Transcriptional memory in skeletal muscle. Don't forget (to) exercise

Transcriptional memory describes an ancient and highly conserved form of cellular learning that enables cells to benefit from recent experience by retaining a mitotically inheritable but reversible memory of the initial transcriptional response when encountering an environmental or physiological stimulus. Herein, we will review recent progress made in the understanding of how cells can make use of diverse constituents of the epigenetic toolbox to retain a transcriptional memory of past states and perturbations. Specifically, we will outline how these mechanisms will help to improve our understanding of skeletal muscle plasticity in health and disease. We describe the epigenetic road map that allows skeletal muscle fibers to navigate through training‐induced adaptation processes, and how epigenetic memory marks can preserve an autobiographical history of lifestyle behavior changes, pathological challenges, and aging. We will further consider some key findings in the field of exercise epigenomics to emphasize major challenges when interpreting dynamic changes in the chromatin landscape in response to acute exercise and training.

The functional interplay of regulatory circuits that orchestrate and transduce this information to the level of gene expression is only partly understood (Hoppeler, 2016). Likewise, there is still no comprehensive understanding of how single-exercise stimuli are memorized, and when and how myofibrillar "decision-making" for stable changes in adaptive gene expression pattern is achieved.
A better understanding of the molecular mechanisms that govern skeletal muscle plasticity, however, is necessary to provide a fundamental basis for the development of individually tailored exercise training programs, not only to improve athletic performance but also to establish exercise-based prevention, rehabilitation, and therapy concepts to counteract deregulated skeletal muscle physiology in conditions of muscle wasting and chronic disease. In recent years, innovative methodological approaches in the fields of transcriptomics, proteomics and metabolomics have been exploited to study large-scale biological processes that participate in metabolic and structural remodeling of human skeletal muscle, both in response to acute exercise and long-term training stimulation (Catoire et al., 2012;Hoffman et al., 2015;Lindholm et al., 2016;Nader et al., 2014;Neubauer et al., 2014;Raue et al., 2012;Schild et al., 2015;Vissing & Schjerling, 2014). In combination, such studies have revealed a complex interplay between a myriad of signaling cascades that are physically and functionally interconnected to build up a continuous framework where each step from gene transcription to active protein feeds back to, depends on, and is influenced by regulatory input from diverse intracellular and extracellular pathways. If there is one lesson to learn from these high-throughput biological data, it is that gene expression networks in skeletal muscle adaptation processes cannot be sufficiently described by traditional views that focus on a linear series of stimulus-response coupling events, solely governed and shaped at the level of transcription factor (TF) activity.
In recent years, the way in which we view gene expression has changed significantly. Groundbreaking studies on the role of chromatin structure in gene regulation have focused attention on so-called "epigenetic mechanisms" that are considered to operate "above" the level of the genetic code by marshalling access of the transcriptional apparatus to the DNA template, and by providing a mitotically heritable memory of cell lineage-specific transcriptional programs as well as of environmentally acquired traits. Epigenetic marks comprise a variety of stable but reversible chemical modifications to DNA and its associated proteins that influence and/or are influenced by chromatin structure and gene expression (Allis & Jenuwein, 2016). Key players in epigenetic control include diverse histone-modifying enzymes, chromatin remodeling proteins, DNA methyltransferases, and an expanding universe of noncoding RNAs.
In this review, we will focus on how cells can make use of specific constituents of the epigenetic machinery to retain a memory of past states and perturbations. Specifically, we will outline how these mechanisms could be exploited by skeletal muscle fibers to store information from previous experience of exercise To adaptively shape transcriptional responses in subsequent exercise bouts (transcriptional memory), thereby providing a flexible transcriptional framework for skeletal muscle plasticity. We will further consider some key findings in the field of exercise epigenomics to emphasize major challenges when interpreting dynamic changes in the chromatin landscape in response to acute exercise and training.

| TRANSCRIPTIONAL MEMORY ENABLES LEARNING FROM EXPERIENCE
Intuitively, we link exercise training-induced skeletal muscle adaptation to beneficial effects, like enhanced performance, overall wellbeing, and healthy longevity. Viewed from a biochemical perspective, skeletal muscle adaptation is a cost-intensive process that consumes considerable amounts of energy and resources. Biological systems function on a maximum economy basis. Technically speaking, the metabolic costs that arise from functional, structural, and metabolic remodeling of skeletal muscle groups must somehow get weighed against the potential benefits to better cope with recurrent exercise. To get the optimal trade-off, skeletal muscle biochemical networks have to integrate mode, duration, and intensity of single-exercise bouts, and somehow have to develop a strategy based on the frequency of recurrent exercise regimes. Thus, skeletal muscle adaptation processes must at least partly rely on a system of adaptive thresholds to integrate the experience of previous bouts of exercise into its decision-making networks, or, simply spoken, skeletal muscle must retain a "memory" of its exercise history.
