H2A.X promotes endosperm-specific DNA methylation in Arabidopsis thaliana

Background H2A.X is an H2A variant histone in eukaryotes, unique for its ability to respond to DNA damage, initiating the DNA repair pathway. H2A.X replacement within the histone octamer is mediated by the FAcilitates Chromatin Transactions (FACT) complex, a key chromatin remodeler. FACT is required for DEMETER (DME)-mediated DNA demethylation at certain loci in Arabidopsis thaliana female gametophytes during reproduction. Here, we sought to investigate whether H2A.X is involved in DME- and FACT-mediated DNA demethylation during reproduction. Results H2A.X is encoded by two genes in Arabidopsis genome, HTA3 and HTA5. We generated h2a.x double mutants, which displayed a normal growth profile, whereby flowering time, seed development, and root tip organization, S-phase progression and proliferation were all normal. However, h2a.x mutants were more sensitive to genotoxic stress, consistent with previous reports. H2A.X fused to Green Fluorescent Protein (GFP) under the H2A.X promoter was highly expressed especially in newly developing Arabidopsis tissues, including in male and female gametophytes, where DME is also expressed. We examined DNA methylation in h2a.x developing seeds and seedlings using whole genome bisulfite sequencing, and found that CG DNA methylation is decreased genome-wide in h2a.x mutant seeds. Hypomethylation was most striking in transposon bodies, and occurred on both parental alleles in the developing endosperm, but not the embryo or seedling. h2a.x-mediated hypomethylated sites overlapped DME targets, but also included other loci, predominately located in heterochromatic transposons and intergenic DNA. Conclusions Our genome-wide methylation analyses suggest that H2A.X could function in preventing access of the DME demethylase to non-canonical sites. Alternatively, H2A.X may be involved in recruiting methyltransferases to those sites. Overall, our data suggest that H2A.X is required to maintain DNA methylation homeostasis in the unique chromatin environment of the Arabidopsis endosperm.


Background DNA methylation regulates gene expression and silences transposable elements (TEs) in plants and
vertebrates [1], and epigenetic reprogramming by DNA demethylation is vital for reproduction in mammals and owering plants [2][3][4][5]. In Arabidopsis thaliana, DNA demethylation during reproduction is catalyzed by the DNA glycosylase DEMETER (DME) [6]. DME is a dual function glycosylase/AP lyase, which actively removes DNA methylation via the Base Excision Repair (BER) pathway [7]. DME-mediated DNA demethylation occurs genome-wide at discrete loci that fall into two groups. The rst consists of relatively euchromatic, AT-rich, small TEs that are nucleosome-poor, and generally interspersed with genes in chromosome arms [8]. The second group of loci require the Facilitates Chromatin Transactions (FACT) complex for DME access, and are longer, heterochromatic TEs prevalent in pericentromeric, gene poor regions, enriched with heterochromatic histone marks and H1 linker proteins [9]. DME and DME-FACT mediated DNA demethylation occurs speci cally in male and female gamete companion cells, the vegetative and central cells, respectively [8, 10], and is vital for Arabidopsis reproduction, whereby loss of maternal DME or FACT results in development abnormalities, loss of genomic imprinting and seed abortion [6-8, 11,12].
In Arabidopsis, H2A.X is essential for the response to DNA damage, whereby the phosphorylation of its SQEF motif by Ataxia Telangiectasia Mutated (ATM) and ATR kinases, serves as a signal for recruitment of DNA repair and checkpoint proteins [16][17][18]. It is not known how FACT is recruited to DME target sites, and the apurinic/apyrimidinic (AP) sites created during base-excision repair (BER) can lead to the formation of double strand breaks [19]. We therefore sought to explore whether recruitment of H2A.X to sites of DME activity during BER may provide a functional link between H2A.X, FACT and DME during Arabidopsis reproduction. In order to investigate this, we analyzed the expression and activity of H2A.X during Arabidopsis reproduction, nding that H2A.X is expressed throughout the plant, particularly in developing tissues and the male and female gametophytes. The loss of H2A.X does not impair DMEmediated DNA demethylation, however, but leads to CG hypomethylation at intergenic regions and transposable elements, speci cally in the endosperm.

