Athletic endurance performance is probably partly under genetic control, but genetic association studies have yielded inconclusive results. The objective of the present study was to evaluate the association of polymorphisms in eight muscle- or metabolism-related genes with endurance performance in participants of the Olympus Marathon running race. We recruited 438 athletes who participated in the 2007 and 2008 annual running events of the Olympus Marathon: a 43.8-km race with an ascent from sea level to 2,690-m altitude and then a descent to 300 m. Phenotypes of interest were the competitive event time at the specific Olympus Marathon where the athlete was enrolled, the fastest reported timing ever achieved in an Olympus Marathon, and how many kilometers per week the athlete ran during the previous year. Eleven polymorphisms in α3-actinin (ACTN3), AMP deaminase-1 (AMPD1), bradykinin B2 receptor (BDKRB2), β2-adrenergic receptor (ADRB2), peroxisome proliferator-activated receptor (PPAR)-γ coactivator-1α (PPARGC1A), PPAR-α (PPARA), PPAR-δ (PPARD), and apoliprotein E (APOE) were evaluated. Hardy-Weinberg equilibrium testing on the overall cohort of male athletes showed a significant deviation for BDKRB2 rs1799722 (P = 0.018; P = 0.006 when limited to 316 habitual male runners) with an excess of the TT genotype. Across all athletes, no associations showed nominal statistical significance for any of the three phenotypes, and the same was true when analyses were limited to men (n = 417). When limited to 316 male athletes who identified running as their preferred sport, ADRB2 rs1042713 had nominally significant associations with faster times for the minor (A) allele for the fastest time ever (P = 0.01). The direction of effect was identical as previously postulated only for BDKRB2 rs1799722 and ADRB2 rs1042713, indicating consistency. BDKRB2 rs1799722 and ADRB2 rs1042713 have some support for being implicated in endurance performance among habitual runners and require further investigation.
physical fitness is a very complex phenotype influenced by numerous genetic and environmental factors contributing to the observed interindividual variation not only in the general population but also in trained athletes (24). Numerous studies have been performed in the last decade to try to establish whether specific genes with postulated functional roles in human physiology influence athletic performance and endurance, and a growing number of candidate gene associations have been proposed (5, 27). Studies have varied in sample size and type of population, and several proposed associations have not been consistently replicated in independent investigations by different teams of investigators.
Multiple genetic loci are likely to underlie the heritability of the complex phenotype of athletic performance. Some aspects of this phenotype have high estimated heritability, for example, the heritability of maximal O2 uptake (V̇o2max) is ∼50% in both the sedentary state and after training (3, 4). Even if heritability is less for other aspects of athletic performance, each genetic variant is likely to explain only a small fraction of the genetic predisposition. Recently, Williams and Folland (44) calculated the probability that exists for an individual to be in possession of any of the most optimal of the 23 selected polymorphisms related to athletic performance. Studies investigating the collective influence of a number of genetic polymorphisms could possibly help unravel and thus possibly better explain the inherent individual variations in athletic performance. In the present study, we studied participants in the Olympus Marathon, an athletic event that requires considerable endurance and for which there are large differences in performance among the participants. The optimal physiological phenotype of athletes competing in long-duration events probably entails an inherent genetic makeup conferring cardiovascular, pulmonary, and skeletal competence to perform during such events and efficient metabolism of available substrates to sustain the performance throughout the event's duration.
We evaluated single-nucleotide polymorphisms (SNPs) in eight genes that have known functional roles, and we based our selection on previous gene association studies investigating aspects of athletic endurance. In particular, the selected genes in our study have been previously associated with appropriate substrate metabolism as a source of energy during exercise, in the cardiovascular and pulmonary responses to exercise that would allow for a much efficient O2 uptake and transfer to the muscles for utilization, and/or in muscle contractility suitable for the purposes of endurance performance. Moreover, the selected polymorphisms had been evaluated in the past in one or more studies, where at least one of the studies had reported significant associations with endurance performance (5, 27).