Retaining an epigenetic memory of past incidents is an efficient strategy of eukaryotic organisms to quantitatively and qualitatively adapt transcriptional responses after repeated encounters with the initial stimulus (D'Urso & Brickner, 2017). This "transcriptional memory" was initially described in yeast cells that, once fed with galactose instead of glucose, can remember this metabolic challenge and reactivate genes for galactose conversion more rapidly when encountering galactose again (Kundu, Horn, & Peterson, 2007). Reciprocally, genes that transiently become repressed during the first encounter with galactose show stronger and more rapid suppression in subsequent rounds of carbonsource shifts (B. B. Lee, Choi, et al., 2018). Intriguingly, these acquired traits become transmitted across cell generations and are maintained through multiple cell divisions. Similar mechanisms are exploited by plants to "remember" abiotic and biotic stresses, as evidenced by modified transcriptional responses, enhanced tolerance, or improved defense capacity when facing the same stressor again (Crisp, Ganguly, Eichten, Borevitz, & Pogson, 2016). In view of the fact that many epigenetic mechanisms and strategies are highly conserved between species, it is hardly surprising that also mammalian cells make use of transcriptional/epigenetic memory to recall past incidents and to use this stored knowledge for flexible adaptation of gene expression programs.
Recent studies have provided evidence that transcriptional memory plays an important role in mammalian immunity, allowing innate immune cells to remember and to adaptively respond when challenged by repeated exposure to pathogenic and inflammatory stimuli, a phenomenon known as "trained innate immunity" (Dunn, McCuaig, Tu, Hardy, & Rao, 2015;Hamon & Quintin, 2016). The concept of "trained immunity" lead to a paradigm shift in immunology, since innate immune cells have traditionally been thought incapable to adapt or to preserve an active memory of prior exposure. Importantly, trained immunity responses have BEITER ET AL.

| 5477
further been shown to become circumstantially readjusted to accommodate changing environmental conditions and challenges (Kamada et al., 2018).

As comprehensively reviewed by Adam Sharples and colleagues,
there is ample evidence that also SCs can retain a mitotically heritable "epi"-memory of prior physical activity, metabolic challenges, as well as of prior maladaptive states (Sharples, Stewart, & Seaborne, 2016). SCs are indispensable to promote muscle repair and regeneration, but their actual role(s) in adult skeletal muscle homeostasis and exercise adaptation is not comprehensively understood. It is commonly acknowledged that exercise-induced hypertrophy of skeletal muscle fibers is initiated/accompanied by SC activation and proliferation, and/or acquisition of extra fiber nuclei (myonuclei) from SC fusion (Blaauw & Reggiani, 2014;Snijders et al., 2015). Experimental studies in rodents revealed that muscle fibers can acquire myonuclei from SCs during a period of resistance exercise or overload, and these extra nuclei can be maintained even during prolonged periods of detraining and atrophy, providing a long-lasting "muscle memory" that confers improved hypertrophic capacity when muscle mass and metabolic properties are reestablished (Bruusgaard, Johansen, Egner, Rana, & Gundersen, 2010;. It should, however, be mentioned that, in experimental models, hypertrophic responses in skeletal muscle have reportedly been observed without myonuclear accretion (Blaauw et al., 2009;McCarthy et al., 2011). Similarly, a recent human study found no increase in myonuclear number that would accompany gains in muscle force and mass in the course of 10 weeks of leg strength training (Psilander et al., 2019). Further complication arises from a recent murine study that reported traininginduced increases in myonuclear density but found no evidence for a retention of these newly recruited myonuclei that would last over a subsequent period of detraining (Dungan et al., 2019). Therefore it does not surprise that the question as to the overall contribution of SC fusion in resistance exercise adaptation is a matter of ongoing debate (Gundersen, 2016;Murach et al., 2018).
Far less questionable is the general notion that skeletal muscle is somehow capable of remembering hypertrophic exercise, even over prolonged periods of subsequent inactivity. One session of maximal eccentric exercise has been shown sufficient to imprint a "lasting impression" in the skeletal muscles of untrained individuals, as evidenced by a modified acute inflammatory response when challenged by a second bout of exercise 4 weeks later (Deyhle et al., 2015). Long-term protective memory effects that provide improved recovery from acute eccentric exercise could be demonstrated to last for up to 9 months when untrained males were initially challenged by a single bout of repeated maximal isometric contractions (Nosaka, Sakamoto, Newton, & Sacco, 2001). At the molecular level, 7 weeks of resistance exercise training have recently been proposed to leave distinct memory marks in the chromatin landscape that persisted over a 7-week period of complete detraining, becoming manifest in an improved capacity to regain muscle mass and strength when training was resumed (Seaborne et al., 2018). In contrast, long-term or lasting memory effects of endurance exercise and training are less well-studied and, due to the multilevel impact of endurance exercise adaptation on whole-body physiology, also less straightforward to assess. In a recent study using a within-subject design with single-and double-leg endurance exercise, no coherent evidence of a skeletal muscle memory at the functional level or global transcriptome level from 3 months of endurance training was found retained after 9 months of detraining, albeit the authors could not altogether exclude residual memory patterns when analyzing the skeletal muscle transcriptome after a second subsequent training period (Lindholm et al., 2016). Due to lack of studies focusing on this topic, it remains to be established whether endurance and resistance exercise are equally remembered, or whether there are differences in the length of memory (or the degree of memory dissipation) that is inflicted in our muscles by different training modes.
