Detection of epistasis between ACTN3 and SNAP-25 with an insight towards gymnastic aptitude identification

Abstract

In this study, we performed an analysis of the impact of performance enhancing polymorphisms (PEPs) on gymnastic aptitude while considering epistatic effects. Seven PEPs (rs1815739, rs8192678, rs4253778, rs6265, rs5443, rs1076560, rs362584) were considered in a case (gymnasts)–control (sedentary individuals) setting. The study sample comprised of two athletes’ sets: 27 elite (aged 24.8 ± 2.1 years) and 46 sub-elite (aged 19.7 ± 2.4 years) sportsmen as well as a control group of 245 sedentary individuals (aged 22.5 ± 2.1 years). The DNA was derived from saliva and PEP alleles were determined by PCR, RT-PCR. Following Multifactor Dimensionality Reduction, logistic regression models were built. The synergistic effect for rs1815739 x rs362584 reached 5.43%. The rs1815739 x rs362584 epistatic regression model exhibited a good fit to the data (Chi-squared = 33.758, p ≈ 0) achieving a significant improvement in sportsmen identification over naïve guessing. The area under the receiver operating characteristic curve was 0.715 (Z-score = 38.917, p ≈ 0). In contrast, the additive ACTN3 –SNAP-25 logistic regression model has been verified as non-significant. We demonstrate that a gene involved in the differentiation of muscle architecture–ACTN3 and a gene, which plays an important role in the nervous system–SNAP-25 interact. From the perspective originally established by the Berlin Academy of Science in 1751, the matter of communication between the brain and muscles via nerves adopts molecular manifestations. Further in-vitro investigations are required to explain the molecular details of the rs1815739 –rs362584 interaction.

About the Authors:

Łukasz Andrzej Płóciennik

Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing

* E-mail: [email protected]

Affiliations Department of Physical Education, Academy of Physical Education and Sport in Gdansk, Gdansk, Pomorskie Voivodeship, Poland, FitnessFitback, Pomorskie Voivodeship, Poland

ORCID logo http://orcid.org/0000-0003-2383-1336

Jan Zaucha

Roles Supervision, Writing – review & editing

Affiliation: Department of Bioinformatics, Wissenschaftszentrum Weihenstephan, Technische Universität München, Freising, Germany

ORCID logo http://orcid.org/0000-0003-3289-4590

Jan Maciej Zaucha

Roles Supervision, Writing – review & editing

Affiliation: Department of Haematology and Transplantation, Medical University of Gdansk, Gdansk, Pomorskie Voivodeship, Poland

ORCID logo http://orcid.org/0000-0002-0986-8936

Krzysztof Łukaszuk

Roles Supervision, Writing – review & editing

Affiliation: Faculty of Health Sciences with Institute of Maritime and Tropical Medicine, Medical University of Gdansk, Gdansk, Pomorskie Voivodeship, Poland

Marek Jóźwicki

Roles Supervision, Visualization

Affiliation: Department of Architecture and Design, Academy of Fine Arts, Gdansk, Pomorskie Voivodeship, Poland

Magdalena Płóciennik

Roles Formal analysis, Writing – original draft

Affiliation: FitnessFitback, Pomorskie Voivodeship, Poland

Paweł Cięszczyk

Roles Data curation, Methodology, Resources, Writing – review & editing

Affiliation: Department of Physical Education, Academy of Physical Education and Sport in Gdansk, Gdansk, Pomorskie Voivodeship, Poland

Introduction

By 1798, Luigi Galvani discovered two phenomena: muscle stimulation by extrinsic electricity and a genuine potential difference between the nerve and the muscle. These findings lead his successors to investigate the details of the electrical influence on nerve function in the context of muscle movement. By now, the scientific community has reached the molecular level of understanding the mechanisms involved and have already honed in on the genomic loci affecting athleticism. As a result, multiple single nucleotide polymorphisms (SNPs) have been implicated in affecting the aptitude for gymnastics. To move beyond simple SNP associations, genetic epistasis modeling may enhance the understanding of sports performance. Authors investigating genetic interactions typically rely only on genotype frequency odds ratios [1–3] or perform Genome-Wide Interaction Analyses (GWIA) employing tests visualized by pseudo-Manhattan plotting. So far, the matter of epistasis has been investigated for: (a) the Body Mass Index (BMI) [4]; (b) physical activity in mice [5]; (c) medical disorders in clinical studies [6]; in ischemic stroke susceptibility [7].

Variant interactions including synergy or redundancy have not yet been considered in the context of predicting athletic performance [8, 9]. Instead, the total genotype score (TGS) for distinguishing athletes has been calculated several times in different research projects [10, 11]. Unfortunately, TGS models do not consider interactions between polymorphisms, i.e., their synergy and redundancy [11]. The main strength of pure epistatic models is their potential for deciphering the genetic variation of predisposed athletes ab initio. Interestingly, ensemble-based classifiers [12], which are free of external attributes, have so far yielded better predictions than alternative approaches incorporating environmental effects into the model.

The genetic foundations of muscle performance are explored by mathematical modeling. While parametric techniques, such as logistic regression (LR) are limited in their ability to characterize the multivariate architecture of complex phenotypes, information theory provides a solution for quantifying the information gain between different statistical models of inference. The relative difference in Shannon entropy i.e. the Kullback-Leibler divergence (also known as information gain—IG) allows selecting the optimal approach for modeling the genetic effects on phenotype. Additionally, Multifactor Dimensionality Reduction (MDR), a non-parametric statistical technique enables detecting interactions between attributes of the model. In this work, we applied this method to detect epistasis in a set of candidate genes, Artistic gymnastics is one of many sport disciplines, which has not been extensively studied with regard to its genetic underpinnings. Notwithstanding the exact definition of the proportion of speed and strength to power output, gymnastics is definitely a highly polygenic anaerobic event, dependent on multiple, potentially interacting genetic variants.