We enrolled athletes participated in the annual Olympus Marathon event set on Mount Olympus in Greece. We only enrolled athletes of self-verified Greek ancestry to avoid confounding data from population stratification. Athletes were approached at the competition site during the 2 days before the race by a member of the research team. They were informed about the aim and purpose of the study and were asked whether or not they would be interested in participating in the study. Written informed consent was obtained from all volunteers. Athletes filled out a short questionnaire on personal identifiers; age, sex, and current height and weight; whether running was their preferred sport (and, if not, which sport was their preferred one); how many kilometers they had run per week in the last week, month, and year before the competition; the highest altitude they had ever reached; and the longest distance they had ever ran without stopping. A saliva sample was finally collected for DNA analysis. The data collection for the study occurred at the 2007 and 2008 annual events. The study protocol was reviewed by the University Hospital of Ioannina Scientific Committee, and ethical approval was granted.
Race course description.
The race course length is 43.8 km. It entails a gradual ascent from the start line located at 3 m above sea level, reaches a maximum altitude of 2,690 m, and then has a downhill route to the finish line at an altitude of 300 m. The race provides an excellent event for assessing variability in endurance performance. The winner usually completes the race in ∼5 h, but times of people that have completed the race vary substantially and can exceed 14 h.
Saliva samples (2 ml) were collected from all volunteers using the Oragene DNA Self Collection Kits (DNA Genotek) according to the manufacturer's instructions; the Oragene collection tube also contained 2 ml of stabilizing agent. All samples were collected at the competition site in Greece and were then sent for analysis at Erasmus University Rotterdam (Rotterdam, The Netherlands). Samples were processed using a PUREGENE DNA Purification kit (DNA Genotek's Oragene DNA Purification protocol, DNA Genotec, Ottawa, Ontario, Canada). For DNA isolation, 3 ml of the saliva mixture were used, and 120 ml of Oragene purifier were added. After a 10-min incubation on ice, the vial was spun down, and the pellet was discarded. Subsequently, an equal volume of ethanol was added to precipitate the DNA. After a 10-min incubation at room temperature, the vial was spun down, the supernatant was discarded, and the pellet was air dried. The DNA pellet was then dissolved in 300 ml MilliQ water.
Polymorphisms under study.
The selected genes, their respective SNP identification numbers, and the implicated function of these genes are shown in Table 1. For all genes under study, one variant has been investigated for associations with endurance performance and related phenotypes except for the peroxisome proliferator-activated receptor (PPAR)-δ gene (PPARD), where three variants have been proposed to affect endurance. For all SNPs, Taqman assays were generated and applied according to the manufacturers's specifications. Results were analyzed with the ABI Taqman 7900HT using sequence detection system 2.22 software (Applied Biosystems, Foster City, CA). To confirm the accuracy of the genotyping results, 5% of the randomly selected samples were regenotyped with the same method. No inconsistencies were observed. All primers and probes used in the present study are available on request.
We defined a priori the following phenotypes: 1) the event completion time at the specific Olympus Marathon where the athlete was enrolled in the study, 2) the fastest reported event completion time ever achieved in the Olympus Marathon race, and 3) how many kilometers per week the athlete ran during the 12 mo before enrollment in the study.
For athletes who ran both years and were included in our sample size, we used for the time to finish the time in the year during which they were first enrolled. However, if they only completed one of the two races, then we used the time when they completed the race. All athletes that exceeded the time limit for finishing the race set by the organizing committee (12 h in 2007 and 10 h in 2008) were personally contacted to provide a self-reported time finish, since the official time was not available.
Genotypes for each tested polymorphism were tested for compliance with the Hardy-Weinberg equilibrium (HWE) law using an exact test. Deviations from HWE may be due to different reasons (31), but, in this design, they may also be used as a crude test of association (36), since the analyzed cohort is a selected population with selection based on a phenotype (participation in a very demanding endurance event) that reflects overall endurance performance.
The main association analyses were performed using model-free ANOVA and allele-based linear regression models for continuous traits including only males. The major allele was considered as a reference. For the apolipoprotein E gene (APOE), the ε4/ε4 versus ε4/other versus other/other genotypes were considered for the ANOVA. Also, a contrast of ε2 versus other alleles was considered. The analysis of the event completion time at the specific Olympus Marathon was adjusted for age and year of the race (to allow for different event completion times due to any different weather conditions in different years). The best time ever achieved in the Olympus Marathon and the kilometers per week analysis were adjusted for age only. A Shapiro-Wilk test was performed to evaluate if the phenotypes were normally distributed. Given that we observed significant deviations from normality, a natural logarithm transformation was performed on all three phenotypes. For significant associations, a pairwise interaction analysis was performed.
