Organisms often respond to environmental changes by producing alternative phenotypes. Epigenetic processes such as DNA methylation may contribute to environmentally induced phenotypic variation by modifying gene expression. Changes in DNA methylation, unlike DNA mutations, can be influenced by the environment; they are stable at the time scale of an individual and present different levels of heritability. These characteristics make DNA methylation a potentially important molecular process to respond to environmental change. The aim of this review is to present the implications of DNA methylation on phenotypic variations driven by environmental changes. More specifically, we explore epigenetic concepts concerning phenotypic change in response to the environment and heritability of DNA methylation, namely the Baldwin effect and genetic accommodation. Before addressing this point, we report major differences in DNA methylation across taxa and the role of this modification in producing and maintaining environmentally induced phenotypic variation. We also present the different methods allowing the detection of methylation polymorphism. We believe this review will be helpful to molecular ecologists, in that it highlights the importance of epigenetic processes in ecological and evolutionary studies.
The interactions between an organism and its biotic and abiotic environment are continuous and diversified. Environmental conditions often change rapidly and in unpredictable ways, challenging the organism’s survival and reproduction. When possible, individuals can move to escape unfavourable conditions. Alternatively, the maintenance of internal equilibrium by physiological homeostasis or individual genetic variability (e.g. Hedrick 1999) can provide a buffer against environmental changes. However, many organisms respond to environmental changes by modifying their original phenotype. This may encompass changes in an individual’s development, morphology, physiology or behaviour (Agrawal 2001; Price et al. 2003).
Alternative phenotypes produced during the lifetime of an individual can be achieved through regulation of the expression of a specific gene or activation of an alternative gene pathway (Schlichting & Pigliucci 1993; Pigliucci 1996). An important property of environmentally induced phenotypes is that the associated variations in gene regulation are not necessarily heritable—the gene is always transmitted but not necessarily its expression state. On the other hand, the capacity to respond to environmental cues is often heritable, indicating a genetic basis for this process. The ability to produce alternative phenotypes could have evolved to maximize the fitness of individuals in variable environments (Dudley & Schmitt 1996; Debat & David 2001).
Although many examples of environmentally induced phenotypes have been described and the associated changes in gene transcription measured (e.g. Mori et al. 2005; Derome et al. 2006; Sumner et al. 2006), the underlying mechanisms responsible for regulating which genes’ expressions should change in response to a specific environmental variation are still poorly understood.
Changes in gene expression may occur through epigenetic modifications (Jaenisch & Bird 2003). Epigenetic modifications refer to changes in gene expression that are stable throughout mitoses but also reversible and that occur without changes in the underlying DNA sequence. The most direct way of tagging a gene for expression or silencing is to place a chemical mark directly on its DNA. DNA methylation is indeed the most studied and probably the best understood type of epigenetic modification (for an overview of other types of epigenetic modifications, see Box 1). Such a mechanism could represent a way to allow phenotypic variability in a changing environment without having to rely on genetic variation.
Box 1. Other epigenetic processes affecting phenotype
DNA methylation is an essential gene regulation process that can influence an individual’s phenotype. However, several taxa, including model organisms such as the fruit fly D. melanogaster, the nematode worm C. elegans, and the yeast S. cerevisiae, have undetectable or very low levels of DNA methylation. These organisms nonetheless display extensive phenotypic variation, indicating that DNA methylation is not the only process responsible for phenotypic variation. In these organisms, as well as in organisms for which DNA methylation is present, other epigenetic processes are important in determining the phenotypic outcome. These other processes may affect gene expression at the transcriptional level, as does DNA methylation, or at the post-transcriptional level.
DNA is intimately associated with histone proteins; modifications of the histone tails are known to control the packaging of DNA, therefore regulating the access to the genes for transcription. Polycomb and Trithorax systems include numerous genes that encode for proteins modifying histone tails to a repressed (barrier to gene transcription) or active (accessible to transcription) chromatin state, respectively. It is thought that these genes are important in regulating the expression of house-keeping genes and developmentally important genes such as the Hox genes (Meissner et al. 2008). This epigenetic system and its effects on phenotype have been extensively studied in Drosophila (e.g. Hauenschild et al. 2008). It is also important to note that DNA methylation and Polycomb/Trithorax are interwoven epigenetic systems (Meissner et al. 2008).
