairs)	DZ (n = 49–73 pairs)	z	pa	MZ (n = 21–50 pairs)	DZ (n = 15–48 pairs)	z	pa
Twin Intraclass Correlations
Percent Body Fat	.90* (.87, .94)	.57* (.45, .67)	4.63	<.01	.91* (.87, .94)	.47* (.30, .61)	4.85	<.01
Physical Activity	.28* (.06, .47)	.43* (.26, .58)	−0.80	.21	.59* (.35, .76)	.77* (.58, .89)	−0.94	.17

Note. MZ = monozygotic; DZ = dizygotic. The z statistic examines the difference between MZ and DZ twin correlations.

^*p < .05, two-tailed. The correlation is significantly different from zero. Significant differences between MZ and DZ correlations are indicated by boldface.

^aThe p value is one-tailed.

Standardized parameter estimates and fit statistics for each model estimated are included in Tables 3–4. As seen there, in the unconstrained ACE model, the shared environmental estimate for %Fat was non-significant (as indicated by confidence intervals that overlapped with zero) in both boys and girls. Since the unconstrained ACE model provided a better fit to the data than the constrained ACE model, the next model examined was the unconstrained AE model. Not surprisingly, the unconstrained AE model, in which the C parameter is dropped from the model, provided a better fit to the data than did the unconstrained ACE model. This finding indicates that while the genetic and environmental variances may vary in boys and girls in terms of their magnitude, they are similar across sex. The best-fitting model allowed the heritability of %Fat to differ across sex with large estimates for both sexes.

Table 3.

Parameter Estimates and Fit Statistics from the Unconstrained and Best-Fitting Twin Models for percent body fat.

	Standardized Parameter Estimates						Fit Statistics
	Girls			Boys
Model	A	C	E	A	C	E	−2lnL (df)	χ2(df)a	Δχ2 (df)b	p	AIC
Baseline	--	--	--	--	--	--	1096.11 (487)	--	--	--	--
ACE, unconstrained	.64 (.41, .92)	.27 (0, .50)	.09 (.06, .13)	.80 (.41, .87)	0 (0, .36)	.20 (.13, .32)	1110.41 (499)	14.30 (12)	--	--	-9.70
ACE, constrained to be equal across sex	.64 (.43, .90)	.25 (0, .46)	.11 (.08, .14)	.64 (.43, .90)	.25 (0, .46)	.11 (.08, .14)	1140.53 (502)	44.42 (15)	30.12 (3)	<.01	14.42
AE, unconstrained	.90 (.86, .93)	--	.10 (.07, .14)	.88 (.81, .92)	--	.12 (.08, .19)	1113.80 (501)	17.69 (14)	3.39 (2)	.18	−10.31
CE	--	.74 (.66, .80)	.26 (.20, .34)	--	.76 (.67, .83)	.24 (.17, .33)	1171.12 (501)	75.01 (14)	60.71 (2)	<.01	47.01
E	--	--	1.0 (1.0, 1.0)	--	--	1.0 (1.0, 1.0)	1379.30 (503)	283.19 (16)	269.89 (4)	<.01	251.19

Note. The unconstrained ACE model allows all genetic (A) shared environmental (C), and nonshared environmental (E) influences were allowed to vary across sex. Unless specified, models allow A, C, and E to differ across sex. Parameter estimates are standardized and range from −1.0 to 1.0. Age was regressed out of both variables prior to model fitting. The best-fitting models are indicated by boldface. Ninety-five percent confidence intervals are in parentheses. 2lnL = −2 times log likelihood of data; df = degrees of freedom; AIC = Akaike’s information criterion.

^aχ² is computed by subtracting the-2lnL of the baseline model from the −2lnL of the submodel; the AIC is calculated using this fit statistic (i.e., AIC = χ² – 2*df)

^bΔχ² compares submodels to the full ACE model

Table 4.

Parameter Estimates and Fit Statistics from the Unconstrained and Best-Fitting Twin Models for physical activity.

	Standardized Parameter Estimates						Fit Statistics
	Girls			Boys
Model	A	C	E	A	C	E	−2lnL (df)	χ2(df)a	Δχ2 (df)b	p	AIC
Physical Activity
Baseline	--	--	--	--	--	--	650.84 (249)	--	--	--	--
ACE, unconstrained	0 (0, 0)	.34 (0, .50)	.66 (.50, .85)	0 (0, 0)	.65 (.13, .80)	.35 (.20, .57)	680.50 (261)	29.66 (12)	--	--	5.66
ACE, constrained to be equal across sex	0 (0, .15)	.51 (.33, .63)	.49 (.37, .63)	0 (0, .15)	.51 (.33, .63)	.49 (.37, .63)	712.60 (264)	61.76 (15)	32.10 (3)	<.01	31.76
CE, constrained to be equal across sex	--	.51 (.37, .63)	.49 (.37, .63)	--	.51 (.37, .63)	.49 (.37, .63)	712.60 (265)	61.76 (13)	32.10 (1)	<.01	35.76
CE, A constrained to 0	--	.33 (.14, .49)	.67 (.51, .86)	--	.66 (.44, .81)	.34 (.19, .56)	680.50 (263)	29.66 (14)	0.00 (2)	.99	1.66

