Noncoding RNAs and DNA Methylation in Plants

Abstract

Cytosine DNA methylation is an epigenetic modification in eukaryotes that maintains genome integrity and regulates gene expression. The DNA methylation patterns in plants are more complex than those in animals, and plants and animals have common as well as distinct pathways in regulating DNA methylation. Recent studies involving genetic, molecular, biochemical and genomic approaches have greatly expanded our knowledge of DNA methylation in plants. The roles of many proteins as well as non-coding RNAs in DNA methylation have been uncovered.

DNA methylation is a type of epigenetic modification that occurs at the bases of the DNA strands. In both animals and plants, DNA methylation is mostly found at the 5^th carbon of the cytosine pyrimidine ring. Methylation on DNA may alter the transcriptional activity of associated genes or maintain genome integrity by repressing transposable elements. The methylation status at some loci may respond to external or internal signals, however, as a type of epigenetic modification, the overall DNA methylation status is heritable. In animals, almost all methylated cytosines occur in the CG sequence context, although non-CG methylation has been found in embryonic stem cells and neurons [1, 2]. The mammalian DNA methyltransferase 3 (DNMT3) family functions in the establishment of DNA methylation during the development of germ cells [3, 4]. Thereafter, DNA methyltransferase 1 (DNMT1) maintains DNA methylation during DNA replication [4]. In contrast to DNA methylation in animals, plant cytosine methylation has been found at all cytosine sequence contexts—CG, CHG and CHH (H=A, T or C). In plants, de novo DNA methylation is mediated by the RNA-directed DNA methylation (RdDM) pathway. After DNA replication, multiple DNA methyltransferases are employed to maintain cytosine methylation at different sequence contexts. In this review, we focus on the progress of plant DNA methylation studies in recent years. We first discuss the DNA methylation landscapes in plants. Then we describe the mechanisms underlying de novo DNA methylation, maintenance of DNA methylation in different sequence contexts and DNA demethylation in plants. We mainly focus on the functions of proteins and non-coding RNAs involved in these processes in the model plant Arabidopsis. For easy reference, Arabidopsis genes that function in DNA methylation are listed in Table 1. Furthermore, the transgenerational inheritance and variation of DNA methylation in plants are also discussed.

Table 1.

