Genetic Variants and Hamstring Injury in Soccer

An Association and Validation Study

Abstract

Purpose 

This study aimed to investigate the association of candidate single nucleotide polymorphisms (SNP) with noncontact hamstring muscle injuries in elite soccer players and to create and validate a model to assess the risk of hamstring injury.

Methods 

A total of 107 elite male outfield players were prospectively followed for six seasons. Players were genotyped for 37 SNP previously investigated in relation to musculoskeletal injuries. The association of SNP, previous injury, age, level of play, position, and anthropometric data with 129 hamstring injuries (413 observations) was investigated in the discovery phase (2010–2015), and a multivariable Cox frailty model was created using forward selection. The model’s discriminative ability was tested in the validation phase (2015–2016, 31 injuries, 98 observations) using Harrell’s C index.

Results 

Five SNP were found to be significantly associated with hamstring injury in a multivariable model: matrix metalloproteinase 3 rs679620 (A vs G, hazard ratio [HR] = 2.06, 95% confidence interval [CI] = 1.51–2.81), tenascin C rs2104772 (A vs T, HR = 1.65, 95% CI = 1.17–2.32), interleukin 6 rs1800795 (GG vs GC + CC, HR = 1.68, 95% CI = 1.11–2.53), nitric oxide synthase 3 rs1799983 (G vs T, HR = 1.35, 95% CI = 1.01–1.79), and hypoxia-inducible factor-1α rs11549465 (CC vs CT, HR = 2.08, 95% CI = 1.00–4.29). Age also entered the model (≥24 vs <24 yr, HR = 2.10, 95% CI = 1.29–3.42). The model showed acceptable discrimination in the discovery phase (C index = 0.74), but not in the validation phase (C index = 0.52).

Conclusion 

Genetic variants appear to be involved in the etiology of hamstring injuries but were not found to have predictive value by themselves. Further research, increasing the number of genetic variants and including environmental factors in complex multifactorial risk models, is necessary.

Soccer injuries affect team performance negatively, have high economic costs, and might induce long-term health consequences (^1,2). Hamstring muscle injury is the most frequent injury in elite male soccer (³), and identifying those at risk and preventing hamstring injuries is a priority. Risk factor studies have revealed previous hamstring injury to be highly associated with the occurrence of hamstring injuries in male soccer players. Other risk factors, such as, older age, reduced hamstring flexibility, decreased hamstring strength or strength imbalances, and fatigue show conflicting or limited evidence (⁴).

Previous research has also suggested that a genetic susceptibility may contribute to the interindividual variation in musculoskeletal injury risk. Several single nucleotide polymorphisms (SNP) located in genes responsible for encoding the structural and regulatory proteins of musculoskeletal soft tissues have been associated in case–control retrospective studies with injuries, such as anterior cruciate ligament (ACL) rupture and chronic Achilles tendinopathy (^5,6). By contrast, very few studies have investigated noncontact muscle injuries (^7,8). In addition, variants associated with exercise-induced muscle damage have been pointed out as potential markers of muscle injury risk (⁹). These polymorphisms might contribute to interindividual variation in the structural and functional properties of muscle and tendon and their response to mechanical loading, thus potentially being implicated in the susceptibility to hamstring injury (⁵).

However, there is no evidence regarding the influence of genetic variants on the risk of hamstring injury. Because statistically significant associations might not be enough to predict players at risk of injury, the predictive ability of any test needs to be validated in independent samples (¹⁰). Thus, the aims of this study were to investigate the association of candidate genetic variants with noncontact hamstring injuries in elite soccer players over several seasons and to create a model to estimate the risk of hamstring injury and test its validity.

METHODS

Participants and study design

This study was approved by the Clinical Research Ethics Committee of the Basque Country (PI2014215). A total of 107 male outfield players from Athletic Club voluntarily agreed to participate after receiving oral and written details outlining the study. Informed consent was obtained from each participant. All players were Caucasian from the Basque Country in Spain. Players were recruited, and saliva samples were collected at the beginning of the 2014–2015 season. Twenty-eight players belonged to the First team, 43 to the two Reserves teams, and 36 to the two U19 teams. All players from the First, Reserves, and U19 teams had been prospectively followed from the 2010–2011 season to the 2015–2016 season, and injury records, exposure time, and anthropometric data were collected by the medical and coaching staff following common procedures. The study was divided in two phases (Fig. 1): 1) the discovery phase, from the 2010–2011 season to the 2014–2015 season, when the association between risk factors and hamstring injuries was investigated, and 2) the validation phase in the 2015–2016 season, when the predictive ability of the risk model was assessed.