Results
Arabidopsis seedlings lacking H2A.X have reduced DNA damage tolerance H2A.X is encoded by two genes in Arabidopsis, HTA3 (AT1G54690) and HTA5 (AT1G08880). To investigate the effect of H2A.X mutations, we generated double mutants lacking both HTA3 and HTA5 (Fig. 1a), veri ed the loss of transcripts using RT-PCR ( Fig. 1b and Additional le 1: Fig. S1a) and analyzed the sporophytic and gametophytic phenotypes of h2a.x plants. h2a.x mutant allele segregation, plant morphology, growth rate and owering time were all normal, except for a signi cant increase in root hair length (Fig. 1c, p < 0.0001). h2a.x root hairs were ~ 15% longer than WT three days after germination (DAG). We then tested whether cell proliferation was normal in h2a.x, using the 5-ethynyl-2′-deoxyuridine (EdU), a thymine analog, and click chemistry to measure incorporation in newly synthesized DNA [20]. We did not observe a difference in EdU-stained cells between wild-type and h2a.x roots (Fig. 1d), indicating that S-phase progression and cell proliferation are normal in h2a.x mutants. We also measured whether there was increased DNA damage in mutant roots using propidium iodide (PI) staining but did not observe any differences between WT and h2a.x (Fig. 1e). These observations are consistent with mutant phenotypes observed in other DNA damage pathway genes, such as ATM or ATR kinases, which only exhibit a phenotype under growth conditions that promote DNA damage [21].
We therefore grew h2a.x and segregating WT seeds on MS plates containing Bleomycin sulphate, which induces double strand breaks (DSB) in DNA. MS Bleomycin concentrations of 0.5 ng/ml were used to test primary root formation and 1 ug/ml to test true leaf formation, as root development was more sensitive to the drug. h2a.x mutant seedlings had a signi cant reduction in root length compared to WT ( Fig. 1f and Additional le 1: Fig. S1b). True leaf formation rate was slightly reduced in h2a.x mutants (Fig. 1g). These data are consistent with previous ndings, also showing aberrant true leaf and root growth in h2a.x double mutant seedlings grown under genotoxic stress [22]. Thus, a lack of H2A.X resulted in increased sensitivity of developing tissues to DNA damaging agents, showing that H2A.X is required for the response to DNA damage in Arabidopsis.
H2A.X is widely expressed across Arabidopsis tissues, including in gamete companion cells To investigate the role of H2A.X in Arabidopsis development, we analyzed its expression pattern in sporophytic and reproductive tissues. We generated translational fusion constructs between the Green Fluorescent Protein (GFP) gene and either the HTA3 or HTA5 genes, including their promoter sequences, and introduced them into WT Col-0 Arabidopsis plants using Agrobacterium mediated transfer, deriving three and four independent lines for each allele, respectively. GFP uorescence was observed using confocal microscopy. Both HTA3 and HTA5 proteins were expressed in dividing cells of the sporophyte: First true leaves ( Fig. 2a and b), the oral meristem (Fig. 2c), the adaxial leaf surface ( Fig. 2d and e), root tips ( Fig. 2f and g), petal tips (Fig. 2h) and root meristem (Fig. 2i), though HTA5 was more strongly expressed. In reproductive structures supporting gametophyte development, such as the ovule primordia ( Fig. 2j), anthers ( Fig. 2k and l), and developing ovules (Fig. 2m), HTA5 was predominant. In the next generation seed, both isoforms were present in the developing embryo (linear cot), but not in endosperm at this stage (heart) ( Fig. 2n and o). Conversely, in gametophytic development, both isoforms were again expressed but HTA3 was the dominant isoform ( Fig. 3). In the male gametophyte, both HTA3 and HTA5 were present in the microspore prior to mitosis. After Pollen Mitosis 1 (PMI) HTA3 was expressed in the generative and vegetative nucleus of bicellular pollen, and following Pollen Mitosis II (PMII), in both sperm cells and the vegetative nucleus of mature, tricellular pollen (Fig. 3a). HTA5 expression was also present in both the generative and vegetative nucleus following PMI, but was lost in the vegetative nucleus following PMII, in tricellular pollen (Fig. 3b). In the female gametophyte, egg cell expression was visible for both HTA3 and 5, but was weak, conversely, HTA3 expression was very striking in the central cell, where it persisted following fertilization in the rst cell divisions of the developing endosperm ( Fig. 3c and e). HTA5 expression was also observed in the central cell, but expression in the surrounding ovule tissue was more striking for this isoform (Fig. 3d). DME does not regulate H2A.X expression in the Arabidopsis gametophyte Since H2A.X expression was prominent in the central and vegetative cells, speci cally where DMEmediated demethylation and related BER activity takes place [10,23], we reasoned that H2A.X expression may be regulated by promoter DNA methylation, whereby DME might demethylate HTA3 and HTA5 promoter sequences in the gametophyte, increasing expression of these transcripts. To test this hypothesis, we utilized wild-type plants hemizygous for the HTA3:GFP transgene, for which strong HTA3 expression could be observed in the central cell in ~ 50% of the developing ovules (Fig. 4). We crossed these plants with DME/dme-2 heterozygotes to derive DME/dme-2 plants that were also hemizygous for the HTA3:GFP transgene. A maternally inherited dme-2 mutation generates embryo abortion and seed lethality, so