The Beneficial Effects of Physical Activity: Is It Down to Your Genes? A Systematic Review and Meta-Analysis of Twin and Family Studies

Search Strategy

We conducted a systematic review and meta-analysis in accordance with the “Preferred reporting items for systematic reviews and meta-analyses” (PRISMA) statement [21]. The protocol for this systematic review has been registered on PROSPERO (Registration No: CRD42015020056). We performed electronic database searches in MEDLINE, CINAHL, EMBASE, SPORTS Discuss, AMED, PsycINFO, WEB OF SCIENCE, and SCOPUS from the earliest records to May 2015. The search was then updated in March 2016. We used a comprehensive key word search strategy (Additional file 1) combining key words relating to PA (e.g., “physical activi*” OR “exercise” OR “resistance training” etc.) and genetics (e.g., “genetic*” OR “herita*” OR “family resemblance” etc.). The search strategy remained sensitive to capture all outcomes related to body composition and cardiorespiratory fitness. To identify additional studies we performed a hand search of the reference lists from included papers.

Study Selection

Two reviewers (TA and JZ) independently performed the selection of studies and consensus was used to resolve any disagreement. Studies were included if they investigated clinically relevant outcome measures of body composition or cardiorespiratory fitness following a PA or exercise intervention (referred to hereafter as PA interventions) amongst twin pairs and/or family members. Studies investigating a PA intervention in combination with other interventions (e.g., diet) were included. We included randomised controlled trials and case series provided they reported a within MZ twin pair correlation (r_MZ), heritability estimate (from a twin study), or maximal heritability estimate (from a family study). Heritability estimates and the r_MZ for the response of an intervention (based on change scores) are commonly reported in studies where twin pairs are considered as clusters, with the treatment effect as a fixed variable [22]. To investigate the intra-pair resemblance in the response to PA it is essential that twin pairs participate in an identical intervention. This is similar to the methodology employed in family studies to obtain a maximal heritability estimate (where the variance explained by genetic and shared environmental factors cannot be teased apart). Therefore, we decided not to use methodological quality as part of the inclusion/exclusion criteria as it is not practical to consider items commonly assessed in systematic reviews of randomized controlled trials (such as allocation concealment, blinding, and intention-to-treat) [23] when considering this study design. It is unlikely results from twin and family studies investigating heritability are subject to publication bias, since the contribution of genetics and shared environment is relevant regardless of whether the estimates are small or large. However, we acknowledge the possibility that individual studies may only report results for traits that demonstrate a high heritability. To minimize the risk of reporting bias, we contacted authors when there was data available on body composition and cardiorespiratory fitness but within twin pair correlations were not reported. Observational studies or studies only assessing the heritability of PA engagement, without a PA intervention, were excluded. There was no restriction on the age or gender of participants, nor the type of PA intervention investigated. We included published conference abstracts and dissertations provided they met the inclusion criteria.

Data Extraction

Two reviewers (DS and JZ) independently performed the extraction of data. A standardized data extraction form was used to collect data on participants’ characteristics (age, gender, and zygosity), sample size, prescribed PA intervention (frequency, intensity, duration, and type), co-prescription of other interventions (e.g., diet), outcomes assessed, loss to follow up, and study type.

Data Analysis

Data on correlation (r), equality of variances (F), heritability (h ²), and maximal heritability were extracted from included studies. In family studies, “heritability” estimates were derived from the familial correlation model and termed “maximal heritability”, since the model is unable to partition the variance explained by genetic and non-genetic sources shared within families [24]. In twin studies, heritability estimates were calculated from the following formula: h ² = 2(r_MZ–r_DZ), where r_DZ is the within DZ twin pair correlation. When h ² was greater than 1 we used r_MZ as the heritability estimate, since it is not possible for genetics to contribute more than 100% to the variance of a phenotype. In addition, if there were no data available for DZ twins, we used r_MZ as an estimate of the upper bound of heritability (including variance from genetic and shared environmental factors). In cases where the F-ratio was reported but the r_MZ was not, we used the following formula to calculate r_MZ as described by Haggard: r = (F–1)/(F + 1) [25]. Authors were contacted when required data were not published. When raw data were obtained from twin studies, we attempted to fit variance components models to change scores in order to estimate the r_MZ and r_DZ simultaneously and formally compare models in which these two parameters were forced to be equal with models in which they were allowed to differ. However, for many phenotypes the models could not be fitted or failed to converge due to small sample sizes (no results shown from these models). Instead, for all phenotypes, and separately for MZ and DZ twin pairs, we performed a one-way (twin pair identifier) analysis of variance with change score as the outcome (calculated from the pre and post-intervention raw data). Specifying change score as a repeated measure within a twin pair in the models enabled calculation of the within twin pair correlation. When possible and applicable, we adjusted the analyses for age, gender, and baseline values [22]. If studies were considered homogenous in terms of outcomes and PA interventions, we performed a meta-analysis using Comprehensive Meta-Analysis Version 3.0. Additionally, if there were enough studies investigating PA interventions of varying durations, the co-prescription of other interventions (e.g., diet), or analyzing data from males and females separately, we sub-grouped our meta-analyses accordingly. If pooling data on heritability/maximal heritability was not possible (from either twin or family studies), we attempted to pool data on the r_MZ. Data on correlation and sample size from each study with greater than or equal to four twin pairs (the minimum number of observations allowed to be entered into the software) was used to provide a pooled estimate of the r_MZ, 95% confidence interval (CI), and p-value. Heterogeneity between studies was assessed using the I ² statistic. An I ² value <25% indicates low heterogeneity between studies. We used fixed-effects where I ² was <50% and random-effects when I ² was ≥50% (moderate heterogeneity). We did not display pooled estimates where the I ² value indicated high heterogeneity (≥75%) [26].