We performed additional secondary analyses: a sex-stratified analysis where both men and women were considered and an analysis where we considered only male athletes who stated in the enrollment questionnaire that running was their preferred sport. Moreover, in the analysis of the event completion time at the specific Olympus Marathon, whenever an athlete did not complete the race the athlete was excluded from the main analysis; however, we also performed a sensitivity analysis where the timing of such dropout athletes was imputed by the time of the slowest athlete finishing the race. Finally, in another analysis, we tested if a specific variant affected the performance of an athlete by testing athletes that finished the race versus athletes who did not complete the event using a per-allele model.
Our study had 80% power to detect at α = 0.05 associations where the per-allele effect would be 1/10th the magnitude of the standard deviation of the phenotype of interest in the population, when the minor allele frequency would be 10%.
Associations that reach formal statistical significance at P < 0.05 do not necessarily mean that they are true. One may consider an approach to correct the P values for eight genes being evaluated and claim significance for P < 0.05/8 = 0.006 or, if all variants are considered independent, for P < 0.05/11 = 0.0045. However, such a correction would be debatable given that these gene variants have already been proposed and have at least one other study suggesting that they may be important. To avoid this controversy, we present inferences based on an alternative, Bayesian approach. Specifically, we estimated the Bayes factors according to a spike and smear model (10), if the mean effect of true associations might be 0.04 in the log scale (corresponding to a 9.6% difference in performance per allele, e.g., completing the marathon in 456 vs. 500 min). The inverse of the Bayes factor tells how many fold the odds of an association to be true increase based on the results of the study compared with what one thought before the study. Thus, if the odds of an association to be true were 1:9 [1/(1 + 9) = 1/10 = 10% likely to be true] before the study and the Bayes factor from the study data is 0.2 (inverse = 1/0.2 = 5), then the odds of the association to be true become 5:9 [5/(5 + 9) = 5/14 = 36% likely to be true]. Given that different investigators may have different beliefs, we also present credibility estimates (the likelihood of an association being true) for prior evidence of 0.1 (with a 1:10 prior chance of the association being true), 0.001 (a more conservative, sceptical approach, with a 1:1,000 chance of the association being true), and 0.000001 (a most sceptical approach, with a 1 in a million chance of the association being true, e.g., if the prior evidence mostly suggests that the association would have an effect in the opposite direction than what is seen in the present study).
All analyses were performed with STATA (version 10.0, College Station, TX). P values were two tailed. Reporting of the study data follows STREGA guidelines (17).
A total of 438 athletes (417 men and 21 women) volunteered for the study during the 2 yr. Two hundred sixty-eight athletes enrolled in 2007 and 229 athletes completed the race; in 2008, 170 athletes enrolled and 156 athletes finished the race. Nine athletes who did not finish the race in the year of enrolment (2007) completed the race in 2008. Hence, there were 44 athletes that did not ever finish the Olympus marathon. Population characteristics are shown in Table 2. Three hundred thirty-three athletes (76%) stated that their preferred sport was running, whereas the other participants mentioned climbing (n = 18), cycling (n = 16), or other sports (n = 72). The training of athletes that do not compete in running may not be focused in the development of an endurance phenotype. Runners completed the race in faster times than the other group of athletes; however, this difference was not statistically different. Runners versus other athletes had a significantly higher number of kilometers run per week in the previous year (mean: 56.4 vs. 37.2, P < 0.0001) and previous month (mean: 63.4 vs. 46.8, P < 0.0001) but did not differ in age, sex, weight, and height.
Distribution of alleles and genotypes.
The frequencies of the polymorphisms are shown in Table 1. The undetermined genotypes were 2% or less for all polymorphisms.
With one exception, the distribution of genotypes for all polymorphisms did not deviate significantly from HWE. Specifically, bradykinin B2 receptor (BDKRB2) rs1799722 deviated from the equilibrium (CC: n = 162, CT: n = 173, and TT: n = 76, exact P value = 0.018) for male athletes because of an excess of the TT genotype. The deviation was also statistically significant when we analyzed men that stated that running was their preferred sport (CC: n = 127, CT: n = 125, TT: n = 59, exact P value = 0.006) as well as all (both male and female) athletes who stated that running was their preferred sport (CC: n = 133, CT: n = 132, TT: n = 63, exact P value = 0.005), always with an excess of the TT genotype. HWE did not deviate significantly among athletes that did not complete the race.