The seven PEPs that were evaluated in this study include: (1) rs1815739, located within the ACTN3 gene is involved in muscle contractions [13]; (2) rs8192678, located within the PPARGC1A gene is responsible for the variability in power output; the substitution of glycine for serine at position 428 was reported to hinder performance in endurance activities [14]; (3) rs4253778, located within the PPARα gene appears to be associated with the hypertrophic effect due to its effects on the cardiac and skeletal muscle substrate utilization [15]; (4) rs6265, located within the BDNF-AS gene is highly correlated with learning and the development of memory-related hippocampal neurons; (5) rs5443, located within the GNB3 gene seems to be a candidate for explaining the variability in exercise phenotypes [16, 17]. Specifically, the proportion of the TT genotype is more pronounced in the top-level endurance athletes as compared with the sprinter group. Hence, G protein activity may affect the likelihood of becoming a highly-qualified endurance athlete [17]; (6) rs1076560, located within the DRD2 gene can predispose athletes to better performance in Australian Rules Football; it allows for specific talent identification and has been linked with motor coordination and learning [18]; (7) rs362584, located within the SNAP-25 gene was found to be associated with cognitive ability [19] and with the cognitive disorder [20]. Furthermore in 2015, Islamov et al. [21] have shown that SNAP-25 is synthesized in the motor nerve endings, and affects motor neurons of the spinal cord. The aforementioned PEPs were analyzed with regard to epistasis in the context of gymnastics and evaluated in terms of their ability to discriminate between athletes and non-athletic individuals.

Results

Quality control of SNPs called

The minor allele frequency (MAF) for every candidate SNP was no less than 16.5%, which was the lowest value for the case of rs4253788 (PPARα)–in the control group (Supplementary Material 1, S2 Table in S2 File). All of the seven genetic polymorphisms were in Hardy-Weinberg equilibrium (HWE; H0: χ2≤6.635(0.01; 1)).

Models adjustment according to genetic markers

All SNPs under consideration were coded according to the values of the odds ratios for heterozygote, homozygote of major allele and for homozygote of minor allele (oddMm, oddMM, and oddmm) extracted from contingency tables [22] (S3 File, p. 2). Data in Table 1 indicates the odds ratios obtained for different genetic models.

[Figure omitted. See PDF.]

Table 1. Model adjustment according to examined SNPs.

https://doi.org/10.1371/journal.pone.0237808.t001

Entropy analysis

Next, the statistical significance has been calculated for each polymorphism’s ability to distinguish between the case (athletes) and control (non-athletes) groups. The strongest effect observed for any single locus was for PPARGC1A. Its normalized information gain (IG) reached the value of 0.0065 bits (0.65%). It was the largest univariate factor reducing entropy with a borderline significance at p = 0.07 (at χ2 = 5.317). Table 2 presents IGs and p-values of all genetic markers in the performed analysis:

[Figure omitted. See PDF.]

Table 2. Information gain values of studied genetic attributes.

https://doi.org/10.1371/journal.pone.0237808.t002

Multifactor dimensionality reduction

Next, a genetic dendrogram has been constructed, using Rajski’s distance, Ward’s method and Lance and Williams recursive algorithm (S3 File, pp. 3–4). As a consequence, synergistic (red connections) and redundant effects have been determined (Fig 1). The analysis shows that polymorphisms are grouped into two clusters and two independent genetic pools of variants, namely: PPARα, PPARGC1A –GNB3 and BDNF, DRD2 –ACTN3 –SNAP-25.

[Figure omitted. See PDF.]

Fig 1. A gene-gene interaction dendrogram in sports gymnastics performancea.

aOrange line indicates weak positive interaction between clusters. Golden connections suggest the independence of PPARα, BDNF.

https://doi.org/10.1371/journal.pone.0237808.g001

Epistasis between pairs of SNPs was evaluated in terms of the interaction information (I) between SNPs A and B in the context of class C: I(A; B; C), with positive values corresponding to synergy while negative values indicating a redundancy (correlation) of the markers [23]. The only strong synergistic effects were found between ACTN3 –SNAP-25 and PPARGC1A –GNB3, represented by 0.0543 bits of interaction information (5.43%) and 0.0364 bits (3.64%), respectively. However, little evidence corroborates other possible two-way interactions. A positive moderation has been detected for twenty out of twenty-one combinations. The highest values regard PPARGC1A –SNAP-25 (0.0523 bits—5.23%), ACTN3 –PPARα (0.298 bits—2.98%) and GNB3 –BDNF (0.027 bits—2.70%). The only negative interaction was between SNAP-25 and PPARα; this pair of SNPs diminishes 0.0001 bits of information about sports gymnastics. The results presented above support the alternative hypothesis stipulating the existence of a synergistic effect (e.g. for ACTN3 and SNAP-25) in the set comprised of twenty-one possible two-way interactions between rs1815739, rs8192678, rs4253778, rs6265, rs5443, rs1076560, rs362584.

Next, a filtering technique (S3 File, Eq 8) has been applied to identify the best epistatic framework The optimal model has been obtained for the combination of ACTN3 –PPARGC1A –PPARα–SNAP-25. Its performance is summarized in Table 3.

[Figure omitted. See PDF.]

Table 3. Test set results obtained for the ACTN3 –PPARGC1A –PPARα–SNAP-25 epistatic model selected to maximize balanced accuracy in 10-fold cross validation.

https://doi.org/10.1371/journal.pone.0237808.t003

MDR analysis confirmed the statistical significance (p = 0.001) of the model by comparing the value of the sign test against 1000 random permutations of the data, assuming no association under the null hypothesis. The model achieved a balanced accuracy (weighting case and control samples so as to simulate an equal sample size in each group) of 0.712. The odds ratio of positivity within the gymnasts’ group relative to the controls is equal to 6.2. Interestingly, the p-value of the model estimated from the χ2-test achieved only borderline significance, confirming previous concerns about the reliability of the p-value obtained from the MDR analysis sign-test [24]. Nevertheless, the precision is above 40% and Cohen’s Kappa at 0.326 indicates a performance, which significantly surpasses naïve guessing. With regard to perfect precision and recall, the classifier is positioned in the middle of the achievable spectrum: F1-measure = 0.525. The training and whole data models are even more convincing (Supplementary Material 1, S5 Table, S6 Table in S2 File), since χ2 p–values retained significance after Bonferroni’s correction for multiple hypothesis testing. Nevertheless, we do not have definitive evidence that that the null hypothesis can be rejected.

