Reduced intronic methylation of IBM1 is associated with ectopic genic hypermethylation in natural A. thaliana accessions

IBM1 is a histone demethylase that removes H3K9me2 from transcribed regions in A. thaliana [33,35,36]. The functional loss of IBM1 results in an accumulation of H3K9me2 and non-CG methylation (mCHG) in a subset of gene bodies [34,36,39,40]. The 7th intron of IBM1 is able to sense genome-wide levels of H3K9me2, which affects IBM1 transcription and H3K9 demethylase activity to maintain H3K9me2 levels genome wide [40]. Mutants deficient in DNA methylation, such as met1 and cmt3, exhibit a decrease in the transcription of full-length IBM1 transcripts along with a loss or reduction of DNA methylation in the IBM1 intron region [40]. In met1, decrease of IBM1 expression induces the ectopic gain of mCHG and H3K9me2 in multiple genes [40], similar to how it occurs in ibm1 [33–36,39]. Intriguingly, our previous research has identified three natural A. thaliana accessions exhibiting a weak ibm1-like phenotype, characterized by ectopic mCHG-gain in a subset of genes [39]. These accessions also showed a decrease in the DNA methylation level within IBM1’s 7th intron, suggesting that the heterochromatin sensor activity of IBM1 is potentially evolving [39].

To further explore the relationship between intron methylation of IBM1 and ectopic mCHG in gene bodies, we used RNA-seq isoform quantification using transcriptomes from the 1,001 genomes project (n = 635 with both RNA-seq and DNA methylation data) [16,43–45]. Our analysis revealed that eleven A. thaliana accessions possess ectopic mCHG in at least 120 genes (Fig 1A, labeled top eleven by name), thereby categorizing them as ibm1-like accessions. Notably, these ibm1-like accessions tend to show lower intron methylation in both CG and CHG contexts compared to the rest of the population (Fig 1B and 1C). The observed decrease in intron methylation is not due to genetic variation among the accessions, as the DNA methylation mapping coverage of the heterochromatin sensing intron was comparable across the accessions (S1A–S1D Fig and S1 Table). Furthermore, in accessions exhibiting low mCHG levels within the IBM1 intron, a positive correlation (p<0.035) was observed between intron DNA methylation levels and IBM1 gene expression (Figs 1D and S1E–S1G). Conversely, there was a negative correlation with the proportion of short isoform IBM1 transcripts (Figs 1E and S1H–S1K). Consistent with this observation, the ibm1-like accessions tend to exhibit reduced expression of IBM1 and an increased proportion of the IBM1-S/IBM1-L transcript isoforms compared to other accessions (Fig 1F and 1G).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Natural variation in intron DNA methylation of IBM1 within A. thaliana.

(A) The scatter plot shows the mCHG levels in IBM1’s long 7th intron against the number of genes with ectopic non-CG methylation in each accession. Accessions with a significant number (n>120) of ectopic methylated genes are labeled by name and marked with red dots, whereas other accessions are represented with blue dots. The box plot displays the distribution of (B) mCG and (C) mCHG levels in the long intron of IBM1, comparing the top 11 accessions (labeled in A) with the highest number of ectopic methylated genes against all other accessions. (D) The scatter plot shows the relationship between mCHG levels in the long intron of IBM1 and IBM1 gene expression for accessions with low IBM1 long intron mCHG levels (< = 0.035). The dots are colored following the same scheme as in (A). (E) Similar to D, this scatter plot presents the mCHG level in the long intron of IBM1 plotted against the ratio of short to long isoform of IBM1. (F) The box plot shows the distribution of IBM1 gene expression by comparing accessions that rank in the top 11 for the highest number of ectopic methylated genes with all remaining accessions. (G) Similar to F, the box plot shows the distribution of the ratio of short to long isoform of IBM1 between the two groups. (H) The browser shows the intron methylation of IBM1 in Col-0 WT, met1, cmt3, suvh456 and ddm1 mutants. (I) The bar plot shows the gene expression levels of the long and short isoforms of IBM1, as well as their combined expression, in ddm1, met1, suvh456 mutants, and Col-0 WT. The isoform expressions were quantified using RNA-seq data sourced from the Saze lab [47].

https://doi.org/10.1371/journal.pgen.1011358.g001

This diminished IBM1 activity likely contributed to the onset of ectopic genic methylation in these accessions. The reduced expression of functional IBM1 in these accessions raises a question regarding the plant’s potential regulatory response, specifically whether there is a compensatory downregulation of CMT3 expression to mitigate the effects of imbalanced genic methylation. However, our investigations reveal that CMT3 expression levels did not show significant alterations in the accessions exhibiting an ibm1-like molecular phenotype (S1L Fig). Collectively, our findings indicate a robust association between the DNA methylation status of the IBM1 intron, the ratio of the short and long transcript isoforms and the gene expression levels of IBM1. This relationship appears to be a significant factor influencing the widespread acquisition of mCHG in the gene bodies in a subset of A. thaliana accessions.