When 417 male athletes were considered, there were generally formally no statistically significant findings observed for the polymorphisms under study regardless of whether a model-free ANOVA or an allele-based model was used, as shown in Table 3. The results were largely unaltered when both men and women were considered and when the missing values were imputed with the slowest time reported (not shown).
When we performed the analysis with male athletes that stated that running was their preferred sport of interest (Table 4), an association in one polymorphism [β2-adrenergic receptor (ADRB2) rs1042713] reached nominal statistical significance, with faster times for the A allele for the fastest time ever (P = 0.01). For the specific association, we performed a sensitivity analysis where the training volume (expressed as kilometers per week the athlete ran during the 12 mo before enrollment in the study) was used as a covariate. The association remained significant for rs1042713 (P = 0.035). Also, a nominal significant P value of 0.04 was found for AMP deaminase-1 (AMPD1) rs17602729 for the fastest time ever using model-free ANOVA (Table 4).
These two variants (AMPD1 rs17602729 and ADRB2 rs1042713) reached statistical significance, both for the time at the specific event and for the fastest time ever, and with similar per-allele effects for the two phenotypes when we performed an analysis including both men and women. For AMPD1 rs17602729, the P values were 0.021 and 0.03 for event completion time and for the fastest time ever, respectively. For ADRB2 rs1042713, the P values were 0.015 and 0.003, respectively.
In a secondary analysis, a pairwise interaction analysis of the significant SNPs did not provide any significant findings (not shown in detail). Moreover, an analysis of athletes that completed the race versus athletes that did not finish the race did not yield any significant findings (data not shown).
In the per-allele models, for the analysis including only male runners, the estimated Bayes factor was 0.22 for ADRB2 rs1042713 for the fastest time ever achieved. When the analysis of the runners included both sexes, the estimated Bayes factors were 0.27 and 0.24 for the time in the specific event and 0.34 and 0.06 for the fastest time ever achieved for AMPD1 rs17602729 and ADRB2 rs1042713, respectively. These correspond, respectively, to 5-, 4-, 4-, 3-, and 16-fold increases in the odds that the association is true compared with the prior belief before the study. For a prior credibility of 10% (1:10 likely to be true), the credibility of the associations was 31% for the fastest time ever achieved in the analysis of male runners. For all runners, the credibility was 27% and 29% for the time in the specific event and 23% and 62% for the fastest time ever achieved for the two polymorphisms, respectively. For a prior credibility of 1:1,000 or 1:1,000,000, the credibility of the associations would be negligible (<0.1% for all four analyses).
We investigated 11 polymorphisms in 8 muscle- and/or metabolism-related genes in a large group of athletes that competed in a special athletic event that is highly demanding for endurance. In this selected cohort, BDKRB2 rs1799722 deviated significantly from HWE with an excess of TT individuals, suggesting that this genotype may be associated with endurance performance in general. Moreover, we found associations with nominal statistical significance for ADRB2 rs1042713 for the best time ever among male athletes where running is their preferred sport and associations for AMPD1 rs17602729 and ADRB2 rs1042713 for the time of the specific event and best time ever among all (both male and female) runners. Given that these genes and variants have clear functional roles and have also been proposed as potentially important by previous studies, modest signals should not be dismissed, if they also agree with other prior evidence.
Our findings are in agreement with prior evidence for ADRB2 rs1042713 (Arg16Gly). ADRB2 rs1042713 encodes the β2-adrenergic receptor that is primarily responsible for increases in bronchodilation, ventricular function, and vasodilation, all of which have direct implications on cardiovascular and pulmonary responses to exercise (36). In a case-control study (n = 313 endurance athletes competing in 8 endurance sporting disciplines vs. n = 297 sedentary controls), the A (Arg16) allele was significantly more prevalent in the endurance athletes (45). ARDB2 regulates the cardiopulmonary response to exercise (36), and this polymorphism has been investigated in many phenotypes, with some evidence of lower mean arterial blood pressure at rest, during and after exercise for Arg16 homozygotes (37), decreased exercise performance in patients with heart failure with Gly16 (42), and decreased thermogenic responses to β2-adrenergic stimulation with Arg16 (26).
