|Home | About | Journals | Submit | Contact Us | Français|
Chromatin is a complex of DNA, RNA, histones and non-histone proteins and provides the platform on which the transcriptional machinery operates in eukaryotes. The structure and configuration of chromatin are manipulated by families of enzymes, some catalysing the dynamic addition and removal of chemical ligands to selected protein amino acids and some directly altering or displacing the basic structural units. The activities of many of these enzymes are sensitive to environmental and metabolic agents and can thereby serve as sensors through which environmental agents can alter gene expression. Such changes can, in turn, precipitate either local or cell-wide changes as the initial effect spreads through multiple interactive networks. This review discusses the increasingly well-understood mechanisms through which these enzymes alter chromatin function. In some cases at least, it seems that the effects on gene expression may persist even after the removal of the inducing agent, and can be passed on, through mitosis, to subsequent cell generations, constituting a heritable, epigenetic change. If such changes occur in germ cells or their precursors, then they may be passed on to subsequent generations. Mechanisms are now known to exist through which an epigenetic change might give rise to a localized change in DNA sequence exerting the same functional effect, thereby converting an epigenetic to a genetic change. If the induced genetic change has phenotypic effects on which selection can act, then this hypothetical chain of events constitutes a potential route through which the environment might directly influence evolution.
Single-celled organisms must sense and respond to their environment, often by detecting changes in nutrient levels and up- or downregulating expression of selected genes. Regulation of the three genes that make up the Lac operon in Escherichia coli provides a classic example, the fundamentals of which were worked out almost 50 years ago and revealed concepts that underpin the basics of gene regulation across the living world (Jacob & Monod 1961; Vilar et al. 2003). Two proteins, the Lac Operator and Repressor, bind to defined sequences at the 5′ end of the Lac operon and, respectively, enhance or repress the coordinated transcription of the three Lac genes. The Lac Repressor protein is able to bind lactose and its derivatives. The bound form is inactive as a repressor, thus allowing transcription of genes encoding lactose metabolizing enzymes in an environment rich in this particular sugar. Single-celled eukaryotes operate in a similar way. The yeast Saccharomyces cerevisiae switches the expression of a defined family of genes depending on whether its primary nutrient source is glucose or galactose (Holstege et al. 1998; Bennett et al. 2008). The switch involves increasingly well-defined signalling pathways that detect the environmental change and transmit an appropriate signal to selected regions of the genome.
It seems likely that most, if not all, cells in multi-cellular organisms have retained the ability to respond to environmental changes with altered programmes of gene expression. In some higher eukaryotes, environmental changes can drive major changes in phenotype. For example sex determination in some fish is temperature sensitive (Fernandino et al. 2008; Marshall Graves 2008; Ospina-Alvarez & Piferrer 2008), while appropriately timed flowering in some plants requires exposure to a cold period, a process known as vernalization (Finnegan et al. 2004; Shindo et al. 2006). Nonetheless, for most cells in higher eukaryotes, the environment in which they exist is determined by the physiology and metabolism of the organism and of the cells in their immediate neighbourhood. Signals from this local environment, either cell–cell contacts or soluble factors, prompt the cell to put in place programmes of gene expression appropriate for replication, differentiation, quiescence or apoptotic death, depending on the cell type and developmental context. Contact with the external environment is filtered by these surroundings, but may still be influential. For example, although the mammalian embryo is protected from the physical and chemical dangers of the outside world by the uterine environment, elements of the maternal diet, drugs or accidentally ingested toxins can find their way into the embryo via the maternal circulation. Such environmental effects can be dramatic, as evidenced by the morphological effects of drugs such as valproic acid (VPA), a simple short-chain fatty acid and an effective anti-epileptic (Duncan 2007; Eadie 2008), or more subtle, as in the complex effects on foetal development associated with high maternal alcohol consumption (Chiaffarino et al. 2006; Disney et al. 2008).
Establishing the extent to which particular environmental factors influence genome function is an essential first step in defining their potential effects on long-term viability of the target organism. If an environmental factor can induce an epigenetic change that is heritable through mitosis, and that persists even in the absence of the original factor, then there is the potential for significant phenotypic effects long after the initiating factor has disappeared. Further, if such mitotically heritable changes are induced in germ cells, then there is the potential for transmission through meiosis to succeeding generations (figure 1). It is well known, through many years of work on imprinted genes, that epigenetic effects (i.e. whether a gene is expressed or not) can be transmitted through the germ line (Jaenisch & Bird 2003; Reik et al. 2003).
An overview of the interacting networks through which environmentally induced changes in gene expression influence cell behaviour and potential. Gene expression is regulated through an interlinked complex of DNA, histones, non-histone proteins (including ...
In this review, I will explore mechanisms by which environmental factors impact on the functioning of the genome in higher eukaryotes. I will focus on molecular mechanisms that operate through families of enzymes that modify chromatin-associated proteins and will use these mechanisms to describe how an environmental agent might induce a change in gene expression that is heritable through mitosis (effectively an epigenetic mutation) or even through the germ line to subsequent generations. Such a picture is necessarily incomplete and the increasingly important families of regulatory RNAs, reviewed recently (Carthew & Sontheimer 2009), will not be dealt with in any detail; their actions are likely to complement the mechanisms discussed here. I will argue that known mechanisms provide the potential for environmental agents to exert a direct influence on evolutionary change, perhaps by accelerating incremental, Darwinian processes.
