Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Affected pathways and transcriptional regulators in gene expression response to an ultra-marathon trail: Global and independent activity approaches

  • Maria Maqueda ,

    maria.maqueda@upc.edu

    Affiliations Department of ESAII, Center for Biomedical Engineering Research, Universitat Politècnica de Catalunya, Barcelona, Catalonia, Spain, CIBER de Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN), Barcelona, Catalonia, Spain

  • Emma Roca,

    Affiliations Summit 2014 S.L., Centelles, Barcelona, Catalonia Spain, Department of Electronic Engineering, Center for Biomedical Engineering Research, Universitat Politècnica de Catalunya, Barcelona, Catalonia, Spain

  • Daniel Brotons,

    Affiliation Catalan Sports Council, Barcelona, Catalonia, Spain

  • Jose Manuel Soria,

    Affiliation Unit of Genomics of Complex Diseases, Institut de Recerca de l'Hospital de la Santa Creu i Sant Pau, Barcelona, Catalonia, Spain

  • Alexandre Perera

    Affiliations Department of ESAII, Center for Biomedical Engineering Research, Universitat Politècnica de Catalunya, Barcelona, Catalonia, Spain, CIBER de Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN), Barcelona, Catalonia, Spain

Affected pathways and transcriptional regulators in gene expression response to an ultra-marathon trail: Global and independent activity approaches

  • Maria Maqueda, 
  • Emma Roca, 
  • Daniel Brotons, 
  • Jose Manuel Soria, 
  • Alexandre Perera
PLOS
x

Abstract

Gene expression (GE) analyses on blood samples from marathon and half-marathon runners have reported significant impacts on the immune and inflammatory systems. An ultra-marathon trail (UMT) represents a greater effort due to its more testing conditions. For the first time, we report the genome-wide GE profiling in a group of 16 runners participating in an 82 km UMT competition. We quantified their differential GE profile before and after the race using HuGene2.0st microarrays (Affymetrix Inc., California, US). The results obtained were decomposed by means of an independent component analysis (ICA) targeting independent expression modes. We observed significant differences in the expression levels of 5,084 protein coding genes resulting in an overrepresentation of 14% of the human biological pathways from the Kyoto Encyclopedia of Genes and Genomes database. These were mainly clustered on terms related with protein synthesis repression, altered immune system and infectious diseases related mechanisms. In a second analysis, 27 out of the 196 transcriptional regulators (TRs) included in the Open Regulatory Annotation database were overrepresented. Among these TRs, we identified transcription factors from the hypoxia-inducible factors (HIF) family EPAS1 (p< 0.01) and HIF1A (p<0.001), and others jointly described in the gluconeogenesis program such as HNF4 (p< 0.001), EGR1 (p<0.001), CEBPA (p< 0.001) and a highly specific TR, YY1 (p<0.01). The five independent components, obtained from ICA, further revealed a down-regulation of 10 genes distributed in the complex I, III and V from the electron transport chain. This mitochondrial activity reduction is compatible with HIF-1 system activation. The vascular endothelial growth factor (VEGF) pathway, known to be regulated by HIF, also emerged (p<0.05). Additionally, and related to the brain rewarding circuit, the endocannabinoid signalling pathway was overrepresented (p<0.05).

Introduction

Previous research has identified mechanisms triggered with the practice of moderate exercise that yield beneficial effects on health, specially on cardiovascular disease [1]. These effects may be explained due to the adaptation of many organs to cope with required musculoskeletal performance [2]. However, health benefits for the case of extreme endurance exercise remain unclear [3,4].

Ultra-marathon trails (UMTs) could be considered as extreme endurance exercise since their running events should be longer than the traditional marathon (42.195 km). Typically, they are run through a mountainous terrain with a considerable accumulated altitude change. Due to their high physical and psychological demand, UMTs are identified as an ideal sport for investigating a wide range of physiological responses [5]. These competitions show a growing popularity as indicated by the, approximately, sevenfold increase in the number of finishers of 100km worldwide ultra-marathons between 1998 and 2011 [6]. In parallel, the amount of scientific contributions focusing on UMT interventions has risen. They cover a varied range of perspectives: some authors detected reactive oxygen species (ROS) promotion, oxidative stress and inflammation in runners (n = 46) participating in a 330km UMT from capillary blood sample using micro-invasive analytic methods [7]; others evidenced a respiratory muscle strength reduction in inspiratory muscles when running a 110km UMT (n = 22) [8] or even, the adversely impact in the cognitive performance after a 168km UMT race (n = 17) [9].

To the best of our knowledge, there are no studies that approach UMT runners’ genome-wide gene expression (GE) response. This methodology has been used in other type of exercise-related interventions such as a single bout of 4-hour stationary cycling (n = 5) [10] or after a specific running endurance training (n = 13) [11]. On the other hand, GE on particular sets of genes has been assessed in shorter distances as marathon races when studying the response of specific interleukins (n = 16) [12] or in toll-like receptors (n = 47) [13].

A better understanding of the immune and inflammatory response has been the main motivation with regard to peripheral blood sample experiments. The link between exercise and immune system has long been studied tracing the beginning back to 1893 when an exercise-induced leukocytosis was described [14]. Prior studies suggest that moderate exercise negatively correlates with upper respiratory tract infections (URTI) incidence among other positive clinical implications [15]. However, this may not be the case in marathon or similar events where the opposite effect is detected [16]. A mechanistic explanation of an increased URTI risk in marathon runners (n = 16) is proposed elsewhere [17]. This study is based on the ratio imbalance of GE values from genes related to T-helper 1 (Th1) and Th2 cells. Likewise, other authors summarized the exercise impact on the GE of common inflammatory markers in a diverse range of exercise disciplines, intensity and duration [18].

In other experiments, the use of skeletal muscle biopsy samples is driven by the understanding of the adaptation of the human skeletal muscle to exercise [19]. In this context, the role of the hypoxia-inducible factor family (HIF), as reviewed in [20], is of great interest since its target genes include the vascular endothelial growth factor (VEGF), which related signalling pathway is one of the events driving the vascular system remodelling known to occur with dynamic exercise [2].

In this study, we obtained the genome-wide GE profiling in a group of runners (n = 16) participating in an 82km UMT race. We report the biological pathways and transcriptional regulators enriched by the list of differentially expressed genes as a result of the UMT intervention. For doing so, we addressed the genetic response from a global perspective and from an independent activity approach after implementing a statistical method capable of extracting independent sources of information.

Materials and methods

Ethical approval

All procedures involved in this study conformed to the Declaration of Helsinki. Ethical approval was granted by the Ethics Committee of the Catalan Sports Council from Government of Catalonia (Approval number 0099S/2046/2013). Written informed consent was obtained from all individuals participating in the study.

