EFFECTOR OF TRANSCRIPTION factors are novel plant‐specific regulators associated with genomic DNA methylation in Arabidopsis

Summary Plant‐specific EFFECTORS OF TRANSCRIPTION (ET) are characterised by a variable number of highly conserved ET repeats, which are involved in zinc and DNA binding. In addition, ETs share a GIY‐YIG domain, involved in DNA nicking activity. It was hypothesised that ETs might act as epigenetic regulators. Here, methylome, transcriptome and phenotypic analyses were performed to investigate the role of ET factors and their involvement in DNA methylation in Arabidopsis thaliana. Comparative DNA methylation and transcriptome analyses in flowers and seedlings of et mutants revealed ET‐specific differentially expressed genes and mostly independently characteristic, ET‐specific differentially methylated regions. Loss of ET function results in pleiotropic developmental defects. The accumulation of cyclobutane pyrimidine dimers after ultraviolet stress in et mutants suggests an ET function in DNA repair.


Introduction
Plant development depends on complex regulatory interactions, including the orchestrated coordination of numerous transcriptional networks. While interactions of transcription factors with DNA are essential for regulating gene expression, these are often modified through epigenetic mechanisms such as DNA methylation and histone modifications (Du et al., 2015).
Previous work on plant embryogenesis has led to the isolation of a plant-specific class of gene regulators (Raventos et al., 1998;Ellerstrom et al., 2005;Ivanov et al., 2008) with the founding members known as EFFECTOR OF TRANSCRIPTION (ET). Overexpression studies have shown that ET factors can affect diverse developmental processes, such as seed germination and xylem differentiation (Ellerstrom et al., 2005;Ivanov et al., 2008). ET proteins share highly conserved cysteine-histidine domains with zinc-and DNA-binding repeats. These repeats are also found in nonflowering plants such as the moss Physcomitrella patens, demonstrating their evolutionary conservation. ET1 and ET2-GFP fusion proteins are detectable in the nucleus (Ivanov et al., 2008). In addition to their functional DNA-binding ET repeats (Ellerstrom et al., 2005), ET factors share a characteristic DNA single-strand nuclease domain (GIY-YIG) with structural similarity to that of bacterial UVRC proteins (Dunin-Horkawicz et al., 2006) and homing nucleases (Stoddard, 2005;Liu et al., 2013). The bacterial UVRC protein is essential for DNA excision repair (Moolenaar et al., 1998a,b). It is targeted to ultraviolet (UV)-induced DNA lesions such as thymidine-dimers, and introduces two single-strand cuts eight bases upstream and four bases downstream of a lesion. The two single-strand cuts are made by two structurally distinct domains: a C-terminal domain consisting of an Endonuclease V (EndoV) and a Helix-hairpin-Helix (HhH) domain, which are required for the 5 0 -cut, and an N-terminal GIY-YIG domain, which inserts the 3 0 -nick (Van Roey et al., 2002). The sequence similarity between plant ET factors and UVRC is restricted to this single-strand cutting GIY-YIG domain, suggesting that an ancestral bacterial GIY-YIG domain has been recruited by ET proteins and combined with the DNA-binding ET repeats to create a novel plant-specific regulatory protein (Ivanov et al., 2008). The single-strand cleavage function of the Arabidopsis ET2 GIY-YIG domain has been confirmed by substitution and complementation of the corresponding domain of the Escherichia coli UVRC protein (Ivanov et al., 2008). On the transcriptional level, a sevenfold upregulation of ET2 was described in response to ionising radiation in Arabidopsis plants (Culligan et al., 2006). This upregulation was not detectable in plants deficient for ATAXIA-TELANGIECTASIA MUTATED (ATM), a sensor for DNA damage. This observation was driving our hypothesis that ET factors are involved in DNA repair.
The HhH domain, the second DNA-nicking domain in the UVRC protein, which is structurally distinct from the GIY-YIG domain, has been considered as the ancestral protein domain for two related epigenetic plant regulators, the DNA glycosylase DEMETER (DME) (Choi et al., 2002) and the REPRESSOR OF SILENCING 1 (ROS1) (Gong et al., 2002;Morales-Ruiz et al., 2006). DME can introduce single-strand nicks as part of a DNA demethylation pathway, whereas ROS1 represses homologydependent transcriptional silencing by demethylation of the target promoter (Gong et al., 2002). The importance of DNA methylation as an epigenetic marker required for several developmental phases such as seed development and germination was described recently (Kawakatsu et al., 2017). Several regions have been identified in which dynamic control of DNA methylation and transcriptional reactivation is contributing to reproductive development (Borges et al., 2012). Although basic molecular analysis of ETs has been performed, the functional context in planta remains elusive.
Considering the structural and functional similarities between DME/ROS1 and the ET factors, we hypothesised that ETs are involved in regulation of DNA methylation based on their singlestrand cleavage function. Here, we performed whole-genome DNA methylation analyses in flowers of et1 and et2 mutants to gain insights into the function of these proteins. From combining methylome data with transcriptional profiles and with extensive phenotypic analyses in different organs and tissues, we propose that Arabidopsis ET factors constitute a new class of epigenetic regulators involved in stable inheritance of DNA methylation patterns.

