Chromatin remodeling of histone H3 variants by DDM1 underlies epigenetic inheritance of DNA methylation

Abstract

Nucleosomes block access to DNA methyltransferase, unless they are remodeled by DECREASE in DNA METHYLATION 1 (DDM1^LSH/HELLS), a Snf2-like master regulator of epigenetic inheritance. We show that DDM1 promotes replacement of histone variant H3.3 by H3.1. In ddm1 mutants, DNA methylation is partly restored by loss of the H3.3 chaperone HIRA, while the H3.1 chaperone CAF-1 becomes essential. The single-particle cryo-EM structure at 3.2 Å of DDM1 with a variant nucleosome reveals engagement with histone H3.3 near residues required for assembly and with the unmodified H4 tail. An N-terminal autoinhibitory domain inhibits activity, while a disulfide bond in the helicase domain supports activity. DDM1 co-localizes with H3.1 and H3.3 during the cell cycle, and with the DNA methyltransferase MET1^Dnmt1, but is blocked by H4K16 acetylation. The male germline H3.3 variant MGH3/HTR10 is resistant to remodeling by DDM1 and acts as a placeholder nucleosome in sperm cells for epigenetic inheritance.

Keywords: DNA methylation; DNA replication; chromatin remodeling; epigenetic inheritance; heterochromatin; histone H3.1; histone H3.3; histone MGH3/HTR10; histone modification; transposable elements.

Figure 1.. Replacement of histone H3.1 by H3.3 in ddm1 mutants.

(A) H3.1(HTR3)-GFP and H3.3(HTR5)-RFP localization in Arabidopsis root tips of wild-type (WT), ddm1, and met1. (B) Ectopic chromocenter localization of H3.3(HTR5)-RFP in ddm1 as compared to WT and met1. Scale bars indicate 2 μm. (C) Male Germline-specific Histone H3.3 variant MGH3-GFP localization in sperm nuclei of Arabidopsis pollen grains from WT and ddm1/+ plants. DAPI staining was used to visualize vegetative (VN) and sperm nuclei (SN). Mis-localization to the nuclear periphery was observed in pollen from ddm1/+. Scale bars indicate 2 μm. (D) Distribution of ChIP-seq marks in WT along chromosome 5, showing preferential localization of H3.3(HTR5) in leaf tissue and MGH3 in pollen on chromosome arms, and H3K27me1 and DDM1 on pericentromeric heterochromatin. The values correspond to the log2 fold change of IP/H3 for H3.3(HTR5) and H3K27me1, and IP/Input for DDM1 and MGH3, normalized in counts per million. Signal tracks were averaged in 50kb windows with negative log2 values shown in grey. (E) Distribution of the log2 ratio of the ChIP-seq coverage between ddm1 and WT, showing an increase in H3.3 and MGH3 in peri-centromeric regions, coupled with a loss of H3K27me1 and DDM1. MGH3 IP was performed on pollen grains from a heterozygote ddm1/+ plant (as in C). In ddm1-2 mutants, DDM1 protein is present at reduced levels (Figure S1E).

Figure 2.. Genetic interactions between ddm1 and histone H3 variants and chaperones impact DNA methylation.

fas2 and hira are mutants in H3.1 (CAF-1), and H3.3 (HIRA) chaperones, respectively. ATRX is a chromatin remodeler required for H3.3 deposition. (A) Siliques of wild-type (WT) and ddm1/+ fas2 plants. Red arrows indicate nonviable seeds (synthetic lethality). (B) F2 ddm1 htr4 htr5 htr8/+ with reduced H3.3 has severe growth phenotypes compared to htr4 htr5 htr8/+. (C) F2 ddm1 hira double mutants from ddm1 and hira parents, compared with WT, hira and ddm1 siblings. (D) F2 ddm1 atrx double mutants from ddm1 and atrx parents compared with WT and ddm1 siblings. ddm1 hira and ddm1 atrx were phenotypically indistinguishable from ddm1 siblings but ddm1 hira were more severe. (E) DNA methylation levels in CG, CHG and CHH contexts in ddm1 and hira mutants on the left, and ddm1 and atrx mutants on the right, determined by whole genome bisulfite sequencing. The DNA methylation levels range from 0 to 100% and are substantially increased in ddm1 hira as compared to ddm1. atrx mutants lose some methylation and fail to rescue methylation loss in ddm1. Metaplots calculated from all 31,189 transposable elements annotated in TAIR10. (F) Levels of H3.3 in WT and ddm1 Chip-seq at differentially methylated regions (DMRs) between ddm1 and ddm1 hira (hyper-methylated in ddm1 hira). The number of DMRs (n) in the different cytosine nucleotide contexts are noted. H3.3 is statistically enriched in ddm1 compared to WT at these DMRs, but not in random regions (**** P<0.0001, ns not significant, t-test). See Table S1 for the list of all DMRs. (G) Representative loci that re-gain DNA methylation in ddm1 hira as compared to ddm1. Ectopic H3.3 in ddm1 is shown above (red track).