Over the course of prolonged exercise training, untrained skeletal muscles develop progressive adjustments in mechanic and energetic characteristics, affecting mitochondrial function and biogenesis, vascular supply, enzymatic equipment, activation patterning, structural composition of the contractile apparatus, and even complete reprogramming of fiber types (Egan & Zierath, 2013;Talbot & Maves, 2016), albeit the extent to which the latter occurs is not entirely clear (Wilson et al., 2012). Analogous to long-term and short-term epigenetic memory effects from previous experience of exercise stress or training, it seems conceivable that during the initiation phase of skeletal muscle adaptation, each exercise regime should leave distinct epigenetic/transcriptional memory marks to accumulate retrospective information that navigates subsequent phenotypic adaptations. Such transcriptional memory effects were recently illustrated for key genes involved in mitochondrial biogenesis, including the most versatile transcriptional coactivator in the control of energy metabolism, named peroxisome proliferator-activated receptor γ coactivator 1-α (PGC-1α). During a 2-week period of high-intensity interval training, acute transcriptional response kinetics progressively changed after each successive training session, with gene-specific gradual adjustments in activation and resolution patterns (Perry et al., 2010).
In the following section, we will outline how skeletal muscle can make use of the epigenetic toolbox to retain a recollection of previous experience of exercise and training, enabling muscle fibers to respond differentially to subsequent bouts of exercise, thereby setting the scene for short-term and long-term adaptation processes.
We have to emphasize that these mechanisms do not act in isolation but are part of a closely interconnected network of endogenous and exogenous signaling pathways, equally complex and fascinating, but still a long way from comprehensive understanding.

| Chromatin modifications
The basic units of chromatin are the nucleosomes consisting of an octamer, formed by two copies each of the core histones H2A, H2B, H3, and H4, around which~146 bp of DNA are wrapped in almost two turns (Luger, Mader, Richmond, Sargent, & Richmond, 1997).
Active transcription depends on and is associated with major changes in chromatin organization to modulate the accessibility of the DNA template for TFs, regulatory factors, and RNA Polymerase II (Pol II), and to facilitate passage of Pol II along the gene body (Venkatesh & Workman, 2015). Alterations of chromatin structure are commonly associated with covalent modifications of DNA and histones which may alter the physical properties of the nucleosome, and serve to recruit downstream effectors that promote, retard or modulate transcriptional activity. For instance, active genes generally are characterized by a lack of DNA methylation at their promoters and carry high levels of acetylated lysines on the tails of core histones H3 and H4 (H3/4Kac). In addition, transcriptional active genomic regions often show histones with di-or trimethylated lysine residues on H3 (H3K4me2/3) at their core promoters, and H3K36me3 along the gene body, whereas active enhancers often are enriched in H3K27ac and H3K4me1. Repressed loci typically show hypoacetylation of both H3 and H4, di-and trimethylation of H3 at K9 and K27 (H3K9me2/3, H3K27me3), and/or contain 5'-methylated DNA cytosine residues (5mC) at CG dinucleotides (CpG sites) in promoter or enhancer regions (Chen & Dent, 2014;Venkatesh & Workman, 2015).
Intriguingly, hypomethylated CpG sites within regulatory regions may circumstantially be associated with "bivalent domains" of histone modifications, harboring both repressive marks, such as H3K27me3, and active marks, such as H3K4me3 (Jadhav et al., 2016;Kinkley et al., 2016;Sodersten et al., 2018;Weiner et al., 2016). Such seemingly "undecided" chromatin states preserve a higher rate of adaptive flexibility, allowing rapid changes in either direction ( Figure 1). DNA methylation and specific histone modifications appear to reciprocally influence each other at multiple levels, albeit the complexity of the underlying mechanisms has yet to be fully characterized (Du, Johnson, Jacobsen, & Patel, 2015). CpG dinucleotides are statistically underrepresented in mammalian genomes but spottily occur in dense clusters, so-called "CpG islands" that are typically defined as CG-rich regions with length greater than 200 bps, GC content greater than 50%, and an observed-to-expected CpG ratio greater than 0.6 (Illingworth & Bird, 2009). CpG islands commonly colocalize with promoters and transcriptional start sites DNA methylation at CpG sites has long been thought to represent a rather static epigenetic mark that is established during early development to define lineage-restriction, genomic imprinting, X-chromosome inactivation, and to silence repetitive regions. In contrast to this prevailing view, a number of recent studies provided hints for dynamic switching of DNA methylation in response to environmental cues, internal signals, and stressors (Emeny et al., 2018;Hartley et al., 2013;Magnusson et al., 2015;Saunderson et al., 2016). This suggests that differential methylation of CpG sites not only serves to establish epigenetic memory during lineage commitment and cell differentiation but might also contribute to adaptive transcriptional memory mechanisms.