analysis of seed development is generally only possible in DME/dme-2 heterozygotes. We then analyzed the incidence of HTA3:GFP expression in DME/dme-2 mutants and their segregating wildtype siblings in the F2 population. In both DME/DME HTA3:GFP/- (Fig. 4a) and DME/dme-2 HTA3:GFP/-F2 (Fig. 4b) siblings we observed that ~ 50% of the female gametophytes within ovules produced a strong GFP signal, indicating that the loss of DME did not alter the expression of H2A.X in the Arabidopsis female gametophyte. Consistent with this, when we compared promoter DNA methylation for the H2A.X variants in Arabidopsis wild-type and dme-2 mutant central cells and endosperm [11,24], we found that H2A.X promoter methylation was low in both tissues, and unchanged in the dme-2 mutant (Additional le 1: Fig. S2a and S2b). Other H2A variant gene loci were also unmethylated in both wild-type and dme-2 mutant central cell and endosperm, except for H2A.Z.4, which exhibited promoter methylation in central cell and endosperm, that increased in dme-2 mutants, a hallmark of a DME-target promoter (Additional le 1: Fig. S2c).
h2a.x mutant endosperm is hypomethylated genome-wide To investigate whether changes in DNA methylation were present in h2a.x, we carried out genome-wide bisul te sequencing (BS-seq) of manually dissected endosperm and embryo from homozygous h2a.x mutant and wild-type F1 seeds and their resulting seedlings, following self-pollination of homozygous h2a.x mutants and wild-type sibling plants. We observed that embryo DNA methylation in the h2a.x mutant was unchanged from wild-type, with the peak of fractional methylation difference at zero (Fig. 5a). However, DNA methylation of h2a.x mutant endosperm was reduced compared to wild-type in the CG context, with the fractional methylation difference peak shifted to the left (Fig. 5b).
To ascertain which endosperm loci were hypomethylated, we aligned our methylome data to the 5' transcriptional start sites (TSS) and 3' transcriptional end sites (TES) of genes and transposons, also including the h2a.x seedling methylome. Hypomethylation was present only in endosperm, and although also present in gene bodies and intergenic regions, was most striking in transposon bodies, (Fig. 5c and d). CHG and CHH methylation was also reduced in endosperm transposon bodies (Additional le 1: Fig.   S3a-S3d). In h2a.x embryos, CHH methylation in TEs was decreased (Additional le 1: Fig. S3d). However, embryo CHH methylation increases steadily with time during embryo development [25] so it is likely the differences observed are technical, whereby mutant seeds were dissected slightly earlier in their development than wild-type. Consistent with this, CHH methylation in h2a.x seedlings, which derive directly from the embryo, was not different from WT (Additional le 1: Fig. S3d).
H2A.X hypomethylation overlaps DME target loci Inheritance of a maternal loss-of-function dme allele or a maternal loss-of-function ssrp1 allele, which encodes one of the proteins in the FACT complex, result in striking phenotypes of seed abortion and developmental delay. Seed viability in homozygous h2a.x mutants, as well as in crosses from maternal h2a.x with wild-type Col-0 pollen, was normal, suggesting that DME-and DME-FACT-mediated DNA demethylation occurred normally in h2a.x mutant seeds, at least at PRC2 genes critical for seed viability. In wild type female gametophytes, the central cell genome undergoes genome-wide epigenetic remodeling. DME/DME-FACT-mediated DNA demethylation leads to a deeply hypomethylated maternal endosperm genome compared to embryo [8, 9,11,24]. To assess whether h2a.x mutant hypomethylated regions overlapped sites of canonical endosperm hypomethylation, we compared differential methylated regions between endosperm and embryo (EN-EM DMRs) in WT and h2a.x mutant seeds. There were 4,451 hypo-EN-EM DMRs between WT endosperm vs embryo, covering about 1.3 M bps. In contrast, 7,526 hypomethylated EN-EM DMRs were identi ed between h2a.x endosperm and embryo, covering 2.7 M bps in length, more than double the area of the wild-type hypomethylated EN-EM DMRs (Fig. 6a).
Since many wild-type EN-EM DMRs are a result of DME activity in the central cell, we measured the overlap between h2a.x EN-EM DMRs and DME targets, nding that of the 7526 h2a.x EN-EM DMRs, 4692 (62%) overlapped with canonical DME target loci [8]. We therefore delineated the hypomethylated h2a.x EN-EM DMRs according to whether they overlapped WT embryo-endosperm DMRs (including DME targets) (n = 3238), and those that were novel h2a.x speci c DMRs (n = 4357; Fig. 6a and b). There was also a group of WT DMRs which were only differentially methylated between WT embryo and endosperm (n = 1213). We delineated DMRs by size (0.1 kb->1.5 kb) and found that h2a.x EN-EM DMRs were represented across all size classes, with similar proportions of size classes to wild-type (Fig. 6c).
Since h2a.x hypomethylation overlapped DME targets, we next assessed whether the h2a.x mutation affected DME activity in the central cell. Maternal h2a.x mutant plants in the Columbia ecotype were pollinated with wild-type Ler pollen, and BS-seq with embryo and endosperm from manually-dissected mutant and segregating wild-type seeds carried out, sorting resulting reads according to their parental ecotype. In this way, the maternal endosperm genome can be used as a proxy for the central cell genome. Here, the kernel density methylation plot peak is close to zero, but slightly negative, indicating that the maternal endosperm genome is hypomethylated (Fig. 6d, green trace and Fig. 5b). We then speci cally looked at maternal DME and shared DME-FACT loci, but these peaks also lay close to zero, with no striking skewing observed (Fig. 6d, red and orange traces, respectively). This indicates that whilst h2a.x hypomethylated loci do overlap some DME and DME-FACT targets, the h2a.x mutation does not affect DME/FACT targeting or activity at their canonical sites.