FIGURE 1: Schematic diagram of the study design.

Injury, exposure time, and anthropometric data recording

Time-loss injuries resulting from soccer training or match play were recorded following the consensus on definitions and data collection procedures outlined by the International Federation of Association Football (¹¹). Noncontact hamstring injuries were recorded when a player was unable to participate in a future training session or match because of an injury to the hamstring muscle group and was considered injured until the medical staff cleared the player for full participation in training and match play. Structural–mechanical injuries, such as total and partial muscle ruptures, and functional injuries, such as fatigue-induced or neurogenic muscle hardening (hypertonia), were included (³). Injuries were confirmed through a clinical examination by the team doctor, and if indicated, the diagnosis was supported by ultrasonography and magnetic resonance imaging. Injuries during national team duties were also registered. Hamstring injuries with a sudden, identifiable onset were defined as acute injuries, whereas those with a gradual onset were defined as overuse injuries. According to the number of days of absence, injury severity was recorded as minimal (1–3 d), mild (4–7 d), moderate (8–28 d), or severe (>28 d). Recurrent hamstring injuries were those occurring in the same leg and during the same season as an index hamstring injury.

Individual player exposure time in training and matches (friendly and competitive), including national team exposure, was daily recorded in minutes. Anthropometric data were collected every 2 months approximately by the team doctor. Height was measured using a stadiometer (Añó Sayol, Barcelona, Spain), and body mass was measured with a portable balance (Seca, Bonn, Germany). Skinfold thicknesses were measured at six sites (triceps, subscapular, abdominal, suprailiac, thigh, and calf) using a skinfold caliper (Harpenden, West Sussex, England), and the sum of these six skinfolds was calculated in millimeters.

Genotyping

Thirty-seven SNP previously investigated in relation to musculoskeletal injuries (^6–8) or exercise-induced muscle damage (⁹) were selected for the study. The full list of SNP and associated injuries are presented in Table, Supplemental Digital Content 1, Associated injuries, genotype frequencies and missing data of the selected candidate SNP, https://links.lww.com/MSS/B62. For a more detailed information on these SNP, readers are referred to recent reviews (^6,9). Saliva samples were obtained using buccal swabs (4N6FLOQSwab, Life Technologies, Carlsbad, CA). DNA was extracted via QIAmp DNA Mini kit (Qiagen, Hilden, Germany) and quantified by fluorometry using Qubit (Life Technologies). DNA samples were genotyped using SNP-type assays in the Biomark HD system (Fluidigm, South San Francisco, CA).

Statistical analysis

The required sample size was calculated using the powerSurvEpi package in R version 3.2.3 (R Core Team 2015, R Foundation for Statistical Computing, Vienna, Austria). With 80% power, a two-sided significance of 0.05, and injuries occurring in 25% of observations, to detect a hazard ratio (HR) of 2, the minimum required number of injuries was 89. Injury incidences are presented as the number of injuries/1000 player hours with 95% confidence intervals (CI). Descriptive data are presented as mean ± SD.

The Cox proportional hazards model with a frailty extension was used to investigate the association between risk factors and hamstring injuries in the discovery phase, using the survival package in R (¹²). This model accounts for varying exposure times between players (measured as total hours of exposure in each season), and the frailty term allows for correlation between observations from the same player to be accounted for (^13,14). Observations started at the beginning of each season. Some players had no occurrence of injury during the season and contributed censored survival times, whereas other players sustained one or more hamstring injuries and had multiple observations.

First, potential risk factors (37 SNP, age, height, body mass, sum of six skinfolds, level of play, position, and previous hamstring injury during the preceding or same season) were individually analyzed adjusting for the players’ match exposure ratio (match hours/total hours of exposure) (¹⁵). Individual analyses were separately performed for all, acute, overuse, severe, and recurrent hamstring injuries. For recurrent hamstring injuries, each observation started when a player suffered an index hamstring injury. The analysis of previous injuries included only prospectively recorded injuries in the club, and hence, the players’ first season in the club was not considered for the analysis (¹⁵). Continuous variables were categorized according to the optimal cutoff value using the CatPredi package in R (¹⁶). Each SNP was analyzed under dominant (aa + Aa vs AA), recessive (aa vs AA + Aa), overdominant (Aa vs AA + aa), and log-additive (aa = 2, Aa = 1, AA = 0) modes of inheritance, and the best mode for each SNP was selected based on the minimum P value.