Gene	Genome Identifier	Gene product	Function	Suggested Reference
DRM2	AT5G14620	DNA methyltransferase	De novo methylation and maintainance of CHH methylation through RdDM	[40]
DRM1	AT5G15380	DNA methyltransferase	A homolog of DRM2	[40]
MET1	AT5G49160	DNA methyltransferase	Maintains CG methylation	[43, 44]
CMT3	AT1G69770	DNA methyltransferase	Maintains CHG methylation	[50, 51]
CMT2	AT4G19020	DNA methyltransferase	Maintains DNA methylation at heterochromatin	[44, 58]
NRPD1	AT1G63020	Largest subunit of Pol IV	Generates 24-nt siRNA precursors in RdDM	[15, 16]
NRPE1	AT2G40030	Largest subunit of Pol V	Generates scaffold transcripts in RdDM	[11, 85, 86]
NRPB2	AT4G21710	Second largest subunit of Pol II	Generaes scaffold transcripts at some RdDM target loci	[27]
CLSY1	AT3G42670	SNF2-like chromatin remodelling protein	Interacts with Pol IV, functions in 24-nt siRNA precursor biogenesis	[18]
DRD1	AT2G16390	SNF2-like chromatin remodelling protein	Part of the DDR complex, functions in the recruitment of Pol V to RdDM target loci	[28, 29]
DDM1	AT5G66750	SNF2-like chromatin remodelling protein	Functions in the maintenance of DNA methylation at heterochromatin	[58, 87]
SUVH2	AT2G33290	SRA domain methylcytosine-binding protein	Functions in the recruitment of Pol V to RdDM loci	[33, 88]
SUVH9	AT4G13460	SRA domain methylcytosine-binding protein	A homolog of SUVH2	[33, 88]
VIM1	AT1G57820	SRA domain methylcytosine-binding protein	Functions in the maintenance of CG methylation	[45, 46]
VIM2	AT1G66050	SRA domain methylcytosine-binding protein	A homolog of VIM1	[46]
VIM3	AT5G39550	SRA domain methylcytosine-binding protein	A homolog of VIM1	[46]
RDM4	AT2G30280	IWR (Interacts with Pol II)-type transcription factor	Interacts with Pol IV, functions in 24-nt siRNA precursor biogenesis	[19, 20]
RDR2	AT4G11130	RNA dependent RNA polymerase	Interacts with Pol IV, functions in 24-nt siRNA precursor biogenesis	[89, 90]
SHH1	AT1G15215	SAWADEE domain and homeodomain containing protein	Interacts with Pol IV, functions in 24-nt siRNA precursor biogenesis	[17, 21]
DCL3	AT3G43920	Dicer-like protein	Generates 24-nt siRNA duplexes in RdDM	[89]
AGO4	AT2G27040	Argonaute protein	Binds to 24-nt siRNAs, functions in RdDM	[25, 42, 91]
HEN1	AT4G20910	Small RNA methyltransferase	Methylates siRNA duplexes	[24]
DMS3	AT3G49250	Structural maintenance of chromosomes (SMC) hinge domain-containing protein	Part of the DDR complex, functions in the recruitment of Pol V to RdDM target loci	[29, 31]
RDM1	AT3G22680	Contains a domain of unknown function	Part of the DDR complex, functions in the recruitment of Pol V to RdDM target loci	[28–30]
STA1	AT4G03430	PRP6-like splicing factor	Functions in scaffold transcript biogenesis	[35]
KTF1	AT5G04290	Transcription elongation factor SPT5-like protein	Functions downstream of scaffold transcripts in RdDM	[36, 37]
IDN2	AT3G48670	dsRNA-binding protein	Functions downstream of scaffold transcripts in RdDM	[38, 92]
KYP	AT5G13960	Histone H3K9 methytransferase	Functions in the maintenance of CHG methylation through CMT3	[93]
SUVH5	AT2G35160	Histone H3K9 methytransferase	A homolog of KYP	[94, 95]
SUVH6	AT2G22740	Histone H3K9 methytransferase	A homolog of KYP	[54]
IBM1	AT3G07610	Histone demethylase	Functions in preventing ectopic DNA methylation at gene bodies	[55, 56]
ROS1	AT2G36490	DNA glycosylase	Demethylates DNA	[60, 62]
DML2	AT3G10010	DNA glycosylase	A homolog of ROS1	[59, 96]
DML3	AT4G34060	DNA glycosylase	A homolog of ROS1	[59, 96]
DME	AT5G04560	DNA glycosylase	A homolog of ROS1 that is active in central cells	[69–71]
ROS4	AT3G14980	MBD domain-containing Histone H3 acetyltransferase	Functions in DNA demethylation at some ROS1 target loci	[61]
ROS3	AT5G58130	RNA-binding protein	Functions in DNA demethylation	[66]
ZDP	AT3G14890	Phosphoesterase	Functions downstream of ROS1 in DNA demethylation	[63]
ATXRCC1	AT1G80420	DNA repair protein	Functions in DNA demethylation	[65]

The DNA methylation landscape in plants

By combining bisulfite conversion and high-throughput sequencing, genome-wide DNA methylation has been profiled in the model plant Arabidopsis [5, 6]. The DNA methylation profiling using five-week-old Arabidopsis plants indicates that cytosine methylation levels in CG, CHG and CHH sequence contexts are 24%, 6.7% and 1.7% respectively [5]. For cytosine sites in Arabidopsis, most CG sites are either unmethylated or methylated at the 80%–100% level [5, 6], suggesting that CG methylation is precisely maintained after DNA replication. In contrast, methylated CHG sites have a wide range of methylation levels, and most methylated CHH sites have a low methylation level [5, 6], suggesting that differentially methylated CHG and CHH loci exist in different cells. In consistency with its function in transposon silencing, DNA methylation is highly enriched at pericentromeric regions where transposon and repeat frequencies are high [5, 6]. At transposons or other repeats, cytosines in all three types of sequence contexts could be methylated; whereas at the bodies of some protein-coding genes, CG methylation exists without repressing the expression of the gene [5]. Although the role CG methylation at gene bodies is not well understood, a study using different Arabidopsis accessions has shown positive correlation between gene body CG methylation levels and gene expression levels [7]. Similar to Arabidopsis, the DNA methylation profiles from soybean [8] and rice [9] also show similar distribution patterns of CG, CHG and CHH methylation in transposable elements and gene bodies.

