Molecular control of endurance training adaptation in male mouse skeletal muscle

A surprisingly low number of genes define the trained muscle

To study differences between untrained and endurance-trained muscles (in the present study, muscle always refers to quadriceps), mice were exercised by treadmill running on 5 d per week for 1 h. After 4 weeks, a significant improvement in running performance was observed (Extended Data Fig. 1a). A proteomic analysis also indicated a substantial remodelling of skeletal muscle (Fig. 1a and Supplementary Table 1). For example, proteins involved in mitochondrial respiration, lipid metabolism, oxygen transport or stress resilience are more abundant in trained than in untrained muscle (Fig. 1a,b, Extended Data Fig. 1b and Supplementary Table 2). In contrast, the levels of proteins linked to catabolic processes related to proteasomal degradation are mitigated by endurance training (Fig. 1a and Extended Data Fig. 1c), which, together with the induction of molecular chaperones, alludes to altered proteostasis. In contrast to the training-induced changes in protein abundance, training has very few effects on the steady-state phosphoproteome (differences in phosphorylation were observed in only 54 proteins). The corresponding proteins are mainly involved in cytoskeletal structure, sarcomere organization and muscle contraction, and are thus most probably linked to long-lasting alterations of contractility (Extended Data Fig. 1d and Supplementary Tables 1 and 2). According to prevailing models, the proteomic changes are brought about by a persistent modulation of gene expression with repeated exercise bouts¹³. To test this, we assessed the transcriptomic landscape of the trained muscle. Intriguingly, <2% of the detected genes were significantly changed in a trained muscle, with most transcripts being downregulated (Fig. 1c). Collectively, these genes define long-term cellular changes, for example, related to fibre-type switch, lipid metabolic processes or decreased inflammation (Fig. 1d and Supplementary Table 3). In line with these observations, Integrated System for Motif Activity Response Analysis (ISMARA)²¹ revealed a modulation of the predicted activity of only 22 transcription factors, for example, higher activity of the Esrrb_Esrra and lower activity of the Rela_Rel_Nfkb1 motifs (Fig. 1e and Supplementary Table 4). The genes that are altered in a trained muscle show only a small overlap with proteomic changes, suggesting that the proteome of a trained muscle is only to a small extent maintained transcriptionally. It is interesting that the subset of proteins with corresponding gene expression changes are predominantly involved in the lipid metabolic process (Extended Data Fig. 1e and Supplementary Table 5). Thus, most of the proteins that define the long-term plasticity of a trained muscle are not directly linked to a corresponding persistent transcriptional response.

a, All functional annotation clusters of up- (orange) and downregulated (blue) proteins in trained muscle with an enrichment score >2. ROS, reactive oxygen species. b, Examples of proteins involved in the response to stress in sedentary untrained (light grey) and unperturbed trained (dark grey) muscle (box plots display the median and the 25th to 75th percentiles and whiskers indicate the minimal and maximal values). c, Number of genes differentially expressed in unperturbed trained muscle (cut-off: FDR < 0.05; log₂(FC) ± 0.6). d, All functional annotation clusters of up- (orange) and downregulated (blue) genes in trained muscle with an enrichment score >2. e, Motifs of transcription factors from ISMARA that are among those with the highest and lowest activity. AU, arbitrary units. f, Number of genes after an acute bout of exhaustion exercise that are up- (orange) and downregulated (blue). g, Venn diagram of all genes that are changed in unperturbed trained muscle (orange is upregulated and blue downregulated) and those that are regulated after an acute bout of maximal exercise (light colour, dashed line). h, Heatmap of all genes differentially expressed in unperturbed trained muscle to visualize the overlap with acutely regulated genes using Euclidean distance hierarchical clustering for rows. The data are from five biological replicates and represent mean ± s.e.m. (if not otherwise indicated). Statistics of proteomics data were performed using empirical Bayes-moderated t-statistics as implemented in the R/Bioconductor limma package and for RNA-seq data with the CLC Genomics Workbench Software. Exact P values of proteomics data and z-scores of ISMARA data are displayed in Source data. The asterisk indicates difference to control (Ctrl; pre-exercise condition) if not otherwise indicated: in b, ^*P < 0.05, in e, ^*z-score > 1.96 (Extended Data Fig. 1 and Supplementary Tables 1–4).