h2a.x mutant endosperm hypomethylation occurs post-fertilisation Since the maternal h2a.x endosperm allele was hypomethylated, yet H2A.X is not expressed in wild type endosperm, (Fig. 2p) we next sought to investigate whether h2a.x-mediated methylation loss originated in the maternal gametophyte. Consistent with this idea, both H2A.X isoforms are strongly expressed in the wild-type central cell ( Fig. 3c and d). Using data from the F1 crosses outlined above, where maternal h2a.x mutant plants in the Col-0 ecotype were pollinated with wild-type Ler pollen, we next plotted both maternal and paternal CG methylation. In h2a.x embryos, both maternal and paternal allele CG methylation is identical to WT (peak aligns on zero, Fig. 7a), consistent with our observations in selfpollinated h2a.x mutants (Fig. 5a). However, in endosperm, a slight shift is visible towards the left, indicating mutant hypomethylation is present on both maternal and paternal endosperm alleles (Fig. 7b). This indicates that whilst maternal allele hypomethylation may be inherited from the central cell, hypomethylation of the wild type paternal allele is also present. Paternal allele hypomethylation must manifest post-fertilization, i.e. due to a reduction in CG methylation e ciency or maintenance. To ascertain which parental loci were hypomethylated, we again aligned our methylomes to the TSS/5' and TES/3' ends of genes and transposons (Fig. 7c-f and Additional le 1: Fig. S4a-S4d ). As in the previous analysis, CG methylation in embryo is not different from wild-type in heterozygous h2a.x mutant gene and transposon bodies, but both maternal and paternal endosperm alleles are hypomethylated in genes, intergenic regions and TEs, with hypomethylation in TE bodies being most visible. CHG methylation is the same in wild-type and H2A.X/h2a.x mutant embryo and endosperm (Additional le 1: Fig. S4e-S4h) whereas CHH methylation is decreased on both parental alleles, in both H2A.X/h2a.x embryo and endosperm (Additional le 1: Fig. S4i-S4l).
H2A.X is encoded by two almost identical isoforms, HTA3 and HTA5. To determine whether one isoform may have an effect independent of the other, we dissected developing seeds from both hta3/hta3 hta5/+ (H2A.X-g3) and hta3/+ hta5/hta5 mutants (H2A.X-g5), crossed to Ler, so that the sporophyte had one remaining copy of one of the isoforms, but both H2A.X isoforms are lost in ½ of the gametophytes. Following BS-seq, we determined that both isoforms act redundantly, whereby endosperm methylation was not substantially affected in either hta3/hta3 hta5/+ or hta5/hta5 hta3/+ seeds (Fig. 7g, H2A.X-g3 and H2A.X-g5, kdensity peaks on zero) compared to hypomethylated h2a.x double mutant endosperm (Fig. 7g, h2a.x peak shifted to the left).
h2a.x hypomethylation is widespread in intergenic DNA To assess if h2a.x endosperm hypomethylated loci are associated with particular chromatin states, we used published histone marks and genomic characteristics that topologically group the Arabidopsis genome into nine distinct chromatin states [26] and used them to compare methylation differences between homozygous h2a.x vs wildtype endosperm. For the hypomethylated EN-EM DMRs speci c to h2a.x, the majority reside in non-coding, intergenic sequences, including distal promoters (chromatin state 4, Fig. 8a) and AT-rich heterochromatic regions (chromatin states 8 and 9, Fig. 8a), consistent with what we observed in TE metaplots ( Fig. 5c and d, Fig. 7e and f). In addition, when we used fractional methylation differences to analyse the chromatin states of hypomethylated loci unique to the h2a.x mutant, i.e. not including those present in between wild-type embryo and endosperm, chromatin states 4 and 8 exhibit the largest shift ( Fig. 8b and c). These data indicate that the novel, h2a.x-speci c EN-EM DMRs lie primarily in chromatin states 4, 8 and 9.
In order to gain resolution on DMR location, we aligned wild-type and h2a.x DMR coordinates according to 5' and 3' ends of genes, which revealed that h2a.x hypomethylation is enriched in intergenic regions, consistent with its enrichment in chromatin states 4 and 8 (Fig. 8d). To further characterize the location of h2a.x EN-EM DMRs, we plotted their co-ordinates across the Arabidopsis genome in 300 kb bins (see Methods and Materials; Fig. 8e). This analysis showed that h2a.x EN-EM DMRs in general mirror the distribution of wild-type EN-EM DMRs, which are enriched in pericentric heterochromatin (Fig. 8e). To determine whether the h2a.x endosperm hypomethylation represented novel sites of demethylation, or resulted from increased demethylation at already demethylated sites (e.g., resulting in longer DMRs), we took a locus-speci c approach, using the IGV genome browser to view aligned methylation data and DMRs (Fig. 8f). The majority of h2a.x-speci c hypomethylation represented stand-alone, novel DMRs (red outline). h2a.x DMRs overlapped DME-mediated wild-type endosperm/embryo DMRs (green outline), but did not make them longer.