Experimental design

We approached and recruited 18 healthy runners who accepted to voluntarily participate in the study. They were athletes with prior experience in UMTs and presented no muscle injuries in the previous six months. One of the subjects dropout so finally, a total of 17 runners participated in the study (12 males aged 38.2 ± 4.3 years, 5 females aged 35.6 ± 2.2 years). All individuals were of Western European descent. Table 1 shows basic anthropometrical measures and weekly training hours per participant. The experiment was conducted in June 2012 at the “Cavalls del Vent” UMT located in the Catalan Pyrenees (Spain). This was a circular 82 km route starting at 760m above sea level and achieving a maximum altitude of 2,520m. The total accumulated altitude change was 12,180 m. In previous edition (2011), male and female winners needed 8.9 hours and 11.6 hours respectively to complete the race while last finishers took approximately 22 hours (no gender differentiation) to cover the total distance.

thumbnail
Table 1. Participants in the study–age, basic anthropometric and training regime.

https://doi.org/10.1371/journal.pone.0180322.t001

Blood samples, RNA extraction and microarray expression data

Venous blood samples were drawn, at rest in a sitting position, from the antecubital vein and collected into PAXgene Blood RNA Tubes according to the manufacturer’s protocol (PreAnalytiX GmbH/QIAGEN, Switzerland/US). Samples were obtained from each subject prior to and immediately after the UMT, with the exception of five participants (ids 1, 9, 13, 15 and 16 as shown in Table 2) from whom only pre-race samples were available. A total of 29 samples, 17 of them corresponding to pre-race and 12 to post-race, were stored at -80°C until assayed in the Hospital de la Santa Creu i Sant Pau (Barcelona, Spain). Samples were tagged with an identifier followed by PRE or POST referring to pre- or post-race sample. Total RNA was isolated using the PAXgene Blood RNA kit (PreAnalytiX GmbH/QIAGEN, Switzerland/US). The concentration of the extracted RNA was measured spectrophotometrically (Nanodrop 1000/ Thermo Fisher Scientific, Wilmington, US). GeneChip WT Plus Reagent kit (Affymetrix Inc., California US) was used for processing 100ng of total RNA per sample. Biotinylated sscDNA was hybridized for 16 hour at 45°C and 60 rpm on HuGene2.0st microarrays in a Hybridization Oven 640, both from Affymetrix. Microarrays were washed and stained in the Affymetrix Fluidics Station 450. Finally, they were confocal scanned using the GeneChip 3000 7G with Autoloader from Affymetrix. Raw fluorescence intensity values were stored in Chip Expression Level (CEL) file types, one per available blood sample. Data is available in the Gene Expression Omnibus database (GSE93945).

thumbnail
Table 2. Completed distance, time achieved and average speed per study participant.

Table indicates the race performance for each participant in the study. Biological sample availability is indicated in terms of pre- or post- race extraction.

https://doi.org/10.1371/journal.pone.0180322.t002

The corresponding expression profiles from the CEL files were background corrected, quantile normalized and summarized using the Robust Multichip Average (RMA) [21] on the R software platform [22] with BioConductor [23] using the oligo package [24]. The expression levels of 53,617 transcript clusters (TCs) were available per sample. Quality control (QC) was performed over pre-processed data to detect possible outliers based on the following metrics: relative log expression [25], normalized unscaled standard error [25], density intensity distributions (histogram and boxplot) and principal component analysis (PCA). Relevant versions of used packages are given in S1 Table.

Differential gene expression analysis (DGEA)

Only those TCs targeting protein-coding RNA molecules were considered for DGEA based on the annotation from the hugene20sttranscriptcluster.db package [26]. A non-supervised filtering [27] was applied to discard low expressed TCs which were assumed to be non-informative. TCs with expression values higher than the overall intensity mean, computed across all arrays, and on more than 12 arrays were selected for DGEA. The genefilter package [28] was used for this purpose. Then, a linear regression model (LM) was fitted to each TC expression value according to Eq (1). (1) where gk is the expression value of TC k, β0k is the LM intercept for TC expression value k, β1k and β2k are the unknown coefficients for the variables gender g and distance d respectively and ϵk are the random errors. The empirical Bayes moderated t-statistics tested whether each individual coefficient was zero using the limma package [29]. Statistically significant differentially expressed TCs (differential TCs) were selected and ranked (adjusted p-value < 5%, FDR) per LM predictor variable. Entrez Gene identifiers (IDs) were mapped from their differential TCs.

The resulting list of differential genes was used as input for the downstream analysis (Fig 1). A heatmap was generated with gplots package [30] for selected TCs including a hierarchical clustering with complete linkage method.

thumbnail
Fig 1. Complete workflow implemented for the study.

The differential genes list obtained from the differential gene expression analysis is taken as the initial step for the workflow. This represents the global response to intervention but can also be decomposed in independent components through an Independent Component Analysis (ICA) to obtain the independent block response. ICA is computed after applying a deflation method to the original expression data. Gene and transcriptional regulator (TR) enrichment analyses are computed over the global and independent response. Results are summarized in overrepresented pathway graphs and overrepresented TRs rankings.

https://doi.org/10.1371/journal.pone.0180322.g001

Independent activity analysis

Microarray expression data could be understood as a linear combination of independent expression sources, each one associated with a particular biological reading [31]. We computed an Independent Component Analysis (ICA) to extract these expression sources [32] according to Eq (2). (2) where X is an n × m matrix of the expression values of n genes under m array samples. The columns of the m × k source matrix S contain k independent components (ICs) and the k × n matrix A represents the linear mixing matrix. The row of matrix A comprises the weights with which the expression levels of the n genes contribute to each kth expression mode.

The list of differential genes was selected to build a matrix X. First, the optimum number of k ICs for X was obtained by estimating the optimal number of components in the PCA using the generalized cross-validation approximation (GCV) and the smoothing method [33], both implemented in the FactoMineR package [34]. Then a deflationary method was applied to X to remove the first component of variance as computed by the PCA. This was applied to eliminate the main response to the intervention characterized by the immune system and the genetic information processes as latter shown. These powerful signals act as a masking effect for the rest of underlying processes making difficult for ICA to detect them. Deflation was applied according to Eq (3): (3) where Y is an n × m matrix which refers to the expression values of n genes from m array samples captured by the first principal component (PC1), zi1 is the scores vector of the ith array sample in PC1 and ϕj1 corresponds to the loadings vector of the jth gene. Lastly, an estimated matrix was built according to Eq (4): (4)

ICA was performed over both the matrixes XT and where k − 1 ICs were considered in the second case due to the applied deflation. The fastICA package was used [35] (εt <1e-4, Glog cosh with α1 = 1 [32], ICs extracted simultaneously). Those genes with absolute weight value included in the ninth decile were considered as the most representative genes for each specific IC.

Gene enrichment analysis (GEA)

A GEA was applied in two stages: (i) globally when considering the list of differential genes and (ii) specifically for each IC derived from ICA and only considering the most representative genes. GEA was computed over Kyoto Encyclopedia of Genes and Genoms (KEGG) PATHWAY [36], Reactome [37] and The Gene Ontology (GO) Biological Processes [38] databases with the package clusterProfiler [39]. For each queried biological pathway or GO term, an adjusted p-value was calculated with a hypergeometric distribution test (adjusted p-value < 5%, False Discovery Rate FDR). The background distribution was defined by all available annotations in the relevant database or by the list of differential genes if the global or ICA stage was considered, respectively.

Transcriptional regulator enrichment analysis (TREA)

To explore overrepresented transcriptional regulators (TRs) as a response to UMT completed distance, a TREA was conducted. We considered differential genes to be potentially regulated by one or more TRs. The TREA was implemented with a hypergeometric model to assess whether the number of differential genes related to a specific TR was larger than expected. TRs were ranked based on their adjusted p-values (<5%, FDR). The TREA was implemented in two stages, globally and specifically per IC. A compilation of interactions between human TRs and target genes (TGs) was obtained from the Open Regulatory Annotation (ORA) database v3.0 [40]. Interactions between 196 regulatory elements and 23,991 TGs were chosen (type of regulation was set to transcription factor binding site, GRCh37/hg19). Background distribution was defined by the complete customized database or by the list of differential genes if the global or ICA stage was considered, respectively.