De novo methylation and the RNA-directed DNA Methylation pathway

The establishment of DNA methylation in plants is mediated by the RNA-directed DNA Methylation (RdDM) pathway. In 1994, Wassenegger et al. first reported that the potato spindle tuber viroid (PSTVd) transgene in tobacco was methylated when the viroid RNA-RNA replication occurred, which indicated that DNA methylation was directed by homologous RNA [10]. It is now clear that the RNAs that direct DNA methylation are 24-nt (nucleotide) small interfering RNAs (siRNAs). In addition to the 24-nt siRNAs, longer non-coding RNAs specifically referred to as the scaffold RNAs also play a very important role in guiding the methyltransferase to target loci [11]. During RdDM, these two types of non-coding RNAs as well as several proteins are involved. The 24-nt siRNAs are generated by DICER-LIKE 3 (DCL3) from long double-stranded RNA (dsRNA) precursors, and then the small RNAs are loaded into AGO4 [12]. The small RNA-AGO4 complex is recruited to the RdDM target loci by the homologous nascent scaffold RNA through sequence complementarity between the siRNA and the scaffold RNA, and following this interaction, the DNA methyltransferase DOMAINS REARRANGED METHYLTRANSFERASE 2 (DRM2) is recruited to the target loci [12] (Figure 1).

Figure 1. The RNA-directed DNA methylation pathway (RdDM) in Arabidopsis.

At the early stage of RdDM, RNA Polymerase IV (Pol IV) is proposed to generate a single-stranded non-coding RNA, which is converted into double-stranded RNA (dsRNA) by RNA-DEPENDENT RNA POLYMARASE 2 (RDR2). SAWADEE HOMEODOMAIN HOMOLOG 1 (SHH1), CLASSY1 (CLSY1), RNA-DIRECTED DNA METHYLATION 4 (RDM4) and RDR2 were found to be in the same complex as Pol IV and may assist the recruitment of, or transcription by, Pol IV. 24-nt siRNA duplexes are produced from dsRNA precursors by DICER-LIKE 3 (DCL3) and are methylated at their 3′ ends by HEN1. The 24-nt siRNAs are exported into the cytoplasm to be loaded into AGONAUTE4 (AGO4). The molecular chaperon HSP90 facilitates the release of the siRNA passenger strand. The AGO4 protein loaded with an siRNA is translocated into the nucleus. RNA POLYMERASE V (Pol V) generates a non-coding scaffold RNA at RdDM loci. SU(VAR)3-9 HOMOLOG 2 (SUVH2) or its homolog SUVH9 contributes to the recruitment of Pol V to methylated DNA. A protein complex containing at least three proteins, DEFECTIVE IN RNA-DIRECTED DNA METHYLATION 1 (DRD1), DEFECTIVE IN MERISTEM SILENCING 3 (DMS3) and RNA-DIRECTED DNA METHYALTION 1 (RDM1), is also required to recruit Pol V to target loci. A splicing factor STABILIZED 1 (STA1) is required for the production of scaffold RNAs, although its mechanism of action is unclear. The scaffold RNA is able to recruit the silencing effector AGO4 as well as KOW DOMAIN-CONTAINING TRANSCRIPTION FACTOR 1 (KTF1), which may help to recruit downstream factors of RdDM. A dsRNA binding protein INVOLVED IN DE NOVO 2 (IDN2) is proposed to recognize the complex of the siRNA and the scaffold RNA. These events further recruit DOMAINS REARRANGED METHYLTRANSFERASE 2 (DRM2) to the RdDM loci through an unknown mechanism.

