Epiallelic Variation in Arabidopsis thaliana

Arabidopsis thaliana is found over a large swath of Eurasia and parts of North Africa exposing individual populations to an array of biotic and abiotic challenges (Koornneef et al. 2004). The significant phenotypic variation that exists between regional populations of this species is reflective of the need to cope with a diverse spectrum of environmental variables (Koornneef et al. 2004). Wild Arabidopsis isolates from specific geographic locations, known as accessions, have proven to be a valuable resource for studying phenotypic variation shedding light on how a plant species diverges in response to varying degrees of isolation and adaptation to local environments(Fournier-Level et al. 2011; Hancock et al. 2011).

While largely similar in regards to basic anatomy, Arabidopsis accessions show phenotypic differences in important developmental traits such as leaf shape, growth rate, biomass, and flowering time (Weigel 2012). Perhaps even more striking is the enormous variation observed in response to specific biotic and abiotic challenges. Dramatic phenotypic variations are observed when accession are exposed to various challenges such as hormones, bioactive small molecules, metals, salts, light, heat, cold, moisture, and pathogens, all of which are environmental variations likely to be encountered by a widely dispersed population (Weigel 2012). Recent sequencing of Arabidopsis accessions genomes reveals substantial intraspecific sequence divergence in the form of single nucleotide polymorphisms (SNPs) and both small and large insertions and deletions (Cao et al. 2011; Gan et al. 2011; Ossowski et al. 2008; Schneeberger et al. 2011; Santuari et al. 2010). Many genes are affected in terms of coding sequence and significant transcriptional variation is observed between the accessions, providing an initial blueprint of the evolutionary forces shaping transcriptome variation, a primary wellspring of phenotypic divergence (Cao et al. 2011; Gan et al. 2011).

While differences in genotype were once viewed as the sole cause of natural phenotypic variation, in the past two decades it has been recognized that epigenetic variation, consisting of heritable modifications not directly encoded in the nucleotide sequence, may provide a distinct mechanism for inter- and intraspecific variation (Paszkowski and Grossniklaus 2011). Heritable epigenetic marks such as cytosine methylation, histone modifications, histone variants, and small RNA have all been shown to play a role in regulation of gene expression (Law and Jacobsen 2010; Roudier et al. 2011). Alleles variant in transcriptional regulation resulting from epigenetic differences are commonly referred to as epialleles. Though only a few examples of natural phenotypic variation caused by epialleles have been identified to date, there is evidence of substantial epigenetic variation between Arabidopsis accessions, suggesting the potential for a major contribution of epiallelic divergence on phenotype (Groszmann et al. 2011; Greaves et al. 2012b) (Shen et al. 2012).

Cytosine DNA methylation, the addition of a methyl group to the fifth position of cytosine in nuclear DNA, is one of the best characterized of these epigenetic modifications and has been shown to have profound effects on regulation of gene and transposon transcription (Slotkin and Martienssen 2007; Matzke et al. 2007). Cytosine DNA methylation is also the most described epigenomic variation in regards to natural and artificially-induced epialleles in plants (Richards 2011). Furthermore, with recent developments in DNA sequencing, single basepair resolution whole-genome methylome maps of Arabidopsis can be generated quickly and relatively inexpensively promising future enrichment of the annotation of DNA methylation in many Arabidopsis accessions (Lister et al. 2008; Cokus et al. 2008). As a result, in this review, we will largely focus on epialleles associated with variation in DNA methylation though other epigenetic marks will be discussed when relevant. We will discuss the generation of artificial epialleles that have proven to be powerful tools for studying the genetic pathways and mechanisms controlling DNA methylation which have proven to be valuable model systems to inform our understanding of natural occurring epialleles. The second half of the review will focus on examples of natural occurring epialleles and epigenomic variation with a focus on phenotypes, associated sequence features, transmission, heritability and the complex interplay that can exist between genotype and epigenotype.

DNA Methylation in Arabidopsis

The single base-pair resolution possible with next-generation sequencing allowed for direct analysis of the extent and frequency of methylation for each of the targeted cytosine contexts in Arabidopsis (Lister et al. 2008; Cokus et al. 2008). Overall, ~6% of cytosines were found to be methylated, in close agreement with bulk methylcytosine content measured by HPLC (Rozhon et al. 2008). In eukaryotes, cytosine methylation is catalyzed by DNA methyltransferase. Arabidopsis has three classes of these genes: DNA Methyltransferase 1 (MET1), Chromomethylase 3 (CMT3), and Domains Rearranged Methyltransferase 1 and 2 (DRM1/2) (Law and Jacobsen 2010). Each DNA methyltransferase is regulated by a distinct genetic pathway and targets a specific sequence context (Law and Jacobsen 2010).