Plant material, mutant characterisation and complementation
Arabidopsis thaliana (L.) Heynh. plants of accession Columbia-0 (Col-0) for et1-1, et2-3, et1-1 et2-3 and Wassilewskija-2 (Ws-2) accession for et2-1 were grown in growth chambers under a 16-h photoperiod at 22°C and 60% humidity. Alternatively, seeds from each line were plated on Murashige and Skoog (MS;Murashige & Skoog, 1962) agar plates, supplemented with the appropriate antibiotic or herbicide when required. Seedlings were grown in growth chambers (Percival Scientific, Perry, IA, USA), under a 16-h photoperiod at 22°C. Green seedlings were transferred to soil and grown under the same conditions as for the mature plants.
T-DNA insertion lines et1-1 and et2-3 generated in the genetic background of Col-0 were obtained from the SALK T-DNA collection (Alonso et al., 2003) and designated as et1-1 (SALK_000422) and et2-3 (SALK_151861). The Ws-2-derived et2-1 mutant (Ivanov et al., 2008(Ivanov et al., , 2012 was isolated from the collection of the Arabidopsis Knock-out Facility (Sussman et al., 2000) at the University of Wisconsin Biotechnology Center, following a pool screening for insertion in the ET2 gene in the Ws-2 background (Krysan et al., 1999) and has been back crossed six-times to the Col-0 ecotype. The positions of the T-DNA insertions are summarized in Fig. 1(a). Homozygous single mutants, et1-1, et2-1 and et2-3, have been isolated and homozygous double mutants, et1-1 et2-1 and et1-1 et2-3, have been generated. Absence of full-length mRNA was confirmed by qualitative PCR spanning the T-DNA insertion site. Double mutants were generated by crossing the mutant et1-1 with the et2-1 and et2-3 mutants, respectively. Primer sequences are provided in Supporting Information Table S1. To complement the et1-1 mutation, a Col-0-derived ET1 genomic fragment including sequences 1228 bp upstream of the start codon and 792 bp downstream of the stop codon was PCR amplified using Platinum Taq High Fidelity (Invitrogen) and resequenced. The gene fragment was cloned into the pDONR/Zeo vector using the BP reaction (Gateway ® BP Clonase ® Enzyme Mix, Invitrogen) and transferred into the pBGW destination vector (Karimi et al., 2007) using the LR reaction (Gateway ® LR Clonase ® Enzyme Mix, Invitrogen). The complementation construct was introduced into Agrobacterium tumefaciens strain GV2260 by freeze-thaw transformation (Chen et al., 1994). The floral dip method (Clough & Bent, 1998) was used for plant transformation.
UV stress was applied using 1-wk-old seedlings and placing them for 15 min at 30 cm from an MBR UV-C mobile room sterilizer, mediating 165 µW cm À2 m À1 UV-C light (253.7 nm).

RNA extraction, cDNA synthesis
Total RNA was isolated from 10-d-old seedlings grown under sterile conditions on solid MS medium and from flower buds at the 12c-14 stage (Smyth et al., 1990).Total RNA was isolated from 100 mg of plant material using the RNeasy Plant Mini kit as described in the manufacturer's protocol (Qiagen) dissolved in 30 ll DEPC-treated water and treated with DNaseI (Roche). Total RNA concentration was quantified using a Nanodrop ® ND-1000 spectrophotometer (NanoDrop Technologies Inc., Waltham, MA, USA) and the quality and integrity was assessed by running 1 ll of every sample on an Agilent 2100 Bioanalyzer (Agilent Technology Inc., Waldbronn, Germany). First-strand cDNA was synthesized by reverse transcription from total RNA using the RevertAid H Minus First strand cDNA synthesis kit (Fermentas, Vilnius, Lithuania).

Quantitative real-time PCR
Quantitative real-time measurements were performed using SYBR Green Master Mix reagent in an ABI Prism 7700 Sequence Detection System (Applied Biosystems), according to the manufacturer's instructions. For each condition, three technical replicates and three biological replicates were used. Transcript levels were determined by quantitative real-time PCR (RT-PCR) and the raw threshold cycle values (C T ) for all samples were normalised against C T values obtained for the reference transcript of the ACTIN11 gene using QBASE software (Biogazelle, Ghent, Belgium). Primers used in this work were designed with the QuantPrime tool (Arvidsson et al., 2008) and are listed in Table S1.