Figure 3.. Structural basis of DDM1-nucleosome interactions.

(A) Protein domain schematic of DDM1. Residue numbers indicate the boundaries of DDM1 and its domains: the N-terminal DEXD ATPase domain and helicase superfamily C-terminal domain (HELICc). Dashed lines represent the coverage of the DDM1 molecular model. (B) Overview of the molecular structure of the DDM1-nucleosome complex as determined by cryo-EM. DDM1 domains, corresponding to the two lobes, are labeled on the side view. (C) DDM1-histone interactions. The experimental density map of the complex shows that DDM1 interacts with histones H3.3 (green) and H4 (yellow). For the inset, a cartoon representation with partially transparent cryoEM map colored by domains is shown. Amino acids along the DDM1-histone interface (6 Å cutoff) are displayed as sticks, and include T80 and D81 of histone H3.

Figure 4.. Structure and function of the Helicase C terminal and ATPase lobes.

(A) A cartoon view of the DNA distortion caused by DDM1 binding. The DNA backbone of the DDM1-nucleosome model (tan) was aligned to a naked nucleosome DNA backbone (grey, PDB code 1KX5), showing the distortion of DNA where DDM1 is bound to the nucleosome, as well as distortion on the other gyre. A transparent surface model of DDM1 is shown for clarity. The green arrow shows the distortion (opening) of the gyre position caused by DDM1 binding to the nucleosome, and the magenta arrow represents the distortion of the DNA backbone. (B) A view of the disulfide bond formed between C615 and C634, connecting two regions in the HELICc domain of DDM1. The molecular model is shown as ribbons, cryo-EM density is shown as a gray volume. The first mutation of ddm1 to be isolated, ddm1-1, substitutes C615 for Y and has a strong DNA methylation defect. (C) Highly conserved glutamine residues Q625 (red) and Q629 (orange) project between the lobes and are mutated to arginine in human HELLS (identified in ICF syndrome proband E) and in Arabidopsis ddm1-9 (where it results in hypomethylation), respectively. The arginine residues are predicted to contact phosphates in the DNA minor groove and are also highlighted in (A). (D) A surface representation of the DDM1-nucleosome complex, highlighting the T80V mutation found in the male germline specific histone H3.3 MGH3 (red). T80 directly contacts DDM1 (Fig. 3C inset). (E) A surface representation of the DDM1-nucleosome complex, showing the surface exposed D382E mutation that results in a hypomethylation phenotype in ddm1-14. (F) A zoomed in view of the refined cryoEM density map, showing the N-terminal tail of histone H4 extending into the density observed for the DEXD ATPase domain of DDM1. Color coding of histone variants and DDM1 as in Figure 3.

Figure 5.. An N-terminal autoinhibitory domain regulates H4 peptide-binding, ATPase and nucleosome remodeling activities of DDM1 with histone variants.

