Cardiorespiratory Fitness and Adiposity as Determinants of Metabolic Health—Pooled Analysis of Two Twin Cohorts

Abstract

Context:

The joint effects of cardiorespiratory fitness (CRF) and body composition on metabolic health are not well known.

Objective:

To examine the associations of CRF, fat-free mass index (FFMI), and fat mass index (FMI) with metabolic health in individual twins and controlling for genetic and shared environmental effects by studying monozygotic intrapair differences.

Design, Setting, and Participants:

Two cross-sectional samples of healthy adult monozygotic and dizygotic twins were drawn from population-based Danish and Finnish national twin registries (n = 996 and n = 309).

Main Measures:

CRF was defined as VO₂max divided by fat-free mass. Insulin sensitivity and acute insulin response indices were derived from an oral glucose tolerance test. A continuous metabolic syndrome score was calculated. Visceral and liver fat were measured in the Finnish sample. Associations were analyzed separately in both cohorts with multivariate linear regression and aggregated with meta-analytic methods.

Results:

Insulin sensitivity, acute insulin response, metabolic syndrome score, visceral, and liver fat amount had strong and statistically significant associations with FMI (|β| 0.53 to 0.79), whereas their associations with CRF and FFMI were at most weak (|β| 0.02 to 0.15). The results of the monozygotic intrapair differences analysis showed the same pattern.

Conclusions:

Although FMI is strongly associated with worsening of metabolic health traits, even after controlling for genetic and shared environmental factors, there was little evidence for the effects of CRF or FFMI on metabolic health. This suggests that changing FMI rather than CRF or FFMI may affect metabolic health irrespective of genetic or early environmental determinants.

The role of adiposity in metabolic health has long been recognized, although recent discussion has concentrated on the possibility that cardiorespiratory fitness (CRF) might diminish the deleterious effects of obesity (1). The definition of metabolic health varies, but we use it to refer to the components of the metabolic syndrome (2) and other associated factors (e.g., insulin sensitivity, ectopic fat deposition). Some earlier studies show that CRF is associated positively with insulin sensitivity (3–5) and negatively with metabolic syndrome risk or individual cardiometabolic risk factors [e.g., waist circumference, high-density lipoprotein (HDL), triglycerides, fasting glucose, and blood pressure] (6–8). In meta-analyses, CRF (9) and adiposity (10) both independently contribute to cause mortality. Additionally, CRF seems to moderate the effects of adiposity on mortality (1, 11) or cardiovascular disease outcomes (12), so that when people have high CRF, the effects of adiposity are minimal or nonexistent.

A major problem with studies comparing the effects of CRF with the effects of adiposity on metabolic health is that many of them define and measure CRF in ways that are confounded with adiposity. First, measurement of CRF with weight-bearing exercise such as walking on a treadmill with progressively increasing speed and/or inclination leads to a poorer outcome (VO₂max estimate) in subjects with higher adiposity, even with equal performance of the cardiorespiratory system (13). Second, the usual method of scaling VO₂max to body size by dividing it with body weight systemically underestimates CRF in obese subjects (14–17) because fat mass is included in total weight, although it does not itself contribute much to VO₂max. Therefore, two possibilities arise to minimize the confounding of CRF with adiposity: the use of a maximal cycle ergometry test, which is much less dependent on weight than treadmill testing, and scaling VO₂max to body size by dividing it with fat-free mass (FFM) rather than with total body weight (14, 15).

Some studies on the relationship between a CRF measure not confounded by adiposity and insulin sensitivity have been carried out. Two large cross-sectional studies in children or adolescents have observed no correlation between VO₂max/FFM and insulin sensitivity (16, 18), whereas in one study, with severely obese 8 to 16 year olds, a modest association (r = 0.36) was observed (4). Two studies in adults have demonstrated significant positive associations between VO₂max/FFM and insulin sensitivity: r = 0.78 (3) and r = 0.29 (5). Regarding the metabolic syndrome, some cross-sectional studies have found a negative association between VO₂max/weight and the odds or risk of metabolic syndrome (7, 19) or a continuous metabolic syndrome score (6), whereas some longitudinal studies find no (or very weak) associations (|r| from 0.00 to 0.10) (20–22). Also, ectopic (visceral or liver) fat deposition has been suggested to mediate the effects of CRF on metabolic health by some researchers (23, 24), but in these studies VO₂max/weight was used as the CRF measure.

