Empirical evidence for epigenetic inheritance driving evolutionary adaptation

Abstract

The cellular machinery that regulates gene expression can be self-propagated across cell division cycles and even generations. This renders gene expression states and their associated phenotypes heritable, independently of genetic changes. These phenotypic states, in turn, can be subject to selection and may influence evolutionary adaptation. In this review, we will discuss the molecular basis of epigenetic inheritance, the extent of its transmission and mechanisms of evolutionary adaptation. The current work shows that heritable gene expression can facilitate the process of adaptation through the increase of survival in a novel environment and by enlarging the size of beneficial mutational targets. Moreover, epigenetic control of gene expression enables stochastic switching between different phenotypes in populations that can potentially facilitate adaptation in rapidly fluctuating environments. Ecological studies of the variation of epigenetic markers (e.g. DNA methylation patterns) in wild populations show a potential contribution of this mode of inheritance to local adaptation in nature. However, the extent of the adaptive contribution of the naturally occurring variation in epi-alleles compared to genetic variation remains unclear.

This article is part of the theme issue ‘How does epigenetics influence the course of evolution?’

1. Introduction

Selection of beneficial traits depends on the faithful inheritance of phenotypic variation. Heritable variation is both created and transmitted by changes in the primary nucleic acid sequence. While DNA sequences directly encode cellular components and harbour a genetically wired programme of development, additional, more indirect modes of inheritance occur. For instance, once genetic programmes have been established, the maintenance of cellular gene expression states or phenotypic traits can become self-propagated. Maintenance of a certain phenotype is, therefore, not only dependent on DNA-encoded instructions but also on the pre-existing cellular state, e.g. the presence of certain previously produced transcription factors or regulatory enzymes. When these causative factors have any means of self-maintenance through either a feedback loop or self-templated mode of propagation, then such states can effectively become heritable, known as epigenetic inheritance or hysteresis. Consequently, different heritable phenotypes can be produced from the same genetic background depending on prior events, and can contribute to heritable variation. Typically, epigenetic states are not as stable as genetic determinants, but they nevertheless can have a significant phenotypic impact and be inherited for several generations. Therefore, epigenetic forms of inheritance have the potential to influence evolutionary trajectories. In this review, we ask three key questions. (i) What is the molecular nature of an epigenetic state, (ii) to what extent can such a state be inherited, and (iii) what is the evidence that an epigenetically heritable phenotypic state impacts the evolution of a population? We will focus on experimental evidence that offer answers to all three questions.

2. History and definition of epigenetic inheritance

The notion that genes are expressed in concert and collectively form a phenotypic state was formalized in the context of development by Conrad Waddington. His epigenetic landscape is a representation of all the possible phenotypes an organism could develop and is shaped by the interactions of the underlying genetic network [1]. During development, an organism would move through this landscape towards its final phenotypic form. While the original notion of ‘epigenetics’ focused on the emergence of the phenotype from the underlying genetic instructions, modern definitions of epigenetic inheritance largely focus on molecular mechanisms that allow gene expression states to be maintained [2]. Self-reinforcing feedback loops are abundant in biological systems and operate within a given genetic background. These include not only transcriptional programmes but many other aspects of cellular life including self-templating proteins (prions) that affect many cell functions [3] or even cytoskeletal components [4]. It is important to point out at the outset that no epigenetic state is operating truly independently from the underlying genomic instructions. The idea that non-genetic variation has any bearing on evolution is contentious as all phenotypes and cellular components can ultimately be traced back to genetic instructions. Indeed, it is implicit that all cellular components are ultimately encoded by genes. Moreover, and perhaps less obviously, nearly all forms of epigenetic inheritance of gene expression states are nucleated at specific sequences (e.g. transcription factor binding sites). However, what is key is the notion that the same sequence elements nucleate distinct heritable functional states such as whether genes are expressed or not. This is the definition of epigenetic inheritance we adhere to in this context.