MET1, a homolog of the mammalian methyltransferase DNMT1, deposits DNA methylation at dinucleotide CG accounting for ~55% of methylated sites in Arabidopsis (Ronemus et al. 1996; Finnegan et al. 1996; Saze et al. 2003; Kankel et al. 2003; Lister et al. 2008; Cokus et al. 2008). MET1 may target hemimethylated DNA that is present after DNA replication allowing for faithful propagation of methylation to the symmetric CG site on the newly synthesized daughter strand. This mechanism has been demonstrated in mammalian systems, where the UHRF1 (Ubiquitin-like containing, PHD, Ring Finger) protein binds to both DNMT1 and hemimethylated DNA providing an interface between the DNA methyltransferase and target DNA (Bostick et al. 2007; Sharif et al. 2007; Hashimoto et al. 2008; Kim et al. 2009). VIM1, an Arabidopisis homologue of UHRF1, is important for maintenance of CG methylation, and has been shown to bind methylated DNA, though not exclusively hemimethylated CG, suggesting that it may play a similar, though perhaps not biochemically identical, function as UHRF1 (Woo et al. 2007; Kraft et al. 2008). UHRF1 has also been shown to bind specific histone methyltransferases and deacetylases suggesting a direct association between DNA methylation and histone silencing marks (Kim et al. 2009). A null mutation in the Arabidopsis HDA6, a histone deacetylase, results in reduced CG methylation and derepression of many of the same genes observed in the met1–3 null mutation (Aufsatz et al. 2007; To et al. 2011; Aufsatz et al. 2002). MET1 and HDA6 have been shown to directly interact in vivo and in vitro supporting a model that co-localization of MET1 and histone modifiers at targeted loci is important in CG methylation in Arabidopsis (Liu et al. 2012).

CMT3, a plant specific DNA methyltransferase, catalyzes DNA methylation in the CHG context and accounts for ~22% of all methylated sites in Arabidopsis (where H = A, C or T) (Bartee et al. 2001; Lindroth et al. 2001; Lister et al. 2008; Cokus et al. 2008). Genetic evidence indicates that the establishment of the CHG methylation is dependent upon the histone mark H3K9me2 and H3K27me, and that CMT3 can directly bind to H3K9me2 (Lindroth et al. 2004). Consistent with this, CHG methylation is dependent on the activity of a family of H3K9 methyltransferases (KRY/SUVH4, SUVH5, and SUVH6) and CHG methylation is associated with H3K9me2 genome wide (Jackson et al. 2002; Bernatavichute et al. 2008; Malagnac et al. 2002). Furthermore, KRY has been shown to itself directly bind to methylated DNA, with a preference for CHG and CHH contexts, providing a potential reinforcing loop between H3K9me2 and CHG methylation (Johnson et al. 2007). The loss of a H3K9 demethylase, IBM1, also has a profound effect as many genes become methylated in the CHG context in the ibm1 null, suggesting that this enzyme is actively demethylating H3K9 from many genes and preventing CHG methylation (Inagaki et al. 2009).

The third class of DNA methyltransferase, DRM1 and DRM2, methylate cytosines in the asymmetric CHH sequence context which comprises 22% of all methylated sites (Cao and Jacobsen 2002; Cokus et al. 2008; Lister et al. 2008). DMR2 is directed to target loci by small RNAs (smRNA) and a specific RNA interference pathway known as RNA-directed DNA methylation (RdDM) (Chan et al. 2004). A number of mutants have been discovered in this pathway allowing for extensive epistatic and molecular analysis (Law and Jacobsen 2010). The general model consistent with the current data begins with expression of a primary transcript by the DNA polymerases Pol IV, a plant specific variant of Pol II (Herr et al. 2005; Kanno et al. 2005; Onodera et al. 2005). This transcript is then converted into double stranded RNA by the RNA-directed RNA polymerase RDR2, processed by the Dicer-like DCL3 and the resultant short interfering RNA (siRNA) is loaded into the Argonaut AGO4(Zilberman et al. 2003; Chan et al. 2004; Xie et al. 2004) The AGO4-siRNA then directs a protein complex that includes the DNA methyltransferase DRM2 to specifically target methylation to a locus via siRNA hybridization to a “guide” transcript generated by Pol V, a distinct DNA polymerase variant to the one which synthesized the primary transcript used to generate siRNA (Henderson and Jacobsen 2007; Matzke et al. 2009). As is the case for CMT3, DRM2 also requires two SUVH methylcytosine binding proteins for CHH methylation activity(Johnson et al. 2008).