Discussion
H2A.X is one of the H2A variants in higher eukaryotes and differs from canonical H2A by its rapid phosphorylation to y-H2A.X in response to DNA double-strand breaks. Unmodi ed H2A.X is ubiquitously expressed and distributed throughout the genome as a component of nucleosome core structure, estimated to represent approximately 10% of H2A variants present in chromatin at any given time [27][28][29][30]. We show that H2A.X is widely expressed in Arabidopsis newly developing tissues and reproductive cells, including the companion cells of the male and female gametophytes the vegetative and central cells, respectively. Loss of H2A.X results in endosperm hypomethylation at intergenic regions and in heterochromatic TEs.
It was previously shown that the HTA3 and HTA5 gene promoters exhibited differences in activity, with HTA5 observed to be less active in the oral bud [31]. Consistent with this, we show that whilst HTA5 is the predominant protein isoform expressed in the sporophyte, HTA3 predominates in the developing gametophytes, though both are highly expressed in pollen. One explanation for H2A3/5 high expression in the vegetative and central cells is that DME activity creates AP sites during BER, that may lead to the formation of double strand breaks, thereby requiring high levels of H2A.X [19]. Intriguingly, however, in heterochromatin, the mechanism of DNA repair is different; an H2A.W variant, H2A.W.7 is phosphorylated by ATM to initiate the response in constitutive heterochromatin to DNA damage [22].
We observed a signi cant increase in root hair length in h2a.x mutants compared to wild-type in the absence of any DNA damaging conditions. Intriguingly, reduction in H2A.Z incorporation into chromatin also results in an increase in root hair length, since the altered chromatin state mimics phosphate de ciency -activating a phosphate de ciency response gene locus [32,33]. Similarly, h2a.x mutations may indirectly affect the expression of root hair-growth genes [34][35][36]. Alternatively, defective H2A.X expression may also cause nutrient-stress, resulting in the modulation of genes involved in root hair growth.
We identi ed hypomethylation on both male and female endosperm alleles in hybrid maternal h2a.x mutants crossed to paternal wild-type F1, meaning that hypomethylation manifests post-fertilization, at least on the paternal allele. On the maternal allele, both pre-and post-fertilization DNA methylation dysregulation may be present. Hypomethylation was also con ned to endosperm, whereas h2a.x embryo was normally methylated. Endosperm is a triploid tissue, with distinct higher-order chromatin structure compared to other tissues, being less condensed, and subsequently featuring increased trans-chromatin interactions, increased expression of TEs, and encroachment of heterochromatin into euchromatic regions [37,38]. Endosperm is also the site of parental competition for generational resources, in part re ected in the activities of DME and FACT in the central cell, which confer deep hypomethylation. As such, the chromatin environment of the central cell and endosperm may in turn impact DNA methylation homeostasis in h2a.x null tissues differently to embryo. The mechanism of DNA methylation loss in h2a.x endosperm remains unclear. The h2a.x hypomethylated DMRs represent regions that are not normally demethylated during seed development. Potentially, H2A.X could function in wild-type cells to exclude DME from these regions; preventing inappropriate remodeling of regulatory DNA and heterochromatic TEs. As such, when H2A.X is lost, DME may demethylate further sites in the central cell, contributing to endosperm hypomethylation, speci cally of the maternal allele. However, in human cells, H2A.X phosphorylation destabilises chromatin structure, increasing its accessibility [16]. Nucleosome cores are crucial for nuclear DNA organization and function, and lost H2A.X is likely quickly replaced by other H2A variants, such as H2A.Z, H2A.W, or by canonical H2A. The replacement of H2A.X with variants that cannot be phosphorylated, may therefore reduce chromatin access, for example, to DNA methyltransferases, resulting in comparative hypomethylation.
In conclusion, we demonstrate that H2A.X is expressed widely in developing Arabidopsis tissues and gamete companion cells, and show that the DNA damage response is impaired in h2a.x mutant roots and seedlings. We show that h2a.x mutant endosperm exhibits DNA hypomethylation at intergenic regions and heterochromatic TEs, creating a large number of endosperm-embryo DMRs, not present in wild-type.
We speculate that h2a.x mediated hypomethylation may be due to an increase in DME targeting to noncanonical sites, or a loss of chromatin accessibility to methyltransferases.