To validate the accuracy of RNA-seq based isoform quantification, we conducted a detailed analysis of IBM1 isoform expression across various DNA methylation mutant lines using different data sources, including met1, suvh456, cmt3, ibm2, ddm1, and drd1 (S2 Table) [11,41,46–48]. MET1 and CMT3 are responsible for maintaining CG and CHG methylation, respectively [6,7]. The SUVH4 (also called KYP), SUVH5 and SUVH6 proteins are involved in the H3K9me2 deposition, crucial for transposable element silencing and heterochromatin formation [13,17,41,42,49]. DDM1, a chromatin remodeler, is essential for maintaining heterochromatin across the genome [11]. DRD1 is also a chromatin remodeler which regulates RNA-directed DNA methylation pathway, which is required for silencing transposable elements through methylation [11,50]. Lastly, IBM2 plays an important role in ensuring RNA polymerase II processing at loci carrying heterochromatic transposable elements including IBM1 locus [41,42]. In Arabidopsis, the interaction between an IBM2/EDM2/AIPP1 complex and the heterochromatic repeat in the IBM1’s intron is essential for regulating IBM1 expression [51–53]. IBM2, an RNA-binding protein with a BAH domain, likely mediates chromatin association within the heterochromatin regions of the IBM1 intron, whereas EDM2 recognizes H3K9me2 within the intron [41,42,52]. AIPP1 serves as a bridge between IBM2 and EDM2, helping to form a protein complex that promotes full-length expression of the IBM1 transcript [53]. This system of interactions ensures precise control of IBM1 expression. The mutants showing a reduction in IBM1 intron methylation, such as suvh456, cmt3, and met1, demonstrated a decrease in full-length IBM1-L expression level (Figs 1H,1I and S2A–S2C). In ibm2, where the gene encoding the enzyme crucial for IBM1’s proper transcription is affected [41,42], also exhibited significant reduction of the IBM1-L transcript (S2D Fig). Conversely, in ddm1 and drd1 where the intronic methylation of IBM1 is unaltered, the transcription of the full-length transcripts remains unaffected (Figs 1I and S2C), indicating a direct correlation between intron DNA methylation status and the expression of functional IBM1-L isoform [40].

To assess the prevalence of the role of intron methylation in transcription, we analyzed all A. thaliana Col-0 genes with long introns (> 1kb). A total of 109 out of the 705 genes identified exhibit non-CG methylation within their introns (S3 Table). We observed that Col-0 genes with mCHG in the intron (n = 109) generally show less expression compared to those without mCHG (S2E Fig). However, for most Col-0 genes with mCHG in the introns, a reduction in methylation—ascertained by comparing suvh456 mutants with wild-type plants—does not influence the production of full-length transcripts (S4 Table). The notable exceptions to this are IBM1 and PPD7 (AT3G05410), a component of the thylakoid lumen proteome essential for the photosystem II oxygen-evolving complex in chloroplasts (S2F and S2G Fig). This association was previously established in ibm2 mutants [53]. Yet, in natural accessions, the pronounced effect of PPD7’s intron methylation reduction on PPD7 transcription observed in mutants does not persist (S2I Fig). For instance, PPD7’s intron methylation is completely absent in the Cnt-1 accession, which surprisingly does not impact the generation of full-length transcript of PPD7. Intriguingly, minimal read mapping occurs within the PPD7’s intron region in Cnt-1, a long region (5.7 kb) with a non-LTR retrotransposon in the Col-0 accession (S2H and S2I Fig). This suggests a possible deletion of the entire region in Cnt-1. Although some accessions exhibit low read coverage, the majority demonstrate adequate coverage; yet, even with low mCHG levels, no substantial effect on the transcription process is evident in most accessions (S2I and S2J Fig). This indicates that factors beyond total intron deletion, such as TE sequences within these large introns, also may contribute to variations in transcriptional responses. This is supported by a recent study showing that intragenic transposable elements can modulate transcription initiation, termination and splicing [54].