Description of Studies

The comprehensive key word search yielded 27,830 results, with one additional study retrieved from hand searching the reference lists of included studies. After removing duplicates and screening titles and abstracts there were 224 full texts which were screened. A total of 14 studies (nine twin and five family studies) were included in this systematic review, with eight twin studies forming the basis for our meta-analyses (Fig. 1). The nine twin studies included data from a total of 83 complete MZ twin pairs, and 15 complete DZ twin pairs, with no twin pairs used in more than one study (as confirmed by authors named in multiple included studies). The five family studies were based off the same sample of 199 families (which did not include any twin pairs). Although there were numerous twin and family studies similar in design and outcomes, we were unable to pool heritability estimates for any outcomes for two main reasons. First, there were an insufficient number of family studies deriving results from independent samples. Second, although we were able to obtain heritability estimates from three twin studies, these estimates were uninformative since the 95% CI covered the whole range (0,1) (apart from Danis and colleagues who estimated heritability without utilizing DZ twins in its design [27]), and differences between the r_MZ and r_DZ were not statistically significant (Table 1).Instead, we were able to pool the r_MZ for selected outcomes, giving us quantitative estimates of the upper bound of heritability. Included studies that reported more than one outcome measure were used in multiple meta-analyses.

Author (year)	Sample	Age [mean (SD)]	Baseline status [mean (SD)]	Within MZ correlation (95% CI)	Within DZ correlation (95% CI)	Between MZ and DZ correlation significanced
Body fat percentage (%)
Hopkins ND (2012)a	6 MZ (1 male and 5 female) and 6 DZ (2 male and 4 female) twin pairs	MZ: 13.5 (0.8) DZ: 13.4 (0.8)	MZ: 27.1 (6.9) DZ: 26.0 (11.3)	0.63 (−0.37 to 0.95)	0.31 (−0.67 to 0.90)	p = 0.606
Afman G (1988)b	18 MZ (2 males and 16 females) and 9 DZ (3 males and 6 females) twin pairs	MZ: 19.0 (1.4) DZ: 19.4 (1.8)	MZ: 21.3 (9.0) DZ: 19.9 (7.2)	0.61 (0.20 to 0.84)	0.50 (−0.25 to 0.87)	p = 0.742
Danis A (2003)	9 MZ male twin pairs	11–14c	E: 17.8 (4.1) C: 16.8 (2.8)	*	*	h 2 = 69%**
BMI
Hopkins ND (2012)a	6 MZ (1 male and 5 female) and 6 DZ (2 male and 4 female) twin pairs	MZ: 13.5 (0.8) DZ: 13.4 (0.8)	MZ: 21.5 (3.5) DZ: 21.9 (3.5)	0.81 (0.00 to 0.98)	0.57 (−0.45 to 0.94)	p = 0.557
Afman G (1988)b	16 MZ (3 males and 13 females) and 6 DZ (2 males and 4 females) twin pairs	MZ:18.6 (1.1) DZ: 19.3 (1.3)	MZ: 21.9 (1.9) DZ: 22.6 (3.7)	0.42 (−0.10 to 0.76)	0.00 (−0.81 to 0.81)	p = 0.485
Weight (kg)
Hopkins ND (2012)a	6 MZ (1 male and 5 female) and 6 DZ (2 male and 4 female) twin pairs	MZ: 13.5 (0.8) DZ: 13.4 (0.8)	MZ: 59.0 (11.5) DZ: 58.9 (12.6)	0.89 (0.28 to 0.99)	0.00 (−0.81 to 0.81)	p = 0.091
Afman G (1988)b	19 MZ (3 males and 16 females) and 9 DZ (3 males and 6 females) twin pairs	MZ: 18.7 (1.0) DZ: 19.4 (1.8)	MZ: 60.4 (10.6) DZ: 67.1 (13.4)	0.53 (0.10 to 0.79)	0.13 (−0.58 to 0.73)	p = 0.337
Fat free mass
Hopkins ND (2012)a	6 MZ (1 male and 5 female) and 6 DZ (2 male and 4 female) twin pairs	MZ: 13.5 (0.8) DZ: 13.4 (0.8)	MZ: 69.9 (6.8)% DZ: 69.9 (6.8)%	0.52 (−0.50 to 0.94)	0.34 (−0.65 to 0.90)	p = 0.785
Afman G (1988)b	19 MZ (3 males and 16 females) and 9 DZ (3 males and 6 females) twin pairs	MZ: 18.9 (1.4) DZ: 19.4 (1.8)	MZ: 48.2 (8.1) kg DZ: 53.2 (13.2) kg	0.40 (−0.07 to 0.72)	0.18 (−0.55 to 0.75)	p = 1.000
Relative VO2 max (mL.kg−1min−1)
Hopkins ND (2012)a	6 MZ (1 male and 5 female) and 6 DZ (2 male and 4 female) twin pairs	MZ: 13.5 (0.8) DZ: 13.4 (0.8)	MZ: 44.4 (8.1) DZ: 45.7 (8.1)	0.43 (−0.59 to 0.92)	0.21 (−0.73 to 0.87)	p = 0.763
Afman G (1988)b	19 MZ (3 males and 16 females) and 9 DZ (3 males and 6 females) twin pairs	MZ: 18.9 (1.4) DZ: 19.4 (1.8)	MZ: 33.3 (7.3) DZ: 37.1 (8.0)	0.44 (0.00 to 0.74)	0.00 (−0.66 to 0.66)	p = 0.324
Danis A (2003)	9 MZ male twin pairs	11–14c	E: 52.1 (3.6) C: 54.0 (3.9)	*	*	h2 = 44%**
Absolute VO2 max (L.min−1)
Afman G (1988)b	20 MZ (3 males and 16 females) and 9 DZ (3 males and 6 females) twin pairs	MZ: 18.9 (1.4) DZ: 19.4 (1.8)	MZ: 2.0 (0.6) DZ: 2.5 (0.9)	0.44 (0.00 to 0.74)	0.00 (−0.66 to 0.66)	p = 0.320
Danis A (2003)	9 MZ male twin pairs	11–14c	E: 2.1 (0.4) C: 2.1 (0.4)	*	*	h 2 = 54%**

