Targeted reprogramming of H3K27me3 resets epigenetic memory in plant paternal chromatin

Abstract

Epigenetic marks are reprogrammed in the gametes to reset genomic potential in the next generation. In mammals, paternal chromatin is extensively reprogrammed through the global erasure of DNA methylation and the exchange of histones with protamines^1,2. Precisely how the paternal epigenome is reprogrammed in flowering plants has remained unclear since DNA is not demethylated and histones are retained in sperm^3,4. Here, we describe a multi-layered mechanism by which H3K27me3 is globally lost from histone-based sperm chromatin in Arabidopsis. This mechanism involves the silencing of H3K27me3 writers, activity of H3K27me3 erasers and deposition of a sperm-specific histone, H3.10 (ref. ⁵), which we show is immune to lysine 27 methylation. The loss of H3K27me3 facilitates the transcription of genes essential for spermatogenesis and pre-configures sperm with a chromatin state that forecasts gene expression in the next generation. Thus, plants have evolved a specific mechanism to simultaneously differentiate male gametes and reprogram the paternal epigenome.

Extended Data Fig. 1

Dynamics of histone H3.1 and H3.3 during pollen development. Expression of H3.1 (a-d) and H3.3 (e-f) isoforms during pollen development. Histone H3.1 is encoded by five isoforms: HTR1 (a), HTR2 (b), HTR3 (c), HTR9 and HTR13 (d). Histone H3.3 is encoded by three isoforms: HTR4, HTR5 (e) and HTR8 (f). Two pairs of genes (HTR4-HTR5 and HTR9-HTR13) are found in tandem at the same locus so a single reporter for each pair was used to monitor expression. Histone H3.1 (HTR2, HTR3 and HTR13) were detectable in the microspore and sperm precursor but this signal disappeared rapidly before sperm mitosis. No pollen expression was detected for HTR1. Histone H3.3 (HTR5 and HTR8) were detected throughout pollen development but had a much reduced (HTR5) or absent (HTR8) signal in sperm. Arrows indicate expression in the microspore or VN while arrowheads distinguish expression in the sperm lineage. The marker line analysis was repeated twice with independent inflorescences. Scale, 5 μm.

Extended Data Fig. 2

Specificity of anti-H3K27 methylation antibodies used in this study. a, Peptide sequences of H3.1, H3.3 and H3.10 surrounding K27 used for testing antibody specificity. Different forms with no methylation (me0), mono-methylation (me1), di-methylation (me2), and tri-methylation (me3) at K27 were used in all dot blots. b-c, Dot blots with serial dilutions of the different forms of histone H3 peptides described in a. The resulting membranes were probed with (b) α-H3K27me1 from Millipore #07-448 and (c) α-H3K27me3 from Millipore #07-449. Importantly, both α-H3K27 methylation antibodies cross react with the correct methylated form of H3.10 peptides, confirming that a lack of H3K27me3 detection in sperm chromatin (Fig. 1a) or on ectopically expressed H3.10-3xHA (Fig. 2e) is not due to poor antibody affinity. The experiment was repeated twice on two independent blots. e, Representative image of T1 htr4;htr5;htr8 plants expressing either untagged H3.10 under control of an H3.3 promoter (left) or endogenous H3.3 (right). Plants devoid of endogenous H3.3 and expressing only H3.10 and H3.1 (left) were developmentally stunted and completely sterile. This was evident in two independent experiments with individual htr4;htr5;htr8 T1 lines. Raw blots are provided in Source Data Extended data fig. 2.

Extended Data Fig. 3

Epigenomic profiling of Arabidopsis sperm chromatin. a, Pearson correlation matrix of the ChIP-seq datasets generated in this study. Each ChIP-seq replicate is indicated in the matrix, which was performed with three biological replicates; two for H3K27me1 and H3K27me3. b, Distribution of repressive (top panel) and active marks (bottom panel) over Arabidopsis chromosome one. Plotted is the ChIP-seq log₂ enrichment of IP DNA relative input calculated in 10kb bins. Pericentromeric heterochromatin is indicated with grey shading. c, Genome browser view of the sperm ChIP-seq datasets. Coverage is represented as the log₂ ratio of IP DNA relative to input. Coloured and grey shading indicate an enriched or depleted signal, respectively. Genes (light grey) and transposable elements (dark grey) are shown below. d, Distribution of sperm histone marks over transposable elements. Plotted is the ChIP-seq log₂ enrichment relative to input. e, Distribution of sperm histone marks over genes sorted by expression level in sperm. f, Genomic distribution of histone mark peaks in sperm. As expected, H3K27ac and H3K4me3 peaks were mostly enriched over the 5’UTR of genes. H3K27me1 and retained H3K27me3 peaks were mostly enriched over exons, while H3K27me1 peaks were also enriched in intergenic regions. g, Overlap of the retained sperm H3K27me3 peaks with somatic H3K27me3 domains. Statistical analysis is based on a one-sided permutation overlap test (n = 100 permutations) compared with random TAIR10 regions. h, Estimated library complexity curves confirmed a sufficient sequencing depth for two independent biological replicates of sperm H3K27me3. The red curve represents the interpolated and extrapolated increase in complexity (i.e. distinct reads) with increased sequencing depth. The grey shading represents the upper and lower 95% confidence interval of the extrapolation. The dashed grey line represents the final sequencing depth of each sample. i, Plot of the pairwise correlation between sperm H3K27me3 biological replicates, which showed high reproducibility. Pearson’s correlation coefficient is shown.

