Cytosine methylation regulates essential genome functions across eukaryotes, but the fundamental question of whether nucleosomal or naked DNA is the preferred substrate of plant and animal methyltransferases remains unresolved. Here, we show that genetic inactivation of a single DDM1/Lsh family nucleosome remodeler biases methylation toward inter-nucleosomal linker DNA in Arabidopsis thaliana and mouse. We find that DDM1 enables methylation of DNA bound to the nucleosome, suggesting that nucleosome-free DNA is the preferred substrate of eukaryotic methyltransferases in vivo. Furthermore, we show that simultaneous mutation of DDM1 and linker histone H1 in Arabidopsis reproduces the strong linker-specific methylation patterns of species that diverged from flowering plants and animals over a billion years ago. Our results indicate that in the absence of remodeling, nucleosomes are strong barriers to DNA methyltransferases. Linker-specific methylation can evolve simply by breaking the connection between nucleosome remodeling and DNA methylation.https://doi.org/10.7554/eLife.30674.001
Living cells add chemical tags to their DNA to regulate which genes are switched on or off at any given time. These tags include methyl groups added to one of the letters of the DNA code called cytosine. Both plants and mammals need cytosine methylation to develop properly. This methylation also keeps sections of foreign DNA that may have invaded the cell in check.
DNA inside the cell is tightly packed, wrapped around proteins to form spool-like structures called nucleosomes. Between each nucleosome is a short DNA segment called a linker region. The DNA wound into nucleosomes is generally inaccessible to other proteins, such as those that add methyl groups. Yet, in flowering plants and mammals, cytosine methylation occurs in both nucleosomes and in linker regions. It was not clear how DNA could be modified in the restrained setting of nucleosomes.
Enzymes called nucleosome remodelers can loosen nucleosomes to allow other proteins to reach the DNA. Lyons and Zilberman asked whether cytosine methylation occurs on the nucleosome-bound DNA or if it requires enzymes like these to free the DNA from the constraints of the nucleosome. The experiments involved a plant called Arabidopsis thaliana and mouse cells grown in the laboratory. In mutant plants lacking a nucleosome remodeler called DDM1, cytosine methylation occurred in the linker regions but not in the nucleosomes. Mouse cells lacking the mouse version of DDM1 also showed less cytosine methylation in the nucleosomes.
These results suggest that nucleosomes are barriers to the enzymes that modify DNA. Nucleosome remodeling enzymes like DDM1 can overcome these obstacles to enable cytosine methylation of nucleosome-wrapped DNA.
These findings imply that cytosine methylation is more easily established and maintained on nucleosome-free DNA. Abnormal patterns of DNA methylation have been linked to medical conditions – such as neurological disorders and cancers – and to plant defects that hamper agriculture. A better understanding of the process may in the future lead to ways to correct problems with cytosine methylation in these different contexts.https://doi.org/10.7554/eLife.30674.002
Cytosine methylation provides a mechanism to heritably alter the genome without permanent modification of the DNA sequence (Kim and Zilberman, 2014; Du et al., 2015). In eukaryotes, Dnmt1 family methyltransferases, called MET1 in plants, rely on selective recognition of hemi-methylated symmetrical CG dinucleotides by an obligate cofactor (Kim and Zilberman, 2014). This mechanism semiconservatively propagates methylation patterns following cell division, and – especially in plants – across generations. Plant and animal genomes also contain cytosine methylation outside CG dinucleotides (Du et al., 2015). In plants, the CMT3 methyltransferase family catalyzes methylation of CNG trinucleotides, conventionally described as CHG (where H is any non-G base) to avoid overlapping CG sites (Law and Jacobsen, 2010). The related CMT2 family can methylate cytosines outside CG and CNG contexts (Zemach et al., 2013; Stroud et al., 2014), a pattern of specificity referred to as CHH. CMT2 and CMT3 preferentially methylate heterochromatic transposable elements (TEs) and rely on a positive feedback loop with dimethylation of lysine 9 of histone H3. Plants also possess an RNA-directed DNA methylation (RdDM) pathway, in which 24 nucleotide RNA molecules guide DRM methyltransferases (homologs of animal Dnmt3) to initiate DNA methylation in all sequence contexts, and to maintain CHH methylation at relatively euchromatic TEs (Matzke and Mosher, 2014).
As implied by the above description of methylation pathways, DNA methylation represses TE transcription and transposition in plants and vertebrates (Law and Jacobsen, 2010). Methylation also regulates endogenous genes: methylation close to the transcriptional start site can cause gene silencing, whereas methylation of other genic regions can promote or counteract expression (Jones, 2012). Genetic defects in the methylation machinery lead to human diseases such as cancer (Baylin and Jones, 2016) and disruption of methylation patterns during clonal propagation of plants causes developmental defects that hamper agriculture (Ong-Abdullah et al., 2015; Springer and Schmitz, 2017). Faithful propagation of DNA methylation is clearly essential, yet our understanding of the underlying processes is far from complete. In particular, we do not know how DNA methyltransferases contend with nucleosomes, which comprise the basic repeating units of chromatin.
Nucleosomes consist of about 147 bp of DNA tightly wrapped around an octameric complex of histone proteins, with the major and minor grooves of the double helix alternately facing toward and away from the histones roughly every 10 bp (Luger et al., 1997). Nucleosomes are separated by short stretches of linker DNA, which are variably bound by histone H1 (Raghuram et al., 2009). Flowering plants and mammals show an overall enrichment of methylation over nucleosomes, suggesting that the nucleosome is the preferred target for DNA methylation (Chodavarapu et al., 2010; Chen et al., 2015). Furthermore, a mild but robust 10 bp DNA methylation periodicity in both lineages suggests that methyltransferases directly modify DNA that is wrapped in nucleosomes (Chodavarapu et al., 2010). However, mammalian methyltransferase activity is inhibited by nucleosomes and H1 in vitro (Robertson et al., 2004; Takeshima et al., 2006; Takeshima et al., 2008; Felle et al., 2011; Schrader et al., 2015), and regions of the human and mouse genomes with reliably positioned nucleosome arrays organized by CTCF binding exhibit preferential methylation of linker DNA (Kelly et al., 2012; Baubec et al., 2015). Methylation is also almost completely confined to linker DNA in several diverse species of marine algae (Huff and Zilberman, 2014), suggesting that nucleosomes are generally refractory to DNA methylation. The reasons for the apparent discrepancy between in vitro and in vivo methyltransferase preferences, and for the observed interspecies differences, remain unknown.
Adding to the mystery, plants and mammals depend on the DDM1/Lsh family of Snf2-class nucleosome remodelers – proteins that can alter nucleosomes in various ways – to achieve normal levels of DNA methylation (Jeddeloh et al., 1999; Zhou et al., 2016). Arabidopsis plants with mutations in DDM1 suffer major methylation losses in all sequence contexts, primarily in heterochromatic TEs, though much smaller losses are also observed in genes (Jeddeloh et al., 1999; Teixeira et al., 2009; Zemach et al., 2013; Ito et al., 2015). Knockout of mouse Lsh likewise causes major depletion of DNA methylation from repetitive heterochromatin (Dennis et al., 2001), with more variable effects on methylation in genes and other sequences (Myant et al., 2011; Tao et al., 2011). Genetic inactivation of Arabidopsis histone H1 partially rescues the ddm1 hypomethylation phenotype, suggesting that DDM1 facilitates methyltransferase access to H1-containing chromatin (Zemach et al., 2013). However, how DDM1 does this is unknown, and the incomplete rescue of DNA methylation in h1ddm1 compound mutants remains unexplained. More generally, how nucleosome remodelers facilitate methylation in species in which methyltransferases are proposed to act on the nucleosome surface is unclear.