While the predictive relationship between specific chromatin modifications and different states of transcriptional activity has been widely described in large-scale profiling studies, whether and in which ways distinct marks are instructive or permissive, cause or consequence of transcriptional activity is still a matter of ongoing debate. Some chromatin marks are lineage-specific or reflect celltype-specific states that are inherited within the cell lineage, allowing the cells of our body to always remember what cell they are and from which cell they derive (Atlasi & Stunnenberg, 2017). Other chromatin F I G U R E 1 Distinct histone modifications define different chromatin states. During active transcription, genes display an open chromatin state marked by activating histone modifications (green circles), while the dense chromatin state of repressed loci is associated with repressive histone marks (red circles). After cessation of the triggering signal, some histone modifications can be retained at altered levels to provide a transcriptional memory, marking poised, bivalent, and suppressed loci. In the poised mode, the gene locus has an open chromatin state and promoters can be pre-loaded with RNA polymerase II (Pol II), but processivity of Pol II is blocked. Such genes are prepared for rapid reactivation. Genes in the bivalent mode are characterized by the simultaneous enrichment of activating and repressive chromatin modifications. Such a chromatin state can change in either direction, providing circumstantial adaptation when the initial signal is encountered again modifications are dynamically associated with acute transcriptional activity and do not persist after the initial stimulus that drives the modification has ceased (Katan-Khaykovich & Struhl, 2002). However, some marks can be retained as a memory of recent transcriptional activity and thereby preserve information for future transcriptional events. Memory-responsive genes can either remain in an activated or poised state for more rapid expression or can be blunted to provide differential responsiveness when the initial stimulus is encountered again (Figure 1).
As reviewed previously, there is ample evidence from human studies and animal models that exercise sustainably alters the chromatin landscape in skeletal muscle, both acutely as well as in the long term (Jacques et al., 2019;Widmann, Niess, & Munz, 2019). As we will exemplify below, one of the most challenging tasks will be to disentangle which chromatin alterations arise from homeostatic perturbation, reflect random fluctuations, are mere passive bystanders of acutely increased or repressed transcription, and which are true epigenetic marks that prelude, establish, or maintain adaptive gene expression programs.

| Nuclear pore proteins
How distal regulatory elements (like enhancers or repressors) identify, recognize, and interact with their target promoters in the three-dimensional nuclear space as yet is not fully understood.
Recent studies indicate that enhancer-promoter contacts can be facilitated and stabilized by components of the nuclear pore complex (NPC) to establish a memory system for signal-dependent transcription (Pascual-Garcia et al., 2017). The NPCs, doughnut-shaped structures that penetrate the nuclear envelope membranes to provide a gateway between the nucleus and the cytoplasm, are modularly assembled from~30 different proteins, termed nucleoporins (Nups), which are broadly conserved among eukaryotes (Grossman, Medalia, & Zwerger, 2012). Early studies in yeast demonstrated that many active genes physically interact with the NPC. Some of the inducible genes that relocate from the nucleoplasm to the NPC upon activation remain at the nuclear periphery after repression and thereby retain a spatial memory of prior exposure that promotes future reactivation when cells re-encounter the initial stimulus ( Figure 2; Brickner & Walter, 2004;Brickner et al., 2007;Taddei et al., 2006). Similarly, hundreds of genes in human HeLa cells have been shown to exhibit interferon γ (IFN-γ)-induced transcriptional memory that persists over several rounds of cell division and requires the interaction of genes with the nucleoporin Nup98, leading to H3K4 dimethylation and binding of poised Pol II to the promoter (Gialitakis, Arampatzi, Makatounakis, & Papamatheakis, 2010;Light et al., 2013). Cells with poised genes react more rapidly and more strongly to IFN-γ than cells that have no memory of IFN-γ exposure. Remarkably, some Nups are not spatially restricted to the NPC but dynamically shuttle on and off the NPC (Rabut, Doye, & Ellenberg, 2004). These mobile Nups retain the ability to regulate gene activity in the nucleoplasm and actively participate in transcriptional control also at genomic loci that do not become positioned at the nuclear periphery (Capelson et al., 2010;Griffis, Craige, Dimaano, Ullman, & Powers, 2004;Kalverda, Pickersgill, Shloma, & Fornerod, 2010;Liang, Franks, Marchetto, Gage, & Hetzer, 2013).
