The 5-methylcytosine DNA glycosylase ROS1 antagonizes parent-of-origin specific DNA methylation in Arabidopsis endosperm

  1. Elizabeth A. Hemenway
  2. Mary Gehring

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

DNA methylation patterning is a consequence of opposing activities of DNA methyltransferases and DNA demethylases. In flowering plants, two distinct female gametes, the egg cell and the central cell, are fertilized, producing what will become the embryo and the endosperm of the seed. In Arabidopsis, a 5-methylcytosine DNA glycosylase, DME, demethylates regions in the central cell genome, leading to methylation differences between maternally- and paternally-inherited endosperm genomes after fertilization. DME is required for endosperm gene imprinting. Homologues of DME include ROS1 , DML2 and DML3 . It is unknown whether any of these DNA glycosylases are required for endosperm methylation patterning. We show that ROS1 prevents hypermethylation of paternally-inherited alleles in the endosperm at regions that lack maternal or paternal-allele methylation in wild-type. Thus, ROS1 promotes epigenetic symmetry between genomes in the endosperm by preventing paternal genome hypermethylation. We propose that ROS1 and DME act in a parent-of-origin-specific manner at shared endosperm targets, and consider implications for the evolution of imprinting mechanisms.

Article activity feed

  1. Review Commons

    Note: This response was posted by the corresponding author to Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Reply to the reviewers

    Reviewer #1 (Evidence, reproducibility and clarity):

    In Arabidopsis, DNA demethylation is catalyzed by a family of DNA glycosylases including DME, ROS1, DML2, and DML3. DME activity in the central cell leads to the hypomethylation of maternal alleles in endosperm. While ROS1, DML2, and DML3 function in vegetative tissues to prevent spreading DNA methylation from TE boundaries, their function in the endosperm was unclear.
    Using whole genome methylome analysis, the authors showed that ROS1 prevents hypermethylation of paternal alleles in the endosperm thus promotes epigenetic symmetry between maternal and paternal genomes.
    The approach and experimental desighs are appropriate, and the key conclusions are adequately supported by the results.
    However, there is not sufficient evidence to support the claim that DME demethylates the maternal allele at ROS1-dependent biallelically-demethylated regions. To clarify the issue, the authors could analyze if there is an overlap between DMRs identified in ros1 endosperm and those identified in dme endosperm using published data. If there is any, the authors could show a genome browser example of DMR including dme data.

    Response: Thank you for your insight on our work. To address your concern and further test our model that DME prevents methylation of the maternal allele at regions where ROS1 is prevents methylation of the paternal allele, we turned to the allele-specific bisulfite-sequencing data published in Ibarra et al 2012. These data were from endosperm isolated at 7-8 DAP from aborting seeds of dme-2 +/- (Col-gl) plants pollinated by L_er_. Our analysis of these data is now included in Figures 6 and 7 and Supplemental Figures 13-17. We show that when the loss-of-function allele dme-2 is inherited maternally, average methylation of the maternal allele increases at ROS1-dependent regions (in the revised version of the paper now referred to as ROS1 paternal, DME maternal regions) from less than 10% CG methylation to approximately 40% CG methylation (Fig. 6D), consistent with our previous analysis using the non-allelic Hsieh et al 2009 data (now moved to Supplemental Figure 15). These results thus provide additional evidence that DME removes maternal allele methylation at regions where ROS1 removes paternal allele methylation (compare Fig. 6B and 6D). We included relevant genome browser examples in Figure 7E and Supplemental Figure 14. In the revised version, the relationship between ROS1 and DME is further expanded upon in the text.

    Reviewer #1 (Significance):

    Endosperm is a tissue unique to flowering plants. Though it is an ephemeral tissue, the endosperm plays essential roles for seed development and germination. The endosperm is also the site genomic imprinting occurs, and it has a distinct epigenomic landscape. This work provides a new insight that ROS1 may antagonize imprinted gene expression in the endosperm. However, it was not shown whether imprinted gene expression is indeed affected in ros1, or whether the ros1 mutation has phenotypic consequences. These results would be useful to discuss the evolution and significance of genomic imprinting.

    Response: We agree that the biological significance of ROS1-mediated paternal allele demethylation is presently unknown. We performed RNA-seq on wild-type and ros1 3C and 6C endosperm nuclei, but these data were unfortunately not of high enough quality to include in the manuscript. In the Discussion we suggest that disrupting ROS1-mediated paternal allele demethylation might lead to a gain of imprinting over evolutionary time. In future work we are planning to address potential relationships to gene imprinting using a molecular, RNA-sequencing approach as well as an evolutionary comparative approach. As expected, given the expectation that imprinted genes are associated with a parent-of-origin specific epigenetic mark, we did not find any relationship between known imprinted genes and ROS1-dependent regions that are biallelically-demethylated regions in wild-type endosperm (see lines 362-372).

    Reviewer #2 (Evidence, reproducibility and clarity):

    SUMMARY

    Hemenway and Gehring present evidence that the paternal genome in Arabidopsis endosperm is demethylated at several hundred loci by the DNA glycosylase/lyase ROS1. The evidence is primarily based on analysis of DNA methylation of ros1 mutants and of hybrid crosses where each parental genome can be differentiated by SNPs. I have some comments/questions/concerns, two of them potentially serious, but I think Hemenway and Gehring can address them through additional analyses of data that they already have available and a bit of clarification in writing.

    Response: Thank you for your thoughtful review of this study. Your insight and suggestions have helped add clarity to the paper.

    MAJOR COMMENTS:

    1. Could the excess methylation in ros1-3 relative to ros1-7 shown in Figures 1A and 1C be explained by a second mutation in the ros1-3 background that elevates methylation at some loci? Any mutation that increased RdDM at these loci, for example could have this effect. This could confound the identification and interpretation of biallelicly demethylated loci.

