A Role for RNAi in the Selective Correction of DNA Methylation Defects

See allHide authors and affiliations

Science  20 Mar 2009:
Vol. 323, Issue 5921, pp. 1600-1604
DOI: 10.1126/science.1165313


DNA methylation is essential for silencing transposable elements and some genes in higher eukaryotes, which suggests that this modification must be tightly controlled. However, accidental changes in DNA methylation can be transmitted through mitosis (as in cancer) or meiosis, leading to epiallelic variation. We demonstrated the existence of an efficient mechanism that protects against transgenerational loss of DNA methylation in Arabidopsis. Remethylation is specific to the subset of heavily methylated repeats that are targeted by the RNA interference (RNAi) machinery. This process does not spread into flanking regions, is usually progressive over several generations, and faithfully restores wild-type methylation over target sequences in an RNAi-dependent manner. Our findings suggest an important role for RNAi in protecting genomes against long-term epigenetic defects.

Cytosine methylation in plant genomes occurs predominantly at CG sites but also at CHG and CHH sites (where H is A, T, or C). Genetic analysis in Arabidopsis has uncovered an interplay between “maintenance” and “de novo” DNA methyltransferases (MTases), DNA demethylases, histone-modifying or remodeling enzymes, and RNA interference (RNAi) components (1, 2). Among many mutants identified, those in the MTase gene MET1 and the adenosine triphosphatase chromatin remodeler gene DDM1 lead to the most severe loss (>70%) of DNA methylation overall (36). This loss persists in F1 plants obtained in crosses with the wild type, despite met1 and ddm1 mutations being recessive (36), as well as in subsequent generations, at least at the few loci that have been examined (2, 7). These and other findings have led to the prevalent view that in plants, DNA methylation cannot be restored once it has been severely compromised (2, 7, 8). However, given the stability of DNA methylation of repeat elements across generations (9), mechanisms must exist that protect plant genomes against the irremediable loss of DNA methylation.

To address this issue, we examined the transgenerational stability of ddm1-induced hypomethylation of a large number (n = 56) of transposable elements and other repeats. Methylation was first investigated five generations after segregation of the ddm1 mutation [the DDM1/DDM1 F5 lines were obtained through one backcross and three selfings (BC1 and S3)] (fig. S1 and table S1) by means of McrBC digestion and quantitative polymerase chain reaction (McrBC-QPCR) (10). Of 47 ddm1-hypomethylated sequences examined in the repeat-rich 500-kb-long hetero-chromatic knob on chromosome 4, 21 remained hypomethylated in DDM1/DDM1 F5 lines that had inherited the knob region from the ddm1 parent (Fig. 1A and table S2). However, another 22 sequences had wild-type (WT) methylation levels, and only four showed inconsistent methylation between these lines. The interspersion of hypo- and WT methylation and the consistency of methylation patterns between lines (Fig. 1A and table S2) as well as in more advanced generations (F6 to F9) (figs. S2 and S3) were all indicative of a robust and targeted remethylation process. This was further illustrated by the fact that even when located close (<2.5 kb) to remethylated sequences, nonmethylated controls remained unmethylated (Fig. 1A and table S2). Remethylation was also observed for five of eight ddm1-hypomethylated sequences tested outside of the knob, indicating that this process acts throughout the genome (fig. S3).

Fig. 1.

DNA methylation and expression analyses in WT, ddm1, and progeny lines. (A) Annotation is at the top (red, DNA transposons; green, retro-elements). The ATENSAT1 tandem array (22.5 copies) is depicted as four bracketed red arrowheads. Vertical bars indicate methylation levels (light gray indicates <30% methylation). (Bottom) Remethylated sequences are marked by arrows and nonremethylated sequences by open circles; sequences that were equally methylated in WT and ddm1 are indicated by an equals sign. Inconsistent DNA methylation levels (±30% or more) between the three F5 lines are indicated by a question mark. (B) Average DNA methylation level of 18 nonremethylatable and 18 remethylatable sequences in WT, ddm1, and F1 progeny of reciprocal crosses. The average between WT and ddm1 is also shown. (C) Average DNA methylation of 8 nonremethylatable and 15 remethylatable sequences in WT, ddm1, and two DDM1/DDM1 progeny lines followed through four selfings (F2 to F5). (D) and (E) reverse transcription PCR (RT-PCR) analysis of two nonremethylatable and three remethylatable transposable elements (TEs). Results are expressed as a percentage of expression relative to controls (10).