AMPD1 (rs17602729), which functions in skeletal muscle metabolism (salvage of adenine nucleotides) and is involved in the regulation of muscle glycolysis during rigorous exercise (29), was evaluated in a case-control study (n = 104 endurance athletes vs. 100 controls). The minor (A) allele was actually significantly less common in endurance athletes than in controls, but endurance indexes such as V̇o2max, ventilatory threshold, and respiratory compensation threshold were not significantly different between athlete carriers and noncarriers of the minor allele (29). Furthermore, AMPD1 rs17602729 was evaluated in a study assessing physiological responses in 503 individuals undergoing a 20-wk training regime, and the minor allele had smaller ventilatory training responses pertaining to endurance (28). Therefore, our results are in the opposite direction for this polymorphism than what prior studies had suggested. Discrepancies may be due to different study design, setting, types of athletes, and phenotype definitions, but most likely they simply represent false positives in opposite directions, and this polymorphism is not credibly associated with endurance performance. Incidentally, there is a case report (19) on a world-class athlete with exceptionally favorable laboratory indexes for endurance performance (V̇o2max) despite carrying the A allele (19), which further documents that these genetic effects, even if present, are subtle and do not doom one's athletic performance.
Variation at the BDKRB2 (rs1799722) gene locus, which encodes a receptor for bradykinin and is implicated in the increase of skeletal muscle glucose uptake during exercise (34), has also been associated with endurance performance in previous studies. In one study (43), athletes (n = 81) competing in longer distance events had a significantly higher frequency of the −9 allele, a 9-bp insertion/deletion polymorphism in exon 1. In a study of 453 athletes who completed the South Africa Ironman Triathlon, the athletes had a significantly higher frequency of the −9/−9 genotype compared with the controls (n = 203) (34). In that same study (34), the −9 allele was not significantly different in three groups of athletes who completed the event in fast, medium, and slow times (34). The −9 allele has been demonstrated to result in increased transcription rates of the gene (6). In our study, we focused only on SNPs rather than insertion/deletion polymorphisms, and we instead genotyped the promoter variant rs1799722 (also known as −58C/T), which is also known to have a functional impact on the gene, with increased transcription rates for the T allele in luciferase experiments (15). We found results consistent with previous evidence: the high transcription allele was overrepresented in this group of endurance athletes and, even more so, among those who are habitual runners. The same promoter variant has also been associated with hypertension, left ventricular hypertrophy, and baroreflex sensitivity (7, 22).
All of the other polymorphisms that showed no nominally significant signals in our study have been examined in the past in at least one other study, but the evidence was often inconclusive or may be stronger for associations with phenotypes that have only indirect bearing on athletic endurance performance. Moreover, we only examined specific polymorphisms for each one of these genes, so we cannot exclude that other variants in the same gene locus may have an effect on athletic performance in particular populations. Finally, our study was powered to detect modest effects, but subtle effects for variants with minor allele frequencies <10% may have been missed.
Specifically, in α3-actinin (ACTN3; rs1815739), which codes for the synthesis of actinin-3 (a major skeletal muscle constituent) and is implicated in fast contraction ability by muscles (46), the T allele was reported at somewhat higher frequencies in endurance athletes than controls in two studies (25, 46), but the differences were not formally significant. Moreover, rs1815739 was not associated with a continuous endurance outcome (based on competitor finish time) during the South Africa Éronman Triathlon competition (n = 457 athletes vs. 143 controls) (33), and no differences were observed in a case-control study (18) (n = 52 Olympic-level runners vs. 50 professional cyclists vs. 123 controls) between groups and when laboratory endurance performance indexes were evaluated.
PPAR-α (PPARA) rs4253778 is an intronic variant that is involved in the regulation of liver, heart, and skeletal muscle lipid metabolism as well as glucose homeostasis (39). This variant was investigated in a multisport case-control study (2) (n = 786 Russian athletes vs. 1,242 controls) stratified by performance (endurance, sprint, and mixed) in which endurance athletes had significantly higher frequency of the G allele than controls. The same polymorphism was found to be associated with triglyceride and apolipoprotein CIII levels in Africa-descended people but not in Caucasians (35). It is unknown whether ancestry-specific genetic effects (11) may also be important in athletic performance.
