Is adult stem cell aging driven by conflicting modes of chromatin remodeling?



Epigenetic control of gene expression by chromatin remodeling is critical for adult stem cell function. A decline in stem cell function is observed during aging, which is accompanied by changes in the chromatin structure that are currently unexplained. Here, we hypothesize that these epigenetic changes originate from the limited cellular capability to inherit epigenetic information. We suggest that spontaneous loss of histone modification, due to fluctuations over short time scales, gives rise to long-term changes in DNA methylation and, accordingly, in gene expression. These changes are assumed to impair stem cell function and, thus, to contribute to aging. We discuss cell replication as a major source of fluctuations in histone modification patterns. Gene silencing by our proposed mechanism can be interpreted as a manifestation of the conflict between the stem cell plasticity required for tissue regeneration and the permanent silencing of potentially deleterious genomic sequences.


DNMT, DNA methyltransferase; H3K4me3, trimethylation at lysine 4 of histone 3; H3K9me3, trimethylation at lysine 9 of histone 3; TF, transcription factor.


Aging is characterized by a gradual functional decline of virtually every tissue system and increased vulnerability and susceptibility to numerous diseases. Regenerative mechanisms are required by the organism to maintain tissue homeostasis throughout life. Thus, restrictions to these mechanisms are fundamental in mammalian aging. In general, regenerative mechanisms rely on an appropriately regulated balance between the self-renewal and differentiation of adult stem cells. Accordingly, attenuated stem cell function has been suggested as representing a key aspect of mammalian aging 1. Understanding the processes that underlie age-related changes in stem cell functioning is therefore crucial to improve diagnostics and develop strategies for efficient prevention and therapy of age-related diseases.

During aging, the gene expression profiles of stem cells undergo changes that are accompanied by extensive remodeling and decreased chromatin stability 2. This kind of deregulation has been shown to impair stem cell functioning, including self-renewal and lineage specification 3, 4. Entire loss of stem cell pools accelerates the aging process dramatically 5.

The pronounced changes observed in the gene expression profiles of aging stem cells suggest that aging can be viewed as a loss of regulatory event coordination 6. Deregulation has been linked to the high metabolic activity of adult stem cells during tissue regeneration. These cells are exposed to high levels of DNA damage, which is induced by metabolic side products such as reactive oxygen species 7, 8. In fact, genomic changes have been considered as the origin of the majority of alterations in aged cells and to constitute the primary cause of aging 9. In contrast, changes in the epigenome have been considered to be merely indicative of permanent genomic changes 10.

Here, we put forward the hypothesis that alterations of the stem cell epigenome and, in particular, of the chromatin structure fundamentally contribute to aging.

Dynamic chromatin structure

Whether chromatin forms transcriptionally active euchromatin or transcriptionally silenced heterochromatin is primarily determined by modifications of histones and DNA. Histone acetylation promotes euchromatin, whereas DNA-methylation is associated with heterochromatin 11. Chromatin undergoes permanent remodeling during development and differentiation. This process includes both regulated and stochastic events, and takes place on different time scales. Histone modifications are subject to fast fluctuations on the time scale of seconds to hours 12. In contrast, DNA methylation confers long-term gene silencing that can be stable for months or even longer 13. Interestingly, intimate cross-talk between histone modifications and DNA methylation has been detected. Trimethylation at lysine 4 of histone 3 (H3K4me3) directly suppresses DNA methylation 14, while trimethylation of lysine 9 of histone 3 (H3K9me3) is preferentially associated with strong DNA methylation 15.

Both histone modification and DNA methylation have been demonstrated to be essential in adult stem cell renewal and differentiation 15–17. These findings have prompted questions about the impact of chromatin remodeling on stem cell aging. Experiments investigating this issue have proven difficult, as only limited numbers of these cells are available and their in vitro expansion can change the epigenome 18. It has been shown for different tissues that aging is correlated with functionally relevant globalized DNA hypomethylation, in parallel with focused DNA hypermethylation 19. In the haematopoietic system, these changes have already been detected in stem and precursor cells 20. The origin of these alterations is largely unknown.