The 24-nt siRNAs could be produced from both exogenous loci and endogenous loci. Exogenous RNAs such as transcripts derived from viruses or transgenes can be processed into siRNAs by the plant RdDM pathway to silence these foreign sequences at the transcriptional level [10, 13]. In addition, lots of endogenous loci such as transposable elements and other repeats also give rise to siRNAs, which mediate the maintenance of CHH DNA methylation through the RdDM pathway after DNA replication[14]. When the 24-nt siRNAs are produced from endogenous loci that are already methylated, the transcription of the target loci depends on the plant-specific RNA polymerase Pol IV. When the NRPD1 gene encoding the largest subunit of Pol IV is mutated, the production of almost all heterochromatic 24-nt siRNAs is abolished [14–16]. By affinity purification of NRPD1, Law et al. [17] identified several proteins that are present in the Pol IV complex, including CLASSY1 (CLSY1), RNA-DIRECTED DNA METHYLATION 4 (RDM4), SAWADEE HOMEODOMAIN HOMOLOG 1 (SHH1) and RNA-DEPENDENT RNA POLYMERASE 2 (RDR2). CLSY1 is an SNF2 domain-containing protein that is required for the accumulation of heterochromatic siRNAs and is likely to mediate the interaction between Pol IV and RDR2 [18]. RDM4 is an IWR (Interacts with Pol II)-type transcription factor that is also required for Pol II and Pol V transcription [19, 20]. The specificity of Pol IV recruitment to some target RdDM loci may be conferred by SHH1, whose SAWADEE domain is capable of recognizing histone H3 tails with unmethylated lysine (K)4 and methylated K9 [21]. RDR2 is proposed to convert the Pol IV transcripts into dsRNAs for Dicer processing. In vitro assays suggested that the activity of RDR2 in dsRNA synthesis requires Pol IV [22]. The production of 24-nt siRNAs from a dsRNA procursor requires the Dicer protein DCL3, which colocalizes with RDR2 in the nucleolus [23]. Like miRNAs in plants, the 24nt-siRNAs are methylated by the small RNA methyltransferase HEN1 at their 3′ ends to be stabilized [24]. In recent years, it was found that that the loading of 24-nt siRNA into AGO4 takes place in the cytoplasm and the maturation of the AGO4/siRNA complex requires the catalytic activity of AGO4 and the molecular chaperon HSP90 [25]. However, how siRNAs are exported to the cytoplasm and recruited to AGO4 is still unknown. After siRNA binding, AGO4 is translocated into the nucleus [25] and recruited to RdDM target loci by scaffold RNAs.

Scaffold RNAs are long nascent transcripts generated from RdDM target loci. These transcripts have triphosphates or 7meG caps at their 5′ ends and are not 3′ polyadenylated [11]. Given that scaffold RNAs physically interact with AGO4 [26], scaffold RNAs have been hypothesized to be bridges that link siRNAs to their target loci through sequence complementarity. The biogenesis of scaffold RNAs is independent of siRNA biogenesis. The scaffold RNAs are mainly produced by another plant-specific RNA polymerase, Pol V [11], and Pol II also contributes to the accumulation of scaffold RNAs at some loci [27]. In addition to the RNA polymerase, a putative chromatin-remodeling complex is also required for the biogenesis of scaffold RNAs. This complex is composed of at least three proteins: a SWI2/SNF2-like chromatin-remodeling protein DEFECTIVE IN RNA-DIRECTED DNA METHYLATION 1 (DRD1), a structural maintenance of chromosomes (SMC) hinge domain-containing protein DEFECTIVE IN MERISTEM SILENCING 3 (DMS3) and a protein without known functional domains named RNA-DIRECTED DNA METHYALTION 1 (RDM1) [28–31]. This complex is therefore termed the “DDR complex” [29]. These three proteins were biochemically co-purified [29] and genetic analysis indicates that all three genes are required for the association of Pol V with certain chomatin loci, mainly gene promoters and some evolutionarily young transposable elements [32]. A chromatin feature that attracts Pol V is the pre-existing DNA methylation, as the occupancy of NRPE1 (the largest subunit of Pol V) was lost at normal RdDM target loci and redistributed to other loci in the met1 mutant defective in the maintenance of CG methylation [33]. SU(VAR)3-9 HOMOLOG 2 (SUVH2) and its homolog SUVH9, which bind methylated DNA through the SET- and RING finger-associated (SRA) domain, recruit Pol V to methylated DNA [33, 34]. Recently, a PRP6-like splicing factor STABILIZED 1 (STA1) has been reported to be involved in the production of Pol V-dependent scaffold RNAs [35]. STA1 does not affect the splicing of most known RdDM genes and colocalizes with AGO4 in the Cajal body [35]. However, how STA1 functions in RdDM remains elusive.