The purpose of the current study was to compare the effects of body composition and CRF on metabolic health in predominantly healthy adult twins. Our initial hypothesis was that adiposity would not have a considerable effect on metabolic health when a measure of CRF is included in the analysis. We also hypothesized that FFM would be associated with metabolic health independently of adiposity. To study these hypotheses, we compared the effects of fat mass, FFM, and CRF (defined as VO₂max/FFM) on metabolic health variables [insulin sensitivity, acute insulin response, and metabolic syndrome components, such as fasting glucose, blood pressure, low-density lipoprotein (LDL), high-density lipoprotein, triglycerides, and a continuous metabolic syndrome score]. We also assessed the associations of fat mass, FFM, and CRF on the accumulation of ectopic (visceral or liver) fat. To estimate the individually acquired environmental effects of adiposity or CRF on metabolic health, we examined the associations between intrapair differences in traits in monozygotic (MZ) twins. The idea is to provide an estimate of the potential modifiability of metabolic health at given influences from genetic and shared familial environmental factors. Therefore, we estimated the extent to which the associations of CRF or body composition with metabolic health variables found in individual-based analyses were also present when examining intrapair trait differences of the MZ twin pairs, which controls for genetic and shared environmental confounding.

Methods

Participants

The first dataset came from the GEMINAKAR twin study recruited through the national, population-based Danish Twin Registry (25) and established during 1997 to 2000 (26). The current sample from GEMINAKAR consisted of 459 complete twin pairs aged 18 to 67 (median 38) at the time of examination. There were 182 MZ, 179 same-sex dizygotic (DZ), and 98 opposite-sex DZ pairs (Table 1). These twins represent 60% of the original GEMINAKAR study sample that underwent more extensive metabolic phenotyping (see Supplemental Methods for more details). The second dataset, TwinFat, is based on two Finnish population-based longitudinal studies of five consecutive birth cohorts of twins: FinnTwin16 and FinnTwin12 (27). The current TwinFat sample consisted of 153 complete twin pairs aged 23 to 32 (median 28) at the time of examination. There were 74 MZ and 79 same-sex DZ pairs (Table 1). Participants for TwinFat were selected according to their reported body mass indexes (BMIs) at the age of 23 to 27 years to represent a wide range of intrapair differences in BMI. Thus, a part of the TwinFat sample was enriched with 20 MZ and 53 DZ pairs discordant for BMI (∆BMI >3 kg/m²) and 18 MZ and 13 DZ pairs concordant for BMI (∆BMI <1 kg/m²); otherwise, it was a random sample of the twin pairs. For both GEMINAKAR and TwinFat, twins were excluded at recruitment if they were pregnant or breastfeeding, reported substance abuse, or had been diagnosed with diabetes or heart disease. Both studies were conducted according to the principles of the Helsinki Declaration. Written informed consent was obtained from all participants.

Independent variables: body composition and CRF

Fat mass and fat-free mass (FFM) were estimated by bioelectrical impedance analysis (103 RJL-System analyzer; RJL-Systems, Detroit, MI) in GEMINAKAR (28) and by dual-energy X-ray absorptiometry (Lunar Prodigy, Madison WI; software version 2.15) in TwinFat. Fat mass and FFM were scaled to height by calculating the fat mass index (FMI = fat mass/height²) and the FFM index (FFMI = FFM/height²).

Participants both in GEMINAKAR and TwinFat completed a maximal exercise test using an electrically braked bicycle ergometer. In GEMINAKAR, VO₂max was estimated from the average workload for the last 2 minutes of cycling before exhaustion (29). In TwinFat, oxygen uptake was measured breath by breath with a Vmax spiroergometer (Sensorimedics, Yourba Linda, CA). The exercise was continued until exhaustion. The criteria for maximality of the exercise were rate of perceived exertion 19 to 20/20 on the Borg scale or the gas exchange ratio VCO₂/VO₂ of >1.10. VO₂max was defined as the mean VO₂ measured during the last 30 seconds at the maximal workload (30). CRF was defined as VO₂max divided by FFM, as this does not correlate with adiposity and subsequently underestimate CRF in obesity, compared with dividing VO₂max by body weight (14–17) (Supplemental Figs. 1 and 2).