Source data

These unexpected results raised the question of whether perturbations evoked by an acute maximal exercise bout activate transcriptional networks that encode the biological programmes observed in trained muscle. To test this hypothesis, untrained mice were exercised to exhaustion by treadmill running and the muscle transcriptome was assessed 0, 4, 6 and 8 h post-exhaustion. Similar to other studies, we found a large number of gene-regulatory events in this context of acute maximal exercise in untrained muscle, peaking 6 h post-exhaustion (Fig. 1f). A subset of these acute changes correlated with the accumulation of proteins in a trained muscle. These proteins were mostly upregulated and predominantly involved in mitochondrial respiration (Extended Data Fig. 1f and Supplementary Table 5). These genes are increased at the 8-h timepoint, suggesting that the induction of mitochondrial genes occurs several hours post-exercise and might even further increase at later timepoints. Intriguingly, the acutely regulated genes only poorly overlapped with persistent transcriptomic changes in trained muscle, because only 21% (57 of the 276 in total, with 44 of 107 up- and 13 of 169 downregulated) of the genes modulated in an unperturbed trained muscle are also regulated acutely in untrained muscle (Fig. 1g,h and Extended Data Fig. 1g). In fact, some of the genes exhibited the opposite regulation (Fig. 1h and Extended Data Fig. 1g), for example, reflected in transcripts related to inflammation (up acutely post-exercise, down in trained muscle).

Acute response to exercise is training status dependent

As the acute maximal exercise response in untrained muscle was not predictive of training adaptation, we next investigated the response of a trained muscle to an acute bout of maximal endurance exercise at the same four timepoints (Fig. 2a). Accordingly, mice that were trained for 4 weeks performed an exhaustive bout of treadmill running (Fig. 2a). Strikingly, the transcriptomic responses of untrained and trained muscle to an acute maximal endurance exercise bout were decisively different, qualitatively and quantitatively, the latter in terms of both amplitude (extent of change, that is, attenuated or exacerbated) and phase (temporal regulation, that is, induction of gene expression at different timepoints) (Fig. 2b,c). First, less than half of the upregulated genes overlapped between these two conditions and an even smaller proportion of the downregulated transcripts, of which a greater number were altered in the trained condition (Fig. 2c). Functionally, many of the acutely regulated genes in untrained muscle cluster with regulation of transcription and various aspects of stress response, damage, axon guidance and extracellular matrix (ECM) organization (Fig. 2d, Extended Data Fig. 2a and Supplementary Table 3). Strikingly, in regard to ECM remodelling and axon guidance, the functional prediction of the acute response of trained muscle was diametrically opposite to that of the untrained muscle (Fig. 2d,e, Extended Data Fig. 2a–d and Supplementary Table 3). ISMARA confirmed the substantial regulatory diversification (Extended Data Fig. 3a–e and Supplementary Table 4). Although approximately 35–43% of the motifs are specific to the training status (Extended Data Fig. 3b), many of the common motifs (n = 77) show altered trajectories and/or amplitudes (Extended Data Fig. 3c–e). In fact, 18 of the 77 motifs in the overlap significantly differed in amplitude. Moreover, in an additional 22 of the 77 motifs, the activity profiles point in the opposite direction. For example, the Wrnip1_Mta3_Rcor1 motif activity is higher in untrained and lower in trained muscle and, based on the association with collagen formation, could contribute to the distinct patterns of ECM remodeling (Fig. 2f). Thus, of the 178 predicted transcription factor motif activities after an acute bout of maximal exercise (in untrained and trained), only 21% (37 out of 178 motifs) exhibited a shared direction and amplitude, implying a strong regulatory diversification between these two conditions.