The recruitment of AGO4-siRNA to RdDM target loci by Pol V-dependent scaffold transcripts initiates DNA methylation at these loci. The KOW DOMAIN-CONTAINING TRANSCRIPTION FACTOR 1 (KTF1) protein is also recruited to target loci in a Pol V-dependent manner and is proposed to function in the recruitment of chromatin modifiers together with AGO4. [36, 37]. Furthermore, a dsRNA-binding protein INVOLVED IN DE NOVO 2 (IDN2) acts downstream of scaffold RNA transcription and may function in nucleosome remodeling at target loci in RdDM [38, 39]. The final key step of RdDM is the recruitment of methyltransferase DRM2 or its homolog DRM1 [40]. However, a direct link between Pol V-dependent transcripts and the recruitment of DRM2 is still missing. Although there has been no evidence of the downstream effectors of RdDM being involved in the production of siRNAs, it should be noted that some of them such as AGO4, Pol V and DRM2 are required in the accumulation of siRNAs at a subset of RdDM loci [21, 41]. Considering that mutations in NRPE1 (encoding the largest subunit of Pol V) do not affect the accumulation of the AGO4 protein whereas those in NRPD1 and RDR2 do [41], the effect of Pol V and DRM2 on siRNA accumulation is likely due to feedback regulation from loss of DNA methylation. AGO4 could contribute to siRNA accumulation at multiple levels. First, the slicer activity of AGO4 is required for the maturation of the AGO4-siRNA complex [25], in which AGO4 may stabilize siRNAs by protecting them from nucleases. Second, AGO4 may promote the production of secondary siRNAs triggered by the slicing of scaffold RNAs by primary siRNAs. It has been found that the siRNA accumulation level at some RdDM target loci requires the slicing activity of AGO4 [41, 42], raising the possibility that secondary siRNAs are produced from these loci. Yet the cleavage products of scaffold transcripts by AGO4 in vivo have not been reported. Finally, the reduced siRNA accumulation at some loci in ago4 mutants could also result from the feedback regulation from reduced DNA methylation.

Maintenance of DNA methylation

After DNA replication, only the template strand is methylated. DNA methyltransferases in plants deposit methyl groups onto the nascent strand in order to maintain the DNA methylation patterns through cell division. It has been well characterized in Arabidopsis that the maintenance of DNA methylation in different cytosine sequence contexts is implemented by different methyltransferases through distinct pathways (Figure 2).

Figure 2. Maintenance of DNA methylation in different sequence contexts in Arabidopsis.

The maintenance of cytosine methylation in the CG sequence context requires DNA METHYLTRANSFERASE 1 (MET1). CHG methylation is mainly maintained by CHROMOMETHYLASE 3 (CMT3), which is guided by the histone H3K9me2 mark. DOMAINS REARRANGED METHYLTRANSFERASE 2 (DRM2) mediates the maintenance of CHH methylation at euchromatic short TEs or the edges of heterochromatic long TEs through the RNA-directed DNA methylation (RdDM) pathway. At the bodies of heterochromatic long TEs where the RdDM machineries cannot access, cytosine methylation in the CHH sequence context is maintained by CHROMOMETHYLASE 2 (CMT2). Crosstalks also exist between different DNA methylation pathways. Maintenance of CHH methylation at some loci requires CMT3 while maintenance of CHG methylation at some loci requires the RdDM pathway.