PPARD (rs1053049, rs6902123, and rs2267668) is involved in fatty acid β-oxidation, glucose utilization, mitochondrial biogenesis, angiogenesis, and muscle fiber type (39), and PPAR-γ coactivator-1α (PPARGC1A; rs8192678) regulates genes involved in energy metabolism and is associated with mitochondrial biogenesis and skeletal muscle fiber type conversion (20). Both genes seem to have independent effects on the effectiveness of aerobic exercise training to increase aerobic physical fitness and insulin sensitivity (39). Furthermore, a whole body MRI study (38) found differential changes after lifestyle intervention in overall adiposity, hepatic fat storage, and relative muscle mass for the rs1053049, rs6902123, and rs2267668 variants of PPARD (38). Moreover, in a multisport case-control study (1) (n = 1,256 athletes vs. 610 controls), PPARD allelic differences were seen in endurance athletes but not in controls (1), and another case-control study (20) found that the minor allele of PPARGC1A rs8192678 was significantly more common in unfit controls (n = 100) than in endurance athletes (n = 104). However, it is unclear whether such associations would result also in noticeable differences in performance among endurance athletes besides differentiating athletes from controls. Similarly, for apolipoprotein E (APOE; rs7412 and rs429358), which facilitates triglyceride clearance by mediating lipoprotein binding to hepatic receptors, thus contributing to the variability in individual response to exercise training (41), the ε4 allele in previous studies (8, 41) was found to be associated with better physiological responses after exercise training, but this may not necessarily translate to better endurance performance. In all, more extensive replication would be essential (10, 15) before making claims that any of these variants have a substantial effect specifically on endurance performance.
Furthermore, three case-control studies (30, 32, 23) examined the representation in athletes and controls of seven polymorphisms, three of which (ACTN3 rs1815739, AMPD1 rs17602729, and PPARGC1A rs8192678) were also investigated in our study. In the first study (3) (n = 46 endurance athletes vs. 123 controls), a point was given for each “favorable allele,” and the overall mean total score was higher in athletes than in controls. The second study (32) (n = 39 world-level athletes vs. 15 national-level athletes) found no differences in genetic profiles between the two groups of athletes. The third study (23) compared many groups (n = 50 professional cyclists vs. 52 Olympic-level runners vs. 39 world-level rowers vs. 123 controls), and no significant differences among the groups were found. Given the relatively limited sample size, it is difficult to draw conclusive inferences.
Some limitations should be discussed. We used a cohort design and did not make a comparison against matched nonathlete controls from the general population. Therefore, we performed two analyses for association: a first crude screening based on the Hardy-Weinberg test and another based on association for continuous outcomes of endurance performance within the cohort of athletes. Hardy-Weinberg testing has low power as an association test, and subtle associations may have been missed in this screening. The analysis of comparative performance is based on a continuous outcome, and it has good power to detect modest differences, but most genetic effects for common variants represent effects with very subtle magnitude (13), and such effects could still have been missed. Conversely, nominally significant associations should be interpreted with caution since they may still represent false positives, and we encourage further replication of our findings in additional studies and settings (14). Third, genes and polymorphisms were selected based on considerable prior knowledge on the function of the proteins encoded by these genes and also some prior epidemiological evidence. With the advent of high-throughput genotyping, the field of athletic performance should also consider the conduct of agnostic genome-wide association studies (21), although for variants emerging from such studies it is often difficult to establish their physiological functional role (10).
Overall, human physical performance is a multifactorial phenotype in which numerous genetic and environmental factors have joint effects. Athletic performance is very much influenced by training programmes and opportunities, residing altitude, recovery, nutrition, sporting equipment, and psychological factors during competition. Some of these factors may also be reciprocally determined by genetic input to some extent, e.g., the propensity to exercise may also be a genetic trait (38). In our study, we also tried to perform analyses limited to athletes that had a strong preference for running, but there was still diversity in some other exposures, e.g., the amount of training for these athletes. Dissecting the complex effects on athletic excellence will require large-scale studies with meticulous measurements and replication of proposed associations in diverse datasets.
No conflicts of interest are declared by the author(s).
The authors thank F. Kavvoura, D. Panagiotopoulos, M. Matziaris, T. Thanopoulos, and K. Gousis for the invaluable assistance during data collection.
- Copyright © 2010 the American Physiological Society