    Response: We propose a simpler explanation for the additional hypermethylation observed in ros1-3: ros1-3 is a loss-of-function (null) allele whereas ros1-7 is likely a hypomorphic allele. For clarity, we have added a diagram of all of the alleles used in this study as Supplemental Figure 1B. The ros1-3 allele was first described in Penterman et al, PNAS, 2007. It is a T-DNA insertion allele that was isolated in the Ws accession and then backcrossed 6 times to Col-0, greatly minimizing the risk of unlinked secondary mutations being present. There is no genetic evidence that there is another T-DNA insertion in this line. The ros1-7 allele was described in Williams et al, Plos Genet, 2015. It was isolated from the Arabidopsis Col-0 TILLING population and is missense mutation (E956K) in a residue in the glycosylase domain that is conserved among the four DNA glycosylases. It is known that ROS1 transcripts are produced from the ros1-7 allele (Williams et al 2015). We observe less hypermethylation in the ros1-7 background compared to the ros1-3 background, and thus propose that the ros1-7 allele is a hypomorphic allele of ROS1. The use of two independent ros1 mutant alleles for initial endosperm methylation profiling strengthens the findings of our study. Importantly, regions that are hypermethylated in ros1-3 are also hypermethylated in ros1-7, but to a lesser extent, and vice versa (Fig 1D, Supplemental Figs. 3 and 4).

    We also use a third allele in this study, ros1-1, which is a nonsense allele in the C24 accession. Notably, we find that the regions are demethylated on both maternal and paternal alleles in wild-type C24 gain DNA methylation primarily on the paternal allele in ros1-1 endosperm (Figure 4C,D and Supplemental Figure 10). This is discussed further in response to your second point.

    Given these lines of evidence, a gain-of-function mutation in a methylation pathway, like RdDM, in the ros1-3 background is an unlikely explanation for increased hypermethylation compared to ros1-7. The use of three independent ros1 alleles for methylation profiling, all of which lead to the same conclusions, is a major strength of our study.

    1. It appears that the main focus of the manuscript, the existence of loci that are paternally demethylated by ROS1, is supported by a set of 274 DMRs. This is a small number relative to the size of the genome and raises suspicions of rare false positives. Even the most stringent p-values that DMR-finding tools report do not guarantee that the DMRs are actually reproducible in an independent experiment. Demonstrating overlap between these 274 DMRs and an independently defined set using a different WT control and different ros1 allele would suffice to remove this concern. It appears that authors already have the needed raw data with ros1-1 and ros1-7 alleles.

    Response: First, we should clarify that paternal demethylation by ROS1 is supported by more than the 274 DMRs. All ros1 CG hyperDMRs show an increase in paternal allele methylation in ros1 (Fig. 4B,D). The 274 DMRs are a distinct subset defined as having less methylation on the maternal allele than the paternal allele in ros1 endosperm and where there is no maternal allele hypomethylation in wild-type endosperm (refer to Fig. 5B).

    We agree with your sentiments about DMR-finders and we are cautious of relying exclusively on DMR calls when making conclusions. We verify the nature of identified DMRs using metaplots and weighted average comparisons throughout the paper, which we think increases confidence in the conclusions and goes beyond a simple DMR-calling approach.

    We argue that we have replicated the major conclusion of the paper, that ROS1 prevents paternal allele hypermethylation at target regions in the endosperm, in the following ways:

    1. In the dataset without allelic-specific methylation information (Figures 1-3), we found that both ros1-3 and ros1-7 CG hyperDMRs have a limited capacity for hypermethylation in the endosperm relative to leaf or sperm (Table 1, Fig 3, Supplemental Fig. 4). In the allele-specific dataset, ros1-3 CG hyperDMRs were revealed to have particularly low maternal mCG relative to paternal mCG in ros1 mutant endosperm (Fig 4A-B, Supplemental Fig. 10).
    2. We found that ros1-3 and ros1-1 hyperDMRs, which we identified using non-allelic data, are biased for paternal allele hypermethylation in the endosperm of F1 hybrids (Fig 4B,D). The replicability of the paternal bias in hypermethylation in both ros1-3 in the Col-0 ecotype and ros1-1 in the C24 ecotype is a critical result, and we have moved the ros1-1 hyperDMR plots from the supplement to main figure 4C-D in the revised version of the manuscript as a result of your comment.
    3. The 274 DMRs identified as “biallelically-demethylated, ROS1-dependent” are by definition replicated between reciprocal cross directions. (Note that we now refer to these regions as ROS1 paternal, DME maternal regions in the revision.) Regions in this category had to be called as maternally-hypomethylated in both ros1-1 x ros1-3 and ros1-3 x ros1-1 endosperm. These regions also had to not be identified as maternally-hypomethylated in both C24 x Col-0 and Col-0 x C24. We hope this is clarified for readers by Table 1, which we have included based on your suggestion in comment #3, as well as other clarifying edits we made in this section of the paper.comparisons between maternal and paternal methylation in endosperm, DMRs defined by comparison between mutants and wildtype, and more. These need clearer descriptions of which sets are being referred to throughout the main text and in figure legends. A table summarizing them might help (not in the supplement). Use of consistent and precisely defined terms would help. Stating the number of DMRs along with the name for each set would help a lot, even though this would make for some redundancy. (The number of DMRs in each set not only helps with interpretation but also act as a sort of ID). The reason I put this as a major concern is because the text and figures are difficult to understand, and it is currently hard to evaluate both the results and the authors' conclusions from those results.

    Response: Thank you for your feedback and suggestions. We have edited the main text so that only one descriptive name is used for each DMR type throughout the paper. We have also renamed regions for greater clarity. The previous “ROS1-independent, maternally demethylated regions” are now referred to as “DME maternal regions”. The previous “_ROS1-_independent, biallelically-demethylated regions” are now referred to as “ROS1 paternal, DME maternal regions”. These changes provide greater clarity and also emphasize the role of DME at regions that are paternally hypermethylated in ros1. We have added Table 1 to summarize the DMR classes of interest.

    MINOR COMMENTS

    1. The sRNA results in Figure 2B are difficult to interpret because they do not reveal anything about the number of TEs that have siRNAs overlapping them or their flanks. While the magnitude of some of the highest endosperm sRNA peaks is higher than the embryo peaks, that could be explained by a small number of TEs with large numbers of sRNAs. To make this result more interpretable, we also need some information about how many TEs have a significant number of sRNAs associated with them in endosperm and embryo in each region (e.g., middle, 5', 3', and flanks of TEs). What a "significant number of sRNAs" is would be up to the authors to decide based on the distribution of sRNA counts they observe for TEs. Perhaps the top quartile of TEs? Combined with the same analysis done in parallel with non-ROS1 target TEs, this would reveal whether there is any evidence for ROS1 counteracting sRNA-driven methylation spread from TEs.