Extended Data Fig. 4

Dynamics of the Polycomb machinery during sperm development. a-f, Expression of MEA-YFP (a), CFP-CLF (b), SWN-GFP (c), LHP1-YFP (d), EMF2-GFP (e) and FIE-VENUS (f) during pollen development. All markers were absent from sperm at mature pollen stage. FIE had an appreciable signal in the sperm precursor but was excluded from the nucleus. Arrows indicate expression in the microspore or VN while arrowheads distinguish expression in the sperm lineage. Marker line analysis was repeated twice with independent inflorescences. Scale, 5 μm. g, Expression of Arabidopsis PRC2 (top panel) and PRC1 (bottom panel) subunits. Expression represents the inverse hyperbolic sine (asinh) transform of the mean RNA-seq TPM values obtained from previously published datasets detailed in Supplementary Table 6. Sperm and egg were profiled with three and four biological replicates, respectively. h, Ectopic expression of SWN-Clover under control of the sperm lineage-specific DUO1 promoter. Predicted insertions were estimated from T2 segregation of RFP fluorescent seeds arising from the pAlligatorR43 selection marker. Expression of SWN-GFP in T1 lines was barely detectable in pollen and well below that predicted from the T2 segregation data. i, Schematic of the action of JMJ proteins, which can demethylate H3K27 di-and tri-methylation but not mono-methylation. Statistical source data are provided in Source Data Extended data fig. 4.

Extended Data Fig. 5

Transcriptional profiling of htr10, elf6;ref6;jmj13 and elf6;ref6;jmj13;htr10 pollen. a, Principal component analysis illustrating the high reproducibility of replicates and variation among the RNA-seq datasets generated from WT (n = 3 replicates), htr10(n = 4 replicates), elf6;ref6;jmj13 (n = 3 replicates) and elf6;ref6;jmj13;htr10(n = 3 replicates) pollen. All the biological replicates indicated (n) were used in the analysis that follows in panels b,c,d of this figure. b,c, Expression of (a) Arabidopsis histone H4 variants and (b) H3K27 demethylases in WT, elf6;ref6;jmj13, htr10 and elf6;ref6;jmj13;htr10 pollen. Expression represents the inverse hyperbolic sine (asinh) transform of the mean RNA-seq TPM values. The mean value of the biological replicates in a is shown, while the asterisks (*) indicate significantly different expression relative to WT pollen (p <0.001) using DESeq differential expression analysis and Benjamin-Hochberg correction to control for multiple comparisons. See source data for p-values. d, Volcano plots summarising significantly (adjusted p-value < 0.1) up-regulated (log2 FC > 0, red) and down-regulated (log2 FC < 0, blue) genes in htr10, elf6;ref6;jmj13 and elf6;ref6;jmj13;htr10 pollen relative to WT. DESeq analysis was used to determine differentially expressed genes from the biological replicates detailed in a and multiple comparisons controlled for using Benjamin-Hochberg correction. See Supplementary Table 3. e, Differentially-expressed genes (DEGs) in htr10(n = 73) and elf6;ref6;jmj13(n = 194) significantly overlap each other. Significance of the enriched overlap (p-value) was determined using a two-sided Fisher’s exact test. f, Clustered heatmap displaying enriched gene ontology (GO) terms associated with the DEGs in htr10(n = 73), elf6;ref6;jmj13(n = 194) and elf6;ref6;jmj13;htr10(n = 468) pollen relative to WT. Significant enrichment was assessed using g:Profiler and controlled for the multiple testing problem using the in-built g:SCS (sets counts and sizes) correction.