a, Schematic representation of the setup (illustration was created using BioRender.com with permission). b, Number of genes differentially expressed immediately (0 h), 4, 6 and 8 h after an acute bout of exhaustion exercise (cut-off: FDR < 0.05; log₂(FC) ± 0.6) in untrained and trained muscle. c, Venn diagram of all significantly up- (orange) and downregulated (blue) genes (all timepoints merged) in untrained (light colour, dashed line) and trained (dark colour, solid line) muscle. d, Dot plot of all functional annotation clusters of up- (orange) and downregulated (blue) genes in untrained and trained muscle post-exercise, as well as unperturbed trained muscle with an enrichment score >2. e, Examples of gene trajectories in untrained (light grey) and trained (dark grey) muscle involved in axon guidance. f, Motif activities from ISMARA and expression changes of a predicted target gene that show an opposite regulation in untrained and trained muscle. The data are from five biological replicates and present mean ± s.e.m. Statistics of RNA-seq data were performed using the CLC Genomics Workbench Software. Exact FDR values of RNA-seq data and z-scores of ISMARA data are displayed in Source data. The asterisk indicates difference to Ctrl (pre-exercise condition): ^*P < 0.05 (for motif activity: ^*z-score > 1.96); ^**P < 0.01; ^***P < 0.001 (Extended Data Figs. 2–4 and Supplementary Tables 3 and 4).

Source data

Many of the predicted functions, including a modulation of ECM remodeling, axon guidance and inflammation, could originate from non-myocytes in muscle tissue. Therefore, we performed cellular deconvolution of the bulk results with published single-cell RNA-sequencing (scRNA-seq) and single-nucleus RNA-sequencing (snRNA-seq) data of untrained muscle (Extended Data Fig. 4a)^22,23. These analyses imply a surprisingly detailed specification of gene expression across different cell types (Extended Data Fig. 4b,c). For example, the cellular origin of ECM remodelling genes could mainly be fibroadipogenic progenitors in untrained muscle, complemented by tenocytes in trained muscle. This type of analysis is only predictive and cannot differentiate between changes in cell composition (that is, reduction in the number of tenocytes that could result in a relative downregulation of tenocyte-specific genes) or selective repression of the specific genes in stable populations (that is, lower number of transcripts per cell). However, regardless of the precise mechanism, such cell type-specific responses presumably result in the correspondingly distinct outcomes for ECM remodeling, axon guidance and potentially other functions after an acute endurance exercise bout in untrained compared with trained muscle. Future studies should therefore dissect the acute exercise and chronic training response of muscle tissue at the single-cell and single-nucleus levels.

Besides qualitative differences in functional gene clusters between untrained and trained muscle, quantitative specification was observed for some common processes induced on an acute perturbation, for example, the regulation of transcription or the response to heat stress in a training status-dependent manner (Fig. 2d). For example, the modulation of the regulatory axis serum response factor–early growth response 1 indicates mitigation of the immediate early stress response in trained muscle (Fig. 3a and Extended Data Fig. 3e). Inversely, the expression of other transcriptional regulators such as PGC-1α is exacerbated, highlighting the specificity of gene-regulatory events in untrained and trained muscle (Fig. 3b). Intriguingly, contradicting the suggested attenuated response of a trained muscle^{3,10,11,13,14,15}, the maximal amplitude of peak expression of most commonly regulated genes is very similar in untrained and trained muscle, specifically 73% of the shared upregulated and almost 90% of the shared downregulated genes (Fig. 3c and Extended Data Fig. 5a). However, a marked shift in the temporal trajectories was observed. For example, a substantially higher number of commonly regulated genes are already elevated at 0 h in trained muscle (Fig. 3d and Extended Data Fig. 5a). Furthermore, peak expression is also shifted towards the 0 h timepoint in all, as well as just the subset of the commonly regulated genes (Extended Data Fig. 5a,b). In fact, almost half of all upregulated genes in trained muscle peak at 0 h, whereas this applies to only ~20% of the upregulated genes in untrained muscle, where most peak after 6 h (Extended Data Fig. 5b). Overall, as opposed to the model of general attenuation of gene expression with training habituation^{3,10,11,13,14,15}, our results suggest a much more complex picture, with noteworthy occurrence of all scenarios: attenuation, exacerbation and selective expression changes in untrained or trained muscle after an acute maximal exercise bout and, probably as important, a temporal shift in gene expression (Fig. 3e).