Results

Only 3 out of the 17 initial participants in the study finished the UMT while the rest of volunteers decided to leave the race at different distances along the trail. This was due to the adverse weather conditions mainly because of low temperatures (from 0.9°C to 13.1°C) and rain presence (from 0 to 6.1mm/h). The corresponding completed distance per participant and respective achieved time is given in Table 2. Additionally, biological sample availability is commented relative to its extraction before and/or after the race (pre- and post-race respectively).

QC excluded one pre-race sample, without post-race counterpart, which showed an abnormal pattern (S1 Fig). Pre-processing and QC was repeated after its removal with positive results. Therefore, a total of 28 samples, 16 of them pre-race and 12 post-race, were kept for further analysis. After filtering by target protein-coding RNA molecules, 25,272 TCs were available for DGEA. This group of TCs interrogated 22,072 different genes. To visualize their main source of variance, a PCA was conducted over their expression values. PC1 captured 25% of the total data variance, this being aligned with the effect of participating in the UMT (Fig 2).

thumbnail
Fig 2. Principal component analysis (PCA) over the expression values of the ranked transcript clusters (TCs).

(a) PCA over 25,272 TCs expression values targeting protein coding RNA molecules included in HuGene20st microarray. First principal component was aligned with the effect of participating in the Ultra Marathon Trail (UMT) (b) Proportion (in %) of captured variance per principal component.

https://doi.org/10.1371/journal.pone.0180322.g002

DGEA reveals a list of 5084 distinct genes responding to intervention

DGEA identified 5,974 differential TCs as a response to UMT (β2 ≠ 0). Among the list of 5,974 differential TCs, 5,499 were unambiguously annotated to a single gene (S2 Table) while 475 had multiple annotations (S3 Table). The list of 5,499 differential TCs corresponded to 5,084 distinct genes which were mainly down-regulated (63%) rather than up-regulated (37%). No TC appeared with statistical correlation with runners’ age. Fig 3 shows an unsupervised clustering analysis of a subset from the latter 5,499 differential TCs. The figure indicates a coherent sorting of samples prior to UMT compared to posterior ones, except for one misclassified sample labelled as 17_POST which corresponds to an individual who only completed 17% of the UMT. DGEA additionally revealed 35 differential TCs (S4 Table) related to gender (β1 ≠ 0, all TCs with single gene annotation). A 43% of these differential genes showed a higher expression level in male than in female runners.

thumbnail
Fig 3. Heatmap of the ranked transcript clusters (TCs) with stronger effect as response to the intervention.

Heatmap of the TCs with stronger effect as response to completed distance in the Ultra Marathon Trail (UMT). Selection was based on β2 values obtained in the linear model. Those TCs with abs2) > were chosen corresponding to 1,115 among the 5,499 differential TCs.

https://doi.org/10.1371/journal.pone.0180322.g003

Global response

Pathways associated with genetic information processing, infectious diseases and immune system are mainly affected.

The 5,804 differential genes were used to conduct a global GEA for each of the three databases: KEGG, Reactome and GO Biological Processes. Results obtained from KEGG revealed a list of 42 statistically overrepresented pathways (Table 3). All of them were connected through 978 out of the 5,084 initial differential genes (Fig 4). According to the database structure, 11 among the 42 induced pathways were involved in genetic information processing with most of their annotated genes down-regulated (mean 86.7% ± standard deviation 11.1%). A total of 11 affected infectious diseases were distributed among bacterial (three), viral (five) and parasitic (three) infection types. Genes annotated to bacterial and parasitic pathways were up-regulated by 61.5% ± 3.6% and 61.2% ± 8%, respectively. Genes annotated to the viral pathways were mostly down-regulated by 59.7% ± 5.3%. Nine pathways from the immune system emerged, including specific signalling pathways (three) and related immune diseases (two). No significant common sense of regulation was observed in this case with the exception of the two immune diseases, both mainly down-regulated (62.1% and 88.9%). Both lymphoid and myeloid cell lines from the hematopoietic cell lineage pathway were impacted (S2 Fig). Cell surface molecules included in this pathway (26 out of 55) were showing up-regulation (CD1, CD11b, CD13, CD14, CD35, CD36, CD42, CD55, CD59, CD114, CD116, CD121, CD124 and CD126) or down-regulation (CD2, CD3, CD5, CD7, CD8, CD20, CD24, CD38, CD49, CD71, CD125 and CD127). Other overrepresented pathways refer to signal transduction such as signalling pathways for HIF-1 and Nuclear Factor NF-κβ, with 58.1% and 53.8% of the annotated genes up-regulated respectively. Several cellular processes, as apoptosis (with 52.6% of annotated genes up-regulated) and cell cycle (with 79.6% of annotated genes down-regulated) were also impacted. The complete list of up-and down-regulated genes per listed pathway is enclosed in S5 Table.

thumbnail
Fig 4. Network of the overrepresented Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways listed in Table 3 connected through their differential annotated genes.

Statistically significant overrepresented KEGG pathways within the 5,084 distinct genes with marked differential expression level as a response to endurance exercise. Pathway’s circle size is proportional to the number of annotated genes (node degree). Pathway’s node color refers to their specific main category according to the KEGG structure. Genes annotated to each pathway are color-coded according to their type of regulation (green codes for down-regulation and red for up-regulation).

https://doi.org/10.1371/journal.pone.0180322.g004

thumbnail
Table 3. List of the 42 overrepresented Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways as a response to the intervention.

KEGG pathway identifier (ID) and description is enclosed in the table. Pathway’s main category and subcategory are shown. Gene:Bg Ratio indicates the number of genes annotated to a pathway (within the specific list of 1,905 out of 5,084 differential genes which appear in KEGG database) versus the number of genes annotated to that specific pathway within the background. Background considers all genes included in the database which corresponds to 6,997 elements. Pathways are sorted based on their adj p-val (FDR correction) coded as *** < 0.001, ** < 0.01 and * < 0.05. Up-reg[%] indicates the percentage of differential genes annotated to the specific pathway being up-regulated.

https://doi.org/10.1371/journal.pone.0180322.t003

A total of 193 Reactome pathways were found statistically overrepresented (S6 Table). Table 4 shows a summary by clustering them into parental superclasses based on the database hierarchy. Gene Expression, Immune System and Disease were top affected superclasses which enclose biological information similar to abovementioned KEGG genetic information processing, immune system and infectious disease. Obtained Reactome pathways related to Disease were all concentrated on viral infectious diseases capturing 21 out of the 193 ranked pathways.

thumbnail
Table 4. Clustering of the 193 statistically overrepresented Reactome pathways into parental superclasses.

Table shows the number of overrepresented pathways annotated to each existing parental superclass according to database structure.

https://doi.org/10.1371/journal.pone.0180322.t004

A total of 1,232 GO terms from Biological Processes ontology were statistically overrepresented (S7 Table). Translation GO term was the most overrepresented based on this list (S3 Fig).

Comparison with the literature linked to common inflammatory markers and Th1/Th2 related genes.

Regarding the immune system, we compared our results with gene expression studies focused on common inflammatory markers after a single exercise intervention in humans (Table 5) as reviewed by other authors [18]. Different intervention types were considered in this review, but none of them referred to an UMT.

thumbnail
Table 5. Differential expressed genes related to common inflammatory markers.

List is based on the review presented by other authors [18]. Gene symbol and name are indicated for each marker in the list. ↓ indicates genes being down-regulated and ↑ genes being up-regulated in Reg column. Prior results refer to studies where the expression of the specific marker was evaluated in a single exercise-related intervention in humans. Genes are sorted in alphabetical order.

https://doi.org/10.1371/journal.pone.0180322.t005

We reproduced the same sense of immune imbalance as in [17] where the Th1/Th2 ratio was assessed one week after a marathon race. Although there is a partial overlap in the ranked genes (Table 6) with regard to prior study, we also observed a down-regulation trend in Th1 cytokines and related genes. Of note is the up-regulation of CEBPB which was previously related to Th2 cell response enhancer [51].

thumbnail
Table 6. Differential expressed genes related to T-helper 1 and T-helper 2 cells from immune system.