Here, we propose that the essential alterations of the chromatin structure with age are caused by limited inheritance of epigenetic information.

Inheritance of epigenetic information

ChIP-chip and ChIP-seq experiments are steadily broadening our knowledge on the function of chromatin structure in the regulation of transcription. However, it is still unclear how epigenetic marks are transmitted from one cell generation to the next. As a consequence, we do not know how 30% of the information stored in a chromatin fiber is transmitted 21. Conservative, semi-conservative and/or random modes for the replication of epigenetic information have been suggested 22, with experimental evidence increasingly supporting random mode hypotheses 23. In a pioneering theoretical study, Dodd et al. 24 demonstrated that cooperation between histone modifications and positive feedback mechanisms are crucial for the coexistence of alternative history-dependent states of histone modification and, thus, for epigenetic memory. They showed that the necessary feedback for such “bi-stable modification states” can be provided by a mechanism where modified histones recruit enzymes that similarly modify nearby histones. Sedighi and Sengupta 25 provided a formal description of a bi-stable switch through a self-sustaining equation, M = f(M), assuming cooperative binding of the histone modifying proteins to chromatin. An equation of this type can have two stable solutions (M1, M3), separated by an unstable solution (M2, see Fig. 1A).

Figure 1.

Fluctuation-induced changes in histone modification levels. A: Dynamics of fluctuation-induced transitions in a bi-stable system. Modification states M1 and M3 represent stable solutions of the self-sustaining equation, M = f(M), while M2 represents an unstable one. They are obtained as intersections of the function f(M) with the diagonal shown. For example, suppose the system is in state M3, then small fluctuations do not change the state. However, large fluctuations can drive the system below the M2 state and induce spontaneous de-modification, i.e. a transition into state M1. The source of such fluctuations can be (B) stochastic (de-)modification dynamics, and incorporation of de novo synthesized histones resulting from (C) ongoing exchange of individual nucleosomes (star symbol) or (D) cell replication.

Note that spontaneous transitions between the states of low (M1) and high (M3) modification levels can be induced by different types of molecular fluctuations and, hence, may corrupt inheritance of the original modification state (Fig. 1A). Accordingly, spontaneous histone demodification in the course of cell divisions has been predicted in theoretical studies 26. Support for these ideas comes from the observation that modification levels of parental histones are indeed larger than those of newly incorporated histones 27.

Propagation of DNA methylation has been extensively studied 28. Three different DNA methyltransferases (DNMTs) have so far been identified, two of which (DNMT3A,B) are involved in de novo methylation and one (DNMT1) is required for the maintenance of DNA methylation. A number of theoretical studies have focused on the cooperative role of de novo and maintenance methylation 29, 30. Recently, this focus has been extended to a fundamental issue in stem cell biology, namely the asymmetric strand segregation hypothesis 31. It has been shown that asymmetric strand segregation leads to systematic increases in methylation levels if parent strands are subject to de novo methylation events. Thus, directional changes in epigenetic information have been suggested as a cost of high genetic fidelity, which has implications for aging and disease.

We expect that understanding the cross-talk between mechanisms of epigenetic information processing and inheritance will be of fundamental importance in unraveling geroid phenotypes.


Here, we aim to explain the experimentally observed, age-related, “local” hyper-methylation at defined loci (typically CpG-rich promoters), paralleled by “global” DNA hypo-methylation at distinct loci 32, that is found in diverse human tissues 19. We hypothesize that this remodeling is a result of the cross-talk between histone modification and DNA methylation processes.

There are two general, experimental observations that have guided the development of our hypothesis. Firstly, complex and surprisingly heterogeneous alterations of the epigenome have been detected during aging. These alterations suggest a strong stochastic component of the ongoing remodeling processes. Secondly, the observed non-homogeneous spatial distribution of epigenetic changes appears to require cross-talk between different mechanisms. This suggests that it cannot simply be explained by a single mechanism affecting the genome at a global scale, such as, for example, the above-mentioned asymmetric strand segregation 31, or the global up-regulation/down-regulation of DNMTs 33, 34.

