Variation in DNA sequence can cause variation in gene expression, which influences quantitative phenotypic variation in organisms and is an important factor in natural variation. Gene expression regulatory networks are comprised of cis-and trans-acting factors, and differences in gene expression are attributable to genetic variation. In eukaryotes, the genome is compacted into chromatin, and the chromatin structure plays an important role in gene expression: Gene expression can be controlled by changes in the structure of chromatin without changing the DNA sequence, and this phenomenon is termed “epigenetic” control. Recently, there have been many reports indicating that epigenetic change can cause phenotypic variation, and thus epigenetic change can be considered as an important factor in understanding phenotypic change. DNA methylation and histone modifications are well known epigenetic modifications. DNA methylation refers to an addition of a methyl group at the fifth carbon position of a cytosine ring, and in plants it is observed not only in the symmetric CG context but also in sequence contexts of CHG and CHH (where H is A, C, or T) [1–3]. DNA methylation is enriched in heterochromatic regions, such as in centromeric and pericentromeric regions, predominantly consisting of transposons [3–7]. Most transposons are immobile to protect genome integrity and are silenced via DNA methylation [3,8–12]. DNA methylation is also observed in euchromatic regions such as gene-coding regions (gene body methylation), and it is widely seen in eukaryotes [3,13,14].

Nucleosomes are formed by a histone octamer containing two of each of the core histones H2A, H2B, H3, and H4, and 147bp of DNA is wrapped around this core. The N-terminal regions of histone proteins are subject to various chemical modifications such as methylation or acetylation, and these histone modifications are associated with gene transcription [15,16]. In plants, DNA methylation, histone deacetylation and histone methylation in H3K9 (9th lysine of H3) and H3K27 are associated with gene repression, and DNA demethylation, histone acetylation and histone methylation in H3K4 and H3K36 are associated with gene activation [15,17]. Histone lysine residues are able to be mono-, di-, or tri-methylated and each methylation state is associated with different functions [15,17]. Epigenetic modifications play important roles in various aspects of the plant life cycle such as genome integrity, transgene silencing, nucleosome arrangement, nucleolar dominance, paramutation, flowering, and parent of origin-specific gene expression (imprinting) [15,18–21].

Genome-wide profiles of epigenetic information (the epigenome) are available in plants using new technologies such as tiling arrays or high-throughput next generation sequencing [15]. High-resolution maps of epigenetic features have been obtained from bisulfite sequencing (bisulfite converted DNA is directly sequenced) or a combination of chromatin immunoprecipitation (ChIP) technology and genomic tiling arrays (ChIP on chip) or ChIP and high-throughput sequencing (ChIP-seq) [15]. Using these technologies, effects of epigenetic modifications in mutants and variations of DNA methylation status between accessions in Arabidopsis thaliana, rice and maize have been shown at the whole genome level [22–26].

In general, heritable variation is a consequence of differences of nucleotide sequence. However, more studies are reporting heritable variation caused by epigenetic variation [27,28]. These epigenetic variations were categorized “obligatory”, “facilitated”, or “pure epialleles” by Richards [27]. “Obligatory” epigenetic variation is entirely dependent on DNA sequence changes, “facilitated” epigenetic variation is caused by stochastic variation in epigenetic status associated with a DNA sequence change, while “pure” epigenetic variation is generated stochastically and is completely independent of DNA sequence [27]. The “pure” epigenetic variations are subcategorized as stably or metastably inherited [29]. Sometimes these heritable epigenetic changes with or without genetic changes accompany phenotypic change, and there is evidence that spontaneous epigenetic changes generate new plant phenotypes in nature or in cultivars [30–32]. In addition, abnormalities in DNA hypo-methylated mutants have been characterized and some of them are due to the change of DNA methylation status without any difference in nucleotide sequence [33]. Increased knowledge about heritable epigenetic change associated with phenotypic variation suggests that heritable epigenetic changes may become a resource in plant breeding or play a role in plant adaptation [34].

In this review, we describe instances of naturally occurring epigenetic variants and how these can affect plant phenotype. We speculate on the possible causes and analyze the molecular basis of many of these variants and where possible, we elaborate on the resulting phenotypes. Most of our examples are from A. thaliana, as its genomics resources are most advanced. We conclude that epigenetic variation is widespread and contributes significantly to the generation of natural variation probably in most species, not just A. thaliana.