Extended Data Fig. 6

Sperm-specific accumulation of H3K4me3 is enriched at somatic H3K27me3 domains. a, Heatmaps centred on H3K4me3 peaks in sperm and leaf. Regions are split based on peaks being sperm-specific, leaf-specific or common to both sperm and leaf. The number of peaks and relative percentage are indicated in the labels to the left. Plotted is the ChIP-seq log₂ ratio relative to input or H3 for sperm and leaf, respectively. ChIP-seq was performed with three biological replicates for sperm; four for leaf. b, Expression of the Arabidopsis SET-domain family of proteins. Expression represents the inverse hyperbolic sine (asinh) transform of the mean RNA-seq TPM values obtained from previously published datasets detailed in Supplementary Table 6. Sperm and egg were profiled with three and four biological replicates, respectively. c, Overlap of somatic H3K27me3 domains with sperm-specific H3K4me3 peaks. Statistical analysis is based on a one-sided permutation overlap test (n = 100 permutations) compared with random TAIR10 genomic regions.

Extended Data Fig. 7

Reprogramming of Polycomb-silenced genes in sperm. a, Heatmap illustrating the developmentally regulated expression of somatic H3K27me3-marked genes. Expression represents z-score normalised RNA-seq TPM values. b, Heatmap of the expression of the genes marked in Figure 4a. Expression represents the inverse hyperbolic sine (asinh) transform of RNA-seq TPM values. c, Averaged DNA methylation signal over MEGs and PEGs in sperm. Plotted is the proportion of methylated cystosines in all contexts (i.e. CG, CHG and CHH). d, Averaged H3K4me3 signal over PEGs with detectable expression (TPM>1, black line) or no expression (TPM<1, grey line) in sperm. Plotted is the ChIP-seq log₂ enrichment relative to input. PEGs accumulate H3K4me3 regardless of expression in sperm, although the level of H3K4me3 enrichment was expectedly higher at sperm-expressed PEGs. Statistical source data are provided in Source Data Extended data fig. 7.

Fig. 1. H3K27me3 marks are globally lost from Arabidopsis sperm chromatin.

a, H3 antibody (α-H3) and 4′,6-diamidino-2-phenylindole (DAPI) staining of Arabidopsis sperm nuclei. Scale, 2 μm. b, Schematic of Arabidopsis pollen development. Microspore nuclei (MN) divide asymmetrically to produce a vegetative nucleus (VN) and germ cell nucleus (GN). The latter divides once again to produce two sperm nuclei (SN). c, α-H3K27me3 and DAPI staining of Arabidopsis vegetative cell nuclei (VN) and sperm nuclei (SN). Scale, 2 μm. The immunostain was repeated three times. d-e,α-H3K27ac (d) and α-H3K27me1 (e) immunostaining of Arabidopsis sperm nuclei alongside DAPI staining. Scale, 2 μm. f, Whole-mount α-H3K27me3 and DAPI staining of Arabidopsis pollen grains at different stages of development. Cell wall autofluorescence is indicated (cwa). Scale, 5 μm. The immunostaining in a and d-f was repeated twice. g, Quantification of α-H3K27me3 levels in VN and SN from WT, htr10, elf6;ref6;jmj13 and elf6;ref6;jmj13;htr10 pollennormalized to total H3 content. The violin curve represents the density of differing H3K27me3 levels. h, Expression of Arabidopsis histone H3 lysine 27 mono- (me1) and tri- (me3) methyltransferases (top two panels) and demethylases (third panel). Expression represents the inverse hyperbolic sine (asinh) transform of the mean RNA-seq TPM value sobtained from previously published datasets detailed in Supplementary Table 6. Sperm and egg were profiled with three and four RNA-seq biological replicates, respectively. i, α-H3K27me3 and α-H3 staining of VN and SN from wild type (WT), elf6;ref6;jmj13 and htr10 pollen. Scale, 2 μm. j, Quantification of α-H3K27me1 levels in VN and SN from WT, htr10,elf6;ref6;jmj13 and elf6;ref6;jmj13;htr10 pollen normalized to total H3 content. In g and j, red bar represents the mean of each genotype; quantification and statistical analysis were based on samples from one representative experiment; sample size (n = total number of nuclei) of each genotype is denoted alongside each violin plot. Statistical analysis was performed using a two-sided Mann-Whitney U-test. The experiment was repeated twice. Statistical source data are provided in Source Data fig. 1.