MZ monozygotic, DZ dizygotic, E experimental group, C control group, SD standard deviation, CI confidence interval, h ² heritability, VO ₂ max maximal oxygen uptake, BMI body mass index

*No reported correlation due to a different method used to estimate heritability

**Unable to calculate the standard error and thus present the 95% CI

^aWithin twin pair correlations (95% CI) extracted from the publication

^bWithin twin pair correlations (95% CI) calculated from raw data

^cDid not report a mean age (SD)

^dUnable to calculate the within MZ and DZ twin pair correlations for Danis A (2003) due to methodology, so the h ² is presented instead

Due to significant between-study variation for the intervention frequency and duration, we were unable to stratify meta-analyses in this way. Instead, we examined the correlations for each outcome to investigate if studies with more frequent bouts of PA, or longer intervention durations reported higher r_MZ, but, we were unable to identify any trends. We were able to stratify our meta-analyses by the co-prescription of a diet intervention, and by gender.

Outcomes of Body Composition

There were 11 studies (nine twin studies [20, 27–30, 32–35] and two family studies [24, 36]) which investigated body composition measures and their response following a PA intervention. Pooling of eight twin studies results (excluding Danis and colleagues [27] due to different methodology) suggest there is a significant r_MZ across the majority of body composition measures (Table 4). The pooled r_MZ was highest for BMI (r_MZ = 0.69, 95% CI: 0.49–0.82, n = 58) and the ratio of fat mass to fat free mass (r_MZ = 0.69, 95% CI: 0.42–0.85, n = 36) (Fig. 2), where “n” represents the total number of twin pairs from all studies. There were significant pooled r_MZ for fat mass (r_MZ = 0.58, 95% CI: 0.13–0.83, n = 48), fat free mass (r_MZ = 0.57, 95% CI: 0.35–0.73, n = 73) (Fig. 3), body fat percentage (r_MZ = 0.55, 95% CI: 0.32–0.72, n = 72), waist circumference (r_MZ = 0.50, 95% CI: 0.09–0.77, n = 27) and hip circumference (r_MZ = 0.51, 95% CI: 0.11–0.77, n = 27) (Fig. 4). However, the pooled r_MZ was lower and not statistically significantly different from 0 for waist-to-hip ratio (r_MZ = 0.29, 95% CI: −0.16–0.64, n = 27) (Fig. 5).

