Endurance training remodels sperm-borne small RNA expression and methylation at neurological gene hotspots

Remodeling of the sperm epigenome by lifestyle factors before conception could account for altered metabolism in the next generation offspring. Here, we hypothesized that endurance training changes the epigenome of human spermatozoa. Using small RNA (sRNA) sequencing and reduced representation bisulfite sequencing (RRBS), we, respectively, investigated sRNA expression and DNA methylation in pure fractions of motile spermatozoa collected from young healthy individuals before, after 6 weeks of endurance training and after 3 months without exercise. Expression of 8 PIWI interacting RNA were changed by exercise training. RRBS analysis revealed 330 differentially methylated regions (DMRs) after training and 303 DMRs after the detraining period, which were, in both conditions, enriched at close vicinity of transcription start sites. Ontology analysis of genes located at proximity of DMRs returned terms related to neurological function at the trained state and, to a much lesser extent, at the detrained state. Our study reveal that short-term endurance training induces marked remodeling of the sperm epigenome, and identify genes related to the development of the central nervous system as potential hot spots for epigenetic variation upon environmental stress. Electronic supplementary material The online version of this article (10.1186/s13148-018-0446-7) contains supplementary material, which is available to authorized users.


Introduction
Gametes contain epigenetic information that plays a fundamental role in embryonic development [1]. Any preconceptional disturbance of the gametic epigenome has thus the potential to alter the phenotype of the next generation offspring through so-called epigenetic inheritance. In spermatozoa, not only methylation of DNA and post-translational modifications of histones carry epigenetic signals but also, sperm-borne small RNA (sRNA) can contribute to epigenetic inheritance, as supported by microinjection experiments of spermatozoal RNA into fertilized oocytes [2][3][4].
Epidemiological evidence indicates that paternal nutrition before conception can alter the phenotype of the following generation [5,6]. Numerous animal studies have now clearly established that environmental factors can remodel the sperm epigenome at the level of post-translational modifications of histones, DNA methylation, and sRNA [7][8][9][10][11][12][13][14]. Previously, we showed that sRNA expression and DNA methylation are altered in sperm cells of obese men after gastric-bypass-induced weight loss [15]. Whether physical exercise, a potent intervention to treat or prevent obesity and related diseases like obesity and type 2 diabetes (T2D) [16,17], can concomitantly remodel the sperm sRNA and DNA methylation profile is unknown.
Here, we hypothesized that endurance training changes the epigenetic profile of human spermatozoa. We used an intervention protocol of exercise training and detraining, to dissociate the effect of exercise to that of time, and to analyze the potential long-term memory of endurance training on the sperm epigenome. We show that exercise training specifically remodels the expression of several sRNAs and changes DNA methylation at specific gene hotspots related to brain function and development.

Results and discussion
Single ejaculates were obtained at three different time points; at baseline (referred to as Untrained), 4 days after ending a six-week endurance exercise intervention (referred to as Trained) and after 3 months without exercise training (referred to as Detrained). Analysis of DNA methylation was performed on the 12 participants while for sRNA, a subset of 6-9 participants was analyzed (see method section for details, see Additional file 1: Figure S1 for an overview). Clinical characteristics of the volunteers at the three time points are presented in Table 1A and B. As expected, aerobic capacity, as measured by VO 2 peak, was increased from a median value of 44.15/46.2 (respectively for the DNA methylation/sRNA subsets) ml/kg/min at the untrained state to 56.2/53.7 ml/kg/min at the trained state (p < 0.001/p < 0.005). Three months after the training program, VO 2 peak decreased to 51.2/50.7 ml/kg/ min, and was no longer significantly higher than baseline. We did not observe any inter-individual differences in sperm quality throughout the course of the exercise intervention, nor the detraining period.