    Response: Thank you for the suggestion. We now present these data and the data for individual TEs underlying the metaplots in Supplemental Figure 7. As suggested by the reviewer, ROS1 TEs do not have uniformly higher levels of sRNA in their flanks in the endosperm compared to the embryo. We have modified our interpretations accordingly.

    1. The statement "we are likely underestimating the true degree of differential methylation among genotypes" should be validated and partially quantified using a methylation metaplot like Figure 2A, but substitute DMRs for TEs. Related to that, Figure 1B needs an indicator of scale in bp.

    Response: We have now included a methylation metaplot over ros1-3 hyperDMRs and ros1-7 hyperDMRs as Supplemental Figure 3 These plots show that indeed there is additional hypermethylation in DMR-proximal regions. We have added a scale bar to Figure 1B and other browser examples in the paper.

    1. The statement "Over half of ROS1 target regions identified in the ros1-3 mutant endosperm were within 1 kb or intersecting a TE (Fig. 1D)" is hard to interpret without some kind of ROS1 non-target regions or whole-genome control comparison. How different are the numbers in Fig. 1D from a random expectation?

    Response: We have now included a control for random regions in Figure 1E. We define these as regions where there was sufficient methylation data coverage and a low enough methylation level in wild-type to detect hypermethylation if it existed.

    1. The sentence at line 262 is confusing. Is the comparison between dme mutant and ros1 mutant or between different types of regions? And it appears that the comparison value is missing in the "3-5% CG methylation gain..." e.g., "3-5% CG methylation vs 10-20%" or something like that.

    Response: This section has been re-written as we now focus on allele-specific dme endosperm methylation data for our comparisons.

    1. The dme mutant data in Figure 5C appear to be key to the model in Figure 7. The relative impact of the dme mutant in the two types of regions should be quantified.

    Response: Thank you for this comment. To further probe our model that DME prevents hypermethylation of the maternal allele at regions where ROS1 is preventing hypermethylation of the paternal allele, we turned to the allele-specific bisulfite-sequencing data published in Ibarra et al 2012 (see also response to reviewer #1). Using these data, we show that when the loss-of-function allele dme-2 is inherited maternally, ROS1 paternal, DME maternal regions (previous referred to as _ROS1-_dependent, biallelically-demethylated regions) are CG hypermethylated on the maternal allele (Figure 6D). Thus, these results both replicate the observations made with the Hsieh et al 2009 data, and provide additional evidence that DME prevents maternal allele hypermethylation at regions were ROS1 is preventing paternal allele hypermethylation. These results have replaced the Hsieh et al 2009 results in Figure 6, and we have moved the analysis of Hsieh et al 2009 data to Supplemental Figure 15.

    1. Looks like sRNA methods are missing.

    Response: Thank you for identifying this. We previously included the reference for the analyzed dataset we used and the method for plotting under an unclear section header. These methods are now in the section “Analysis of average methylation and 24-nt sRNA patterns for features of interest”, and we have added additional reference to the specific dataset we used.

    1. Supplemental Figure 1 is hard to interpret since it only list gene IDs, not gene names.

    Response: As suggested, we have added gene names to this figure.

    The last comments are suggestions for increasing the impact of this study:

    1. Figure 2A and 3B suggest that ROS1 target TEs show demethylation in their flanks but not in the TE themselves. This is an interesting result. If it is true, more DMRs would be expected in the ROS1 target flanks than in the ROS1 target TEs. Reporting how many ROS1 target TEs have DMRs in them and what proportion have DMRs in their flanking 1-Kb regions would answer this question. Given the significance of this result, it also deserves a bit more context: Is the magnitude of increased methylation flanking TEs in ros1 mutant endosperm different than in ros1 mutant leaves or other tissue? Does methylation in TE flanks behave the way in dme mutant endosperm?

    Response: We define “ROS1 target TEs” (now referred to more simply as ROS1 TEs) as TEs within 1kb or intersecting a ros1-3 hyperDMR. Consistent with your interpretation, 80% of the TEs in this category do not have a DMR overlapping them, instead they have a TE within 1kb. We now mention this in the text on line 150.

    The total level of DNA methylation at ROS1 TEs is lower in the endosperm than in leaf, as DNA methylation levels are overall lower in endosperm than in leaf. The magnitude of increased methylation flanking TEs in ros1 mutant endosperm is not different between the two tissues. This is observable in Supplemental Fig. 5 in the revised version of the paper, and we report this result in the revised text. In the revision we also present methylation profiles of DME TEs in WT and ros1 endosperm (Fig. 7B-D). DME TEs are hypomethylated in both the body and flanks in WT and ros1.

    1. The idea of biallelic demethylation has been theoretically suggested in maize to explain weak overlap between endosperm DMRs and imprinting (Gent et al 2022). If that were true in Arabidopsis, then ROS1 target, biallelicly demethylated loci would be less likely to have imprinted expression than maternally demethylated loci. This prediction could be tested using available data in Arabidopsis.

    Response: Indeed, as you hypothesize, there are no known imprinted genes (Pignatta et al 2014) associated with biallelically-demethylated, ROS1-dependent regions (now referred to as ROS1 paternal, DME maternal regions). Expectedly, there are imprinted genes associated with maternally-demethylated regions (now referred to as DME regions). 23 imprinted genes identified in the Pignatta et al 2014 study are within 1 kb or intersecting a DME region. This is discussed on lines 364-374.

    1. There is currently no evidence for biological significance of biallelicly demethylated loci. Knowing where they are in the genome might give some hints. A figure like Fig. 1D but specifically showing the biallelicly demethylated DMRs would be valuable.

    Response: This is now included in Figure 7A.

    1. It is hard to make the comparisons between genotypes and parental genomes in Figure 6 and know what they mean. Maybe a different way of displaying the data would help. Or maybe even a different labeling system could make it a little more accessible.

    Response: We have revised this figure (now Fig. 8) in the following ways, which we believe address your comments and clarify the main conclusions:

    Figure 8C is now a boxplot comparing methylation of the paternal allele of ROS1 paternal, DME maternal regions (previously referred to as biallelically-demethylated, ROS1-dependent regions) across endosperm ROS1 genotypes. This plot shows increased methylation of paternal alleles when the paternal parent is a ros1 mutant, regardless of whether the resultant F1 endosperm is homozygous or heterozygous for ros1 (columns 3, 4, 6).