3. Cellular machinery involved in epigenetic inheritance

For phenotypic states to be heritable and impact evolutionary trajectories, the causative cellular components require transmission through meiosis or mitosis in the case of vegetatively growing unicellular organisms. Protein feedback loops represent a major mechanism of heritable gene regulation in which a protein acts as an indirect or direct transcriptional activator of the gene that encodes its own synthesis (see [5] and extensively reviewed in [6]) or as inducer of conformational change of proteins in a self-templating manner, e.g. prions [3]. In this case, the protein itself is the heritable factor that can be passed on through the cytoplasm both in the cases of meiosis and mitosis (figure 1a). Regulatory RNAs follow a similar scenario. siRNAs, miRNAs or piRNAs are short 20–25 nucleotide-long RNAs that bind to the mRNA during transcription through complementary binding, which results in mRNA cleavage via the RISC ribonuclease complex [8,9]. During meiosis, regulatory RNAs can be maternally inherited via the cytoplasm of the oocyte or even paternally in mature sperm [10].

Figure 1.

Molecular mechanisms of epigenetic inheritance. (a) In protein feedback loops, a protein (blue circle) directly or indirectly induces the transcription of its own gene (green box), resulting in positive feedback. (b) Maintenance of DNA methylation is enabled through the recruitment of DNA methyltransferases (light brown) that bind hemi-methylated DNA (darker brown circles), which methylate the corresponding unmethylated cytosine on the complementary strand. (c) Unacetylated histones, shown as nucleosome complexes (orange circles), recruit silent information regulator (SIR) protein complexes that deacetylate neighbouring nucleosomes thereby maintaining and spreading the hypo-acetylated state. Conversely, methylated histones recruit histone methyltransferases (dark grey) that propagate the methyl mark (light grey) to neighbouring nucleosomes, countering the SIR complex activity. (d) Basic structure of a generic bistable switch. ON and OFF states represent gene expression states. TF1 and TF2 are hypothetical transcription factors that have a negative and positive effect on transcription, respectively (diagram templated on [7]).

In addition to these trans-acting soluble factors, local chromatin structure (e.g. DNA methylation or histone modifications) can be heritable. Cytosine methylation is a classic epigenetic modification as it is stable, faithfully replicated and has a clear phenotypic impact [11,12]. Methylation by DNA methyltransferases is targeted to specific genomic loci through, for example, transcription factors [13]. Once established, the palindromic nature of target sequences (such as CpG sites) allows faithful copying of methylation onto the nascent DNA strand during DNA replication and cell division (figure 1b) [14]. DNA methylation strongly correlates with gene silencing through the recruitment of a host of factors that modify and compact chromatin. Examples include the classic case of X-inactivation [15], involving DNA methylation as well as layers of synergizing chromatin silencers. Similarly, genes can be marked in the germline by DNA methylation, called imprinting, resulting in a silent state that persists in the early embryo in a parent of origin-dependent manner [16–18].

Although more poorly understood, transmission of modifications of histones is similar, in principle, to DNA methylation. Histones are proteins that form heterotetrameric complexes, called nucleosomes, that are stably associated with DNA and transmitted through cell division cycles [19,20]. Some of these histones (particularly H3 and H4) have extensive N-terminal extensions protruding from the globular complex that can be modified post-translationally by addition of, among others, phosphoryl, acetyl, methyl groups or ubiquitin [21]. The effect of histone modifications on gene expression is highly context-dependent and varies with both the position and type of modification [22]. Acetylation events on histone H3 typically correlate with active gene expression, while methylation can have opposite effects depending on the residue. Most prominently, modification of lysine 9 or 27 is associated with repression, while the same modification on the lysine 4 or 36 is linked with gene activation [23]. The principle of inheritance of histone modifications is similar to DNA methylation. The modified histones tend to be distributed randomly between cells during DNA replication [19] and act as sites of recruitment for histone modifying protein complexes that, in turn, modify neighbouring histones, which reestablishes the local chromatin state [24] (figure 1c, left). For instance, in the case of H3K9 methylation, the core of the feedback mechanism resides in the ability of the writer of the mark to also read it [25], although other readers contribute as well for efficient spreading and inheritance [26]. Furthermore, sequence elements are key to target silencers to specific domains. In the case of the fission yeast heterochromatin, a key recruitment platform is the nascent mRNA that, when targeted by small regulatory RNAs, forms a platform for histone modifiers that locally modify chromatin [25]. A similar RNAi-mediated mechanism for targeting DNA methylation has been identified in plants [27]. Maintenance of the silent state is typically not achieved by one modification but rather involves a combination of some or all of the components mentioned above [28–30], within a domain across the affected gene.