Exercise training modulates sperm sRNA expression in a reversible fashion
We first investigated small RNA (sRNA) expression in purified sperm from the same subjects at the Untrained, Trained and Detrained state by sRNA sequencing. Consistent with previous reports, PIWI-interacting RNAs (piRNA) are expressed at higher levels than tRNA fragments (tRF) and microRNAs in human sperm [15,18] (Fig. 1a, Additional file 2: Figure S3, Additional file 3: Figure S4, Additional file 4: Figure S5). To analyze the effect of training on sRNA expression, we identified 5 piRNAs and 27 fragments of repetitive elements that were differentially expressed between the Trained and the Untrained state. Between any two time-points, a total of 3 miRNAs, 2 tRNAs, 6 piRNAs, and 38 repetitive elements were differentially expressed (Fig. 1b, Additional file 5: Figure S6), false discovery rate [FDR] < 0.1; Additional file 6: Tables S1-S4, Additional file 7: Table  S5). To identify reversible sRNA expression changes, we compared the changes observed between Untrained/ Trained with the changes at the Trained/Detrained and Untrained/Detrained. We found several piRNAs underwent a transient change in expression (Fig 1b). The expression of five of the six piRNA was changed between the Untrained and Trained state, while no piRNA was differentially expressed when comparing the Untrained and Detrained state, indicating a specific response to exercise training. On the other hand, no miRNAs were differentially expressed between the Untrained and Trained state, but one out of three were found between Untrained and Detrained and two of three were found between Trained and Detrained, suggesting changes in miRNA expression are not primarily triggered by exercise training. Expression of tRNAs followed a similar pattern where all changes were exclusively detected when comparing the Untrained to the Detrained state. Lastly, repetitive elements did not follow a specific pattern (Additional file 8: Figure S2). Taken together, our results suggest exercise induces acute response in piRNA expression, which is reverted after cessation of training. Expression of miRNA and tRNA, however, seems to be more stable with time.
In silico target prediction for piR-hsa-28,160 returned multiple copies of the ILF3/NF90-interacting RNA Small ILF3/NF90-associated RNA (SNAR), a regulator of the let7 family member let7a, a miRNA family well documented in inflammation, glucose metabolism [19][20][21][22][23] and, more recently, epigenetic inheritance [11]. Of the remaining piR-NAs, one has no predicted target, one targets FAM225A and B, two ncRNAs with no known function which are highly expressed in testis and one targets NSD1, a transcription factor that has been associated with Sotos syndrome [24], the symptoms of which include mild intellectual impairment and co-occur with autism [25]. The remaining two piRNAs target small nucleolar RNAs (snoRNAs). Both of those snoRNAs are predicted to regulate ribosomal RNA (rRNA) maturation. snoRNAs have been implicated in both cancer and lipotoxicity, and thought to exert miRNA-like function, notably for the regulation of alternative splicing [26,27]. It was previously shown that a loss of snoRNA leads to a variety of diseases, such as the Prader-Willi syndrome, which is characterized by morbid obesity and intellectual impairment [28]. Thus, it is possible that changes in the expression of sperm-borne snoRNAs after endurance training influences the developmental programming of the embryo and predispose/protect from disease. Altogether, our data demonstrate that sperm-borne sRNA content can be dynamically affected by a 6-week endurance training intervention. The functional relevance of the exerciseinduced sRNA differential expression in human sperm on the developmental programming of the embryo remains to be investigated.