Here we explore how DNA methylation is regulated by DDM1/Lsh, H1 and nucleosomes. We find that nucleosomes and H1 are barriers to DNA methylation and that DDM1/Lsh remodelers are required for methylation of nucleosomal DNA. At heterochromatic loci with reliably positioned arrays of nucleosomes, methylation in h1ddm1 plants is strikingly periodic, with nearly wild-type levels in linkers and nearly ddm1 levels in nucleosomes. Our results indicate that methyltransferases generally require remodeling activity to access nucleosomal DNA. Lack of such remodeling biases methylation toward linker DNA, providing a straightforward explanation for the linker-specific methylation patterns found in diverse eukaryotes.
To investigate the relationship between nucleosomes and DNA methylation, we deep sequenced unamplified DNA isolated from micrococcal nuclease (MNase)-digested chromatin of Arabidopsis plants with inactivating mutations in both canonical linker histone H1 genes (h1), ddm1 mutants, compound h1ddm1 mutants, and in WT controls. We selected mononucleosomes in silico to avoid inconsistencies stemming from DNA electrophoretic mobility. Using biological replicates, we defined four groups of nucleosomes per genotype, from well-positioned to poorly-positioned (see Materials and methods), and anchored methylation analysis at the presumptive dyad (nucleosome center) (Chen et al., 2014). We focused our initial analysis on heterochromatic sequences (see Materials and methods), where H1 is most abundant (Rutowicz et al., 2015) and DDM1 is most important for maintenance of DNA methylation (Zemach et al., 2013).
CG methylation shows a mild but clear depletion over well-positioned nucleosomes in WT and h1 mutant plants, and an overt depletion in ddm1 and h1ddm1 plants (Figure 1A, Figure 1—figure supplement 1, loci with well-positioned nucleosomes on the left). Poorly positioned nucleosomes by definition do not have a well-delineated core region, and these loci accordingly exhibit a relatively flat level of methylation in all genotypes (Figure 1A, Figure 1—figure supplement 1). The difference between core and linker methylation is much greater in h1ddm1 than in ddm1 (Figure 1B), an effect that is observable at individual loci (Figure 1C, Figure 1—figure supplement 2), where methylation of linker regions – much stronger in h1ddm1 than in ddm1 – alternates with weak or absent methylation of nucleosome cores.
For the above analyses, we defined quality of nucleosome positioning separately for each genotype, and therefore the nucleosomes that comprise each of the four positioning groups vary between genotypes. To control for any potential sequence bias across genotypes, we directly compared the best positioned nucleosomes shared by WT and h1ddm1, or by h1 and h1ddm1 (Figure 1—figure supplement 3; see Materials and methods). This analysis produced the same pattern as in Figure 1, at nearly the same scale. Interestingly, using only WT well-positioned nucleosomes severely attenuates the core-to-linker CG methylation difference in ddm1 and essentially eliminates any difference in h1ddm1 (Figure 1—figure supplement 3), illustrating the importance of genotype-specific nucleosome mapping.
Our results indicate that nucleosomes and H1 are obstacles to DNA methylation in vivo. Nucleosomes, which form much more stable associations with DNA than H1 (Misteli et al., 2000), are more powerful obstacles. In WT an h1 plants, DDM1 activity allows DNA methyltransferases robust access to DNA. In ddm1 mutants, nucleosomes are inaccessible and H1 blocks linker DNA, but the greater accessibility of linkers allows higher methylation levels at these sequences (Figure 1A,B). In the absence of DDM1 and H1, linker DNA is accessible to methyltransferases, but nucleosome cores are not, producing the robust methylation periodicity in h1ddm1 plants (Figure 1A–C).
Interestingly, non-CG methylation shows a pronounced nucleosome core depletion at well-positioned nucleosomes in all genotypes (Figure 1D,E, Figure 1—figure supplement 3). In the CHG context, h1ddm1 has the most exaggerated core-to-linker methylation differential, followed by h1 (Figure 1B,D, Figure 1—figure supplement 3). In the CHH context, h1ddm1 and h1 show a similarly strong core-to-linker methylation differential (Figure 1B,E, Figure 1—figure supplement 3). Thus, whereas DDM1 allows for near-perfect efficiency of nucleosome core CG methylation (Figure 1A), non-CG methylation of nucleosomal DNA is well below that of linkers even in WT (Figure 1D,E). Loss of H1 increases the contrast between linker and core methylation levels (Figure 1A–E). Taken together, the observed CG and non-CG methylation patterns strongly argue that in the absence of nucleosome remodeling by DDM1, DNA within heterochromatic nucleosomes is unavailable to methyltransferases.
DDM1 belongs to an ancient protein family that includes mouse Lsh, inactivation of which causes extensive DNA hypomethylation (Dennis et al., 2001; Law and Jacobsen, 2010). To assess whether Lsh, like DDM1, permits methylation of DNA wrapped in nucleosomes, we examined published methylation data from Lsh null mouse embryonic fibroblasts and WT controls (Yu et al., 2014) in relation to nucleosome positions (Teif et al., 2012) (Figure 1F). Because this analysis requires reliably well-positioned nucleosomes, we assessed regions of the mouse genome immediately flanking CTCF binding sites, which are surrounded by positioned nucleosome arrays (Fu et al., 2008). As previously reported, we find that WT DNA methylation is periodic around CTCF binding sites, with clear depletion at nucleosome core regions (Kelly et al., 2012; Baubec et al., 2015) (Figure 1F). Cells with inactivated Lsh show a much stronger methylation periodicity, with far greater depletion at nucleosome cores (Figure 1F). This contrast is most evident in the linkers that extend from the most reliably positioned nucleosomes on either side of the CTCF binding sites (emphasized with arrows in Figure 1F), which show nearly WT levels of DNA methylation in Lsh mutant cells. Our results indicate that nucleosomes present barriers to mammalian as well as plant DNA methyltransferases, and that the conserved function of the DDM1/Lsh family of remodelers is to enable methylation of nucleosomal DNA.
Arabidopsis TEs can be roughly grouped into heterochromatic and euchromatic elements (Zemach et al., 2013). Heterochromatic TEs, which we analyzed in Figure 1, tend to be longer, are methylated at CHH sites primarily by CMT2 and require DDM1 for methylation in all contexts (Zemach et al., 2013; Stroud et al., 2014). Euchromatic TEs are much shorter, less dependent on DDM1, and CHH methylation at these elements is maintained by DRM1 and DRM2 (Zemach et al., 2013; Stroud et al., 2014). Because DRM1/2-dependent loci are short and often bordered by unmethylated DNA, analysis of methylation anchored to the centers of differentially methylated regions (DMRs; see Materials and methods) between drm1drm2 mutants and wild-type produces a sharp peak (plots in the right column of Figure 2A–C). Nonetheless, anchoring methylation analysis to well-positioned nucleosomes shows that CG and CHG methylation at DRM-dependent loci is depleted over nucleosomes much as it is in heterochromatin (Figure 2A–B, Figure 2—figure supplement 1), demonstrating that nucleosomes in euchromatic TEs are refractory to DNA methylation. CHH methylation is also depleted over nucleosomes, but unlike in heterochromatin, the depletion is not obviously enhanced in any of the mutant genotypes (Figure 2C, Figure 2—figure supplement 1), probably because the RdDM pathway that guides DRM methyltransferases is largely independent of DDM1 but is associated with other remodelers (Kanno et al., 2004; Smith et al., 2007; Zemach et al., 2013; Han et al., 2015). As expected, CMT2-dependent loci recapitulate the behavior of heterochromatic TEs (Figure 2D–F, Figure 2—figure supplement 1). Due to the strong linker enrichment of methylation in h1ddm1 heterochromatin (Figure 1A–E), even analyses centered on DMRs between cmt2 and wild-type without regard to nucleosome position produce periodic h1ddm1 methylation patterns (plots in the right column of Figure 2D–E).