In addition to positive feedback to maintain a transcriptional state, negative feedback co-occurs that counteracts the opposite state (e.g. as in figure 1c, right). In such cases, activators and silencers not only counteract each other but also regulate each other (figure 1d). This renders feed-forward loops strongly self-reinforcing. Such double negative feedback results in either state being self-reinforcing, creating bistability where effectively genes are in a heritable ON or OFF state [31]. In effect, just as genetic variation creates different gene alleles, these heritable gene expression states create stable epi-alleles, although the rate of switching (epimutation) between different gene expression states is typically several orders of magnitude higher than the genetic mutation rate.

(a). Transgenerational transmission and phenotypic effect of DNA methylation

Methylation of DNA is one of the most extensively studied mechanisms of epigenetic inheritance. Studies in plants were one of the first to identify the heritable nature of DNA methylation. Plants lack a clear separation between soma and germ cells early in development, unlike most animal phyla. Possibly because of this, epigenetic marks that are established in somatic cells are more readily transmitted to germ cells and to the offspring. Studies in the toadflax, Linaria vulgaris, offer a classic example of a heritable phenotypic effect of DNA methylation [32]. This plant can exist in two distinct morphological forms caused by different expression levels of Lcyc gene. The active, unmethylated state of the gene results in the canonical development of corolla tube flowers with clear dorsoventral asymmetry, while an inactive, methylated Lcyc results in the formation of flowers with radial symmetry [33]. Without knowing the underlying cause of the phenotypic switch from one form to the other, de Vries [34], a pioneer in mutation theory, estimated the rate of the switching to be 1% per generation, which represents the first estimate of an epigenetic switching rate. More recent time course experiments in Arabidopsis found transition rates on the order of 10⁻⁴, where the rate of demethylation is three times higher than the rate of methylation [35]. The highest rates are observed in genes, while the lowest are in the transposable elements (TEs), possibly reflecting the necessity for the silencing of TEs and preventing their movement within the genome.

In addition to natural occurrence of stable variants of DNA methylation, experimental creation of variants has helped to discover the contribution of DNA methylation to heritable phenotypic variation. To this end, epigenetic recombinant inbred lines (epiRILs) were constructed by inbreeding Arabidopsis lines that lack DNA methylation machinery with a wild-type strain [36]. Through successive rounds of inbreeding, an epigenetically diverse, yet genetically similar population is created. By mapping genomic sites of DNA methylation after each round of inbreeding, differentially methylated loci were found to be maintained for up to seven generations [37]. More importantly, particular epigenetic loci explained up to 90% of the broad-sense heritability of flowering time, and more than 50% of the observed phenotypic variance [38]. Moreover, experimental alteration of DNA methylation was shown to alter many ecologically important traits, such as flowering time, in Arabidopsis thaliana [39]. However, it still remains unclear to what extent these epigenetic marks are truly independent from any residual variation in the underlying sequence [40]. Differentiating this is a major challenge going forward.