RdDM targeted loci are characterized by dense methylation in all sequence contexts, not just CHH, and show a high incidence of overlap with siRNA, particularly of the 24bp size class (Lister et al. 2008). Thus, though the establishment of methylation by RdDM at newly inserted transgenes has been shown to be dependent upon the presence of DRM2 and an intact siRNA pathway, the other two methyltransferases must be recruited to these sites as well since methylation is found in all sequence contexts (Cao and Jacobsen 2002; Chan et al. 2004). It has yet to be clearly established by what mechanisms all three pathways converge on this spot and will certainly be an important future direction of study. It is interesting to note, however, that several of the DNA methylation “readers”, VIM1–3 and the SUVH2–9 family members, which have been shown to have effect specific DNA methylation context, show much lower specificity in regards to binding of specific contexts (Johnson et al. 2008; Woo et al. 2008; 2007). It will be interesting to see if recruitment of histone methylatransferases and deacetylases to methylated DNA may provide a potential “crosstalk” between the distinct methyltransferase pathways. The potential roles for histone modifiers in RdDM is an active field of investigation (Saze et al. 2012).

Different RdDM loci show substantial heterogeneity in the relative contribution of specific genes in the methyltransferase, siRNA and histone modifying pathways. The variation in the role of the methylatransferases in silencing at different loci can clearly be seen by the fact that of the over 500 genes that are derepressed in the met1 and ddc backgrounds less than 10% are upregulated in both (based on a comparison from data from http://neomorph.salk.edu/~julian/clusters_simple.php) (Lister et al. 2008). Different histone methyltransferases have similarly been observed to have loci-specific effects on DNA methylation as different members of the SUVH3–9 show overlapping effects at some loci and independent effects at others. In the RNAi pathway a number of the key genes are members of gene families and in some cases mutants in specific paralogs affect some RdDM loci more than others. For example, while AGO4 is required to targeting RdDM at many loci (Zilberman et al. 2003; 2004), AGO6 has also been shown to contribute to RdDM in a partly redundant manner at a subset of loci (Zheng et al. 2007). Additionally, while a Pol V mutant has been shown to disrupt RdDM at several loci (Wierzbicki et al. 2009), potentially due to lose of the scaffolding transcript, there is evidence that Pol II may contribute to generation of scaffold transcripts and/or is involved in the establishment or stabilization of the interaction of the AGO-siRNA at some low copy number loci (Zheng et al. 2009). Further integration of multiple levels of epigenomic information in various null mutants may provide a clearer picture of the contribution of different methyltransferase pathways, RNAi paralogs and histone-modification enzymes that contribute to loci specific regulation.

DNA transposons and retrotransposons are heavily targeted by RdDM in plants and a primary function of this epigenetic regulation appears to be transcriptional silencing to limit transposon mobilization (Law and Jacobsen 2010). One study found that 74% of annotated transposons are methylated and the remaining 26% of unmethylated transposon were enriched for short, highly degenerate relics many of which might be expected to have lost transcriptional and transposable competency (Ahmed et al. 2011). Transcriptome sequencing of met1 has shown that hundreds of transposon/pseudogenes are derepressed upon the loss of CG methylation (Lister et al. 2008). Furthermore, mobilization of several classes of transposons have been seen in met1, as well as in other RdDM deficient backgrounds, clearly demonstrating a protective role for RdDM silencing of transposons on genomic integrity (Kato et al. 2004; Mirouze et al. 2009; 2009; Reinders et al. 2009; Tsukahara et al. 2009; Matsunaga et al. 2012). Our understanding of how transposons are targeted for RdDM is still a work-in-progress, though several features common to transposons have been proposed to be have a role in attracting RdDM including enrichment for repeat sequences, high copy number, convergent and sense-antisense transcription, and high expression levels observed when silencing is derepressed and transposition is occurring (Slotkin and Martienssen 2007).