Logistic regression analysis

For a simultaneous examination of the first and second order effects in the ACTN3 –PPARGC1A –PPARα–SNAP-25 interaction, logistic regression with backward variable selection has been adopted. Since this analysis yielded empty combinations, two-way interactions were considered first. Contrasts between genotype categories were expressed in terms of cross-partial derivatives. To ensure the interpretability of the results for unbalanced classes, we used weighted effect coding (WEC). Interestingly, none of the other known mathematical and statistical coding structures apart from WEC allows detecting pure genetic interaction (Supplementary Material 1, S1 Table in S2 File). In particular, such phenomenon has been confirmed between ACTN3 and SNAP-25, when setting the homogenous derived (alternative) allele category as the reference (Table 4):

[Figure omitted. See PDF.]

Table 4. The full ACTN3 –SNAP-25 model with the derived allele reference category.

https://doi.org/10.1371/journal.pone.0237808.t004

The baseline OR for being a highly qualified gymnast equals 0.24, when carrying the most common genotype. Maximal log-likelihood for the estimated model totalled -133.857 with χ2-score of 34.344 (df = 8) and p-value ≈ 0.000. Although the model explains genetic foundations for sub-elite versus elite gymnasts’ recognition in just 11% (pseudo R2 = 0.114), we accept the global alternative hypothesis–H1e, which states that at least one product term between PEPs is significantly different than zero. Considering the WEC data arrangement, the main effects of the model can be considered as non-significant being an order of magnitude less than the interaction weights, which are all below or equal 0.05*. Thus, individual beta weights (bi) for ACTN3 and SNAP-25 are ≈ 0 and obeying statistical parsimony, we reject the null hypothesis. Next, we performed logistic regression for rs1815739 and rs362584 without first-order effects. Typically, in WEC, weights of regression coefficients do not change when the reference category is switched. The same applies to maximal log-likelihood statistics. Hence, we present different models (grouped according to reference genotype category) of interactions between genotypes in Table 5:

[Figure omitted. See PDF.]

Table 5. The ACTN3 –SNAP-25 interaction models.

https://doi.org/10.1371/journal.pone.0237808.t005

In agreement with previous results, all interaction effects from the model for ACTN3 –SNAP-25, with the derived (minor allele) genotype set as the weighted reference category are significant. Moreover, G-G homogenous derived genotype, ancestral-derived and heterozygous (XX,GA) interaction genotypes also show considerable effects, at the edge of the p-value threshold for statistical significance. Maximal log-likelihood for the interaction model for the homogenous derived allele reference category has reached the value of -134.150. The χ2 statistic was equal to 33.758 (df = 4) and pseudo R2 = 0.112 giving a p-value < 0.00001. According to the model, the pure minor allele (XX,AA) genotype has the strongest negative influence. Thus, it determines the context for the other interactions. In our analysis, b1,1, b1,2, b2,1, b2,2 reached the p-value of 0.05 for the derived allele reference category (Table 5). The statistical significance was retained after applying Bonferroni’s correction for multiple tests (p-valueα/2 = 0.001). In the light of this fact, three-way and multi-way interactions have not been examined.

Particularly noteworthy is that the pure epistatic logistic regression model achieved much better performance as compared with the additive-only model. When removing all second-order derivatives, the maximal log-likelihood for the rs1815739 + rs362584 combination is -150.688 and becomes non-significant with a p-value of 0.409.

The results obtained from the MDR and LR analyses revealed a remarkable crosstalk between ACTN3 –SNAP-25 polymorphisms. Disappointingly, the bheterozygous,heterozygous and bancestral,ancestral coefficients are attributed with negative weights; presumably, in both cases a low ratio of gymnasts to sedentary individuals (5/49 and 6/70, respectively) cause these effects (Supplementary Material 1, S4 Table in S2 File). Nevertheless, homogenous minor allele (XX,AA) genotype hosts represent the lowest chance of classification to the gymnast group: 0.059. Taking this genotype as the reference, the modeled ACTN3 –SNAP-25 interaction effects allow rejecting the null hypothesis of no interaction.

Based on the training set, the classification performance for the interaction model without additive terms, with the XX–AA allele reference category and multiplicative entries arranged according to WEC achieved the area under the ROC curve (AUC-ROC) of 0.715 (95% CI: 0.647–0.782; Z-score = 38.917, p-value ≈ 0.000) with a standard error (Se) of AUC-ROC = 0.034. The cut-off point was selected by maximizing the Youden index = TPF-FPF and was equal to 0.379 (Fig 2). Although the achieved classification accuracy offers good specificity and is already satisfactory to aid gymnasts’ recognition, the Cohen’s Kappa statistic is fair (27.2%) and F1-measure totals 0.498.

[Figure omitted. See PDF.]

Fig 2. The area under the curve (AUC-ROC) and cut-off point for the epistatic rs1815739 * rs362584 model based on the training dataset.

https://doi.org/10.1371/journal.pone.0237808.g002

When applied to the test hold-out dataset (n = 36), our classifier has correctly classified four athletes and fifteen sedentary individuals, yielding an accuracy of 52.78%. This is unsatisfactory for the purpose of supporting decision-making in sub-elite or elite gymnasts’ identification. The observed AUC-ROC (0.715) and measure of Se AUC-ROC (0.034), despite being highly significant (p-value ≈ 0.000) has limited potential to confer these genetic variants as predictors for athlete’s discrimination in the light of the obtained Kappa statistics and F1-measure. Further studies comprising larger samples may assert the status of these variants as informative for the task of gymnasts’ identification. However, our results do not allow rejecting the null hypothesis.