The maintenance of CG methylation is mediated primarily by DNA METHYLTRANSFERASE 1 (MET1)—the homolog of the mammalian DNMT1 [43, 44]. As a type of symmetric DNA methylation, methylated CG sites become hemi-methylated after DNA replication and MET1 is recruited to restore the target loci to a fully methylated state. In addition to MET1, the maintenance of CG methylation also requires three VARIATION IN METHYLATION family proteins: VIM1, VIM2 and VIM3. These proteins contain a plant homeodomain (PHD) domain, a SRA domain and two RING domains [45, 46]. From in vitro assays, it has been shown that VIM1 is a methylcytosine-binding protein [45] that preferentially binds hemi-methylated CG sites [47]. In the vim1 vim2 vim3 triple mutant, a global loss of DNA methylation in the CG sequence context that strongly resembled the methylation profile in met1 mutants was observed [48], suggesting that these VIM proteins may function upstream of MET1. Introduction of a wild-type MET1 into the met1 mutant restored CG methylation at transposons. In contrast, DNA methylation at gene bodies could not be restored when MET1 was introduced back into met1 mutants [48]. Since non-CG methylation maintained by other methyltransferases is preserved at transposons but not gene bodies in the met1 mutant, these different behaviors of MET1 target loci during re-methylation suggested that maintenance of DNA methylation by MET1 depends on pre-existing DNA methylation. Recently, Ruscio et al. [49] reported that in human cells, a non-coding transcript generated from the gene locus CEBPA associated with DNMT1 and prevented DNA methylation at this locus. Genome-wide profiling of DNMT1-associated RNAs revealed that the DNMT1-RNA interaction occurred at thousands of loci [49]. It remains to be tested whether Arabidopsis MET1 is regulated by non-coding RNAs in a similar manner.

The maintenance of CHG methylation mostly depends on a plant-specific DNA methyltransferase named CHROMOMETHYLASE 3 (CMT3), which is guided by the dimethylated histone H3K9 (H3K9me2) mark [50, 51]. The H3K9me2 modification in the Arabidopsis genome is distributed at loci of transposable elements and other repeats of mainly pericentromeric regions as well as some euchromatic regions [52]. The maintenance of histone H3K9me2 is mediated by the histone methyltransferase KRYPTONITE (KYP)/SUVH4 and its homologs SU(VAR)3-9 HOMOLOG 5 (SUVH5) and SUVH6 in Arabidopsis. In the kyp single mutant or the kyp suvh5 suvh6 triple mutant, not only was the H3K9me2 level reduced [53], DNA methylation in the CHG sequence context at those loci was also decreased [48]. Du et al. [50] showed that CMT3 could bind to the methylated histone H3 tails through its Chromo and BAH domains and therefore be recruited to the H3K9me2 sites. Interestingly, the histone methyltransferase KYP could preferentially bind to methylated cytosines through its SRA domain, and loss of DNA methylation also resulted in decreased H3K9me2 accumulation [54]. These lines of evidence suggest that a self-reinforcing loop exists between DNA methylation and histone H3K9me2 to silence transposable elements and repeats. Since the SRA domain of the histone H3K9 methyltransferases binds to methylated cytosines in all sequence contexts [54], a histone demethylase INCREASED IN BONSAI METHYLATION 1 (IBM1) is required to prevent ectopic H3K9me2 and DNA methylation at genes [48, 55, 56]. Besides the CMT3/SUVH pathway, the DNA methyltransferase DRM2 also contributes to the maintenance of DNA methylation in the CHG sequence context at some loci through the RdDM pathway [48]. So far, no RNA was found to directly function in the CMT3/SUVH pathway. However, by investigating a natural inverted repeat locus formed by two PHOSPHORIBOSYLANTHRANILATE ISOMERASE (PAI) genes, Enke et al. [57] found that although non-CG DNA methylation at this locus was only dependent on the CMT3/SUVH pathway and was independent of DRM2, the maintenance of non-CG DNA methylation and H3K9me2 was compromised in the dcl2 dcl3 dcl4 triple mutant, indicating that small RNAs function in the maintenance of non-CG methylation in a DRM2-independent mechanism and prevent the loss of H3K9me2 during Pol II transcription. Further mechanistic studies will be required to provide insights to the role of small RNAs in this process.

Cytosine methylation in the CHH sequence context is asymmetric and methylation at a CHH locus will be lost in one daughter DNA molecule after replication. The previously introduced RdDM pathway is responsible for the maintenance of CHH methylation at target loci, mainly small TEs at euchromatic arms or the edges of long TEs. The histone remodeler in the RdDM pathway, DRD1, may have lower efficiency in remodeling nucleosomes at highly condensed heterochromatic regions [58] and therefore these regions cannot be accessed by the RdDM machinery. At those heterochromatic loci, another SWI2/SNF2 type chromatin remodeling protein DECREASED DNA METHYLATION 1 (DDM1) functions in the remodeling of nucleosomes, allowing methyltransferases to access the DNA [58]. At these loci, the maintenance of DNA methylation in CG and CHG sequence contexts requires MET1 and CMT3 respectively, and CMT2, a CMT3 homolog, mediates the maintenance of cytosine methylation in the CHH sequence context [58].