Subsequently, and only for all hamstring injuries, variables with P ≤ 0.25 were entered in a multivariable Cox frailty model using forward selection. At each step, variables with P ≤ 0.05 were separately added to the model, and the model with the smallest Akaike information criterion value was retained until no variable showed P ≤ 0.05. HR and 95% CI were calculated. The proportional hazard assumption was assessed using the cox.zph function in R. Kaplan–Meier survival curves were plotted to illustrate the probability of remaining injury-free during a season using GraphPad Prism v.6.0c (GraphPad Software, La Jolla, CA). The significance level was set at P ≤ 0.05.

Finally, a risk score for each player relative to the average player within the data set was estimated from the model. The discriminative ability of the model was tested separately in the discovery and validation phases calculating Harrell’s C index. Harrell’s C (for concordance) index estimates the probability that, of two randomly chosen players, the player with the higher risk score will be more likely to sustain an injury compared with the player with the lower risk score. Values of C index near 0.5 indicate that the model is as good as a random guess, whereas a value of 1.0 indicates that the model always discriminates players with a higher risk (¹⁷).

RESULTS

A total of 107 players (20 ± 4 yr, 179 ± 5 cm, 72 ± 6 kg, 51 ± 12 mm of skinfolds) were followed up for at least one season for a total of 356 player-seasons (3 ± 1 seasons per player). Descriptive data on player exposure and hamstring injuries are presented in Table 1. The discovery phase consisted of 413 observations and 129 hamstring injuries (107 players), whereas 98 observations and 31 hamstring injuries (67 players) were included in the validation phase (Fig. 1). Genotype frequencies are presented in Table, Supplemental Digital Content 1, Associated injuries, genotype frequencies and missing data of the selected candidate SNP, https://links.lww.com/MSS/B62. Two SNP had >5% missing data, COL1A1 rs1800012 (10%) and COL5A1 rs12722 (10%).

TABLE 1: Descriptive data on player exposure and hamstring injuries.

Analysis of individual SNP revealed seven polymorphisms significantly associated with the risk of hamstring injury (Table 2). Age was the only nongenetic variable significantly associated with hamstring injuries, even after adjusting for the level of play (≥24 vs <24 yr, HR = 3.33, 95% CI = 1.38–8.02, P = 0.01). Matrix metalloproteinase 3 (MMP3) rs679620 remained statistically significant when acute, overuse, severe, and recurrent hamstring injuries were separately analyzed (Table 3; see full Tables, Supplemental Digital Content 2, Association between acute, overuse, severe, and recurrent hamstring injuries and genetic and nongenetic factors in elite soccer players, https://links.lww.com/MSS/B63). Previous hamstring injury was significantly associated only with acute hamstring injuries. In a multivariable model, five SNP and age were significantly associated with hamstring injury (Table 4). Kaplan–Meier survival curves for these variables are shown in Figure 2. These results show a higher hamstring injury risk for players older than 24 yr, and for players with the MMP3 rs679620 AA, tenascin-C (TNC) rs2104772 AA, interleukin 6 (IL6) rs1800795 GG, nitric oxide synthase 3 (NOS3) rs1799983 GG, and hypoxia-inducible factor 1α (HIF1A) rs11549465 CC genotypes. All significant variables met the proportional hazards assumption. Finally, the C index of the model was 0.74 in the discovery phase and 0.52 in the validation phase.

TABLE 2: Individual analysis of genetic and nongenetic risk factors for hamstring injuries in elite soccer players.

TABLE 3: Individual analysis of genetic and nongenetic risk factors for hamstring injury subtypes in elite soccer players.

TABLE 4: Multivariable Cox frailty model for the association between risk factors and hamstring injuries in elite soccer players.

FIGURE 2: Kaplan–Meier survival curves illustrating the probability of remaining hamstring injury free during a season for the risk factors significantly associated with hamstring injury in a multivariable Cox frailty model.