RNA deep sequencing
Strand-specific cDNA libraries for Illumina Next Generation Sequencing were generated from triplicate biological samples (10-d-old seedlings and flower buds at the 12c-14 stage). For detailed description see Methods S1.

DNA methylation analysis
For detailed description of DNA methylation analysis see Methods S1.

Immunodetection of cyclobutane pyrimidine dimers (CPDs)
After 3 h of regeneration samples were taken from the aerial tissue and genomic DNA was extracted using a DNeasy Plant Mini kit as described in the manufacturer's protocol (Qiagen). In total, 1 lg of DNA was spotted on an Amersham Hybond N + Nylon membrane (GE Healthcare, Little Chalfont, UK). Immunodetection was performed according to the manufacturer's description in TBST/5% milk powder using the Anti-Thymine Dimer primary antibody (H3) from Abcam (ab10347, Cambridge, UK) and ECL anti-mouse IgG horseradish peroxidase-linked whole secondary antibody (NA931V, GE Healthcare). Signals were detected using Clarity Western ECL substrate (Bio-Rad) and Amersham Hyperfilm ECL (GE Healthcare). Quantification of signals was performed from three independent experiments using the QUANTITY ONE 4.5.2 software (Bio-Rad). For quantification the adjusted volume intensity 9 mm 2 /unstressed Col-0 background signal was calculated.

Microscopy techniques
For detailed description of microscopy techniques see Methods S1.

Data accessibility
Results of the whole genome bisulphite sequencing have been deposited at the European Nucleotide Archive under accession number PRJEB12413. DNA methylation data have been uploaded to the epigenome browser of the EPIC Consortium (https://www.plant-epigenome.org/; https://genomevolution. org/wiki/index.php/EPIC-CoGe) and can be accessed at http:// genomevolution.org/r/939v. The flower and seedlings transcriptome data have been deposited at the European Nucleotide Archive under accession numbers PRJEB19779 and PRJEB14889, respectively. All data deposited will be made publicly available upon publication.

Gene family evolution
The ET gene family in A. thaliana (Fig. 1a) comprises three members, ET1 (AT4G26170), ET2 (AT5G56780) and ET3 (AT5G56770). ET1 and ET2 encode all characteristic ET sequence motifs, including the typical cysteine-rich ET repeats and the GIY-YIG domain, whereas ET3 is a partial tandem duplication of ET2 lacking the C-terminal ET repeats. This 3 0 truncated gene is located downstream of ET2 and considered a nonfunctional pseudogene. Here, we focused our analysis on ET1 and ET2, which contain the name-giving ET-domain. ET genes are exclusively found in plants, suggesting their involvement in plant-specific processes. A phylogenetic tree identified the ET2-type gene as ancestral, dating back to the common ancestor of mosses and seed plants (Fig. 1b). ET2 consists of three exons of which the second encodes the GIY-YIG domain and the third contains the characteristic ET repeats. ET1 probably resulted from deletion of the second intron of ET2 and an insertion of a complete exon into the first intron of ET2 (Fig. 1c). Acquisition of the second exon, characteristic for ET1 genes, is only found in species of the family Brassicaceae (Fig. 1c). The evolutionary origin of ET1 might be the a-whole genome duplication event in this family (Hohmann et al., 2015).