Dependent variables: blood tests and oral glucose tolerance test

Total cholesterol, HDL cholesterol, and triglycerides were measured from fasting venous blood samples. LDL was estimated with the Friedewald Formula (LDL = total cholesterol − HDL − triglycerides/5). Subjects underwent a standardized 75-g oral glucose tolerance test. For more detailed information regarding laboratory measurements in GEMINAKAR see (26), and in TwinFat see (31). Glucose and insulin values from the oral glucose tolerance test were used to calculate the BIGTT-SI (BIGTT insulin sensitivity index) and BIGTT-AIR (BIGTT acute insulin response index, an indicator of β cell function) with formulas derived by Hansen et al. (32). Homeostatic model assessment insulin resistance index (HOMA-IR) was calculated with the following formula: fasting glucose × fasting insulin × 22.5⁻¹.

Dependent variables: visceral fat and liver fat

Visceral fat volume and liver fat percentage were measured from 83 individual twins in TwinFat by magnetic resonance imaging and spectroscopy, respectively, as described previously (33).

Dependent variables: metabolic syndrome risk score

Waist circumference was measured midway between the anterior superior iliac spine and the lower rib margin, and scaled to height by expressing it as the waist-to-height ratio. Blood pressure was measured with a sphygmomanometer supine after at least 5 minutes’ rest with a mean of three consecutive measures. A continuous score depicting the risk for metabolic syndrome (metabolic syndrome score), based on the National Cholesterol Education Program Adult Treatment Panel III definition of metabolic syndrome (2), was calculated from sex-specific Z-scores of variables in each country with the following formula: metabolic syndrome score = waist circumference + (systolic blood pressure + diastolic blood pressure)/2 − HDL_log + fasting glucose_log + triglycerides_log. For a validation of a similar score, see ref. 34.

Statistical analyses

All statistical analyses were performed in R for OS X (version 3.3.0). The following variables were natural log transformed prior to analysis due to positive skewness: HOMA-IR, BIGTT-AIR, HDL, triglycerides, fasting glucose, and visceral fat volume. Due to extreme skewness, liver fat percentage was transformed as −1/sqrt(liver fat %). Differences between samples were assessed with t tests. The two samples were analyzed separately at first (see Supplemental Tables 1 and 2), and for the Results section the effect sizes (β), confidence intervals, and P values were combined with a random effects generic inverse variance meta-analysis (metagen package for R, version 1.0).

We performed classical twin linear regression models, with ACE decomposition, for quantitative traits (35) (twinlm function of mets package for R, version 1.1.1) on all available individual twins. These models are analogous to linear regression models, but the clustered sampling by twin pairs and their differing degrees of genetic relatedness must be taken into account. These models shown in Table 2 were controlled for age and sex.

For the analyses in Table 3, we repeated the same regression models as in Table 2, but with intrapair differences of MZ twin pairs and bootstrapping (see Supplemental Methods). Because all of the differences between MZ twins reared together are due to nonshared environmental differences, this approach controls for genetic and environmental factors shared between the cotwins, and thus reflects the nonshared/unique environmental associations between traits.

Results

Characteristics of the study subjects are summarized in Table 1. Compared with the TwinFat sample, the GEMINAKAR twins were older and had lower BMIs, with better insulin sensitivity (BIGTT-SI) but worse LDL cholesterol, triglyceride values, and VO₂max/FFM (all P < 0.001). These differences (apart from BMI) probably result from subjects in GEMINAKAR being older. A quantity amounting to 4.9% of subjects from GEMINAKAR and 8.7% of subjects from TwinFat met the National Cholesterol Education Program Adult Treatment Panel III criteria for metabolic syndrome (2).

Associations of fat mass and VO₂max/FFM with outcome variables in individuals

The meta-analysis of the results from individual twins (Table 2) shows that FMI was significantly associated with all of the metabolic outcome variables, having the strongest associations with HOMA-IR (β = 0.67, P < 0.001), BIGTT-SI (β = −0.79, P < 0.001), BIGTT-AIR (β = 0.53, P < 0.001), metabolic syndrome score (β = 0.69, P < 0.001), waist-to-height ratio (β = 0.85, P < 0.001), visceral fat volume (β = 0.75, P < 0.001), and liver fat percentage (β = 0.53, P < 0.001). Conversely, FFMI was significantly associated only with waist-to-height ratio (β = 0.29, P < 0.001). The associations between VO₂max/FFM and outcome variables were weak or nonsignificant (|β| from 0.00 to 0.16, P < 0.001 to 0.987), with the strongest association (β = 0.15, P < 0.001) with HOMA-IR. Separate estimates from the two different samples are shown in Supplemental Table 1.