a, Example of a possible transcriptional cascade including a top predicted transcription factor by ISMARA and one of the downstream targets (gene expression and motif activities). b, Example of a transcriptional regulator with distinct trajectories in untrained (light grey) and trained (dark grey) muscle. c, Proportion of commonly regulated genes with the same maximal amplitude (grey), higher amplitude in untrained muscle (light colour) or higher amplitude in trained muscle (dark colour). d, Visualization of the temporal trajectories of the commonly regulated genes (overlap from Fig. 2c) in untrained (light colour) and trained (dark colour) muscle (orange is upregulated and blue downregulated). e, Examples of different gene trajectories in untrained and trained muscle after an acute maximal exercise bout representing the different training status-specific transcriptional scenarios. f, Number of DMRs in an unperturbed trained muscle (hypermethylated is shown as a solid bar and hypomethylated as an open bar) compared with untrained sedentary WT muscle. g, Bar Venn diagram of DMRs of an unperturbed trained muscle (white) and DEGs after acute maximal exercise in trained muscle (dark grey) and the functional annotation clusters of the overlap (light grey, n = 120) with an enrichment score >2. h, Example of a transcription factor that is differentially methylated in trained muscle and more highly expressed after exercise in trained compared with untrained muscle. The data are from five biological replicates and represent mean ± s.e.m. Statistics of RNA-seq data were performed using the CLC Genomics Workbench Software. Exact FDR values of RNA-seq data and z-scores of ISMARA data are displayed in Source data. Differences in relative expression changes presented in d were calculated using a two-tailed Student’s t-test. The asterisk indicates difference to Ctrl (pre-exercise condition): ^*P < 0.05 (for motif activity: ^*z-score > 1.96); ^**P < 0.01; ^***P < 0.001 (Extended Data Fig. 5 and Supplementary Tables 4, 6 and 7).

Source data

Priming of regulatory genes by DNA methylation changes

The divergent specification and more rapid induction of gene expression in trained muscle suggest a priming to react to recurrent perturbations. Epigenetic changes, which can induce such a poised state, have been reported in training adaptation^17,19. To test this, we performed reduced representation bisulphite sequencing (RRBS) to catalogue DNA methylation events in trained muscle (Fig. 3f and Supplementary Table 6). A very low number of differentially methylated regions (DMRs) are associated with gene expression changes in unperturbed trained muscle (n = 9). Intriguingly, a subset of these DMRs are in the immediate genomic vicinity of a group of genes (120 out of 2,387) that are regulated after an acute maximal exercise bout in trained muscle (Fig. 3g). These epigenetic modulations thus might not be associated with the persistent gene expression pattern in unperturbed trained muscle, but are more likely to contribute to a priming of transcriptional regulation to an acute bout of exercise. It is interesting that these genes enrich in functions related to the regulation of transcription, Wnt signalling and axon guidance signalling effectors (Fig. 3g,h, Extended Data Fig. 5c–e and Supplementary Table 7). For example, the induction of nuclear receptor 4A3 (Nr4a3), which is associated with DMRs in trained muscle, is not only greatly exacerbated in trained compared with untrained muscle, but also displays a phase shift towards a peak immediately post-exercise (Fig. 3h). Thus, epigenetic modifications could contribute to the different gene expression of a trained muscle to an acute perturbation, primarily affecting regulatory genes, with subsequent downstream consequences independent of DNA methylation changes.

Overall, very distinct transcriptomes were found in unperturbed trained, acutely exercised untrained and acutely exercise trained muscle. Although the correlation between the persistent proteomic and transcriptome changes in trained muscle is low, better results are achieved when integrating all gene expression changes, including those observed in acutely exercised untrained and trained animals. Collectively, these transcriptomic events correlate with the trained proteome up to 53% for upregulated and 30% for downregulated proteins for which transcript data are available (Extended Data Fig. 5f), highlighting the importance of broad comparisons between transcriptomes and proteomes²⁴. So far, it is unclear whether the rest of the trained proteome (47% of upregulated proteins, which, for example, cluster in the tricarboxylic acid (TCA) cycle (Supplementary Table 5), and 70% of downregulated proteins) is evoked by post-translational processes, or linked to transcriptional events at other timepoints post-exhaustion and/or intermediate exercise bouts not included in the present study. Overall, these findings allude to a complex regulatory network, including transient and persistent transcriptional as well as post-translational events, mediating long-term proteomic adaptations.