(A) Disorder predictions for DDM1 were calculated with PrDOS95. The red line indicates the threshold corresponding to a false positive rate of 5%. The autoinhibitory domain (1–132 residues) AutoN is indicated in the diagram along with DEXD ATPase and HELICc domains. (B) Binding affinities between DDM1 and H4 peptides shown as the fraction bound at peptide concentrations measured by microscale thermophoresis (MST). KD values were estimated by fitting algorithms provided by the supplier (Methods). Binding was not detected (N.D.) for H4K5K8K12K16Ac (quadruple acetylation), but was detected for unacetylated H4, H4K20me1, H4K20me2, and H4K20me3 peptides. Truncation of AutoN in the DDM1(Δ1–132) enzyme resulted in higher binding affinity consistent with AutoN competing with the H4 tail. (C) DNA-dependent ATPase activities for recombinant DDM1, DDM1(Δ1–132) and DDM1 C615S. ATPase activities are given as luminescence with relative light units (RLU). The X-axis indicates ATPase reaction time. Error bars represent standard deviations from two independent replicates. The relative rate enhancement for ATPase activity between DDM1(Δ1–132) and DDM1 is 6.4x. DDM1C615S has a further reduction of 2.3x relative to DDM1. (D) Nucleosome remodeling assays with 0N60 mono-nucleosomes (147bp Widom 601 DNA plus 60bp linker) were performed with octamers of H2B, H4 and combinations of H3 and H2A variants as shown. Center-positioned nucleosomes (arrows) were incubated with DDM1, DDM1 C615S, or DDM1(Δ1–132) at t=0 mins, and then remodeled upon addition of ATP by putative sliding (slower migration) and unwrapping (faster migration) activities. (E) Quantification of remodeling activities for DDM1, DDM1 C615S and DDM1(Δ1–132) are shown below each assay series as the fraction of intact nucleosomes (arrows) remaining at each timepoint relative to t=0. Error bars indicate standard deviation, and are too small to be resolved for H3.3 H2A.W. DDM1 C615S had little or no remodeling activity and was used as a control.

Figure 6.. DDM1 remodels H3.3 and H3.1 during the cell cycle at differentially methylated targets of DDM1.

(A) Subnuclear localization of DDM1-mCherry in root tip cells as compared to H3.1(HTR13)-CFP during presumptive late S-phase (marked by H3.1-labeled chromocenters). (B) Subnuclear localization of DDM1-GFP as compared to H3.3(HTR5-RFP) during interphase. The scale bar indicates 2 μm. See also Supplemental Videos S1 and S2. (C) Conserved amino acids of the ATP binding sites for DDM1 and orthologs, LSH (mouse) and HELLS (human). K233Q Walker A mutation disrupts ATP binding and causes enhanced chromocenter localization of DDM1-mCherry fusion in a WT background. Scale bar indicates 2 μm. (D) Western blot of DDM1-mCherry (DDM1-mCh) for the wild-type (WT) and mutant form (K233Q) from soluble (S) and chromatin/pellet (P) fractions indicates failure to release catalytic mutant from chromatin. H3 was used as loading control. Non-transgenic WT was used as negative control. (E) Comparisons of DDM1, H3K27me1, H4K16ac and H3.3(HTR5) chromatin association by ChIP-seq in WT and ddm1 leaves, as well as MGH3 in pollen from WT and ddm1/+ plants. Stable and revertant differentially methylated regions (DMRs) lost DNA methylation in ddm1 mutants, and stable DMRs never regained methylation when DDM1 was reintroduced. Thus, DMRs represent epigenetic targets of DDM1. Heatmaps and metaplots were generated using DeepTools, where each region was scaled to 2kb with 5kb upstream and 5kb downstream with a binsize of 10bp, and sorted based on DDM1 levels in WT. Metaplots above each heatmap show the mean value for each region. Random regions are revertant DMRs reshuffled randomly in the genome, whereas random genes correspond to the same number of protein coding genes selected at random (see Table S1). DDM1 and H3.1 (H3K27me1) are specifically enriched at DDM1 targets (DMRs), while H4K16Ac, H3.3 and MGH3 are specifically depleted, except in ddm1 mutants. Similar analysis was performed on all transposable elements for comparison (Figure S5).

Figure 7.. A model for epigenetic inheritance of unmethylated transposons.

Active transposable elements (TEs) from male and female gametes are unmethylated and comprise MGH3 H2A and H3.3 H2A nucleosomes, respectively. In the zygote, H3.3 H2A nucleosomes are remodeled by DDM1 before replication, allowing deposition of H3.1 in S phase by CAF-1. Unwrapping of H3.1 H2A by DDM1 permits access to the methyltransferase MET1 allowing CG methylation. Subsequent incorporation of H2A.W stabilizes the nucleosome, possibly promoting H3K9 di-methylation (not shown) and CHG methylation by the chromomethylase CMT3. MGH3 H2A nucleosomes, inherited from pollen, are resistant to remodeling after fertilization. They are eventually replaced in the embryo by H3.3 H2A nucleosomes, but acetylation of histone H4 and other marks of active euchromatin prevent recognition by DDM1.