Note. The unconstrained ACE model allows all genetic (A) shared environmental (C), and nonshared environmental (E) influences were allowed to vary across sex. Unless specified, models allow A, C, and E to differ across sex. Parameter estimates are standardized and range from −1.0 to 1.0. Age was regressed out of both variables prior to model fitting. The best-fitting models are indicated by boldface. Ninety-five percent confidence intervals are in parentheses. 2lnL = −2 times log likelihood of data; df = degrees of freedom; AIC = Akaike’s information criterion.

^aχ² is computed by subtracting the-2lnL of the baseline model from the −2lnL of the submodel; the AIC is calculated using this fit statistic (i.e., AIC = χ² – 2*df)

^bΔχ² compares submodels to the full ACE model

Table 4 includes standardized parameter estimates and fit statistics for the PA models. As shown in the table, in the unconstrained ACE model, genetic effects were estimated at zero for both boys and girls, while shared environmental influences were moderate in magnitude. As with %Fat, the unconstrained ACE model provided a better fit to the data than the constrained ACE model for PA. Thus, subsequent models allowed C and E to vary and tested whether the best-fitting model included A constrained across sex or A constrained to zero in both sexes. Not surprisingly, the best fitting model was one in which the A estimate was constrained to be zero in both sexes, but both C and E were estimated and allowed to vary across sex. This model had a non-significant change in chi-square from the unconstrained ACE model and had a lower AIC than the unconstrained model (see Table 4) as well as a CE model constraining estimates across sex. Results from this best fitting model indicated that although individual differences in PA were due to environmental influences only, the magnitude of environmental effects varied across sex, with greater shared environmental influences in boys and greater nonshared environmental influences in girls.

DISCUSSION

Overall, our results showed sex differences in estimates of genetic and nonshared environmental contribution to percent body fatness. Similarly, sex differences were indicated in the environmental influences on total MET-minutes; with the best-fitting model indicating no genetic influences on total MET-minutes. Shared environmental influences on PA were greater in boys compared to girls, whereas nonshared environmental influences were significantly larger in girls as compared to boys. In contrast, for both boys and girls, the genetic influences were greatest for percent body fatness.

Findings indicating that genetic influences of similar magnitude were primarily influential for %Fat in both sexes corroborated previous research^2,26. Our results are in line with those found in a study of 66 twin pairs aged 3–17 yr in which they also found genetic factors to be the most influential (75–80%) in the determination of percent body fatness as determined by BIA²⁷. In a sample (n = 5,092) of younger twins (8–11 yr), waist circumference and BMI were also found to be highly genetically heritable (77% for both)²⁸. This result has also been found in adult twins (79% in males and 78% in females for BMI and 56% in males and 71% in females for WC), but higher levels of PA attenuated the genetic component, thus demonstrating that those who are more genetically susceptible to obesity could benefit most from a PA intervention²⁹. In addition, we found an overall negative correlation between PA and percent body fatness; however, this did not hold true when separated by sex. These results corroborate other research finding a protective or null effect of PA on percent body fatness in adolescents^14,30. Consequently, PA programs should continue to be used not only for improving PA levels, but also for reducing percent body fatness as there may be a potential relationship. While percent body fatness is influenced primarily through genetics, environmental influences, especially the nonshared, should still be given consideration in the development of PA interventions as they do appear to make a minor impact and is considered a modifiable behavior.

Sex differences were suggested for environmental influences on PA, with higher shared environmental influences in boys, and higher nonshared environmental effects in girls. The lack of genetic influence in both sexes corroborates other research showing nonshared environmental factors providing the strongest influence on one hundred 4–10 year old twins² and same-sex adult twins’ PA levels¹¹. Collectively, these results may imply that environmental factors play a stronger role in PA levels than genetic factors across the developmental spectrum; however, the latter studies did not analyze PA by sex which may account for part of the variation.