We suggest that age-related changes in DNA methylation at particular genomic loci are a consequence of the major fluctuations in histone modifications. In particular, dilution of histone marks can eliminate or weaken the protective action of histone modifications and, thus, enable DNA (de-)methylases to become active at the genomic loci concerned.

Since the fluctuations in histone modifications due to underlying stochastic modification dynamics (Fig. 1B) are, in principle, symmetric and reversible, the question arises as to how they could trigger an age-related, unidirectional trend towards histone de-modification. We suggest that fluctuations due to the incorporation of de novo synthesized histones during histone exchange processes and DNA replication play a key role (Fig. 1B–D).

Dilution of modified histones has been postulated for nucleosome exchange processes during transcription and DNA-repair 35. This is supported by evidence for the incorporation of new histones at DNA repair sites in human HeLa cells 36. However, these processes can potentially induce only a limited amount of dilution and, hence, de-modification may result for only a small subset of relatively unstable modified regions. During DNA replication, the modified parental histones appear to be randomly distributed along the daughter strands and are complemented by unmodified histones that are synthesized de novo. These processes have been studied in detail in human HeLa cell lines 23. On average, DNA replication leads to dilution of each strand's modified histones down to half of their equilibrium value in the parent cell. Such strong dilution potentially results in spontaneous de-modification of all bi-stably modified genomic loci. We consider such de-modification itself to be the major prerequisite of both focused DNA hyper-methylation and global DNA hypo-methylation. In general, ongoing fluctuations in the histone modification level can reverse de-modification. However, this scenario at least allows a time window for long-term changes of the DNA methylation status. These changes in turn impact histone modification dynamics (see below).


Cross-talk between histone modifications and DNA-methylation have been observed in numerous studies (reviewed in 37). A well-studied interaction, which supports our hypothesis, is that between the histone methyltransferase MLL1 triggering H3K4me3 and the machinery of de novo DNA-methylation. On the one hand, the DNMT DNMT3a and its accessory protein DNMT3L interact with un-methylated H3K4 through an ADD domain. Using mouse embryonic stem cells (ESCs), this interaction has been demonstrated to become suppressed for tri-methylated H3K4 by direct interaction 14. On the other hand, it has been shown that mouse MLL1, as part of the MLL1 complex, contains a CpG-interacting CXXC domain that couples the H3K4 methylation reaction to unmethylated DNA 38. Together, these results suggest a feedback loop whereby H3K4 methylation protects local CpGs from becoming methylated, while high levels of local DNA methylation prevent H3K4 methylation. Thus, de-modification of a chromatin region, e.g. associated with a CpG-rich promotor, would increase the accessibility of DNMT3a and hence facilitate local CpG methylation. In further rounds of replication, the pattern of de novo DNA methylation would be faithfully copied by DNMT1 and, thus, prevent further H3K4 de novo methylation. A schematic description of the proposed model is given in Fig. 2A.

Figure 2.

Sketch of the proposed model of age-related chromatin remodeling. H3K4me3 cross-talk with the de novo DNA methylation machinery is used as an example. A: In young individuals, many active genes with CpG-rich promoters are associated with H3K4me3. This histone modification protects the DNA from becoming methylated de novo by methylase DNMT3a. Spontaneous de-modification of the histones during aging enables the DNMT3a to become active and to silence the genes by DNA methylation. B: An assumed consequence of gene silencing is globally reduced activity of DNMT1 and/or DNMT3a. This leads to partial DNA de-methylation, as predominantly observed for promoters of CpG-poor genes or repetitive units.

Note that experimentally observed de novo DNA methylation during aging has been implicated in the reduced phenotypic plasticity of, for example, human haematopoietic progenitor cells 39.