Although plants have specialized pathways to maintain DNA methylation in different sequence contexts, genome methylation profiles in various methylation pathway mutants showed that crosstalks exist between different pathways in maintaining DNA methylation in different cytosine sequence contexts [48]. In met1 mutants, loss of both CG and non-CG methylation was observed at some loci, suggesting that non-CG methylation at these loci may depend on CG methylation. Mutations in the CMT3/SUVH pathway genes such as CMT3, KYP and SUVH5/6 also lead to loss of CHH methylation at some loci, while the mechanism of KYP- and SUVH5/6-mediated CHH methylation maintenance at some of these loci may be independent of CMT3 [48].

DNA Demethylation

The methylation status of DNA is dynamically regulated. At certain developmental stages or in certain cell types, DNA demethylation occurs, either through passive loss of DNA methylation after DNA replication, or active DNA demethylation by DNA glycosylases.

In somatic cells, a DNA glycosylase named REPRESSOR OF SILENCING 1 (ROS1) together with its homologs DEMETER-LIKE 2(DML2) and DML3 actively remove DNA methylation to prevent hyper-methylation at their target loci [59, 60]. So far, only a few proteins involved in demethylation have been identified in Arabidopsis. A histone acetyltransferase named REPRESSOR OF SILENCING 4 (ROS4) is known to act at a subset of ROS1 target loci [61]. In addition to the N-Acyltransferase domain, ROS4 also contains a MBD domain that could bind to methylated cytosines, and a PHD finger domain that could bind to histone proteins without H3K4 di- or trimethylation [61]. Thus ROS4 may recognize chromatin loci with certain features and provide a favorable environment for DNA glycosylases by changing histone modifications. ROS1 and other glycosylases demethylate DNA by excising methylated cytosines through their glycosylase/lyase activity [60, 62]. Following base excision, the 3′ phosphate generated by ROS1 is removed by the DNA phosphatase ZDP [63] and the gap is filled with an unmethylated cytosine by the DNA repair system [64]. The Arabidopsis homolog of animal X-RAY REPAIR CROSS COMPLEMENTING 1 (XRCC1) –ATXRCC1 may function in stimulating the activity of ROS1 and ZDP [65]. An RNA-binding protein REPRESSOR OF SILENCING 3 (ROS3) is also involved in DNA demethylation [66]. ROS3 colocalizes with ROS1 in vivo and may guide ROS1 to its target loci.

Demethylation of DNA also takes place during the development of reproductive cells. During male gametogenesis, DDM1 expression is down-regulated in the vegetative nuclei, resulting in loss of DNA methylation and activation of transposable elements [67]. 21-nt siRNAs generated from activated TEs are thought to move into the sperm nuclei to reinforce the DNA methylation at target loci [67]. Thus demethylation in the vegetative nuclei ensures trans-generational silencing of transposable elements. On the other hand, MET1 is down-regulated during female gametogenesis [68], causing passive DNA demethylation through cell cycles. In addition, another DNA glycosylase DEMETER (DME), which is homologous to ROS1, actively demethylates DNA in the central cell [69–71]. As a result, the CG methylation level is decreased in the central cell and the eventual endosperm cells. It has been suggested that the activated transposable elements in the endosperm also give rise to siRNAs that move into the embryo where they reinforce the silencing status of their homologous sequences [72, 73]. However, the proposed movement of siRNAs from non-germ line cells to germ-line cells has yet to be shown experimentally.