Whole genome analysis of differentially methylated regions in et mutants
The single-strand cleavage function on DNA mediated by the GIY-YIG domain in the N-terminal half of the ET factors (Ivanov et al., 2008) suggested an effect on DNA methylation patterns analogous to described demethylases ROS1 and DME (Choi et al., 2002;Gong et al., 2002). Therefore, whole genome bisulphite sequencing (WGBS) of Col-0, et1-1 and et2-3 single mutants and the et1-1 et2-3 double mutant was performed. As ET factors show their maximum of expression in reproductive tissues (Ivanov et al., 2008), the analysis was focused on flower buds (12c-14 stage; Smyth et al., 1990). Principal component analysis (PCA) of differential DNA methylation showed clear separation of the genotypes and high similarity between biological replicates (Fig. 2a). Methylated regions (MRs) were identified in every sample using a previously published algorithm (Hagmann et al., 2015). Comparisons of Col-0 and mutant lines revealed 352 highly differentially methylated regions (hDMRs) for et1-1, 373 for et2-3 and 275 for the double mutant (see Table S2 for a list of hDMRs:). Cluster analysis of the hDMRs revealed preferential loss of methylation in the mutants compared to Col-0 ( Fig. 2b), mainly in the symmetric CG context (see Fig. S2a). Although genomic regions covered by hDMRs coincided mainly with transposable elements (TEs), hDMRs were proportionally over-represented 2 kb upstream and 2 kb downstream of protein-coding sequences (Fig. 2c). Methylated regions that were classified as non-DMRs showed minor variation in methylation, confirming the specificity of our algorithm (see Fig. S2d). Among identified hDMRs, MPF (hDMR686) was found, showing et-1-1-specific hyper-methylation. MPF (Methylated region near Flowering locus C, AT5G10140) was described as a marker region for loss of demethylation function (Penterman et al., 2007a;Zhai et al., 2008). To validate the identified hDMRs the available et-T-DNA insertion mutant lines et1-1, et2-1 and et2-3 ( Fig. 1a) were tested by clonal bisulphite sequencing analysis (Fig. S1). AT1G26400 (FAD-Berberine-binding protein), AT1G34245 (EPF2, Epidermal Protein Factor2) hypermethylated in ros1 and dme mutants, and AtSN1 as a reference region for RdDM (Kuhlmann & Mette, 2012) were tested. At AT1G26400 and AT1G34245, a significant increase of cytosine methylation was detectable for both alleles of the et2 mutant (et2-1 and et2-3). This increase was preferentially caused by an increase of methylation in the symmetric CG context.
The largest hDMR (hDMR180: 1159 bp) was located on chromosome 1 : 28515015, completely covering a HELITRON1 element (AT1TE93275); this locus was depleted of methylation in all three mutant lines. Demethylation of this region was characteristic for nerd mutant plants (Pontier et al., 2012). Loss of this GYF-and zinc-finger (CCCH-type) domain-containing protein function led to definition of a plant-specific chromatin-based RNA silencing pathway depending on RDR1/6. The second region defining the NERD pathway was psORF (AT5G35935). This region was also detected as hDMR750 in the et mutants.
The identified hDMRs in the et1 and et2 single mutants overlapped substantially (Fig. 2d), suggesting similar regulatory function of ET1 and ET2 at these shared loci. In turn, c. 50% of hDMRs were specific to either mutant, indicating an additional gene-specific influence of either of the two factors on DNA methylation. Differential DNA methylation with respect to Col-0 was similar in both mutants for a large fraction of hDMRs (Fig. 2c). In total, 70% of hDMRs showed the same directional methylation change. However, distinct roles of ET factors in DNA methylation were also apparent: 15% of et1-1 hDMRs were hypo-methylated in et1-1 but hyper-methylated in et2-3, while 6% showed the opposite pattern, which suggests antagonistic roles of ET1 and ET2 for methylation of these loci.

New Phytologist
For the vast majority of hDMRs, DNA methylation in the et1-1 et2-3 double mutant either reflected the situation in one of the single mutants, or showed additive effects, corroborating the combination of overlapping and specific function of ET1 and ET2 that we had already derived from the hDMR overlap analysis. Intriguingly, a small subset of hDMRs that showed loss of methylation in either et1-1 or et2-3 did not show methylation changes in the double mutant, suggesting epistatic interaction of ET1 and ET2 at these loci.
To gain insights into the methylation pathway that ET1 and ET2 might be involved in, we next used the hDMR between Col-0 and the et1-1 et2-3 double mutant as a proxy to investigate DNA methylation at the same loci in a collection of previously published epigenetic mutants (Stroud et al., 2013). As CG methylation was the most prominently affected in et1-1 et2-3, we focused our analysis on this context. Analysis of hDMRs with gain of methylation in et1-1 et2-3 revealed the closest similarity to methylation patterns of rdd mutant plants (Fig. 3a). rdd is a triple mutant defective for ROS1, DM2 and DML3 (Penterman et al., 2007b).
Although prior in vitro studies (Ivanov et al., 2012) showed that ETs bind to DNA irrespective of the sequence context, we investigated whether any sequence feature could be identified using the set of identified DMRs. Therefore, we choose the 136 hDMRs detected in et2-3 flower tissue which showed gain of methylation. We applied the motif-based sequence analysis tool MEME (Bailey et al., 2006) on these potential ET2 DNA binding motifs. No motif could be identified, suggesting that the DNA binding is not sequence-specific. The DIMONT approach (Grau et al., 2013), which includes sorting of the sequences according to intensities, did not reveal any binding pattern either. Importantly, when using a motif length of 10 (bgOrder = 0, motifOrder = 0, other parameters = default), we detected two adjacent pyrimidines (TT, CT, TC) as a recurring motif (Fig. S3).