Previous work demonstrated a global enrichment of DNA methylation on Arabidopsis nucleosomes (Chodavarapu et al., 2010). We have so far described the opposite pattern, but our analysis has been confined to TEs. To identify regions of the genome that show the reported preferential methylation of nucleosomal DNA, we expanded our analysis to include all nucleosomes. We find that nucleosomes within genes possess a marked enrichment of CG methylation in WT, whereas TEs exhibit no depletion, so that methylation is enriched in WT nucleosomes on average (Figure 3A; see Materials and methods). Separating genic nucleosomes by degree of positioning, as we did for heterochromatin (Figure 1A), reveals an enrichment of DNA methylation in the nucleosome core that diminishes with increased positioning uncertainty (Figure 3—figure supplement 1). The correspondence between nucleosomes and DNA methylation within genes can be easily seen at individual loci (Figure 3B). Interestingly, nucleosomes in the exons of genes show a clear enrichment of DNA methylation, whereas intronic nucleosomes do not (Figure 3C), indicating that methylation is not intrinsically targeted to genic nucleosomes.
The enrichment of methylation over nucleosomes in genes does not preclude the possibility that this methylation requires a remodeling activity. The ddm1 and h1ddm1 mutations do not eliminate methylation of genic nucleosomes (Figure 3A, Figure 3—figure supplement 1), which is consistent with the preferential requirement of DDM1 for heterochromatic methylation (Zemach et al., 2013). However, because DDM1 does contribute to DNA methylation in genes (Zemach et al., 2013), we asked whether DDM1 is needed for methylation of a subset of genic nucleosomes. Indeed, we find that lowly-expressed genes depend on DDM1 to maintain nucleosome methylation, whereas more highly expressed genes do not, on average, need DDM1 (Figure 3D). Our definition of genes excludes TEs or sequences with TE-like methylation patterns (Zemach et al., 2013), therefore this result is not caused by TE contamination of our gene anotations. Genes with lower levels of expression undergo less transcription-coupled nucleosome displacement (Workman, 2006), and their methylation may therefore be more dependent on DDM1-mediated remodeling than more highly expressed counterparts. Our results demonstrate that DDM1 facilitates methylation of nucleosome-wrapped DNA in a subset of genes as well as in heterochromatin.
Perhaps the most compelling evidence that DNA methylation occurs on the surface of the nucleosome is the reported subtle 10 bp periodicity of methylation in all sequence contexts (Chodavarapu et al., 2010) – a periodicity that reflects the alternate facing of the major and minor grooves of DNA toward and away from the histone octamer. If the 10 bp periodicity is indeed caused by differential accessibility of DNA on the nucleosome surface, it should not be affected by mutation of a nucleosome remodeler. To assess whether DDM1 is required for this pattern, we examined DNA methylation periodicity in ddm1 plants and sibling controls at heterochromatic TEs. We anchored methylation analysis to the ends of 147 bp MNase fragments isolated in silico from our paired-end libraries, and calculated the strength of periodicity in the average per-base methylation signal with fast Fourier transform (FFT) computation (Chodavarapu et al., 2010; see Materials and methods).
As expected, WT plants display a 10 bp methylation periodicity in all sequence contexts (Figure 4A, Figure 4—figure supplement 1A–C). Lack of H1 does not appreciably alter the periodicity in non-CG contexts; CG methylation periodicity is weaker than WT (Figure 4A), probably because the very high levels of heterochromatic CG methylation in h1 plants (Zemach et al., 2013) leave little room for oscillations. Importantly, the 10 bp periodicity is virtually absent in ddm1 heterochromatin in all contexts (Figure 4A). In contrast, methylation of genic nucleosomes, which in most cases does not require DDM1, displays a clear 10 bp periodicity in all genotypes examined (Figure 4B, Figure 4—figure supplement 1D).
To confirm our Arabidopsis results, we examined whether the 10 bp periodicity also declines in Lsh mutants. Because our analysis of Lsh methylation is based on published nucleosome data from WT cells (Teif et al., 2012), and corresponding data from Lsh null cells is not available, we based our calculation of methylation periodicity on nucleosomal fragments mapping around the CTCF sites used in Figure 1F, where nucleosomes positioning is highly reliable (Fu et al., 2008). We find a clear 10 bp periodicity in WT CG methylation, but in Lsh-/- the signal is noisy and does not exhibit a maximum at 10 bp (Figure 4C, Figure 4—figure supplement 2). Although this phenotype may be caused by anchoring the Lsh-/- methylation analysis to WT rather than Lsh-/- nucleosomes, bound CTCF sites generally possess very well positioned flanking nucleosomes (Fu et al., 2008; Stadler et al., 2011), suggesting that methylation periodicity is attenuated in the absence of Lsh. Thus, our results indicate that the 10 bp methylation periodicity observed in Arabidopsis and mouse requires a nucleosome remodeler, providing further support for the hypothesis that DNA methyltransferases cannot directly methylate nucleosomes.
The depletion of nucleosome methylation in ddm1 and h1ddm1 plants (Figure 1A–E, Figure 1—figure supplement 3) is reminiscent of the linker-specific DNA methylation patterns observed in three lineages of marine algae (Huff and Zilberman, 2014) that diverged from one another, and from flowering plants and animals, over a billion years ago (Parfrey et al., 2011). However, the contrast between the full methylation of linker DNA and the absence of nucleosomal methylation in marine algae (Huff and Zilberman, 2014) is much greater than that observed even in the best positioned group of h1ddm1 nucleosomes (Figure 1A). A potential explanation is that the relevant algal species have exceptionally well-positioned nucleosomes (Huff and Zilberman, 2014), whereas even the best positioned Arabidopsis nucleosome group we identified is likely quite heterogeneous.
To test this hypothesis, we identified arrays of well-positioned Arabidopsis nucleosomes by using an unsupervised clustering algorithm (Stempor et al., 2016) to organize h1ddm1 CG methylation data centered on heterochromatic nucleosomes in the best-positioned group. One of the resultant clusters (C2, 4849 loci) exhibits highly structured CG methylation that alternates with nucleosome positions (Figure 5A). All genotypes show methylation periodicity at C2 in all sequence contexts (Figure 5A–D). Remarkably, h1ddm1 linker CG methylation reaches near-WT levels in this cluster, whereas nucleosome core methylation is close to ddm1 levels (Figure 5B). The periodicity of h1ddm1 non-CG methylation at C2 is also striking, with ddm1-like core methylation and linker methylation that is even higher than WT (Figure 5C,D, Figure 5—figure supplement 1). The h1ddm1 methylation patterns at C2 closely resemble those seen in algae over a billion years removed from the Arabidopsis lineage.
We have shown that nucleosomes are strong impediments to DNA methylation and that the DDM1/Lsh remodeling proteins facilitate nucleosome methylation in vivo. It remains formally possible that DDM1/Lsh remodelers permit methyltransferases to work directly on the nucleosome surface or promote catalysis without affecting methyltransferase access. However, these interpretations are not parsimonious, and are inconsistent with the accumulating evidence that naked DNA is the preferred template for most DNA modifying enzymes. Such enzymes include cytidine deaminases (Kodgire et al., 2012), the TET1 dioxygenase (Kizaki et al., 2016), CAS9 (Horlbeck et al., 2016), and a bacterial cytosine methyltransferase that has the same catalytic mechanism as eukaryotic enzymes (Kelly et al., 2012). In conjunction with these observations, our data strongly support the hypothesis that cytosine methyltransferases generally require naked DNA, and that DDM1/Lsh nucleosome remodelers provide this substrate.