Th1 and Th2 cytokines and related genes with differential expression are listed.

https://doi.org/10.1371/journal.pone.0180322.t006

Identified overrepresented TRs related to hematopoietic cell lineage proliferation, gluconeogenesis and hypoxia situation.

A TREA was computed with 4,772 among the 5,084 differential genes which were annotated as TGs to any of the 196 available regulatory elements. Table 7 shows the 27 statistically overrepresented TRs. Only 10 among the 27 ranked TRs had been previously prioritized by the DGEA. From the list, RBL2, RB1 [52] or CTCF [53] are directly involved in chromatin structure modifications. Elements capable of interacting appeared simultaneously. E2F4 binds with high affinity to RBL2 and possibly binds with RB1 which interacts with E2F1 [54]. Eight known transcription factor (TF) families emerged significant (E2F, ETS, FOS, STAT, EGR, GATA, HIF and RUNX). Most of them are related to general processes such as cell cycle, cell proliferation and development. RUNX1, GATA2 and GATA3 act in the development and proliferation of the hematopoietic cell lineage where GATA2 has been considered elsewhere as the master regulator of hematopoietic progenitor cells [55]. TAL1, which collaborates with GATA1, is implicated in several aspects of the final differentiation of red blood cells [56]. HNF4, EGR1, CEBPA and YY1 are TRs described in the gluconeogenesis program in response to a fasting state [57]. HIF1A and EPAS1 are members of the HIF family whose respective signalling pathways were overrepresented. YY1 and EPAS1 are the most selective TFs obtained with, respectively, 90 and 265 out of 23,991 annotated TGs.

thumbnail
Table 7. List of the 27 statistically overrepresented transcriptional regulators (TRs) as a response to the intervention.

TR symbol and name are indicated for each TR in the list. Gene:Bg Ratio indicates the number of target genes regulated by the TR (within the specific list of 4,772 out of 5,084 differential genes which appear in customized TR database obtained from Open Regulatory Annotation database) versus the number of target genes regulated by the TR within the background. Background considers 23,991 genes included in the customized database for TR Enrichment Analysis (TREA). TRs are sorted based on their adj p-val (FDR correction) coded as ***<0.001, ** < 0.01 and—in case > 0.05. Last column indicates the adj p-val obtained from Differential Gene Expression Analysis (DGEA).

https://doi.org/10.1371/journal.pone.0180322.t007

Independent response activity

ICA was computed over a PCA projection at six components determined by the smooth and GCV methods. A matrix with the expression levels of the 5,084 differential genes was used for this purpose. The selection of the number of components was based on the mean error obtained for each number of PCs when applying GCV or smooth method (S4 Fig). First PC is capturing a 52% of data variance and threshold corresponding to 80% of cumulative percentage is achieved by six components (S5 Fig).

ICA decomposed the input expression matrix of 28 array samples × 5,084 differential genes into the mixing matrix A (6 × 5,084) and source matrix S (28 × 6). The mixing matrix contained the weights of 5,804 differential genes for each six independent response blocks to exercise (S6 Fig). A total of 509 main contributors per component were selected corresponding to the highest weight values. S8 Table indicates the number of matches between ICs and respective unique representatives which ranged between 22% (IC6) and 44% (IC3).

Dominance of the immune system.

First IC was capturing the induced responses both in the innate and in the adaptive immune system according to GEA results conducted over KEGG and Reactome databases (S9 Table and S10 Table respectively). A subset of surface cell markers found in global GEA (CD2, CD3, CD7, CD8, CD14, CD36, CD59, CD116 and CD121) plus new CD28 and CD40LG from hematopoietic cell lineage was affected. First line of defense for pathogen recognition arisen with toll-like receptors TLR2, TLR4 and TLR5 in different infectious diseases such as malaria (adj pval 0.014), amoebiasis (adj pval 0.027) and legionellosis (adj pval 0.039) according to GEA over KEGG database. They were also present in Reactome overrepresented pathways MyD88 deficiency (adj pval 0.041) and IRAK4 deficiency (adj pval 0.048). Ribosome pathway from KEGG was enriched from third IC group of genes, aligned with a considerable number of overrepresented Reactome pathways related to translation. Sixth IC was mainly involved with cell cycle and translation process again according to GEA over Reactome. There were not overrepresented pathways in the rest of ICs.

As a result of TREA, 11 regulator elements were found overrepresented from the group of genes from first IC (S11 Table). Nine of them were already obtained with the global list of differential genes. GATA2 was found in first and third ICs. ETS1 and SMARCA4, also known as BRG1, were found in fourth IC. There were no overrepresented TRs in the rest of ICs.

Removal of first line of variance.

Previous results provided similar biological insights as the global analysis where all the differential genes were considered. The first line of data variance, accounting for 52% as determined by PC1, featured the immune system response to the intervention. To avoid this, ICA was again computed over the PCA projection at five components after considering the deflationary method over the initial matrix X of 5,084 differential GE levels. Five IC sets were obtained and their 509 main contributors were selected for applying GEA and TREA on each one. The number of matches between them and respective unique elements now ranged between 65% (IC5) and 72% (IC2) (Table 8).

thumbnail
Table 8. Main contributors of the independent components (ICs) after applying deflationary method to differential genes expression matrix.

Main contributors were selected based on their highest weight values after subtracting one-dimensional data approximation (PC1) to differential genes expression matrix. Those located in the ninth decile were chosen, obtaining a total of 509 genes per IC. Table shows the number of matches between components and the elements that were unique per IC.

https://doi.org/10.1371/journal.pone.0180322.t008

Terms relative to electron transport chain, a complex signal transduction network and nervous system.

According to GEA results over KEGG (Table 9), first IC elucidated four down-regulated genes responsible for encoding major histocompatibility complex (MHC) class II proteins (HLA-DPA1, HLA-DPB1, HLA-DMA and HLA-DRA) (Fig 5A). These, together with the up-regulated TNF gene, matched in seven out of the nine overrepresented KEGG pathways, related to immune, autoimmune or alloimmune responses. The second IC presented three neurodegenerative diseases: Parkinson’s, Alzheimer’s and Huntington’s diseases (Fig 5B). All of them shared 10 down-regulated genes from the electron transport chain (ETC) in the mitochondrion (Table 10). Genes involved in signal transduction stood out among the 24 KEGG pathways enriched from third IC (Fig 5D). There were two main hubs of signal communication. The up-regulated MAPK3, MAPK13, PIK3R5, PIK3CD genes and the down-regulated MAPK8 and MAPK9 genes characterized one hub. Among them, MAPK3, PIK3R5 and PIK3CD were common for 20 out of the 24 pathways. The other hub was featured by the up-regulated PRKACA and ADCY4, related to cAMP second messengers, and GNAI2 gene. Two pathways associated with the nervous system, retrograde endocannabinoid signalling and morphine addiction, showed up-regulation in most of their annotated differential genes with a 70% and a 85.7% respectively. GABRD gene which encodes for a neurotransmitter GABA receptor was one of these up-regulated genes. OSCAR and AGER genes, both up-regulated, were respectively specific elements for osteoclast differentiation and AGE-RAGE signalling pathway in diabetic complications pathways. Five infectious diseases were overrepresented, four of them being of viral origin: HTLV-I infection, hepatitis C, hepatitis B and influenza A. The up-regulated SERPINB1 gene was annotated to, inter alia, the overrepresented amoebiasis KEGG pathway from fourth IC (Fig 5C). The role of this gene has been previously related to the mitigation of inflammation in pulmonary influenza infections [58]. Genes encoding ribosomal proteins, including mitochondrial ribosomal proteins, characterized the fifth IC.

thumbnail
Fig 5. Network of the overrepresented Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways listed in Table 9.