Fig. 2. Sperm-specific histone H3.10 is immune to K27 methylation.

a, Expression of Arabidopsis histone H3 variants. Expression represents the inverse hyperbolic sine (asinh) transform of the mean RNA-seq TPM values obtained from previously published datasets detailed in Supplementary Table 6. Sperm and egg were profiled with three and four biological replicates, respectively.. b, Expression of H3.10-Clover during pollen development. Cell wall autofluorescence is indicated (cwa). The experiment was repeated twice. Scale, 5 μm. c, N-terminal tail alignment of histone H3.1, H3.3 and H3.10-like variants from Arabidopsis thaliana (AtH3.10), Arabidopsis lyrata (AlH3.10), Arabidopsis halleri (AhH3.10), Capsella rubella (CrH3.10), Boechera stricta (BsH3.10), Lilium davidii (LdH3.10) and Oryza sativa (OsH3.10). Variants verified as sperm-specific in the literature are marked with a black dot. Black shading shows residue differences. The divergent domain and K27 are highlighted in red. Alanine 31 is highlighted in green. d, In vitro histone methyltransferase assays using recombinant H3.1 and H3.10 nucleosomes and Arabidopsis mono- (ATXR5/6) and tri- (MEA-PRC2 and CLF-PRC2) K27 methyltransferases. HMT assays with H3.10K27A nucleosomes confirm that residual MEA-PRC2 activity is non-specific to K27. The assay was performed once for ATXR5/6 and twice for PRC2. e, Western blot analysis of leaf histones isolated from two independent ProHTR5:HTR10-HA transgenic lines. No form of K27 methylation is detected on HA-tagged H3.10. This was not due to poor antibody affinity (Extended Data Fig. 2a-c) nor failure of the tagged variant to be modified in vivo, as indicated by the detection of H3K4me3. The experiment was repeated twice. f, Expression of Arabidopsis histone H3 variants in WT, htr10, elf6;ref6;jmj13 and elf6;ref6;jmj13;htr10 pollen. Expression represents the asinh transform of the mean RNA-seq TPM values of three biological replicates; four in the case of htr10. * indicates significantly different expression relative to WT pollen (p < 0.001) using DESeq differential expression analysis and Benjamin-Hochberg correction to control for multiple comparisons. See source data for precise p -values. Statistical source data and raw blots are provided in Source Data fig. 2.

Fig. 3. H3.10 deposition in sperm correlates with the loss of H3K27me3.

a-c, ChIP-seq tracks of the region surrounding HTR10 (a) BBM (b) and FLC (c). All three genes (blue shading) are embedded within broad somatic H3K27me3 domains. Coverage is represented as the log₂ ratio of IP DNA relative to input. Coloured and grey shading indicate an enriched or depleted signal, respectively. d, ChIP-seq profiles of leaf H3K27me3 (top), sperm H3K27me3 (middle) and sperm H3.10 (bottom) over somatic H3K27me3 domains. In sperm, H3K27me3 is lost over these regions, which instead become enriched for H3.10. e-f, Similar profiles as shown in (d) but centered on regions enriched with H3K4me3 (e) and H3K27ac (f) in leaf. In sperm, these regions remain enriched with each mark together with H3.10. g, Number of H3K27me3 peaks called in leaf, seedling and sperm. Somatic datasets were subsampled to read depth in sperm prior to peak calling. h, Box plot of the size distribution (in kb) of somatic and sperm H3K27me3 domains indicating the minimum and maximum values as well as the 25th, 50th and 75th quartiles. The number of peaks considered were n=6,235 (soma) and 478 (sperm). i-j, ChIP-seq profiles showing how H3.10 is depleted over retained H3K27me3 peaks in sperm. k-l, ChIP-seq profiles over de novo peaks of H3K27me1 (k) and H3K27me3 (l) in WT and htr10 sperm. m, ChIP-seq tracks of the region surrounding three representative genes down-regulated in htr10 pollen relative to WT. Coloured and grey shading indicate an enriched or depleted signal, respectively. H3K27me1 and H3K27me3 signal is increased in the flanking promoter region (blue shading). n, Overlap of de novo H3K27me1 and H3K27me3 peaks that accumulate in htr10. Statistical p-value of the peak overlap is based on a one-sided permutation test (n = 100 permutations) compared with random TAIR10 genomic regions. o, H3K27me1 and H3K27me3 signal in WT and htr10 sperm over genes down-regulated in elf6;ref6;jmj13;htr10 pollen relative to WT. ChIP-seq was performed with two and three biological replicates for WT and htr10 sperm, respectively.