We propose that DDM1/Lsh-type remodelers render a region of nucleosomal DNA accessible to DNA methyltransferases in a manner analogous to the remodeling mechanism proposed for Swi2/Snf2 (Zofall et al., 2006). These remodelers, which are closely related to DDM1, bind the nucleosome and translocate DNA from linker toward dyad, generating a nucleosome-free DNA loop (Figure 6) (Flaus et al., 2006; Zofall et al., 2006; Zhou et al., 2016). Such a loop would allow methyltransferases access to the DNA that would be biased by its orientation to the histone octamer surface, plausibly producing the observed subtle 10 bp methylation periodicity. The 10 bp periodicity could also arise if the formation or stability of nucleosomes is affected by the orientation of 5-methylcytosine toward the histone octamer (Jimenez-Useche et al., 2013).
Despite the overall importance of DDM1 for DNA methylation, many nucleosomes, particularly those in genes, are methylated normally without DDM1 (Figure 3). A likely explanation is that other nucleosome remodelers facilitate methylation of euchromatic loci. This hypothesis is consistent with the observations that the RdDM pathway requires remodelers other than DDM1 (Kanno et al., 2004; Smith et al., 2007; Zemach et al., 2013; Han et al., 2015), although their specific functions remain unknown. The partial restoration of the 10 bp methylation periodicity in h1ddm1 heterochromatin (Figure 4A) also suggests that remodelers other than DDM1 can mediate methylation of heterochromatic nucleosomes in the absence of H1. This idea is supported by the reported ability of H1 to prevent nucleosome remodeling in vitro for some Swi2/Snf2-type remodelers (Lusser et al., 2005) and could explain why heterochromatic nucleosomes – and not just linkers – show higher methylation in h1 compared to WT, and in h1ddm1 compared to ddm1 (Figures 1B and 5B–D).
The likely involvement of multiple remodelers in euchromatic DNA methylation does not by itself explain the robust enrichment of methylation within exonic nucleosomes (Figure 3). The origins and biological functions of gene body methylation remain quite mysterious (Zilberman, 2017), and therefore any explanation of its features is inherently speculative. Given that introns do not exhibit preferential methylation of nucleosomes (Figure 3C), the pathways that target methylation to genes are unlikely to favor nucleosomes, especially considering that methylation in genes and TEs is catalyzed by the same methyltranferases (Law and Jacobsen, 2010). Preferential methylation of nucleosomes within exons may be related to the function of gene body methylation. DNA methylation is known to affect the properties of nucleosomes, such as positioning and stability (Davey et al., 1997; Choy et al., 2010; Jimenez-Useche et al., 2013; Huff and Zilberman, 2014). If gene body methylation influences splicing, other aspects of RNA processing, or the stability of transcript elongation (Zilberman, 2017), it may do so in part by altering the properties of exonic nucleosomes. Enrichment of methylation over exonic nucleosomes may therefore be a product of selection. Despite their preferential methylation, genic nucleosomes – at least those experiencing low levels of transcription – are refractory to MET1 methyltransferase activity, as evidenced by their decreased methylation in ddm1 mutants (Figure 3D).
The ability of nucleosomes and H1 to block DNA methylation in the absence of remodeling is dramatically illustrated by a set of heterochromatic loci with arrays of positioned nucleosomes (Figure 5, Figure 5—figure supplement 1). Here, linker methylation that is close to or even above WT in h1ddm1 plants is juxtaposed with severe depletion of nucleosomal methylation. Methylation patterns at these loci closely resemble the natural patterns of several diverse species in which methylation is confined to linker DNA throughout the genome (Huff and Zilberman, 2014). This resemblance is unlikely to be coincidental. Species with linker-specific methylation are interspersed on the tree of life with species that do not limit methylation to linkers (Figure 5—figure supplement 2; Huff and Zilberman, 2014). This evolutionary history has been difficult to explain, particularly if one assumes that transitions to or from linker-specific methylation require changing the intrinsic ability of DNA methyltransferases to access nucleosomes.
Our ability to impose an essentially linker-specific methylation pattern on a substantial fraction of the Arabidopsis genome by inactivating two chromatin proteins provides a plausible solution to this quandary. Species in which linker-specific methylation has been described have well-positioned nucleosomes (Huff and Zilberman, 2014) and all but one of these species (Micromonas pusilla) lack the winged-helix motif of canonical H1 (Figure 5—figure supplement 3) (Kasinsky et al., 2001; Worden et al., 2009). A transition to linker-specific methylation in a species with reliably positioned nucleosomes and low or absent linker histones would only require the uncoupling of nucleosome remodeling from DNA methylation maintenance. This could occur through inactivation of the relevant remodeler, or – more likely – by a mutation that severs the spatiotemporal connection between DNA methylation and remodeling. Such a mutation, if sufficiently advantageous, could also be fixed in a species with complex nucleosome positioning and high H1. The linker-specific methylation patterns almost certainly offer a selective advantage to the species that have them, in part by contributing to nucleosome positioning (Huff and Zilberman, 2014). Thus, loss of H1 could be advantageous in the presence of a linker-specific methylation system, and reliable nucleosome positioning a consequence of such a system. Considering the many independent instances of complete DNA methylation loss among eukaryotes (Zemach and Zilberman, 2010), fixation of advantageous mutations that separate methylation from remodeling could plausibly occur multiple times during eukaryotic evolution. Such a model would easily explain the diverse methylation patterns observed among eukaryotes by providing the mechanism by which apparently radically different patterns can repeatedly evolve.
ddm1-10 (Zhang et al., 2016), h1.1 and h1.2 (Zemach et al., 2013) Arabidopsis thaliana (Columbia ecotype) were crossed to generate h1ddm1 and siblings for this study. Mutation of DDM1 causes permanent methylation losses at many loci that are not recovered after restoration of DDM1 activity (Teixeira et al., 2009). We therefore used the ddm1-10 allele, which, unlike the ddm1-2 allele we used in a previous study (Zemach et al., 2013), was not homozygous at any point prior to the generation of the compound h1ddm1 mutant reported here. For MNase-seq, RNA-seq, and bisulfite-seq samples, plants were germinated on agarose plates supplemented with Gamborg’s B-5 growth media (GBP07, Caisson Labs). Following germination, they were transferred to soil and grown on a 16hr-light/8hr-dark long-day schedule. For all experiments described here, rosette leaves of 1 month old plants were used.
Genomic DNA (gDNA) was extracted from 1-month-old Arabidopsis thaliana rosette leaves with the DNeasy plant mini kit (Qiagen, cat. no. 69104) per the manufacturer’s instructions. Libraries were prepared from roughly 500 ng of purified gDNA that was sheared to approximately 400 bp on a Misonix water bath sonicator, then purified using 1.2X volume of Agencourt Ampure beads (referred to as ‘beads’ henceforth, cat. no. A63881). Following ligation of methylated Truseq sequencing adapters (Illumina Hayward, CA), bisulfite conversion of DNA was carried out according to manufacturer’s protocol (Qiagen Epitect Kit, cat. no. 59104) except without using carrier RNA. DNA was purified twice with 1.2X beads and converted a second time to ensure complete bisulfite conversion of unmethylated cytosine. NEB next indexing primers (cat. no. E7335S) were used for generating multiplexed libraries during PCR amplification of libraries.
Leaves were flash frozen in liquid nitrogen, pulverized with mortar and pestle on dry ice, and the resulting material was subjected to vortexing in Trizol (Invitrogen, cat. no. 15596–026). Chloroform was then added at one-fifth the total volume and further vortexing was carried out until the solution appeared homogenous. RNA was subsequently pelleted in ice-cold isopropanol. The resuspended RNA was subjected to rRNA removal with Ribo-zero plant kit (Illumina, MRZPL1224) according to the manufacturer’s protocol. 50 ng of ribo-depleted RNA was used for library preparation with the Scriptseq kit (Epicentre, cat. no. SSV21124) following the manufacturer’s protocol but with the following modifications: the RNA fragmentation step was extended to 10 min, and the temperature was increased to 90°C.