PGC-1α is indispensable for normal training adaptation

Notably, many transcriptional regulators that are engaged strongly and early after acute maximal exercise exhibit a diversification between the first bout (in an untrained muscle) and after a period of training, including PGC-1α (Fig. 3b). This coregulator protein has been implicated in the acute response by integrating various signalling pathways and subsequently affecting the activity of numerous transcription factors, thereby controlling a complex transcriptional network²⁵. Our observation, recapitulating previous results in human muscle²⁰, of a quantitative difference of PGC-1α on exercise in trained compared with untrained muscle, would indicate that PGC-1α not only controls an acute stress response, but also might affect the transcriptome of exercised muscle in the trained state. Nevertheless, gene expression changes of this regulatory nexus are only transient and not preserved in unperturbed trained muscle. Thus, the relevance of adequate regulation and function of PGC-1α in long-term training adaptations has been questioned and, at least in part, conflicting findings have been reported^26,27,28,29. To obtain comprehensive information on muscle PGC-1α in training, we therefore repeated the exercise study with muscle-specific PGC-1α knockout (mKO) mice (Fig. 4a). In agreement with previous work³⁰, mKO mice exhibit a reduced endurance capacity, running approximately 40% less than wild-type (WT) controls (Fig. 4b). Despite these limitations, the PGC-1α loss-of-function animals substantially improved maximal performance after 4 weeks of training, in relative and absolute terms, reaching the levels of untrained WT mice, thus still significantly less than the trained WT counterparts (Fig. 4b). Importantly, blood lactate levels post-exercise were higher in mKO compared with WT animals, which implies a higher reliance on anaerobic processes to generate ATP (Extended Data Fig. 6a). Moreover, maximal oxygen consumption (VO_2max) failed to improve in mKOs (Fig. 4c), alluding to an alternative adaptation of endurance capacity in these mice. Such an abnormal endurance training adaptation was substantiated by the proteomic analysis of trained muscle of WT and mKO mice (Extended Data Fig. 6b and Supplementary Table 1). First, many of the training-regulated proteins involved in mitochondrial respiration, the lipid metabolic process and the TCA cycle are already found at lower levels in sedentary mKO compared with sedentary WT animals (Fig. 4d–f, Extended Data Fig. 6c,d and Supplementary Tables 1 and 2). Although many of these proteins can be modulated in mKO mice by training, most do not even reach levels normally seen in sedentary WT muscle. Similar to trained WT muscle, relatively few proteins show altered phosphorylation levels (103 proteins; Supplementary Table 1). These proteins are predominantly involved in sarcomere organization and muscle contraction (Extended Data Fig. 6e and Supplementary Table 2).

a, Schematic representation of the setup (illustration was created using BioRender.com with permission). b, Performance of untrained (light colour) and trained (dark colour) WT (grey) and mKO (blue) animals (WT-trained versus WT-untrained: mean difference (MD) = 1,242, 95% confidence interval (CI) = 946.1–1,539, P < 0.0001; mKO-trained versus mKO-untrained: MD = 500.9, 95% CI = 204.5–797.3, P = 0.0002; mKO-untrained versus WT-untrained: MD = −478.9, 95% CI = −775.3 to −182.5, P = 0.0003; and mKO-trained versus WT-trained: MD = −1,220, 95% CI = −1,517 to −924.1, P < 0.0001) and relative improvement of WT and mKO animals after 4 weeks of progressive treadmill training (MD = −0.3368, 95% CI = −0.6574 to −0.01625, P = 0.0399) (n = 25 biological replicates per group). c, Changes in VO_2max before (light colour) and after (dark colour) training (WT post-training versus WT pre-training: MD = 6.833, 95% CI = 0.5067–13.16, P = 0.350; mKO post-training versus mKO pre-training: MD = 3.667, 95% CI = −2.66 to 9.993, P = 0.2926; mKO-untrained versus WT-untrained: MD = −11.00, 95% CI = −17.87 to −4.132, P = 0.0051; and mKO-trained versus WT-trained: MD = −14.17, 95% CI = −25.67 to −2.659, P = 0.0207) (n = 6 biological replicates per group). d, Dot plot of all functional annotation clusters of significantly altered proteins with an enrichment score >2. e,f, Examples of proteins involved in mitochondrial respiration (e) and TCA cycle (f) in WT-trained (grey; n = 5), mTG-untrained (pink; n = 5), mKO-untrained (dark blue; n = 6) and mKO-trained (blue; n = 5). Values are expressed relative to untrained WT sedentary controls (n = 5). Statistics of proteomics data were performed using empirical Bayes-moderated t-statistics as implemented in the R/Bioconductor limma package. Exact P values are displayed in Source data. To assess differences between untrained and trained animals and between genotypes, two-way ANOVA followed by Šídák’s multiple-comparison test (b and c) or two-tailed Student’s t-test was performed (relative improvement in b and c). The asterisk indicates difference to Ctrl (pre-exercise condition) if not otherwise indicated; hashtag indicates differences to the same condition in WTs: ^*/#P < 0.05; ^**/##P < 0.01; ^***/###P < 0.001 (Extended Data Fig. 6 and Supplementary Tables 1 and 2).