    Figure 8B remains as a scatterplot, where we can observe significant correlation between individual ROS1 paternal, DME maternal regions in homozygous ros1 endosperm and heterozygous ros1/+ endosperm. Note that paternal allele methylation is higher in homozygous ros1 endosperm for most regions.

    Reviewer #2 (Significance):

    Demethylation of the maternal genome in endosperm has been the subject of much research because it can result in genomic imprinting of gene expression. The enzymes responsible, DNA glycosylases/lyases, also demethylate DNA in other cell types as well, where DNA methylation is not confined to one parental genome (biallelic or biparental as opposed to uniparental demethylation). To the best of my knowledge, the extent or even existence of biallelelic demethylation in endosperm has not been studied until now (except for a superficial look in a bioRxiv preprint, https://www.biorxiv.org/content/10.1101/2024.07.31.606038v1). Hemenway and Gehring have carried out a thoughtful and detailed analysis of the topic in Arabidopsis at least as far as it depends on the DNA glycosylase ROS1.

    A limitation is that the study design would miss biallelic demethylation by any of the other three DNA glycosylases in Arabidopsis. A second limitation is that there is no clear biological significance, just some conjecture about evolution. Nonetheless, given the novelty of the topic, biological significance may follow.

    The audience for biallelic DNA demethylation in Arabidopsis endosperm is certainly in the "specialized" category, but its relevance to the larger topic of gene regulation in endosperm will attract a larger audience.

    Response: With regard to the other demethylases, note that we also profiled methylation in ros1 dml2 dml3 triple mutant endosperm. We did not find evidence for many DMRs that were present in the triple mutant that were not present in the ros1 single mutant. We do not rule out a function for DML2 or DML3 in the endosperm, but this is not observed at the level of bulk endosperm.

    The reviewer is correct that we have shown a molecular phenotype (paternal allele hypermethylation) and not a developmental or morphological phenotype. A function that occurs in one parent but not the other is, to us, exciting. Our thoughts about how this finding might relate to imprinting are indeed speculative, but not wildly so.

    Reviewer #3 (Evidence, reproducibility and clarity):

    DNA demethylases play a key role in DNA methylation patterning during flowering plant reproduction. The demethylase DME, in particular, is critical for proper endosperm development. While the function of DME in endosperm development has been explored, the contributions of the other demethylases in the same family, ROS1, DML2 and DML3 in Arabidopsis, have not yet been investigated. In vegetative tissues, ROS1 prevents hypermethylation of some loci. In this work, Hemenway and Gehring explore whether ROS1, DML2 and DML3 also affect DNA methylation patterns in endosperm. Using EM-seq of sorted endosperm nuclei, they show that loss of ROS1 indeed causes hypermethylation of a number of loci, particularly the flanks of methylated transposons, while loss of DML2 and DML3 has minimal additional effect. By obtaining allele-specific EM-seq data through crosses of Col and C24, the authors show that ros1 endosperm hypermethylation is mostly restricted to the paternal allele. The authors propose that at some sites, ROS1 helps bring down paternal methylation levels to match maternal methylation levels, which are typically reduced in endosperm due to DME activity in the female gametophyte prior to fertilization. In a ros1 mutant with paternal hypermethylation, these sites become differentially methylated on the maternal and paternal alleles, resembling imprinted loci. This work convincingly establishes a function for ROS1 in DNA methylation patterning in endosperm. However, I struggled with the clarity of the writing and reasoning in a few places, and would suggest clarification of a few points and additional analyses below.

    Response: Thank you for your thoughtful review of our paper. Your questions and suggestions have been invaluable in revising the work.

    I think making a few simple changes to streamline nomenclature would improve readability. For example, in the section starting on line 129, the same set of genomic features are called ROS1 target-proximal TEs, TEs that are near a ROS1 target region, and ROS1 target-associated TE regions. Also for example in line 254 "regions that are maternally-demethylated in wild-type endosperm, and are not dependent on ROS1 for proper demethylation" - are these the same as the "ROS1-independent, maternally-demethylated" regions in Fig. 5a? Given how complex these terms are, being consistent throughout the manuscript really helps the reader.

    Response: We edited the text and figures so that only one descriptive name is used for each DMR class or region throughout the paper. Thank you for this feedback; these edits have made the paper much clearer.

    Is there any notable effect of ros1 on gene expression in endosperm? Endosperm is a terminal tissue, so maintaining DNA methylation boundaries as ROS1 does in vegetative tissues seems less important. It begs the question of why ROS1 is doing this in endosperm, is it just because it's there, or is there an endosperm-specific function? Exploring effects on imprinting would be particularly interesting (does loss of ROS1 'create' imprinted loci at these newly asymmetrically methylated sites?) but probably beyond the scope of the present work.

    Response: We agree, the question of the functional consequence of ROS1 activity in the endosperm is something we are keen to address in future work. We performed RNA-seq on wild-type and ros1 3C and 6C endosperm nuclei, but these data were unfortunately not of high enough quality to include in the manuscript. We are in particular interested in this question you have proposed – if loss of ROS1 can ‘create’ imprinted loci. We are planning to address this both using a molecular, RNA-sequencing approach as well as an evolutionary comparative approach. This is an important and exciting future direction.

    Is DME expressed in sperm, or is expression of DME affected in ros1 sperm or endosperm? One other explanation for ros1 hypermethylation occurring primarily on the paternal allele is that, potentially, DME can substitute for ROS1 in the central cell where DME is already very active, but not in sperm cells. Related, how well expressed is ROS1 vs. DME in sperm cells?

    Response: This is an important series of questions, and something we are very interested in as well. Studies of Arabidopsis pollen have shown that both ROS1 and DME, while they prevent some hypermethylation in sperm, are more active in the vegetative nucleus of pollen than in sperm. ROS1 is expressed at a low level in the microspore and bicellular pollen and DME is expressed at a low level throughout pollen development. We have included Supplemental Fig. 17 with available expression data to make this point in the paper. Likely, any effects of loss of ROS1 or DME on sperm DNA methylation are inherited from precursor cells (Ibarra et al 2012, Calarco et al 2012, Khouider et al 2021). Your proposal that perhaps DME can sub in for ROS1 in the central cell but not in sperm is intriguing. Unfortunately there’s not enough data in the central cell to convincingly address this at this time.