Approximately 1 g of pulverized flash frozen rosette leaf tissue from F4 generation plants was resuspended on ice in nuclei isolation buffer (0.25 M sucrose, 15 mM PIPES pH 6.8, 5 mM MgCl2, 60 mM KCl, 15 mM NaCl, 0.9% Triton X-100, 1 mM PMSF, 1X protease inhibitor cocktail) and allowed to thaw for 15 min. The suspension was then strained once through Miracloth (EMD-Millipore, cat. no. 475855) and the nuclear pellet was centrifuged in a swinging-bucket rotor at 2000 g in pre-chilled 5 ml conical tubes. Buffer was carefully aspirated from the loose pellet, which was subsequently resuspended in TM2 (10 mM Tris–HCl, pH 8, 2 mM MgCl2) and washed twice in TM2. The pellet was then resuspended in pre-warmed MNase digestion buffer (50 mM Tris-HCl, pH 8, 5 mM CaCl2) and 25 units of MNase were added per 0.1 g initial starting material and incubated for 10 min at 37°C with periodic agitation; reactions were stopped with EGTA (20 mM final concentration). This was followed by the addition of SDS to a final concentration of 0.625%, increasing the temperature to 75°C, addition of RNase A for 10 min, and the addition of proteinase-K for an additional 15 min. DNA was then extracted with phenol/chloroform, pelleted in 70% EtOH, then resuspended in TE pH 8.0.
Verification of MNase digestion was carried out by fractionating one-tenth of the total reaction on a standard TAE gel and checking for a majority mononucleosomal fraction. Following removal of short DNA with beads, libraries were synthesized using total DNA (without PCR amplification) using the Nugen Ovation kit (Cat. no. 0319) following the manufacturer’s protocol.
All sequencing was carried out as single-end 100 bp reads, except MNase libraries, which were sequenced in paired-end 100 bp reads on Illumina HiSeq 2500 or −4000 at the QB3 Vincent Coates Genomic Sequencing Lab at UC Berkeley.
Bisulfite sequencing reads were mapped with Bowtie (Langmead et al., 2009) using the bs-sequel pipeline (available at http://dzlab.pmb.berkeley.edu/tools), with Bowtie settings allowing for up to two mismatches in the seed and reporting up to 10 matches for a given read. 100 bp paired-end MNase-seq reads were mapped using Bowtie2 (Langmead and Salzberg, 2012) allowing for up to one mismatch in the seed region and only the best match reported. All alignments were made to the TAIR10 Arabidopsis thaliana genome assembly.
The Kallisto (Bray et al., 2016) quant command was invoked with the following settings: --single --fr-stranded -b 100 l 320 s 30. Kallisto output was fed into the R environment for processing with the Sleuth software package (Pimentel et al., 2017), which was used to perform expression quantification and derive expression deciles.
For initial identification of nucleosomes, the Python program iNPS (Chen et al., 2014) was used on our paired-end MNase-seq reads (120 to 180 bp, not inclusive), which were filtered using SAMtools (Li et al., 2009). iNPS produces a number of nucleosome types depending on the shape of read distribution: only ‘main peaks’ from this output were taken from our biological replicates and used in identification of positioned nucleosomes. The other output from this program, the ‘like_Wig’ files, were used in visualization of MNase-seq data in the genome browser as well as for calculating average nucleosome enrichment across loci.
We took iNPS nucleosome peaks and removed all that were wider than 140 bp. Remaining nucleosome overlap was calculated using the built-in ‘-wo’ option of bedtools (Quinlan and Hall, 2010) ‘intersect’ function. The percent overlap of peaks from replicate one with peaks from replicate two was then used to define regions of the genome with reliably positioned nucleosomes. Nucleosome peaks that overlapped reciprocally more than 75% were classified as well-positioned (left-most panel in Figure 1A, for example). Less-well-positioned nucleosomes were further defined in 25% increments of overlap, such that the next category overlapped more than 50% but less than 76%, and so on, generating four groups, plus a group that was comprised of other nucleosomes. To isolate differentially-methylated regions (DMR) outside of positioned nucleosomes (see below for definition of DMRs), we used DMRs not overlapping any of the four groups defined above.
The presumptive centers of these nucleosomal loci were calculated as the arithmetic means of the combined 5’- and 3’-most ends of the overlapping nucleosomes, allowing for multiple overlaps for a given nucleosome. Nucleosome groups per genotype are provided in GEO (GSE96994). Shared nucleosomes (Figure 1—figure supplement 2) are defined as those nucleosomes whose calculated dyad is within 20 bp (inclusive) of the other genotype. For h1ddm1, three biological replicates were generated for visualization but only two were used as input to determine nucleosome position because addition of a third replicate did not improve the positioning analysis. The third replicate is available in the GEO accession associated with this work.
‘Heterochromatic sequences’ in Figure 1 refers to merged TAIR10 TEs exhibiting >5% CG methylation that are in the upper two quintiles for H3K9me2 enrichment (as calculated from Stroud et al., 2014) and are longer than 30 bp (complete merged TE bed file provided in GEO accession GSE96994). TEs referred to related to Figure 3 are all merged TEs, regardless of H3K9me2 level. The TSS (Figure 3A) was defined as ±1 kb from the annotated TAIR10 TSS; while gene body was considered the portion downstream of 1 kb. Genes were defined to exclude annotated TEs or sequences methylated like TEs, as in (Zemach et al., 2013).
50 bp windows with at least 10% WT CHH methylation (mCHH) and with more than 30% loss of mCHH in drm1drm2 or cmt2 were used to define DRM1/2 and CMT2 dependent DMRs, respectively. These windows also exhibited statistically significant decrease in mCHH from WT (p<0.01, Fisher’s exact test). We further filtered out DMRs that exhibited significant mCHH differences relative to WT in both genotypes, ensuring the DMRs are exclusive to the given genotype. After merging adjacent 50 bp DMR windows, there were 54,106 CMT2 DMRs and 8640 DRM1/2 DMRs. These coordinates are available in GEO accession GSE96994. All DMRs overlap at least one annotated TE.
Methylation plots in Figure 2 are centered either on ‘group1’ genotype-specific well positioned nucleosomes that overlap the indicated DMR by at least 1 bp or, for the ‘DMR’ plots in the same figure, we used the arithmetic center of DMRs that do not overlap with any of the nucleosomes identified in groups 1 through 4, also in a genotype-specific manner (e.g. h1 DMR plots are anchored to those DMRs which do not overlap any of the h1 group 1–4 nucleosomes).
We aligned sequences with T-coffee (Notredame et al., 2000) in normal mode and plotted the resulting alignment file with Boxshade (http://www.ch.embnet.org/). The following sequences were used to perform multiple sequence alignment shown in Figure 5—figure supplement 3: Aureococcus anophagefferens, protein ID 72830, scaffold_256:8187–9341; Ostreococcus lucimarinus, protein ID 18993, Chr_20:191746–192084 (+); Emiliana huxleyi protein ID 448806/460675, scaffold_1716:1449–2181 (+)/scaffold_11:31908–32637 (-); Micromonas pusilla, protein ID 1714, scaffold_11:410445–411786, Arabidopsis thaliana H1.1 (AT1G06760), Neurospora crassa hH1 (ORF name NCU06863), and Mus musculus H1a (MGI:1931523).