Phenotypic variation may also be due to the release of previously hidden genetic (Rutherford & Lindquist 1998) or epigenetic (Sollars et al. 2003) variation in response to stressful environmental conditions. For instance, under normal environmental conditions, the protein chaperone Hsp90 assures correct folding and function of proteins in spite of mutations the proteins might contain in their sequence. The variability buffered by such a mechanism can be revealed if Hsp90 function is compromised (for instance, following an environmental stress), therefore resulting in an increase in phenotypic variation.
Phenotypic variation is central in ecology and evolution and often plays a role in adaptation, niche shift, population dynamics, and evolutionary diversification (West-Eberhard 1989; Agrawal 2001; Debat & David 2001; Price et al. 2003; Pigliucci et al. 2006). The objective of this review is to present the role of DNA methylation in creating phenotypic variation driven by environmental changes. DNA methylation exists in all living organisms with important differences among and even within taxa. We describe the extent and the differences in DNA methylation across taxa. We then examine the processes responsible for creating variation in DNA methylation and how they link the environment with phenotypic changes through modulation of gene expression. We present the tools and framework available to measure DNA methylation polymorphism in natural populations and to assess its evolutionary importance. Finally, we present how concepts related to phenotypic change in response to the environment, namely the Baldwin effect and genetic accommodation, can be explained by heritable (or not) changes in DNA methylation patterns and we discuss the evolutionary relevance of these epigenetic processes.
DNA methylation, the incorporation of a methyl group (CH3) to specific nucleotides, is the most widespread epigenetic modification. Indeed, DNA methylation is detected throughout all domains of life, in Eubacteria, Archea, and Eukaryotes. The establishment and maintenance of DNA methylation is achieved by specific enzymes known as DNA methyltransferases. The sequence similarity of these enzymes in bacteria, plants, and animals suggests a common origin (Ponger & Li 2005). Among them, the DNMT3 family is responsible for the establishment of methylation patterns on DNA (de novo methylation) and the DNMT1 family ensures the maintenance of these marks during DNA replication. The role of other methyltransferases is less clear (e.g. DNMT2) or is specific to given taxa (e.g. chromomethylases [CMT] in plants or DNMT 4, 5 in fungi; Chen & Li 2004; Ponger & Li 2005).
Though DNA methylation may appear to be a hallmark of all living organisms, numerous differences have been reported among and even within taxa. First, there are differences in which nucleotides are methylated and at what molecular position. Depending on the organism, the methyl group may be incorporated on the N6 position of the adenine or at different positions on the molecular structure of the cytosine (N4 or C5) by using distinct DNA methyltransferases. Methylation of adenine is found in Eubacteria and Archea, but in Eukaryotes it is restricted to some unicellular organisms (e.g. Tetrahymena; Hatmann 2005) and the chloroplastic genome of land plants. However, methylation at the C5 position of cytosine is common throughout all domains of life and is the only DNA modification convincingly reported in multicellular Eukaryotes (Suzuki & Bird 2008 and references therein). The distribution of 5-methylcytosine is not random in the genome, and organisms with DNA methylation differ in the type of sequences that are methylated. For example, methylation occurs mainly at the CpG dinucleotide (adjacent cytosine and guanine linked by a phosphate) in vertebrates but also at CpNpG sequences in plants and at CpNpN (N could be A, C, G, or T) sequences in plants and fungi (Ito et al. 2003).
Organisms also differ in the pattern of DNA methylation across their genome (Suzuki & Bird 2008 and references therein). For instance, in vertebrates, methylated sites are distributed globally across the genome: all types of DNA sequences (genes, transposable elements, intergenic DNA) are subject to methylation. The exception to this global methylation is short unmethylated regions, the CpG islands, that represent only a small fraction of the genome (1–2%) and that are generally associated with housekeeping genes. In other animals, methylation has a mosaic pattern, with methylated domains interspersed with unmethylated domains. The highest methylation levels are observed in plants, where up to 50% of cytosines can be methylated in certain species. In maize, for example, these high levels of methylation are associated with the large number of transposons present in the genome. However, other plants such as Arabidopsis thaliana display mosaic patterns of DNA methylation similar to what is seen in non-vertebrate animals (Chan et al. 2005).