Worth reporting are other insights shed by the LR and WEC data organization for the ACTN3 –PPARα, PPARGC1A –SNAP-25, PPARGC1A –GNB3, GNB3 –BDNF interactions. The contingency table for ACTN3 –PPARα and GNB3 –BDNF exposed empty cell or singular representatives in genotype categories. Consequently, data were not processed any further for these models. Fortunately, the same did not apply, when PPARGC1A –SNAP-25 and PPARGC1A –GNB3 were considered. Both pairs of SNPs were annotated with four statistically significant weights (p-value ≤ 0.05) for the same second-order product terms: PPARGC1A –SNAP-25: bGlyGly,GA (SerSer,GG reference (ref.) genotype: favorable), PPARGC1A –GNB3: bGlyGly,CT (SerSer,CC ref. group: favorable), PPARGC1A –SNAP-25: bGlyGly,GG (GlySer,GA ref. heterozygous), PPARGC1A –GNB3: bGlyGly,CC (GlySer,CT heterozygous reference group), PPARGC1A –SNAP-25: bGlySer,GG, bSerSer,GA (GlyGly,AA ref. disfavorable), PPARGC1A –GNB3: bGlySer,CC, bSerSer,CT (GlyGly,TT reference group: disfavorable). The maximal log-likelihood value was -129.97 and -139.52, respectively. Nevertheless, the first-order effects remain insignificant for all possible pairwise combinations of SNPs. Further non-trivial effects of cross-partial G-G interactions obtained from eighteen other coding schemes applied to LR are in S2 File.

Discussion

The biological and sport science perspective

The ultimate goal in sport is the athletic outcome, which correlates strongly with the level of physical fitness (with psychological effects playing a secondary role). An important theoretical aspect of predicting, which individuals are genetically predisposed to athleticism regards establishing which allele encoding schemes allow for the most faithful discrimination between athletically-gifted and ungifted individuals. Apart from fundamental, molecular types of genotype ordering, we evaluated nineteen classic (statistical and mathematical) notations to describe SNPs (list available in S2 File). On the basis of planned contrasts [25], taking the trend and non-trend approaches [25], all possible ways of raw genetic data encoding have been processed to detect epistatic interactions. So far, there have been no studies in which genetic epistasis has been investigated using so many different encoding schemes. Most authors do not recognize this possibility and are reporting G-G interactions by means of LR but without considering cross-partial derivatives and using unspecified coding schemes [26, 27]. Nonetheless, a growing body of literature has discussed ways of combining non-parametric and parametric techniques with the goal of examining epistasis. A comprehensive attempt at investigating molecular interactions has been performed by Manuguerra et al. [28]. Similar to our research, these authors have presented, apart from a measure of CVC and p-values, a prediction error percentage of low and-high risk instances for given G-G models and odds ratio reports to determine the probability of false-positive predictions. Besides, it is worth noting that Wu et al. [29] have performed an analysis considering relationships between genotypes internally but also with environmental variables. Unfortunately, no information has been given on the categorical coding scheme. Only a general linear assignment was presented, which enabled us to determine the class that was used as the reference. Also, Dasgupta et al. [30], inform on gene–environmental interaction odds ratios based on MLR without considering regression coefficients. Nevertheless the essential result summarizing protective and risk-conferring alleles has been delineated. Bottema et al. applied LR to confirm interactions identified by means of MDR. Of the epistatic interactions they identified, MDR indicated that most interactions were synergistic [31]. However, the negative gene–gene interactions in the logistic regression of two-locus models suggest that polymorphisms of these genes counteract the effect of one another.

In this study we provide multiple lines of evidence indicating an interaction between ACTN3 and SNAP-25. To the best of our knowledge, no previous study has reported such a relationship. Furthermore, notwithstanding the context of gymnast recognition, no data suggesting any kind of interaction between ACTN3 and SNAP-25 is available in String-db [32]. However, based on the outcome of the multidimensional stimulation therapy—MST intervention, neurophysiological studies have indicated the possibility of epistatic interactions between APOE and SNAP-25 [33]. Interestingly, the interaction between ACTN3 and APOE has been studied to explain the potential for exceptional longevity [34]. So far, with regard to sports science, an epistasis of ACE ID and ACTN3 R577X polymorphisms has been determined, e.g. in swimmers–sprint and endurance performance [2].

In order to detect epistatic interactions Wei et al. [4] applied MLR and demonstrated two-way G-G effects affecting the body mass index (BMI) based on a genome-wide analysis. Specifically, interactions between the 19 shared epistatic genes (defined as these, which represent significant SNP interactions across cohorts) and those involving BMI candidate loci were tested across five populations (p-value < 5.0E-08). Ultimately, eight replicated SNP pairs were found in at least one cohort (p-value < 0.05) and no beta coefficients were detailed.

An interaction can also be recognized as product term, e.g. second-order parameter in logistic model under the assumption of linear coding. This technique has been used by Lee et al. [35] for testing the interaction between EOT-2 and CCR3 genes. The authors found that an EOTAXIN-2 gene variant: EOT-2+304C>A (29L>I), was significantly associated with blood eosinophilia (p = 0.0087) by the effect of CCR3 = -0.68. Nevertheless, no information was presented on logistic regression main effects. Potentially, an analysis of first-order parameters in the LR model may be essential to verify pseudo R2 performance. In comparison all marginal weights of the full ACTN3 –SNAP-25 model are insignificant and the benefit from applying the additive–multiplicative paradigm to gymnasts recognition is just 2‰. Likewise, the subject of interaction has been studied for the rs12722 and the rs13946 in COL5A1 gene to assess a risk of the anterior cruciate ligament rupture in soccer players and controls [36]. Unfortunately, with regard to sportsman diagnosis or prognosis no details have been given on classification accuracy.

The ACTN3 –SNAP-25 interaction allows explaining 11% of the variance between high-level sports gymnasts. Bearing in mind that genetic factors typically explain between 20% - 80% variation in a wide variety of traits relevant to athletic performance [37], the G-G epistasis detailed in this paper should not be neglected in future investigations.