Transcriptome analysis to identify differentially expressed genes in et mutants
Although previous studies showed that a strong overlap of DMRs and differentially expressed genes (DEGs) cannot be expected (Havecker et al., 2012;Kawakatsu et al., 2016), we analysed the transcriptomes of flower buds from the same tissue used for WGBS. Triplicate strand-specific cDNA libraries of Col-0, et1-1, et2-3 and the double mutant et1-1 et2-3 yielded between 13.7 and 23.3 million short reads (107 nt), of which, after adapter and quality trimming, 7.2-12.3 million reads mapped in sense orientation onto annotated, nuclear gene models in the genome of A. thaliana (TAIR10, Table S3). PCA of normalised and mapped read counts revealed reliable separation of the mutant samples and showed that mRNA abundance of the double mutant was more similar to et2-3 than to et1-1 (Fig. 4a). DEGs were identified for pairwise comparisons between Col-0 and et mutants. In total, 337, 330 and 486 DEGs with a false discovery rate (FDR) ≤ 0.01 and an absolute log 2 fold change (lg2FC) ≥ 1 were found for the comparisons of Col-0 vs et1-1, Col-0 vs et2-3, and Col-0 vs et1-1 et2-3, respectively (Fig. 4b). The number of transcripts downregulated in mutants (et1-1, 193; et2-3, 240; et1-1 et2-3, 329) was always larger than the number of upregulated ones (et1-1, 144; et2-3, 90; et1-1 et2-3, 157). Similar to hDMRs, we detected DEGs shared between the two single mutants (142) as well as DEGs private to either et1-1 or et2-3 (195 and 188, respectively), indicating partial functional redundancy of ET1 and ET2. The 185 transcripts differentially regulated in the et1-1 et2-3 double mutant demonstrate that interactions between  (129) showed a consistent direction of change in the different lines. lg2FC values of the 129 transcripts significantly influenced in all three mutants showed that for 56 transcripts, the effects of et1-1 and et2-3 were additive, while for 72 transcripts the influence of one mutation was modulated by the other . This suggested epistatic interactions, similar to what we observed for DNA methylation effects.

Functional analysis of DEGs
A gene ontology (GO) term analysis for DEGs focusing on GOslim terms using BINGO (Maere et al., 2005) showed that the terms 'plastid' and 'thylakoid membrane' in the category 'cellular components' were significantly enriched (FDR < 0.01) in all three et mutant comparisons against Col-0 (Table 1). Also, the significant enriched terms in the category 'biological process' indicated that light-regulated and light-dependent photosynthesis was strongly affected in et mutants, even though there was no visible phenotype with respect to leaf colour and vitality. This result prompted us to inspect light-regulated processes in greater detail, which led to the discovery that certain clock and flowering time genes were differentially expressed. Because care had been taken to harvest flower samples always at the same time of the long day light-dark cycle (3-5 h after lights came on), we were able to analyse DEGs in relation to their usual diurnal peak phase of expression (Mockler et al., 2007; Fig. 5a, et1; Fig. 5b, et2; Fig. 5c, et1-1 et2-3). DEGs with a peak phase in late night (zeitgeber time (ztg) 20-22) were generally overexpressed in et mutants, while DEGs with a peak phase during early morning (ztg 3-6) were generally downregulated. This indicated a delay of the clock phases in the et mutants.
Among DEGs were several key regulators involved in the photoperiodic pathway (Table S3) Based on the focus of our study, several genes were selected for individual inspection. This includes the ET-gene family and the top 10 DEGs (Table 2). We inspected the top 10 DEGs for correlation of DNA methylation difference. hDMR and DMR lists from Table S2 were used, and the respective genes were additionally inspected for reproducible minor changes in DNA methylation among the triplicates (vDMR, visually detected differential methylated regions and single methylation polymorphisms (SMPs), visually detected single methylation polymorphisms). ET1 expression is lower than ET2 and ET2 shows a peak of expression in flower tissues. In the et1-1 T-DNA insertion line as well as in the double mutant, 0-2 reads per million (RPM) are detected which were located downstream of the insertion site, confirming the absence of functional mRNA. Consequently ET1 was found in the list of downregulated genes in et1-1. An increase of DNA methylation upstream of the second intron associated with the et1-1 T-DNA insertion was detectable (DMR1409). In the et2-3 mutant expression of ET1 was not significantly different from that in Col-0 (7-11 RPM).
In contrast to the qPCR results, ET2 was found in the top 10 lists of upregulated genes in et2-3 and the double mutant (lg2FC = 2.8). This might be caused by the integrated pROK2derived T-DNA in the used SALK_151861 line leading to 35S promotor-driven ectopic transcription (Daxinger et al., 2008). Inspection of reads and subsequent sequencing of the et2-3 ET2 gene revealed a 24 bp deletion at position 1203 in the third exon and confirmed the T-DNA insertion located in the first exon 85 bp after the start ATG. A potential alternative translation start 869 bp after start ATG of the gene might lead to expression of a truncated version without DNA cleavage domain. Therefore, absence of functional full-length ET2 mRNA in the analysed et2-3 T-DNA insertion plants could be confirmed. The differential expression was associated with the DMRs 1890 and 1891, located within the coding region of ET2 showing reduction of methylation.
One gene not present among the top 10 DEGs (sense), but being lg2FC = 4.2 up-regulated in all mutants (Fig. 6e), was associated with hDMR153 (Fig. 6f). This region is referred to as RITA (AT1G64795, encoded in antisense orientation upstream of ILYTHIA, AT1G64790 and not in the TAIR10 dataset) already described as a metastable DMR (Havecker et al., 2012).