In mammalian cardiomyocytes, Nups have emerged to participate in controlling the cardiac hypertrophic growth program. Here, the nucleoporin Nup155 can interconnect with the repressive chromatin modifier histone deacetylase 4 (HDAC4) to keep sarcomeric genes and Ca 2+ -handling genes in a poised state that can rapidly be resolved by removal of HDAC4 in response to mitogenic signaling (Kehat, Accornero, Aronow, & Molkentin, 2011). Similarly, Nupchromatin interactions are also crucial for temporal and spatial regulation of structural gene expression programs in skeletal muscle.
During skeletal muscle differentiation, the muscle-specific Nup210 has been shown to become integrated into the NPC to scaffold the TF myocyte-specific enhancer factor 2C (Mef2C) at the NPC, promoting activation of target genes involved in sarcomere assembly, myofiber maturation, and muscle growth (Raices et al., 2017).
Remarkably, this muscle-specific NPC-tethered gene expression network not only encompasses protein-coding genes but also distinct species of muscle-specific microRNAs.

| microRNAs microRNAs (miRNAs) are small endogenous noncoding RNAs of~22
nucleotides that are processed from the stem-loop structures of F I G U R E 2 Dynamic associations of genomic loci and regulatory regions with the nuclear pore complex (NPC) provide an architectural framework for transcriptional memory responses. Initial gene activation leads to translocation of the genomic locus from the nucleoplasm to NPCs through association with nucleoporins (Nups) in a manner dependent upon transcription factor (TF) binding, promoter-enhancer interactions, and chromatin modifiers. After cessation of the transcriptional trigger, the gene remains at the nuclear periphery in a memory state that can be rapidly reactivated. This way, NPCs provide a scaffold for topological genome organization and memory function precursor transcripts by the concerted action of nuclear and cytoplasmic enzyme complexes, crucially involving two ribonuclease In mammals, miRNA action has long been thought restricted to cytoplasmic PTGS. However, there is accumulating evidence that some fraction of mammalian miRNAs may also execute important nuclear regulatory functions by controlling gene expression in the nucleus at both transcriptional and posttranscriptional levels, as well as by affecting alternative splice site selection (Roberts, 2014). In a set of elegant experiments, a significant fraction of miRNAs and several RISC components, including AGOs, have been identified to become located and functionally operative within the nucleus of human cells (Avivi et al., 2017;Bottini et al., 2017;Castanotto et al., 2018;Gagnon, Li, Chu, Janowski, & Corey, 2014;Ohrt et al., 2008;Sarshad et al., 2018;Weinmann et al., 2009). Nuclear localized miRNAs were found capable both to suppress and to stimulate transcriptional expression at distinct gene loci, involving direct and indirect interference pathways. Target RNA levels in the nucleus can be reduced through site-specific cleavage by AGO slicer activity (Gagnon et al., 2014), enabling nuclear RISC not only to mediate degradation of mRNAs but also of diverse noncoding nuclear RNA species that serve gene regulatory functions (H. Liu et al., 2018). By repressing the repressors, specific nuclear miRNAs thus can indirectly promote transcriptional gene activation, for example, via F I G U R E 3 Overview of microRNA (miRNA) processing, mode of action, and cellular localization. Mature miRNAs are generated from hairpin-containing primary miRNAs (pri-miRNA) transcripts that are initially processed into ∼70-nucleotide stem-loop precursor miRNAs (pre-miRNA) by the nuclear DROSHA complex. After cytoplasmic translocation, pre-miRNAs are further cleaved by the DICER complex into ∼21nucleotide miRNA duplexes. Eventually, the miRNA duplex is loaded onto an Argonaute (AGO) protein to form the RNA-induced silencing complex (RISC). One strand of the duplex (passenger strand) becomes removed. The remaining RNA strand (guide strand) confers specificity to mature miRISC that now recognizes its mRNA targets and mediates posttranscriptional gene silencing by target-specific mRNA deadenylation/ decapping and decay. Circumstantially, mRNA targets can specifically be protected from degradation by association with distinct RNA-binding proteins. Mature miRNAs localize in multiple subcellular locations in the cytoplasm. On the rough endoplasmic reticulum, miRISC can interfere with the translation initiation process. Complexes of miRISCs and polysome-bound mRNAs can shuttle to the early/late endosomes for storage and/or degradation. Under certain cellular conditions, miRISCs can be selectively incorporated into multivesicular bodies (MVB) that act as transport intermediates between early and late endosomes, but can also fuse with the plasma membrane to release their intraluminal vesicles into the extracellular milieu. AGO-associated miRNAs can also translocate to the mitochondrion to promote translational activation or mRNA translational inhibition and decay. Finally, AGO-associated miRNAs can relocate to the nucleus to influence alternate splicing decisions and to control gene expression, at both transcriptional and posttranscriptional levels BEITER ET AL.