Methodological aspects

Several details of our analysis deserve particular attention. Firstly, considering the multiplicative–over-dominant scheme of epistasis between ACTN3 and SNAP-25, the theoretically desirable ancestral–ancestral (bancestral,ancestral) or heterozygous–heterozygous (bheterozygous,heterozygous) genotype carries a negative value. However, assuming disordinal interactions, there may be a region of non-significance [38], wherein there is a range of values for which no epistatic effect occurs. Secondly, possible signs change might occur for non-linear models even in the absence of an interaction [39]. These exist rational explanations for our results concerning bheterozygous,heterozygous and bancestral,ancestral. The third aspect concerns the data distribution. There were very few instances of gymnasts, who carried two heterozygous or dominant alleles for ACTN3 and SNAP-25. An additional corroboration of our results is the fact that the gene * gene interaction at the rs1815739 and rs362584 loci was detected by means of both: non-parametric and parametric tests. Here, after correction for multiple testing, statistical significance was far below the restrictive threshold. Finally, in terms of probability calculus, an additive only model: ACTN3 + SNAP-25 is not significant. Consequently, our results have interesting implications, which explain the underlying molecular details coordinating the neuromuscular system, which has been first studied by Luigi Galvani in the 18th century. Finally, we would like to stress that further studies concerning the ACTN3 * SNAP-25 interactions should be conducted while considering two other levels of epistasis (suppressive, co-suppressive) [40].

The gymnasts identification context

Despite significant results corroborating the identified genetic interaction, the resultant model for discriminating between athletes and non-athletes does not yet allow for making fully reliable predictions (Fig 2). In terms of prognosis, even a single genotype of a genetic polymorphism may be introduced as a biomarker of prevalence risk, like has been done for ischemic stroke [7]. Similarly, in our opinion, the PPARGC1A gene (Table 2) might be considered for diagnostic purposes. However, its usefulness in the context of gymnasts recognition has not been so far confirmed. Finally, we also observed a nominal statistical G*G partial interaction of PPARGC1A –SNAP-25 and PPARGC1A –GNB3 based on the gymnast status, which is interesting in the context of the studies that have associated these loci with effects relating to sport [14, 16, 17, 19–21]. Lastly, it should be acknowledged that apart from the PEPs, which we considered, interactions between other genetic loci could occur. However, expanding the analysis to include all tag SNPs (tSNPs) does not guarantee robustness for stochastic models in the aspect of predicting a predisposition to become a professional gymnast. Of note, till 2016 only twelve genetic markers have shown a positive correlation with the athlete status it at least three or more studies [41].

Conclusions

Our analysis of seven PEPs (ACTN3, PPARGC1A, PPARα, BDNF-AS, DRD2, GNB3, SNAP-2), allows us to state with 93% confidence that the rs8192678 provides as much as 0.0065 bit of information on sports gymnastics. The molecular dendrogram of gymnastics aptitude indicated the strongest connection between rs1815739 and rs362584: 5.43% with a significant threshold of ≈ 0.000, when the homogenous derived allele category is set as the reference group. According to the findings, the best MDR epistatic model of sports gymnastics comprises of: ACTN3 –PPARGC1A –PPARα–SNAP-25 (the cross validation consistency equals 100%). Manifestly, when considering all pairwise combinations between ACTN3, PPARGC1A, PPARα, BDNF-AS, DRD2, GNB3, SNAP-25, the results confirm that only the second order terms of sports gymnastics epistatic models are non-zero. Lastly, out of the set of ACTN3, PPARGC1A, PPARα, BDNF-AS, DRD2, GNB3 and SNAP-25 genes, the most informative epistatic classifier–rs1815739 x rs362584 is statistically significant in the context of sportsman recognition.

Materials and methods

Ethic committee

The study was approved by The Pomeranian Medical University Ethics Committee, Poland (Approval number 09/KB/IV/2011). Research procedures were run according to the World Medical Association Declaration of Helsinki. An informed consent form was completed by each participant or obtained from a parent / legal guardian (in the case of minors) in accordance with current Polish, Italian and Lithuanian law.

Participants

A Seventy three sportsman and two hundred forty five sedentary, non-active individuals met the inclusion criteria and comprised a group for this study. They had no records of metabolic, cardiovascular diseases or musculoskeletal injuries. The subjects were non-smokers and did not take any medications. The cohort participants volunteered in Poland, Italy, Lithuania between 2012 and 2017. All participants were unrelated European men (59.4%) or women (40.6%), and all of European descent (as self-reported) for ≥ 3 generations. Therefore, the influence of an ethnically-induced genetic skew has been minimized and the potential population stratification issues have been controlled (Study protocol, p. 4, 5 in S1 File). The study sample included 34 females and 39 males in two homogenous athletes groups–elite (25.2 ± 2.8 years old): ngymnasts (1,1) = 18 (24.7%), who had competed at an international level (European or World Championships or Olympic Games) and sub-elite–national-level athletes (19.4 ± 3.5 years old): ngymnasts (1,2) = 55 (75.3%), who performed sports gymnastics at a national level only. Contestants were classified according to the highest-level contest they had appeared in. The gymnasts were only included if they had never been tested positive by an anti-doping agency. A control group of healthy individuals ncontrols = 245; 150 males and 95 females; 22.6 ± 2.5 years old was also selected from the Polish, Italian and Lithuanian population (college students) with no background in the sport.

Controls were matched to gymnasts in ca. 1:4 ratio; adjustment consideration has been specified in the Study protocol (S1 File).

Methods, aims and hypotheses

In the paper, a quantitative approach to analyses has been conducted. The methods of observation and diagnostic survey were used. To gather the molecular data, PCR and RT-PCR techniques have been applied.