Gene expression is also affected by RdDM, frequently resulting in partial or complete transcriptional silencing. In both the met1 and ddc background hundreds of genes are up-regulated, significantly changing the transcriptome of these mutants (Lister et al. 2008). In the following sections we will explore artificially induced epialles and discuss some of the lessons they have taught us regarding the phenotypic effects, genetic pathways, stability and unique properties of RdDM on gene regulation.

Hypomethylation mutants and induced epialleles

Segregation analysis of a met1 loss-of-function mutant shows that homozygous individuals are recovered at only 2% of the expected frequency indicating that MET1 activity is critical in early embryo development (Saze et al. 2003). This may, at least in part, be due to the demonstrated role of methylation in controlling imprinted genes. Surprisingly, however, the met1 null plants that do germinate do not display gross morphological defects, as plants appear to tolerate the complete loss of CG methylation. However, there are a number of stable heritable phenotypes that segregate independently of the met1 allele (Reinders et al. 2009). Phenotypically, the triple mutant drm1 drm2 cmt3 (hereafter ddc) is less severe than the met1 though reduced stature and leaf curling is observed in this background (Chan et al. 2006a; Cao and Jacobsen 2002). DDM1, a chromatin remodeling protein, is not directly responsible for deposition of methyl groups, yet a null ddm1 mutation cause a 70% drop in global methylation primarily due to the loss of CG methylation, similar to what is seen in met1 (Jeddeloh et al. 1998; 1999; Vongs et al. 1993). ddm1 has some of the same phenotypic effects as met1 including late flowering and reduced stature that appear to be due to the some of the same hypomethylated epialleles (Johannes et al. 2009; Reinders et al. 2009). Unlike met1, however, ddm1 plants require several generations of propagation to fully establish the hypomethylated state and likewise the hypomethylated phenotypes become more pronounced in later generations of inbred plants (Brzeski and Jerzmanowski 2002; Jeddeloh et al. 1999; 1998).

Stable, heritable phenotypes in these hypomethylated backgrounds have been mapped by genetic linkage. These induced epialleles have been instrumental in determining the components of the genetic pathway and identifying sequence features associated with RdDM. One extensively studied induced epimutant is in the FLOWERING WAGENINGEN (FWA) gene, a homeodomain transcription factor. In wild type plants, FWA is an imprinted gene expressed only by the maternal allele and only in the developing endosperm (Kinoshita et al. 2004). The epiallele fwa1, on the other hand, is unmethylated and expressed ectopically in vegetative tissues resulting in a late flowering phenotype (Soppe et al. 2000b; Kankel et al. 2003; Saze et al. 2003)). A hypomethylated fwa allele has been found to be responsible for the late flowering phenotype seen in both the ddm1 and met1 backgrounds as well as the original fwa1 allele identified from a chemically mutagenized population(Soppe et al. 2000a; Kankel et al. 2003; Saze et al. 2003).

While the late flowering phenotype is likely an artifact of ectopic expression, fwa has been a valuable tool for identifying many of the components of the RdDM pathway. Insertion of an additional copy of an FWA sequence by Agrobacterium-mediated transformation into a wild type plants results in silencing of the newly introduced transgene (Cao and Jacobsen 2002). However, when RdDM is compromised, the transgene remains unmethylated and expressed, resulting in the late flowering phenotype. Screening for mutants incapable of silencing the FWA transgene in this manner allowed for identification of the de novo methyltransferases DRM2 and provided the first demonstration for a role for RNAi silencing genes in RdDM (Cao and Jacobsen 2002; Chan et al. 2004).

In addition to its use as a reporter for mutants in the RdDM pathway, the FWA transgene system has been an informative model for understanding the types of primary sequence features responsible for attracting RdDM. The methylated region in wild-type FWA overlaps with an annotated retrotransposon of the SINE family (Chan et al. 2006b; Kinoshita et al. 2007). Constructs containing truncated versions of the FWA gene and upstream sequences have been used to demonstrate that the SINE element as necessary and sufficient for RdDM targeting (Chan et al. 2006b; Kinoshita et al. 2007). As the majority of the SINE element sequence consists of a pair of direct repeats, a common target for RdDM, a transgene that contained only the tandem repeat sequence was also tested and found to be a sufficient target for methylation(Chan et al. 2006b; Kinoshita et al. 2007). Furthermore, when the two direct repeats were reduced to just a single copy of each, no methylation occurred indicating that repeat-character and not the intrinsic sequence they encode is responsible for the targeting of RdDM (Chan et al. 2006b; Kinoshita et al. 2007).