Stable long-term heritability of DNA methylation extends to yeasts. A recent study reported spectacular transmission rates in Cryptococcus neoformans [41], maintained by a single methyltransferase, Dnmt5 [42]. This protein exclusively recognizes and binds hemi-methylated DNA and maintains DNA methylation in cis without the ability to establish methylation de novo. De novo methyltransferases are lost in this species since its divergence from the common ancestor carrying such activity between 50 and 150 mya ago [41,43]. Methylation has been maintained through the action of both strong epigenetic stability and selection. The rate of loss and gain of methylation marks in the populations of C. neoformans were estimated to be in order of 10⁻⁵ and 10⁻⁶, respectively.

In mammals, the extent of inheritance of DNA methylation appears to be quite limited owing to the efficient removal of DNA methylation, initially in the germline and in the early embryo [44]. Nevertheless, germline transmission of DNA methylation is observed, albeit at dramatically lower transmission rates. A classic case involves the agouti locus in mice. A transposon insertion upstream of this gene results in differential methylation, causing variegating expression of the agouti gene and associated changes in fur coat colour [45–47]. The particular epigenetic state is maternally transmitted for two generations [48]. Importantly, it was shown that the degree of gene silencing and the corresponding fur colour can be modified by methyl donors present in the diet [49,50], highlighting a case of an environmentally induced heritable phenotype. A heritable effect of diet has also been observed in rats. Supplementing diet with variable quantities of protein causes differential DNA methylation patterns of a glucocorticoid receptor gene that persist for two generations [51,52]. In addition to physiology, behaviour can also be affected transgenerationally. Odour-associated fear conditioning in the parental generation was reported to cause a fear response in F2 progeny even though the offspring never encountered the odour-associated threat [53]. The inheritance of this particular behavioural trait is transmitted both paternally and maternally. Although the molecular mechanism of inheritance is largely unknown, DNA methylation was implicated as the underlying cause.

(b). The role of chromatin proteins in epigenetic transmission of transcriptional states

In addition to DNA methylation, local chromatin structure can also be maintained through successive cell division cycles. Particularly, histone modifications can be maintained in a self-perpetuating manner through cell division, as well as, transgenerationally, at least in the case of silent chromatin (see box 1). For example, in fission yeast (Schizosaccharomyces pombe), transcriptional inactivation of a reporter gene inserted in the mat locus is dependent on the methylation of lysine 9 of histone H3, as well as on the presence of chromatin remodelling complexes. This transcriptional state and its causative epigenetic marks are inherited in cis even after a round of meiotic division [66,67]. Artificial reconstitution of an epigenetic state at an ectopic site through insertion of genetic elements [67] or tethering of the lysine 9 histone methyltransferase Clr4 [68] to a reporter gene causes methylation of the surrounding histones and silencing of the reporter, a phenomenon also partially recapitulated in mammalian cells [69]. The silent state is inherited even upon the removal of the initiating sequence or the tethering protein itself. In fission yeast, this particular histone methylation mark is found to be maintained for 20 generations, though this maintenance can be extended for up to 50 generations once a key histone demethylase is removed [68,70]. The observation that histone methylation is reversible indicates that the degree in which it is transmitted may be regulated depending on environmental conditions.

Box 1. A curious histone variant that epigenetically maintains centromere position.

In addition to modification of histones, different allelic variants index the genome into functional domains. For example, the H3 variant CENP-A determines the formation of centromeres [54], chromosomal loci that ensure correct separation of chromatids during cell division. Interestingly, centromere position and maintenance are strongly driven by a protein-based self-propagating mechanism that operates largely independently from underlying DNA sequences that are neither sufficient nor necessary for the formation of the centromere. The most striking evidence comes from artificial seeding of CENP-A at a naive chromatin locus that was found to be sufficient to initiate a centromere [55,56] and to be mitotically maintained through subsequent generations, independently of the initial seed [55]. Central to inheritance is a CENP-A histone-based feedback loop in which CENP-A recruits its own maintenance machinery that ensures further local deposition in subsequent cell divisions [57,58], including through the germline in mammals, as well as Drosophila [59,60]. While CENP-A does not have a direct effect on the transcription, it can impact evolutionary trajectories through alternative mechanisms. Stronger centromeres, i.e. centromeres with more CENP-A, were shown to be preferentially retained in the oocytes during meiosis [61–63]. This causes non-Mendelian inheritance of the genetic variants located on the chromosome with the stronger centromere. This could potentially lead to hitch-hiking of maladaptive alleles and their maintenance in the population despite their deleterious effect on fitness [64]. Moreover, centromere repositioning was shown to create reproductive barriers in fission yeast, suggesting a role of centromeres in the process of speciation [65].