Pathways are connected through their differential annotated genes for each Independent Component (IC) after removing first line of variance. (a) IC1 (b) IC2 (c) IC4 and (d) IC3. Pathway’s node size is proportional to the number of annotated genes (node degree). Genes annotated to each pathway are color-coded according to their type of regulation (green codes for down-regulation and red for up-regulation) together with its official gene symbol. SP stands for signalling pathway.

https://doi.org/10.1371/journal.pone.0180322.g005

thumbnail
Table 9. List of the statistically overrepresented Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways obtained for each independent component (IC) after removing first line of variance.

KEGG pathway ID and description is enclosed in the table. Pathway’s main category and subcategory are shown. Gene:Bg Ratio indicates the number of genes annotated to a pathway within the specific list of differential genes among the 509 major contributors that are included in the database (i.e. 201 for IC1) versus the number of genes annotated to a pathway within the background. Background considers all differential genes included in KEGG database which corresponds to 1905 elements among 5084 differential genes. Pathways are sorted based on their adj p-val (FDR correction) coded as *** < 0.001, ** < 0.01 and * < 0.05. Up-reg indicates the percentage of differential genes annotated to the specific pathway being up-regulated.

https://doi.org/10.1371/journal.pone.0180322.t009

thumbnail
Table 10. Down-regulated genes from the electron transport chain as a response to the intervention.

https://doi.org/10.1371/journal.pone.0180322.t010

A summary of the GEA results over Reactome is included in S12 Table. Overrepresented pathways were only found for second and fifth ICs. In line with KEGG results, ETC was also shown in 2 among the 31 enriched pathways in the second IC: the citric acid (TCA) cycle and respiratory electron transport and the respiratory electron transport, ATP synthesis by chemiosmotic coupling, and heat production by uncoupling proteins pathways.

Overrepresented TRs among ICs are aligned with pathway enrichment results.

A list of 19 TRs was found overrepresented in IC2, IC3 and IC4 as a result of TREA (Table 11). Most of them, 17 out of the 19, were already prioritized when considering a global response (Table 7). In this case, ETS1 and E2F4 transcription factors were enriched in the second IC. E2F4 belong to E2F family which is known by its dual role in cell proliferation and its contribution to cell death in response to cell stress [59]. The third IC showed five enriched TRs (EGR1, VDR, ZNF263, TFAP2C and CTCF) where, as best we know, the last three have an unspecific TR role. EGR1 was connected to MAPK and vascular endothelial growth factor (VEGF) signalling, both highlighted GEA results for IC3 (Table 9), when studying the relationship between insulin sensitivity and exercise-induced gene expression [60]. The VDR gene is known to be connected to bone homeostasis [61] which is compatible with the presence of osteoclast differentiation pathway (Table 7 –IC3). From fourth IC, TP53 and SOX2 genes were the new hits found.

thumbnail
Table 11. List of the statistically overrepresented transcriptional regulators (TRs) obtained per independent component (IC) after removing first line of variance.

Three ICs (IC2, IC3 and IC4) among the computed five components show enriched TRs. TR symbol and name are indicated for each TR in the list. Gene:Bg Ratio indicates the number of target genes (TGs) regulated by the specific TR among the 509 major contributor genes versus the number of TGs regulated by the TR within the background. Only those major contributors that appear in the customized TR database obtained from Open Regulatory Annotation (ORA) database per IC are considered (i.e. for IC2, 489 out of the 509 contributors). Background considers 4,772 genes included in the customized ORA database for TR Enrichment Analysis (TREA) among 5,084 differential genes. TRs are sorted based on their adj p-val (FDR correction) coded as *** < 0.001, ** < 0.01, * < 0.05 and—in case > 0.05. Last column indicates the adj p-val obtained from Differential Gene Expression Analysis (DGEA).

https://doi.org/10.1371/journal.pone.0180322.t011

Discussion and conclusions

Previous studies have accumulated evidence about the health risk reduction as a result of moderate physical activity [1]. Nevertheless, an U-curve pattern has been previously described when considering the effect of high intensity and prolonged exercise over cardiovascular [3] or URTI [16] risks. In this sense, an UMT is of interest due to its extreme conditions [5] and its consequences on the whole body homeostasis. To our knowledge, the present study is the first genome-wide investigation aiming an expression profiling in response to a UMT race.

Our results show that gene expression is heavily impacted by the intervention based on the 5,084 protein-coding genes, among 23,557 initially tested, with significant differential expression. The global gene enrichment analysis reveals extensive alterations in human biology mainly concentrated around the immune system, infectious diseases and genetic information processing.

A 36% of the enriched infectious diseases terms (Table 3) are caused by parasitic (Toxoplasmosis) and viral pathogens (Epstein-Barr virus infection, Herpes simplex infection and Influenza A) associated with URTI [62], An additional 27% implicates pathogens responsible of other respiratory infections such Legionellosis, Measles and Tuberculosis, the latter primarily attacking the lungs, while the rest were unrelated respiratory infections. These results do not necessarily imply that subjects presented a particular infection, but its genetic mechanisms triggered by the strenuous exercise.

We interpret protein synthesis as repressed based on the systematic down-regulation of the genes annotated to the related intracellular processes. This response is compatible with two opposite situations: the negative energy balance due to the high-demanding exercise [63] and, as defined by other authors, the maintenance of protein levels is a bioenergetically expensive process [64]. In a similar experiment using muscle biopsy samples, authors found an activation of muscle protein degradation in addition to muscle protein repression [65]. The autophagy-lysosomal and the ubiquitin proteasome pathways (UPP) mainly control protein degradation in skeletal muscle [66]. We report an overrepresentation of the Lysosome pathway with a general up-regulation of up to 65% of its annotated differential genes. Controversially, overrepresented pathways related to UPP clearly emerged down-regulated, Ubiquitin mediated proteolysis and Proteasome, with the 78% and 91% of their annotated differential genes respectively.

HIF-1 signalling pathway enrichment is aligned with the TREA results where EPAS1 (aka HIF-2α) and HIF-1A were found. We identified the up-regulation of genes related to the increase oxygen delivery (TIMP-1, HMOX-1), oxygen consumption reduction (HK, ALDOA and PFK2) and associated TR (HIF-1β aka ARNT). In human skeletal muscle studies, HIF-1 has been held to be responsible for, among other functions, a reduction in mitochondrial activity [20] and VEGF regulation [67]. Its activation has been previously reported after a single exercise [68]. On the other hand, the EPAS1 gene is a TF that plays a key role in the HIF pathway by activating genes in response to hypoxia [69], specifically those involved in erythropoiesis and angiogenesis [70]. While several studies have evaluated the influence of EPAS1 genetic variants in individual aerobic capacity [70] and athletic performance [71]; to our knowledge no specific studies have explored the EPAS1 response to exercise from a transcriptomics approach. Additionally, there is prior evidence of the collaboration of ETS1, another overrepresented TF, with HIF-1 in regulating hypoxia-inducible genes in pathological situations [72].