The ectopic mCHG in genes in Cnt-1 is reduced by IBM1 overexpression

The A. thaliana Cnt-1 accession stands out as an exceptional case, exhibiting the largest number of ectopic mCHG-gain genes among the natural accessions along with lower levels of intron methylation (Fig 1A). As this phenotype resembles that of a weak ibm1 mutant, we explored isoform quantification analysis of Cnt-1 to estimate IBM1’s expression level. Notably, Cnt-1 displayed a substantially lower expression level of the functional IBM1-L isoform in comparison to the Col-0 reference accession, with only a marginal reduction in the expression of the IBM1-S isoform (Fig 2A). This result suggests that decreased expression of IBM1-L in Cnt-1 leads to an elevation in mCHG within these genes. Moreover, TE methylation in Cnt-1 is relatively high among all accessions (S1M Fig and S1 Table). This combination of subdued IBM1 activity and vigorous CMT-KYP methylation may account for the high number of mCHG-gain genes in Cnt-1 and supports that this heterochromatin sensing mechanism is dynamic in populations to shape the epigenome.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The ectopic mCHG in genes in Cnt-1 is reduced by IBM1 overexpression.

(A) Expression level of IBM1-L and IBM1-S in Col-0 and Cnt-1. (B) Expression level of IBM1 in each transgenic plant. Primers were designed to target the conserved region of IBM1 between Col-0 and Cnt-1. RT-qPCR was performed to estimate the expression level of IBM1 in these plants. (C) A genome browser view shows that ectopic mCHG in genes in Cnt-1 is reduced by introducing the IBM1 transgene. (D) Metaplots show the average changes to mCHG over gbM and UM genes in IBM1 transgenic lines of Cnt-1. (E) The number of mCHG-gain genes is reduced in IBM1 transgenic lines compared to Cnt-1. (F) Heatmaps show reduction of mCHG on genes in IBM1 transgenic lines of Cnt-1. The 636 mCHG-gain genes from (E) are shown.

https://doi.org/10.1371/journal.pgen.1011358.g002

Next, we examined the effect of restoring IBM1 expression in the Cnt-1 accession. The UBQ10 cis-regulatory sequences were used to express the Col-0 IBM1 coding sequence in Cnt-1, resulting in the generation of three independent transgenic lines. These lines exhibited a significant increase in IBM1 expression, ranging from 20-100-fold higher than the control (Fig 2B). All three of the IBM1-ox lines displayed a decrease in mCHG in a subset of gene bodies compared to the Cnt-1 control (Fig 2C and 2D), indicating that overexpression of IBM1 reduces ectopic mCHG in gene bodies presumably by decreasing H3K9me2, although this was not tested. In Arabidopsis, H3K9me2 and mCHG form a self-reinforced loop, thus, measuring mCHG levels can indirectly reflect H3K9me2 levels [8,16,17]. This suggests that the observed reduction in mCHG are likely indicative of decreased H3K9me2 levels. This reduction was further supported by a decrease in the number of mCHG-gain genes when compared to the Cnt-1 control (Fig 2E). A heatmap analysis of mCHG-gain genes from Cnt-1 revealed a substantial decrease in mCHG levels in the IBM1-ox lines (Fig 2F). Despite the substantial increase in expression levels (Fig 2B), complete mCHG reduction was not achieved (Fig 2C–2F), which could be attributed to promoter choice. Although we used the UBQ10 promoter for its strong and constitutive expression, this may not fully replicate the native expression pattern of IBM1, particularly in tissue-specific contexts where IBM1 function is crucial, such as during meiosis [38]. This discrepancy could explain the partial reduction in mCHG levels observed, suggesting that the expression driven by UBQ10 does not fully mimic the native promoter’s regulatory effects. Considering the conservation of IBM1 coding sequences between Col-0 and Cnt-1 (99.61% sequence identity), these findings collectively indicate that increasing IBM1 expression in Cnt-1 significantly reduces the aberrant mCHG accumulation in the gbM genes. We extended the analysis of methylation beyond gene regions to include TEs and TE genes (S3A Fig). When comparing the methylation profiles of the Col-0 and Cnt-1 lines, both generally exhibited similar methylation levels. However, the IBM1-ox lines displayed methylation patterns similar to Cnt-1 for mCG, while mCHG and mCHH levels were slightly reduced. Furthermore, we assessed global DNA methylation levels (S3B Fig), revealing that, aside from minor differences in mCG levels compared to the Cnt-1 control, there were notable reductions in mCHG and mCHH levels across the pericentromeric regions of the genome. This suggests that IBM1 overexpression has a broad impact on methylation, affecting regions well beyond gene bodies.