The additional layer of information provided by DNA methylation alters neither the DNA sequence of the gene nor that of the RNA or the protein, but can actually regulate the expression of the gene. First, the level of DNA methylation in the promoter region—a sequence upstream of a gene required for its transcription—is generally negatively correlated to levels of gene expression. Repression of transcription is expected to occur when specific proteins—the methyl-CpG-binding proteins—bind to the methylated promoter instead of the transcription factors and subsequently recruit chromatin remodelling complexes; this action eventually closes the chromatin to gene transcription (Boyes & Bird 1991; Weaver et al. 2004). However, DNA methylation was also found to be targeted on the transcription units of actively transcribed genes in A. thaliana, where it is likely to reduce transcriptional noise by preventing spurious initiation of transcription (Bird 1995; Weber et al. 2005; Zilberman et al. 2006). In addition, it seems that not only the amount but also the pattern of methylation (in terms of which specific CpG dinucleotide is methylated) are important in determining levels of gene expression. For instance, in a study by Weaver et al. (2004), where they compared the methylation status of CpG dinucleotides in the promoter of the glucocorticoid receptor gene when it was expressed and when it was not expressed, the variation in the methylation status of a single CpG dinucleotide was found to be relevant in determining whether the gene would be expressed. Similarly, comparisons between active and inactive X chromosomes in female humans revealed that the active X chromosome was overall more methylated than the inactive X chromosome. This observation first seemed counterintuitive since higher levels of methylation are generally associated with an increase in silencing. However, a closer look at the methylated sequences revealed that the promoters of the inactive chromosome were hypermethylated whereas the extra methylation on the active X chromosome was actually located on the gene bodies (Hellman et al. 2007).
Gene regulation by DNA methylation is involved in different functions (Colot & Rossignol 1999; Suzuki & Bird 2008). First, DNA methylation can serve as a protection system against transposable elements. This role has been convincingly reported in plants and fungi and it might also be present in mammals (Suzuki & Bird 2008). Also, the genes required for transposition of mobile elements are generally heavily methylated for repression. In mammalian dosage compensation (Avner & Heard 2001), DNA methylation is involved in the inactivation of one of the two female X chromosomes, which leads to the expression of a single X chromosome, therefore mimicking the situation prevailing in males. DNA methylation is also involved in genomic imprinting (Li et al. 1992; Wilkins 2005). Genomic imprinting has been detected in mammals and in some plants and is established during gametogenesis, where sex-specific methylation alters the expression of hundreds of genes. In the zygote, imprinted genes are either expressed only from the allele inherited from the mother or from the allele inherited from the father. Most imprinted genes are required for normal development. Finally, the role of DNA methylation is not restricted to endogenous gene regulation. In bacteria, for example, DNA methylation serves to protect the bacterial genome from invasion by extracellular DNA. Indeed, while bacterial restriction endonucleases cleave the foreign DNA, they do not recognize the methylated sequences of the bacterial genome.
How does DNA methylation affect phenotype?
Epigenetic processes are crucial in coordinating changes in gene expression leading to cell lineage differentiation during an organism’s development (Bird 2002). However, DNA methylation is not exclusively influenced by intrinsic signals during development. Indeed, examples of spontaneous or environmentally induced changes in methylation profiles are increasingly reported, and these are the object of the current review. Numerous studies have highlighted the relevance of such processes in creating phenotypic variation (Weaver et al. 2004; Blewitt et al. 2006; Manning et al. 2006; Crews et al. 2007; Kucharski et al. 2008; Jablonka & Raz 2009). Through DNA methylation, the genome can integrate environmental signals and as a result, these extrinsic signals can potentially directly modify the phenotype without changing the underlying DNA sequence.
Changes in gene expression through DNA methylation may have profound phenotypic repercussions. For example, studies with plants in which the establishment and the maintenance of methylation marks were disrupted by mutagenesis or chemical treatment yielded phenotypically aberrant individuals, thus giving evidence for a correlation between DNA methylation and the phenotype (Fieldes & Amyot 1999; Kalisz & Purugganan 2004 and references therein). The list of alternative phenotypes resulting from alternative DNA methylation states of a same gene is continuously growing (e.g. Jablonka & Raz 2009). Alternative methylation states of a same gene, known as epialleles, have been associated with variations in individual behaviour, physiology, and morphology as well as development (Weaver et al. 2004; Anway et al. 2005; Jeon 2008). How DNA methylation can cause phenotypic variation through the modification of gene or gene pathway expression is exemplified by the Colourless non-ripening (Cnr) gene, a component of the regulatory network controlling fruit ripening in tomato. The Cnr phenotype differs from the wild-type phenotype by a colourless and mealy pericarp. Comparison of the cytosine methylation patterns of the wild-type and Cnr phenotypes revealed that cytosines at the promoter of the Cnr gene (SQUAMOSA promoter binding protein-like genes) are extensively methylated in all individuals carrying the Cnr phenotype whereas they are largely unmethylated in wild-type fruits (Manning et al. 2006).
How do variations in DNA methylation appear?