CTCF sites (Baubec et al., 2015) were used as anchors for plotting averaged DNA methylation and MNase-seq data. Lsh mutant and WT MEF DNA methylation data (Yu et al., 2014) and WT MNase-seq data (Teif et al., 2012) were obtained through GEO (GSE56151 and GSE40896, respectively). Bin widths for methylation and MNase analysis were both set at 10 bp.
For Arabidopsis, 147 bp MNase-seq fragments were isolated from one of the biological replicates for each genotype. The resulting reads were used as anchors to calculate mean DNA methylation using bisulfite data (Chodavarapu et al., 2010); this methylation vector was used as input to the TSA R program (function ‘periodogram’) (Chan and Ripley, 2012), which generated the power spectrum over a range of frequencies. ‘spec’ output was plotted against the inverse of frequency (in this case, bp), truncated to 30 bp and plotted. For mouse, 147 bp fragments from Teif et al. (2012) that overlapped CTCF sites (±1 kb) (Baubec et al., 2015) were used as anchors for methylation averaging, and the strength of periodicity was calculated as above.
For MNase-seq plotting, in order to compare reads mapped across different genotypes and multiple replicates, we normalized by multiplying the smoothed output of iNPS by the reads per million per nucleotide (RPM). For instance, a 10 bp bin score of MNase-seq in sample A would be multiplied by 1 × 106/(number of mapped reads in sample A). Perl scripts (http://dzlab.pmb.berkeley.edu/tools) were used to generate enrichment score matrices of mapped data around genomic features of interest. These matrices were imported to R (R Core Team , 2017; Davey et al., 1997) for further processing and visualization using base plotting functions and the ggplot2 library (Wickham, 2009). Heatmaps were generated with Seqplots (Stempor et al., 2016) using the self-organizing map (SOM) clustering algorithm. Genome tracks are manual screenshots of our data displayed in IGV (Robinson et al., 2011).
DNA methylation increases nucleosome compaction and rigidityJournal of the American Chemical Society 132:1782–1783.https://doi.org/10.1021/ja910264z
CpG methylation remodels chromatin structure in vitroJournal of Molecular Biology 267:276–288.https://doi.org/10.1006/jmbi.1997.0899
Lsh, a member of the SNF2 family, is required for genome-wide methylationGenes & Development 15:2940–2944.https://doi.org/10.1101/gad.929101
DNA methylation pathways and their crosstalk with histone methylationNature Reviews Molecular Cell Biology 16:519–532.https://doi.org/10.1038/nrm4043
Nucleosomes protect DNA from DNA methylation in vivo and in vitroNucleic Acids Research 39:6956–6969.https://doi.org/10.1093/nar/gkr263
Identification of multiple distinct Snf2 subfamilies with conserved structural motifsNucleic Acids Research 34:2887–2905.https://doi.org/10.1093/nar/gkl295
Roles and activities of chromatin remodeling ATPases in plantsThe Plant Journal 83:62–77.https://doi.org/10.1111/tpj.12877
Maintenance of genomic methylation requires a SWI2/SNF2-like proteinNature genetics 22:94–97.https://doi.org/10.1038/8803
Functions of DNA methylation: islands, start sites, gene bodies and beyondNature Reviews Genetics 13:484–492.https://doi.org/10.1038/nrg3230
DNA methylation as a system of plant genomic immunityTrends in Plant Science 19:320–326.https://doi.org/10.1016/j.tplants.2014.01.014
Preferential 5-methylcytosine oxidation in the linker region of reconstituted positioned nucleosomes by Tet1 proteinChemistry - A European Journal 22:16598–16601.https://doi.org/10.1002/chem.201602435
Nucleosome stability dramatically impacts the targeting of somatic hypermutationMolecular and Cellular Biology 32:2030–2040.https://doi.org/10.1128/MCB.06722-11
Establishing, maintaining and modifying DNA methylation patterns in plants and animalsNature Reviews Genetics 11:204–220.https://doi.org/10.1038/nrg2719
Distinct activities of CHD1 and ACF in ATP-dependent chromatin assemblyNature Structural & Molecular Biology 12:160–166.https://doi.org/10.1038/nsmb884
RNA-directed DNA methylation: an epigenetic pathway of increasing complexityNature Reviews Genetics 15:394–408.https://doi.org/10.1038/nrg3683
T-Coffee: A novel method for fast and accurate multiple sequence alignmentJournal of Molecular Biology 302:205–217.https://doi.org/10.1006/jmbi.2000.4042
R: A language and environment for Statistical ComputingR Foundation for Statistical Computing, Vienna: Austria.
Effects of chromatin structure on the enzymatic and DNA binding functions of DNA methyltransferases DNMT1 and Dnmt3a in vitroBiochemical and Biophysical Research Communications 322:110–118.https://doi.org/10.1016/j.bbrc.2004.07.083
Exploiting induced and natural epigenetic variation for crop improvementNature Reviews Genetics 18:563–575.https://doi.org/10.1038/nrg.2017.45
Non-CG methylation patterns shape the epigenetic landscape in ArabidopsisNature Structural & Molecular Biology 21:64–72.https://doi.org/10.1038/nsmb.2735
Distinct DNA methylation activity of Dnmt3a and Dnmt3b towards naked and nucleosomal DNAThe Journal of Biochemistry 139:503–515.https://doi.org/10.1093/jb/mvj044
Mouse Dnmt3a preferentially methylates linker DNA and is inhibited by histone H1Journal of Molecular Biology 383:810–821.https://doi.org/10.1016/j.jmb.2008.03.001
Genome-wide nucleosome positioning during embryonic stem cell developmentNature Structural & Molecular Biology 19:1185–1192.https://doi.org/10.1038/nsmb.2419
Ggplot2 : Elegant Graphics for Data AnalysisNew York: Springer-Verlag. ISBN 978-0-387-98140-6.
Evolution of eukaryotic DNA methylation and the pursuit of safer sexCurrent Biology 20:R780–R785.https://doi.org/10.1016/j.cub.2010.07.007
Mechanisms of ATP-dependent chromatin remodeling motorsAnnual Review of Biophysics 45:153–181.https://doi.org/10.1146/annurev-biophys-051013-022819
Chromatin remodeling by ISW2 and SWI/SNF requires DNA translocation inside the nucleosomeNature Structural & Molecular Biology 13:339–346.https://doi.org/10.1038/nsmb1071
Richard AmasinoReviewing Editor; University of Wisconsin, United States
In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.
Thank you for submitting this interesting study "DDM1 and Lsh remodelers allow methylation of DNA wrapped in nucleosomes" for consideration by eLife. Your article has been favorably evaluated by Detlef Weigel (Senior Editor) and three reviewers, one of whom, Richard Amasino (Reviewer #1), is a member of our Board of Reviewing Editors. The following individual involved in review of your submission has agreed to reveal their identity: Vincent Colot (Reviewer #2).
The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision below to help you prepare a revised submission.
In the consolidated review below, some parts are directly from the first round of review and a ** was added to those parts when a comment to address was within a paragraph.
In this manuscript, Lyons and Zilberman examine the implication of the chromatin remodeler DDM1 and canonical histone H1 in the relationship between nucleosomes and DNA methylation in Arabidopsis. In previous work, the Zilberman lab showed that removal of canonical histone H1 leads to an increase in DNA methylation over heterochromatic TEs and a partial rescue of the well-documented loss of DNA methylation seen in ddm1. These observations lead to the conclusion that DDM1 provides DNA methyltransferases (MTases) with access to H1-containing heterochromatin (Zemach et al., Cell 2013).
Numerous in vitro and in vivo studies using mammalian MTases and de novo DNA methylation assays have indeed established that the nucleosome core is an obstacle to DNA methylation. The authors should cite Felle et al., NAR 2011 as well as Baubec et al., Nature 2015 in their introductory paragraph (Baubec et al. is only cited later, in a different context). In addition, the authors should cite Takeshima et al., J. Mol. Biol. 2008; in this paper it was shown that histone H1 blocks DNA methylation of linker DNA in vitro.