Gene silencing as a direct consequence of proximal repeat sequence has been observed in a number of other induced epialleles. One well-characterized example is an epiallele of Suppressor of ddc (SDC), an F-box protein, which was identified as the cause of a dwarf phenotype in the ddc background (Henderson and Jacobsen 2008). In wild-type plants, SDC is methylated in a 2kb region of its promoter and no expression of the gene can be detected at any stage of development (Henderson and Jacobsen 2008). In the ddc background, this promoter methylation is lost and the gene is expressed in all tissues (Henderson and Jacobsen 2008). As is the case with FWA, methylation of the SDC promoter is associated with a tandem direct repeat, further indicating that repeat sequences by themselves can be sufficient for RdDM targeting (Figure 1). Unlike FWA, however, where methylation is largely restricted to the tandem repeats region, methylation and siRNA production in the SDC promoter is spread many hundreds of base pairs upstream and downstream of the repeats (Henderson and Jacobsen 2008). The spreading observed into SDC is lost in the met1 background, though the methylation at the repeat site is retained suggesting a possible role for MET1/CG methylation at least at this loci(Henderson and Jacobsen 2008). Interestingly, it was observed that siRNA and DNA methylation are capable of spreading independently from one another based on experiments with a set of RNAi and DNA methyltransferase null mutants(Henderson and Jacobsen 2008). In a bioassay using a two-component silencing system where one loci with a hairpin was used to target a second transgenic site, Daxinger et. al. observed that spreading of RdDM at the target site required a functional Pol IV, RDR2 and DCL3, suggesting the possibility that an Pol IV primary transcript was generated extending beyond the sequence shared by (Daxinger et al.). A deeper understanding of the genetic pathways and the underlying sequences that promote or contain spreading may be important for understanding natural variation in epialleles as this commonly observed phenomena can provide a mechanism by which methylation at repeats and transposons can impinge upon regulatory regions in neighboring genes.

Figure 1. FWA and SDC in wild type and hypomethylated mutant backgrounds.

The FWA gene is shown with tracks containing the methylcytosines (CG,tan; CHG blue; CHH,red), smRNA and mRNA maps of Col-0 and the met1 background (Lister et. al. 2008). The SDC gene is shown with the same tracks for the Col-0 and ddc background. The purple bar in the “Repeats” track represents a region containing direct repeats.

Transmittance of an epiallelic state from an endogenous gene to an unlinked homologous sequence demonstrates another important feature of epialleles that distinguishes them from genetic mutations. Methylation of the FWA transgene is strongly affected by the epiallelic state of the endogenous FWA. When wild-type plants are transformed with FWA the transgene is always silenced, while transformation of a plant containing the endogenous fwa1 epiallele results in silencing only 7% of the time (Chan et al. 2006b). This dependency of methylation at the transgene as a function of the epigenetic state of the endogenous gene strongly suggests that a direct epigenetic interaction exists between these unlinked loci. One model consistent with this would be that the siRNA produced at the endogenous copy is required for trans-RdDM targeting of the transgene, and that the fwa1 has a reduced siRNA level, an effect frequently observed at many hypomethylated sites (Lister et al. 2008). However, a northern blot of the siRNA specific to fwa1, an epiallele isolated from a chemically mutagenized population, did not show a significant reduction in siRNA levels compared to the wild-type FWA (Chan et al. 2006b). It is worth noting, though, that the same blot showed no reduction in of siRNA from an fwa generated in the met1 background, while another study had shown by LNA pull down experiments that the siRNA level in met1 decreases substantially (Lippman et al. 2004). A similar substantial decrease in met1 siRNA is also evident in bisulfite and smRNA sequencing maps of met1 (Figure 1) (Lister et al. 2008). It might be possible, therefore, that the smaller dynamic range of the northern blot failed to capture a small, but potentially significant decrease in the fwa siRNA that could explain the reduced RdDM targeting in trans. However, it is impossible to rule out that some other unidentified epigenetic state of the fwa1 unrelated to its siRNA levels was responsible for limiting its ability to transmit methylation to the endogenous fwa copy. Understanding whether a simple siRNA dosage effect or other unidentified mechanism dictates the efficacy of trans-RdDM of the transgenic FWA could be very relevant for studying naturally occurring epialleles where trans-RdDM is frequently observed (Greaves et al. 2012a).