There are key events at which such variations in DNA methylation patterns may occur. First, the DNMT1 enzymes responsible for copying the methylation marks during DNA replication have an error rate of 5% of CpG per cell division (Riggs et al. 1998; Bird 2002) compared to 10−9 per nucleotide per cell division for DNA polymerase. Errors in the replication of the initial epigenetic state of a gene lead to epigenetic variations among cells of the same tissue, a process that can lead to phenotypic variegation. Such spontaneous and random epigenetic errors may provide a large spectrum of alternative methylation states for the same genetic sequence. For instance, a dominant mutation at the Agouti gene (Avy allele) in mice results in an extensive variation of coat colouration (Morgan et al. 1999). The Avy allele displays a variable degree of expression that is linked to the level of methylation of a transposable element inserted upstream of the Agouti gene: if the transposable element is hypomethylated across cells, Avy is ectopically expressed and the coat colour is more yellow; if the transposable element is hypermethylated, Avy is expressed normally and the coat colour is normal. Because the methylation level of the transposable element displays extensive variation, the expression of the Agouti gene varies from cell to cell, which leads to a large variation in the coat colouration of an individual (e.g. mottling; Morgan et al. 1999). The extent of the variegated yellow and normal coat is linked to the random methylation of the Avy allele among cell lineages.
Secondly, a nonmethylated sequence can be de novo methylated (or vice versa). De novo methylation is induced by intrinsic developmental signals. However, it may also appear randomly (as a spontaneous variation) or be induced by environmental signals throughout the life of an individual. For example, in an isogenic strain of mice containing the Avy allele, extensive phenotypic variation can be observed among individuals, indicating inter-individual variations in random de novo methylation at the Agouti gene (Morgan et al. 1999; Blewitt et al. 2006). Another spectacular example of alternative de novo methylation leading to the production of distinct phenotypes can be observed in social insects. In the honeybee (Apis mellifera), phenotype is determined environmentally via the feeding of royal jelly to larvae meant to be fertile queens but not to larvae meant to be sterile workers. Kucharski & collaborators (2008) induced the inactivation in bee larvae of the enzyme responsible for de novo methylation and thereby demonstrated that the normal developmental pathway was to provide sterile individuals, but that interruption of the spread of methylation at a precise moment during the development led to the production of fertile individuals.
Interestingly, the variations observed in agouti mice can also be influenced by the environment. Maternal nutrient supplementation with a diet rich in methyl donors during gestation will globally increase the levels of DNA methylation at the transposable element associated with the Agouti gene and increase the proportion of the progeny with a normal phenotype (Wolff et al. 1998). Inversely, neonatal exposure to bisphenol A decreases methylation, therefore shifting the coat colour toward the mutant phenotype (Dolinoy et al. 2007). However, these effects of environment are not similar to those observed in social insects in that not all individuals exposed to these conditions will exhibit a given alternative phenotype. In addition, this epigenetically induced phenotypic variation has profound consequences on the individual’s fitness because the effects of the Agouti gene are pleiotropic, and mice with a hypomethylated transposable element will be obese, have a non-insulin-dependent diabetic-like condition, and have a propensity to develop a variety of tumours.
These previous examples highlight how the environment and DNA methylation may trigger different gene expression patterns to produce different phenotypes from genetically identical individuals. In these examples, the epigenetic state is important in determining the phenotype; the genotype by itself cannot explain the phenotype.
Heritability of methylation
One of the properties of DNA methylation marks is their transmission through mitosis: these marks are conserved during DNA replication by the action of DNMT1. This situation refers to mitotic epigenetic inheritance and concerns transmission within an individual’s lifetime (Crews 2008). However, for epigenetic variation to affect inheritance, meiotic transmission is also required. Some examples of meiotically transmitted methylation marks have indeed been reported. The transgenerational effects of DNA methylation require either a direct or indirect alteration of methylation in the germ line. This is termed meiotic epigenetic inheritance (Crews 2008). Therefore, in addition to perpetuating a change in gene expression throughout an individual’s life, DNA methylation could also transmit the effect of the environment on gene expression to further generations, even in the absence of initial stimulus (Crews et al. 2007; Jablonka & Raz 2009 and references therein). An interesting characteristic of DNA methylation is that inheritance of the variants appears to be highly variable among affected genes as well as among taxa (Rakyan et al. 2002). Depending on the genotypic context (see Richards 2006 for discussion) and the taxon, some methylation marks will be transmitted across generations whereas others will be limited to the lifetime of an individual.