| 5481
miRNA-directed degradation of miRNA precursors, long noncoding RNAs, and natural antisense transcripts (NATs) that convey epigenetic silencing of distinct gene loci in either a cis or a trans manner (Beiter, Reich, Williams, & Simon, 2009;H. Liu et al., 2018;Roberts, 2014). However, it appears that the functions of miRNAs in the nucleus extend beyond RNA degradation pathways. Through base-pairing interactions with nascent transcripts, as well as with noncoding enhancer RNAs (eRNAs) and promoter-associated RNAs (paRNAs), miRNA-AGO complexes in the nucleus may serve as target guides for effector proteins, providing a molecular scaffold to attract chromatin remodelers and epigenetic enzymes towards destined genomic locations (Catalanotto, Cogoni, & Zardo, 2016;Holoch & Moazed, 2015;Weinberg & Morris, 2016). Reciprocally, availability of chromatin-modifying enzymes and DNA methyltransferases can be affected via PTGS in the cytoplasm, which in turn will also modulate specific downstream effects on the chromatin landscape (Singh & Campbell, 2013). This complex interplay has recently been found crucial in the functional reprogramming of macrophages that develop endotoxin tolerance after repeated or chronic exposure to endotoxin (Seeley et al., 2018). Here, immune memory is patterned via endotoxin-induced miRNAs that promote adaptive transcriptional silencing of a subset of inflammatory genes by repressing a chromatin remodeler that governs the reactivity of these genes.
Not surprisingly, miRNAs appear to be among the prime candidates to fine-tune the networks for transcriptional memory to past stimuli. Studies on the spatiotemporal distribution and functionality of miRNAs indicate that miRNAs and associated proteins can shuttle between different subcellular compartments with localized functionality, including cytoplasmic-processing bodies (P-bodies), endoplasmic reticulum (ER), endosomes, multivesicular bodies, mitochondria, and nucleus, and may differentially be stored for "on demand" use as needed (Leung, 2015;Pitchiaya, Heinicke, Park, Cameron, & Walter, 2017;Trabucchi, 2019). Moreover, reversible posttranslational modifications of RISC components allow flexible and context-specific regulation of miRNA activity and miRNA localization in response to endogenous and exogenous stimulation (Leung, 2015).
Several studies reported quantitative changes of skeletal muscle miRNA profiles in response to exercise and training, as well as in the wake of ageing and disease. Moreover, experimental approaches provided links for specific miRNAs species to be crucially involved in skeletal muscle differentiation, regulation of skeletal muscle growth, and fiber-type transformation. In these fields, the reader is referred to excellent reviews published recently (Domanska-Senderowska et al., 2019;Kirby, Chaillou, & McCarthy, 2015;Margolis & Rivas, 2018;Ultimo et al., 2018). At present, deciphering the true implications of miRNAs in skeletal muscle adaptation and memory modules is puzzling as each miRNA potentially targets a large number of genes, indicating substantial pleiotropic functional redundancy (Eichhorn et al., 2014;Liufu et al., 2017). Many miRNAs belong to multigene families, which are predicted to target the same (or overlapping) sets of genes, and multiple miRNAs may work as miRNA modules to synergistically regulate common target mRNAs (Ding, Li, & Hu, 2015). Moreover, in high-throughput quantitative studies, target prediction is primarily based on bioinformatic identification of phylogenetically conserved miRNA-binding sites, a methodological approach that is highly prone to produce false positive results (Pinzon et al., 2017).

| SKELETAL MUSCLE, EXERCISE, AND FLEXIBLE EPIGENOME
In recent years, there is growing awareness that chromatin evolved with multiple functions that reach beyond the storage, propagation, and expression of genetic information. It has become evident that the cellular chromatin state is dynamically linked to metabolic and homeostatic perturbations, not only by providing a flexible platform for metabolic gene expression programs but also by acting as an intrinsic rheostat of carbon flux and cellular pH (van der Knaap & Verrijzer, 2016). As has been reviewed extensively elsewhere, the activity of histone-and nucleic acid-modifying enzymes relies on (and is influenced by) the availability of specific substrates, intermediates, and products from diverse metabolic pathways, including glycolysis, tricarboxylic acid cycle, fatty acid β-oxidation, methionine cycle, and glutamine metabolism (Etchegaray & Mostoslavsky, 2016;Li, Egervari, Wang, Berger, & Lu, 2018;Nieborak & Schneider, 2018;Reid, Dai, & Locasale, 2017;Schvartzman, Thompson, & Finley, 2018;Sharma & Rando, 2017;van der Knaap & Verrijzer, 2016).
