Our results demonstrate consistent evidence that shared familial factors (whether genetic or environmental) play a role in the response of body composition and cardiorespiratory fitness following PA, despite varying on the outcome being assessed, particularly when results were stratified by gender. The pooled rMZ were generally >0.5, and the bulk of most CIs also exceeded 0.5. Shared familial factors appear to play a larger role in the response of body composition when compared to cardiorespiratory fitness, and may have more influence on the response for most outcomes when considering a combined PA and diet intervention.
Heritability Estimates and the Within MZ Twin Pair Correlation
Only a few studies included DZ twins (n = 2) [29, 32], so we pooled the rMZ 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 , and chronic pain ) 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 rMZ 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 , resulting in a rMZ 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  and cardiorespiratory fitness [19, 24] following a PA intervention. Although some suggest this indicates a greater influence of shared environmental factors , 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 rMZ = 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 rMZ 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 ). However, other cross-sectional studies have reported heritability estimates for waist circumference (66%) and body fat percentage (68%)  that appear to be slightly higher than our rMZ in response to exercise (pooled rMZ = 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  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 . Our pooled rMZ 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 rMZ 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 , and single gender cohorts . 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) , 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 rMZ 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 rMZ = 0.7 and rDZ = 0.5, with 80% power. If both correlations are lower (e.g., rMZ = 0.5 and rDZ = 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 , 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 . However, only one study included in this review assessed zygosity using only a self-reported questionnaire , with another failing to describe how zygosity was assessed . 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 ) and on the total variation within a population, which may differ between populations and between the same population measured at different times . 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 .
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 . 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 . 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 . 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 . For example, the insulin-like growth factor-1 (IGF-1) gene marker was strongly linked to response of fat-free mass following PA , with linkage also present for a polymorphism in the S100A gene  (predominantly found in slow-twitch skeletal and cardiac muscle fibers ). 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.