Conclusions
In this study, we examined the effect of h2a.x double mutants on genome-wide methylation patterns between embryo and endosprm, as well as between maternal and paternal alleles, compared to wild type. Hypomethylation was observed in the h2a.x endosperm, but not in the embryo or seedling. This hypomethylation was present on both maternal and paternal alleles, suggesting that hypomethylation occurs after fertilization, at least on paternal allele of h2a.x mutant endopserm. While the expression patterns and methylation targets of H2A.X ovelap with those of DME, other loci including heterochromatic transposons and intergenic sequences that are non-DME or non-DME-FACT canonical sites, are also hypomethylated in h2a.x mutant endopserm. In summary, our ndings suggest that H2A.X palys a role in preserving the balance of DNA methylation within its unique target sites, which represent distinctive chomomation states in Arabidopsis endopserm.

Methods and Materials
Plant materials and growth conditions Wild type and h2ax mutant Arabidopsis seeds were bleached and sown onto Murashige and Skoog plates, followed by cold treatement in the dark at 4 degrees C for 3 days, and two weeks growth in a light chamber, before transplantation onto soil. Seedlings were grown in a greenhouse with a long-day photoperiod (16 h light, 8 h dark). Seed stocks of T-DNA insertion mutants (SALK_012255 in HTA3 and SAIL_382_B11/CS873648 in HTA5, Fig. 1a) in the Columbia-0 (Col-0) background were obtained from the ABRC stock center. Mutant alleles were, backcrossed ve times to wild-type, and nally crossed to obtain double hta3/hta3; hta5/hta5 null plants, designated h2a.x, as well as segregating wild-type siblings. Both T-DNA insertion alleles have been studied and validated in recent reports [22,39].
Edu cell proliferation assay 5-ethynyl-2'-deoxyuridine (EdU) staining using an Invitrogen Click-iT™ EdU Alexa Fluor™ 488 HCS Assay (C10350) was performed based on Kotokany et al. (2010) [20] to detect S phase cells. Seeds were grown in MS media vertically for 3 days. Seedlings were collected in MS solution containing 1µM Edu and incubated at 22°C for 30 minutes. Samples were xed in 4%(w/v) formaldehyde solution in phosphatebuffered saline (PBS) with 0.1% Triton X-100 for 30min, and washed three times with PBS each for 5 minutes. The samples were incubated in Edu detection cocktail solution at room temperature for 30 minutes in the dark, and washed with the Click-iT® rinse buffer and then three times with PBS. The photos were taken using confocal microscopy (LSM700, Zeiss).