Outcome	All studies	Studies including a combined physical activity and diet intervention	Studies only including a physical activity intervention
Body fat percentage (%)	0.55 (0.32–0.72)*** (n = 72) Poehlam A et al. (1987) Koenigstorfer J et al. (2011) Hopkins N et al. (2012) Bouchard C et al. (1994) Hainer V et al. (2000) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.61 (0.28–0.82)** (n = 33) Poehlam A et al. (1987) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000)	0.49 (0.16–0.73)** (n = 39) Hopkins N et al. (2012) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)
BMI	0.69 (0.49–0.82)*** (n = 58) Koenigstorfer J et al. (2011) Hopkins N et al. (2012) Bouchard C et al. (1994) Hainer V et al. (2000) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.79 (0.54–0.91)*** (n = 27) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000)	0.58 (0.23–0.79)** (n = 31) Hopkins N et al. (2012) Prud’Homme D et al. (1984) Afman G et al. (1988)
Fat free mass (kg)	0.57 (0.35–0.73)*** (n = 73) Poehlam A et al. (1987) Koenigstorfer J et al. (2011) Hopkins N et al. (2012) Bouchard C et al. (1994) Hainer V et al. (2000) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.71 (0.43–0.87)*** (n = 33) Poehlam A et al. (1987) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000)	0.43 (0.09–0.68)* (n = 40) Hopkins N et al. (2012) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)
Fat mass (kg)	0.58 (0.13–0.83)* (n = 48) Poehlam A et al. (1987) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000) Hamel P et al. (1986) Prud’Homme D et al. (1984)	0.68 (0.16–0.90)* (n = 33) Poehlam A et al. (1987) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000)	0.27 (−0.36–0.73) (n = 15) Hamel P et al. (1986) Prud’Homme D et al. (1984)
Fat mass to fat free mass ratio	0.69 (0.42–0.85)*** (n = 36) Bouchard C et al. (1994) Hainer V et al. (2000) Hamel P et al. (1986) Prud’Homme D et al. (1984)	0.82 (0.58–0.93)*** (n = 21) Bouchard C et al. (1994) Hainer V et al. (2000)	0.30 (−0.33–0.75) (n = 15) Hamel P et al. (1986) Prud’Homme D et al. (1984)
Waist circumference (cm)	0.50 (0.09–0.77)* (n = 27) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000)		–
Hip circumference (cm)	0.51 (0.11–0.77)* (n = 27) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000)		–
Waist to hip ratio	0.29 (−0.16–0.64) (n = 27) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000)		–
Sum of skin folds (cm)	0.67 (0.37–0.85)*** (n = 30) Bouchard C et al. (1994) Hainer V et al. (2000) Prud’Homme D et al. (1984)	0.73 (0.39–0.89)*** (n = 21) Bouchard C et al. (1994) Hainer V et al. (2000)	0.49 (−0.26–0.87) (n = 9) Prud’Homme D et al. (1984)
Trunk fat	0.52 (0.12–0.78)* (n = 27) Hopkins N et al. (2012) Bouchard C et al. (1994) Hainer V et al. (2000)	0.56 (0.13–0.82)* (n = 21) Bouchard C et al. (1994) Hainer V et al. (2000)	0.30 (−0.68–0.89) (n = 6) Hopkins N et al. (2012)
Extremity skin fold (cm)	0.54 (−0.39–0.92) (n = 21) Bouchard C et al. (1994) Hainer V et al. (2000)		–
Trunk to extremity ratio	0.48 (−0.30–0.88) (n = 21) Bouchard C et al. (1994) Hainer V et al. (2000)		–
Weight (kg)	0.67 (0.48–0.79)*** (n = 73) Poehlam A et al. (1987) Koenigstorfer J et al. (2011) Hopkins N et al. (2012) Bouchard C et al. (1994) Hainer V et al. (2000) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.73 (0.47–0.88)*** (n = 33) Poehlam A et al. (1987) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000)	0.61 (0.32–0.79)*** (n = 40) Hopkins N et al. (2012) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)
Absolute VO2 max (L.min−1)	0.38 (0.04–0.64)* (n = 42) Bouchard C et al. (1994) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.52 (−0.38–0.92) (n = 7) Bouchard C et al. (1994)	0.36 (−0.01–0.64) (n = 35) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)
Relative VO2 max (mL.min−1.kg−1)	0.39 (0.07–0.64)* (n = 48) Hopkins N et al. (2012) Bouchard C et al. (1994) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.48 (−0.43–0.91) (n = 7) Bouchard C et al. (1994)	0.38 (0.04–0.64)* (n = 41) Hopkins N et al. (2012) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)

n number of twin pairs, VO ₂ max maximal oxygen uptake, BMI body mass index

*p < 0.05; **p < 0.01; ***p < 0.001

When we pooled data from twin studies that included a combined PA and diet intervention (four studies [20, 28, 30, 33], there was a trend for the r_MZ to be higher across all measures of body composition compared to twin studies that only involved a PA intervention (four studies [29, 32, 34, 35]) (Table 4). The r_MZ for BMI was higher when results were pooled for studies including a combined PA and diet intervention (r_MZ = 0.79, 95% CI: 0.54–0.91, n = 27), compared to studies only involving a PA intervention (r_MZ = 0.58, 95% CI: 0.23–0.79, n = 31) (Fig. 6), although confidence intervals were wide. Meta-analyses for each outcome were stratified by gender. The r_MZ was variable between males and females, depending on the outcome assessed (Table 5), with wide confidence intervals observed for both males and females. The pooled r_MZ for the response of fat mass following PA was higher and statistically significant in females (r_MZ = 0.85, 95% CI: 0.63–0.94, n = 25) compared to males (r_MZ = 0.40, 95% CI: −0.26–0.81, n = 17) (Fig. 7). However, the pooled r_MZ for fat free mass was higher in males (r_MZ = 0.80, 95% CI: 0.39–0.95, n = 17) compared to females (r_MZ = 0.52, 95% CI: 0.19–0.75, n = 38) (Fig. 8), both being statistically significantly different from 0 but not from each other.