In a similar vein, it has been observed that maintenance of a hypomethylated epiallele is strongly dependent upon how much of the wild-type methylation and siRNA are retained in the hypomethylated background. In a study of remethylation of loci in the ddm1 background, after reintroduction of a wild type DDM1 by outcrossing, it was observed that loci that showed higher retention of siRNA and methylation were more likely to remethylate than loci where most of the siRNA and methylation were lost in the hypomethylated parent (Teixeira et al. 2009). Though not examined in this study, fwa generated in ddm1 and met1 show significantly different likelihood to be remethylation after outcrossing, and this may be due to the degree of siRNA/methylation depletion seen at the fwa loci in each background. The met1-generated fwa shows a much greater depletion of siRNA levels compared to the ddm1-generated fwa (Lippman et al. 2004), but in both backgrounds in the first generation after an outcross the fwa remains unmethylated. However, after three additional generations the ddm1-generated fwa was silenced in most lines, while in the met1 background, by contrast, no silencing was observed (Reinders et al. 2009; Johannes et al. 2009). This suggests that the degree of hypomethylation and related loss of siRNA in the two backgrounds may be important determinate of the stability of their respective fwa epialleles. Additionally, it supports a model where an siRNA dose-dependency may be important for the stability and reestablishment of RdDM.

These induced epialleles have been informative models, providing tractable experimental system for exploring the pathways and sequence requirements involved in cis-and trans-silencing. Much of what has been observed in these systems has also been observed in several naturally occurring epialleles that have been described over the past decade. In addition to providing confirmation of many of the phenomena observed in the induced epialleles, these natural epialleles provide concrete examples of how epigenetic and structural variation in divergent accessions can interact to establish cis and trans methylation, ultimately effecting traits in a non-Mendialian fashion.

Natural-occuring epialleles

One well-characterized example of a natural epiallele occurs in the phosphoribosylanthranilate isomerase (PAI) gene family Figure 2). In Col-0, as well the majority of other accessions examined, PAI is encoded by a three member gene family (Bender and Fink 1995). However, in some accession, including Wassilewskija (Ws) a fourth paralog, PAI4 is found in a tail-to-tail arrangement with PAI1 (Bender and Fink 1995). The large inverted repeat resulting from this PAI1/4 rearrangement results in targeting of RdDM over the entire length of both genes and results in lower expression of PAI1 relative to Col-0 (Bender and Fink 1995). In addition to silencing at the PAI1/4, the unlinked PAI2 and PAI3 genes are also methylated in Ws, but not in Col-0 (Bender and Fink 1995). When Ws was crossed to Col-0, all three unmethylated copies of the Col-0 PAI genes become methylated in the F1. In F2 segregants that lose the Ws PAI1/4 methylation of the PAI2 and 3 genes are also lost demonstrating a direct requirement of the inverted repeat for maintaining the trans RdDM (Bender and Fink 1995).

Figure 2. Effect of PAI1/4 rearrangement on RdDM in trans.

The three membered phosphoribosylanthranilate isomerase (PAI) gene family in Col-0 is depicted with the corresponding cartoon showing the absense of RNA-directed DNA methylation (RdDM) in the track labeled “mC”. The Ws accession contains a fourth copy (PAI4) in a tail-to-tail arrangement with PAI1 and all four genes in this accession are targeted by RdDM. The Col-0 × Ws show the consequence of an F1 cross between Ws and Col-0 which results in RdDM targeting of Col-0 PAI gene family from siRNA produced at the Ws PAI1/4 direct repeat.

Though the inverted repeat in PAI1/4 is due to a gene duplication, a similar requirement for repetitive sequence for RdDM targeting as is seen in FWA and SDC suggests that structural variation in repeat sequences may be a powerful trigger for RdDM. As gene duplications and rearrangements are also a common in plant genome evolution this type of feature may be also be a common mechanism by which epialleles are established. Furthermore, PAI illustrates how once a rearrangement is established, RdDM in trans may influence transmission of this epiallelic state to close paralogs. However, at least in this case, genetic segregation of the triggering repeat results in loss of the epiallelic state indicating that at some epialleles the genetic and epigenetic states are linked.