To determine when remethylation occurs and whether it might be parent-of-origin–specific, we performed reciprocal crosses between wild type and ddm1 plants. Irrespective of the direction of the cross, methylation levels were constant in the different F1 plants and were systematically intermediate between the wild type and ddm1, whether “remethylatable” or “nonremethylatable” according to the initial analysis of F5 lines (Fig. 1B, fig. S4, and table S2). These results are in agreement with global measurements of cytosine methylation (11) and indicate that remethylation does not occur detectably in the F1 progeny of ddm1–wild type crosses. F1 plants were selfed or reciprocally backcrossed with the wild type, and F2 individuals of DDM1/DDM1 genotype were propagated for several generations. Eight of the 21 nonremethylatable sequences originally identified (Fig. 1A) were tested, and none had regained DNA methylation in these new F2 to F5 lines (Fig. 1C and table S2). On the other hand, 12 of 15 remethylatable sequences had increased methylation by the F2 and three by the F3 (Fig. 1C and table S2). With three exceptions, remethylation was progressive and reached WT levels after one to three additional generations (Fig. 1C and table S2). Remethylation efficiencies were similar in the progeny of reciprocal backcrosses, ruling out overt parent-specific effects. Furthermore, remethylation did occur when either one or both copies of the knob were derived from the ddm1 parent, indicating that this process differs from paramutation (12) or meiotic transfer of methylation (13), which involves a transfer of epigenetic information between alleles.

Many transposable elements are reactivated in ddm1 (14, 15), and several nonremethylatable and remethylatable sequences were of this kind. However, nonremethylatable elements remained active after the restoration of DDM1 function, whereas remethylatable elements became resilenced (Fig. 1, D and E). This suggests an important function for remethylation in protecting the genome against the deleterious effects of transposable element activity.

Three remethylatable and four nonremethylatable sequences were then analyzed by means of sequencing of bisulfite-treated DNA (10). These two groups of sequences, which had comparable CG, CHG, and CHH compositions (fig. S7A), exhibited similar methylation in WT plants (Fig. 2A and fig. S7B), which is consistent with McrBC-QPCR (Fig. 1, A to C, and table S2) and genome-wide bisulphite sequencing data (fig. S7C) (16). In the DDM1/DDM1 F5 lines, faithful remethylation was observed at CG, CHG, and CHH sites for remethylatable sequences, whereas nonremethylatable sequences retained their ddm1 hypomethylation profiles (Fig. 2A and figs. S5 and S6). Bisulfite sequencing also indicated that the more pronounced loss of methylation in ddm1 over nonremethylatable sequences (Fig. 1, A to C, and table S2) was caused by differences at CHH and CHG but not CG sites (Fig. 2B and figs. S5 and S6). Indeed, although the loss of methylation in ddm1 was severe at CG sites for the two groups of sequences (around 80%), as well as at CHG and CHH sites for nonremethylatable sequences (80 and 95%, respectively), it was less so at CHG sites and only marginal at CHH sites for remethylatable sequences (60 and 25%, respectively) (Fig. 2B and fig. S6). These results revealed a strong dependence of CHH methylation on DDM1, but only at nonremethylatable sequences.

Fig. 2.

Bisulfite sequencing analysis of DNA methylation in WT, ddm1, and progeny lines. (A) The locations of At4g03650 (remethylatable) and At4g03826 (nonremethylatable) are shown at the top. Vertical bars indicate the percentage of methylation for each cytosine, and n is the number of clones sequenced. (B) Percentage of CG, CHG, and CHH methylation present in ddm1 as compared with that in WT. The total number of sites analyzed is shown above each bar. Differences were statistically significant for CHG and CHH sites (two-tailed t test, P < 0.01) but not for CG sites.

Given the involvement of RNAi in CHH methylation (1, 2), small RNA (sRNA) deep-sequencing data available for the wild type (17) were examined and compared with similar data obtained for ddm1 (10). Contrary to nonremethylatable sequences, remethylatable sequences were characterized in WT plants by an abundance of corresponding sRNAs of mostly 24 nucleotides (nt) (Fig. 3, A and B, fig. S8, and table S1), which is consistent with genome-wide data indicating that 24-nt small interfering RNAs (siRNAs) match only a subset of methylated repeats (18). Analysis of the ddm1 data revealed a change as compared with that of the wild type in the relative proportion of the different size classes of sRNAs other than known microRNAs (miRNAs) or trans-acting small interfering RNAs (tasiRNAs) (Fig. 3C). This change was mainly caused by a specific and massive accumulation of 21-nt siRNAs matching ATHILA elements (Fig. 3A, fig. S8, and table S1) (10), which is similar to that observed in met1 mutant plants (18) as well as in tissue culture (19). Nonetheless, 18 of the 27 remethylatable sequences, including most ATHILA sequences, were still characterized by an abundance of matching 24-nt siRNAs (Fig. 3, A and B, fig. S8, and table S1) (10), which is in agreement with previous studies (15, 20). This suggests that the RNAi machinery is responsible for the persistence of CHH methylation specifically at remethylatable sequences in ddm1 and hence in their distinction from nonremethylatable sequences. Consistent with this, transcript levels of genes involved in the 24-nt siRNA production pathway (2123), including DICER-LIKE 3 (DCL3), RNA-DEPENDENT RNA POLYMERASE 2 (RDR2), NUCLEAR RNA POLYMERASE D1A (NRPD1A), and NRPD1B, were unchanged in ddm1 (Fig. 3D). As anticipated from the disappearance of 24-nt siRNAs in rdr2 and dcl2dcl3 mutants (22, 23), methylation was almost completely lost over remethylatable sequences when ddm1 was combined with rdr2 or dcl2dcl3 (Fig. 3E and table S2). Reduction of methylation at nonremethylatable sequences was also observed in ddm1rdr2 or ddm1dcl2dlc3 mutant seedlings (Fig. 3E and table S2), suggesting a low input from the RNAi pathway in the methylation of these sequences. Although global methylation levels were unaffected in rdr2, dcl2dcl3, and nrpd1a&1b mutant seedlings (fig. S9 and table S2), a significant loss of CHH methylation occurred specifically over remethylatable sequences (Fig. 3F and figs. S5 and S6). These results establish a crucial role for RNAi in the distinction between remethylatable and nonremethylatable sequences.