Similar to S. pombe, gene silencing in budding yeast involves histone modification and is observed at three distinct genomic loci: telomeres, mating type loci and the rDNA locus. In this case, silencing is dependent on the activity of the silent information regulator (SIR) protein complex. Importantly, while sequence elements are required to target silencing [71], silencing itself is subject to stochastic switching between active and inactive gene expression state, both of which are heritable for multiple divisions. The silencing itself involves deacetylation of neighbouring histones upon SIR complex recruitment [72]. This, in turn, results in binding of additional SIR complexes leading to spreading of deacetylation and suppression of local transcription [73,74] (figure 1c). The spread of the deacetylation is counteracted by the activity of histone acetyltransferases and histones methyltransferases that modify histones preventing further recruitment of the SIR complex. This is an example of a self-enforcing feedback loop between the activities of histone methylation and deacetylation (figure 1c,d) [75–78]. These stable, yet counteracting forces lead to stochastic switching between active and repressed transcriptional states that define the rate of epigenetic switching [24,79]. The silenced state conferred by Sir activity within the subtelomeric region can be maintained for up to 20 generations as measured by an ADE2 reporter gene that confers colony colour [80]. Further, more recent work that employed a Cre recombinase reporter embedded in the silent yeast mating type locus coupled to a floxed reporter to measure the silencing loss rate estimated the rate to be between 10⁻³ and 10⁻⁴, depending on the mating type locus [81]. In sum, the epigenetic switching rates observed in yeast are several orders of magnitude higher than mutation rates that are estimated to be in the range of 10⁻⁷–10⁻⁸ [82].

Additionally, in budding yeast, a recent discovery found a chromatin-modifying histone deacetylase complex to have prion properties. The prion state induces phenotypic changes that last for over a hundred generations. These include activation of subtelomeric gene expression, linking cytoplasmic prion inheritance directly to chromatin-mediated gene expression [83,84].

Importantly, it should be noted that the rate of gene silencing is not constant as the silencing machinery is itself influenced by environmental factors. For instance, in yeasts or Drosophila, heterochromatin is modulated by temperature changes or nutrient starvation [85–88]. This creates an opportunity for the environment to induce adaptive changes.

In animals, such as Caenorhabditis elegans (see [89]), gene expression patterns can also be transmitted even through meiosis. Transient exposure of these worms to high temperatures leads to derepression of a gene array comprising several copies of HSP90 gene. This activation persists for 14 generations, even after the population is transferred to lower temperature [90]. The phenomenon is dependent on the methylation of lysine 9 at histone H3. Furthermore, C. elegans exhibits a phenomenon of transgenerational inheritance of behavioural patterns [91,92]. Once exposed to pathogenic bacteria, worms learn to avoid it upon re-exposure. This avoidance behaviour is transmitted to the next generation, can last up to four generations and is dependent on Piwi small RNA pathway and histone modifier complexes [92], which engage in self-reinforcing loops as elegantly shown in fission yeast [93,94]. It should be noted that a recent study that analysed the formation and transmission of small RNA epi-alleles in an unbiased manner found most to be transient with limited heritability [95]. Finally, in mammals, despite extensive reprograming of epigenetic marks, certain histones with particular modifications are selectively retained at the promoters of specific genes [96,97], offering an opportunity to prime specific genes for expression in the next generation, although a role in adaptation is yet to be established.