DNA methylation has most extensively been studied in mammals, where methylation patterns are erased (to some extent) and reset twice during development. There is a first global genome demethylation during gametogenesis (but see Flanagan et al. 2006) and a second one during the period following fertilization. Erasure of methylation patterns has also been shown to occur during zebrafish development (Mackay et al. 2007), suggesting a common pattern in vertebrates. Therefore it seems that the DNA methylation marks that these organisms acquire during their life will not be transmitted to their progeny. However, the extent of erasure of epigenetic marks was found to vary among multicellular organisms, and even in mammals this erasure is not absolute (Richards 2006; Hitchins et al. 2007). Indeed, in organisms where the gametes appear later in development and where there is a less extensive erasure of epigenetic marks, such as plants, a higher propensity for methylation mark transmission is expected (Richards 2006). In support of this, most epialleles have been detected in plants (reviewed in Kalisz & Purugganan 2004).
Though it seems that DNA methylation by itself is a process whose effects are in many cases limited to the lifetime of an individual, DNA methylation can interact at several levels with other mechanisms that may indirectly promote its transmission (Blewitt et al. 2006). Indeed, it seems that DNA methylation influences and could be influenced by small regulatory RNAs. For example, previous observations that the environment can affect microRNA expression could be accounted for by the fact that environmentally induced DNA methylation regulates the expression of microRNA genes. Therefore, the effects of environmentally induced methylation could be indirectly propagated across generations via RNA molecules that are transmitted in the cytoplasm of gametes. Evidence for such indirect effects of DNA methylation has not yet been identified, but inheritance of a phenotypic variant via RNA can be observed in paramutation (Rassoulzadegan et al. 2006; Chandler 2007). Paramutations are defined as the modification of the effective expression of an allele (paramutated allele) by another homologous allele (paramutator allele). An example of this is found in mice where an engineered allele of the kit locus (paramutator allele) leads to the production of small interfering RNAs (siRNA) that degrade messenger RNA produced by the wild-type allele (paramutated allele) (Rassoulzadegan et al. 2006). The wild-type allele’s DNA sequence remains unchanged, but there is loss of effective expression of the gene. This silencing of the wild-type allele leads to a white-tipped tail and white feet phenotype. Interestingly, through transmission of these siRNAs in the cytoplasm of the gametes, progeny that inherited the wild-type allele can also display the alternative phenotype. These siRNA will then interfere with the expression of the wild-type alleles in the next generation. However, in absence of the paramutator allele, the phenotype is diluted each generation.
How to study DNA methylation
While the field of cancer epigenetics demonstrated more than a decade ago that changes in DNA methylation and gene regulation occur in cancer, there is still a dearth of studies addressing the importance and the frequency as well as the heritability of epigenetic variation in natural populations (Kalisz & Purugganan 2004; Richards 2008). Most current knowledge on DNA methylation comes from the comparison of epigenetic profiles of individuals of a same species with highly divergent phenotypes. The examples presented in the previous paragraphs are isolated and spectacular cases. There is as yet no sense of how widespread these types of phenomena are or of the importance of the challenge they present the current evolutionary theory. The implications of a widespread heritable environmentally induced epigenetic variation are potentially quite important. This section is meant as an overview of the methods available for quantifying DNA methylation polymorphism and for investigating the effects of methylation on phenotypic variation.
The amounts of DNA methylation in a genome have traditionally been analysed by high-performance liquid chromatography (HPLC). Although this technique has successfully been used to assess global changes of methylation in different experimental contexts (e.g., Cai & Chinnapa 1999), it does not allow the detection of the methylation state at the single gene level.
The tools for investigating variations in DNA methylation at the gene level are available and can easily be incorporated into any laboratory studying DNA polymorphism (reviewed in Liu & Maekawa 2003; Suzuki & Bird 2008). Methylation is a chemical mark added to the DNA, and there is no complementary nucleotide specific to methylated cytosines. It is not possible to detect the presence of methylation by directly using classic PCR-based analyses or sequencing because methylated and non-methylated cytosines are indistinguishable. However, methylated cytosines can be labelled prior to PCR amplification. Two different approaches can be envisaged: methods using endonucleases with different sensitivities to methylation or methods where non-methylated cytosines are chemically altered.
The presence of a methyl group on their restriction site can affect the capacity of certain bacterial endonucleases to recognize this site. Methylation sites can be identified by comparing the restriction fragment patterns generated by enzymes that have the same restriction site but different sensitivities to the methylation of this site. The isoschizomeric enzymes HpaII and MspI, for example, both recognize the CCGG sequence, but HpaII is unable to cut the DNA when the internal cytosine is methylated. Surveys using Methylation Sensitive Amplified Polymorphism (MSAP) (Xiong et al. 1999), a variant of the AFLP (Vos et al. 1995), can then be performed without further treatment. A limitation of this technique is that the endonuclease only detects differences in methylation that occur at its restriction site.