The operon model in which activator and repressor proteins bind to specific DNA sequences in ways that are sensitive to the concentrations of metabolites and environmental nutrients is still the paradigm on which our understanding of gene regulation is based. Transcription factors (TFs) (i.e. operators, repressors, etc.), metabolic sensing and positive and negative feedback loops can act together as sophisticated control networks regulating families of genes. This model works well for small genomes, but encounters problems when confronted with the large genomes common in multicellular eukaryotes. It is a surprising fact that, as organisms have increased in complexity through evolutionary time, genome size has increased out of all proportion to the relatively modest increase in gene number. For example, while E. coli has 4200 genes in 4.5 × 106 bp of DNA (approx. 1.1 kb per gene), Homo sapiens has approximately 24 000 genes in 3.3 × 109 bp of DNA (approx. 140 kb per gene). The reasons for this remain a matter of conjecture, but it is clear that this excess of non-coding DNA makes mechanisms of gene regulation based solely on TF binding untenable in higher eukaryotes. For example, a typical 6 bp consensus TF-binding sequence will occur by chance once every 4096 (46) base pairs, or about 700 000 times in the human genome. Even a relatively large binding sequence of 10 bp, as used by the Lac repressor, would still occur about 3000 times by chance alone. As most TFs bind to several sequences similar to the consensus sequence, the number of potential binding sites will be even higher. One way round this numerical problem would be for higher eukaryotes to have evolved much larger TF-binding sequences, but in fact, most binding sequences have remained relatively short at around 6–8 bp (Wray et al. 2003). What distinguishes those sites that bind TFs in a functionally meaningful way and the vast majority that do not? Mechanisms by which the TF-binding site problem might be resolved include the selective packaging of DNA as chromatin, a structure unique to eukaryotes, and the enzyme-catalysed, post-translational modification of chromatin and TFs themselves. Both these processes are generally classed as ‘epigenetic’ and are considered in more detail below.
Of the various protein–DNA, and RNA–DNA interactions that mediate genome function, that between the histones and DNA occupies a special place, owing to both the abundance of the histones and the intimacy of their association with DNA. The basic unit of DNA packaging is an octamer of histones (two each of H2A, H2B, H3 and H4) around which are wrapped 146 bp of DNA in 1¾ superhelical turns (Luger et al. 1997). This structure, the nucleosome, is found in virtually all eukaryotes and is the first in a complex series of folding steps that, in higher eukaryotes, package approximately 2 m of genomic DNA into a cell nucleus of approximately 10 µm diameter. Packaging DNA in this way inevitably influences its ability to bind TFs and other DNA-binding proteins. Experiments over many years have shown that placing a nucleosome over a TF binding site can, in itself, block factor binding (Fragoso et al. 1995), and directed nucleosome positioning is a potentially important control mechanism (see below). But the nucleosome can also influence genomic functions in more subtle ways. The four core histones are subject to over 100 different post-translational modifications to defined amino acids, including acetylation of lysine, methylation of lysine and arginine, phosphorylation of serine and threonine and attachment of the short peptide ubiquitin (reviewed in Turner 2005; Kouzarides 2007). All are put in place by specific enzyme families, and removed by others (figure 2). The most extensively studied modifications are found along the N-terminal histone tails, regions that contain little secondary structure and are exposed on the surface of the nucleosome (Luger et al. 1997). The histone tails, or their modifications, have little direct effect on nucleosome structure, though they may influence internucleosome interaction and hence higher-order chromatin structure (Luger & Richmond 1998). However, recent analyses, often by mass spectrometry, have shown that numerous amino acids within the globular histone domains inside the nucleosome are subject to modifications exactly analogous to those on the histone tails (Cosgrove 2007). The functional significance of these internal modifications remains to be defined, but they are likely to exert a direct effect on nucleosome structure, and it has been plausibly suggested that they influence nucleosome mobility and perhaps positioning (Cosgrove 2007).
Metabolic and environmental influences on histone-modifying enzymes. The N-terminal tail domains of the core histones are acted upon by a variety of enzyme families, many of whose members require common metabolites as substrates or cofactors. The level ...
The enzymes known to be involved in setting and removing histone modifications are increasingly numerous; for example, there are 18 histone deacetylases (HDACs) in humans and mice (Gregoretti et al. 2004; Frye et al. 2007) and 28 different methyltransferases known to act on histones, at least in vitro, and no doubt more remain to be discovered (Allis et al. 2007). Some histone modifications (e.g. lysine acetylation) reduce the net positive charge of the histone tails and thereby reduce histone–DNA binding, perhaps with functional consequences. Alternatively, specific modified residues, or combinations thereof, can form binding sites for non-histone proteins, which in turn influence chromatin structure and function. This concept was first proposed over 15 years ago (Turner 1993), but only recently has the true diversity of the range of binding options available, and their functional outcomes, become apparent (de la Cruz et al. 2005; Taverna et al. 2007; Turner 2007). For the present discussion, two points are particularly important. The first is that the steady-state level of each modification represents a dynamic balance between the effects of the modifying and de-modifying enzymes, with turnover likely to vary from one part of the genome to another and between cell types. The second is that many, if not all, of the enzymes are dependent upon, or influenced by, metabolites or components present in the intra- or extracellular environment (figure 2). Thus, the nucleosome, through the array of histone modifications it carries and the enzymes that put them in place, can be seen as a finely tuned sensor of the metabolic state of the cell and the composition of its environment. It provides a potential platform through which environmental variables can influence genomic function.
As might be expected of a system in which combinations of modifications determine the specificity of binding of effector proteins, there is interaction (also called cross-talk) between modifications that helps determine the pattern of modifications to be generated. For example, treatment of cells with a variety of HDAC inhibitors (figure 2) not only leads, as expected, to global hyperacetylation of core histones, but also generates hypermethylation of H3 lysine 4 (Nightingale et al. 2007). Methylation at other lysines is unaffected. The explanation for this may lie in the properties of the methyltransferase catalytic domain. The catalytic (SET) domain of the enzyme responsible for H3 lysine 4 methylation, MLL1, shows enhanced methylation activity against highly acetylated histone tail substrates (Nightingale et al. 2007). Other factors may also play a role (Lee et al. 2006), but the important point is that an inhibitor can generate changes in histone modification beyond those expected from the catalytic activity of the enzyme it initially inhibits.
Chromatin-associated non-histone proteins, including many TFs, are subject to a similar variety of enzyme-catalysed post-translational modifications to those on the nucleosome. For example, androgen receptor (AR), cMYC and p53 can all be phosphorylated at specific serines and threonines, acetylated at specific lysines and ubiquitinylated, also at specific lysines (Glozak et al. 2005; Vervoorts et al. 2006; Popov et al. 2007; Vousden & Lane 2007). The modifications are put in place and removed by the same enzyme families that are involved in histone modifications, often with several different enzymes acting on the same factor; e.g. the acetyltransferases GCN5, TIP60 and CBP/p300 all act on cMYC and have been associated with distinct functional outcomes (Popov et al. 2007).