Outcome	All studies	Females	Males
Body fat percentage (%)	0.55 (0.32–0.72)*** (n = 72) Poehlam A et al. (1987) Koenigstorfer J et al. (2011) Hopkins N et al. (2012) Bouchard C et al. (1994) Hainer V et al. (2000) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.63 (0.36–0.80)*** (n = 41) Koenigstorfer J et al. (2011) Hainer V et al. (2000) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.58 (−0.04–0.87) (n = 17) Poehlam A et al. (1987) Bouchard C et al. (1994) Prud’Homme D et al. (1984)
BMI	0.69 (0.49–0.82)*** (n = 58) Koenigstorfer J et al. (2011) Hopkins N et al. (2012) Bouchard C et al. (1994) Hainer V et al. (2000) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.63 (0.36–0.80)*** (n = 41) Koenigstorfer J et al. (2011) Hainer V et al. (2000) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.63 (−0.13–0.93) (n = 11) Bouchard C et al. (1994) Prud’Homme D et al. (1984)
Fat free mass (kg)	0.57 (0.35–0.73)*** (n = 73) Poehlam A et al. (1987) Koenigstorfer J et al. (2011) Hopkins N et al. (2012) Bouchard C et al. (1994) Hainer V et al. (2000) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.52 (0.19–0.75)** (n = 38) Koenigstorfer J et al. (2011) Hainer V et al. (2000) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.80 (0.39–0.95)** (n = 17) Poehlam A et al. (1987) Bouchard C et al. (1994) Prud’Homme D et al. (1984)
Fat mass (kg)	0.58 (0.13–0.83)* (n = 48) Poehlam A et al. (1987) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000) Hamel P et al. (1986) Prud’Homme D et al. (1984)	0.85 (0.63–0.94)*** (n = 25) Koenigstorfer J et al. (2011) Hainer V et al. (2000) Prud’Homme D et al. (1984)	0.40 (−0.26–0.81) (n = 17) Poehlam A et al. (1987) Bouchard C et al. (1994) Prud’Homme D et al. (1984)
Fat mass to fat free mass ratio	0.69 (0.42–0.85)*** (n = 36) Bouchard C et al. (1994) Hainer V et al. (2000) Hamel P et al. (1986) Prud’Homme D et al. (1984)	0.85 (0.61–0.95)*** (n = 19) Hainer V et al. (2000) Prud’Homme D et al. (1984)	0.62 (−0.15–0.92) (n = 11) Bouchard C et al. (1994) Prud’Homme D et al. (1984)
Waist circumference (cm)	0.50 (0.09–0.77)* (n = 27) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000)	0.36 (−0.15–0.72) (n = 20) Koenigstorfer J et al. (2011) Hainer V et al. (2000)	0.83 (0.21–0.97)* (n = 7) Bouchard C et al. (1994)
Hip circumference (cm)	0.51 (0.11–0.77)* (n = 27) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000)	0.58 (0.13–0.83)* (n = 20) Koenigstorfer J et al. (2011) Hainer V et al. (2000)	0.25 (−0.62–0.84) (n = 7) Bouchard C et al. (1994)
Waist to hip ratio	0.29 (−0.16–0.64) (n = 27) Koenigstorfer J et al. (2011) Bouchard C et al. (1994) Hainer V et al. (2000)	0.27 (−0.24–0.66) (n = 20) Koenigstorfer J et al. (2011) Hainer V et al. (2000)	0.35 (−0.55–0.87) (n = 7) Bouchard C et al. (1994)
Sum of skin folds (cm)	0.67 (0.37–0.85)*** (n = 30) Bouchard C et al. (1994) Hainer V et al. (2000) Prud’Homme D et al. (1984)	0.78 (0.46–0.92)*** (n = 19) Hainer V et al. (2000) Prud’Homme D et al. (1984)	0.51 (−0.30–0.89) (n = 11) Bouchard C et al. (1994) Prud’Homme D et al. (1984)
Trunk fat	0.52 (0.12–0.78)* (n = 27) Hopkins N et al. (2012) Bouchard C et al. (1994) Hainer V et al. (2000)	0.67 (0.22–0.89)** (n = 14) Hainer V et al. (2000)	0.15 (−0.68–0.81) (n = 7) Bouchard C et al. (1994)
Extremity skin fold (cm)	0.54 (−0.39–0.92) (n = 21) Bouchard C et al. (1994) Hainer V et al. (2000)	0.78 (0.43–0.93)** (n = 14) Hainer V et al. (2000)	0.00 (−0.75–0.75) (n = 7) Bouchard C et al. (1994)
Trunk to extremity ratio	0.48 (−0.30–0.88) (n = 21) Bouchard C et al. (1994) Hainer V et al. (2000)	0.70 (0.27–0.90)** (n = 14) Hainer V et al. (2000)	0.00 (−0.75–0.75) (n = 7) Bouchard C et al. (1994)
Weight (kg)	0.67 (0.48–0.79)*** (n = 73) Poehlam A et al. (1987) Koenigstorfer J et al. (2011) Hopkins N et al. (2012) Bouchard C et al. (1994) Hainer V et al. (2000) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.70 (0.46–0.84)*** (n = 41) Koenigstorfer J et al. (2011) Hainer V et al. (2000) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.45 (−0.20–0.83) (n = 17) Poehlam A et al. (1987) Bouchard C et al. (1994) Prud’Homme D et al. (1984)
Absolute VO2 max (L.min−1)	0.38 (0.04–0.64)* (n = 42) Bouchard C et al. (1994) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.74 (−0.18–0.97) (n = 21) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.49 (−0.33–0.89) (n = 11) Bouchard C et al. (1994) Prud’Homme D et al. (1984)
Relative VO2 max (mL.min−1.kg−1)	0.39 (0.07–0.64)* (n = 48) Hopkins N et al. (2012) Bouchard C et al. (1994) Hamel P et al. (1986) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.51 (0.06–0.79)* (n = 21) Prud’Homme D et al. (1984) Afman G et al. (1988)	0.40 (−0.43–0.86) (n = 11) Bouchard C et al. (1994) Prud’Homme D et al. (1984)