The gold standard for the detection of methylation polymorphisms remains sodium bisulfite treatment of DNA prior to PCR analyses. This chemical treatment allows the conversion of unmethylated cytosines to uracil while methylated cytosines remain unchanged (Frommer et al. 1992). Sequencing of treated and untreated DNA allows the identification of all methylated cytosines in a given sequence. To screen for variations in DNA methylation at the scale of a population, bisulfite treatment can be used prior to SSCP (Maekawa et al. 1999), methylation-allele–specific PCR using primers ending on a CpG dinucleotide (methylation-sensitive PCR) or microarray analysis (Yamamoto 2004).
However, except in model species or well-known gene pathways (e.g. Lister et al. 2008; 2009; Meissner et al. 2008), the question is not so much how but where to look for methylation differences in the genome. Surveys for candidate sequences can be achieved with AFLP-based techniques by comparing across individuals restriction fragment patterns generated by enzymes with different sensitivities to methylation or by comparing DNA treated or untreated with sodium bisulfite.
Screening for variations in DNA methylation patterns is different from screening for variations in DNA sequence because methylation patterns are time and tissue specific. Indeed, even though the cells of multicellular organisms are genetically identical, they present structural and functional heterogeneity. A developmental program may lead to the production of more than 200 different cell phenotypes, most of which can be accounted for by variations in DNA methylation (Bird 2002; Meissner et al. 2008). Epigenetic regulation of gene expression is also thought to be a dynamic process, with the methylation status of a gene potentially changing in response to developmental and environmental cues and aging (Fraga et al. 2005). Therefore, not only are different cell types within a given organism likely to have very different DNA methylation patterns (different epigenomes or methylomes), but fluctuations in time can also be expected even within the same cell. For instance, the analysis of the methylation polymorphism of the human major histocompatibility complex (MHC) revealed that a significant proportion of these genes show variegation (tissue-specific methylation profiles) in addition to inter-individual epigenetic variation (Rakyan et al. 2004).
DNA methylation can be influenced or not by the environment, is more or less independent of the associated genetic background, and displays different levels of heritability. Once inter-individual variations in DNA methylation have been detected, it is of interest to characterize these elements to determine the evolutionary significance of these variations. DNA methylation is affected by the environment and the genotype as well as by their interaction (Richards 2006), and can therefore be considered as a phenotypic trait (Gorelick 2004, 2005). The quantitative genetics framework can thus be used to establish the relative importance of the environment, the genotype, and their interaction on this phenotypic trait. The variation of a given phenotype can be measured by controlling for genetic background and environment; for instance, by using an experimental design involving several replicates of a genotype (clones, full siblings) maintained in controlled environmental conditions. A common garden experiment can assess developmental flexibility while reciprocal translocation of environmentally induced phenotypes can be used to investigate plasticity (e.g. Cai & Chinnappa 1999). Following this kind of approach, ‘pure’ epigenetic marks are those whose presence correlates with the environmental conditions whereas marks that are independent from the environment are referred to as genetically obligate marks (Richards 2008).
A similar challenge is faced when trying to study the effect of DNA methylation on the phenotype since the phenotype is influenced by DNA variations, the environment, and DNA methylation as well as by the interactions of these elements (Gorelick 2004, 2005; Bossdorf et al. 2008; Richard 2008). Variations in DNA methylation can be correlated with phenotypic variation by using approaches similar to those discussed in the previous paragraph. In addition, in the case of heritable methylation marks, the environment experienced by previous generations must also be known since it can affect the offspring’s phenotype (Gorelick 2004). Alternative approaches can also be considered to investigate the role of DNA methylation on phenotype, such as those involving treatments that affect the establishment or the maintenance of methylation marks (Fieldes & Amyot 1999; Kucharski et al. 2008).
Methylation, ecology, and evolution
The relevance of phenotypic variation in the domains of ecology and evolution is now widely accepted. Environmentally induced phenotypes are coined either phenotypic plasticity or developmental flexibility. Phenotypic plasticity refers to the capacity of an individual to change its phenotype throughout its life in response to a change in environments (Callahan 1997). On the other hand, developmental flexibility is the production of different phenotypes from individuals harbouring a similar genotype, depending on the environmental conditions experienced during their development (Bradshaw 1965). The effects of developmental flexibility (the reaction norm) are most readily observed when comparing the phenotypes of genetically identical individuals reared in different environments (Thoday 1953). The ability to produce such phenotypic variation is an evolving property, and evolutionary changes can also be mediated by phenotypic variation (Debat & David 2001; Young & Babyaev 2007).