Consequently, shifts in substrate selection and energy provision in the exercising muscle may have multiple acute implications on chromatin dynamics and epigenetic modifications. An illustrative presentation of how metabolic activity can dynamically shape the overall chromatin landscape was recently provided by a comparative study of 11 non-tumorigenic and tumorigenic human cell lines, revealing that the cellular rate of glycolysis appeared strikingly reflected by the global level of histone acetylation (X. S. Liu, Little, & Yuan, 2015). There is emerging evidence that glycolytic flux directly feeds into histone acetylation, and the availability of acetyl-CoA, the preferred acetyl donor used by histone acetyltransferases (HATs), profoundly determines the abundance of histone lysine acetylation at a global scale (Cluntun et al., 2015;Simithy et al., 2017). Reciprocally, in cancer cells, global deacetylation of histones, mainly accomplished by HDACs, was found crucial to maintain intracellular pH homeostasis, mechanistically being coupled to the activity of monocarboxylate transporters that preserve physiological pH by pumping protons in a cotransport together with chromatin-derived acetate anions out of the cell (McBrian et al., 2013). It seems compelling to translate these findings to the exercising muscle when enhanced glycolytic flux becomes manifest in an accumulation of acetyl-CoA and a decrease of pH during energetically demanding contractile activity (Juel, 2008;Lundsgaard, Fritzen, & Kiens, 2018). Unfortunately, no experimental data are currently available concerning the kinetics and dynamics of global chromatin alterations during the course of prolonged contraction. What is evident is that exercise and skeletal muscle contraction provoke profound shifts in the balanced actions of HATs and HDACs, which have been conclusively linked to be involved in SC differentiation, metabolic adaptation, and regulation of muscle mass (Gaur et al., 2016;Hong et al., 2017;Mal, Sturniolo, Schiltz, Ghosh, & Harter, 2001;McGee et al., 2008;McGee, Fairlie, Garnham, & Hargreaves, 2009;McKinsey, Zhang, & Olson, 2001;Moresi et al., 2010;Niu et al., 2017;Yang, Menconi, Wei, Petkova, & Hasselgren, 2005). Nuclear export of HDACs and temporally increased global acetylation levels of H3K36 were detectable by western blot analysis immediately after a 60 min bout of intensive cycling exercise (McGee et al., 2009). Of note, acetylation of histones not only facilitates destabilization of DNAnucleosome interactions for active transcription but, when provoked at a global scale, also promotes rearrangements in the threedimensional genome architecture, thereby enabling extensive translocation of genomic loci and regulatory regions to the NPCs (Brown, Kennedy, Delmar, Forbes, & Silver, 2008). It remains yet to be deciphered how these multiple functional layers of chromatin conspire together, and how they can be integrated into current models of metabolic and adaptive programming in skeletal muscle.
In addition to global impacts on histone patterns, acute exercise also appears to affect the overall DNA methylation status in the contracting muscle. By the use of a methylation-sensitive restriction assay, a significant net decrease in whole-genome CpG methylation was observed in skeletal muscle samples from sedentary men and women after an acute bout of exhaustive cycling exercise (Barres et al., 2012). At the single-locus level, same authors reported acutely enhanced transcriptional activity and temporally diminished methylation levels at the promoter regions of a selected set of known metabolic driver genes, including peroxisome proliferator-activated receptor γ (PPARG) and its "Jack of many trades" cofactor PGC-1α (PPARGC1A; Barres et al., 2012).
Similarly, BeadChip profiling of bisulfite-converted DNA from skeletal muscle before and after an acute bout of resistance exercise revealed extensive changes in the DNA methylome, with a global trend towards acute hypomethylation of CpG sites (Seaborne et al., 2018). The same group analyzed differential basal methylation patterns provoked by chronic hypertrophy training, and deciphered several candidate genes that appear to become benchmarked at the start or in the course of resistance exercise training to become remembered in subsequent rounds of exercise (Seaborne et al., 2018). By comparing their differential methylation data with publicly available transcriptomic data sets, the authors further provided a list of putative "memory genes" that may exhibit differential responsiveness to recurrent resistance exercise stress (Turner, Seaborne, & Sharples, 2019). Genome-wide DNA methylation profiles provided supportive evidence for DNA methylation/demethylation being also an integral part of the maintenance and memory programs that govern skeletal plasticity in response to endurance exercise training, revealing dynamic methylation signatures in enhancer regions that harbor putative binding motifs for several TFs known to control basic adaptive gene expression networks (Lindholm et al., 2016). Further, by use of methylated DNA immunoprecipitation method coupled with a promoter tiling array, altered methylation levels at promoter sites of several key genes attributed to skeletal muscle homeostasis have been reported to manifest after 6 week of endurance exercise (Nitert et al., 2012). A general trend towards reduced promoter methylation levels of genes involved in energy metabolism, myogenesis, contractile activity, and oxidative stress resistance became apparent when global methylation patterns in skeletal muscle of healthy aged men with a lifelong history of physical activity were compared to "couch potatoes" of the same age group (Sailani et al., 2019). Reciprocally, also "bad memories" inflicted by physical inactivity, unhealthy diet, and prenatal stress have been reported to become reflected by differential DNA methylation patterns in skeletal muscle tissue (Alibegovic et al., 2009;Jacobsen et al., 2012;Jacobsen et al., 2014;Nilsson & Ling, 2017;Nitert et al., 2012;Sharples, Polydorou et al., 2016;Sheppard et al., 2017).