Cross-talk between histone modification and DNA-methylation has also been observed for H3K9 methylation and DNA maintenance methylation. These relationships have been most clearly established in Neurospora crassa 40 and Arabidopsis 41, where DNA methylation directly depends on H3K9 methylation. However, in mammals, the dependence of DNMT1 recruitment on the local presence of H3K9 methylase G9a has also been suggested based on experimental findings on human cell lines 42 and mouse ESCs 43. Presumably, this provides a mechanism for coordinated maintenance of DNA and H3K9 methylation during cell division. Thus, if fluctuations in histone modification patterns dilute existing H3K9me3 methylation marks, the heterochromatic structure of the regions involved could switch into euchromatin. The resulting reduction in cooperative recruitment of histone methylases (like G9a) would then, as a side effect, locally reduce DNMT1 recruitment and thereby promote passive DNA de-methylation.

However, there is another explanation for the experimentally observed global trend of hypo-methylation. In this new scenario, parts of the DNA-methylation genes become silenced by the mechanisms described above, resulting in globally reduced DNA methylation activity. This scenario is experimentally supported by the findings of reduced DNMT1 activity in human fibroblasts 44, as well as reduced DNMT3a activity in mouse brain tissue 4 during aging. Accordingly, we consider the observed global DNA hypo-methlyation to be a side-effect of the described mechanisms of DNA hyper-methylation (see Fig. 2B).

Remarkably, the subset of genes involved in DNA hypo-methylation in aged human hematopoietic precursor cells has been observed to show significant overlap with genes that become hypo-methylated during differentiation of young cells. Consequently, age-related DNA hypo-methylation has been linked to a reduced stem cell differentiation potential 39.

A decrease in nucleosome density with age has been observed and linked experimentally to changes in chaperon activity 35. As cooperative action is critical for bi-stability, a decreasing number of nucleosomes will render originally stable, modified, chromatin states more sensitive to fluctuations. Thus, over time, more genes will change their DNA methylation status, meaning that the proposed aging process is progressive.

Dynamic model of the aging process

To further elaborate on our hypothesis, we have developed a simple mathematical model. This model builds on an artificial genome model, as introduced by Binder et al. 45. The explicit representation of the genome in this model enables a straightforward definition of large transcription factor (TF) networks (Fig. 3A and B). As a result, it allows different modes of transcriptional regulation to be considered. We have included regulation by TFs and histone modifications. In order to describe histone modifications, we have identified the genes of the artificial genome with the response elements required for recruitment of histone modifiers (Fig. 3A) and assume that all histones associated with the response elements act cooperatively. We focus our model on H3K4me3 dynamics. We assume that interaction between CpGs and the CXXC domain of the H3K4 methyltransferase MLL1 controls recruitment of the MLL1 complex. Cooperative binding is introduced by assuming that recruitment is improved by the interaction of the MLL1 complex with modified histones 46. This leads to a system of histone modification that shows bi-stable solutions, M = f(M,{a,b,c…}), for the defined parameter sets {a,b,c…}. In particular, CpG density and the number of cooperatively acting histones determine whether a response element becomes bound by the MLL1 complex and, thus, if the associated histones become bi-stably modified. Details of our model assumptions can be found in 47. These assumptions are supported by increasing evidence that higher eukaryotes, and in particular mammalians, do indeed exploit epigenetic feedback circuits for bi-stable switches in transcription states. For example, the discovery of H3K27 demethylases in 2007 strongly supported the idea of an epigenetic memory that relies on the concept of multistability 48. Recent evidence has demonstrated that UTX, a H3K27 demethylase, acts as a critical switch for activating the cardiac developmental program in mice 49. Similar epigenetic, bi-stable systems have been described in the context of the human Ring1B/Bmi1 ubiquination system 50, for a histone H3 phosphomethyl switch regulating TFIID-mediated transcription during the human cell cycle 51, and an epigenetic switch involved in the activation of pioneer factor FOXA1-dependent enhancers during human neural development 52. Interestingly, this latter factor exploits cross-talk between local DNA methylation and the H3K4 methylation status, similar to the process we describe above.

Figure 3.