Because of its role in gene regulation and the creation of phenotypic variation as well as its versatility, DNA methylation is expected to contribute to the persistence and evolution of populations in multiple ways. Indeed, epigenetic variation creates phenotypic differences that have an effect on individual fitness and therefore can be acted upon by natural selection (e.g. Crews et al. 2007). In addition, DNA methylation may allow individuals to use different strategies in fluctuating environments depending on its degree of inheritance. DNA methylation, unlike genetic modifications, may occur rapidly in response to environmental changes and could therefore represent a potential way to cope with environmental stress on very short time scales, possibly even during the lifetime of an individual (Rando & Verstrepen 2007). Because it could then allow individuals to produce alternative phenotypes in response to environmental change, DNA methylation would be a relevant process even in the absence of inheritance.
Following changes in the environment, alternative methylation patterns may be established on certain sensitive alleles, possibly giving the individuals that possess them an alternative phenotype (Fig. 1a). The frequency of advantageous methylation-sensitive alleles will therefore increase in subsequent generations, thereby increasing the number of individuals apt to react to environmental fluctuations. In this scenario, there is no inheritance of the methylation marks: DNA methylation modulates gene expression in response to the environment, while the genetic background provides heritability of the genes required for flexibility. It is the flexibility of the phenotype that is selected for rather than the result of the flexibility itself (Fig. 1b). This type of scenario, known as the Baldwin effect (Simpson 1953, reviewed in Crispo 2007), could be advantageous in unstable or highly heterogeneous environments.
Production of phenotypic variants through DNA methylation may help in exploring alternative environments and consequently in providing a wider niche. For example, Pal & Miklos (1999) showed how environmentally induced phenotypes (heritable or not) may initiate the transition from one adaptive peak to another by allowing the exploration of the adaptive landscape without leaving the high fitness peak linked to the underlying genotype. DNA methylation could provide an additional process to induce a peak shift that, unlike genetic drift, does not require demographic reduction or small population size (shifting balance theory; Wright 1932). Regulation of gene expression through DNA methylation may also provide asexual organisms with developmental flexibility, possibly explaining observations where populations of clonal organisms were found to display as much phenotypic variability as closely related sexual species (e.g. Doeringsfeld et al. 2004; Gorelick et al. 2010).
In what circumstances is the inheritance of an environmentally induced phenotype advantageous? Predictability (variability across generations) and reliability (variability within generation) of the environmental conditions are components expected to influence selection on phenotypic variation (West-Eberhard 2003). While a fluctuating environment should favour a high level of plasticity (Baldwin effect, Fig. 1b), new environmental conditions that are stable both within and across generations should favour genetic assimilation (Young & Badyaev 2007). Genetic assimilation refers to a reduction of the variability around a new phenotype following environmental changes (Fig. 1c). For example, the heritability of an environmentally induced phenotype becomes favourable if the variability around the new phenotype is low and the phenotype is close to its optimum (Waddington 1961; Pal & Miklos 1999; Crispo 2007). However, this process requires long-term heritability of the marks responsible for the apparition of the new environmentally induced phenotype. The efficiency of transmission of such marks or cellular memory is essential to enhance the strength of the phenotypic selection (Pal 1998). Because of the delay between the induction and the selection of the new phenotype, the environment must be stable for a period at least as long as the organism’s generation time in order for it to be adaptive (Lachmann & Jablonka 1996). In completely random or highly fluctuating environments, the persistence of induced phenotypes for several generations even in the absence of the inducing environmental conditions that produced the phenotype seems to represent the optimal strategy when compared to strictly fixed (genetic) or completely inducible (plastic) strategies (Jablonka et al. 1995). An epigenetic inheritance system could ensure the transmission of the marks responsible for the environmentally induced phenotype from parents to their progeny. Such a mechanism may facilitate the transition between individual plasticity and long-term evolutionary innovations and allow adaptation at an intermediate time frame.
When inherited across generations, variation in DNA methylation becomes similar to genetic variation and serves as a template to natural selection and other evolutionary forces such as drift and migration. However, because of its reversibility and more labile nature, the persistence of DNA methylation is not expected to be as stable as DNA mutations over a long period of time. During the process of genetic assimilation, the environmentally induced phenotype becomes genetically assimilated, and the environmental signal as well as the epigenetic marks are no longer required to produce it (Waddington 1953). While the role of DNA methylation may only be transient, it remains crucial in initiating the exploration of the adaptive landscape by inducing phenotypic variation in response to the new environmental conditions (Fig. 1c).