Phenotypic characterisation of mutants
The phenotypic analysis of et mutants revealed a series of pleiotropic anomalies during plant development, similar to many other epigenetic pathway mutants (Kakutani et al., 1996;Ronemus et al., 1996).
One of the phenotypic defects observed in et mutant plants became apparent during endosperm differentiation. The
In et mutants immature seeds started to germinate as early as in the silique. While during regular germination the radicle penetrates the seed coat first, in et mutants the cotyledon emerged first along the side of the seed coat (Fig. 7c,d). No precocious germination was detected in the Col-0 control. Precocious germination of the et mutants was also observed when immature seeds were germinated in vitro (Fig. 7e,f), and the in vitro germination rate of mutants was elevated (Fig. S6). Yellow areas, daytime (light); black areas, night phase (darkness); grey areas, genes which are nonrhythmic (nr) or not detected in a previous study based on the ATH1 chip (Mockler et al., 2007). Each point represent one DEG, and change is given in log 2 -fold change (lg2FC). Colour intensity towards red indicates the level of significance of the number of DEGs for each zeitgeber (zgt) time-phase.

New Phytologist
The male gametophytes, which develop within the anther, consist of two sperm cells encased within a vegetative cell. Pollen of Col-0 and et mutants was analysed using DAPI staining (n = 200 each line). In Col-0, the vegetative nucleus and the two generative nuclei were clearly distinguishable, whereas many abnormal and collapsed pollen grains were detectable in et mutants (Figs 7l-o, S7). The female gametophyte or embryo sac develops within the ovule and consists of two synergids, one egg cell, one central cell and three antipodal cells which degenerate at the mature stage before fertilization (Drews & Koltunow, 2011). In the et mutants, the fusion of the two polar nuclei was partially compromised. The single mutants et1-1 and et2-3 showed c. 10% female gametophytes with distortions of polar nuclei fusion (n = 40 and n = 45, respectively), et2-1 c. 5% (n = 24) and et1-1 et2-1 c. 15% (n = 63; Fig. S8).
The et single and double mutant plants also exhibited reduced fertility. Seed set studies of five plants (10 siliques per plant) revealed in et1-1 and et2-3 mutants that 10% and 20% of ovules were infertile, respectively. In the et2-1 mutant as well as in the double mutant et1-1 et2-1, a striking 60% of ovules were infertile ( Fig. 7g-k).
The flowers of single and double et mutants had aberrant organ numbers, with all whorls being affected (Fig. 8). Flower morphology was quantified from 180 flowers from eight plants per genotype. In et1-1, 20% of the sepals, 60% of the stamens and 30% of the petals showed anomalies. In et2-3, 10% of the sepals, 65% of the stamens and 10% of the petals displayed defects (Fig. S9). Homeotic transformations were also found. The most frequent transformation was the conversion of the stamen into carpel-like structures (Fig. 8). The et2-1 mutant showed a homeotic transformation rate of c. 50%, whereas the double mutant et1-1 et2-1 reached c. 80%. The double mutant showed formation of multiple ovules (up to five per transformed anther) and several stigma-like structures (up to four per transformed anther) (Fig. S9). To characterise the stamen-derived ovules in more detail, the tissue was cleared and analysed by DIC microscopy. The ectopic ovules contained a normal gametophyte with fully developed egg cell, synergids and central cell. The normal polarity (synergids localized next to the micropyle, followed by egg cell and central cell) was distorted in the ectopic ovules (Fig. 8).

ET2 mutant plants accumulate mutations
Based on our hypothesis and supported by the results obtained from phenotypic inspection (enlarged nucleoli, pleiotropic phenotypes), methylation (DMRs) and transcriptional (flowering time, delay in circadian rhythm) analyses, we hypothesised that ET factors are involved in DNA damage repair.
We used the RNA-seq data to search for new mutations compared to the A. thaliana reference sequence (TAIR10). Although this approach is hampered by the fact that RNA editing appears as sequence variations (Shao et al., 2014), this method was chosen to quantify differences. Detectable sequence variation can be classified into deletions, insertions, multinucleotide variations (MNVs), replacements and single nucleotide variations (SNVs). The group of SNVs is the most abundant class of mutations. Compared to the reference genome, a similar number of SNVs were detected among the samples in the Col-0 reference plants. Interestingly, the number of detectable SNVs in the et1-1 plants was slightly lower, indicating a closer similarity to the TAIR10 reference genome, probably due to the Col-0 initially used for generation of the ET-1 T-DNA insertion. In the et2-3 plants a stronger deviation between the tested samples was recognised, with the highest accumulation of SNVs in et2-3 sample 3. A similar variation was detected in the et1-1/et2-3 samples (Fig. 9c).