Epigenetic variation can arise in a number of ways. One way is through mutations in the genes responsible for maintaining epigenetic modifications such as DNA methylation. In A. thaliana, DNA methylation in the CG context is maintained by MET1 (METHYLTRANSFERASE 1), while non-CG contexts are maintained by DRM (DOMAINS REARRANGED METHYLTRANSFERASE) and CMT3 (CHROMOMETHYLASE 3) [3,35]. In addition to DNA methyltransferase, a chromatin remodeling factor| DDM1 (DECREASE IN DNA METHYLATION 1), histone methyltransferase| SUVH4/KYP (SU(VAR)3-9 HOMOLOG 4/KRYPTONITE) (hereafter KYP) and SUVH5/6, or SRA-domain methylcytosine-binding protein VIM1/2/3 (VARIANT IN METHYLATION 1/2/3) are also involved in the maintenance of DNA methylation [35]. The process of de novo DNA methylation is triggered by 24-nt siRNAs produced by the RNAi (RNA interference) pathway, termed RdDM (RNA-directed DNA methylation) [36]. Two plant specific RNA polymerases, Pol IV and Pol V, RDR2 (RNA-DEPENDENT RNA POLYMERASE 2), DCL3 (DICER-LIKE 3), and AGO4 (ARGONAUTE 4) proteins function in this RNAi pathway [36,37].

Plant developmental abnormalities have been detected in mutants with disturbed epigenetic modifications, some of which are heritable. An allele of a heritable variant, which is caused by a change in an epigenetic modification without a change in the DNA sequence, is termed an “epiallele”. In A. thaliana, ddm1 hypo-methylated mutants showed only slight morphological changes in the early generations, but morphological abnormalities increased after repeated self-pollination over several generations [38,39]. Some developmental abnormalities are heritable and are not linked to the DDM1 gene [40]. Some of these mutants are a consequence of the mobilization of transposons due to removal of DNA methylation from the transposon. Genes responsible for these abnormal phenotypes have been able to be identified by map-based cloning, because these phenotypes are heritable and indistinguishable from genetic mutations [33]. One of them is clm (clam), which showed a lack of elongation in shoots and petioles. This clm mutant is caused by an insertion of a CACTA1 transposon in the DWF4 gene, which encodes 22-α-hydroxylase in the brassinosteroid biosynthetic pathway. In wild type, CACTA1 is silent, but it can transpose in ddm1 [8]. This transposition has also been observed in met1 cmt3 double mutants, indicating that DNA methylation is important for the silencing of CACTA1 [41]. Another mutant, wvs (wavy-sepal), is also caused by an insertion of a transposon into the FASCIATA1 gene. This transposon is a member of LTR (Long-terminal repeat) retrotransposon class, AtGP3-1. AtGP3-1 is silent in wild type, but it can transpose in ddm1 [11]. clm and wvs are genetic mutants caused by epigenetic changes. Another transposon is mobilized in the hypo-methylated mutants, ddm1 or met1, and has the potential to generate a new genic mutant [11,42].

Other mutants can be caused directly by changes in DNA methylation affecting transcription of the gene. The late flowering mutant fwa (FLOWERING WAGENINGEN) caused by ectopic expression of the FWA gene, encodes a homeodomain-containing transcription factor. In wild type, the promoter region of FWA is DNA methylated and FWA is not expressed in vegetative tissues, this DNA methylation is removed in the ddm1 mutant and FWA is expressed in vegetative tissues (Figure 1) and causes late flowering [43]. This late flowering phenotype is also observed in the met1 mutant [44,45], but not in the drm1 drm2 cmt3 triple mutant [46], suggesting that silencing of FWA is mainly dependent on CG methylation. The DNA hypo-methylation in the promoter region of FWA and the late flowering phenotype are stable in the normal DDM1 background, indicating that fwa is a gain of function epigenetic mutant.