Lyons and Zilberman used MNase-seq and BS-seq to investigate directly DNA methylation in relation to nucleosome position in h1, ddm1 and h1ddm1 mutant plants as well as in WT controls. Key to their analysis is the definition of four groups of nucleosomes, from well-positioned to poorly-positioned. This distinction enabled the authors to report several new findings, which clarify the role of DDM1 in the maintenance of DNA methylation over H1-enriched nucleosomes.
1) First, they show that heterochromatic nucleosomes, which are enriched in histone H1, have invariably high average CG methylation across the core and linker in WT controls. This is strong evidence that, in vivo, core and linker DNA are equally accessible to MET1, which is responsible for the maintenance of CG methylation. In marked contrast, average CHG and CHH methylation, which is much lower than average CG methylation, is preferentially located over linker DNA for these same nucleosomes. Thus, the MTases responsible for the perpetuation of this non-CG methylation across cell divisions preferentially access linker DNA in vivo.
2/3) The second and key set of findings in this manuscript is that at presumably the same (**this is not clear from the text) well-positioned heterochromatic nucleosomes, the increase and decrease in average CG, CHG and CHH methylation observed in h1 and ddm1 mutant plants, is most and least pronounced over linker DNA, respectively. As could then be predicted, the partial rescue of methylation in the double mutant is also more pronounced over linker DNA. These findings are entirely consistent with the inhibitory effect in vitro of histone H1 on linker DNA methylation (Takeshima et al., J. Mol. Biol. 2008) and as noted above Takeshima et al. should be cited in this context. In addition, they suggest that DDM1 facilitates – to varying extents – the access of the different MTases responsible for CG, CHG and CHH methylation: not only to the core, but also to the linker, of H1-enriched, heterochromatic nucleosomes. **However, in the absence of genomic binding data for these MTases in WT and mutant backgrounds, indirect effects cannot be ruled out. This notwithstanding, the authors also used published nucleosome positioning and methylome data to show that loss of Lsh, the mammalian homolog of DDM1, causes a much greater depletion of DNA methylation over nucleosome cores than over linkers in mouse embryonic fibroblasts.
4) Coming back to Arabidopsis, the authors then show that in contrast to TEs (which comprise most heterochromatic nucleosomes), genes have higher average CG methylation over nucleosome cores than over linker DNA in WT plants. This pattern is evident for exons but not for introns (**an intriguing observation, which would merit being discussed by the authors) and it remains largely unaltered in the three mutant backgrounds, except in the case of lowly expressed genes, perhaps because these genes undergo less transcription-coupled nucleosome displacement (it is not clear though if these genes are not part of TEs). The 10 bp methylation periodicity seen over nucleosome cores in all sequence contexts is also analyzed and results indicate that in the case of heterochromatic nucleosomes, it is dependent on DDM1. Based on this last result, the authors propose a model whereby DDM1 acts by releasing a loop of DNA from the surface of the histone octamer.
In a last part, the authors show that a subset of heterochromatic nucleosomes that are well-positioned in the h1ddm1 double mutant exhibit the same DNA methylation pattern as seen in marine algae, namely a perfect alternation of low and high methylation over the nucleosome core and linker, respectively.
In sum, this manuscript reports many new findings that are of broad interest. It adds nicely to the two other comprehensive studies by the Zilberman lab (Zemach et al., Cell 2013; Huff and Zilberman 2014) and to that of Baubec et al. (Nature 2015) indicating that nucleosomes inhibit DNA methylation in vivo. In addition, it provides novel and compelling evidence that the linker histone H1 inhibits DNA methylation in vivo mainly by preventing access to linker DNA, consistent with in vitro data (Takeshima et al., J. Mol. Biol. 2008). Thus, as the authors suggest, the distinct patterns of DNA methylation observed among eukaryotes are likely determined in large part by whether or not a given species has well positioned nucleosomes and possesses the linker histone H1 and/or chromatin remodelers that can provide DNA MTases access to nucleosomes.
5/6) Overall, the data as well as the analyses presented in the manuscript are solid and comprehensive. **However, most analyses are described too succinctly and it is often difficult to understand what exactly was done (see below for a list of specific points that the authors need to address in addition to those raised in the general comments). Furthermore, the manuscript is difficult to read in places and I would recommend that the authors **subdivide it into three clearly identified sections (Introduction, Results, Discussion) as well as several sub-sections in Results. This division will also help separate what is a fair interpretation of the findings reported in the manuscript, from the more speculative conclusions.
7) One of the authors' main claims is that mutation of both DDM and Lsh1 bias methylation towards linker DNA. To support, and fully evaluate, this claim the following points should be addressed:
a) The conclusion from Figure 1 is that nucleosomes are refractory to DNA methylation without chromatin remodelers. However, it is unclear how the correlations between DNA methylation and nucleosomes were conducted. For panels A and C, either the methylation in all the mutants is shown relative to the nucleosome position in WT plants, in which case potential differences in nucleosome position in the mutants should be accounted for, or the methylation is shown relative the nucleosome positioning in each mutant, in which case the sequence differences in the nucleosome cores vs linker regions should be accounted for. Likewise, for the mammalian data, where the methylation data in the WT and lsh mutants is relative to the nucleosome positions in WT samples, changes in nucleosome positions in the lsh mutant should be accounted for in the analysis. Without such information it remains unclear whether the differences in methylation observed in the mutants are associated with changes in nucleosome positions, DNA sequence biases, and/or changes in the ability of DNA methyltransferases to methylate nucleosomal or linker DNA.
b) For the lsh mutant, a meta-analysis is presented showing decreases in DNA methylation over well positioned nucleosomes and near WT levels of methylation in the linker regions; thus, methylation is skewed towards linkers. For ddm1/h1ddm1 mutants a similar trend is observed. However, more in-depth analyses should be presented to strengthen the assertion that LSH and DDM1 are acting similarly. For example, is Lsh activity also mainly associated with heterochromatin? Is the methylation periodicity in mammalian cells disrupted in lsh mutants?
8) Another major point is that DDM1 enables the methylation of nucleosomal DNA by remodeling the chromatin such that DNA methylation actually occurs on a DNA loop and not directly on nucleosome bound DNA.
Here the authors suppose (without any data) that this loop would still retain a bias in DNA methylation resulting in a periodic pattern matching previous in vivo findings. To test their model, the periodicity of DNA methylation was then assessed in a ddm1 mutant background. However, as this mutant causes global decreases in DNA methylation, disruption of the periodic behavior is hard to assess. Furthermore, the decrease in periodicity could either be because the DNA loops aren't created so methylation is blocked or because DDM1 serves a different function that allows MET1 to directly methylate nucleosomal DNA. Thus, these in vivo experiments remain inconclusive. To support their conclusions, in vitro remodeling assays showing that DDM1 activity creates accessible DNA loops that can be methylated by MET1 would be highly informative, although such assays are clearly not trivial. In the absence of such assays, the authors should note the alternative interpretations noted above.
9) A final point of the paper is that the decoupling of chromatin remodeling and DNA methylation could explain some of the differences in global methylation patterns observed between species.
Here the authors find that a subset of well positioned, heterochromatin-localized nucleosomes (cluster C2) have strong linker methylation and low nucleosome methylation in the h1ddm1 background. Based on the similarity of this DNA methylation pattern with those observed in algae, it is proposed that decoupling methylation and remodeling is a mechanism employed to generate such patterns in organisms with well positioned nucleosomes. However, this model is not further vetted. In this case, a phylogenetic analysis correlating the presence or absence of DDM1 or H1 orthologs with linker-specific methylation patterns might provide additional support for the authors claims.