    To investigate the relationship between DME and ROS1 in the male germline, we used the bisulfite-sequencing data generated in sperm cells in Khouider et al 2021. We calculated average DNA methylation levels in dme/+, ros1, dme/+;ros1, and wild-type Col-0 sperm cells at ROS1 paternal, DME maternal regions, shown in Supplemental Fig. 18A. We observed little increase in mCG methylation in dme/+ sperm relative to wild-type Col-0 sperm. This is consistent with your proposed model that DME is unable to demethylate these regions outside of the female germline. As expected, there is increased mCG in ROS1 paternal, DME maternal regions in ros1-3 mutant sperm relative to wild-type Col-0 sperm. DME maternal regions are highly methylated in wild-type Col-0 sperm.

    Fig 2b shows that ROS1 target-associated TEs are enriched for sRNAs in endosperm relative to embryo, whereas the reverse is true for non-ROS1-assoc TEs. Since TEs are not always well annotated and some may be missing from this analysis, what about trying the reverse analysis - are regions enriched for 24nt sRNAs in endosperm significantly hypermethylated in ros1 endosperm? All regions or only some?

    Response: We performed an analysis to address your inquiry and observed a low magnitude increase in DNA methylation in ros1 mutant endosperm at regions defined by Erdmann et al as more sRNA producing in the endosperm relative to the embryo (endosperm DSRs). Endosperm DSRs are generally lowly methylated in wild-type endosperm, as was observed originally in Erdmann et al 2017. Small increases in DNA methylation are observed at endosperm DSRs in all sequence contexts in ros1 endosperm. Overall, this is consistent with ROS1 targets being a subset of sRNA-producing regions in the endosperm. This analysis is now included in Supplemental Fig. 7C.

    What is the relationship between previously-defined DME targets and ROS1 targets identified in this paper? DME tends to target small euchromatic TE bodies, whereas Fig. 3 suggests that ROS1 helps prevent methylation spreading on the outer edges of the TEs, rather than in the TE body. Do all DME targets tend to be adjacent to or flanked by ROS1 target sites? Or are the TEs affected by DME (in body) and by ROS1 (at edges) largely nonoverlapping? Fig. 5a suggests that the ROS1-dependent, biallelically-demethylated sites are both DME and ROS1 targets, but how often do these really appear to overlap? More than by chance?

    Response: We have sought to address your comments through a series of analyses that we have included in Fig. 7 and Supplemental Fig. 16. We found that ROS1 paternal, DME maternal regions (formerly referred to as ROS1-dependent, biallelically-demethylated regions) and DME maternal regions (formerly referred to as ROS1-independent, maternally-demethylated regions) do not occupy the same genomic regions. However, we do observe some evidence for ROS1 activity in flanking regions of DME targets (Fig. 6A, Fig. 7B-D). To look at TEs specifically, as you suggest, we first identified TEs that were within 1kb or intersecting a DME maternal region. Based on our characterization of these regions, we assume these to be DME-targeted TEs. We then performed ends analysis to see if there was evidence of ROS1 activity at the ends of these TEs. Indeed, at a global level there is a slight hypermethylation of the paternal allele in a ros1 mutant at the end of these DME TEs (Fig. 7B). To better visualize how many DME TEs are showing ROS1 activity at their ends, we then plotted the difference between the median ros1-3 methylation and median Col-0 values in the non-allelic endosperm for each TE in a clustered heatmap (Fig. 7C). The parent-of-origin data does not have enough coverage for clustering in this way, so we used the non-allelic data. A small fraction of “DME TEs” gain methylation in the ros1 mutant endosperm relative to wild-type (Fig. 7C-D).

    Are the TEs whose boundaries are demethylated by ROS1 more likely to be expressed in vegetative or endosperm tissues than TEs not affected by loss of ROS1? Expressed TEs likely produce more sRNAs, which would increase RdDM in a way that might need to be more actively countered by ROS1 than transcriptionally silent or evolutionarily older TEs.

    Response: This is an interesting line of inquiry, although perhaps out of the scope of our present study. It has been shown that TEs demethylated by ROS1 are targeted by the RdDM pathway in Arabidopsis vegetative tissue (Tang et al 2016). Using data from Erdmann et al 2017, we looked at 24 nt sRNAs at ROS1-TEs in the endosperm and embryo (Supplemental Fig. 7). sRNA production at ROS1 TE-flanking regions is observed in both embryo and endosperm, but clearly not all ROS1 TEs produce 24 nt sRNA production in the seed. Future work comparing sRNA profiles in a ros1 mutant to those of wild-type could inform our understanding of TE spreading in a ros1 mutant, as would a comprehensive analysis of TE expression, again in both a ros1 mutant and in wild-type. It’s unclear to us if the endosperm would be the most informative or useful tissue to perform such analyses in.

    Fig6 - as noted in the text, one way to test whether demethylation by ROS1 occurs before or after fertilization is to provide functional ROS1 through only one parent via reciprocal WT x ros-1 crosses, so that the endosperm always has ROS1 but either sperm or central cell does not, and see if this can rescue the paternal hypermethylation. If ROS1 acts prior to fertilization, then paternal ROS1 will rescue ros1 hypermethylation, but maternal ROS1 won't. If after fertilization, then either maternally or paternally supplied ROS1 will rescue the hypermethylation phenotype (assuming both are well expressed). Thus, to distinguish the two, it is sufficient to test whether maternally supplied ROS1 in an otherwise mutant background can rescue the hypermethylation phenotype, which is what is shown in Fig. 6. However, I think it's also important to show that paternally supplied ROS1 can also rescue the hypermethylation phenotype, which is not currently shown. The plots showing no effect on maternal mCG aren't as informative, since maternal methylation levels are mostly unaffected by ros1 anyway. Instead of comparing pairs of samples in a scatterplot, it might be clearer to show paternal mCG across all four comparisons (WT x WT, WT x ros1, ros1 x WT, and ros1 x ros1) side by side in a heatmap, using clustering to group similar behavior.

    Response: We have revised this figure, now Fig. 8, in the following ways, which we believe addresses your comments and clarify the main conclusions (see same response to reviewer 2 for point 14):

    Figure 8B remains as a scatterplot, where we observe significant correlation between individual ROS1 paternal, DME maternal regions in homozygous ros1 endosperm and heterozygous ros1/+ endosperm. Note that paternal allele methylation is higher in homozygous ros1 endosperm for most regions.