As we have seen, the extent and stability of epigenetic inheritance of gene expression states depend on the organism, the mechanism of inheritance and the environment. Maintenance rates vary considerably, from thousands of generations in fungi to perhaps just a few in mice. Particularly, germline reprogramming in mammals limits the extent of transgenerational effects. However, epigenetic memory, even if short-lived for a few generations, can have profound evolutionary effects. In the next section, we will cover its adaptive value.

4. Evolutionary mechanisms underlying adaptation via epigenetic means

The effect of non-genetic inheritance on populations has been extensively studied from a theoretical standpoint by modelling approaches [98–101]. In principle, phenotypic heterogeneity, established in the population owing to frequent switching between epigenetic states, can provide fitness advantage in fluctuating environments offering a bet-hedging strategy for survival [102]. A fitness disadvantage from producing a maladaptive phenotype in the pre-existing environment is compensated by the potential advantage the phenotype might provide upon the environmental change [103,104] (figure 2). The cost–benefit is balanced by the rate of epigenetic switching. Rates that correspond to the periodicity of the environmental shifts are expected to be positively selected [101]. Indeed, experimental evidence in yeast points in this direction. Populations with high rates of epigenetic switching grow faster in environments with short periodicity, while lower rates of epigenetic switching are beneficial in environments with longer periodicity [105]. Moreover, stochastic epigenetic inheritance can explain population dynamics of C. elegans in fluctuating environments [106].

Figure 2.

Contribution of epigenetic inheritance to adaptation in a fluctuating environment. Ability to quickly switch between two phenotypes caused by epigenetic mechanisms of inheritance (epi-allele 1 and epi-allele 2) may provide a fitness benefit during adaptation to fluctuating environments. The beneficial phenotype in one environment (epi-allele 1 in environment 1 or epi-allele 2 in environment 2) would, owing to constant switching, produce a maladaptive phenotype in a given environment creating a fitness disadvantage. However, this fitness cost would be compensated by the fitness advantage of the phenotype upon the environmental change. Owing to the slow rate of genetic changes, mutations are expected to be less effective for adaptation in a rapidly changing environment, compared to epigenetic switching.

In stable environments, the adaptive advantage of epigenetic switching is less intuitive. Nevertheless, when a population meets a single but sudden change in environmental conditions, epigenetic switching could provide a buffering mechanism. This enables higher survival of a population in the new environment, until the acquisition of a more beneficial mutation, a form of genetic assimilation (figure 3). This might especially be important in scenarios where a population is exposed to a deleterious environment and would go extinct if it remained at its current phenotype. (For more indirect effects of epigenetic inheritance see box 2.)

Figure 3.

Contribution of epigenetic inheritance to adaptation in a novel stable environment. During the process of selection, a maladaptive phenotype (epi-allele 2) would be purged out of the population, causing diminishing in population size. Due to high rate of switching between gene expression states, epigenetic inheritance could provide a switch to a more beneficial phenotype (epi-allele 1), rescuing the population from extinction. The newly acquired phenotype is expected to be unstable and switch back to the maladaptive state frequently, creating a phenotypic heterogeneity in the population. However, the population increase caused by the epi-allele would increase the probability of acquisition of a more stable genetic change (mutation) fixing the beneficial phenotype (genetic assimilation).

Box 2. Indirect effects of chromatin machinery on adaptation.

In addition to modifying gene expression, chromatin components also directly affect mutation rates, thereby increasing the probability of acquiring beneficial mutations. Methylated cytosines are known to convert to thymidine through the process of spontaneous deamination at a higher rate than unmethylated ones [107]. Consequently, methylated DNA has a higher propensity of acquiring mutations [108]. Overexpression of a DNA methyltransferase was shown to increase the mutation rate as a consequence of impaired recruitment of mis-match repair machinery [109]. On the other hand, DNA methylation inactivates TEs and prevents mutagenesis by transposon insertion across the genome, as shown in plants [110,111]. Further, histone modifications of chromatin regions are also associated with differential mutation rates. For example, in human cancer cell lines, the variation in histone methylation can account for 40% of variation in mutation rate [112]. Additionally, chromatin structure is known to affect recombination rate. In yeast, the absence of epigenetic silencing increases recombination events [113], indicating a role for chromatin structure in the maintenance of genome integrity (for a more detailed account, see [114]).