DNA methylation and modern evolutionary synthesis
Cases of environmentally induced heritable variations in DNA methylation causing phenotypic differences with an impact on fitness have been reported (Crews et al 2007; Crews 2008). These findings challenge the existing theory of evolution (modern synthesis) that supports the view that the information that is transmitted changes only at random without any direction from the environment toward any phenotypic outcome. Indeed, it rather seems the environment could be responsible for both the apparition of variation and its selection (Richards 2006; Jablonka & Lamb 2008).
A major source of confusion regarding the effect of environment on phenotype is the belief that all environmentally-induced changes produce a phenotypic adaptation. The effect of larval diet on the methylation patterns and the phenotype of social insects (Kucharski et al. 2008) and how the nursing by mother rats reduces stress in their offspring (Weaver et al. 2004) are examples of how the environment, can act as an intrinsic cue to modify development, resulting in a predetermined phenotypic change (Fig. 2a).
However, environmental changes do not always result in a predictable phenotypic outcome (Fig. 2b). Individuals as well as genes are likely to display different thresholds or sensitivities to a given environmental signal (Sollars et al. 2003), and a given gene could be affected differently by environmentally induced methylation in different individuals, resulting in distinct phenotypes on which selection can act. The resulting phenotypes are thus not always favourable. For example, induction of aberrant DNA methylation by environmental toxicants during critical developmental periods has been known to lead to inappropriate gene expression and disease pathogenesis in later life (Dolinoy et al. 2007; Perera et al. 2009).
DNA methylation and other epigenetic processes appear not to be in disagreement with the theory of modern evolutionary synthesis. Such processes allow organisms to use the environment to modify development as a ‘signal of a programmed change’ (or modify their development in response to environment). But environmentally driven methylation changes are expected to occur as randomly as mutations, and according to the outcomes of these changes, the phenotype will be selected for or against. Even though environment appears to be responsible for both the creation of and selective pressures affecting variation, these two processes are independent.
Many of its characteristics make DNA methylation a versatile mechanism for modifying gene expression and phenotype. Methylation marks are enzymatically modifiable, can change rapidly, and are reversible, which is not the case with DNA mutations. They are also conservatively replicated through mitosis and, in some cases, through meiosis as well. It is unclear which characteristics of DNA methylation prevail in natural populations and how they impact evolutionary processes. However, because of the variety of genes, genotypes, and organisms involved, and because of the spatial and temporal heterogeneity of the environment, it is expected that a broad range of conditions exists where non-heritable methylation marks permit rapid adjustment to the environment or where, in certain circumstances, transgenerational marks lead to local adaptation and promote divergence until speciation (Jablonka & Lamb 1991, 1995; Pal & Miklos 1999; Pigliucci & Murren 2003; West-Eberhard 2003; Schlichting 2004; De Jong 2005; Bonduriansky & Day 2009). This also suggests that all organisms do not react in the same way to environmental changes due to evolutionary differences in DNA methylation among taxa or to different genetic backgrounds among individuals and populations.
DNA methylation and other epigenetic processes also suggest that a reconsideration of the nature of the heritable material is needed (e.g. Gorelick & Laubichler 2008; Bonduriansky & Day 2009), but they are not otherwise in disagreement with the modern evolutionary synthesis. Though it may seem to add to the complexity of carrying out biological research, taking into account the additional layer of information provided by DNA methylation has already provided insightful explanations to biological phenomena that could not be accounted for by variations in DNA sequence. The study of DNA methylation and other epigenetic processes may soon be an integral component in the study of any biological process.
We are grateful to Pierre Magnan, Laure Devine, Root Gorelick and anonymous reviewers for constructive comments on the manuscript. This work was supported by a research grant from NSERC to B. Angers.
B.A. is a professor at the Université de Montréal. His research programme focuses on the understanding of interactions between ecological and evolutionary processes in generating the distribution of biodiversity. E.C. is a PhD candidate at the University of Edinburgh interested in the influence of epigenetic processes in regulating cell division. This work was part of her MSc in B.A.’s team. R.M. is a PhD candidate under the supervision of B.A. and her thesis is related to the evolutionary implications of epigenetic variation for asexual organisms. The authors have a particular interest in the use of unisexual fish as model organisms to study epigenetic processes.