Identified DMRs in the et mutants indicate similarities to mutants with impaired demethylation
Based on our hypothesis that ET factors act on DNA by singlestrand cleavage, gain of methylation is expected for ET-target regions in the ET loss of function mutants. Therefore, identified DMRs were separated for gain and loss of methylation and analysed individually. Approximately one-third of the identified hDMRs showed gain of methylation in the mutants and twothirds loss of methylation. The detectable hypomethylation might result from complex feedback regulation, also reported for ros1 (Zhu et al., 2007) and dme mutant plants (Ortega-Galisteo et al., 2008). The evolutionary neo-functionalisation of the ET2based gene duplication might explain the identified ET1-and ET2-specific DMR and DEGs.  PCA of hypomethylated hDMRs in et mutants indicated similarities with met1 and vim123 mutants (Kim et al., 2014;Shook & Richards, 2014). This observation indicated the unspecific loss of methylation in genomic regions which are under control of the DNA methylation maintenance pathway. The hypermethylated hDMRs exhibit a high similarity to DMRs detected in the genome of the rdd mutant, a triple mutant defective for ROS1 and DeMeter Like-2 and 3 (DML2 and DML3) (Penterman et al., 2007b), all involved in demethylation of DNA. Moreover, selected regions known to be affected by active demethylation such as AT1G26400, AT1G34245 and AT5G10140 were found to be hypermethylated in all sequence contexts in the et mutants. Together this implies either a function of ETs in demethylation via deregulating ROS1, DME and DML2/3 or ETs represent another novel component of the active demethylation pathway. Because DME and DMLs are not found among the differentially expressed genes in et mutants, we favour the second view. As regulation at the post-translational level or upregulation in a different context cannot be excluded, the detailed molecular mechanism remains to be investigated.

Rare overlap between DEGs and regions with strong methylation difference (hDMRs)
Using our definition of hDMRs and DEGs, an overlap of regions and expression was barely detectable: AT2G09187 (overlapping with the Athila6A retrotransposon AT2TE15880) with hDMR165, AT5G56780 (AtET2) with hDMR517 and AT5G66300 (VND3) overlapping with hDMR865. Such rare overlap between DMRs and DEGs has been reported by several other studies (Havecker et al., 2012;Kawakatsu et al., 2016). The position of the DMR, located in a gene body or promotor, together with its genomic environment make it difficult to predict whether the change in  Fig. S5(c,d).

Research
New Phytologist methylation is a cause or consequence of differential expression. The rare overlap might indicate that our criteria defining an hDMR definition are very strict. As shown in the top list of DEGs, there are small regions of differential methylation as well as a number of SMPs in the genomic environment of DEGs (Table 2). These might lead to a difference in the amount of detectable transcripts of the respective genes in the mutants.
The over-representation of detectable hDMRs associated with coding regions indicates a mechanism which might be associated with histone modifications such as H3K9 acetylation (H3K9ac) and H3K4 trimethylation (H3K4me3) (Ha et al., 2011), and these chromatin marks might also improve the recognition of putative ET binding sites. Such influence has been described for the histone acetyltransferase IDM1 which is required for demethylation (Qian et al., 2012). In addition, the identification of the NERD-pathway target genes (Pontier et al., 2012) HELITRON1 (AT1TE93275) and psORF (AT5G35935) might indicate an association of H3K4 histone modification with ET function.

Specific cases of metastable DMRs and their associated change in gene expression
Our studies identified two genic regions with a strong correlation between differential methylation and associated gene expression in the et mutants. AT1G64795 (RITA) was described as a metastable DMR (Havecker et al., 2012). The identified DNA methylation pattern was found to correspond to Col-0 6000, which was the corresponding genotype used for the SALK T-DNA collection. Although expression of this region was not reduced by transgenic reconstitution of ET1 function in the et1-1 mutant, we cannot conclude whether ETs affect this metastable locus or whether the identified loss of methylation is derived from the use of Col-0 6000 as SALK mutant background.
The second region, specific for et1-1, corresponds to the Athila6 transposon belonging to the clade of the Ty3/Gypsy family (Pelissier et al., 1995;Wicker et al., 2007;Slotkin, 2010). Transcriptional activation of Athila6A was reported for mutants of the MOM1 gene encoding a regulator of transcriptional gene silencing (Yokthongwattana et al., 2010), for mutants of the gene MORC6 which is required for heterochromatin condensation and gene silencing (Moissiard et al., 2014), and for mutants of ARABIDOPSIS TRITHORAX RELATED PROTEIN ATX5 and ATX6 (Jacob et al., 2014).
Similar to RITA, the expression level of Athila6 was not reduced by transgenic reconstitution of ET1 function. Again, we could not determine whether loss of ET function mediates a heritable, noncomplementable modification.