Naturally occurring epialleles with visible phenotypic consequence have been identified in other plants including the tomato Colorless non-ripening (Cnr) locus, peloric in Linaria and paramutation of the Booster (b’) locus in maize (Cubas et al. 1999; Manning et al. 2006; Stam et al. 2002). While all three provide interesting examples of natural epimutations, the b’ locus is the best characterized in terms of the genetic requirements and sequence features associated with the establishment of methylation (Stam et al. 2002). The b’ locus encodes a transcription factor that regulates genes involved in pigmentation (Chandler and Stam 2004). Two alleles at the b’ locus, the B-I and the B’, result in a dark or light pigmentation phenotype and this difference corresponds to a higher or lower expression levels of each allele, respectively. Furthermore, the expression is inversely correlated to different amounts of methylation at seven direct repeats upstream of the gene. When a cross is made between plants containing the B-I and B’, the methylation at the direct repeat of B-I allele increases and expression decreases, which is subsequently referred to as the paramutated B’* allele. B’* acts like any B’ allele as it is able to transmit this methylated state to other B-I alleles. Phenotypically, the consequence of paramutation is that Mendelian segregation is violated as only B-I (lightly pigmented) progeny are produced from a parent with B’ (dark) and B-I (light) alleles.

Like PAI, the B’* is a similar case of trans-methylation associated with repeat sequences, though it is distinct from PAI in that once the B’* methylation is established it no longer requires the B’ allele that initiated it to maintain it. This is an important distinction as it shows that epiallelic states can be transmitted independently of the feature which triggers them. This could allow for independent selection on the source and target of a trans-RdDM event. As is the case with fwa, functional components of the RNAi pathway have been shown to be required indicating that paramutation is likely to be another example trans RdDM (Sidorenko et al. 2009; Erhard et al. 2009)

A recent study further illustrates how the interplay between structural variation and cis- and trans-RdDM can affect important traits (Figure 3). Previously it had been observed that in a segregating population derived from a Col-0/Shandra (Sha) hybrid cross that one of the four possible homozygous combinations of two unlinked loci, K4 and K5, were significantly underrepresented (Durand et al. 2012). This deviation from expected Mendelian segregation would be consistent with selection against plants where Col-0 is homozygous at K4 and Sha is homozygous at K5. Fine-mapping of these regions determined that Col-0 and Sha each have only one expressed folate transporter at K4 and K5 respectively (Durand et al. 2012). Thus, the underrepresented combination lacks of a functional copy of the folate transporter gene which results in a stunted, less-fertile plant.

Figure 3. Genetic incompatibility between Col-0 and Sha due to genetic and epigenetic variation.

An expressed copy of AtFolta gene is found at K4 locus in Sha accession (AtFolta2) and at the K5 locus of Col-0 accession (AtFolta1). The Sha accession also contains what appears to be a functional copy of AtFolta1 at K5 but it is silenced by DNA methylation. The source for the silencing signal appears to come from a rearranged AtFolta-like sequence at K4 in Sha (solid arrow). In a single F3 of a Col-0 × Sha cross, the Col-0 AtFolta1 is silenced, likely due to RdDM in trans originating from the Sha AtFolta rearrangement (dashed arrow).

In a surprising twist, however, it was discovered that the Sha accession does in fact contain what appears to be a functional ortholog of AtFolta1 at the K4 loci, but that it is silenced by methylation (Durand et al. 2012). The source of the silencing signal was found to come from a rearrangement of an AtFolta-like sequence in Sha at K5 proximal to the functional Sha AtFolta2. Thus, the Sha AtFolat-like rearrangement appears to silence the Sha AtFolat1 in trans, resulting in the establishment of these incompatible loci (Durand et al. 2012). Further confirmation of this model comes from the observation that at a low frequency the Sha AtFolta-like rearrangement can silence the Col-0 AtFolta1 (Durand et al. 2012). Thus, a phenotype which may be an important first step on the road towards speciation is directly due to a complex interaction involving gene duplication and rearrangement induced trans RdDM.

Clearly, repeat sequences and transposons, proximal to genes will contribute significantly to natural epigenomic variation as they have been implicated in all examples known to date. As such, variation in the epigenome between accessions is likely to be tightly associated with structural variation. However, a recent pair of papers suggest another potential source of epigenetic variation that is not directly the consequence of changes in primary sequence (Becker et al. 2011; Schmitz et al. 2011). In these two studies the methylomes were sequenced for a set of five Arabidopsis lines derived from a single ancestor propagated by single seed descent for 30 generations (Shaw et al. 2000). Both studies found that several percent of the methylcytosines were distinct between the descendants when compared to the ancestral lines and even greater variation existed between the descendants themselves (Becker et al. 2011; Schmitz et al. 2011). As these lines had previously been sequenced and only a small number of SNPs were observed (average of 30 per line) these large changes in methylation appear to be occurring in the absence of changes in primary sequence (Ossowski et al. 2010). While most of the differential methylcytosines were found to occur in gene bodies, several hundred RdDM loci were found to have diverged in at least one line (Becker et al. 2011; Schmitz et al. 2011). In four cases where the loss of methylation occurred proximal to genes derepression of silencing was observed indicating that these stochastic change do impact transcriptional regulation (Schmitz et al. 2011).