DISCUSSION

Five SNP and age were associated with hamstring injury in a Cox frailty model

The most strongly associated SNP was MMP3 rs679620 G/A, with each copy of the A allele increasing the risk of hamstring injury twice compared with the GG genotype. It was also the only SNP significantly associated with acute, overuse, severe, and recurrent hamstring injuries. MMP3 plays an important role in the maintenance of myofiber functional integrity by breaking down components of the extracellular matrix and in the regulation of skeletal muscle cell migration, differentiation, and regeneration (¹⁸). MMP3 rs679620 is in linkage disequilibrium with MMP3 rs3025058 5A/6A (¹⁹), a functional promoter polymorphism. The 5A allele, which is linked to the A allele of rs679620, has been shown to result in a higher MMP3 expression compared with the 6A allele (²⁰). Conversely, this SNP was not associated with noncontact skeletal muscle injuries in a previous study in elite soccer players, although there are large methodological differences with the present study in terms of player ethnicity, statistical analysis, and injury definition (⁷). Moreover, the GG genotype was overrepresented in individuals with Achilles tendinopathy compared with asymptomatic controls (²¹), but these findings were not replicated in another cohort and no association was found with the risk of ACL rupture (¹⁹).

Among the other significant SNP, each A allele of TNC rs2104772 A/T was associated with 1.65 times higher risk of hamstring injury. Tenascin C is a glycoprotein that regulates cell–matrix interactions, plays an important role in the muscle damage–repair cycle, and provides strength and elasticity to withstand mechanical forces. It is expressed in regenerating myofibers and in response to mechanical loading in the myotendinous junction, the most vulnerable site to injury (²²). The T > A substitution results in a leucine to isoleucine amino acid change in the fibronectin type III-D domain region of TNC that could cause structural instability and alter the molecular elasticity of the domain (²³). The A allele was previously associated with Achilles tendinopathy (²⁴), but not with noncontact muscle injuries (⁸).

The GG genotype of IL6 rs1800795 G/C was associated with 1.68 times higher risk of hamstring injury compared with the GC and CC genotypes. The cytokine IL6 is produced by skeletal muscle following exercise, and it also targets skeletal muscle, paradoxically, as both stimulator of hypertrophic muscle growth and myogenesis and promoter of atrophy and muscle wasting (²⁵). The G allele appears to increase IL6 gene transcription and plasma levels in response to stress stimuli (²⁶), and it has been previously associated with Achilles tendinopathy (²⁷), lumbar disc degeneration (²⁸), and power/strength athlete status (²⁹). By contrast, the CC genotype was associated with higher creatine kinase levels after eccentric exercise in healthy individuals (^9,30).

Each G allele of NOS3 rs1799983 G/T was associated with 1.35 times higher risk of hamstring injury. NOS3 is the rate-limiting enzyme for nitric oxide production. Nitric oxide has many relevant biological functions, such as, regulation of blood flow, muscle contractility, mitochondrial respiration, and skeletal muscle injury repair (³¹). NOS3 produced from the G allele seems to be less susceptible to proteolytic cleavage, which might result in increased NOS3 activity and higher NO production (³²). This SNP was not previously associated with Achilles tendinopathy (³³).

The risk of hamstring injury was twice as high in players with the HIF1A rs11549465 CC genotype in comparison with those with the CT genotype. Hypoxia-inducible factor 1α is a transcription factor regulating several genes in response to hypoxia, stimulating angiogenesis and glycolytic metabolism (³⁴). It can also be induced by mechanical loading, and it is an important component of matrix remodeling and skeletal myogenesis (^34,35). Previously, the T allele was linked with higher transcriptional activity of HIF1A (³⁶) and power/strength athlete status (²⁹), but this SNP was not associated with ACL injury (³⁷) or lumbar disc degeneration (³⁸).

Collectively, these five variants, or other closely linked polymorphisms, might influence musculotendinous integrity and function and its response to mechanical loading. Nonetheless, mechanistic studies are required to unravel the molecular mechanisms behind these associations (⁵).

Lastly, players older than 24 yr had two times higher risk of injury compared with younger players, and the association was independent of the level of play. This association was observed also in overuse, but not acute hamstring injuries. Previous studies show conflicting evidence with regard to the effect of age, which may be due to differences in mean age and level of play between study cohorts (⁴). Older players might be at a higher risk of injury because of age-related physical changes or a greater likelihood of having suffered a previous hamstring injury (^4,15).