n number of twin pairs, VO ₂ max maximal oxygen uptake, BMI body mass index

*p < 0.05; **p < 0.01; ***p < 0.001

Outcomes of Cardiorespiratory Fitness

There were nine studies (six twin studies [27, 29, 30, 32, 34, 35] and three family studies [19, 37, 38]) which investigated cardiorespiratory fitness measures and their response following a PA intervention. Pooling of five twin studies results (excluding Danis and colleagues [27] due to different methodology) suggests there are significant pooled r_MZ for absolute VO₂max (L.min⁻¹) (r_MZ = 0.38, 95% CI: 0.04–0.64, n = 42) and relative VO₂max (mL.min⁻¹.kg⁻¹) (r_MZ = 0.39, 95% CI: 0.07–0.64, n = 48) (Table 4).

There was one twin study which investigated the response of cardiorespiratory fitness following a combined PA and diet intervention [30] and four twin studies which investigated the response of cardiorespiratory fitness following an isolated PA intervention [29, 32, 34, 35]. The r_MZ for absolute and relative VO₂max in the study (n = 7) which combined PA with diet (r_MZ = 0.52, 95% CI: −0.38–0.92, and r_MZ = 0.48, 95% CI: −0.43–0.91, respectively) was higher than the pooled r_MZ from the studies which only investigated a PA intervention (r_MZ = 0.36, 95% CI: −0.01–0.64, n = 35, and r_MZ = 0.38, 95% CI: 0.04–0.64, n = 41, respectively) (Fig. 9) although the confidence intervals overlapped, and the r_MZ from the individual study was not statistically significantly different from 0 (with 95% CIs generated from the meta-analysis software). Meta-analyses for absolute and relative VO₂max were stratified by gender, with the pooled r_MZ being higher in females (Table 5). The pooled r_MZ for the response of absolute VO₂max following PA was 0.74 in females (n = 21) and 0.49 in males (n = 11), although neither were statistically significantly different from 0 (Fig. 10).

Heritability estimates for the response of VO₂max from two twin studies [29, 32] were not reported, as there were no statistically significant differences between the r_MZ and r_DZ (Table 1). Danis and colleagues [27] used different methodology to calculate heritability and we reported the estimates in Table 1 .Maximal heritability estimates from the three included family studies [19, 37, 38] were variable, ranging from 22–57% depending on race and when VO₂max was measured (e.g., ventilatory threshold, pre-determined power levels, etc.) (Table 6).

Heritability Estimates and the Within MZ Twin Pair Correlation

Only a few studies included DZ twins (n = 2) [29, 32], so we pooled the r_MZ to provide an estimate of the upper bound of heritability. Traditionally, twin and family studies investigating the heritability of a phenotype (e.g., PA engagement [15, 16], BMI [39], and chronic pain [40]) have done so using a cross-sectional design, with twin studies dividing the variance of a phenotype into components or proportions due to additive genetic factors (heritability), shared environmental factors, and unique environmental factors. Our pooled estimates represent the upper bound of heritability, including variance from additive genetic and shared environmental factors. However, our study investigated how shared familial factors influence the response to PA, with the r_MZ derived from the change in outcome status following an intervention. Since interventions were implemented over a specified timeframe, with training parameters controlled, it has been suggested that unique and shared environmental factors would make minor contributions to the variance of the response to PA [22], resulting in a r_MZ that would give a close estimate of heritability. However, family studies included in this review found significant correlations between spouses for the response of body composition [38] and cardiorespiratory fitness [19, 24] following a PA intervention. Although some suggest this indicates a greater influence of shared environmental factors [19], this correlation may equally be due to shared genes (assortative mating), so without making strong assumptions as to which is occurring in spouses, this is unlikely to indicate a greater influence of shared environmental factors.