Exercise training remodels methylation of brain genes
To investigate the effect of exercise training on DNA methylation, we performed Reduced Representation Bisulfite Sequencing (RRBS) on the pure sperm fractions collected at each time point. In total, 119,624 CpG clusters covering more than 1.4 million individual CpGs were interrogated by the RRBS protocol. Results were analyzed using a FDR 5 or 10% cut-off (Additional file 7: Table S6). With a FDR 10% cut-off, compared to the Untrained state, the Trained state returned 330 differentially methylated regions (DMRs), while 303 DMRs were detected 3 months after the last training session of the training program  Untrained, n = 6 Trained, n = 6 Detrained, n = 6 Ageyears 23 ( Analysis of median methylation showed that, while the clusters investigated followed a bi-modal distribution of low or high methylation, the DMRs were almost entirely located in a low methylated context (Fig. 2a). In both the Trained and Detrained state, DMRs were enriched at promoter regions over exon, intron and distal intergenic regions (Fig. 2b, c). Closer analysis of promoter regions revealed that DMRs were most preferably located in a 10 kb region centered on the transcription start site (TSS) (Fig. 2d, e). Collectively, these data show that DNA methylation changes in response to exercise occur at specific genomic elements, and strongly suggests a role in the control of transcription initiation.
To gain insight into the functional relevance of differential methylation after exercise, we performed a gene ontology analysis for the genes proximal to exercise DMRs at 5 or 10% FDR using g:Profiler [29]. Regardless of the FDR cut-off, at the Trained state, significant enrichment was found for the ontology terms related to the development of the central nervous system such as "neurogenesis", "neuron differentiation", and "axon guidance" (Additional file 9: Tables S7 and S8 and Fig. 3a). It is noteworthy that the gene ontology term "neurogenesis" contains all genes of the aforementioned terms, except the gene PPP1R13L. While some DMRs survived 3 months after training, gene ontology analysis of the Detrained state only returned the more generic term "regionalization" (Fig.  3a). Study design of published intervention studies (including from our group) did not establish if epigenetic variation in sperm is simply time-related or specifically triggered by the intervention itself [15,30]. Here, the loss of both ontology term specificity and number at the Detrained state infers that exercise triggers specific DNA methylation changes and that these changes are not caused by simple time-related effect.
Genes related to the development of the central nervous system were previously identified as genes escaping epigenetic reprogramming in human primordial germ cells and early embryogenesis [31]. While we did not find, using a hyper-geometric test, a statistically significant overlap between our exercise-responsive DMRs and regions escaping epigenetic reprogramming during human gametogenesis [31], we found seven genes proximal to our DMRs in common; SMCO1, PCDH10, FAM160A1, TRIML1, ABL1, SETX, and TSPY3 (see Table 2). Testing for overlap between our gene list and past reports investigating sperm DNA methylation changes in health and disease, we also did not detect specific enrichment between genes at proximity of our exercise DMRs and genes reported in sperm from obese [15], from an autism cohort [32] or after 3 months of endurance training [30]. These results may imply that while genes related to the development of the central nervous system are epigenetically susceptible to environmental influences in male gametes, each environmental insult triggers changes on a specific subset of genes. Alternatively, the difference in regions covered by each of these studies could explain the lack of overlap across cohorts. Nevertheless, our results strengthen that a subset of genes involved in the development of the central nervous system represents a genomic hotspot for epigenetic variation under environmental influences that has potential to convey reprogramming signals to the embryo.
To identify if epigenetically variable genes in sperm carry common sequence features, we searched for motifs surrounding the exercise DMRs found in a low methylation context using the MEME suite [33]. We discovered that two motifs cover the majority of DMRs (96% of DMRs contained at least one motif, and 64% contain both). Prediction of transcription factor binding returned putative binding site for the transcription factors EHF, MAZ, STAT1, and MNT (Fig. 4a), and SP2, SP3, SP4, and KLF16 for the respective motifs. Most importantly, a genomic scan for genes containing each respective motif returned that genes containing the motif of Fig. 4a were enriched for the term "nervous system development" (Fig. 4c) while motif of Fig. 4b did not (Fig. 4d). This observation indicates that we identified motifs located at proximity of epigenetically-variable genes and reinforces our finding that epigenetically variable genes in human sperm relate to the development of the central nervous system. This study contains a few limitations worth noting. The sample size is relatively small (n = 6 for sRNA and n = 12 for RRBS), due to a technical limitation when extracting DNA and RNA from the on single ejaculate, and in all three time-points (Untrained, Trained, and Detrained). However, the fact that each participant were assessed at all of the three time points considerably increases statistical power [34]. The lack of a control group accounting for time-related epigenetic alterations during the 6 week of exercise intervention can be seen as a limit to this study; however, others have reported no significant alterations in total methylation levels in spermatozoa of a control, nonexercising group, in a 3-month sampling interval [30].
In conclusion, our data provide evidence that endurance training remodels sRNA expression and the DNA methylation profile at close proximity of transcription start sites, specifically, at genes related to neurological development and function. These findings highlight the dynamic nature of the spermatozoal epigenome in response to environmental or lifestyle factors in humans.  Table 2 Selected DMRs at the trained state. DMRs which are found to be responsive to exercise and whose nearest gene has been found to escape epigenetic reprogramming, Chr is chromosome, Difference is the median of methylation differences observed within the cluster, upon exercise training Future studies will determine the role of environmentally induced epigenetic changes in sperm on the development of the embryo and phenotype of the offspring.

Subjects and sample collection
The study was approved by the Ethics Committee from the Capital Region of Denmark (reference H-1-2013-064) and informed consent was obtained from all participants. A portion of this cohort has previously been described [35]. Clinical characteristics of participants at all three time-points are presented in Table 1A and B. For the analysis of sRNA nine participants were analyzed at the untrained and trained, six participants were analyzed at the detrained state. Analysis of DNA methylation was performed on 12 participants at all three time points. All recruited participants were young, healthy, sedentary Caucasian males in their reproductive age (18-35 years). Exclusion criteria were regular smoking, alcohol consumption of > 14 units per week, presence of chronic or acute disease as well as daily intake of medicine. Men exercising more than twice per week, or who within the last 2 years had performed exercise on competitive levels, were excluded. VO 2max tests were performed by incremental exercise to volitional fatigue on an electromagnetically braked cycle ergometer (Monark Ergomedic 839E, Sweden) under fasting conditions. Pulmonary gas exchange was measured during the test breathby-breath with a gas analyzing system (Oxycon Pro, Jaeger, Germany). All participants were fecund and cleared for testicular and andrological abnormalities by inspection, anamnesis, and palpation. Microscopy was used to rule out spermatozoal morphological abnormalities and to count sperm concentration. Semen samples were delivered by masturbation after an overnight fast and a period of minimum 3 and maximum 7 days of ejaculative abstinence. Ejaculates were immediately stored at 37°C. Venous blood was drawn at fasting conditions.