Associations of MZ intrapair differences (Δ)

In the analyses controlling for genetic and shared environmental factors within 256 MZ twin pairs (Table 3), ΔFMI was significantly associated with all metabolic outcome Δ-variables in twin pairs, except with ∆HDL and ∆triglycerides. The associations of ΔFMI were strongest with ∆HOMA-IR (β = 0.59, P = 0.002), ΔBIGTT-SI (β = −0.68, P < 0.001), ΔBIGTT-AIR (β = 0.48, P = 0.001), Δmetabolic syndrome score (β = 0.55, P < 0.001), Δwaist-to-height ratio (β = 0.79, P < 0.001), ∆LDL (β = 0.47), ∆visceral fat volume (β = 0.98, P < 0.001), and ∆liver fat percentage (β = 0.77, P < 0.001). In contrast, ΔFFMI was significantly associated only with ∆waist-to-height ratio (β = 0.19, P < 0.001) and ΔVO₂max/FFM only with ∆HDL (β = 0.13, P = 0.016). Separate estimates from the two different samples are shown in Supplemental Table 2. There was statistically significant heterogeneity between the two samples for some of the estimates in Tables 2 and 3, indicating that the differences in the estimates are not only due to random variation in sampling (Supplemental Tables 1 and 2).

Discussion

Our findings from two large Nordic twin samples show that adiposity (FMI) is significantly and unfavorably associated with metabolic health variables (negatively with insulin sensitivity and HDL, and positively with fasting glucose, insulin resistance and acute insulin response indices, blood pressure, LDL, triglycerides, metabolic syndrome risk, visceral fat volume, and liver fat percentage). Except for HDL, these associations persist even after controlling for genetic and shared environmental confounding with the MZ twin intrapair differences analysis. In contrast, FFMI was associated only with the waist-to-height-ratio, and CRF (VO₂max/FFM) was only weakly or nonsignificantly (|β| ≤ 0.16) associated with the metabolic health variables, in both individual and MZ twin intrapair differences analyses.

There has been extensive discussion on whether adiposity or CRF matters more for metabolic health. Our study clearly shows that whereas adiposity is strongly negatively associated with metabolic health, both FFMI and VO₂max/FFM have much smaller (or nonsignificant) effects. One motivation for the present analyses was to use a CRF variable (VO₂max/FFM) that is not confounded by adiposity instead of the widely used VO₂max/weight, which is confounded by adiposity (14–17) (also see Supplemental Figs. 1 and 2). We know of only a few previous studies that have used the same approach. Some of them show an association between insulin sensitivity and VO₂max/FFM (3–5), whereas others do not (16,18). McMurray et al. (16) and Henderson et al. (18) studied predominantly normal weight children or adolescents, and demonstrated no significant association between VO₂max/FFM and insulin sensitivity while controlling for adiposity. However, Morinder et al. (4), studying predominantly severely obese children, demonstrated that VO₂max/FFM was a bit stronger univariate predictor of insulin sensitivity (r = 0.36) than BMI SDS (r = −0.22), or fat mass (r = −0.31). The few studies on adults with CRF measured as VO₂max/FFM (3, 5) show positive associations between insulin resistance and adiposity, and negative associations between VO₂max/FFM and adiposity. Huth et al. (3) studying 53 men report very high correlations for VO₂max/FFM with insulin sensitivity (r = 0.78) and with adiposity (r = −0.62 to −0.74). They, however, calculated correlations across heterogenous groups, which may inflate the estimates. Sævarsson et al. (5) report an association between HOMA-IR and VO₂max/FFM (r = −0.29), although the association between HOMA-IR and VO₂max/weight was not significant after controlling for fat percentage. In a recent study on adult subjects with normal or impaired glucose tolerance or manifest type 2 diabetes, Solomon et al. (36) observe that in a regression model with three overlapping adiposity variables (BMI, weight, and fat percentage), age, and sex, VO₂max/weight (β = 0.34) significantly predicts insulin sensitivity. However, the β’s from these regression models are hard to interpret because of the highly correlated predictors and the unique variance explained by different predictors not being reported.

Adding to this research, our results indicate that the independent relationships between CRF (VO₂max/FFM) and BIGTT-SI (β = 0.10), HOMA-IR (β = −0.15), or fasting glucose (β = 0.03) are weak, especially when compared with the effects of adiposity (FMI; |β| from 0.59 to 0.79). It is not entirely clear why some of these studies find a moderate to large effect (3–5) and some (16, 18), including ours, do not (or find weak effects). However, the studies finding no (or weak) effects tend to have larger and more representative study samples and more consistently control for confounding factors such as age, sex, and adiposity. Results from longitudinal studies with dietary and/or exercise interventions looking at the effect of changes in CRF to metabolic health are consistent with our results showing a weak effect. These studies show very small associations (corresponding to a |r| of from 0.00 to 0.10) between improvements in CRF and improvements in glucose homeostasis, lipids, or blood pressure (20–22).