Specific modifications have selective effects on TF function. For example, acetylation of the AR enhances its ligand-dependent activation of a subset of target genes, leading to enhanced cell growth, but so far has been found to have little or no effect on its repressive or apoptosis-inducing activities (Fu et al. 2004; Popov et al. 2007). Acetylation of p53 is essential for its role as a mediator of the stress response, though other functional effects seem less acetylation dependent (Tang et al. 2008). As with histones, modifications of TFs are interdependent and interactive. At the simplest level, acetylation of the ε-amino group of a lysine blocks ubiquitinylation of that same lysine. Attachment of a single ubiquitin molecule can enhance TF function, while a ubiquitin polymer at the same residue can target the protein to the proteasome for degradation (Vervoorts et al. 2006). This is an important mechanism for downregulating the transcriptional response to TFs. Interactions can be more subtle. For example cMYC can be phosphorylated at threonine 58, but only if serine 62 is phosphorylated first (Yeh et al. 2004; Arnold & Sears 2006). Ubiquitinylation of cMYC at K48 by the SCF-FBW7 complex requires phosphorylation of threonine 58, but occurs only if serine 62 is first dephosphorylated by the PP2A phosphatase (Vervoorts et al. 2006).
Chromatin-modifying enzymes are susceptible to environmental agents and metabolite fluctuations. For example, a combination of genetic and biochemical experiments has shown that vernalization in flowering plants requires methylation of specific histone arginine and lysine residues (Finnegan & Dennis 2007; Schmitz et al. 2008), revealing a link between temperature and chromatin state. Chromatin-modifying enzymes are also susceptible to the concentrations of various metabolites. The kinases, acetylases and methylases that act on histones and TFs are all dependent on high-energy cosubstrates (figure 2), the levels of which can affect their activities. However, for one enzyme family at least, specific mechanisms are in place that allow a more subtle response to metabolic change. The NAD-dependent class III deacetylase SIRT1 has been shown to act on both histones and TFs such as P53 (Vaziri et al. 2001) and AR (Fu et al. 2006) and provides an intriguing link with the metabolic state of the cell (Rodgers et al. 2005). A high NAD/NADH ratio enhances SIRT1 activity, with deacetylation of AR and diminution of its growth-promoting activity. Conversely, low levels of NAD, or high levels of the inhibitor (and SIRT1 product) nicotinamide, suppress SIRT1 activity and hence can enhance the acetylation-dependent activities of AR. SIRT1 may act as a sensor of the redox state of the cell (Fulco et al. 2003). This is particularly significant in light of the long-standing observation that many cancers, including AR-dependent prostate cancer, show enhanced glycolysis, even under aerobic conditions, with a consequent diminution in NAD and enhanced lactate levels (Altenberg & Greulich 2004; Baron et al. 2004).
Intriguingly, a second product of the SIRT1-catalysed deacetylation reaction, the NAD metabolite O-acetyl-ADP ribose (OAADPR), binds selectively to the macro-domain of the histone variant macroH2A, a marker of heterochromatin (Borra et al. 2004; Kustatscher et al. 2005; Tong et al. 2009). OAADPR binding induces only a subtle structural change and the functional outcomes remain uncertain. However, the finding that OAADPR is bound by the splice variant macroH2A1.1 and not by a second variant, macroH2A1.2, suggests a high level of binding specificity (Kustatscher et al. 2005). The catalytic mechanism of SIRT1 is far more complex than necessary to deacetylate proteins; the 11 class I and class II deacetylases are straightforward hydrolases without obligatory cosubstrates (Marmorstein & Trievel 2008). However, it is interesting that when these deacetylases are assayed against native histone substates (rather than the commonly used synthetic peptides), their activity is enhanced by ATP and chaperone proteins such as the stress response protein HSP70 (Johnson et al. 2002). Collectively, these findings suggest that the mechanisms of action of at least some protein deacetylases have evolved to provide a link between intermediary metabolism, or environmental components, and gene function.
For some TFs, there is evidence to suggest that histone-modification state is somehow linked to their selective binding. cMYC is a member of the MYC/MAX/MAD family and forms a heterodimeric complex with MAX to activate the expression of a diverse range of genes. Deregulated expression of c-MYC has been documented in a wide range of human malignancies (Vita & Henriksson 2006). Like many TFs, MYC has the potential to target a large proportion (11%) of all genes in the human genome (Fernandez et al. 2003), but the set of genes to which it actually binds in any particular cell is much more restricted and regulated by a variety of factors, including interacting proteins. For example, the MAD family of transcriptional repressors are, like MYC, MAX-binding proteins and antagonize the activity of MYC by competing for MAX binding at E-box sequences in target gene promoters, actively repressing transcription of MYC target genes (Adhikary & Eilers 2005). The specificity and affinity of MYC binding may also be influenced by the configuration of the chromatin packaging at potential binding sites, and particularly by patterns of histone modification (Guccione et al. 2006). MYC was found to bind E-boxes in regions enriched for several histone modifications including acetylated H3 (specifically H3 acetylated at lysines 9, 14 and/or 18), but showed the strongest association with H3 tri-methylated at lysine 4 (H3K4me3). All these modifications are generally associated with relatively ‘open’ euchromatin. Reciprocally, MYC binding was inversely correlated with the repressive polycomb group mark H3K27me3 (Guccione et al. 2006). Whether these correlations reflect a specific underlying mechanism or are simply due to overall differences in chromatin compaction, and hence accessibility, remains to be seen.
A second example is provided by the pioneer factor FoxA1, a factor central to certain oestrogen receptor (ERα) functions (Carroll et al. 2005; Laganiere et al. 2005). FoxA1 binds with a high specificity to a genomic consensus sequence, but, as with other factors, only a small proportion of possible sites (3.7%) are actually occupied and chromatin immunoprecipitation (ChIP) analyses show that these occupied sites are significantly enriched in H3K4me1 and H3K4me2 (Lupien et al. 2008). Knock-down of FoxA1 does not alter the levels of these modifications, indicating they are present prior to FoxA1 binding, presumably to facilitate preferential recruitment. Significantly, over-expression of the histone demethylating enzyme KDM1/LSD1 decreased levels of H3K4me2 and significantly inhibited FoxA1 binding to chromatin (Carroll et al. 2005; Lupien et al. 2008).