Exercise intervention
Before the exercise intervention, all participants delivered a semen sample, blood sample and performed a VO 2max test (Untrained). The 6-week exercise program was performed by five weekly 1-h sessions for 6 weeks with supervised spinning classes by a certified instructor. The spinning classes were kept at an intensity of 70% of the participants' individual reserve capacity of their max pulse, determined by a max test performed before the exercise intervention. All participants participated in all sessions within the 6 weeks, and performance at each session was in accordance with the required intensity, as monitored individually by personal pulse monitors. After the 6-week exercise intervention, participants rested for 4 days before delivering the Trained ejaculate and performing VO 2 max test. Participants returned to their habitual Untrained exercise level for the following 3 months until their last session of semen sample delivery, blood sampling and VO 2 max were performed (Detrained). Compliance with the detraining program was verified by self-reporting at regular check-ups by investigator. Changes in clinical parameters were only analyzed on respective subset of participants included for DNA methylation or sRNA expression. Thus, sRNA was calculated based on data from six participants. Tests for the RRBS cohort was based on 12 participants. Clinical parameters were tested for normality using a Shapiro-Wilk test, p values for normally distributed parameters were calculated with a paired t-test, while non-normal variables were tested with a Wilcoxon signed rank test. All p values were corrected for family-wise error rate by the Holm-Bonferroni method.

Isolation of motile spermatozoa
A "swim-up" procedure was performed to exclude somatic cells and to isolate motile spermatozoa, which resulted in the isolation of the spermatozoa with the highest fertilization potential: 0.5 ml of semen was overlaid with 2 ml of medium (Earle's Balanced Salt Solution (Sigma) with 3.2 mg/ml Human Serum Albumin (Sigma) and 25 mM Hepes) in round-bottom tubes and incubated at 37°C at a 45°angle for 2 h. The upper fractions were pooled per ejaculate, and the spermatozoa counted by microscopy. The potential presence of somatic cells was inspected under a photon microscope.

sRNA-Seq and RRBS
Total RNA was isolated by the TRIzol® method (Life Technologies) from the sperm cells of six men before and after the 6-week exercise intervention, and after 3 months of detraining. The sRNA libraries were prepared using the NEBNext® Multiplex Small RNA Library Prep Set for Illumina (New England Biolabs), according to the manufacturer's instructions. Molecules of 20-50 nucleotides were separated by acrylamide gel electrophoresis, extracted, and sequenced on a HiSeq2500 Illumina instrument as 50 bp single-end reads, and processed by CASAVA 1.8.2.
The filter for including a feature in the sRNA analysis was dependent on the type of feature being analyzed. Due to the vast differences in sequencing depth of the different sRNA types, a single count per million (CPM) cutoff would have either included features with almost no reads or excluded features with many reads. Instead, a dynamic cutoff was used that depended on the total number of reads for that feature. The formula used to calculate the cutoff was: cutOff ¼ 5 Â median n reads 10 6 Where n reads is a vector containing the total number of reads assigned to this type of sRNA for each sample. This translates to a cut-off of 0.2 for miRNA, 0.5 for tRNA, 2.7 for piRNA and 2.9 for Repetitive Elements (Additional file 6: Tables S2-S4, Additional file 7: Table  S5). Features with more than the cut-off in 1/3 or more of the samples (eight or more samples) was included in the test for differential expression.
For DNA methylation analysis, genomic DNA was extracted from the sperm cells of 12 men before and after the exercise intervention, as well as after the detraining period, with the Nucleon™ BACC Genomic DNA Extraction Kit (GE Healthcare, Life Sciences). The protocol was modified for processing of sperm, according to the manufacturer's recommendations. Reduced Representation Bisulfite Sequencing libraries were constructed as previously described [15]. Briefly, 200 ng of genomic DNA was digested with 40 U of MspI enzymes (New England Biolabs) and ligated to TruSeq (Illumina) sequencing adaptors. Bisulfite conversion was conducted once with the EpiTect bisulfite kit (Qiagen) in accordance with the manufacturer's instructions, and the converted DNA was amplified by PCR and sequenced on a HiSeq2500 Illumina instrument as 50 bp single end reads, and processed by CASAVA 1.8.2.
Analysis of sequencing data sRNA reads were aligned to hg19 using the subread aligner using the recommended settings for miRNA mapping with the exception that only uniquely mapping reads were kept [35]. Unmapped reads were aligned first ribosomal sequences allowing for up to 10 mappings per read, reads that were still unmapped were aligned to miRNA-, piRNA-, tRNA-and repeatmasker-sequences