Shared Familial Influence on Changes of Body Composition

Factors shared within MZ twin pairs appear to play a strong role in the response of BMI (pooled r_MZ = 0.69) following a PA intervention (Fig. 2), although they appear to be less influential in the response of other outcomes (e.g., waist-to-hip ratio) (Fig. 5). Although we were unable to pool heritability estimates, our pooled r_MZ for the response of BMI following PA appears to be within the range of previous studies reporting the cross-sectional heritability of BMI (ranging from 47–90% in twins studies [39]). However, other cross-sectional studies have reported heritability estimates for waist circumference (66%) and body fat percentage (68%) [41] that appear to be slightly higher than our r_MZ in response to exercise (pooled r_MZ = 0.50 and 0.55, respectively), especially considering our results represent the upper bound of heritability. Therefore, by comparing our results to those of previous investigations, it appears the genetic influences on an individual’s body composition (cross-sectional association) might be different, and perhaps higher, than the way their body composition responds to PA.

Previous cross-sectional twin studies have reported gender differences for the heritability of body composition, although they appear to vary depending on the outcome of interest. A twin study by Schousboe and colleagues [42] reported that males have higher heritability estimates compared to females for body fat percentage (63 and 59%, respectively), sum of skin folds (65 and 61%, respectively), waist circumference (61 and 48%, respectively) and waist-to-hip ratio (22 and 10%, respectively). However, other studies have reported higher heritability estimates in females across a variety of body composition measures [43, 44]. The variability between genders for the heritability of body composition has been supported in various twin studies, regardless of the sample size, methods of analyses or ethnicity [43, 45–47]. Our results extend the understanding that gender influences the role shared familial factors, including genes, play in the variation of body composition (cross-sectional association), and suggests gender influences how shared familial factors influence the response of body composition measures following PA. In particular, shared familial factors appear to have a greater influence on changes in fat mass for females engaged in PA (Fig. 7) and fat free mass for males engaged in PA (Fig. 8). Therefore, to better understand how both genetics and shared environmental factors impact an individual’s response to PA, it may be important to take into consideration the gender of the individual, and the outcome of interest.

Shared Familial Influence on Changes in Cardiorespiratory Fitness

The heritability of VO₂max assessed in cross-sectional studies ranges from 40–71% in twin studies [41, 48] and has been reported at 50% (maximal heritability) in the HERITAGE Family Study [49]. Our pooled r_MZ were 0.38 and 0.39 for absolute and relative VO₂max, respectively, and appear to be smaller than heritability estimates for an individual’s pre-training VO₂max, although the CIs for our results include the cross-sectional estimates. This suggests genetics may be more influential in determining an individual’s cardiorespiratory fitness, compared to their fitness response following PA, although the biological explanation for this is unclear.

The point estimates of the r_MZ for the response of cardiorespiratory fitness following PA appear slightly greater in females (Fig. 10), although the CIs for both the male and female correlations cover almost all the possible range of values due to small sample sizes in the original studies. Similarly, existing studies investigating the heritability of cardiorespiratory fitness have been limited in their ability to analyze the effect of gender due to small sample sizes [41], and single gender cohorts [48]. Therefore, our results should be viewed as preliminary with this area deserving attention in future studies.

Strengths and Limitations

Our study demonstrated considerable strengths in its design. First, previous studies have predominantly focussed on investigating the heritability of PA engagement (cross-sectional association) [15], without considering how genetics and shared environmental factors impact an individual’s response to PA. From a health-care perspective, it may be more important to investigate how genetics and environmental factors influence the response to PA. It is likely the response to PA would be more dependent on unique environmental factors, such as training parameters (frequency, intensity, duration, type), adherence, therapeutic alliance, and many more. However, neither training frequency nor duration appeared to influence the r_MZ for either body composition or cardiorespiratory fitness, which may suggest the role genetics plays in response to PA is independent of these parameters. Quantifying the influence of genetics and environmental factors on the response to PA may serve to explain why certain individuals do not respond as well to a structured PA program across a variety of outcomes, with implications for how we can modify the training environment to achieve a positive response. Second, twin studies which have investigated how genetics influence the response to PA have been limited in their ability to draw firm conclusions due to small sample sizes. Small sample sizes of the included studies explain cases where our pooled CIs were wide, even though we were able to pool results for up to 83 MZ twin pairs, improving the precision around these estimates. To obtain 95% CIs of sufficiently small width to be informative (e.g., a total width of 0.1), in studies that include only MZ twins, approximately 400 twin pairs are required if the correlation is moderately high (0.7), and greater than 1000 twin pairs if the correlation is 0.4. For studies including both MZ and DZ twins, 150 twin pairs of each zygosity would be required to detect a significant difference (p = 0.05) between r_MZ = 0.7 and r_DZ = 0.5, with 80% power. If both correlations are lower (e.g., r_MZ = 0.5 and r_DZ = 0.3), 275 twins pairs of each zygosity would be required. Many of the studies which reported heritability or maximal heritability also failed to report confidence intervals for their estimates, or provide sufficient information to enable these to be estimated accurately (Tables 1 and 6). Although point estimates are available, there is clearly a substantial information difference between a heritability of 47% with a 95% CI of 44–50% and the same heritability with a 95% CI of 10–85%, and we expect that studies included in this review are more like to the second situation, limiting the utility of the reported estimates. Third, raw data were used to re-analyse previously reported correlations in four twin studies [29, 30, 34, 35] and adjust for age, gender (if applicable), and baseline values. This provided a more precise estimate for quantifying the role genetics plays in the response to PA.