Recent evidence indicates that information encoded in the DNA itself, beyond the consensus-binding sequences, also plays a role in determining TF binding. Odom and colleagues mapped the binding sites of selected TFs across human chromosome 21 (chr.21), first when it was present in wild-type human fibroblasts and then when it was the only human chromosome in mouse × human hybrid fibroblasts (Wilson et al. 2008). In the latter case, TFs and epigenetic control elements originated almost exclusively from the mouse. Surprisingly, the distribution of TFs across chr.21 in a mouse background closely resembled that seen in exclusively human cells, and differed from the distribution across extensive regions of the mouse genome homologous to chr.21 (Wilson et al. 2008). The authors conclude that information encoded in DNA, beyond the genetic code itself, is a major determinant of TF positioning. The code(s) involved remain to be deciphered.
The simplest way in which the nucleosome can influence TF binding is by being positioned such that the recognition sequence is tightly associated with the histone core and inaccessible to the TF. It is known that nucleosomes can occupy defined positions at the promoters and control regions of certain genes (Mellor 2005; Montecino et al. 2007; Tirosh et al. 2007; Jiang & Pugh 2009) and that certain DNA sequences strongly favour the placement of a nucleosome while others discourage it, probably because they are less ‘bendable’ (Tirosh et al. 2007). Nucleosome positioning is determined by multiple factors, but there seems no doubt that DNA sequence plays an important role, though the sequences involved are complex and the formulation of rules is challenging (Segal et al. 2006; Kaplan et al. 2009). Thus, at least for those TF-binding sites, and indeed those gene promoter regions, whose accessibility is regulated by nucleosome positioning, it is likely to be the DNA sequence that is the primary determinant of nucleosome placement. The histone octamer itself is likely to play a functional role that is modulated, in some instances at least, by post-translational modifications. DNA sequence, histones and chromatin-modifying enzymes are all involved in producing the functional end result.
TF function, and thus correctly regulated gene expression, seems to require an integrated network of genetic and epigenetic components, comprising a triumvirate of DNA, TFs and chromatin, with none acting in isolation or having priority over the other two (figure 1). Thus, while it is common to distinguish between genetic and epigenetic processes, the former being based directly on information encoded in the DNA sequence and the latter being those necessary for the interpretation of this information (Holliday 2006; Ptashne 2007), the two processes are often so closely interlinked and interdependent that attempts to tease them apart are problematic and potentially misleading.
Epigenetic effects are often heritable, in the sense that they are passed on from one cell generation to the next. Proof of principle comes from X-inactivation, the process by which most genes on just one of the two X chromosomes in female cells are inactivated early in development (Heard et al. 1997). The X to be inactivated is chosen at random in each of the 20 or so cells of the (mouse) inner cell mass (cells that will go on to form the complete embryo). Once chosen, the X undergoes a series of epigenetic changes, including various changes in histone modification and composition and increased DNA methylation at selected regions. Once in place, these changes persist, in all the progeny of the original cell, though the lifetime of the organism. Under normal circumstances, reactivation occurs only in the primordial germ cells. Thus, epigenetic silencing, particularly if it involves multiple layers of epigenetic changes, can be both heritable and stable.
X inactivation can be seen as the answer to a very specific situation (i.e. a chromosome imbalance between males and females), but all cells must have access to mechanisms by which epigenetic properties can be passed on to daughter cells, if only as a means of retaining cell identity. It seems reasonable to assume that cells are defined by the tissue-specific genes that they express and, if a cell is to retain its identity through DNA replication and mitosis, then this characteristic pattern of gene expression must be maintained. This is sometimes called ‘cellular memory’. Many characteristics distinguish active from inactive genes, and it is not unreasonable to propose that transcriptional states are faithfully transmitted, by default, with genes that are on staying on and genes that are off staying off, unless (differentiation) signals tell them to do otherwise. Formulated in this way, the issue of cell memory becomes rather simple. Unfortunately, this analysis avoids some complicating facts.
A cell can usefully be seen as a robust, dynamic system in which the concentration of each component (protein, RNA transcript metabolite, etc.) will vary, through time, within set limits. These limits are determined by the homeostatic checks and balances that operate across the system, as well as by stochastic variations in gene expression levels, sometimes called transcriptional ‘noise’ (Raj et al. 2006; Maamar et al. 2007). Thus, there is not one state that defines, for example, a fibroblast, but a vast number of different states in which the components vary, but within the limits that operate for the (fibroblast) system as a whole. This situation was illustrated some years ago by microarray expression analysis to quantify RNA transcript levels in yeast. Transcript levels for individual genes varied over a wide range and, remarkably, 80 per cent of genes classified as ‘active’ in the culture as a whole, had transcript levels between 1 and 0.1 on a per cell basis (Holstege et al. 1998; Holland 2002). PCR analysis shows that many transcripts are present at even lower levels (Holstege et al. 1998; Holland 2002). This could be due to the extremely low stability of some mRNAs, with their protein products being rather more stable, but it also raises the possibility that many transcripts are present in some cells, but not in others. On this interpretation, some active genes have no transcript in most cells because the gene is simply not being transcribed in these cells because, in turn, the gene product is already present in sufficient quantity.
This interpretation of low transcript levels suggests that there should be a wide variation in levels of individual transcripts (or their products) from one cell to another. Such analyses are experimentally challenging, but recent results show that cell–cell variability can be enormous for both RNA transcripts and protein products (Raj et al. 2006). For example, flow cytometry has been used to measure levels of individual GFP-tagged proteins in yeast on a cell by cell basis (Newman et al. 2006). There were dramatic differences, from one gene to another, in the extent to which their protein products varied from cell to cell. This variability (noise) was closely correlated with both the mode of transcription and the function of the protein product. Proteins that respond to environmental changes were particularly ‘noisy’.
The enormous variability in transcript levels from one cell to another can be interpreted in two general ways. The first is that gene transcription is inherently variable (noisy), governed by chance and perhaps occurring in temporally discrete bursts (Raj et al. 2006; Raj & van Oudenaarden 2008). While this seems at first an unlikely strategy, it can, by generating a variable population, provide a colony of cells with a cost-effective way of protecting itself against the vagaries of environmental change (see below for an example). On the other hand, cell–cell variability could be the inevitable result of a carefully controlled process of cellular homeostasis. Under this interpretation, many, perhaps most, genes are transcribed only when their product (RNA or the protein translated from it or a dependent metabolite) has fallen below a set minimum level. Transcription continues until the maximum level is reached, when the gene is switched off again. Depending on a variety of factors, such as the stability of the mRNA, protein product or metabolite, the on–off cycle may happen several times in a single cell cycle, or once in several cell cycles. The latter presents a situation where the overall (population-wide) transcript level may be less than one per cell.
