Genome-wide negative feedback drives transgenerational DNA methylation dynamics in Arabidopsis

Abstract

Epigenetic variations of phenotypes, especially those associated with DNA methylation, are often inherited over multiple generations in plants. The active and inactive chromatin states are heritable and can be maintained or even be amplified by positive feedback in a transgenerational manner. However, mechanisms controlling the transgenerational DNA methylation dynamics are largely unknown. As an approach to understand the transgenerational dynamics, we examined long-term effect of impaired DNA methylation in Arabidopsis mutants of the chromatin remodeler gene DDM1 (Decrease in DNA Methylation 1) through whole genome DNA methylation sequencing. The ddm1 mutation induces a drastic decrease in DNA methylation of transposable elements (TEs) and repeats in the initial generation, while also inducing ectopic DNA methylation at hundreds of loci. Unexpectedly, this ectopic methylation can only be seen after repeated self-pollination. The ectopic cytosine methylation is found primarily in the non-CG context and starts from 3' regions within transcription units and spreads upstream. Remarkably, when chromosomes with reduced DNA methylation were introduced from a ddm1 mutant into a DDM1 wild-type background, the ddm1-derived chromosomes also induced analogous de novo accumulation of DNA methylation in trans. These results lead us to propose a model to explain the transgenerational DNA methylation redistribution by genome-wide negative feedback. The global negative feedback, together with local positive feedback, would ensure robust and balanced differentiation of chromatin states within the genome.

Fig 1. DNA methylation in ddm1 mutants before and after repeated self-pollination.

DNA methylation patterns of WT, 1G ddm1, and 9G ddm1 mutants for cellular genes (A-C) or genes within transposable elements (transposable element genes, or TEGs, D-F). “WT” is a DDM1/DDM1 plant segregating as a sibling of the 1G ddm1/ddm1 plants. The black bars in the bottom represent transcribed regions. A chromosome-wide view of DNA methylation is also shown in S2 Fig.

Fig 2. Change of CG methylation during self-pollination of ddm1 mutants.

(A) CG methylation level compared for each transcription unit. Each dot represents the DNA methylation level in a gene (black dot) or a transposable element gene (TEG, red dot). The top half shows effects in four different 1G ddm1 plants, while the bottom half shows effects in four different 9G ddm1 plants. Each of the 9G plants was originated from independent self-pollinations. Comparison of the 9G ddm1 plants to independently self-pollinated 9G DDM1 plants (S1 Fig) is shown in S3 Fig. “WT” is a DDM1/DDM1 plant segregating as a sibling of the 1G ddm1/ddm1 plants. (B, C) Genome browser views of loci with CG methylation reduced in 9G ddm1 using the Integrated Genome Browser [74]. FWA locus (B) and AT2G04350 locus (C) are shown. The FWA gene has dense CG methylation around the 5’ end, which is lost during self-pollination of the ddm1 mutant. (D) H3K9me2 level of differently hypo-methylated regions (hypo-DMRs) in CG context. Left (1G - WT): Distribution of 119,883 DMRs between WT and 1G ddm1 mutant. Center (9G - specific): Distribution of 25,861 DMRs between WT and 9G ddm1, excluding DMRs between WT and 1G ddm1. Distribution of 100,000 randomly chosen 100 bp regions is also shown as a control (right). H3K9me2 level of wild type is shown by reads per million (RPM) in ChIP-seq data obtained from GEO (GSE28398 [72]). (E) Change in CG methylation level in 1G ddm1 (left) and 9G ddm1 (right) from wild type, plotted against enrichment of H3K9me2 in wild type (data from Inagaki et al. 2010).

Fig 3. Change of non-CG methylation during self-pollination of ddm1 mutants.

(A, B) Effects of 1G and 9G ddm1 mutation on CHG methylation (A) and CHH methylation (B). The format is as shown for CG sites in Fig 2A. Comparison of the 9G ddm1 plants to independently self-pollinated 9G DDM1 plants is shown in S3 Fig. (C) The number of genes that gained non-CG methylation in ddm1 mutant (methylation level < 0.1 in WT and ≥ 0.1 in ddm1). Results for the four 1G and four 9G of ddm1 mutants are shown for CHG and CHH sites. (D) Coordinated hypermethylation of CG, CHG and CHH sites. “Genes CHG-hypermethylated in 9G ddm1” are those with methylation level < 0.1 in 1G ddm1 and ≥ 0.1 in 9G ddm1. DNA methylation levels for three contexts are shown for WT, 1G ddm1, and 9G ddm1. On the right, total genes are shown as controls. Although CHG hypermethylated genes tend to have more CG methylation in wild type, the body methylation is not an absolute requirement; even genes without CG methylation occasionally non-CG hypermethylated in 9G ddm1 (S4 Fig). Pattern of CG methylation change from 1G ddm1 to 9G ddm1 is further characterized in S5 Fig.

Fig 4. BONSAI hypermethylation in self-pollinated ddm1 mutants is associated with H3K9 methylation.

