Systematic Assessment of Blood-Borne MicroRNAs Highlights Molecular Profiles of Endurance Sport and Carbohydrate Uptake

Abstract

Multiple studies endorsed the positive effect of regular exercise on mental and physical health. However, the molecular mechanisms underlying training-induced fitness in combination with personal life-style remain largely unexplored. Circulating biomarkers such as microRNAs (miRNAs) offer themselves for studying systemic and cellular changes since they can be collected from the bloodstream in a low-invasive manner. In Homo sapiens miRNAs are known to regulate a substantial number of protein-coding genes in a post-transcriptional manner and hence are of great interest to understand differential gene expression profiles, offering a cost-effective mechanism to study molecular training adaption, and connecting the dots from genomics to observed phenotypes. Here, we investigated molecular expression patterns of 2549 miRNAs in whole-blood samples from 23 healthy and untrained adult participants of a cross-over study, consisting of eight weeks of endurance training, with several sessions per week, followed by 8 weeks of washout and another 8 weeks of running, using microarrays. Participants were randomly assigned to one of the two study groups, one of which administered carbohydrates before each session in the first training period, and switching the treatment group for the second training period. During running sessions clinical parameters as heartbeat frequency were recorded. This information was extended with four measurements of maximum oxygen uptake (VO 2 max) for each participant. We observed that multiple circulating miRNAs show expression changes after endurance training, leveraging the capability to separate the blood samples by training status. To this end, we demonstrate that most of the variance in miRNA expression can be explained by both common and known biological and technical factors. Our findings highlight six distinct clusters of miRNAs, each exhibiting an oscillating expression profile across the four study timepoints, that can effectively be utilized to predict phenotypic VO 2 max levels. In addition, we identified miR-532-5p as a candidate marker to determine personal alterations in physical training performance on a case-by-case analysis taking the influence of a carbohydrate-rich nutrition into account. In literature, miR-532-5p is known as a common down-regulated miRNA in diabetes and obesity, possibly providing a molecular link between cellular homeostasis, personal fitness levels, and health in aging. We conclude that circulating miRNA expression can be altered due to regular endurance training, independent of the carbohydrate (CHO) availability in the training timeframe. Further validation studies are required to confirm the role of exercise-affected miRNAs and the extraordinary function of miR-532-5p in modulating the metabolic response to a high availability of glucose.

Keywords: circulating biomarker; full-blood measurements; glucose nutrition; homeostasis; microRNA; microarray; physical exercising; randomized cross-over study; sncRNAs.

Figure 1

Overview on study design. Healthy and untrained participants were randomly assigned to any of two training groups, each performing of eight weeks of $4\times 45$ min training, followed by a wash-out phase, again followed by eight weeks of endurance training. In the first period participants of one group orally administered glucose-solution 15 min before each running session, while participants of the second group administered carbohydrates in their second training period (cross-over). At four timepoints ($E_1$, $A_1$, $E_2$, and $A_2$) a blood-sample was taken and measured using complementary DNA (cDNA) microarrays probed with 2549 human microRNAs.

Figure 2

Analysis of variance and factors explaining it using measured miRNA expression data. (a) Sample distribution within the first two principal components obtained from principal component analysis (PCA) along with the percentage of variance explained in each dimension; (b) Results from principal variance component analysis (PVCA) showing estimates of variance in the expression data that can be explained with both known and unknown (hidden) sample annotation factors. Each bar corresponds to one factor, where mixed interactions between two variables are also possible and marked respectively by a colon; (c) two-dimensional uniform manifold approximation and projection (UMAP) embedding using the miRNA-sample matrix $X\in {\mathbb{R}}^{90\times 307}$; (d) two-dimensional t-distributed stochastic neighbor embedding (t-SNE) embedding using the miRNA-sample matrix X. In each dimension reduction panel a single point corresponds to one sample that is colored according to the timepoint of blood extraction and relative to the training period, i.e., before (blue) or after (red) one of the training intervals.

Figure 3

Volcano plots for six comparison setups using 307 microRNAs. The x-axis indicates log${}_2$ fold change, while the y-axis refers to the negative decade logarithm of p-values from student’s t-tests (unadjusted). Dashed horizontal lines indicate a p-value of $0.05$, while dashed vertical lines indicate a log${}_2$ fold change of 2. (a) Pooling the samples from the two participant groups and using timepoints of the first training interval ($E_1$ and $A_1$); (b) pooled sample approach analogous to (a) but with timepoints from the second training period ($E_2$ and $A_2$); (c) analysis for timepoints like in (a) only using samples from the first treatment group; (d) analysis for timepoints like in (b) only using samples from the first treatment group; (e) analysis for timepoints like in (a) only using samples from the second treatment group; (f) Analysis for timepoints like in (b) only using samples from the second treatment group. MiRNAs that exceed both axis thresholds (dashed lines) are colored according to their direction of dys-regulation, i.e., red for up-regulation and green for down-regulation after training. For each row of panels, significantly de-regulated miRNAs on the left side (panels (a,c,e) are labelled and displayed on the corresponding right side (panels b,d,f).

Figure 4

Distribution of Z-scores into six miRNA clusters $C_1$–$C_6$ corresponding to panel (a–f), along the four study timepoints using all samples. Each single grey line corresponds to one miRNA expression profile for one participant. Thick black lines display cluster-specific and smoothed curves of a cubic b-spline basis ($b=3$). The distinct clusters contain microRNAs that exhibit similar expression patterns over time. Although every cluster exhibits a wave-like expression pattern, miRNAs in $C_1$, $C_3$, $C_5$ show a tendency to be up-regulated after training (up, down, up) while those in $C_2$, $C_4$, and $C_6$ tend to be down-regulated after a period of endurance exercise (down, up, down).

Figure 5

Top 20 positive and negative miRNA regression coefficients. Feature coefficients stem from the best linear model with respect to $R^2$ predicting the measured VO${}_2$ max values. In total 86 out of 307 miRNAs were assigned a coefficient unequal from zero. Each miRNA bar is colored according to its cluster identity ($C_1$–$C_6$) from Figure 4, highlighting a differential distribution of these clusters among the most important regression coefficients.

Figure 6

Assignment procedure conducted to find CHO recommendation groups and group-wise expression of candidate marker miR-532-5p. (a) Schematic workflow of devised procedure to classify each participant p into one of the two recommendation groups $r\left(p\right)\in \{0,1\}$. The decision process primarily depends on the personal change in VO${}_2$ max values during the first training period, ${\Delta }_1=V{O}_2\left(A_1\right)-V{O}_2\left(E_1\right)$ and ${\Delta }_2=V{O}_2\left(A_2\right)-V{O}_2\left(E_2\right)$ from the second training period. In a second step, participants are assigned to exactly one of the two recommendation groups based on which interval the better $\Delta V{O}_2$ occurred; (b) Paired barplots showing the expression of miR-532-5p across training period timepoints, i.e., $T1\in \{E_1,E_2\}$ and $T2\in \{E_2,A_2\}$, compared between the two CHO treatment recommendation groups, once using training periods without and once with an oral administration of glucose. Recommendation groups are highlighted by distinct colors where blue corresponds to a positive and red to a negative recommendation on glucose uptake. Each black point belongs to one sample in the distribution. Bar heights display the mean expression values and black error bars mean ± standard deviation. Smoothed quadratic b-splines ($b=2$) are drawn as black curves with a contrast margin.