Whether gene expression is essentially stochastic or carefully regulated through maintenance of cellular homeostasis, or a combination of both, the issue of ‘cell memory’ changes its character when viewed from this systems biology perspective. Heritability must be seen as a property of the whole system, with the transcription of each gene being determined by the requirements of the system, perhaps over a period of several cell cycles. It may be that genes that are classed as active when the whole cell population is examined, are transcriptionally silent in many individual cells, perhaps for one or more complete cell cycles. What is inherited cannot be the transcriptional state itself, but perhaps the ability to respond appropriately to signals that reflect the level of specific components of the system. The components and signals involved inevitably vary from one gene to another and there may be no unifying mechanism.
Cells are programmed to respond to specific environmental signals. Patterns of gene expression in single-celled eukaryotes change in a specified way depending on available nutrients (Holstege et al. 1998; Bennett et al. 2008) while the cells of higher eukaryotes progress down pathways of differentiation in response to specific signals, often from their neighbours. Once a cell is primed to respond, then even a simple chemical such as retinoic acid can initiate a dramatic change in gene expression patterns and cell phenotype (Muller 2007). Conversely, cells seem to be remarkably resistant to change when exposed to agents to which they have not been primed, even when these agents generate what appears to be a major epigenetic change. As an example of this, I will explore how cells respond to growth in the presence of HDAC inhibitors such as sodium butyrate, reagents that cause global hyperacetylation of all four core histones (figure 2). These reagents are not just of experimental interest. The salts of various short-chain fatty acids, including sodium butyrate, are present at millimolar concentrations in the large intestine in humans and rodents, largely produced by endogenous bacteria (Pryde et al. 2002; Louis & Flint 2009). There is evidence to suggest that their intra-intestinal concentrations are influenced by diet and they have been implicated in protection against colon cancer (Dashwood & Ho 2007; Waldecker et al. 2008). The branched chain analogue VPA, also an HDAC inhibitor, is an effective and widely used anti-epileptic, well tolerated by healthy adults, but a known teratogen (Phiel et al. 2001).
Increased levels of histone acetylation are characteristic of actively transcribed genes and one might expect HDAC inhibitors to cause a major upregulation of gene expression and serious disruption of the properties of the cell. In fact, a variety of studies have shown that only a small proportion of genes show altered transcription in response to HDAC inhibitors, and among the genes that do change, downregulation is as common as upregulation (Peart et al. 2005). Consistent with this, effects on cell behaviour are usually modest, with slowed cell cycle progression and, on prolonged exposure, increased frequency of apoptotic cell death being common responses (Phiel et al. 2001).
A possible explanation for this limited response to environmentally induced histone hyperacetylation comes from the analysis of histone-modification levels at individual gene promoters by ChIP. In a variety of cell types, exposure to HDAC inhibitors caused no increase in histone acetylation across the great majority of genes tested, even those that showed increased expression (VerMilyea et al. 2009; M. D. VerMilyea 2008, unpublished data). It seems that the global histone hyperacetylation detected by western blotting of bulk histones is confined largely to non-genic regions, with the majority of genes remaining unaffected. The reasons for this remain to be defined, but may reflect differences in the turnover of histone acetates from one genomic region to another. The results illustrate the ability of adult and embryonic cells to retain their characteristic phenotypes even in the face of what seems to be a major, environmentally induced epigenetic change.
However, in specific cell systems, the physiological response to HDAC inhibitors can be more dramatic. Studies of epidermal stem cells have provided an intriguing illustration of the relationship between a TF, chromatin modifications and adult stem cell differentiation (Frye et al. 2007). Quiescent stem cells are induced to leave their niche in the interfollicular epidermis and hair follicle bulge by activation of cMYC, an oncogene and TF. The process is accompanied by globally increased H4 acetylation and di-methylation of H3K9 and H4K20. Remarkably, induction of histone hyperacetylation by treatment with the HDAC inhibitor Trichostatin A (TSA), either amplified the differentiation promoting effects of cMYC, or substituted for it in inducing epidermal differentiation. As noted earlier, if a cell is primed to respond to a defined and specific signal, then a generalized environmentally induced change might also trigger that response.
It may be that only a subpopulation of cells in any group of cells are ‘primed’ to respond in a particular way, possibly as a result of cell–cell variation or transcriptional noise. While this is pure speculation in higher eukaryotes, there is evidence for just such an effect in Bacillus subtilis. When a culture of B. subtilis moves into quiescence, usually through nutrient depletion, a proportion of cells become able, through expression of several specific genes, to take up foreign DNA from the surroundings, a state described as ‘competence’. The switch to competence is controlled by a master regulator protein, ComK. Expression of ComK varies widely from one cell to another, i.e. its transcription is noisy. As the culture moves into quiescence, cells that happen, by chance, to be expressing relatively high levels of the ComK competence regulator can move into a state where ComK enhances its own transcription through a positive feedback loop. This leads to a rapid increase in ComK levels, which pushes the cell into a stable ‘competent’ state. The opportunity for cells in the culture to switch to competence persists for approximately 2 h, by which time approximately 15 per cent of cells have become competent. Thereafter, these cells remain competent, but no further switching occurs (Maamar et al. 2007). Thus, noisy transcription had provided a subpopulation of ‘primed cells’ that could respond to a generalized environmental stimulus by differentiating into an altered state.