ICA identified further biological pathways including key alteration in mitochondrial activity and endocannabinoid signalling. Several genes from the ETC were systematically down-regulated as a obtained from the GEA applied to the ICs (Table 10). Most of them (NDUFA9, NDUFAB1, CYC1, UQCRQ and ATP5A1) are reported as a direct effect of ETS1 in cancer cells in its role of mitochondrial stress and dysfunction regulation [73]. TP53 gene was one of the additional transcriptional regulators retrieved after applying ICA. TP53 stands as an stress sensor of the cell such as oxidative stress, hypoxia and nutrient depravation [74], signals compatible with the experiment. Additionally, the TP53 gene has been related to the regulation of mitochondrial respiration [75] and possible exercise-induced mitochondrial biogenesis [2,76] through interactions with TFAM in the mitochondria. However, we have not observed any differential expression in TFAM. With regard to the endocannabinoid-signalling pathway, recent studies in mice describe the so called “runner’s high” dependence on the endocannabinoid system in response to wheel running [77] and how this exercise-induced effect is intensity-modulated in humans [78,79].

The current study has, however, certain limitations. First, the small sample size could limit the results validation in newer cohorts. However, the findings are consistent with existing literature in exercise-related studies. Secondly, the physical effort of each runner may be heterogeneous for the same completed distance. Nevertheless, this feature may be difficult to include in the linear regression model beyond the runner’s subjective perception.

In conclusion, the present study points to almost one fourth of all protein-coding genes affected by running an UMT, with a substantial number of human biology pathways overrepresented. In agreement with prior exercise-related studies, the global physiological approach is predominantly associated with immune system, infectious diseases and genetic information processing. The independent activity approach revealed additional pathways beyond the abovementioned which will require tailored investigations in larger sample sizes. Biological pathways and transcriptional regulators overrepresentation analysis offered a complementary interpretation of the results.

Supporting information

S1 Fig. Quality control–PCA over the expression values of the pre-processed microarray data from the initial 29 samples.

https://doi.org/10.1371/journal.pone.0180322.s001

(PDF)

S2 Fig. Overrepresented hematopoietic cell lineage extracted from KEGG pathway database with differential genes highlighted.

https://doi.org/10.1371/journal.pone.0180322.s002

(PDF)

S3 Fig. Summary of the 1232 statistically overrepresented GO terms from Biological Processes ontology.

https://doi.org/10.1371/journal.pone.0180322.s003

(PDF)

S4 Fig. Estimation of the optimal number of components in PCA with GCV and smooth methods.

https://doi.org/10.1371/journal.pone.0180322.s004

(PDF)

S5 Fig. Cumulative percentage of variance in PCA computed over the expression matrix of 5,084 differential genes.

https://doi.org/10.1371/journal.pone.0180322.s005

(PDF)

S6 Fig. Histogram of each kth row of the mixing matrix A representing the weights of the 5,084 differential genes.

https://doi.org/10.1371/journal.pone.0180322.s006

(PDF)

S1 Table. Software and package versions used in the study.

https://doi.org/10.1371/journal.pone.0180322.s007

(PDF)

S2 Table. List of 5,499 differential expressed transcript clusters with single gene annotation.

https://doi.org/10.1371/journal.pone.0180322.s008

(CSV)

S3 Table. List of 475 differential expressed transcript clusters with multiple gene annotations.

https://doi.org/10.1371/journal.pone.0180322.s009

(CSV)

S4 Table. List of 35 differential expressed transcript clusters related to gender contribution.

https://doi.org/10.1371/journal.pone.0180322.s010

(CSV)

S5 Table. Detailed list of the 42 overrepresented pathways from KEGG as a result of GEA.

https://doi.org/10.1371/journal.pone.0180322.s011

(CSV)

S6 Table. Detailed list of the 193 overrepresented pathways from Reactome as a result of GEA.

https://doi.org/10.1371/journal.pone.0180322.s012

(CSV)

S7 Table. Detailed list of the 1233 overrepresented GO terms from GO–Biological Processes ontology as a result of GEA.

https://doi.org/10.1371/journal.pone.0180322.s013

(CSV)

S8 Table. Number of main contributors to each IC based on their highest weight values.

https://doi.org/10.1371/journal.pone.0180322.s014

(PDF)

S9 Table. List of the statistically overrepresented KEGG pathways obtained per IC after ICA.

https://doi.org/10.1371/journal.pone.0180322.s015

(PDF)

S10 Table. List of the statistically overrepresented Reactome pathways obtained per IC after ICA.

https://doi.org/10.1371/journal.pone.0180322.s016

(PDF)

S11 Table. List of the statistically overrepresented transcriptional regulators (TRs) obtained per IC after ICA.

https://doi.org/10.1371/journal.pone.0180322.s017

(PDF)

S12 Table. List of the statistically overrepresented Reactome pathways obtained for IC2 and IC5 after removing first line of variance.

https://doi.org/10.1371/journal.pone.0180322.s018

(PDF)

Acknowledgments

The study was supported by the Bioinformatics and Biomedical Signals Laboratory (B2SLab), a consolidated research group of the Generalitat de Catalunya, Spain (2014SGR-1063). The authors would like to thank the runners who volunteered to participate in this research study.

Author Contributions

  1. Conceptualization: ER DB JS AP.
  2. Data curation: MM AP.
  3. Formal analysis: MM AP.
  4. Funding acquisition: JS AP.
  5. Investigation: MM ER DB JS AP.
  6. Methodology: MM AP.
  7. Project administration: MM ER DB JS AP.
  8. Resources: ER DB JS.
  9. Software: MM AP.
  10. Supervision: DB JS AP.
  11. Validation: MM.
  12. Visualization: MM.
  13. Writing – original draft: MM AP.
  14. Writing – review & editing: MM ER DB JS AP.