Cross-sectional studies finding a negative association between CRF and the odds of metabolic syndrome (7, 19) or a continuous metabolic syndrome score and metabolic syndrome components (6) in general have not measured CRF independently of obesity but with a weight-bearing treadmill exercise (6), by estimated VO₂max divided by weight (7), or by categorizing VO₂max/weight into groups by age-specific cutoff points (19). In contrast to these studies, our results show no significant independent association between VO₂max/FFM and the metabolic syndrome score. Similarly to our study, a large cross-sectional study on 8 to 18 year olds measuring CRF with VO₂max/FFM found no association between a metabolic syndrome risk score and CRF, after controlling for adiposity (16). Results of longitudinal studies support this view (20–22). One longitudinal study demonstrates an association between change in VO₂max/weight and a metabolic syndrome risk score, but they only controlled for visceral fat and not for overall adiposity (37).

Ectopic fat deposition into the liver or to visceral compartments has been considered a mediator for the deleterious effects of adiposity on cardiometabolic outcomes (38). Additionally, CRF has been thought to possibly improve metabolic health through its effects on ectopic fat deposition. Haufe et al. (24) hypothesized that liver fat might mediate the effects of CRF on insulin sensitivity, because in their study the correlation between VO₂max/weight and insulin sensitivity is smaller and nonsignificant after controlling for liver fat amount. But as liver fat percentage and VO₂max/weight are both correlated with adiposity, this finding can be explained even with no independent effect of CRF on liver fat percentage. Arsenault et al. (23) studied the accumulation of visceral fat in tertiles of VO₂max/weight and conclude that men in the low CRF group have more visceral fat (and the associated metabolic dysfunction), even when matched for BMI. However, in BMI-matched groups, men with low CRF were older and had less muscle mass than men with high CRF, consistent with adiposity confounding the grouping by VO₂max/weight. In our study, only FMI, not CRF or FFMI, was associated with the accumulation of liver and visceral fat.

Because there seems to be a disagreement between the general assumption that physical fitness protects from ill health and our results showing weak associations between CRF and metabolic health, the exact physiological basis of VO₂max warrants some discussion. VO₂max is equal to maximal cardiac output multiplied by the maximal arteriovenous oxygen difference (Ca-VO₂max) during exercise (39). To this end, differences in VO₂max between healthy individuals must result mainly from differences in cardiac output or the ability of muscles to extract and consume oxygen from blood (Ca-VO₂max) during maximal exercise. Reviews or meta-analyses of cross-sectional (40) and longitudinal intervention studies (39) show that differences in VO₂max in athletes and healthy subjects result from differences in maximal cardiac output rather than peripheral factors (Ca-VO₂max). Thus, it is not surprising that VO₂max is not necessarily correlated with metabolic health independently of adiposity, because differences in maximal cardiac output during exercise result mainly from central cardiovascular factors that are not directly connected to peripheral tissues relevant to metabolism (skeletal muscle, adipose tissue, the liver). This might explain our results and the results of previous studies finding weak or no associations between adequately scaled CRF and metabolic health. However, these results do not undermine the message from studies showing beneficial effects of physical activity on metabolic health.

Our study has multiple strengths. First, the measure of CRF (VO₂max/FFM) is probably not confounded by adiposity compared with estimates of CRF from weight-bearing exercise (e.g., treadmill test) or VO₂max/weight (see Supplemental Table 3 for VO₂max/weight analyses). Second, the analyses in MZ twins allow the estimation of the associations between adiposity, CRF, and metabolic health, independently of genetic or shared environmental confounding. Because any differences between MZ cotwins follow from nonshared environmental factors, in this analysis the association between intrapair differences of traits models the unique environmental association between the variables. Although not direct evidence of causal associations, we believe that an advantage in studying correlations between MZ intrapair differences is that they outline opportunities for environmental modifications of variables, given that the fixed genetic and shared environmental factors are controlled for in the associations. The reported significant associations between ∆FMI and metabolic health variables support the notion that interventions or long-term environmental changes reducing adiposity would also improve metabolic health. This does not, however, remove the need for longitudinal studies with long follow-up.