Recent work has shown that Homeotic genes, specifically the mouse Hoxb cluster, are unusual in that, in both embryonic stem (ES) cells (Chambeyron & Bickmore 2004) and the pre-implantation embryo itself (VerMilyea et al. 2009), they show increased histone acetylation following treatment with HDAC inhibitors, including VPA. The increased acetylation is not accompanied by any immediate increase in transcription, which remains undetectable in both embryos and ES cells. Histone hyperacetylation is clearly not sufficient to override mechanisms responsible for Hox gene silencing in the early embryo, but it is interesting that, despite their silent state, histone acetates across the Hoxb loci are turning over rapidly. Given the crucial importance of the Hox genes as determinants of positional and temporal gene expression in the embryo, and their ability to induce a major morphological change when their function is subverted, the finding that their chromatin is unusually susceptible to environmentally induced change is of some interest. Remarkably, when mouse embryos were cultured from the 8-cell stage to morula in the presence of VPA, and then further cultured, in the absence of inhibitor, to the blastocyst stage, acetylation at Hox gene promoters was always higher in blastocysts derived from VPA-treated morulae than in their untreated counterparts. Thus, the environmentally induced change in histone acetylation has been passed on, through mitosis, to a later developmental stage in the absence of any change in transcription. Whether this change affects the timing or location of Hox gene expression later in development (i.e. at stages when Hox genes are normally induced) remains to be seen and requires reimplantation of the cultured embryos. However, the observation shows that mechanisms for the inheritance, through mitosis, of induced histone modification are present in the early embryo and are not transcription dependent. In contrast to the situation in the pre-implantation embryo, experiments in the author's laboratory have so far shown that the VPA-induced hyperacetylation of Hox gene promoters in cultured ES cells, derived from the inner cell mass of the blastocyst, is not heritable (H. Stower & B. M. Turner 2008, unpublished data). Thus, as with the transcriptional responses to HDAC inhibitors described earlier, heritability seems to be dependent on both cell type and/or developmental stage.
The ability of HDAC inhibitor to induce a mitotically heritable change in histone acetylation and gene expression was first demonstrated in the yeast Schizosaccharomyces pombe (Ekwall et al. 1997). Growth for several cell cycles in the presence of the HDAC inhibitor TSA induced hyperacetylation and transcription in normally silent test genes inserted into centric heterochromatin. The active, hyperacetylated state, though spontaneously reversible at low frequency, was retained through many cell cycles in the absence of inhibitor. However, because acetylation and transcription remained closely linked throughout these experiments, it was not possible to determine which of these two factors was the primary determinant of heritability. More recently, nuclear transplantation in Xenopus has been used to show that some genes (e.g. the endodermal gene edd) can retain a memory of an active gene state, even in an inappropriate (e.g. non-endodermal) cell lineage; the memory can be transmitted through up to 24-cell generations from zygote to tadpole (Ng & Gurdon 2008). What makes this particularly significant is that through the first 12 cleavage divisions of the Xenopus embryo, there is no genomic transcription, so the memory mechanism involved does not require active transcription. Chromatin seems to have a role in this memory in that the variant histone H3.3 and specifically its methylatable lysine 4 residue, seems to be necessary for re-expression (memory) of the active state after progression through the early cleavage cycles. H3.3 associates preferentially with active genes (Ahmad & Henikoff 2002; McKittrick et al. 2004) and may play a role in the maintenance of an active state, even in the absence of ongoing transcription.
If mitotically heritable changes are induced in germ cells, then there is the potential for transmission through meiosis to succeeding generations (figure 1). It is well known, through many years of work on imprinted genes (i.e. genes that are differentially expressed in offspring depending on whether they were transmitted through the maternal or paternal germ line) that epigenetic effects can be transmitted through the germ line, though the mechanisms remain mysterious (Jaenisch & Bird 2003; Reik et al. 2003; Santos & Dean 2004). DNA methylation is likely to be involved, but seems not to provide a complete explanation.
Attempts to demonstrate experimentally the germ-line inheritance of induced phenotypic changes are fraught with difficulty. It has been claimed that exposure to the fungicide and endocrine disruptor vinclozolin at specific stages of embryonic development can trigger changes in male fertility and reproductive behaviour that are heritable, over several generations, through the male germ line (Anway et al. 2005, 2008). Perhaps inevitably, the interpretation of these difficult experiments remains controversial (Schneider et al. 2008). In a different approach to the same issue, statistical analysis of disease susceptibilities in an isolated human population in northern Norway revealed an intriguing correlation between age of death from specified diseases and the nutritional status (i.e. success or otherwise of the harvest) of the grandparental generation (Kaati et al. 2007).
Transmission through the male germ line presents additional problems for epigenetic inheritance. Sperm DNA is in a particularly condensed state, with the great majority of histones replaced by protamines. Within minutes of fertilization, sperm DNA is repackaged with maternal histones, followed by an overall loss of methylated cytosine. However, it is likely that some regions (imprinted genes perhaps) are protected from demethylation (Santos & Dean 2004), while careful analyses have shown retention of a small amount of histone in sperm chromatin (Gatewood et al. 1987, 1990), with enrichment of selected variants, such as H3.3 and H2AZ (Ooi & Henikoff 2007). Further, H3.3 in sperm is rich in modifications associated with transcriptionally active chromatin, such as methylation of lysine 4 (Ooi & Henikoff 2007). Sperm histones may be associated with specific genes, perhaps those that need to be expressed very early in zygotic development, but this remains to be definitively shown.
While acknowledging the complexity and experimental challenges posed by work on epigenetic inheritance, there is now no reason to dismiss it because (potential) mechanisms do not exist. If environmental agents can induce a heritable change in, for example, histone modification in somatic cells, then it is likely that it can also happen in germ cell precursors and be transmitted to the germ cells themselves and thence to the zygote. Effects on the developing embryo and the adult organism will depend on the genes involved, but in the case of the Hox genes could be far reaching. It has been suggested that the teratogenic effects of VPA might be mediated, in part at least, through disruption of Hox gene expression (Faiella et al. 2000; Duncan 2007). While the VPA effects studied so far are exerted in the (early) embryo itself, it is conceivable that a heritable change induced in the germ cells of either parent and transmitted to the zygote could exert an effect. If the epigenetic change were to persist, through multiple mitoses, in the germ cell of the next generation, then one has true epigenetic inheritance.
It has been shown that induced epigenetic changes can be inherited through mitosis, and plausible mechanisms exist by which epigenetic changes could be inherited through either the male or female germ lines. However, if environmental changes are to lead to heritable changes that persist over many generations, and perhaps even influence evolutionary change, then they must surely, at some stage, lead to changes in the DNA sequence itself that mimic, functionally, the initiating epigenetic change. Are there any mechanisms by which a conversion from epigenetic to genetic information might occur? One possibility is presented by the enzyme-catalysed methylation of cytosines in DNA.