Fig. 3.

Involvement of RNAi in the distinction between remethylatable and nonremethylatable sequences. (A) sRNAs (blue, 20 to 21 nt; green, 22 to 23 nt; red, 24 to 25 nt; pink, other sizes) matching 700–base pair (bp) regions centered on two remethylatable (R) and two nonremethylatable (NR) knob sequences, as probed by use of McrBC-QPCR and bisulfite sequencing (BS) (black bars). (B) Normalized density (10) of 24-nt RNAs for 14 unmethylated (U), 24 nonremethylatable, and 27 remethylatable sequences. Data for WT was derived from (17). (C) Frequency plot of 19- to 26-nt RNAs (minus miRNAs and tasiRNAs). (D) RT-PCR analysis of ARGONAUTE 4 (AGO4), DCL2, DCL3, NRPD1A, NRPD1B, and RDR2 transcripts. Results are expressed as a percentage of expression relative to controls (10). (E) Average DNA methylation of two unmethylated, 15 nonremethylatable, and 20 remethylatable sequences in different genetic backgrounds. (F) Average percentage of methylation in ddm1 relative to WT for three remethylatable and four nonremethylatable sequences (figs. S5 and S6). Differential loss of CHH methylation in ddm1 is statistically significant (two-tailed t test, P < 0.01).

To determine whether RNAi is causal in remethylation, ddm1rdr2 plants were crossed with rdr2 single mutants and the F1 progeny was selfed to obtain F2 and F3 progeny with impaired RNAi but restored DDM1 function. We observed rare and nonprogressive methylation, which affected both remethylatable and nonremethylatable sequences (Fig. 4 and table S2). This mostly sporadic low-level methylation is reminiscent of that observed in mutant plants homozygous for a null allele of the maintenance MTase gene MET1 (8) and suggests that when RNAi is compromised, an inefficient mechanism of compensatory methylation may come into play. This demonstrates an essential function of RNAi in robust and specific remethylation.

Fig. 4.

Involvement of RNAi in faithful remethylation. DNA methylation was measured by means of McrBC-QPCR in F2 and F3 progeny lines with a ddm1-derived knob region. Results are summarized as illustrated.

We have shown that contrary to the prevalent view (2, 7, 8), numerous repeat elements can efficiently regain WT methylation after a severe loss in Arabidopsis. This remethylation is guided by the RNAi machinery, does not spread into nearby sequences, and, in the case of reactivated transposable elements, is associated with their resilencing. Thus, two main types of methylated repeat elements may be distinguished (fig. S10A). Although methylation of some repeat elements depends almost exclusively on maintenance MTases and cannot be regained once compromised, other repeat elements are efficiently targeted in parallel by the RNAi-dependent de novo methylation machinery, which contributes marginally to their overall methylation but is critical for their faithful remethylation. A third, minor group of repeats is targeted solely by the RNAi-dependent de novo methylation machinery (18, 24) and therefore remains unaffected in ddm1 or met1 (fig. S10B).

The progressivity of remethylation parallels that of the de novo RNA-dependent DNA methylation that is frequently observed with transgenes (1, 2). This suggests that remethylation may be most effective for recently inserted transposable elements and that nonremethylatable repeats were likely to have been remethylatable when first inserted. Although the frequency with which DNA methylation can be lost in natural settings is unknown, our discovery of a corrective mechanism reveals an important role for RNAi in protecting the genome against transgenerational epigenetic defects. This mechanism also has potential adaptive and evolutionary implications (25) because it allows for the generation of epialleles with differences in transgenerational stability.

Supporting Online Material

Materials and Methods

Figs. S1 to S10

Tables S1 to S4


References and Notes

View Abstract

Navigate This Article