Another mutant phenotype seen in the ddm1 background, change of plant structure (short and compact inflorescence with reduced plant height), is bns (BONSAI), which is unstably inherited in the presence of the DDM1 gene. The BNS gene encodes a protein with similarity to the mammalian cell cycle regulator Swm1/Apc13. In wild type, the BNS gene is normally expressed and not methylated (Figure 1). However, in a self-pollinated ddm1 mutant, the BNS gene is methylated and stochastically silenced (Figure 1), indicating that bns is a loss of function epigenetic mutant. The BNS gene is flanked by a LINE (Long interspersed repeated element) sequence in a tail-to-tail orientation, and in the ddm1 mutant DNA methylation in the BNS coding region spreads from the LINE (Figure 1). In ddm1, the DNA methylation level in BNS gradually increases over generations and a phenotype develops. There are two types (with or without LINE sequence) of variation in the BNS gene among 96 accessions of A. thaliana, 70 of these 96 accessions have LINE sequences at the BNS locus. Cvi that lacks the LINE sequence does not show DNA methylation at the BNS locus even in a ddm1 background, indicating that the LINE is essential for the spread of DNA methylation in the ddm1 mutant background [47]. This shows there is ectopic local DNA hyper-methylation of a specific locus in the global DNA hypo-methylation mutant, ddm1. Although small RNAs corresponding to the BNS locus accumulate in the ddm1 mutant, ectopic induction of de novo DNA methylation at the BNS locus in the ddm1 background was independent of the RdDM pathway because mutations in RdDM components such as RDR2, DCL3, AGO4, PolIV, and PolV did not affect ddm1-induced DNA methylation at the BNS locus [48]. However, KYP and CMT3 were essential for this ectopic DNA hyper-methylation at the BNS locus. In addition, meDIP (methylated DNA immnoprecipitation)-chip analysis revealed that BNS-like loci were widespread within the A. thaliana genome, and that they are DNA hyper-methylated in the ddm1 mutant background in a CMT3-KYP-dependent manner. Although CMT3 is known for the maintenance of DNA methylation in the CHG context, CMT3-KYP dependent alternative de novo DNA methylation was found in all three contexts [48].

The met1 mutants in A. thaliana also show developmental abnormalities such as reduced apical dominance, alterations in flowering time, floral abnormalities, curled leaves, embryogenesis, and formation of viable seeds [44,45,49,50], some of which are inherited even when the wild type allele is present. Genome-wide inheritance of hypo-methylation status even in the presence of the MET1 wild type locus has been observed in an F₈ population derived from hybrids between met1 and wild type [51]. The floral abnormalities in the met1 mutant or MET1 antisense lines are due to DNA hyper-methylation and silencing of SUP (SUPERMAN) and/or AG (AGAMOUS) [52,53]. DNA hyper-methylation occurs at CT-rich repeats in the promoter of SUP or in the promoter and second intron of AG. This shows global DNA hypo-methylation by the met1 mutation, which causes local DNA hyper-methylation: Stochastic non-CG methylation has been observed in the met1 mutant [54].

The drm1, drm2 or cmt3 single mutants did not show any apparent phenotypes, but drm1 drm2 cmt3 triple mutants showed pleiotropic phenotypes including developmental retardation, reduced plant size, and partial sterility [46,55]. Unlike ddm1 or met1, the drm1 drm2 cmt3 phenotype is completely recessive: Pleiotropic phenotypes are not inherited independently of the drm and cmt3 mutations [55]. The misexpression of SDC (Suppressor of drm1 drm2 cmt3) was observed in the drm1 drm2 cmt3 triple mutant, and it is sufficient for pleiotropic phenotypes in drm1 drm2 cmt3 triple mutants. The promoter region harboring tandem repeat regions is densely methylated in all contexts in wild type, but DNA methylation in the promoter region is eliminated in drm1 drm2 cmt3 triple mutant. F₁ progeny between drm1 drm2 cmt3 triple mutant and wild type show reversion of developmental phenotypes and the promoter region of SDC becomes methylated and SDC expression is lost. siRNAs corresponding to tandem repeat regions are expressed in wild type leading to DNA methylation of the SDC promoter region dependent on the RdDM pathway. Taken together, the pleiotropic phenotypes in the drm1 drm2 cmt3 triple mutant are due to SDC misexpression caused by the elimination of DNA methylation in its promoter region [56].