Source data

Next, we investigated how the altered phenotypic and proteomic adaptations of trained muscles lacking PGC-1α are reflected in the transcriptome. In sedentary mice, the lack of PGC-1α causes a pronounced transcriptional suppression of genes involved in the lipid metabolic process (Extended Data Fig. 6f and Supplementary Table 8). Then, even when compared with the already constrained transcriptional changes in unperturbed trained WT muscle, far fewer genes were regulated by training in the absence of PGC-1α (Fig. 5a). Of note, 90% of all upregulated and 87% of all downregulated, transcriptional events were dependent on the presence of PGC-1α in WT muscle (Fig. 5a). Many of these genes encode proteins of lipid metabolism and the fast-to-slow muscle fibre transition (Fig. 5b and Supplementary Table 8). Even more impressive, in the unperturbed trained muscle, almost all (91%) transcription factor motif activities were affected by the loss of function of PGC-1α (Extended Data Fig. 6g and Supplementary Table 4). Most notably, ISMARA analysis revealed that the significant training-linked increase in Esrrb_Esrra motif activity, a binding site for the oestrogen-related receptor-α, was completely blunted in mKO mice (Fig. 5c). In line with this, the activity of this motif was highly increased in muscle-specific PGC-1α gain-of-function transgenic mice (mTG) (Fig. 5c). The phenotypic, proteomic and transcriptomic data thus strongly indicate that PGC-1α is indispensable for a normal, physiological training response, even though this factor is only transiently engaged in acute exercise bouts.

a, Bar Venn diagram of the genes altered in unperturbed trained WT (grey) and mKO (blue) muscle. b, All functional annotation clusters of genes that are only up- (orange) and downregulated (blue) in trained muscle of WT animals (up: n = 96; down: n = 147) with an enrichment score >2. c, Motif of the transcription factors from ISMARA with the most significant activity change in trained WT animals and the comparison of the activity in trained mKO muscle (left blue), gain-of-function model (sedentary muscle-specific PGC-1α transgenics (mTG), purple) and loss-of-function model (sedentary mKO, dark blue). d, Number of genes that are up- and downregulated 0, 4, 6 and 8 h after an acute maximal exercise bout in untrained WT (light grey), trained WT (dark grey), untrained mKO (light blue) and trained mKO (dark blue) animals. e,f, Examples of gene trajectories with the peak expression immediately post-exercise (e) or at a later time (f) in untrained WT and mKO animals. g, Venn diagrams of all up- and downregulated genes after an acute bout of exercise in untrained WT (light grey) and mKO (light blue) mice. h, All functional annotation clusters of up- (orange) and downregulated (blue) genes that are regulated only in untrained WT mice (745 genes up- and 314 genes downregulated) with an enrichment score >2. i, Examples of genes involved in ECM organization, microglial cell proliferation and Wnt signalling that are regulated only in WT muscle. j, Prediction of the activity of a motif using ISMARA that is changed only in WT muscle and might be involved in the regulation of ECM-related genes. The data are from five biological replicates and represent mean ± s.e.m (if not otherwise stated). Statistics of RNA-seq data were performed using the CLC Genomics Workbench Software. Exact FDR values of RNA-seq data and z-scores of ISMARA data are displayed in Source data. The asterisk indicates difference to Ctrl (pre-exercise condition): ^*P < 0.05 (for motif activity: ^*z-score > 1.96); ^**P < 0.01; ^***P < 0.001 (Extended Data Fig. 7 and Supplementary Tables 4 and 8).