Methylation of cytosine at carbon 5 of the pyrimidine ring (5′meC) is a relatively frequent modification of DNA in many, though not all, higher eukaryotes. It is put in place by well-characterized DNA methyltransferases and is generally a stable modification, though rapid demethylation occurs in specific physiological situations. For example, in the mouse zygote the paternal genome is demethylated shortly after fertilization (Morgan et al. 2005). Demethylation is problematic owing to the very high energy required to split the C–C bond and the mechanism remains controversial. It may involve complete removal of methylated cytosine and replacement by the unmethylated base using enzymes of the DNA repair system (Ooi & Bestor 2008; Gehring et al. 2009). In mammals, cytosine methylation occurs almost exclusively at CpG dinucleotides, reflecting the specificities of the enzymes involved. In addition, some DNA methyltransferases (e.g. Dnmt1 in mice) preferentially methylate the cytosine of a CpG dinucleotide if the cytosine on the complementary DNA strand is already methylated. They are referred to as maintenance methylases. This catalytic preference is complemented by proteins that bring the enzymes to hemi-methylated sites (Sharif et al. 2007) and by interactions between different methyltransferases (Liang et al. 2002), collectively ensuring that patterns of DNA methylation are retained through DNA replication. Thus, a mechanism for the inheritance of DNA methylation through the cell cycle is built into the enzymology of the system.
In invertebrates, DNA methylation is confined to a small fraction of the genome or cannot be detected at all. In contrast, in vertebrates, DNA methylation is distributed throughout the genome and is generally associated with regions in which transcription is suppressed (Bird 1993; Bird & Tweedie 1995). It has been suggested that DNA methylation has evolved to allow the increased efficiency of gene silencing demanded by larger genomes (Bird 1995) and, at least in its role as a silencing mechanism, it seems to have evolved rather later than histone modification (Bird 1995; Weber et al. 2007; Mohn & Schubeler 2009). The mechanisms by which DNA methylation leads to transcriptional silencing remain to be clarified, and it is clear that an increased level of DNA methylation across a region does not, in itself, guarantee that the genes within that region will be silenced (Mohn & Schubeler 2009). However, the correlation between elevated CpG methylation, particularly at promoter regions, and transcriptional silencing remains strong overall. A family of methyl-DNA-binding proteins recognize and bind to methylated CpGs and thereby, often through attracting other proteins, alter the conformation and functional state of the DNA (Dhasarathy & Wade 2008). Thus, like histone modifications, DNA methylation is often likely to exert its functional effects indirectly through the actions of binding proteins.
Methyl cytosine can be regarded as a fifth base in DNA and constitutes a powerful, intrinsically heritable, epigenetic mark. It can also, over evolutionary time, influence DNA sequence. Spontaneous hydrolytic deamination of cytosine, yielding uracil, is an inevitable and frequent mutation. When confronted with a G=U mismatch, repair enzymes recognize the uracil as inappropriate and replace it. On the other hand, deamination of 5′meC yields thymidine and the resulting G=T mismatch is less easy to resolve correctly and will, on occasion, result in the replacement of the G with an A, leaving an A=T base pair in place of the original G=C. This process will lead not only to new mutations, but also to loss of cytosines from CpG dinucleotides. It is likely to explain the unexpectedly low frequency of CpG dinucleotides across vertebrate genomes (Mohn & Schubeler 2009). Remarkably, this low frequency is not genome wide, with selected regions retaining the expected CpG frequency. These regions, referred to as CpG islands (Bird 1986), often incorporate gene promoters and are generally free of CpG methylation (Mohn & Schubeler 2009). It seems probable that the lack of CpG methylation has protected CpG islands from cytosine depletion through deamination, but what has prevented their methylation in the first place? It may be that the evolutionarily more ancient histone-modification system is involved and there is biochemical evidence to suggest that the high levels of H3K4me3 present in many CpG islands (Weber et al. 2007) protect these regions from DNA methylation by blocking the action of DNA methyltransferases (Ooi et al. 2007).
Work in model systems has established strong links between DNA methylation and histone modification, specifically methylation of H3 lysine 9. This was first noted in the filamentous fungus Neurospora crassa where mutation of a gene encoding a histone methyltransferase abolished DNA methylation (Tamaru & Selker 2001). Later work showed that DNA methyltransferase was brought to chromatin in which H3 was tri-methylated at lysine 9 (H3K9me3) by the heterochromatin protein HP1, which binds selectively to H3K9me3 through its chromodomain (Tamaru et al. 2003; Freitag et al. 2004). In the flowering plant Arabidopsis, DNA methyltransferases are also targeted by chromatin, but in this case, the mechanism seems to be more direct with the methyltransferase itself binding to H3 tails methylated at lysines 9 and/or 27 (Lindroth et al. 2004). In both these organisms, DNA methylation is not CpG based, nor is it genome wide, but it seems that a similar link between H3 lysine 9 methylation and DNA methylation exists in the mouse and probably involves HP1, though the details remain to be worked out (Lehnertz et al. 2003; de la Cruz et al. 2007). In many human cancers, silencing of key regular genes has been linked to hypermethylation of CpG island promoters. Whether DNA methylation, in any particular circumstance, is a cause or consequence of transcriptional changes remains uncertain, but it remains an intriguing possibility that specific histone modifications are determinants of DNA methylation levels (Ohm et al. 2007). In all these systems, the interaction between the histone H3 tail and the methylating enzyme is likely to be mediated by other modifications to the local chromatin, and other histone-modifying enzymes, including deacetylases, have been implicated (Lawrence et al. 2004; Probst et al. 2004; Smith et al. 2008).
The complexity of the situation in mammals is exemplified by recent work on silencing of the master regulatory gene Oct4 in mouse ES cells. The histone methyltransferase G9a methylates H3K9 in ES cells, leading to regional heterochromatin formation (involving binding of the HP1 protein) and silencing of early embryonic genes, including Oct4 (Feldman et al. 2006). There is subsequent DNA methylation at these genes and long-term silencing. It was originally thought likely that, as in Neurospora, the DNA methyltransferase was brought to the gene promoter by association with HP1, which bound H3K9me3. However, a catalytically inactive G9a mutant (with a point mutation in its SET domain) did not allow heterochromatinization, but did attract Dnmt3a/3b (de novo DNA methyltransferases) via its ANK domain and did trigger increased DNA methylation and long-term silencing (Epsztejn-Litman et al. 2008). This more detailed analysis shows that while G9a is essential for DNA methylation and long-term silencing, its catalytic activity is not. Perhaps the methylation of H3K9 by G9a was originally the sole mechanism of Oct4 silencing, but was superseded, later in evolution, by the advent of DNA methylation.