References

  1. 1. Lee D- C, Pate RR, Lavie CJ, Sui X, Church TS, Blair SN. Leisure-time running reduces all-cause and cardiovascular mortality risk. J Am Coll Cardiol 2014;64:472–81. pmid:25082581
  2. 2. Hawley JA, Hargreaves M, Joyner MJ, Zierath JR. Integrative Biology of Exercise. Cell 2014;159:738–49. pmid:25417152
  3. 3. O’Keefe JH, Patil HR, Lavie CJ, Magalski A, Vogel RA, McCullough PA. Potential Adverse Cardiovascular Effects From Excessive Endurance Exercise. Mayo Clin Proc 2012;87:587–95. pmid:22677079
  4. 4. Knez WL, Coombes JS, Jenkins DG. Ultra-endurance exercise and oxidative damage : implications for cardiovascular health. Sports Med 2006;36:429–41. pmid:16646630
  5. 5. Millet GP, Millet GY. Ultramarathon is an outstanding model for the study of adaptive responses to extreme load and stress. BMC Med 2012;10:77. pmid:22812424
  6. 6. Cejka N, Rüst CA, Lepers R, Onywera V, Rosemann T, Knechtle B. Participation and performance trends in 100-km ultra-marathons worldwide. J Sports Sci 2014;32:354–66. pmid:24015856
  7. 7. Mrakic-Sposta S, Gussoni M, Moretti S, Pratali L, Giardini G, Tacchini P, et al. Effects of mountain ultra-marathon running on ROS production and oxidative damage by micro-invasive analytic techniques. PLoS One 2015;10. pmid:26540518
  8. 8. Wüthrich TU, Marty J, Kerherve H, Millet GY, Verges S, Spengler CM. Aspects of respiratory muscle fatigue in a mountain ultramarathon race. Med Sci Sports Exerc 2015;47:519–27. pmid:25033264
  9. 9. Hurdiel R, Peze T, Daugherty J, Girard J, Poussel M, Poletti L, et al. Combined effects of sleep deprivation and strenuous exercise on cognitive performances during The North Face(R) Ultra Trail du Mont Blanc(R) (UTMB(R)). J Sports Sci 2015;33:670–4. pmid:25333827
  10. 10. Nakamura S, Kobayashi M, Sugino T, Kajimoto O, Matoba R, Matsubara K. Effect of exercise on gene expression profile in unfractionated peripheral blood leukocytes. Biochem Biophys Res Commun 2010;391:846–51. pmid:19945435
  11. 11. Dias RG, Silva MSM, Duarte NE, Bolani W, Alves CR, Junior JRL, et al. PBMCs express a transcriptome signature predictor of oxygen uptake responsiveness to endurance exercise training in men. Physiol Genomics 2015;47:13–23. pmid:25465030
  12. 12. Ostrowski K, Rohde T, Zacho M, Asp S, Pedersen BK. Evidence that interleukin-6 is produced in human skeletal muscle during prolonged running. J Physiol 1998;508:949–53. pmid:9518745
  13. 13. Nickel T, Emslander I, Sisic Z, David R, Schmaderer C, Marx N, et al. Modulation of dendritic cells and toll-like receptors by marathon running. Eur J Appl Physiol 2012;112:1699–708. pmid:21881949
  14. 14. Shephard RJ. Development of the discipline of exercise immunology. Exerc Immunol Rev 2010;16:194–222. pmid:20839500
  15. 15. Nieman DC. Clinical implications of exercise immunology. J Sport Heal Sci 2012;1:12–7.
  16. 16. Nieman DC. Upper respiratory tract infections and exercise. Thorax 1995;50:1229–31. pmid:8553291
  17. 17. Xiang L, Rehm KE, Marshall GD. Effects of strenuous exercise on Th1/Th2 gene expression from human peripheral blood mononuclear cells of marathon participants. Mol Immunol 2014;60:129–34. pmid:24853398
  18. 18. Gjevestad GO, Holven KB, Ulven SM. Effects of Exercise on Gene Expression of Inflammatory Markers in Human Peripheral Blood Cells: A Systematic Review. Curr Cardiovasc Risk Rep 2015;9:34. pmid:26005511
  19. 19. Egan B, Zierath JR. Exercise metabolism and the molecular regulation of skeletal muscle adaptation. Cell Metab 2013;17:162–84. pmid:23395166
  20. 20. Lindholm ME, Rundqvist H. Skeletal muscle hypoxia-inducible factor-1 and exercise. Exp Physiol 2016;101:28–32. pmid:26391197
  21. 21. Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U, et al. Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 2003;4:249–64. pmid:12925520
  22. 22. Team RC. R: A language and environment for statistical computing. R Foundation for Statistical Computing. n.d.
  23. 23. Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S, et al. Bioconductor: open software development for computational biology and bioinformatics. Genome Biol 2004;5:R80. pmid:15461798
  24. 24. Carvalho BS, Irizarry RA. A framework for oligonucleotide microarray preprocessing. Bioinformatics 2010;26:2363–7. pmid:20688976
  25. 25. Bolstad BM, Collin F, K.M. S, Irizarry RA, Speed TP. Experimental Design and Low-Level Analysis of Microarray Data. Int Rev Neurobiol 2004;60:25–58. pmid:15474586
  26. 26. MacDonald JW. hugene20sttranscriptcluster.db: Affymetrix hugene20 annotation data (chip hugene20sttranscriptcluster). R package version 8.3.1. 2013.
  27. 27. Bourgon R, Gentleman R, Huber W. Independent filtering increases detection power for high-throughput experiments. Proc Natl Acad Sci 2010;107:9546–51. pmid:20460310
  28. 28. Gentleman R, Carey V, Huber W, Hahne. F. genefilter: genefilter: methods for filtering genes from high-throughput experiments. R package version 1.50.0. n.d.
  29. 29. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 2015. pmid:25605792
  30. 30. Warnes GR, Bolker B, Bonebakker L, Gentleman R, Huber W, Liaw A, et al. gplots: Various R Programming Tools for Plotting Data. R package version 2.17.0. n.d.
  31. 31. Kong W, Mou X, Hu X. Exploring matrix factorization techniques for significant genes identification of Alzheimer’s disease microarray gene expression data. BMC Bioinformatics 2011;12 Suppl 5:S7. pmid:21989140
  32. 32. Hyvärinen A, Oja E. Independent component analysis: algorithms and applications. Neural Netw 2000;13:411–30. pmid:10946390
  33. 33. Josse J, Husson F. Selecting the number of components in principal component analysis using cross-validation approximations. Comput Stat Data Anal 2012;56:1869–79.
  34. 34. Lê S, Josse J, Husson F, others. FactoMineR: an R package for multivariate analysis. R package version 1.31.4. J Stat Softw 2008;25:1–18.
  35. 35. Marchini JL, Heaton C, Ripley BD. fastICA: FastICA Algorithms to perform ICA and Projection Pursuit. R package version 1.2.0. 2013.
  36. 36. Kanehisa M, Goto S. KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res 2000;28:27–30. pmid:10592173
  37. 37. Croft D, Mundo AF, Haw R, Milacic M, Weiser J, Wu G, et al. The Reactome pathway knowledgebase. Nucleic Acids Res 2014;42:D472–7. pmid:24243840
  38. 38. Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 2000;25:25–9. pmid:10802651
  39. 39. Yu G, Wang L-G, Han Y, He Q-Y. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS 2012;16:284–7. pmid:22455463
  40. 40. Lesurf R, Cotto KC, Wang G, Griffith M, Kasaian K, Jones SJM, et al. ORegAnno 3.0: a community-driven resource for curated regulatory annotation. Nucleic Acids Res 2015;44:D126–32. pmid:26578589
  41. 41. Connolly PH, Caiozzo VJ, Zaldivar F, Nemet D, Larson J, Hung S-P, et al. Effects of exercise on gene expression in human peripheral blood mononuclear cells. J Appl Physiol 2004;97:1461–9. pmid:15194674
  42. 42. Ulven SM, Foss SS, Skjolsvik AM, Stadheim HK, Myhrstad MC, Raael E, et al. An acute bout of exercise modulate the inflammatory response in peripheral blood mononuclear cells in healthy young men. Arch Physiol Biochem 2015;121:41–9. pmid:25720858