5. Experimental evidence for the role of epigenetic inheritance in adaptation

Thus far, experimental evidence for the evolutionary impact of epigenetic inheritance comes from studies primarily in unicellular organisms. For instance, in Clamydomonas, genetic or pharmacological perturbation of histone acetylation and DNA methylation resulted in impairment in adaptation to different stresses (salt stress, nitrogen starvation and high concentration of carbon dioxide). The genomic and epigenomic data show differences in methylation patterns in the evolved clones, indicating that new epigenetic patterns arose during the experiment [115]. However, the principal challenge in quantifying the adaptive value of epigenetic transmission of phenotypes is disentangling the effects of a pure epigenetic state from any possible underlying genetic changes. Easily controllable clonal populations, such as in budding yeast, offer a possibility to make such a distinction. In Saccharomyces cerevisiae, self-enforcing, protein-based positive feedback loops, in which a protein is directly or indirectly activating its own transcription, increase the adaptation to selective stressors [116,117]. Moreover, adaptation of a yeast strain with this mode of epigenetic regulation evolved to higher levels of drug resistance. The adaptive mutations acquired during this process depended on the presence of the feedback loop [117].

Further, in a recent study, conducted also in budding yeast, a counter-selection system using a URA3 reporter gene was used to determine the role of subtelomeric SIR-dependent epigenetic silencing in adaptation. As expected, the extent of epigenetic silencing scaled with better survival of the population in the adverse environment [118,119]. However, not all rates of epigenetic switching turned out to be beneficial for long-term evolution. Excessive rates delayed the appearance and fixation of any beneficial mutations, whereas low rates decreased the probability of survival. Therefore, what emerged from this study is that there is an optimal epigenetic switching rate that enables faster acquisition of beneficial mutations, thus increasing the rate of adaptation [119]. Moreover, epigenetic silencing was shown to change the spectrum of beneficial mutations. In the evolved clones, capable of epigenetic silencing, adaptive mutations were found in genes involved in epigenetic regulation itself that modulate the rates of epigenetic switching [119].

Similarly, in fission yeast, adaptation experiments using caffeine as selective agent showed that populations can adapt by epigenetically silencing genes known to be involved in caffeine sensitivity [120]. While the predominant means of adaptation was through mutation, heritable islands of histone H3 lysine 9 methylation at particular resistance-associated loci were shown to have adaptive potential [121]. Indeed, introduction of such a mark through artificial tethering of histone methyltransferases in the ancestral background confirmed the causative effect on adaptation. Moreover, the methylation mark was shown to precede adaptation by mutation that further fixed the resistant phenotype [120], offering evidence that heritable silencing can act as a phenotypic buffer (figure 3). Evidence of epigenetically mediated drug resistance has also been observed in the human fungal pathogen Mucor circinelloides. Resistance to an antifungal drug develops either through acquisition of mutations or via epigenetic means [122]. The latter involves the production of small regulatory RNAs leading to transient silencing that reverts upon removal of the drug (i.e. epimutants). This mechanism of resistance is dependent on the RNAi machinery. Sequencing showed that epi-mutants lacked genetic changes, further illustrating the power of epigenetic mechanisms in conferring resistance to harmful drugs.

These examples indicate that epigenetic inheritance can significantly shape adaptation dynamics in both fluctuating and stable environmental conditions. Nevertheless, the functional significance of these findings for explaining the patterns observed in natural populations still remains unclear. In the next section, we will briefly cover the current knowledge of the ecological epigenomics and its evolutionary relevance.