Whatever the mechanisms involved, it seems that chromatin and histone modifications can influence the targeting of DNA methylation, and that DNA methylation itself can influence DNA sequence by facilitating C to T substitutions. Thus, it is possible to construct a chain of events, based on experimentally verified biochemical mechanisms, through which an environmentally induced change in the activity of chromatin-modifying enzymes can lead to a change in DNA sequence. If the change occurs in somatic cells, then the resulting mutations might be significant in triggering abnormal cell behaviours and disease states, such as cancer. If the change occurs in germ cells, then the resulting DNA mutations will be passed on to subsequent generations where they might exert a selectable phenotypic change. If the environmental agent that precipitates this chain of events were to be persistent over many generations (e.g. through a period of progressive warming or accumulation of an environmental toxin), then in each generation, the same chain of epigenetic events would be triggered, leading to progressive DNA change in selected regions of the genome. It has been appreciated for many years that environmentally driven, epigenetic processes are a potentially significant force in evolution (Jablonka & Lamb 1995) and the recent characterization of enzyme-catalysed processes regulating chromatin and gene expression is providing molecular mechanisms by which this potential may be realized.
Figure 3 shows the chain of interconnected events whereby an environmental inhibitor of an H3K9 demethylase (e.g. a member of the KDM4 family) might lead to a change in DNA sequence in germ cell precursors. The chain itself is entirely speculative, though the individual links are all well-established mechanisms. Thus, inhibition of KDM4 increases local levels of H3K9me3, leading both to targeted gene silencing (through HP1 recruitment and chromatin condensation) and to DNA methylation, which will itself further enhance silencing. Once started, the sequence of events is likely to be self-supporting, with transcriptionally silent chromatin regions being more susceptible to further DNA methylation. As discussed above, the rare but inevitable hydrolytic deamination of 5′ methyl cytosine generates thymidine and a G–T mismatch that is not always correctly repaired. Thus, there is an enhanced rate of DNA mutation at a genomic region targeted by an environmental agent. These mutations might, in turn, exert an effect on DNA methylation levels by enhancing or suppressing binding of DNA methyl transferases (Dnmt) or on gene silencing or activation by altering nucleosome positioning or TF binding. Again, these are potentially self-supporting cycles. Crucially, over evolutionary time, it may be that the altered DNA sequence itself becomes sufficient to drive, or at least support, the gene silencing that was originally a purely epigenetic event.
A possible chain of epigenetic events through which an environmental agent might trigger a change in DNA sequence. The chain of events shown is speculative but the individual elements are all based on established biochemical mechanisms. The process starts ...
It is interesting to note that changes in nucleosome positioning, dependent on the altered DNA sequence, have been shown to accompany the constitutive activation or silencing of sets of genes during evolution of yeast species (Field et al. 2009). Thus, in an aerobic yeast such as Candida albicans, the DNA sequence of the promoters of normally active respiration genes is such as to inhibit nucleosome assembly, keeping the promoter region accessible for binding of TFs and the transcriptional machinery. In contrast, in anaerobic (fermenting) species such as S. cerevisiae, the DNA of the promoter regions of orthologous respiration genes favours nucleosome assembly, rendering these promoters less accessible and transcriptionally less active (Field et al. 2009). These exciting results establish the potential importance of sequence-dependent nucleosome positioning in evolution. Unfortunately for the hypothetical pathway presented in figure 3, there is no DNA methylation in yeast, so an alternative mechanism would have to be postulated to link epigenetic and genetic changes during evolution of this organism.
The pathway outlined in figure 3 does not require meiotic epigenetic inheritance, though inheritance of the induced epigenetic changes through the germ line, via positioned histones or DNA methylation, would increase the time over which localized DNA methylation could be converted into changes in DNA sequence. Nor does the pathway necessarily support ‘inheritance of acquired characteristics’, at least as conventionally defined. However, if a phenotypic change (an acquired characteristic) were itself to precipitate an epigenetic change of the sort outlined, then a similar series of molecular events could conceivably lead to a targeted change in DNA sequence. The molecular events shown in figure 3 are driven by environmental changes, and provide a molecular adaptation to such change. They provide a possible mechanism by which environmental factors can bring about a targeted (potentially heritable) epigenetic change that can generate an altered DNA sequence. If this triggers, in turn, a phenotypic change, then the usual processes of Darwinian selection will operate. Such environmentally driven epigenetic–genetic changes might generate variation across a relatively narrow range of phenotypes, more or less adapted to cope with a particular environment. Pathways such as the hypothetical example in figure 3 should reduce the number of generations required to fix an adaptive phenotypic change within a population and their existence could, in itself, offer a selective advantage.
Despite the irrefutable power of Darwinian natural selection, it is generally accepted that there are aspects of evolutionary change that are not easily explained by the progressive accumulation of small genetic and phenotypic changes (Muller 2007; Stevens 2009). Chance alone (‘genetic drift’) or dramatic environmental events triggering periods of rapid change may also play a role. The importance of gene control elements as drivers of evolutionary change, and particularly how they might operate during embryonic development, has been emphasized (Muller 2007; Carroll 2008; Stevens 2009). A recent review considers various ways in which an epigenetic change might exert evolutionary effects, with an emphasis on how spreading of suppressive chromatin might generate quantitative changes in gene expression (Zuckerkandl & Cavalli 2007). Perhaps the main value of speculation on evolutionary processes is that it can sometimes suggest experimentally testable mechanisms, in the present case, those by which environmental factors can influence epigenetic/genetic processes leading to a heritable change. Such mechanisms not only have long-term implications for evolutionary change itself, but are of immediate relevance to human and animal health.
I am grateful to Adrian Bird and Steve Busby for comments on the manuscript. Experimental work in the author's laboratory is funded by Cancer Research UK and the EU (FP6, ‘Epigenome’ NoE).
One contribution of 11 to a Theme Issue ‘Impacts of environmental change on reproduction and development in wildlife’.