Ethical approval and study participants
Eight national level ski athletes took part in this study. None of them suffered from acute or chronic diseases or reported intake of medication. Participants were informed about the nature, purpose, and potential risks of the experiments and signed an informed consent statement approved by the ethics committee of Scientific Research Center Bioclinicum (Moscow, Russia).
Anthropometric measurements
Height, weight, medical historical data and resting vital signs were recorded at the time of enrolment.
Exercise test protocol
In order to determine the VO
2max values, each subject performed a treadmill test with an incremental step protocol until exhaustion as described previously [
47]. VO
2max was calculated as described [
49]. The anaerobic threshold was calculated using the standard V-slope method [
50].
Two weeks later, athletes participated in the main exercise, consisting of running at 80% VO2 peak for 30 min on a treadmill. The exercise was performed during the morning hours (between 8 and 11 a.m.) keeping the exact test time for each participant constant.
Blood sampling
Venous blood was collected at four time points during the exercise. A 20-gauge intravenous catheter was placed antiseptically into a dorsal hand vein or a vein in the distal forearm as dictated by favourable anatomy using the Seldinger technique and then secured with tape. During ME 2.5 ml of blood was collected in PAXgene blood RNA tube, 7 ml in a Serum separator tube (BD, USA) and 4.5 ml in a tube containing buffered tri-sodium citrate (BD, USA) for flow cytometry analysis at baseline (prior to exercise testing) and immediately post-exercise (within 1 min of completion of exercise testing). After 30 min of rest and after 60 min of rest following exercise testing 2.5 ml of blood was collected in PAXgene blood RNA tubes.
According to the Affymetrix Manual P/N 701880 Rev. 4 total RNA was extracted using the PAXgene Blood RNA kit as recommended by the manufacturer. RNA concentrations were determined by the Nanodrop photometer (NanoDrop, USA). RNA quality was checked using the Agilent Bioanalyser 2100 System (Agilent Technologies, USA). For all samples RNA integrity number (RIN) was greater than 7.
Flow cytometry analysis
Flow cytometry analysis of blood samples was performed using fluorescently labeled antibodies against B and T cell receptors and natural killer (NK) cell markers. Cells were labeled with antibodies against CD3 (FITC), CD4 (PE), CD8 (PE), CD16 (FITC), CD56 (PE), CD19 (FITC) (Sorbent, Russia), where NK cells were distinguished from the rest of the lymphocytes via positive expression of CD56 and negative expression of CD3.
The samples were analyzed on a FACScan Calibur flow cytometer (BD Biosciences, USA) and leukocytes were gated based on forward and side scatter properties. Events in the range of 40,000–200,000 were collected depending on the occurrence of the investigated leukocyte population, and analyzed with CELLQuest Pro analysis software (BD Biosciences, USA). To ensure flow cytometric standardization, the voltage settings were updated daily using ‘Calibrate Beads’ (BD Biosciences, USA).
Microarray data processing
Microarray data was processed using bioconductor [
53] xps package implementation of RMA [
54]. At the first step background correction was performed based on a global model for the distribution of probe intensities [
54]. Then a quantile normalization algorithm [
55] (so-called probe-level normalization) was applied to the preprocessed data. Finally, fitting a robust linear model using Tukey’s median polish procedure [
56] was done to convert probe intensities to the expression levels of probesets.
The statistical analysis of microarray data was performed using bioconductor [
53] package limma [
57]. The analysis was based on a generalized linear model [
57,
58] approach. In this approach one constructs a linear data model with a structure determined by the experiment layout, and then fits this model to the actual data. The linear model is defined in terms of a so-called design matrix. The number of rows in this matrix coincides with the number of experiment samples, and the number of columns coincides with the number of factors that have an essential influence on the measurable values. The value at the i-th row and j-th column of a design matrix specifies an effect of a factor j on a sample i. Each measurable value (i.e., each probeset) in this approach is analysed independently. For each probeset a vector of its expression values E is represented in the form E = Dβ + ϵ, where D is a design matrix, β is a vector of coefficients indicating values of each factor’s actual influence on the analyzed probeset, ϵ is a vector of error, and model fitting consists in the minimization of “error term” ϵ by finding optimal coefficients β. After the coefficients β are computed for each probeset, one can test various hypotheses on the structure of considered factors. For example, in order to find probesets that are affected by a factor i, one should search for the probesets with βi statistically different from zero.
The general linear model was applied in the analysis of the studied transcriptome changes as follows. The model that takes into account both experiment time points and athletes individual features was used. Thus, the total number of factors was eleven: c1, …, c8 correspond to an expression level of each athlete in a normal state (for all samples the effect of these factors is set to 1 if a sample and a factor correspond to the same athlete, and set to 0 otherwise), and d1, d2, d3 correspond to the changes induces by the exercises, exercises and 30-minutes relaxation, and exercises and 60-minutes relaxation respectively (for all samples the effect of these factors is set to 1 if a sample and a factor correspond to the same experiment time point, and set to 0 otherwise). The total number of samples was 32: 4 samples for each athlete.
For each pair of experiment time points the detection of probesets with reliable difference between time points was performed. The probeset was considered to have a reliable difference between time points k, m if an adjusted p-value of an equality βk = βm (where βl was an actual influence of exercises at experiment time point for l = 1,2,3, and β0 = 0) was less than 0.05, and a log-fold change was greater than 0.484 (this threshold corresponds to an intensity change by more than 40%). The Benjamini-Hochberg [
59] algorithm was used for multiple testing adjustment. The minimum adjusted p-value for all pairs of time points is indicated for each differentially expressed probeset in the Table of differentially expressed mRNAs and miRNAs (Additional file
1) as a statistical significance value.
Pathway analysis
Bioinformatic analysis was performed using DAVID online tool (
http://david.abcc.ncifcrf.gov) as described elsewhere [
60]. So all the analyzed genes were classified into several functional groups, and the groups that may be potentially associated with physiological stress were considered and listed in the tabs of excel document. P-values on the tabs are modified Fisher Exact P-Values. When members of two independent groups can fall into one of two mutually exclusive categories, Fisher Exact test is used to determine whether the proportions of those falling into each category differs by group. In DAVID annotation system, Fisher Exact test is adopted to measure the gene-enrichment in annotation terms.