    Figure 8C is now a boxplot comparing methylation of the paternal allele of ROS1 paternal, DME maternal regions (previously referred to as biallelically-demethylated, ROS1-dependent regions) across endosperm ROS1 genotypes. This plot shows increased methylation of paternal alleles when the paternal parent is a ros1 mutant, regardless of whether the resultant F1 endosperm is homozygous or heterozygous for ros1 (columns 3, 4, 6).

    I would also suggest including a little more information in the main plots rather than only in the figure legends. For example, in Fig 2 including a label of 'ROS1-associated TE' for the two plots on the left, and 'TEs not associated with ROS1' on the right. Or for example in Fig. 3a indicating 'ros1-3 CG hyperDMRs' somewhere on the plot. This would just help make the figures easier to read at a glance. Please add common gene names to figures, instead just the ATG gene ID (Fig. S1a).

    Response: Thank you for this feedback, we have made the suggested edits and additional edits of a similar nature.

    Minor:

    • Fig. 1E is referenced in the text before Fig. 1D
    • Fig. S4 and S5 - there are more lines in the plot than the 6 genotypes listed in the legend, do these represent different replicates? If so that should be noted in the legend
    • Fig. 1B has no color legend for the different methylation sequence contexts (looks like same as 1A,C but should indicate either in plot or legend)
    • Line 42 should be "correspond to TE ends"
    • Line 93 "Based on previous studies..." should have references to those studies
    • When referring to the protein (rather than the genetic locus or mutant), ROS1 should not be italicized - for example line 130
    • Line 150 "we conclude that the loss"
    • Should add a y=x line to scatterplots, like those in Fig. 6
    • In fig. 1d, it's hard to evaluate the significance of the overlap of ROS1 targets with genes and TEs. Comparing these numbers to a control where the ROS1 targets have been randomly shuffled would help.

    Response: We have made edits and additions where requested.

    Reviewer #3 (Significance):

    In this work, Hemenway and Gehring explore whether ROS1, DML2 and DML3 also affect DNA methylation patterns in endosperm. Using EM-seq of sorted endosperm nuclei, they show that loss of ROS1 indeed causes hypermethylation of a number of loci, particularly the flanks of methylated transposons, while loss of DML2 and DML3 has minimal additional effect. By obtaining allele-specific EM-seq data through crosses of Col and C24, the authors show that ros1 endosperm hypermethylation is mostly restricted to the paternal allele. The authors propose that at some sites, ROS1 helps bring down paternal methylation levels to match maternal methylation levels, which are typically reduced in endosperm due to DME activity in the female gametophyte prior to fertilization. In a ros1 mutant with paternal hypermethylation, these sites become differentially methylated on the maternal and paternal alleles, resembling imprinted loci. This work convincingly establishes a function for ROS1 in DNA methylation patterning in endosperm. However, I struggled with the clarity of the writing and reasoning in a few places, and would suggest clarification of a few points and additional analyses.

    Response: Thank you for your comments. We have worked on streamlining the text and analysis.

  2. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #3

    Evidence, reproducibility and clarity

    DNA demethylases play a key role in DNA methylation patterning during flowering plant reproduction. The demethylase DME, in particular, is critical for proper endosperm development. While the function of DME in endosperm development has been explored, the contributions of the other demethylases in the same family, ROS1, DML2 and DML3 in Arabidopsis, have not yet been investigated. In vegetative tissues, ROS1 prevents hypermethylation of some loci. In this work, Hemenway and Gehring explore whether ROS1, DML2 and DML3 also affect DNA methylation patterns in endosperm. Using EM-seq of sorted endosperm nuclei, they show that loss of ROS1 indeed causes hypermethylation of a number of loci, particularly the flanks of methylated transposons, while loss of DML2 and DML3 has minimal additional effect. By obtaining allele-specific EM-seq data through crosses of Col and C24, the authors show that ros1 endosperm hypermethylation is mostly restricted to the paternal allele. The authors propose that at some sites, ROS1 helps bring down paternal methylation levels to match maternal methylation levels, which are typically reduced in endosperm due to DME activity in the female gametophyte prior to fertilization. In a ros1 mutant with paternal hypermethylation, these sites become differentially methylated on the maternal and paternal alleles, resembling imprinted loci. This work convincingly establishes a function for ROS1 in DNA methylation patterning in endosperm. However, I struggled with the clarity of the writing and reasoning in a few places, and would suggest clarification of a few points and additional analyses below.

    I think making a few simple changes to streamline nomenclature would improve readability. For example, in the section starting on line 129, the same set of genomic features are called ROS1 target-proximal TEs, TEs that are near a ROS1 target region, and ROS1 target-associated TE regions. Also for example in line 254 "regions that are maternally-demethylated in wild-type endosperm, and are not dependent on ROS1 for proper demethylation" - are these the same as the "ROS1-independent, maternally-demethylated" regions in Fig. 5a? Given how complex these terms are, being consistent throughout the manuscript really helps the reader.

    Is there any notable effect of ros1 on gene expression in endosperm? Endosperm is a terminal tissue, so maintaining DNA methylation boundaries as ROS1 does in vegetative tissues seems less important. It begs the question of why ROS1 is doing this in endosperm, is it just because it's there, or is there an endosperm-specific function? Exploring effects on imprinting would be particularly interesting (does loss of ROS1 'create' imprinted loci at these newly asymmetrically methylated sites?) but probably beyond the scope of the present work.

    Is DME expressed in sperm, or is expression of DME affected in ros1 sperm or endosperm? One other explanation for ros1 hypermethylation occurring primarily on the paternal allele is that, potentially, DME can substitute for ROS1 in the central cell where DME is already very active, but not in sperm cells. Related, how well expressed is ROS1 vs. DME in sperm cells?

    Fig 2b shows that ROS1 target-associated TEs are enriched for sRNAs in endosperm relative to embryo, whereas the reverse is true for non-ROS1-assoc TEs. Since TEs are not always well annotated and some may be missing from this analysis, what about trying the reverse analysis - are regions enriched for 24nt sRNAs in endosperm significantly hypermethylated in ros1 endosperm? All regions or only some?