Our study has a few limitations which need to be considered when interpreting the results. First, samples from included twin studies differed in their age, baseline values, PA interventions, and diet interventions. Furthermore, one study recruited twin pairs admitted to an obesity unit for a 40-day physical activity and diet program [28], a sample not representative of the general population. However, we conducted a number of sensitivity analyses and the exclusion of any single study did not significantly affect the results for any of the outcomes (Additional file 2). In addition, we performed separate meta-analyses for studies which included a diet intervention, to better understand how the difference between interventions impacted our results. Second, two twin studies [29, 30] reported drop outs on the basis of twin pairs failing to complete the training protocol (Table 2), while the family studies only analyzed data from participants who completed the training protocol (Table 3). We acknowledge that this may limit the generalizability of the results, as participants who completed the training protocol are likely to be more motivated to engage in PA than the general population. Third, although using a classical twin design to estimate heritability is a widely reported method to investigate how genetics contributes to the variation of a phenotype, it does have some limitations, and together with the fact that individual twin studies had small sample sizes, is the reason we did not focus our results on these estimates. The use of self-reported zygosity measures, based on the difficulty of being told apart by parents, is often criticized. MZ twins who differ in their height and weight can be mistakenly classified as DZ twins when using self-reported measures, resulting in an underestimation of heritability [50]. However, only one study included in this review assessed zygosity using only a self-reported questionnaire [32], with another failing to describe how zygosity was assessed [20]. The remaining twin studies (n = 7) verified questionnaire-based zygosity through DNA mapping. In addition, not considering the genotype-environment interaction is a limitation of the classical twin design, since genetic factors can influence an individual’s choice/exposure to the environment. However, studies included in this review utilized a controlled training environment, reducing the likelihood that an individual’s genetics would impact their environment for the experimental period. Furthermore, the use of heritability as a measure, although widely reported, has some limitations; it is dependent on the modeling of the mean, on the amount of variance and measurement error (which may be larger in studies of changes in outcomes compared with cross-sectional studies of outcomes [51]) and on the total variation within a population, which may differ between populations and between the same population measured at different times [52]. Finally, when estimated from classic twin studies, this estimate depends on the assumption that environments are shared to the same extent by MZ and DZ pairs—an assumption that is rarely considered or tested in practice [53].

Clinical Implications

The results of this current investigation are consistent with a substantial influence of genes on the response of body composition and cardiorespiratory fitness following PA. These results have implications for conditions which utilize PA as a management strategy, for example, diabetes, and low back pain. If an individual’s response to a PA intervention is partially dictated by genetic factors this could potentially explain why some individuals fail to respond to increased PA. This has implications for changing the modifiable training environment to achieve a desired effect (e.g., increased intensity, frequency, or duration), or excluding people who demonstrate a poor response to reduce treatment costs and consumer disappointment. Furthermore, if genetic factors are involved in the poor response to PA as an intervention, this has implications for the selection of alternative management strategies, or a modification to the outcome investigated, since individuals who show a low training response to one parameter (e.g., VO₂max) might in fact respond positively to another (e.g., BMI).

Research linking genetic markers to a specific phenotype (quantitative trait locus analysis) have aided the mechanistic understanding of how genetics influence the response of body composition and cardiorespiratory fitness following PA, although more genetic research needs to be done. A family study investigated over 300,000 single-nucleotide polymorphisms (SNPs) and identified 21 SNPs which accounted for 49% of the variance in the response of VO₂max following a PA intervention, with one SNP (rs6552828) accounting for ~6% of the variance [54]. The variance explained by these 21 SNPs is similar to the maximal heritability of VO₂max response from the family study included in this review (47%), although this study observed significant spouse correlations which some consider consistent with shared environmental effects, thereby reducing the variance explained by genetics [19]. Similarly, nine SNPs were found to explain 20% of the variance of submaximal heart rate in response to PA, with one SNP (rs2253206) accounting for ~5% of the variance [55]. Earlier studies have identified candidate genes that are strongly linked to or associated with the response of BMI, fat mass, fat-free mass, and body fat percentage following a PA intervention [56]. For example, the insulin-like growth factor-1 (IGF-1) gene marker was strongly linked to response of fat-free mass following PA [57], with linkage also present for a polymorphism in the S100A gene [56] (predominantly found in slow-twitch skeletal and cardiac muscle fibers [58]). Research identifying genetic markers is promising and may aid the prediction of how an individual’s body composition and cardiorespiratory fitness will respond following PA, although it is essential these results are replicated in larger samples, and through a variety of genetic analyses before definite conclusions are reached [59, 60]. Furthermore, research investigating practical and cost-effective methods to identify those who will respond positively to a PA intervention would be of significant interest from a public health and clinical perspective. For example, information regarding how family members have previously responded to PA may help to predict how an individual will respond to a similar intervention, potentially reducing the need for costly genetic testing.