Sex differences in the type and magnitude of environmental effects on PA is interesting and worth exploration. On average, boys spend 40% more time in vigorous PA, which includes both physical education classes in school and PA outside of school³¹. Data in our sample corroborate these results, as boys participated in more total PA than girls (201.3 ± 53.1 MET-minutes vs. 184.2 ± 30.3 MET-minutes, respectively) and MVPA blocks across 3 days (12.7 ± 11.0 blocks vs. 9.3 ± 7.3 blocks, respectively). It is possible that twin girls were more likely to explore different sport and PA opportunities than their co-twins (resulting in increased nonshared environmental influences), whereas the boys may have participated in similar levels and types of sports (resulting in increased shared environmental effects).

However, previous twin study results are contradictory with regard to the relative influence of shared versus nonshared environmental influences on PA. One previous study by Franks et al. corroborated our results in a combined-sex sample of 4–10 year old boy and girl twins where, using an adjusted model of PA energy expenditure (adjusted for body weight, age, study date, season, ethnicity, and sex), shared environmental influences accounted for 69% while nonshared environmental influences accounted for 31% of the variance². In contrast, in a study of 12–25 year old Portuguese twins that examined leisure time PA³, nonshared environmental factors accounted for 37% of the variance, while shared influences approximated zero, and genetic influences approximated 63% of the variance in male twins. In females, shared environmental influences explained 38% of the variance, while nonshared influences accounted for 30% influence³. The differential influences in PA may be in part due to differences in ages among the samples as well as how PA was measured (PA energy expenditure² and leisure time PA³ versus total PA in the current study). While the present data are equivocal with other research, overall, environmental factors (shared and/or nonshared) appear to be the most significant contributors to PA behaviors in in our and other samples.

Our results support the notion that environmental factors play an integral role in PA levels among adolescents and therefore should be considered through school-based PA programs to combat obesity and for general well-being. Further, involvement of peers and the community improves results³², thus suggesting that school-based PA programs need to extend into the family and community. When extending PA programs from the school to the community they tend to be more effective, which in turn can impact body size variables and the other positive health benefits associated with increased PA.

Although this study had several strengths, there were also some limitations. One limitation was recall of PA for 3 consecutive days. To reduce errors in the PA measurements, each twin completed the 3D-PAR with a trained technician and in a separate room from their twin to reduce bias between the answers. Additionally, although the data were recalled over a 3-day period, using the 3D-PAR and analyzing total MET-minutes allowed us to capture numerous activities ranging from low to high intensity. Therefore, we likely eliminated bias in our results by having the subjects report the specific activity and assigning a MET value to it, rather than asking if the activity in the 30-minute block was “light,” “moderate,” or “vigorous,” which could differ based on the participant’s fitness level. In addition, while we estimated percent body fatness using laboratory methods, there are inherent limitations to accurately assessing body composition on adolescents overall due to maturation status (which was not collected).

In addition, power analyses indicated that for boys and girls (as well as both sexes together), more than 1000 twins are needed to attain 90% power to detect shared environmental variances of 20% for body fat and physical activity. Specifically, for girls, samples of 1140 and 1017 were required for body fat and physical activity, respectively. Similarly, for boys, samples of 1113 and 1325 were required for body fat and physical activity, respectively. The sample size needed for a model constraining the shared environmental variance across sex is also above 500 twin pairs. This is not unexpected, as previous researchers have argued that very large sample sizes are needed to detect shared environmental effects^33,34. Indeed, this is a limitation of the classical twin design. Importantly, however, using continuous data, fewer pairs are needed than with categorical variables, and findings with smaller samples provide initial indications of the impact of genetic and environmental factors on a variable of interest. Although our sample is smaller than the size suggested by these power analyses, our estimates of A, C, and E correspond to estimates based on our correlation coefficients as well as parameter estimates from some previous twin studies examining these variables in adolescents ^2,26. Further, the best fitting models for physical activity had large estimates of C, suggesting that even with a smaller sample, significant estimates of C were detected. Nonetheless, a sample of greater than 500 twin pairs is often necessary to detect the “true” variance underlying a trait^33,34. Thus, larger samples are needed to confirm our findings.

The difference between the magnitude of nonshared environmental influence between boys and girls for total MET-minutes was unexpected. Future research is needed to confirm or refute these results among larger, more diverse samples. Future studies should also be designed to assess a wide range of PA, including low, moderate, and vigorous activity with objective devices, such as accelerometry. In this way, investigators will possibly be able to more clearly delineate the genetic contribution to the intensity of PA which may shape how schools can best implement PA programs among this age group.

In conclusion, percent body fatness is greatly influenced by genetic factors for both boys and girls. However, PA is determined primarily via environmental influences in both boys and girls with higher non-shared influences for girls and higher shared influences for boys. Thus, when developing interventions to increase PA in adolescents, it is important to consider the environment which is the primary contributor to PA levels.