Propidium Iodide (PI) staining
Propidium Iodide (PI, P-4170, sigma) staining was used to detect cell death and show anatomy of the roots. The samples were stained with working PI solution (5ml PI solution in 1ml of distilled water) at room temperature for 30s and washed with distilled water on slide glass.

Observation of root hair phenotypes
A stereomicroscope (M205 FA, Leica) was used for the observation of root hair phenotypes. Root hair length was measured as previously described by [40] with slight modi cations as in the [41].
H2A.X expression localization HTA3 and HTA5 GFP fuson proteins were cloned alongside a hygromycin resistance casette using a Gibson assay (Invitrogen) and F1 seeds screened on MS containing hygromycin. F1 plants were screened manually using a uorescence microscope and seeds collected from plants expressing GFP. F2 seeds were grown on hygromycin and selected if we identi ed segregation of the resistance allele, indicating the presence of a single copy transgene. F3 plants were then used for confocal microscopy.

DNA damage assay
Segregating WT and h2a.x homozygous mutant Arabidopsis seeds were planted on MS containing 0.5ug/ml bleomycin sulphate and grown vertically for 14 days under long day conditions, before measuring root length. MS without bleomycin was used as a control. Values are from three independent experiments each including 15 seedlings for each genotype. True leaf assay was performed as previously described with 10-day-old seedlings [42].

Isolation of Arabidopsis endosperm and embryos
Segregating WT and h2a.x mutant Arabidopsis ower buds were either allowed to self, or emasculated at ower stage 12-13 using ne forceps and pollinated with Ler pollen 48 hours later. Eight to ten days after pollination (DAP) developing F1 seeds (linear to bending cotyledon stage) were immersed in dissection solution ( lter-sterilized 0.3 M sorbitol and 5 mM pH 5.7 MES) on sticky tape and dissected by hand under a stereo-microscope using ne forceps (Fine Science Tools, Inox Dumont #5) and insect mounting pins. The seed coat was discarded, and debris removed by washing collected embryos or endosperm ve to six times with dissection solution under the microscope. Bisul te sequencing library construction As described previously, genomic DNA was isolated from endosperm and embryo [11]. Bisul te sequencing libraries for Illumina sequencing were constructed as in [8] with minor modi cations. In brief, about 50 ng of genomic DNA was fragmented by sonication, end repaired and ligated to customsynthesized methylated adapters (Euro ns MWG Operon) according to the manufacturer's instructions for gDNA library construction (Illumina). Adaptor-ligated libraries were subjected to two successive treatments of sodium bisul te conversion using the EpiTect Bisul te kit (Qiagen) as outlined in the manufacturer's instructions. The bisul te-converted library was split between two 50 ul reactions and PCR ampli ed using the following conditions: 2.5 U of ExTaq DNA polymerase (Takara Bio), 5 µl of 10X ExTaq reaction buffer, 25 µM dNTPs, 1 µl Primer 1.1 and 1 µl multiplexed indexing primer. PCR reactions were carried out as follows: 95ºC for 3 minutes, then 14-16 cycles of 95 ºC 30 s, 65 ºC 30 s and 72 ºC 60 s. Enriched libraries were puri ed twice with AMPure beads (Beckman Coulter) prior to quanti cation with the Qubit uorometer (Thermo Scienti c) and quality assessment using the DNA Bioanalyzer high sensitivity DNA assay (Agilent). Sequencing on either the Illumina HiSeq 2000/2500 or HiSeq 4000 platforms was performed at the Vincent J. Coates Genomic Sequencing Laboratory at UC Berkeley.
Bisul te data analysis Sequenced reads were sorted and mapped to Col-0 and Ler genomes in cases of seeds derived from Col x Ler crosses, or not sorted and mapped to Col-0 for selfed samples. Gene and TE ends analysis and kernel density plots were generated as previously described [8], using only windows with at least 10 informative sequenced cytosines, and fractional methylation of at least 0.5 (CG), 0.4 (CHG) or 0.08 (CHH) in at least one of the samples being compared. In addition, the bisul te conversion rate was calculated using chloroplast DNA methylation ratio (Additional le 2: Table. S1).