The model did not have predictive ability in a subsequent independent season

The multivariable Cox model can be used to estimate the risk of injury of each player relative to the average player within the data set. This might be useful to classify players into risk groups or to create a risk profile of each player if the risk of various injuries could be estimated. Unfortunately, despite an acceptable internal concordance (C index = 0.74), the model was as good as a random guess in a subsequent independent season (C index = 0.52). This means that of two random players, the player with the higher risk score was the one that would get injured only half of the time (¹⁷). This result shows the importance of appropriate validation studies, as statistically significant associations might not translate into accurate predictive tests (¹⁰).

The lack of predictive ability may be due to several reasons. Sample size was small in the validation phase, and replication in larger samples might be necessary. However, the accuracy of a screening test in any given season is relevant for soccer clubs. The candidate gene approach is also limited, and the number of genetic variants investigated needs to be increased to understand the influence of genetics on musculoskeletal injury risk (^5,39). Most importantly, injuries are multifactorial disorders, and the use of genetic tests is very limited without considering other potential risk factors (e.g., training load, fatigue, hamstring activation, eccentric strength and fascicle length, fixture congestion, high intensity running, and compliance with preventive training) (^4,5). In this regard, preventive exercises were performed routinely by all players in the study, including Nordic hamstring exercises, core exercises, and strength training. However, this information was not registered, and it is a limitation of the study. Lastly, accurately identifying players at risk is challenging, and there are currently no screening tests available to predict sports injuries (¹⁰). Therefore, to understand such a complex phenomenon, a complex system approach may well be necessary, investigating the influence of genetic variants in interaction with other environmental risk factors (^5,40).

In light of the present findings, the use of genetic testing for hamstring injuries in soccer seems premature. Because of the inaccuracy of current screening tests and the high frequency of hamstring injuries in elite soccer, all players should be included in hamstring injury prevention programs. Research in genetics, overcoming the limitations of the present study, still holds great potential for injury risk screening and prevention. Although genetic testing will never be prognostic or predictive, it may provide information about the baseline injury risk of an individual. As an important piece of the multifactorial injury model, genetic information might be used together with all other risk factors to identify those at high risk of injury and individualize preventive strategies (^5,10).

Other methodological considerations

This is the first study to prospectively investigate genetic risk factors for hamstring injuries, in an ethnically homogeneous sample of elite soccer players, and using previously recommended statistical methods accounting for individual player exposure time and correlation between injuries (^13,14). Nevertheless, the study has limited external validity as only players from one club were investigated, and findings remain to be replicated in other populations. Moreover, the study had adequate power to detect moderate HR when including all hamstring injuries in the analysis, but sample size was insufficient for the analysis of specific types of hamstring injury. In this sense, the influence of genetics might be different for hamstring injuries with different mechanism, size, and location, and hence, a larger sample of well-defined injuries is required. Finally, two important SNP, COL1A1 rs1800012 and COL5A1 rs12722, had 10% missing genotype data because of problems with the genotyping, which might have influenced the results.

CONCLUSION

Five SNP (MMP3 rs679620, TNC rs2104772, IL6 rs1800795, NOS3 rs1799983, and HIF1A rs11549465) and older age were significantly associated with the risk of hamstring injury in a Cox frailty model over five seasons in elite soccer players. MMP3 rs679620 was the only variable individually associated with acute, overuse, severe, and recurrent hamstring injuries. However, the model could not identify players at higher risk of injury in a subsequent independent season, and genetic testing for hamstring injury risk seems premature at the moment. Further research in larger cohorts, increasing the number of genetic variants, and including environmental risk factors would appear to be necessary to understand the influence of genetics on musculoskeletal injuries. Such evidence might be used in the future to assess the injury risk of a player and to make informed decisions about preventing hamstring injuries in soccer.

The study was funded by the genetics company Baigene. J. L. was supported by a Ph.D. studentship from the Vice-Chancellorship for Basque of the University of the Basque Country UPV/EHU (Euskararen arloko Errektoreordetza). IB acknowledges financial support from the Basque Government (IT620-13). S. M. G. acknowledges financial support from the Basque Government (IT922-16) and the University of the Basque Country (PPG17/34). Genotyping was conducted by the Sequencing and Genotyping Unit of the University of the Basque Country (SGIker, UPV/EHU).

D. C., J. R. F. L., R. N., and J. M. A. are members of Baigene. For the remaining authors, no conflicts of interest were declared. The results of the present study do not constitute endorsement by the American College of Sports Medicine. The results of the study are presented clearly, honestly, and without fabrication, falsification, or inappropriate data manipulation.