Dynamic model of the aging process. A: The random genome model defines genes on a random string of four digits representing the genome. A defined sequence is used to identify a base promoter of a gene (dark blue); a defined number of digits downstream represents the coding region (orange); and the sequence upstream of the next coding region defines its regulatory region (white). Transcribed genes (1) are translated into TFs (2) that bind to the DNA (3) with a binding sequence that is calculated from the coding sequence. TFs regulate the transcription of the gene upstream from their binding position (4). The genes are identified with response elements for histone modifiers, defining the extension of cooperative chromatin regions. B: Example of a TF-network obtained from the random genome model. Activated genes are shown in red, repressed genes in green. C: Simulated impact of H3K4me3 de-modification during cell replication on gene expression and DNA methylation. Each line represents a cooperative chromatin region of a respective gene. After a few cell divisions, bi-stably modified chromatin regions become randomly de-modified. Accordingly, the expression of the associated genes decreases. Moreover, histone de-modification allows for methylation of the associated DNA (hyper-methylation) on a longer time scale. Assuming that part of the genes encoding the methylation machinery also becomes silenced, all genes that are methylated in the initial state decrease their methylation level over time (hypo-methylation).

Our model of bi-stable histone modification dynamics is used to calculate the expression of all genes. This expression pattern depends on the action of TFs and the histone modification status. While TFs are randomly chosen to be either activators or repressors 45, for histones we only consider the activating mark H3K4me3. Gene expression is assumed to be directly proportional to the modification level M of the associated histones. In order to accentuate the proposed mechanisms, we have adjusted the model parameters to observe a genome where the majority of genes is bi-stable. Moreover, we assume that all bi-stable genes are modified in the initial configuration (young cell). The initial expression and modification statuses are shown for zero divisions in Fig. 3C.

We have analyzed the inheritance of modified states during activated cell division, assuming that the modified histones of the mother strand are randomly distributed onto the daughter strands. All other sources of fluctuations of the modification level are ignored. Most of the bi-stable regions become de-modified within a few rounds of cell divisions (<10) (see Fig. 3C), leading to a decrease in gene expression of the associated genes. As a further consequence of the de-modification process, the DNA associated with de-modified histones becomes de novo methylated.

We have applied the basic model introduced by Sontag et al. 30 in simulations of this process, assuming a 95% efficacy of maintenance methylation and an 8% initial probability of de novo methylation in accordance with the authors. As shown in Fig. 3C, during about 40 cell divisions the methylation status of the de-modified genes changes from un-methylated to highly methylated, i.e. they become hyper-methylated. However, the resulting distribution of the methylation levels of individual genes turns out to be rather broad.

We consider genes encoding the de novo DNMTs to be part of the pool of silenced genes. As a consequence, the de novo methylation rate decreases to about 4%. Accordingly, all genes initially DNA methylated become hypo-methylated over time, i.e. their methylation levels decrease. Note that we have not assumed any DNA methylation feedback on gene expression in the example shown. Comparable results are observed if genes become silenced by DNA methylation instead of histone de-modification. However, in the case of the former, changes in gene expression occur over the course of the entire simulation (not shown). The system reaches an equilibrium state, which simultaneously accounts for changes in histone methylation, changes in the accessibility of the DNA for methylation, and for changes in the gene expression of DNA-methylation genes in particular.

The model demonstrates that although the de-modification process for most of the genes may be complete after a few cell divisions, changes in DNA methylation state take much longer to manifest. Thus, a relationship between both processes may be hard to detect experimentally, because this would require correlation of changes occurring in young individuals (histone de-modification) with changes becoming prominent only in aged individuals (DNA hypermethylation).

Tissue specifics of the aging process

Our model predicts genome-wide, age-related changes in DNA methylation of bi-stable genes. Whether genes exhibit bi-stable expression states depends on intrinsic (sequence-specific) and extrinsic (environmental) aspects. In particular, bi-stability is influenced by the number of cooperatively acting histones, the local CpG density and the presence of response elements, as well as the activity of the (de-)modification molecules.