The goals of the research were: (a) to measure the magnitude of informative entropy of sport PEPs in artistic gymnastics with subsequent analysis of synergistic effects or redundancy between genetic variants; (b) to determine marginal effects and cross-partial derivatives at the level of 2-way gene-gene interactions; and (c) to investigate quality measures of MDR and logistic regression epistatic models for athletes recognition.

The aims implicate the following questions: (a) How much information will be gained on artistic gymnastics after quantifying Shannon entropy of a single genetic variant? (b) Does at least one two-attribute synergistic or redundant effect exist between sport performance enhancing polymorphisms? (c) Will the best MDR epistatic model of sports gymnastics achieve an outcome greater than 55% in cross validation consistency test? (d) For which combination of gene-gene models are the first and second order terms different than zero? (e) Are genetic classifiers statistically significant in the context of sportsman recognition? These questions concern six alternative hypotheses H1:

(a) H (Smax)<1; (b) ⋁IG(A;B;C)∈IG(A;B;C)I(A; B; C)≠0; (c) CVCmax>55%; (d) and; (e) when two SNPs are investigated in 2-way interaction model; (f) AUCi>0,7 for i = 1,…,m; i-th Kappa statistic > 0.6,

where:

H (Smax) is the maximal value of Shannon entropy in the set of genetic polymorphisms j = 1,…,k, IG is the information gain; I(A;B;C) is the vector of multiple mutual information results from all possible combinations in the analysis; CVCmax−the highest value obtained in cross-validation consistency (count) for epistatic models; bi−SNP marginal effect; bii is a 2-way G-G interaction product term; AUCi−area under the curve for model i; ⋁IG(A;B;C)∈IG(A;B;C) is the existential quantifier.

Biological sample collection and DNA extraction

The buccal cells donated by the participants were acquired using the Oragene–DNA isolation kit (DNA Genotek, Kanata, ON, Canada). The subjects abstained from drinking, and eating for 2 hours prior to saliva collection. Each participant was asked to perform a 2-min mouth rinse with water 30 min before retrieving the DNA sample. Samples were collected by passive drooling in sterile 50 ml tubes. Tubes were filled up to 4 ml, then vigorously mixed and transported to a laboratory for further processing. All samples were stored in the same conditions at −25°C until subsequent steps were performed.

DNA was extracted according to the producer’s protocol. Briefly, the DNA material located in the Oragene tubes was incubated at 50°C overnight. Afterward, the probes were opened and divided into four equal parts. Each one was treated with 40 μl of buffer solution supplied by the manufacturer. After a period of 10 minutes of ice incubation, centrifugation for 3 minutes at 13,000 rpm was performed. The resulting supernatant (DNA) was assessed for both purity and integrity by using spectrometric and electrophoretic methods, respectively.

Determination of genotypes

DNA isolation and genotyping were performed in the molecular laboratory of Gdansk University of Physical Education and Sport, Poland. The genotyping error was assessed as 1%, while the call rate was above 95%. Details on PEPs genotyping can be verified in S1 File. Briefly, six gene variants (ACTN3 –rs1815739, PPARGC1A –rs8192678, PPARα–rs4253778, BDNF-AS–rs6265, GNB3 –rs5443, DRD2– rs1076560) were assessed by PCR. In accordance with [2], amplification was performed in a total volume of 10 μl PCR reaction mix containing 1.5 mM MgCl2, 0.75 nM of each deoxynucleoside triphosphate–dNTP (Novazym, Poland), 4 pM of specific primer (Genomed, Poland) in TE (pH = 8.0; Thermo Fisher Scientific), 0.5 U DNA recombinant Taq polymerase in buffer (pH = 8.0; Sigma, Germany), 1x PCR buffer (pH = 8.7; Sigma, Germany) and 1 μl (30–50 ng) of template DNA (isolate). The thermal-time PCR amplification cycling profile conditions consisted of 10 min of preincubation at 95°C (activation of the Taq DNA polymerase), followed by 40 cycles of denaturation at 95°C for 15 s, and primer annealing, and extension for 1 min at 60°C, followed by a final elongation cycle at 72°C for 3 min. The PCR fragments were subsequently digested with the appropriate restriction enzyme. The PCR products were separated by electrophoresis at 80mV on a 2% agarose gel, stained withn DMS in DMSO ethidium bromide (250ng / ml), and visualized in UV light. The SNAP-25 (rs362584) was genotyped in two replicates with TaqMan fluorescent oligonucleotide probes. Likewise, following [42], a BioRad CFX96 Touch™ RT-PCR Detection System in tandem with the Bio-Rad CFX Manager Software was used to detect the fluorescent signals and to produce a graphical representation which allowed for A / G allelic discrimination. Freshly purified / sterile water was used as a negative control for PCR.

Statistical analyses

From 318 observations, 36 (roughly 10%) of instances were included into the test set (hold-out dataset). Minor allele frequencies were computed for each of the seven SNPs and Hardy-Weinberg equilibrium was tested. In the standard–linear approach, genotypes were coded as ‘1’: potentially disfavorable for strength / power sports activities, ‘2’: heterozygotes, or ‘3’ (Supplementary Material 1, p. 8 in S2 File). Next, the most commonly used six subject-level gene models including: recessive, multiplicative, additive / harmonic, dominant, and over-dominant models [22] were computed to select the best one to the given data distribution of each SNP. After quality control of alleles and model selection, the information gain (IG) of every SNP was computed with standard coding and with the adjustment for the optimal genetic model. Next, the Multifactor Dimensionality Reduction (MDR) and logistic regression algorithms were applied.

All statistical analyses were run in MS Excel on a standard PC and in MDR program available on the Internet (https://www.multifactordimensionalityreduction.org/). The threshold for statistical significance was set to p-value ≤ 0.05, with two-sided Bonferroni correction for multiple comparisons. Formulae used for data processing have been compiled in (Theoretical background–data analysis in S3 File), for further inspection.

Supporting information

[Figure omitted. See PDF.]

S1 File. Study protocol.

This protocol has been provided by the authors to give readers additional information about the research work.

https://doi.org/10.1371/journal.pone.0237808.s001

(PDF)

S2 File. Supplementary material 1.