It is well known that DNA sequence polymorphisms at a single locus or multiple loci cause phenotypic variation, and that they are important sources of variation in plants during evolution. In addition to DNA sequence polymorphisms, epigenetic variation has the potential to contribute to the natural variation of plant traits. Epigenome analysis aids in explaining how natural epigenetic variation causes phenotypic differences in plants. The DNA methylation status at the whole genome level has been examined in several species [1,2,4–7,13,14]. Accessions of a species may have been sourced from different environments and different epigenetic modifications selected over time to ensure optimum adaptation to specific environments. Variation of DNA methylation between accessions in A. thaliana occurs with gene-body methylation being more variable than DNA methylation of transposable elements among 96 natural accessions [22,25,26,57,58]. Differentially methylated regions have been detected in a comparison of whole genome DNA methylation statuses between two lines of rice or maize [23,24]. In the case of maize, differentially DNA methylated regions were generally observed in intergenic regions. Stable inheritance of DNA methylation was exhibited using near-isogenic lines of maize, though trans-acting control of DNA methylation was detected at a few regions [24]. These differences in DNA methylation could have consequences for differential expressions of genes.

A comparison of DNA methylation statuses between parental lines and their progenies generated from single seed descent over 30 generations showed that larger regions of DNA methylation were stable and changes of DNA methylation accumulated through generations [59,60]. The rate of spontaneous changes of DNA methylation is higher than the rate of spontaneous genetic mutations [59–61], suggesting that sequence-independent epialleles play important roles in phenotypic diversity (Figure 2) [59,60]. To identify loci causing phenotypic variation, populations of epigenetic recombinant inbred lines (epi-RILs) between parents, which differed only in epigenetic marks, have been established in A. thaliana, and plant complex traits caused by epigenetic variation are observed [29,51,62]. In A. thaliana, two sets of epi-RILs were generated from ddm1 or met1 mutants that were crossed with wild type [29,48]. Stable inheritance of complex traits such as flowering time and plant height has been observed in these epi-RIL populations, providing important evidence that epigenetic variation can contribute to complex traits [29,51]. Heritable variation that was segregating in epi-RILs is similar to the phenotypic diversity observed in natural populations, suggesting that epigenetic variation in complex traits may drive some portion of natural variation (Figure 2) [63].

FWA is responsible for a late flowering phenotype in A. thaliana that is caused by the inhibition of FT function by protein-protein interaction between ectopically expressed FWA and FT [97,98]. FWA is expressed only in the central cell and endosperm in A. thaliana and reciprocal crosses between Col and Ler have shown that only the maternal allele of FWA is expressed in the endosperm, indicating that FWA is an imprinted gene in A. thaliana [99]. The maternal allele is demethylated in the central cell by the demethylase, DME (DEMETER), which also acts on other imprinted genes in A. thaliana [99,100]. In vegetative tissues, DNA methylation of FWA occurs in the promoter region, which harbors two pairs of tandem repeats and a SINE (short interspersed nuclear element). This DNA methylation is reduced in the endosperm of A. thaliana, suggesting that methylation of this region participates in silencing of FWA [43,99]. Small RNAs are produced from the promoter region of FWA, suggesting that DNA methylation in this region is mediated by the RdDM pathway [101,102]. Indeed, DNA methylation of the promoter region of a “transgene” of FWA is dependent on the function of DRM2, RDR2, DCL3, and AGO4 [103]. Transformation of a double stranded RNA construct, which can cause de novo DNA methylation directed to a target region, into the fwa mutant has shown that DNA methylation in the region harboring the two pairs of tandem repeats and SINE region is sufficient for the silencing of FWA expression in vegetative tissues of A. thaliana [104].