Declarations Supplementary information
Additional les 1: Figure S1 - Figure S4 Additional les 2: Table S1 Ethics approval and consent to participate The seeds utilized in this study were sourced from the ARABIDOPSIS BIOLOGICAL RESOURCE CENTER (ABRC), eliminating the need for eld permissions to collect samples. As a result, specimens have not been deposited as vouchers. The authors have made a declaration that the experimental research conducted on the plants outlined in this paper adheres to institutional, national, and international guidelines.

Consent for publication
Not applicable

Availability of data and materials
The datasets generated and analyzed during the current study are in the NCBI GEO under accession number GSE233920 (reviewer link; https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE233920).

Competing interests
The authors declare that the research was conducted in the absence of any commercial or nancial relationships that could be construed as a potential con ict of interest.  Both HTA5 and HTA3 fused to GFP showed expressions in the rst true leaves (a and b), the oral meristem (c), the adaxial leaf surface (d and e), root tips (f and g), petal tips (h), and secondary root meristems (i). In reproductive structures supporting gametophyte development such as the ovule primordia (j) and anthers (k and l), and ovules (m) both isoforms were expressed. In the next generation seeds, both isoforms were present in the developing embryos (n and o).

Figure 3
HTA3 expression is more dominant in gametophytic development.
In the male gametophyte, both HTA3 and HTA5 are present in the microspore prior to mitosis. After Pollen Mitosis I (PMI) HTA3 was expressed in the generative and vegetative nucleus of bicellular pollen, and following Pollen Mitosis II (PMII), in both sperm cells and the vegetative nucleus of mature, tricellular pollen (a). HTA5 expression was also present in both the generative and vegetative nucleus following PMI, but was lost in the vegetative nucleus following PMII, in tricellular pollen (b). In the female gametophyte, egg cell expression was visible for both HTA3 and HTA5, but was weak. Conversely, HTA3 expression was very striking in the central cell, where it persisted following fertilization in the rst cell divisions of the developing endosperm (c and e). HTA5 expression was also observed in the central cell, but expression in the surrounding ovule tissue was more prominent for this isoform (d).  Genome-wide methylation analysis of selfed double h2a.x mutant developing embryo, endosperm and seedling. a Fractional methylation difference between h2a.x double mutant and WT CG methylation from embryo (linear-bending cotyledon) is plotted, data in 50 bp windows with >10x sequence coverage. Data are from h2a.x Col selfed plants and segregating wild type siblings. b As for a, but with endosperm. c Ends analysis of h2a.x mutant genomic methylation in genes, with genes aligned according to their 5' and 3' ends. d Ends analysis of h2a.xmutant genomic methylation in transposons, with transposons aligned according to their 5' and 3' ends.
Analysis of h2a.xmethylomes, comparing DMRs between endosperm and embryo. a Venn diagram illustrating that WT embryo and endosperm harbour 4451 DMRs, the majority of which (3238) are shared with h2a.x embryo and endosperm. h2a.xembryo and endosperm have an additional 4357 DMRs. b Box plots showing the relative methylation level of DMRs in embryo and endosperm, in wild type, h2a.xand dme-2 mutants. c Characterization of h2a.x-speci c embryo-endosperm DMRs; wild-type and h2a.x Endosperm-Embryo DMRs grouped by size, with the cumulative total length they covered shown, whereby they are represented across all DMR sizes, and represent an overall increase in size distribution. d Fractional methylation difference between h2a.x mutant and WT CG methylation from the maternal endosperm allele is plotted, data in 50 bp windows with >10x sequence coverage, for all loci (green), DME target loci (red) and DME-FACT target loci (orange)