Indications for a function of ET factors in DNA repair
The transcriptional induction of ET1 by ionising radiation (Culligan et al., 2006) supported the initial hypothesis (Ivanov et al., 2008) that ETs are involved in DNA repair mechanisms. Inspection of the genevestigator database (Zimmermann et al., 2004) identified the nucleoside antagonist Cordycepin (3 0 -deoxyadenosine) as the strongest inducer of ET2 transcription (Fig. S11), also reported to affect the DNA damage response (Lee et al., 2012).
The enlarged nuclei detected in the endosperm of et mutants indicates activated DNA damage repair (Kobayashi, 2008;Shaw & Brown, 2012). Activation of retroelements as detected in particular for the et1-1 mutant might cause subsequent mutations. Also, the enhanced DNA damage can delay expression of circadian genes (Chung et al., 2016), which in turn might affect DNA methylation patterns and expression of downstream target genes (Chow & Ng, 2017). The reduced expression of AT4G27330 (SPL, SPOROCYTELESS; Yang et al., 1999) is a molecular feature associated with the precocious germination and incorrect orientation of the female gametophyte resulting in germination with cotyledons first. In addition, incorrect organisation of organ number and failures in development observed in correlation with the reduced expression found in et1-1 and et2-3 resemble the described phenotypes based on SPL reduction (Ito et al., 2004;Liu et al., 2009). The misexpression and phenotypic prominences could be complemented by expression of ET1 arguing for a direct effect of ET function. A stress-sensing mechanism including DNA damage regulating SPL expression has already been discussed (Zhao et al., 2017).
Here we show the accumulation of CPDs in et mutants after UV stress. As the loss of ET function resulted in an accumulation of unrepaired thymidine dimers, we propose that ET factors are involved in the mechanism of DNA repair. A similar observation was reported for ros1 and ddm1 (Questa et al., 2013). ROS1 acts as glycosylase and loss of its function also results in hyper-methylation of specific genomic regions (Morales-Ruiz et al., 2006). PCA of hyper-methylated hDMRs, representing ET sites of action, revealed a close similarity to rdd mutants, which are also affected by a disturbed DNA damage repair mechanism. Close inspection of hDMRs in the ET2 mutant revealed no specific binding sequence, but suggested the presence of two adjacent pyrimidine nucleotides. It was reported that methylated cytosines are more susceptible to UV-induced CPD formation (Martinez-Fernandez et al., 2017). Based on the structure and in planta phenotypes we propose that ETs bind at DNA regions including CPDs with preferential histone modification. Here ETs might act by DNA cleavage and by initiating DNA repair.
The identification of clock-related DEGs using GO analysis further suggests the DNA repair mechanism. The interplay of clock genes and UV-B response has already been described (Sancar et al., 2000;Thompson & Sancar, 2002;Horak & Farre, 2015). The detected delay in the circadian rhythm in et mutants is in agreements with CRY1-related repair mechanisms (Sancar et al., 2000;Thompson & Sancar, 2002). The differential expression of HYH (et1-1), a key regulator of the UV-B response (Binkert et al., 2014), and the linker Histone1-3 (AT2G18050) supports the proposed function of ETs. H1-3 is required for stress adaption on the chromatin level (Rutowicz et al., 2015). RNA interference directed against H1-3 affected the imprinting mechanism and DNA methylation (Rea et al., 2012). A search for DME downstream targets by analysis of DME overexpression revealed strong upregulation of H1-3 (Ohr et al., 2007). Therefore, downregulation of H1-3 might also contribute to detection of ET-mediated DNA methylation differences.
The increased number of detectable SNPs in et2-3, derived from the analysis of RNA sequencing data, supports the idea that ET2 acts in DNA repair. The detectable pleiotropic phenotypes, which occur at random and are heritable, might result from a higher mutation rate as a consequence of reduced DNA damage repair. In summary, all indications point toward a role of ETs as novel factors involved in DNA methylation in A. thaliana.

Supporting Information
Additional Supporting Information may be found online in the Supporting Information section at the end of the article.           Methods S1 Detailed information on RNA deep sequencing, DNA methylation analysis and Microscopy techniques.
Table S1 Primers used in this study.