Source data

Normal transcriptional exercise response depends on PGC-1α

Next, we assessed whether the marked differences in the long-term adaptation to training in muscles lacking PGC-1α are reflected in the response to acute maximal exercise (Fig. 4a). First, immediately post-exercise (0 h), the response of WT and mKO animals is relatively similar in terms of the number of DEGs as well as the amplitude of the gene expression (Fig. 5d,e). However, in untrained mKO muscle, a massive blunting of transcriptional induction at the later timepoints (4–8 h post-exercise) was found, so subsequent to the physiological PGC-1α elevation in WT muscle (Fig. 5d,f). Taken together, substantial qualitative differences in gene expression emerged, with 56% (745 out of 1,325) of all upregulated, and 65% (314 out of 482) of all downregulated genes being dependent on the presence of PGC-1α (Fig. 5g). Functional annotation revealed that many of these genes encode proteins involved in ECM organization, signal transduction, cell cycle/proliferation and other processes (Fig. 5h,i). In untrained mKO muscle, the transcriptomic response to acute maximal exercise was characterized by a modulation of genes related to inflammation and an inverse regulation of genes involved in axon guidance (up in WT, down in mKO) (Extended Data Fig. 7a). Finally, the divergent transcriptomic response was linked to a substantial regulatory rewiring: 52% (62 out of 120) predicted motif activities associated with the acute maximal exercise response of untrained WT muscle were lost in mKOs (for example, Wrnip1_Mta3_Rco1, linked to ECM remodelling) (Fig. 5j, Extended Data Fig. 7b and Supplementary Table 4).

Next, we compared the acute exercise response of trained mKO muscle with trained WT muscle. First, the temporal shift of gene expression towards 0 h was observed in both WT and mKO muscles (Fig. 5d). Second, even though only 39% (487 out of 1,254) of the upregulated genes were PGC-1α dependent in trained muscle, the proportion of commonly PGC-1α-dependent, downregulated genes (62%, 755 out of 1,219) remained similar to that found in untrained muscle (Fig. 6a). Functionally, these genes encode proteins involved in transcription, and metabolism of lipids and carbohydrates, as well as ECM remodelling (Fig. 6b and Supplementary Table 8). Intriguingly, the increase and decrease in ECM remodelling in acute maximal exercise of untrained and trained WT muscles, respectively, both seem to be dependent on the presence of this coregulator (Figs. 5h and 6b–d and Extended Data Fig. 7a,c). Of note, there is a prominent difference in the number of downregulated genes immediately post-exercise (0 h) between WT and mKO muscles (Fig. 5d). The genes that are reduced only in acutely exercised, trained WT muscle, and not in the corresponding mKO counterpart, were associated with inflammation (Extended Data Fig. 7d and Supplementary Table 8), in line with the higher activity-dependent muscle damage and inflammation that have previously been reported in mKO muscles³⁰. In the trained muscle, acute exercise exhibited 39% (52 out of 135) of predicted transcription factor activities to be absent in the mKO muscles, for example, that of Irf3 and Irf2_Irf1_Irf8_Irf9_Irf7, regulating inflammation-related genes (Fig. 6e, Extended Data Fig. 7e and Supplementary Table 4).