    What is the relationship between previously-defined DME targets and ROS1 targets identified in this paper? DME tends to target small euchromatic TE bodies, whereas Fig. 3 suggests that ROS1 helps prevent methylation spreading on the outer edges of the TEs, rather than in the TE body. Do all DME targets tend to be adjacent to or flanked by ROS1 target sites? Or are the TEs affected by DME (in body) and by ROS1 (at edges) largely nonoverlapping? Fig. 5a suggests that the ROS1-dependent, biallelically-demethylated sites are both DME and ROS1 targets, but how often do these really appear to overlap? More than by chance?

    Are the TEs whose boundaries are demethylated by ROS1 more likely to be expressed in vegetative or endosperm tissues than TEs not affected by loss of ROS1? Expressed TEs likely produce more sRNAs, which would increase RdDM in a way that might need to be more actively countered by ROS1 than transcriptionally silent or evolutionarily older TEs.

    Fig6 - as noted in the text, one way to test whether demethylation by ROS1 occurs before or after fertilization is to provide functional ROS1 through only one parent via reciprocal WT x ros-1 crosses, so that the endosperm always has ROS1 but either sperm or central cell does not, and see if this can rescue the paternal hypermethylation. If ROS1 acts prior to fertilization, then paternal ROS1 will rescue ros1 hypermethylation, but maternal ROS1 won't. If after fertilization, then either maternally or paternally supplied ROS1 will rescue the hypermethylation phenotype (assuming both are well expressed). Thus, to distinguish the two, it is sufficient to test whether maternally supplied ROS1 in an otherwise mutant background can rescue the hypermethylation phenotype, which is what is shown in Fig. 6. However, I think it's also important to show that paternally supplied ROS1 can also rescue the hypermethylation phenotype, which is not currently shown. The plots showing no effect on maternal mCG aren't as informative, since maternal methylation levels are mostly unaffected by ros1 anyway. Instead of comparing pairs of samples in a scatterplot, it might be clearer to show paternal mCG across all four comparisons (WT x WT, WT x ros1, ros1 x WT, and ros1 x ros1) side by side in a heatmap, using clustering to group similar behavior.

    I would also suggest including a little more information in the main plots rather than only in the figure legends. For example, in Fig 2 including a label of 'ROS1-associated TE' for the two plots on the left, and 'TEs not associated with ROS1' on the right. Or for example in Fig. 3a indicating 'ros1-3 CG hyperDMRs' somewhere on the plot. This would just help make the figures easier to read at a glance. Please add common gene names to figures, instead just the ATG gene ID (Fig. S1a).

    Minor:

    • Fig. 1E is referenced in the text before Fig. 1D
    • Fig. S4 and S5 - there are more lines in the plot than the 6 genotypes listed in the legend, do these represent different replicates? If so that should be noted in the legend
    • Fig. 1B has no color legend for the different methylation sequence contexts (looks like same as 1A,C but should indicate either in plot or legend)
    • Line 42 should be "correspond to TE ends"
    • Line 93 "Based on previous studies..." should have references to those studies
    • When referring to the protein (rather than the genetic locus or mutant), ROS1 should not be italicized - for example line 130
    • Line 150 "we conclude that the loss"
    • Should add a y=x line to scatterplots, like those in Fig. 6
    • In fig. 1d, it's hard to evaluate the significance of the overlap of ROS1 targets with genes and TEs. Comparing these numbers to a control where the ROS1 targets have been randomly shuffled would help.

    Significance

    In this work, Hemenway and Gehring explore whether ROS1, DML2 and DML3 also affect DNA methylation patterns in endosperm. Using EM-seq of sorted endosperm nuclei, they show that loss of ROS1 indeed causes hypermethylation of a number of loci, particularly the flanks of methylated transposons, while loss of DML2 and DML3 has minimal additional effect. By obtaining allele-specific EM-seq data through crosses of Col and C24, the authors show that ros1 endosperm hypermethylation is mostly restricted to the paternal allele. The authors propose that at some sites, ROS1 helps bring down paternal methylation levels to match maternal methylation levels, which are typically reduced in endosperm due to DME activity in the female gametophyte prior to fertilization. In a ros1 mutant with paternal hypermethylation, these sites become differentially methylated on the maternal and paternal alleles, resembling imprinted loci. This work convincingly establishes a function for ROS1 in DNA methylation patterning in endosperm. However, I struggled with the clarity of the writing and reasoning in a few places, and would suggest clarification of a few points and additional analyses

  3. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #2

    Evidence, reproducibility and clarity

    Summary

    Hemenway and Gehring present evidence that the paternal genome in Arabidopsis endosperm is demethylated at several hundred loci by the DNA glycosylase/lyase ROS1. The evidence is primarily based on analysis of DNA methylation of ros1 mutants and of hybrid crosses where each parental genome can be differentiated by SNPs. I have some comments/questions/concerns, two of them potentially serious, but I think Hemenway and Gehring can address them through additional analyses of data that they already have available and a bit of clarification in writing.

    Major comments:

    1. Could the excess methylation in ros1-3 relative to ros1-7 shown in Figures 1A and 1C be explained by a second mutation in the ros1-3 background that elevates methylation at some loci? Any mutation that increased RdDM at these loci, for example could have this effect. This could confound the identification and interpretation of biallelicly demethylated loci.
    2. It appears that the main focus of the manuscript, the existence of loci that are paternally demethylated by ROS1, is supported by a set of 274 DMRs. This is a small number relative to the size of the genome and raises suspicions of rare false positives. Even the most stringent p-values that DMR-finding tools report do not guarantee that the DMRs are actually reproducible in an independent experiment. Demonstrating overlap between these 274 DMRs and an independently defined set using a different WT control and different ros1 allele would suffice to remove this concern. It appears that authors already have the needed raw data with ros1-1 and ros1-7 alleles.
    3. Because of the multiple sets of DMRs identified and used throughout the paper, it is hard to follow which one is which. There are DMRs defined solely by one sequence context, DMRs defined by all three contexts merged, DMRs defined by comparisons between maternal and paternal methylation in endosperm, DMRs defined by comparison between mutants and wildtype, and more. These need clearer descriptions of which sets are being referred to throughout the main text and in figure legends. A table summarizing them might help (not in the supplement). Use of consistent and precisely defined terms would help. Stating the number of DMRs along with the name for each set would help a lot, even though this would make for some redundancy. (The number of DMRs in each set not only helps with interpretation but also act as a sort of ID). The reason I put this as a major concern is because the text and figures are difficult to understand, and it is currently hard to evaluate both the results and the authors' conclusions from those results.