Both studies observed that some of the same variant methylcytosines and methylated loci arose independently in different descendants, suggesting that some of these sites may represent hot-spots of epigenetic variation (Becker et al. 2011; Schmitz et al. 2011). Supporting this observation, remethylation was observed at several demethylated loci when an additional generation was tested (Becker et al. 2011). How these rapid changes in the epigenetic landscape in the absence of genetic variation might contribute to longer term divergence in natural strains is a open question. However, these findings raise the interesting possibility that “epigenetic drift” may be operating as a source of variation on which selection could possible be acting independent of the genome sequence.

Natural epigenomic variation

While sequence variation between divergent Arabidopsis accessions has been evident for many years, it has recently come to light that extensive epigenetic variation also exists in natural plant populations (Greaves et al. 2012b; Shen et al. 2012; Groszmann et al. 2011; Vaughn et al. 2007). Methylome maps generated with next-generation bisulfite sequencing for the C24 and Ler by two different groups showed that ~20% of methylated cytosines are not shared between these two accessions (Greaves et al. 2012b; Shen et al. 2012). Similarly ~20% of genomic loci targeted by RdDM were also different between the accession indicating substantial variation in the epigenetic landscape which may, in part, be responsible for gene expression variation.

In addition to the Ler and C24 parents, the methylomes and smRNAomes from reciprocal crosses between the two accessions were also examined by both groups. When crossed, these two accessions show significant heterosis, a phenomenon whereby hybrid offspring grow more vigorously than their parents. Both studies explored a potential role of hybrid effects on RdDM in this process. The first study observed a decrease of 24nt siRNAs in the F1 hybrids relative to the parental strains and observed that this corresponded to a decrease in CHH methylation (Greaves et al. 2012b; Groszmann et al. 2011). These findings suggest that RdDM is compromised in the hybrids relative to their parental strains and the authors proposed that this could contribute to increased vigor observed in this hybrid due to effects on gene regulation (Groszmann et al. 2011). In the second study, however, no significant change in the 24nt siRNAs levels in the hybrids was observed and an fact an increase in overall methylation at siRNA loci was reported (Shen et al. 2012). Additionally they showed that a global decrease in smRNA in the hen1 background was associated with a loss of heterosis (Shen et al. 2012). In recent study of siRNA levels in maize hybrids observed a decrease in the siRNA levels similar to Grozzmann et. al. (Barber et al. 2012). However, in the mop1 mutant, which reduces global 24nt siRNA levels, heterosis was unaffected (Barber et al. 2012). This led the authors to conclude that MOP1 specific smRNAs may not play a role in the establishment of hybrid vigor (Barber et al. 2012). It is certainly an interesting hypothesis that epigenomic interactions may play some role in hybrid vigor and considering the incredibly agronomical value of this poorly understood phenomena, there will likely be future investigations into this possibility.

Concluding Remarks

As we can see from the many of the examples of both induced and natural occurring epialleles, epigenetic variation is tightly linked to sequence and in particular transposons and repeats. In the past year, resequenced genomes for 100 Arabidopsis accessions have been published, and though the read size limited the length of de novo assembly possible, there is evidence for substantial structural variation between the accessions (Cao et al. 2011; Gan et al. 2011). For example, Cao et al. predicted that as much as 80% of the TAIR10 annotated transposons are partially or completely deleted within at least one of the 80 sequenced accessions (Cao et al. 2011). As longer reads at lower cost have been the current trend, it is reasonable to assume that in the near future, genome assemblies of accessions comparable in quality to Col-0 reference genome will be available. High quality assemblies should be extremely informative as to how frequently genomic divergence, and in particular structural variation in the form of transposons and repeat features, dictate epigenomic divergence or alternatively, how frequently epigenomic differences may be due to the type of “drift” that is seen in the 30 generations experiments. Finally, complete genome assemblies may allow us to address the frequency, mechanisms, and sequence features underlying RdDM transmission between unlinked homologous sequence, both in individual parents and in hybrid crosses which may potentially be an important source of trait variation displaying non-Mendelian behavior.