In experimental studies, tissue-specific DNA methylation profiles have been analyzed in detail. It has been demonstrated that such profiles are predictors of the tissue type 19. Consequently, the proposed aging-process of the epigenome has to be considered as tissue-specific, with the particular tissue defining the stability of the epigenetic states. As loci in CpG-rich regions mostly gain methylation and loci in CpG-poor regions mainly loose methylation with age, a general mechanism involving silencing of bi-stable, CpG-rich genes is likely to occur 19.

In our model, the processes of chromatin remodeling depend on proliferation activity in the tissue. Accordingly, the model applies only to somatic stem cells. Moreover, this dependence prompts the question as to whether tissues with high turnover age faster. Here, we examine how far chromatin remodeling results in aging phenotypes. Age-related changes in epigenetic states are not expected to directly result in severe functional consequences for the tissue, but to increase the risk of pathological transitions 19, 32. In general, predictions on phenotypic consequences are difficult given that high tissue turnover can be observed for different types of stem cells. For example, stem cells in the gut have been shown to proliferate continuously 53, while hematopoietic stem cells are quiescent most of the time 54. So, are there tissue-specific strategies that enable stem cell plasticity without permanent gene silencing? That stem cells display different degrees of plasticity has been long recognized, with stromal (or mesenchymal) stem cells being potentially the most plastic population 55, 56. How do these strategies function in the presence of an aging environment and, in particular, in the presence of an aging stem cell niche? These questions clearly require further research.

Only recently has it been shown that age-related hyper-methylation has the same targets as the hyper-methylation observed in cancer 57, thus explaining why age is the risk factor in cancer development. Different mechanisms have been suggested as to how epigenetic changes affect genomic stability. It has been shown that histone methylation can trigger recruitment of components of the DNA repair machinery 58. Changes in the modification status, as assumed here, may thus directly impair genomic stability. Moreover, genes mediating DNA damage-induced apoptosis have been found to become silenced during aging 57. Consequently, damaged DNA may accumulate. Both mechanisms can potentially contribute to cancer and have to be considered as tissue-specific. A recent review on this issue has been provided by 59.

Conclusions and prospects

We have formulated the hypothesis that age-related, epigenetic drifts originate in the limited cellular capability to inherit epigenetic information. Specifically, we suggest that the spontaneous loss of histone modification at bi-stably modified chromatin regions due to fluctuations gives rise to long-term changes in the DNA methylation status and, thereby, modifies the expression of the associated genes. Experiments particularly support our aging model with respect to long-term gene silencing by DNA-methlyation subsequent to a loss of H3K4me3. In contrast, long-term gene activation by DNA de-methylation following a loss of H3K9me3 (or other marks) remains more speculative. Given the deleterious consequences of the described aging scenario, it prompts the question as to why these stochastic mechanisms have been conserved in evolution.

Fluctuations in epigenetic states are a central issue of our hypothesis. There is increasing evidence that stem cells exploit such fluctuations to control their fate decisions 60, 61. A candidate for these kinds of fluctuations is stochastic dilutions of histone modifications, as discussed in this paper. Note that these fluctuations are always present, even if a differentiation context is missing. Accordingly, they continuously expose previously protected genes to permanent silencing by DNA methylation, which itself is continuously active, to ensure silencing of potentially hazardous sequences such as retrotransposons 62.

Thus, there is an accumulating probability of dysfunctional gene silencing. We conclude that this may be interpreted as a manifestation of a conflict between the stem cell plasticity required in tissue regeneration and the permanent silencing of deleterious genomic sequences.

Remarkably, current knowledge about the establishment of epigenetic patterns during development suggests a similar sequence from transient regulatory events to increasingly stable epigenetic marks, with DNA methylation marking the most stable “programs” of differentiated cells 63.

So, is aging just an inevitable, dysfunctional continuation of development? Taken together, this evidence would support the view that life, ultimately, always has to “pay back its credits to the accounts of chance” 64 and that resetting the epigenome back to its “ground state” in the germ line 65 is as essential for evolution as the transmission of genetic information.


This work was supported by the BMBF grant MAGE (grant number 50500541). We thank H. Binder for assistance in preparing Fig. 3.