Transgenerational inheritance and variation of DNA methylation

The patterns of DNA methylation are not only maintained through mitosis, but also transmitted to the next generation. Studies on the epigenetic inheritance of DNA methylation using soybean recombinant inbred lines (RILs) [8] and maize RILs [74] showed that most differentially methylated regions between the parents of the RILs behaved like genetic polymorphisms and cosegregated with the genetic background. Yet, unusual epigenetic inheritance such as paramutation may also be related to DNA methylation. Paramutation refers to the interaction between two alleles in which one allele is able to induce a heritable change in the other. A classic case of paramutation is the interaction of the two alleles B-I and B′ at the maize b1 locus. When two plants carrying B-I and B′ respectively were crossed together, the highly expressed B-I allele would be converted into the silenced B′ allele [75]. A region containing seven tandem repeats upstream of b1 is required for paramutation [76] and the methylation status of this regulatory region is different between B-I and B′ [77]. Furthermore, MEDIATOR OF PARAMUTATION 1 (MOP1) encoding an RNA-dependent RNA polymerase was shown to function in paramutation [78], suggesting that non-coding RNA is involved. It has been proposed that RdDM is involved in the conversion of highly expressed alleles into silenced alleles in paramutation [77].

Like genetic variations, spontaneous changes in DNA methylation also happen during genome replication. In recent years, studies on the methylomes of Arabidopsis progenies from a common ancestor that was propagated for 30 generations revealed that although the overall DNA methylation patterns were well maintained over generations, a lot of epigenetic variations in DNA methylation were observed among descendent lines [79, 80]. Based on differentially methylated regions among different lines, protein-coding genes showed a higher frequency of epigenetic change compared to transposable elements that are targets of RdDM [79, 80]. Some DNA methylation changes at protein-coding genes created epialleles and changed gene expression levels [79]. Thus epigenetic variations also contribute to phenotypic diversity and adaptation.

Conclusions and future perspectives

DNA methylation is an important type of epigenetic modification in eukaryotes to maintain genome integrity, silence the transcription of exogenous DNA and regulate endogenous gene expression. Although plants and animals show similarities in the maintenance of CG methylation, plants have evolved distinct pathways in the maintenance of DNA methylation in CHG and CHH sequence contexts. The establishment of DNA methylation requires non-coding RNAs, especially small RNAs. It is noteworthy that piwi-interacting RNAs (piRNAs), a class of small RNAs found in animals, have been implicated in guiding DNA methylation in mice [81, 82] although the underlying mechanisms are unknown. It is possible that piRNAs cause DNA methylation in a similar manner as 24-nt siRNAs in plants, for which the underlying mechanism is well established. Furthermore, active DNA demethylation by glycosylases prevents ectopic methylation in the genome. Through the regulation of both DNA methylation and demethylation pathways, the plant genome maintains proper methylation levels and patterns through cell divisions and responds correctly to developmental or environmental signals.

During DNA methylation or demethylation, DNA methyltransferses or demethylases are precisely recruited to their target loci, due to the guidance from other components in their pathways such as modified histone proteins and non-coding RNAs. Yet many missing links remain to be established to fully understand how proteins in the DNA methylation pathways are accurately guided to and restricted within their target loci without disturbing neighboring regions. For example, what is the signal that triggers de novo methylation at some transgenes; and how methylcytosine-binding proteins distinguish RdDM target loci and CG methylation at gene bodies. In addition, the downstream events that affect transcription after DNA methylation are unknown. To further understand the molecular mechanisms underlying DNA methylation in plants, more factors need to be identified. Although many genes were identified through forward genetic screens using DNA methylation sensitive reporter lines in the last decade, this approach limited the discovery of genes that are functionally redundant or not stringently required in DNA methylation. Other methods such as biochemical or proteomic approaches may lead to the discovery more proteins in DNA methylation pathways.

The application of high-throughput sequencing in DNA methylation studies also greatly contributed to our knowledge of DNA methylation in the genomic scale. Future applications of some newly developed high-throughput sequencing approaches such as RNA-immunoprecipitation sequencing (RIP-seq) [83] or Crosslinking immunoprecipitation sequencing (CLIP-seq) [84] to profile RNAs associated with heterochromatin or DNA methylation related proteins would provide new insights into the role of non-coding RNAs in DNA methylation. Although the DNA methylome has been profiled in plants, however, the mixing of different cell types in methylome studies may mask some methylation patterns in specific cell types that are underrepresented in the samples. Future improvements of cell type-specific DNA isolation or methylome sequencing from a single cell or a few cells would help understand the spatial and temporal dynamics of DNA methylation in plants.