    Minor comments

    1. The sRNA results in Figure 2B are difficult to interpret because they do not reveal anything about the number of TEs that have siRNAs overlapping them or their flanks. While the magnitude of some of the highest endosperm sRNA peaks is higher than the embryo peaks, that could be explained by a small number of TEs with large numbers of sRNAs. To make this result more interpretable, we also need some information about how many TEs have a significant number of sRNAs associated with them in endosperm and embryo in each region (e.g., middle, 5', 3', and flanks of TEs). What a "significant number of sRNAs" is would be up to the authors to decide based on the distribution of sRNA counts they observe for TEs. Perhaps the top quartile of TEs? Combined with the same analysis done in parallel with non-ROS1 target TEs, this would reveal whether there is any evidence for ROS1 counteracting sRNA-driven methylation spread from TEs.
    2. The statement "we are likely underestimating the true degree of differential methylation among genotypes" should be validated and partially quantified using a methylation metaplot like Figure 2A, but substitute DMRs for TEs. Related to that, Figure 1B needs an indicator of scale in bp.
    3. The statement "Over half of ROS1 target regions identified in the ros1-3 mutant endosperm were within 1 kb or intersecting a TE (Fig. 1D)" is hard to interpret without some kind of ROS1 non-target regions or whole-genome control comparison. How different are the numbers in Fig. 1D from a random expectation?
    4. The sentence at line 262 is confusing. Is the comparison between dme mutant and ros1 mutant or between different types of regions? And it appears that the comparison value is missing in the "3-5% CG methylation gain..." e.g., "3-5% CG methylation vs 10-20%" or something like that.
    5. The dme mutant data in Figure 5C appear to be key to the model in Figure 7. The relative impact of the dme mutant in the two types of regions should be quantified.
    6. Looks like sRNA methods are missing.
    7. Supplemental Figure 1 is hard to interpret since it only list gene IDs, not gene names.

    The last comments are suggestions for increasing the impact of this study:

    1. Figure 2A and 3B suggest that ROS1 target TEs show demethylation in their flanks but not in the TE themselves. This is an interesting result. If it is true, more DMRs would be expected in the ROS1 target flanks than in the ROS1 target TEs. Reporting how many ROS1 target TEs have DMRs in them and what proportion have DMRs in their flanking 1-Kb regions would answer this question. Given the significance of this result, it also deserves a bit more context: Is the magnitude of increased methylation flanking TEs in ros1 mutant endosperm different than in ros1 mutant leaves or other tissue? Does methylation in TE flanks behave the way in dme mutant endosperm?
    2. The idea of biallelic demethylation has been theoretically suggested in maize to explain weak overlap between endosperm DMRs and imprinting (Gent et al 2022). If that were true in Arabidopsis, then ROS1 target, biallelicly demethylated loci would be less likely to have imprinted expression than maternally demethylated loci. This prediction could be tested using available data in Arabidopsis.
    3. There is currently no evidence for biological significance of biallelicly demethylated loci. Knowing where they are in the genome might give some hints. A figure like Fig. 1D but specifically showing the biallelicly demethylated DMRs would be valuable.
    4. It is hard to make the comparisons between genotypes and parental genomes in Figure 6 and know what they mean. Maybe a different way of displaying the data would help. Or maybe even a different labeling system could make it a little more accessible.

    Significance

    Demethylation of the maternal genome in endosperm has been the subject of much research because it can result in genomic imprinting of gene expression. The enzymes responsible, DNA glycosylases/lyases, also demethylate DNA in other cell types as well, where DNA methylation is not confined to one parental genome (biallelic or biparental as opposed to uniparental demethylation). To the best of my knowledge, the extent or even existence of biallelelic demethylation in endosperm has not been studied until now (except for a superficial look in a bioRxiv preprint, https://www.biorxiv.org/content/10.1101/2024.07.31.606038v1). Hemenway and Gehring have carried out a thoughtful and detailed analysis of the topic in Arabidopsis at least as far as it depends on the DNA glycosylase ROS1.

    A limitation is that the study design would miss biallelic demethylation by any of the other three DNA glycosylases in Arabidopsis. A second limitation is that there is no clear biological significance, just some conjecture about evolution. Nonetheless, given the novelty of the topic, biological significance may follow.

    The audience for biallelic DNA demethylation in Arabidopsis endosperm is certainly in the "specialized" category, but its relevance to the larger topic of gene regulation in endosperm will attract a larger audience.

  4. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #1

    Evidence, reproducibility and clarity

    In Arabidopsis, DNA demethylation is catalyzed by a family of DNA glycosylases including DME, ROS1, DML2, and DML3. DME activity in the central cell leads to the hypomethylation of maternal alleles in endosperm. While ROS1, DML2, and DML3 function in vegetative tissues to prevent spreading DNA methylation from TE boundaries, their function in the endosperm was unclear.

    Using whole genome methylome analysis, the authors showed that ROS1 prevents hypermethylation of paternal alleles in the endosperm thus promotes epigenetic symmetry between maternal and paternal genomes.
    The approach and experimental desighs are appropriate, and the key conclusions are adequately supported by the results.

    However, there is not sufficient evidence to support the claim that DME demethylates the maternal allele at ROS1-dependent biallelically-demethylated regions. To clarify the issue, the authors could analyze if there is an overlap between DMRs identified in ros1 endosperm and those identified in dme endosperm using published data. If there is any, the authors could show a genome browser example of DMR including dme data.

    Significance

    Endosperm is a tissue unique to flowering plants. Though it is an ephemeral tissue, the endosperm plays essential roles for seed development and germination. The endosperm is also the site genomic imprinting occurs, and it has a distinct epigenomic landscape. This work provides a new insight that ROS1 may antagonize imprinted gene expression in the endosperm. However, it was not shown whether imprinted gene expression is indeed affected in ros1, or whether the ros1 mutation has phenotypic consequences. These results would be useful to discuss the evolution and significance of genomic imprinting.

  5. Version published to 10.1101/2024.11.05.622036 on bioRxiv