This work contains all supplemental text, figures, and tables.

https://doi.org/10.1371/journal.pone.0237808.s002

(PDF)

S3 File. Supplementary material 2.

Theoretical background–data analysis.

https://doi.org/10.1371/journal.pone.0237808.s003

(PDF)

S4 File. Data input.

https://doi.org/10.1371/journal.pone.0237808.s004

(XLSX)

Citation: Płóciennik ŁA, Zaucha J, Zaucha JM, Łukaszuk K, Jóźwicki M, Płóciennik M, et al. (2020) Detection of epistasis between ACTN3 and SNAP-25 with an insight towards gymnastic aptitude identification. PLoS ONE 15(8): e0237808. https://doi.org/10.1371/journal.pone.0237808

References

1. Eynon N, Alves AJ, Sagiv M, Yamin C, Sagiv M, Meckel Y. Interaction between SNPs in the NRF2 gene and elite endurance performance. Physiol Genomics. 2010; 41(1): 78–81. pmid:20028934

2. Grenda A, Leońska-Duniec A, Kaczmarczyk M, Ficek K, Król P, Cięszczyk P. Interaction Between ACE I/D and ACTN3 R557X Polymorphisms in Polish Competitive Swimmers. J Hum Kinet. 2014; 42: 127–36. pmid:25414746

3. O’Connell K, Knight H, Ficek K, Leonska-Duniec A, Maciejewska-Karlowska A, Sawczuk M, et al. Interactions between collagen gene variants and risk of anterior cruciate ligament rupture. Int J Sports Sci. 2015; 15(4): 341–350.

4. Wei W-H, Hemani G, Gyenesei A, Vitart V, Navarro P1, Hayward C, et al. Genome-wide analysis of epistasis in body mass index using multiple human populations. Eur J Hum Genet. 2012; 20(8): 857–862. pmid:22333899

5. Leamy LJ, Pomp D, Lightfoot JT. An epistatic genetic basis for physical activity traits in mice. J Hered. 2008; 99(6): 639–46. pmid:18534999

6. Hsu P-C, Yang U-C, Shih K-H, Liu C-M, Liu Y-L, Hwu H-G. A protein interaction based model for schizophrenia study. BMC Bioinform. 2008; 9 (Suppl 12): S23: 1–9.

7. Liu D., Liu L., Song Z., Liu J., & Hou D. Genetic Variations of Oxidative Stress Related Genes ALOX5, ALOX5AP and MPO Modulate Ischemic Stroke Susceptibility Through Main Effects and Epistatic Interactions in a Chinese Population. Cell Physiol Biochem. 2017; 43, 1588–1602. pmid:29041000

8. Ruiz JR, Arteta D, Buxens A, Artieda M, Gómez-Gallego F, Santiago C, et al. Can we identify a power-oriented polygenic profile? J Appl Physiol. 2010 a; 108(3): 561–566. pmid:20044471

9. Hughes DC, Day SH, Ahmetov II, Williams AG. Genetic of strength and power: polygenic profile similarity limits skeletal muscle performance. J Sports Sci. 2011; 29(13): 1425–34. pmid:21867446

10. Williams AG, Folland JP. Similarity of polygenic profiles limits the potential for elite human physical performance. J Physiol. 2008; 586(1): 113–21. pmid:17901117

11. Grealy R, Herruer J, Smith CLE, Hiller D, Haseler LJ, Griffiths LR. Evaluation of a 7-Gene Genetic Profile for Athletic Endurance Phenotype in Ironman Championship Triathletes. PloS One. 2015; 10(12): e145171: 1–20.

12. Sun YV. Multigenic Modeling of Complex Disease by Random Forest. In: Moore JH, Dunlap JC, editors. Computational Methods for Genetics of Complex Traits. San Diego, CA: Elsevier; 2010. pp. 73–97.

13. Ben-Zaken S, Eliakim A, Nemet D, Meckel Y. Genetic Variability Among Power Athletes: The Stronger vs. the Faster. J Strength Cond Res. 2019; 33(6): 1505–1511. pmid:26840443

14. Tharabenjasin P, Pabalan N, Jarjanazi H. Association of PPARGC1A Gly428Ser (rs8192678) polymorphism with potential for athletic ability and sports performance: A meta-analysis. PLoS ONE. 2019; 14(1): e0200967: 1–18. pmid:30625151

15. Ahmetov II, Fedotovskaya ON. Sports genomics: Current state of knowledge and future directions. CMEP. 2012; 1(1): 1–24.

16. Ruiz J, Eynon N, Meckel Y, Fiuza-Luces C, Santiago C, Gόmez-Gallego F, et al. GNB3 C825T Polymorphism and Elite Athletic Status: A Replication Study with Two Ethnic Groups. Int J Sports Med. 2010 b; 32:151–153. pmid:21110287

17. Eynon N, Oliveira J, Meckel Y, Sagiv M, Yamin C, Sagiv M, et al. The Guanine Nucleotide Binding Protein Beta Polypeptide 3 Gene C825T Polymorphism Is Associated With Elite Endurance Athletes. Exp Physiol. 2009; 94: 344–349. pmid:19139061

18. Jacob Y, Chivers P, Anderton RS. Genetic predictors of match performance in sub-elite Australian football players: A pilot study. J Exerc Sci Fit. 2019; 17(2): 41–46. pmid:30740132

19. Gosso MF, De Geus EJC, van Belzen MJ, Polderman TJC, Heutink P, Boomsma DI, et al. The SNAP-25 Gene Is Associated With Cognitive Ability: Evidence From a Family-Based Study In Two Independent Dutch Cohorts. Mol Psychiatr. 2006; 11(9): 878–886.