A number of fascinating questions arise from these studies that will drive future work on skeletal muscle plasticity in health and disease. However, at present state, the overlap between published data sets is difficult to reconcile, and the establishment of causation in vivo is challenging. Issues that hamper data comparison include differences in study cohorts and study designs (mode, intensity, and duration of exercise and training), and a vast array of different methodological, analytical, and statistical approaches to detect, localize, and quantify physical activity-dependent methylation signatures. Data interpretation is further complicated by cell-type heterogeneity of biopsy material that may confound observed muscle-specific epigenetic patterns at variable degrees. This is specifically an issue when a huge mixed fiber-type muscle is represented by a small biopsy sample. Sample-to-sample variations may mimic or overshadow true methylation changes due to the differential contribution of fiber-type specific methylation marks (Begue, Raue, Jemiolo, & Trappe, 2017). To complicate matters further, distinct exercise regimes as well pathologies have distinct effects predominantly on particular fiber types, and the relative composition and size of muscle fibers can dramatically change in response to exercise, aging, or disease. Currently, it is even unclear whether all myonuclei in a muscle fiber share identical epigenetic signatures. But even within homogenous cell populations, the epigenome is known to exhibit stochastic cell-to-cell variation and also allelic variation between the two chromosomes within the same nucleus (Onuchic et al., 2018). Multiple patterns of methylation in regions of interest may simply arise from a dynamic equilibrium between constant gain and loss of methylation rather than reflecting meaningful biological differences (Edwards, Yarychkivska, Boulard, & Bestor, 2017;Lovkvist, Dodd, Sneppen, & Haerter, 2016). It is further crucial to distinguish changes on average methylation levels at a genomic locus from changes in methylation of specific individual CpGs that may instructively affect gene transcription function, that is when directly located within TF recognition sequences (Han, Shi, & Spivack, 2013;Lioznova et al., 2019;Maeder et al., 2013;Yin et al., 2017). Reciprocally, there is increasing evidence reinforcing the notion that DNA sequence polymorphisms that affect TF recognition seem to account for a substantial part of DNA methylation and chromatin pattern variability (Ball et al., 2009;Bell et al., 2011;Hellman & Chess, 2010;Kasowski et al., 2013;Onuchic et al., 2018).
However, whether methylation at enhancers, promoters, and TSSs BEITER ET AL.

| 5483
generally controls or reacts to TF binding remains largely unexplored, and the ability to correctly assign a functional consequence to the local presence of DNA methylation so far has remained surprisingly limited (Luo, Hajkova, & Ecker, 2018;Schubeler, 2015). In the future, comprehensive screening approaches that integrate epigenetic, transcriptomic, and genotypic information (Birney, Smith, & Greally, 2016), as well as novel technologies for targeted methylome editing in experimental cell culture and animal models (Lei, Huang, & Goodell, 2018), hold promise for a deeper understanding of exercisedependent genomic methylation patterning and its contribution to adaptive skeletal muscle pathways.

| PERSPECTIVES
A better understanding of the molecular mechanisms involved in the control of skeletal muscle plasticity is fundamental to implement efficient and individually tailored exercise training programs that ameliorate or even restore disturbed energetics and metabolism in chronic disease states, and counteract diminished muscle strength and loss of muscle mass under physiological and pathophysiological conditions (including e.g. aging, malnutrition, disuse, obesity, diabetes, cancer, inflammation, neuromuscular disorders, and myopathies).
Chromatin dynamics constitute a crucial part of the decision-making processes that enable the skeletal muscle to structurally and functionally adapt to variations in working demand, nutritional state, and environmental factors. By establishing epigenetic memory marks of recent and remote experience, muscles can preserve an autobiographical history that determines how an individual responds and adapts to lifestyle behavior changes, pathological challenges, and aging.
Unquestionably, the reversible nature of epigenetic alterations provides novel opportunities and challenges for nonpharmacological and pharmacological intervention strategies in prophylaxis and therapy.
In this context, one primary goal of exercise research will be to find suitable answers as to how targeted exercise regimes can be exploited to inflict good memories into our muscle and to erase bad memories from times of physical inactivity, overnutrition, or disease.

CONFLICT OF INTERESTS
No conflicts of interest, financial or otherwise, are declared by the authors.