Although in general we appreciate brevity, some elaboration here would be useful to the reader. Do you think this phenomenon is a result of genetic drift in which in some organisms the loss of one component is not selected against or might it provide some selective advantage? In this section is the sentence: "These species are interspersed on the tree of life with species that do not limit methylation to linkers (Figure 4—figure supplement 1)." Figure 4—figure supplement 1 (does this mean Figure 4 and figure supplement 1?) does not seem to have much relation to the sentence that cites them – i.e., nothing to do with the phylogenetically "interspersed" nature of the phenomenon because both figures are about Arabidopsis.
1) Main text, first paragraph: Replace Jones and Baylin, 2007 with Baylin and Jones, CSH Persp. Biol. 2016, which is more appropriate and recent.
2) Main text, first paragraph: To be consistent with the first part of the sentence, I would be more general here and provide a review instead of Ong-Abdullah et al., 2015.
3) Main text, fourth paragraph: It is not clear if the four groups of nucleosomes were determined for each of the four genotypes or solely for the WT and then maintained for the analysis of the three mutant genotypes? I suspect that it is the latter, but the description in Materials and methods suggests otherwise. How can meaningful comparisons be made if the four groups of nucleosomes differ for each genotype?
4) The authors could easily complement their analysis of heterochromatic nucleosomes by considering separately those that have CMT2-dependent CHH methylation and those that are targeted by DRM1/2 instead. This additional analysis could also prove very informative.
5) Main text, fourth paragraph: Make also reference to the Materials and methods section ("Description of Arabidopsis genome features").
7) Main text, ninth paragraph: Are all genes non TE-genes? And what about TEs? Please provide about how the TE class was defined and how it differs from the heterochromatic nucleosomes analyzed in the first part.
8) Main text, tenth paragraph: The fact that gene methylation is enriched over nucleosomes and that DDM1 is not required for this type of methylation except in the case of lowly expressed genes suggests that other chromatin remodelers are involved. However, at this stage this is only a hypothesis and the authors should therefore tone down the last sentence of this paragraph.
9) Materials and methods:
- Why was the mutant allele ddm1-10 chosen over the much more commonly studied allele ddm1-2, used for instance in Zemach et al., 2013.
- Which generation was used for ddm1 and h1ddm1? This is important to know as the demethylation of heterochromatic sequences increases with the number of generations.
- Were DNA, RNA and nucleosomes extracted from the same starting biological material?
10) Figure 1A and C and Figure 1—figure supplement 2: I would recommend that the supplement be moved to Figure 1, as it is clear that CHH methylation is also affected in the three mutant backgrounds. I would also recommend for each of the three sequence contexts merging the four graphs, so as to make more apparent the gains and losses of methylation. This is particularly important for CHH methylation, which is low to start with compared to CHG and especially CG methylation and which show modest fluctuations in absolute values. Indicating levels relative to WT could improve visualization further.
11) The legend of Figure 4A is confusing as a result that figure is difficult to understand. The authors should also show mCHG and mCHH data or else justify why they chose not to show them.
In Figure 4B it is not clear what data is being shown in the right panels.
12) It is not clear from the methods if the same stage leaf tissue was used for the plant DNA methylation and nucleosome profiling experiments. As these features can vary during development, this point should be clarified.
[Editors' note: further revisions were requested prior to acceptance, as described below.]
Thank you for resubmitting your work entitled "DDM1 and Lsh remodelers allow methylation of DNA wrapped in nucleosomes" for further consideration at eLife. Your revised article has been favorably evaluated by Detlef Weigel (Senior Editor), a Reviewing Editor, and two reviewers.
There are a few minor points to address that the reviewers and I think would make your paper even more clear to the eLife readership.
1)We suggest a more direct formulation. "DNA methylation at DRM1/2-dependent loci is actually enriched around well-positioned nucleosomes in WT (Figure 2A-C). This enrichment is likely caused by the small size of DRM1/2-dependent TEs and hence the high proportion of non-methylated adjacent sequences over these nucleosomes. Consistent with this interpretation, an analysis anchored to the centers of differentially methylated regions (DMRs; see Methods)[…]"
2) Discussion, fourth paragraph, last sentence: This sentence is ambiguous. Gene body methylation is affected by MET1, with no contribution from the other methyltransferases. The ddm1 mutation affects only a small subset of body methylated genes.
3) Re the statement in the Abstract "We find that DDM1 enables methylation of DNA bound to the nucleosome, thus demonstrating that nucleosome-free DNA is the preferred substrate of eukaryotic methyltransferases in vivo." The first half of the statement is well supported, but the data do not demonstrate that nucleosome-free DNA is the preferred substrate. Rather, the data support this hypothesis. It is up to you how to re-phrase, but an example of re-phrasing would be "We find that DDM1 enables methylation of DNA bound to the nucleosome, consistent with a model in which nucleosome-free DNA is the preferred substrate of eukaryotic methyltransferases in vivo."
4) In several cases, you draw specifically on the effects of h1 on CG methylation (see quoted text below). However, the effects of h1 on CG methylation are minimal. Indeed, when plotted relative to WT levels in Figure 1B, the line is flat. However, h1 clearly has effect on nucleosome core methylation levels in the CHG and CHH contexts, and it also enhances the ddm1 mutant. In light of these observations the text should be modified to more accurately reflect the data presented.
"CG methylation is relatively evenly distributed with respect to nucleosomes in WT plants, and shows a weak nucleosomal depletion in h1 mutants (Figure 1A)."
"Without H1, linker DNA becomes even more accessible, producing the observed slight core-to-linker methylation differential in h1 mutants (Figure 1A).”
5) The text now makes clear that the methylation data shown in Figure 1 is relative to the specific nucleosome positions in each genotype. The authors also account for possible affects due to differences in the sequences in the nucleosomes and the linkers through their analysis in Figure 1—figure supplement 2. However, to fully evaluate this data, it would be informative to know how many well positioned nucleosomes there are in total and how many are shared between WT and h1ddm1 or between h1 and h1ddm1. In the response to reviewers it is stated that "only a limited subset of WT well-positioned nucleosomes are also well positioned in the mutants." If this overlap is very small, then it remains unclear if this subset of regions can be used to extrapolate information regarding global effects.
6) The implications of the following statement are unclear.
"However, this appearance is almost certainly caused by the small size of DRM1/2-dependent TEs, because an analysis anchored to the centers of differentially methylated regions (DMRs; see Methods) between drm1drm2 mutants and wild-type that occur in regions with very poorly positioned nucleosomes (those outside the four groups defined in Figure 1; see Methods) shows an even sharper peak (plots in the right column of Figure 2A-C)."
7) To avoid confusion the word "can" should be added to this sentence of the Introduction: "Methylation also regulates endogenous genes: methylation close to the transcriptional start site can cause gene silencing[…]"https://doi.org/10.7554/eLife.30674.027
- Daniel Zilberman
- Daniel Zilberman
- David B Lyons
- David B Lyons
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
We thank X Feng for helpful comments on the manuscript and T Baubec and D Schubeler for the list of CTCF binding sites, as well as P-H Hsieh for help in defining DMRs. We also thank S McDevitt for help with sequencing and C Wistrom for greenhouse assistance. This work was supported by a fellowship from the Helen Hay Whitney Foundation and NIH grant T32HG000047 to DBL, and European Research Council grant MaintainMeth (725746) and a Faculty Scholar grant from HHMI and the Simons Foundation (55108592) to DZ. Data generated from this study is available under GEO accession GSE96994. The authors declare no competing financial interests.
- Richard Amasino, Reviewing Editor, University of Wisconsin, United States
© 2017, Lyons et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.