a, Venn diagrams of all up- and downregulated genes after an acute bout of maximal exercise in trained WT (dark grey) and mKO (dark blue) mice. b, All functional annotation clusters of up- (orange) and downregulated (blue) genes that are regulated only in trained WT mice (487 genes up- and 755 genes downregulated) with an enrichment score >2. c, Dot plot of all functional annotations clusters of up- (orange) and downregulated (blue) genes after an acute bout of maximal exercise in untrained and trained WT and mKO animals. d, Examples of genes involved in ECM organization in trained WT (grey) and mKO (blue) mice. e, Prediction of the activity of motifs using ISMARA that are changed only in WT muscle and linked to inflammation. f, Number of DMRs in trained mKO compared with untrained mKO muscle (hypermethylated is shown as a solid bar and hypomethylated as an open bar). g,h, Number of hyper- (solid bars) and hypomethylated (open bars) regions 0 and 4 h after exhaustion in untrained WT (g) and untrained mKO (h) animals compared with untrained sedentary animals of the respective genotype. i, All functional annotation clusters of genes that are differentially methylated and transcriptionally regulated after an acute bout of exercise in untrained WT (grey) and mKO (blue) mice. The data are from five biological replicates. Statistics of RNA-seq data were performed using the CLC Genomics Workbench Software. Exact FDR values of RNA-seq data and z-scores of ISMARA data are displayed in Source data. The asterisk indicates difference to control animals of the respective genotype: ^*P < 0.05 (for motif activity: ^*z-score > 1.96). (Extended Data Figs. 7 and 8 and Supplementary Tables 3, 4 and 6–8).

Source data

PGC-1α controls exercise-linked DNA methylation events

In WT muscle, we have associated the transcriptomic acute exercise response of a trained muscle with epigenetic modulations of the unperturbed trained muscle (Fig. 3f–h). Therefore, we next investigated whether DNA (de-)methylation events are linked to the massive transcriptional differences in the acutely exercised, trained mKO animals. In unperturbed trained muscle, a markedly higher proportion of hypermethylated regions was found, with little overlap with DMRs in WT quadriceps that are characterized by more hypomethylation (Fig. 6f, Extended Data Fig. 8a and Supplementary Table 6). Similarly, the DEGs after acute maximal exercise associated with DMRs of trained muscle exhibited only a small overlap between the genotypes (Extended Data Fig. 8b,c). Nevertheless, many of these genes in the mKO animals partitioned to regulation of transcription, functionally similar to the results in WT animals (Extended Data Fig. 8d,e and Supplementary Table 7). Based on the largely different transcriptome of trained muscle, a divergence in DMRs might not be unexpected. However, it was surprising that absence of muscle PGC-1α also substantially altered transient epigenetic modulations in untrained muscle after an acute maximal exercise bout. In both phenotypes, little overlap exists between these transient DNA (de-)methylation events in an acute maximal exercise bout and the persistent epigenetic adaptations in unperturbed trained muscle (Extended Data Fig. 8f and Supplementary Table 6). However, although the absolute number of events after an acute maximal exercise bout at 0 h between untrained WT and mKO muscle is comparable (483 in WT, 475 in mKO), most of these DMR–gene associations are distinct (only 109 are the same). Moreover, the absolute number of DMRs in mKO muscles 4 h post-exercise is dramatically smaller than that in WT muscles (646 in WT, 80 in mKO), again with little commonality (Fig. 6g,h, Extended Data Fig. 8g,h and Supplementary Table 6). Despite all these differences between WT and mKO muscles, whenever differentially affected DMRs could be associated with corresponding genes, a strong functional cluster ‘transcription’ emerged in either phenotype, indicating that these transient DNA methylation events are closely linked to acute transcriptional regulation (Fig. 6i and Supplementary Table 7). Collectively, these results imply that PGC-1α is directly involved in the regulation of DNA methylation associated with gene expression. To further test this hypothesis, we analysed the epigenetic, transcriptomic and proteomic changes elicited in a muscle-specific PGC-1α gain-of-function model. Indeed, a substantial number of DMRs were detected in mTGs. Similar to trained WT muscle, and mirroring the outcome in mKO animals, DMRs in mTGs skewed towards hypomethylation (Extended Data Fig. 8i and Supplementary Table 6). However, the overlap between DMRs of trained WT and sedentary mTG mice was very small and only 2.8% of the transcriptionally regulated genes could be associated with DMRs (Extended Data Fig. 8j,k). In line with previous observations³¹, the transcriptome of mTGs differs substantially from the chronically and acutely training- and exercise-regulated genes in WT muscle (Extended Data Fig. 8l). A better functional representation of training adaptation is, however, provided by the mTG proteome, in which a strong accumulation of mitochondrial proteins, including members of the TCA cycle and respiratory chain, lipid metabolism and a depletion of inflammation and proteasomal catabolic processes, recapitulates many of the changes observed in trained WT muscle (Extended Data Fig. 8m and Supplementary Tables 1 and 2).