However, our study also has some limitations. The measures for VO₂max (maximal cycle ergometry) and body composition (bioelectrical impedance analysis) in GEMINAKAR probably contain more measurement error than the gold standard measures (spiroergometry and dual-energy X-ray absorptiometry) in TwinFat. For some of the analyses, there was heterogeneity between the estimates obtained from GEMINAKAR and TwinFat (see Supplemental Tables 1 and 2), which could drive the aggregate estimates toward the null hypothesis. This heterogeneity might reflect differences in VO₂max and body composition measurement, sampling procedures (age or BMI), or other factors. Because there is no conclusive reason to assume that the estimates of one study are more correct, the aggregated estimates from the two samples were reported and interpreted, despite possible heterogeneity in some estimates. The associations between intrapair differences in BMI-related traits in TwinFat might be stronger due to BMI discordance sampling. Also, our subjects were predominantly healthy and young adults, so our results might not generalize to patient populations with established metabolic disease (e.g., metabolic syndrome, diabetes) or the elderly.

In summary, our results from two large Nordic twin studies show that, in a multivariate model containing FMI, FFMI, and CRF, adiposity is strongly associated with the studied metabolic health variables (increased insulin resistance, components of metabolic syndrome, visceral fat volume, and liver fat percentage). CRF and FFMI were weakly or nonsignificantly associated with these metabolic health variables. This does not indicate, however, that physical activity itself is not important for metabolic health, because VO₂max is not a measure of exercise, but is closely related to maximal cardiac output. Finally, the associations between adiposity and metabolism, independently of CRF, persisted even after controlling for genetic and environmental factors shared between MZ cotwins, underscoring the importance of lifestyle and other environmental factors on metabolic health.

Abbreviations:

 
• BIGTT-AIR

  BIGTT acute insulin response index

 
• BIGTT-SI

  BIGTT insulin sensitivity index

 
• BMI

  body mass index

 
• CRF

  cardiorespiratory fitness

 
• DZ

  dizygotic

 
• FFM

  fat-free mass

 
• FFMI

  FFM index

 
• FMI

  fat mass index

 
• HDL

  high-density lipoprotein

 
• HOMA-IR

  homeostatic model assessment insulin resistance index

 
• LDL

  low-density lipoprotein

 
• MZ

  monozygotic.

Acknowledgments

Berit L. Heimann is acknowledged for help in collection of data for GEMINAKAR. Linda Mustelin and the Obesity Research Unit team at the University of Helsinki are acknowledged for help in the collection of the TwinFat data. Richard J. Rose, Indiana University, is thanked for his contributions to FinnTwin16 and FinnTwin12.

The GEMINAKAR study was supported by grants from the Danish Medical Research Fund, Danish Diabetes Association, NOVO Foundation, Danish Heart Foundation, and Apotekerfonden. Personal grants were received from the Academy of Finland (Grant 266286); Academy of Finland Centre of Excellence in Research on Mitochondria, Metabolism, and Disease (Grant 272376); University of Helsinki, Helsinki University Hospital (government funds); Novo Nordisk Foundation; Biomedicum Helsinki Foundation; Finnish Diabetes Research Foundation; Jalmari and Rauha Ahokas Foundation; Finnish Foundation for Cardiovascular Research (to K.H.P.); and Academy of Finland Centre of Excellence in Complex Disease Genetics (to J.K.). Data collection in FinnTwin16 and FinnTwin12 was supported by the National Institute of Alcohol Abuse and Alcoholism (Grants AA-12502 and AA-09203 to R. J. Rose), Academy of Finland (Grants 44069, 205585, and 118555 to J.K.), European Union–funded projects TORNADO (FP7-KBBE-22270) and ENGAGE (FP7- HEALTH-F4-2007- 201413).

Author contributions: K.O.K., C.D., T.I.A.S., and K.H.P. conceived the current study and the scientific questions behind it. S.J., R.H., C.D., K.O.K., T.I.A.S, and K.H.P. developed the study design and methodology. S.J. and R.H. implemented the statistical analyses. K.O.K. and T.I.A.S. were involved in the design, implementation, and data collection of the GEMINAKAR study. A.R., J.K., and K.H.P. were involved in the design, implementation, and data collection of the TwinFat study, with help from P.P., N.L., J.L., and A.H. in data collection. S.J. wrote the initial draft of the manuscript. R.H., C.D., P.P., N.L., J.L., A.H., A.R., J.K., K.O.K., T.I.A.S., and K.H.P. reviewed, commented, and edited the manuscript.

Disclosure Summary: The authors have nothing to disclose.

References

Author notes