(A) Genome browser views of CHH and CHG methylation in AT1G73177 (BONSAI) locus. (B) H2K9me detected by chromatin immunoprecipitation (IP). “input” is the sample before IP; “mock” denotes samples after IP procedure without antibody. H3mK9me1 and H3K9me2 are samples after IP with the respective antibodies. Amplified regions around the BONSAI locus are indicated in (A). LINE and ACT2 are used as positive and negative controls, respectively. 11G plants are the ddm1 mutants in the 11th generation. 11G #1 and 11G #3 samples are prepared from progenies of direct sibling of 9G ddm1 #1 and #3 plants (shown in A), respectively. Results for other loci are shown in S6 and S7 Figs. Although the BONSAI locus accumulated both CHG and CHH methylation, some of the CHG hypermethylated loci have less CHH methylation than others (S6A Fig). In our preliminary analyses, H3K9me1 is more prevalent in those loci than H3K9me2 (S6 and S7 Figs).

Fig 5. Spread of non-CG methylation in self-pollinated ddm1 mutants.

(A-E) Genome browser views of loci with non-CG methylation in the 9G ddm1 plants. AT5G16880 (A-B), AT3G06480 (C-D), and AT4G07518 loci (E) are shown for CHG (A, C) and CHH (B, D, E) contexts. Direction of transcription is shown by an arrow in A-D. (F) Histogram of correlation coefficient between the CHG methylation level and the relative centroid position of CHG methylation within the DMR. The centroid position was determined by averaging relative position of the methylated cytosine weighed with the methylation level for each residue. The coefficient was calculated among the four 9G ddm1 plants in each conDMR for CHG methylation between 9G and 1G ddm1 (details in Materials and Methods section) overlapping with genes. The coefficient becomes negative when the centroid moves to the 5’ regions as the average level of CHG methylation in the conDMR increase. A large proportion of the contiguous DMRs with the coefficient near -1 reflects spread of CHG methylation from 3’ to 5’ regions as the CHG methylation levels increase.

Fig 6. Hypermethylated regions in ddm1 and ibm1 mutants.

(A) Increase of CHG methylation in 1G and 3G ibm1 mutants. Genes hypermethylated in 1G ibm1 (CHG methylation level < 0.1 in WT and ≥ 0.1 in 1G ibm1) are shown (right) with total genes (left). Profiles for multiple 1G and 3G ibm1 mutant plants are shown in S15 Fig. (B) Comparison of regions CHG hypermethylated in ibm1 and 9G ddm1. DMRs between 9G and 1G ddm1 (blue), between 1G ibm1 and WT (orange), and between 3G ibm1 and WT (red) are shown. Heat map of CHG methylation for these DMRs are shown in S16B Fig. (C) DNA methylation profile for the genes CHG hyper-methylated in 9G ddm1 (shown in Fig 3D). The top and bottom half represent genes and TEGs, respectively. In these regions, CHH methylation also increased in 9G ddm1.

Fig 7. Effects of disrupted heterochromatin in the DDM1 wild type background examined by IP.

(A, C, E) Changes in local DNA methylation plotted against the global level of DNA hypomethylation in 123 epigenetic recombinant inbred lines (epiRILs). Each dot represents the value for one line. Three loci, AT1G73177 (BONSAI) (A), AT5G52480 (C), and AT5G35510 (E) are shown. Results with four other loci are shown in S17 Fig, with values for the F₀ ddm1 and WT parents. (B, D, F) WT (light green) / ddm1 (dark blue) haplotype for epiRILs that showed increase of cytosine methylation for each locus (numbered 1–6 for each locus, the line names can be different among the panels). In each panel, the chromosome including the target locus (arrowhead) is shown. Haplotypes of all five chromosomes are shown in S18–S23 Figs. The filled circles indicate centromere positions. The haplotypes are predicted from stably hypomethylated markers [46]. The regions not covered by any markers are indicated in gray. Names of epiRILs numbered 1–6 in each panel are in Materials and Methods. Data of epiRILs were obtained from GEO (GSE37284 [46]).

Fig 8. Effects of disrupted heterochromatin in the DDM1 wild type background examined at single base resolution.

(A) Methylation level was compared for each transcription unit in CG, CHG, and CHH contexts. The format is as shown in Fig 2A. A globally hypomethylated epiRIL (epiRIL98: plant #3 in Fig 7A and 7B and plant #2 in Fig 7E and 7F) and two epiRILs with lower level of hypomethylation (epiRIL260 and epiRIL480) are shown. Global hypomethylation indexes of epiRIL98, epiRIL260, and epiRIL480 are 0.38, 0.04, and 0.09, respectively. “WT” data are from the parental wild-type Col plant used to generate the epiRILs. (B) CHG methylation levels in the genes that were not methylated in WT but methylated in epiRIL98 (methylation level < 0.1 in WT and ≥ 0.1 in epiRIL98: n = 232). For these transcription units, distributions of the methylation levels were compared among the parental WT, the parental 4G ddm1 plant, and the epiRIL98. (C-D) Ectopic CHG methylation in epiRIL98 compared to wild type. Each gene was assigned to the inferred haplotypes in epiRIL98: WT-like (C) or ddm1-like (D). The ectopic methylation could be detected in genes of the WT-like haplotype. Examples of such genes are shown in S25 Fig.

Fig 9. A model for the transgenerational heterochromatin redistribution.

The cylinder indicates a nucleosome. Red dots above the nucleosome indicate methylation of H3K9. Red and blue lines indicate DNA with and without non-CG methylation, respectively. The CMTs are non-CG methylases, such as CMT3 and CMT2 [10,11]. SUVHs are H3K9 methylases, such as SUVH4/KYP, SUVH5 and SUVH6 [75]. In both WT and ddm1 mutant plants, the histone demethylase IBM1 removes H3K9me from transcribed genes.