Fig. 4. Paternal resetting of H3K27me3 facilitates sperm specification

a, Chromatin state of Polycomb-silenced genes in leaf and sperm clustered based on sperm H3K4me3. Number of genes is n=1,866 (cluster 1), 2,220 (cluster 2) and 3,105 (cluster 3). Sperm differentiation genes (blue), embryonic factors (red) and post-embryonic regulators (grey) are marked. b, Percentage of cluster 1, 2 and 3 genes expressed in gametes, embryos and vegetative tissues (TPM > 5). TPM represents the mean obtained from previously published datasets detailed in Supplementary Table 6. Sperm and egg were profiled with three and four biological replicates, respectively. c, Sperm transcript levels (log₂ RNA-seq TPM)for genes expressed from each cluster. Sample size (n= genes with TPM > 0.1)is denoted on each boxplot, which indicates minimum and maximum values as well as 25th, 50th and 75th quartiles. Statistical analysis was performedusing a two-sided Mann-Whitney U-test. d, Percentage Polycomb targets among genes with enriched expression in sperm (n =463 genes), early embryos (n=279 genes) and early endosperm (n =463 genes)with colours corresponding to the clusters defined in panel a. Statistics is based on a one-sided permutation test (n = 100 permutations) compared with random TAIR10 regions. e, ChIP-seq tracks of three sperm differentiation genes - SHORT SUSPENSOR (SSP), HAPLESS 2 (HAP2) and DUO1-ACTIVATED ZINC FINGER 3 (DAZ3). Coverage represents log₂ ratio of IP relative to input. Coloured and grey shading indicate enriched or depleted signal, respectively. In panels a and e, ChIP-seq was performed with three biological replicates; two for sperm H3K27me3. f, Differential expression between htr10, elf6;ref6;jmj13 and elf6;ref6;jmj13;htr10 pollen relative to WT, with minimum and maximum values as well as 25th, 50th and 75th quartiles indicated. Sample size (n) of pollen-expressed (grey) and sperm-enriched genes (black) is shown. Statistical analysis was performed using two-sided Mann-Whitney U-tests. g, Overlap of differentially-expressed genes in htr10, elf6;ref6;jmj13 and elf6;ref6;jmj13;htr10 with sperm-enriched and cluster 1, 2, 3 genes. Statistical enrichment was determined using pairwise two-sided Fisher’s exact tests. Statistical source data including sample sizes (n) and precise p-values are provided in Source Data fig. 4.

Fig. 5. Sperm chromatin state forecasts gene expression in the next generation

a, Percentage of MEGs (n = 57 genes) and PEGs (n = 66 genes) overlapping with somatic H3K27me3 domains. Significance was determined by chi-square analysis compared to proportion in Arabidopsis (n = 28,775 genes). b-c, H3K27me3 signal over MEGs and PEGs in leaves (b) and sperm (c). d, H3K4me3 signal over MEGs and PEGs in sperm. Plotted in c,d is the ChIP-seq log₂ enrichment relative to input; b relative to H3. e, ChIP-seq tracks of three PEGs - SU(VAR)3-9 HOMOLOG 7 (SUVH7), VARIANT IN METHYLATION 5 (VIM5) and YUCCA 10 (YUC10). f, Heatmap of PEG expression representing the asinh transform of the mean RNA-seq TPM values obtained from previously published datasets detailed in Supplementary Table 6. Sperm and egg were profiled with three and four biological replicates, respectively. g, Overlap enrichment of paternal or maternal biased genes in early zygotes with cluster 1, 2, 3 genes. Statistical enrichment was determined using pairwise two-sided Fisher’s exact tests. See source data for sample sizes (n) and precise p-values. h, 14h zygote transcript levels (log₂ RNA-seq TPM) for genes expressed from cluster 1, 2, 3. Sample size (n = genes with TPM > 0.1) is denoted on each boxplot, which indicates minimum and maximum values as well as 25th, 50th and 75th quartiles. Statistical analysis was performed using two-sided Mann-Whitney U-tests. i, ChIP-seq tracks of three embryonic regulators - LEAFY COTYLEDON 1 (LEC1), FUSCA3 (FUS3) and WUSCHEL-RELATED HOMEBOX 2 (WOX2). In panels e and i, coverage represents log₂ enrichment relative to input while coloured and grey shading indicate enriched or depleted signal, respectively. ChIP-seq data in b,c,d,e,i was performed using three biological replicates; two for sperm H3K27me3. j, Model for the H3K27me3 resetting mechanism in Arabidopsis sperm. This involves silencing of multiple PRC2 subunits, H3K27me3 demethylase activity and sperm-specific deposition of histone H3.10, which is immune to K27 methylation. The conjunction of these pathways leads to chromatin and transcriptional reprogramming of genes expressed during sperm differentiation and in the next generation. Statistical source data are provided in Source Data fig. 5.