  43. 43. Henriquez-Olguin C, Diaz-Vegas A, Utreras-Mendoza Y, Campos C, Arias-Calderon M, Llanos P, et al. NOX2 Inhibition Impairs Early Muscle Gene Expression Induced by a Single Exercise Bout. Front Physiol 2016;7:282. pmid:27471471
  44. 44. Fehrenbach E, Niess AM, Schlotz E, Passek F, Dickhuth H- H, Northoff H. Transcriptional and translational regulation of heat shock proteins in leukocytes of endurance runners. J Appl Physiol 2000;89:704 LP–710.
  45. 45. Vance DD, Chen GL, Stoutenberg M, Myerburg RJ, Jacobs K, Nathanson L, et al. Cardiac performance, biomarkers and gene expression studies in previously sedentary men participating in half-marathon training 2014;6. pmid:24552436
  46. 46. Rodriguez-Miguelez P, Fernandez-Gonzalo R, Almar M, Mejias Y, Rivas A, de Paz JA, et al. Role of Toll-like receptor 2 and 4 signaling pathways on the inflammatory response to resistance training in elderly subjects. Age (Dordr) 2014;36:9734. pmid:25427999
  47. 47. Kimsa MC, Strzalka-Mrozik B, Kimsa MW, Gola J, Kochanska-Dziurowicz A, Zebrowska A, et al. Differential expression of inflammation-related genes after intense exercise. Prague Med Rep 2014;115:24–32. pmid:24874932
  48. 48. Büttner P, Mosig S, Lechtermann A, Funke H, Mooren FC. Exercise affects the gene expression profiles of human white blood cells. J Appl Physiol 2007;102:26–36. pmid:16990507
  49. 49. Jenkins NT, Landers RQ, Prior SJ, Soni N, Spangenburg EE, Hagberg JM. Effects of acute and chronic endurance exercise on intracellular nitric oxide and superoxide in circulating CD34(+) and CD34(-) cells. J Appl Physiol 2011;111:929–37. pmid:21700895
  50. 50. Joro R, Uusitalo A, DeRuisseau KC, Atalay M. Changes in cytokines, leptin, and IGF-1 levels in overtrained athletes during a prolonged recovery phase: A case-control study. J Sports Sci 2016:1–8. pmid:27966392
  51. 51. Huber R, Pietsch D, Panterodt T, Brand K. Regulation of C/EBPbeta and resulting functions in cells of the monocytic lineage. Cell Signal 2012;24:1287–96. pmid:22374303
  52. 52. Gonzalo S, García-Cao M, Fraga MF, Schotta G, Peters AHFM, Cotter SE, et al. Role of the RB1 family in stabilizing histone methylation at constitutive heterochromatin. Nat Cell Biol 2005;7:420–8. pmid:15750587
  53. 53. Ong C- T, Corces VG. CTCF: An Architectural Protein Bridging Genome Topology and Function. Nat Rev Genet 2014;15:234–46. pmid:24614316
  54. 54. The UniProt Consortium. UniProt: a hub for protein information. Nucleic Acids Res 2014;43:D204–12. pmid:25348405
  55. 55. Bresnick EH, Katsumura KR, Lee H-Y, Johnson KD, Perkins AS. Master regulatory GATA transcription factors: mechanistic principles and emerging links to hematologic malignancies. Nucleic Acids Res 2012;40:5819–31. pmid:22492510
  56. 56. Kassouf MT, Hughes JR, Taylor S, McGowan SJ, Soneji S, Green AL, et al. Genome-wide identification of TAL1’s functional targets: insights into its mechanisms of action in primary erythroid cells. Genome Res 2010;20:1064–83. pmid:20566737
  57. 57. Goldstein I, Hager GL. Transcriptional and Chromatin Regulation during Fasting—The Genomic Era. Trends Endocrinol Metab 2015;26:699–710. pmid:26520657
  58. 58. Gong D, Farley K, White M, Hartshorn KL, Benarafa C, Remold-O’Donnell E. Critical role of serpinB1 in regulating inflammatory responses in pulmonary influenza infection. J Infect Dis 2011;204:592–600. pmid:21791661
  59. 59. Iaquinta PJ, Lees JA. Life and death decisions by the E2F transcription factors. Curr Opin Cell Biol 2007;19:649–57. pmid:18032011
  60. 60. McLean CS, Mielke C, Cordova JM, Langlais PR, Bowen B, Miranda D, et al. Gene and MicroRNA Expression Responses to Exercise; Relationship with Insulin Sensitivity. PLoS One 2015;10:e0127089. pmid:25984722
  61. 61. Anderson PH, Lam NN, Turner AG, Davey RA, Kogawa M, Atkins GJ, et al. The pleiotropic effects of vitamin D in bone. J Steroid Biochem Mol Biol 2013;136:190–4. pmid:22981997
  62. 62. Walsh NP, Gleeson M, Shephard RJ, Gleeson M, Woods JA, Bishop NC, et al. Position statement. Part one: Immune function and exercise. Exerc Immunol Rev 2011;17:6–63. pmid:21446352
  63. 63. Pikosky MA, Smith TJ, Grediagin A, Castaneda-Sceppa C, Byerley L, Glickman EL, et al. Increased protein maintains nitrogen balance during exercise-induced energy deficit. Med Sci Sports Exerc 2008;40:505–12. pmid:18379214
  64. 64. Smiles WJ, Hawley JA, Camera DM. Effects of skeletal muscle energy availability on protein turnover responses to exercise. J Exp Biol 2016;219:214–25. pmid:26792333
  65. 65. Jamart C, Francaux M, Millet GY, Deldicque L, Frère D, Féasson L. Modulation of autophagy and ubiquitin-proteasome pathways during ultra-endurance running. J Appl Physiol 2012;112:1529–37. pmid:22345427
  66. 66. Sandri M. Signaling in Muscle Atrophy and Hypertrophy. Physiology 2008;23:160 LP–170.
  67. 67. Hoier B, Hellsten Y. Exercise-induced capillary growth in human skeletal muscle and the dynamics of VEGF. Microcirculation 2014;21:301–14. pmid:24450403
  68. 68. Ameln H. Physiological activation of hypoxia inducible factor-1 in human skeletal muscle. FASEB J 2005. pmid:15811877
  69. 69. Patel SA, Simon MC. Biology of hypoxia-inducible factor-2alpha in development and disease. Cell Death Differ 2008;15:628–34. pmid:18259197
  70. 70. Voisin S, Cieszczyk P, Pushkarev VP, Dyatlov DA, Vashlyayev BF, Shumaylov VA, et al. EPAS1 gene variants are associated with sprint/power athletic performance in two cohorts of European athletes. BMC Genomics 2014;15:382. pmid:24884370
  71. 71. Henderson J, Withford-Cave JM, Duffy DL, Cole SJ, Sawyer NA, Gulbin JP, et al. The EPAS1 gene influences the aerobic–anaerobic contribution in elite endurance athletes. Hum Genet 2005;118:416–23. pmid:16208515
  72. 72. Salnikow K, Aprelikova O, Ivanov S, Tackett S, Kaczmarek M, Karaczyn A, et al. Regulation of hypoxia-inducible genes by ETS1 transcription factor. Carcinogenesis 2008;29:1493–9. pmid:18381358
  73. 73. Verschoor ML, Wilson LA, Verschoor CP, Singh G. Ets-1 regulates energy metabolism in cancer cells. PLoS One 2010;5:e13565. pmid:21042593
  74. 74. Bieging KT, Mello SS, Attardi LD. Unravelling mechanisms of p53-mediated tumour suppression. Nat Rev Cancer 2014;14:359–70. pmid:24739573
  75. 75. Lago CU, Sung HJ, Ma W, Wang P, Hwang PM. p53, aerobic metabolism, and cancer. Antioxid Redox Signal 2011;15:1739–48. pmid:20919942
  76. 76. Bartlett JD, Louhelainen J, Iqbal Z, Cochran AJ, Gibala MJ, Gregson W, et al. Reduced carbohydrate availability enhances exercise-induced p53 signaling in human skeletal muscle: implications for mitochondrial biogenesis. Am J Physiol Regul Integr Comp Physiol 2013;304:R450–8. pmid:23364526
  77. 77. Fuss J, Steinle J, Bindila L, Auer MK, Kirchherr H, Lutz B, et al. A runner’s high depends on cannabinoid receptors in mice. Proc Natl Acad Sci 2015;112:13105–8. pmid:26438875
  78. 78. Tantimonaco M, Ceci R, Sabatini S, Catani MV, Rossi A, Gasperi V, et al. Physical activity and the endocannabinoid system: an overview. Cell Mol Life Sci 2014;71:2681–98. pmid:24526057
  79. 79. Raichlen DA, Foster AD, Seillier A, Giuffrida A, Gerdeman GL. Exercise-induced endocannabinoid signaling is modulated by intensity. Eur J Appl Physiol 2013;113:869–75. pmid:22990628