Using species related to A. thaliana, the structures that cause DNA methylation, imprinting, and vegetative silencing of FWA have been examined [105]. FWA genes are conserved in the genus in Arabidopsis, as there is high sequence homology not only in exon regions but also in the intron and promoter regions among species (Figure 4a). The SINE sequence is found in all species examined, A. arenosa, A. halleri, A. lyrata, A. suecica (allotetraploid between A. thaliana and A. arenosa) and A. kamchatica (allotetraploid between A. halleri and A. lyrata), suggesting that the SINE insertion is an ancient event (Figure 4a). In contrast, the structure of the tandem repeats is different among species: A. halleri and A. halleri allele of A. kamchatica have no tandem repeat in the SINE region, while A. arenosa, A. lyrata, A. arenosa and A. thaliana alleles of A. suecica, and the A. lyrata allele of A. kamchatica have tandem repeats like A. thaliana (Figure 4a). The sizes of the repeated and duplicated regions are different between species, suggesting that duplications occurred after speciation (Figure 4a) [105]. The ancient species, Arabis glabra, has a SINE region but no tandem repeat, supporting the hypothesis that the tandem repeat was not in the original structure (Figure 4). The FWA genes of A. lyrata and A. halleri show imprinted expression in immature seeds. DNA methylation of FWA in vegetative tissues in the SINE region is observed in all species. In A. halleri subsp. gemmifera, which lacks the tandem repeat structure, FWA shows imprinted expression, silencing in vegetative tissues, and DNA methylation in the SINE region, suggesting that the SINE sequence per se is important for epigenetic regulation of the FWA gene and FWA may have evolved silencing mechanism for transposable elements [101,105]. Transposable elements are extensively demethylated in endosperm, and the flanking regions of imprinted genes involving repetitive sequences are also demethylated, suggesting that imprinted genes evolved from targeted DNA methylation of transposable elements in A. thaliana [106,107].

Vegetative silencing of FWA varies not only between species but also within species [105,108]. In A. thaliana, 93 out of 96 accessions have two pairs of tandem repeats (termed Type-A), and three have large tandem repeats but not short tandem repeats (termed Type-B). All 96 natural accessions have DNA methylation in the SINE region [22]. FWA is not expressed in all 21 accessions of Type-A that we selected randomly from 93 accessions, but two of three accessions of type-B, Fab-4, Var2-1, and Var2-6 showed a low level of FWA expression. However the DNA methylation level in the SINE region is almost the same among Type-B accessions (Figure 5). Though it is still unknown what the difference in the silencing stability among the three Type-B accessions is, two pairs of tandem repeats stabilize FWA silencing. Indeed, both large and small tandem repeats are involved in silencing FWA [105,108]. In A. lyrata, FWA is expressed in two strains of subsp. lyrata that has three tandem repeats and FWA is not expressed in subsp. petraea that has four tandem repeats. The FWA expression level tended to be inversely correlated with the DNA methylation level of the SINE in A. lyrata [105]. Another repeat in the subsp. petraea enlarged the DNA methylated region, suggesting that more tandem duplications might lead to greater stabilization of FWA silencing, similar to the indications from A. thaliana. In A. halleri, FWA was expressed in vegetative tissues of subsp. halleri, tatlica, and ovirensis and in several strains of subsp. gemmifera, while FWA was silenced in the majority of strains of subsp. gemmifera. One strain, IK, showed variation of FWA expression level among ten individual plants in spite of a perfect match of the promoter sequences, and there is a negative correlation between FWA expression level and DNA methylation level, especially with the non-CG methylation level in the region just upstream of the TSS (transcription start site) [105,108]. From these results, silencing of FWA is stable in Type-A of A. thaliana and A. lyrata subsp. petraea, but unstable in other species. This difference might be due to the number of tandem repeats, which can expand the DNA methylated region.

The results from inter-specific hybridization support this suggestion. In the inter-specific hybrid between A. thaliana (Col or Ler) and A. lyrata subsp. lyrata (pn3 or MN47), only the A. lyrata allele of FWA was expressed in vegetative tissues and this expression level was higher than the expression level of the parent A. lyrata subsp. lyrata (Figure 6). The DNA methylation level in the SINE region of the A. lyrata allele in the inter-specific hybrid was reduced in vegetative tissues [105]. In the inter-specific hybrid between A. thaliana (Col or Ler) and A. halleri subsp. gemmifera, only the A. halleri allele of FWA was expressed in vegetative tissues in spite of no FWA expression in the parents (Figure 6). The DNA methylation level in the SINE region of the A. halleri allele of the inter-specific hybrids was also reduced in vegetative tissues, especially in the non-CG methylation of the region upstream of TSS (Figure 7). Though up-regulation of the A. lyrata or A. halleri allele in inter-specific hybrids was detected by RT-PCR, this up-regulation could not be detected by microarray analysis using ATH1 [109]. The A. thaliana FWA allele in two inter-specific hybrids was silenced and non-CG DNA methylation was slightly reduced in vegetative tissues (Figures 6 and 7). In the inter-specific hybrid between A. thaliana and A. lyrata subsp. petraea, there was no FWA expression in vegetative tissues as in their parents (Figure 6). The DNA methylation level in the A. lyrata allele of the inter-specific hybrid did not change (Figure 8). These results suggest that silencing of FWA might be affected by inter-specific hybridization, if the silencing level of the parent is unstable. These results also support the possibility of enhancement of silencing by tandem duplications.