6. Ecological evidence for the adaptive role of epigenetic marks

Current ecological data addressing the adaptive advantage of epigenetic inheritance come largely from analysis and manipulation of DNA methylation patterns in natural populations (e.g. [123]). These studies show varying levels of methylation that are associated with different ecological niches. For example, comparison of fresh-water snail (Potamopyrgus antipodarum) populations from river and lake habitats revealed extensive differences in DNA methylation between populations that correlated with adaptive phenotypic differences [124]. Furthermore, invasive Japanese knotweed (Fallopia) shows changes in methylation in response to new habitats [125] despite very few genetic differences between different locations, highlighting the potential adaptive importance of the DNA methylation marks. Genetic and epigenetic analysis of populations of salt marsh perennials from different sites showed a higher correlation between epigenetic marks and the habitat compared to genetic variants [126]. Similarly, DNA methylation is implicated in adaptation to salinity in Baltic sticklebacks (Gasterosteus aculeatus) [127]. However, the correlation between habitat and epigenetic variants appears species-specific [128,129]. Finally, DNA methylation has also been implicated in eye degeneration in cavefish. Expression of eye development genes and corresponding DNA methylation levels are markedly different between cavefish and zebrafish, despite no inactivating mutations being found in these genes [130]. Interestingly, perturbation of DNA methylation in cavefish through the addition of inhibitors partially recovered the eye size, highlighting the importance of this process in eye development. Based on these findings, one can speculate that eye degeneration in caves, where eyes represent a fitness cost, may be accelerated via epigenetic gene silencing.

Despite growing support from correlative studies in natural populations, the causal link between adaptation in these populations and particular epigenetic marks is not yet established. Furthermore, DNA methylation patterns can be induced by the environment or be determined by the underlying sequence, which complicates the interpretation of the results from these studies. Indeed, the comparison of genetic and epigenetic loci between populations of violet (Viola cazorlensis) showed that 90% of epigenetic variation can be linked to distinct genetic loci [131]. Therefore, in order to disentangle truly epigenetic effects from the genetic ones, studies of natural populations should include analysis of the underlying genomic sequences and establish how genomic variation directs DNA methylation patterns.

7. Conclusion and the way forward

We have discussed several examples where epigenetic control of gene regulation can have profound impact on phenotype and long-term transmission thereof. DNA methylation is most widely studied, both owing to its ease of experimental interrogation and its unusual stability, especially in ecological and evolutionary contexts. Nevertheless, it is important to point out that not every modification is functionally relevant. Recent selection experiments performed in A. thaliana showed significant changes in methylation patterns in the evolved offspring. However, only few of these epigenetic marks were associated with changes in gene expression [132], indicating the majority of methylation marks had no effect on the phenotype. Therefore, further research on the mechanisms of epigenetic control of gene expression is key and analysis of other chromatin states should be included. Nevertheless, it is clear that some epigenetic changes can have significant impact on the phenotype and be inherited for several generations. These epigenetic mechanisms of gene regulation can alter the adaptation dynamics and significantly change the evolutionary outcomes, at least in the case of asexually reproducing organisms. To understand how general this contribution of epigenetic inheritance to evolution is, it is important to perform similar studies on more complex, sexually reproducing organisms. Sexual reproduction likely adds another level of complexity owing to the erasure of epigenetic marks during the process of gametogenesis. Finally, to better understand the contribution of epigenetic phenotypic switching to evolution, it is crucial to quantify the fitness effects of epigenetically determined phenotypes across environments. It is likely that the effect size of functional epigenetic changes tends to be smaller than that of functional genetic mutations. This is because genetic mutations are more likely to alter protein function, whereas epigenetic changes typically alter expression levels. Nevertheless, even levels can have profound effects within gene networks and the resulting phenotype.

Data accessibility

This article has no additional data.

Authors' contributions

This is a review article that has been jointly drafted, edited and written by D.S. and L.E.T.J.

Competing interests

We declare we have no competing interests.

Funding

L.E.T.J. is funded by a Wellcome Senior Research Fellowship. Funding for D.S. is provided by Wallenberg Foundation (project grant no. 2017.0163).