20. Liu Y-S, Dai X, Wu W, Yuan F-F, Gu X, Chen J-G, et al. The Association of SNAP25 Gene Polymorphisms in Attention Deficit / Hyperactivity Disorder: A Systematic Review and Meta-Analysis. Mol Neurobiol. 2017; 54(3): 2189–2200. pmid:26941099

21. Islamov RR, Samigullin DV, Rizvanov AA, Bondarenko NI, Nikolskiy EE. Synaptosome-associated protein 25 (SNAP25) synthesis in terminal buttons of mouse motor neuron. Dokl Biochem Biophys. 2015; 464: 272–274. pmid:26518545

22. Horita N, Kaneko T. Genetic model selection for a case-control study and a meta-analysis. Meta Gene. 2015; 5: 1–8. pmid:26042205

23. Hu T, Chen Y, Kiralis JW, Collins RL, Wejse Ch, Sirugo G, et al. An information-gain approach to detecting three-way epistatic interactions in genetic association studies. J Am Med Inform Assoc. 2013; 20: 630–636. pmid:23396514

24. Milne RL, Fagerholm R, Nevanlinna H, Benítez J. The importance of replication in gene–gene interaction studies: multifactor dimensionality reduction applied to a two-stage breast cancer case-control study. Carcinogenesis. 2008; 29(6): 1215–1218. pmid:18482998

25. Sundström S. Coding in Multiple Regression Analysis: A Review of Popular Coding Techniques. U.U.D.M. Project Report. Uppsala University. 2010. Available from: http://uu.diva-portal.org/smash/get/diva2:325460/FULLTEXT01.pdf

26. Lekman M, Hössjer O, Andrews P, Källberg H, Uvehag D, Charney D, et al. The genetic interacting landscape of 63 candidate genes in Major Depressive Disorder: an explorative study. BioData Min. 2014; 7(19): 1–18.

27. Jyothi KU, Reddy BM. Gene–gene and gene–environment interactions in the etiology of type 2 diabetes mellitus in the population of Hyderabad, India. Meta Gene. 2015; 5: 9–20. pmid:26042206

28. Manuguerra M, Matullo G, Veglia F, Autrup H, Dunning AM, Garte S, et al. Multifactor dimensionality reduction applied to a large prospective investigation on gene-gene and gene-environment interactions. Carcinogenesis. 2007; 28(2): 414–22. pmid:16956909

29. Wu Y, Zhang L, Liu L, Zhang Y, Zhao Z, Liu X, et al. A multifactor dimensionality reduction-logistic regression model of gene polymorphisms and an environmental interaction analysis in cancer research. Asian Pac J Cancer Prev. 2011; 12(11): 2887–2892. pmid:22393959

30. Dasgupta S, Reddy BM. The role of epistasis in the etiology of Polycystic Ovary Syndrome among Indian women: SNP-SNP and SNP-environment interactions. Ann Hum Genet. 2013; 77(4): 288–298. pmid:23550965

31. Bottema RW, Postma DS, Reijmerink NE, Thijs C, Stelma FF, Smit HA, et al. Interaction of T-cell and antigen presenting cell co-stimulatory genes in childhood IgE. Eur Respir J. 2010; 35(1): 54–63. pmid:19574333

32. Szklarczyk D, Gable AL, Lyon D, Junge A, Wyder S, Huerta-Cepas J, et al. STRING v11: protein–protein association network swith increased coverage, supporting functional discovery in genome-wide experimental datasets. Nucleic Acids Res. 2019; 47 (D1):D607–D613: pmid:30476243

33. Guerini FR, Farina E, Costa AS, Baglio F, Saibene FL, Margaritella N. et al. ApoE and SNAP-25 Polymorphisms Predict the Outcome of Multidimensional Stimulation Therapy Rehabilitation in Alzheimer’s Disease. Neurorehab Neural Re. 2016; 30(9): 883–893.

34. Fuku N, Díaz-Peña R, Arai Y, Abe Y, Zempo H, Naito H. et al. Epistasis, physical capacity-related genes and exceptional longevity: FNDC5 gene interactions with candidate genes FOXOA3 and APOE. BMC Genomics. 2017; 18(Suppl 8): 803: 126–131. pmid:29143599

35. Lee J-H, Jang A-S, Park S-W, Kim D-J, Park C-S. Gene-Gene Interaction Between CCR3 and Eotaxin Genes: The Relationship With Blood Eosinophilia in Asthma. Allergy Asthma Immunol Res. 2014; 6(1); 55–60. pmid:24404394

36. Lulińska-Kuklik E, Rahim M, Domańska-Senderowska D, Ficek K, Michałowska-Sawczyn M, Moska W, et al. Interactions between COL5A1 Gene and Risk of the Anterior Cruciate Ligament Rupture. J Hum Kinet. 2018; 62: 65–71. pmid:29922378

37. MacArthur DG, North KN. ACTN3: A Genetic Influence on Muscle Function and Athletic Performance. Exerc Sport Sci Rev. 2007; 35(1): 30–34. pmid:17211191

38. Jakulin A. Attribute Interactions in Machine Learning. M.Sc. Thesis. Second Edition. University of Ljubljana. 2003. p. 37. Available from: http://stat.columbia.edu/~jakulin/Int/interactions_full.pdf

39. Ch Ai, Norton EC. Interaction terms in logit and probit models. Econ Lett. 2003; 80(1): 123–129.

40. Kim J, Yum S, Kang C, Kang S-J. Gene-gene interactions in gastrointestinal cancer susceptibility. Oncotarget. 2016; 7(41): 67612–67625. pmid:27588484

41. Ahmetov II, Egorova ES, Gabddrakhmanova LJ, Fedotovkaya ON. Genes and Athletic Performance: An Update. Med Sport Sci. 2016; 61: 41–54. pmid:27287076

42. Leońska-Duniec A, Jastrzębski Z, Jażdżewska A, Moska W, Lulińska-Kuklik E, Sawczuk M, et al. Individual Responsiveness to Exercise-Induced Fat Loss and Improvement of Metabolic Profile in Young Women is Associated with Polymorphisms of Adrenergic Receptor Genes. J Sport Sci Med. 2018; 17(1): 134–144.