From these results, two possibilities arise. (1) Tandem duplications stabilize FWA silencing, especially in Type-A of A. thaliana; (2) Non-CG DNA methylation in the region upstream of TSS is important for FWA silencing in the species related to A. thaliana. To confirm these possibilities, critical methylated residues controlling FWA silencing were examined using a double-stranded RNA to direct DNA methylation to target regions. In A. thaliana, DNA methylation in both short (region upstream of the TSS) and large tandem repeats (region downstream of the TSS) played a role in FWA silencing. In contrast, DNA methylation in the region upstream of the TSS played a role in FWA silencing in A. lyrata and A. halleri, but DNA methylation in the region downstream of the TSS was not sufficient for FWA silencing in A. lyrata. In A. thaliana, expression of small RNAs corresponding to the SINE region with two pairs of tandem repeats was confirmed, but few small RNAs were detected in A. lyrata, suggesting that DNA methylation in the SINE region is independent of the RdDM pathway in A. lyrata, unlike A. thaliana [101,102,104,108]. From these results, the critical methylated region for FWA silencing is different between A. thaliana and A. lyrata/A. halleri, and tandem duplications in A. thaliana enlarged the critical DNA methylated regions, which can stabilize the FWA silencing.

There is the question why the silencing mechanism is different between A. thaliana and species related to A. thaliana. This could be due to the ability of FWA to inhibit flowering in A. thaliana but not in A. lyrata. Over-expression of FWA from A. lyrata does not cause late flowering in an A. thaliana background, suggesting that A. lyrata FWA cannot inhibit FT function. Over-expression of both A. thaliana FWA and A. thaliana FT did not show any obvious developmental abnormality in flowers, but over-expression of both A. thaliana FWA and A. lyrata FT reveal occasional floral defects, which are due to misexpression of AP1 (APETALA1) and LFY (LEAFY) [110]. A. thaliana shows amino acid changes in the C-terminal region of FWA close to the region important for binding of FT [98,108], suggesting that A. thaliana FWA might have gained the ability to interact with the FT protein after speciation. Thus ectopic FWA expression caused by DNA demethylation might be disadvantageous for both summer and winter annual natural accessions of A. thaliana, so FWA is stably silenced in A. thaliana. Tajima’s D test showed negative selection against mutations in the C/G site, suggesting that silencing of FWA mediated by DNA methylation plays an important role in adaptation of A. thaliana [105]. In A. thaliana, a more stable FWA silencing mechanism (spreading of critical methylated regions by tandem duplications) has been selected during the process of evolution. This can prevent late flowering caused by a newly generated FWA function involving spontaneous substitutions, which enable FWA to interact with FT and inhibit FT function.

Increasing numbers of epialleles are being reported in various species, and it is clear that epi-mutations can affect plant phenotypes. Some naturally occurring epialleles affect genes involved in plant fitness; some epialleles are stably or metastably inherited [34]. There are multiple causes of epi-mutations such as change of epigenetic status without genetic changes or via genetic changes such as transposon insertion or tandem repeat formations [34]. As plants are sessile organisms, they rely on adaptation mechanisms to withstand environmental stress. Phenotypic modifications by DNA sequence changes cannot respond quickly to environmental stresses. Metastable inheritance may be more useful in adaptation than genetic mutations because metastable epigenetic changes are more flexible and may contribute to phenotypic plasticity under environmental stress conditions [111,112]. Natural variation of epigenetic status has been found among accessions in several plant species, and this variation might be a consequence of the different growing condition in nature [15,20,112]. The higher epi-mutation rate has the potential to contribute to natural variation [59,60], and results using epi-RILs support the idea that complex epigenetic variations are one of the factors of natural variation [29,51]. More research focusing on naturally occurring epigenetic changes will increase our understanding of how epigenetic variation has contributed to natural variation.