Methylation of the intronic heterochromatin sensor is a common feature in IBM1 orthologs across flowering plants

We noted a heightened mutation frequency within the methylated intron of IBM1 compared to the unmethylated non-coding regions and the coding sequences of IBM1 across A. thaliana populations as well as at the species level (S4A and S4B Fig and S5 Table). In addition, methylated long introns in A. thaliana (> 1kb) exhibit a higher mutation frequency, suggesting the inherent mutagenicity of DNA methylation in A. thaliana populations. (S4C Fig). Given these observations, further exploration into the variability of IBM1 intronic DNA methylation and the conservation of its regulatory role across different species presents an intriguing line of inquiry. To investigate the patterns of intronic DNA methylation of IBM1 across angiosperm species, we first identified the orthologous genes of IBM1 in various plant species. This was achieved through the construction of a Maximum Likelihood (ML) gene tree, using 1,516 JmjC family homologous genes derived from 597 different species (see Methods and S6 Table). The orthologs of IBM1 were defined as those genes that clustered within the same clade as A. thaliana’s IBM1, specifically clade 5 (S4D–S4F Fig).

We analyzed the DNA methylation patterns on IBM1 orthologous genes in 34 plant species for which DNA methylome data were available (S7 Table) [19]. This set includes A. trichopoda, the most basal species in our study, which was used as an outgroup. Our investigation focused on identifying introns within these IBM1 orthologs that exhibit significantly higher levels of mCHG methylation compared to the species-specific background genic methylation levels (S8 Table). Our findings reveal that intronic CHG methylation is prevalent in IBM1 orthologous genes across both eudicots and monocots (Figs 3A,3B and S5A–S5C). Among the 65 IBM1 orthologs analyzed, which include species with multiple gene copies, 27 IBM1 orthologs were significantly enriched for mCHG within their introns (Fig 3C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Intron methylation is widely present in IBM1 homologous genes of angiosperms.

(A) The maximum likelihood gene tree displayed here represents the IBM1 orthologs from 34 angiosperm plants that have DNA methylome data. Due to space constraints, this is a truncated version of the tree; the complete version is available in S5 Fig. The left panel presents the gene tree, while the right panel depicts the corresponding gene structures. UTRs, CDS, and the JmjC domain are highlighted in different colors, as indicated in the legend, with introns represented by thin black lines. The locations of methylated sites on the gene sequences are marked with vertical bars, with distinct colors denoting the different methylation contexts: CG, CHG, and CHH. (B) Zoomed-in version of (A). A few species were enlarged to highlight the intron methylation. (C) The bar plot shows the number of genes out of the 65 IBM1 orthologs investigated that are either with or without mCHG in their introns. To assess the significance of mCHG levels within these introns, a binomial test was applied. (D) The bar plot displays the count of introns with mCHG relative to their position within, before, or after the JmjC domain. It should be noted that a single IBM1 ortholog may contain multiple methylated introns. (E) The boxplot shows the distribution of intron lengths between IBM1 gene copies with methylated introns and those with unmethylated introns. (F) The boxplot shows the distribution of expression levels between IBM1 gene copies with methylated introns and those with unmethylated introns.

https://doi.org/10.1371/journal.pgen.1011358.g003

Within the Brassicaceae family, intronic DNA methylation is present in the IBM1 orthologs in six out of nine species, encompassing all species from the A. thaliana subclade (Fig 3A). The intronic methylation observed in these IBM1 orthologs is predominantly found within the JmjC domain, a pattern that is consistent with what has been observed in A. thaliana (Fig 3B). A varied pattern emerges in the E. salsugineum subclade: B. rapa and B. oleracea exhibit significant mCHG enrichment in the long intron region within the JmjC domain, whereas T. parvula and the species with shorter introns, E. salsugineum and T. arvense, show a loss of mCHG (Fig 3A and 3B). This is noteworthy, as the species with no or reduced DNA methylation in the IBM1 intron region containing the heterochromatin sensor are the species that have lost gbM [28,55]. For species outside the Brassicaceae family, the patterns of intronic methylation in IBM1 orthologs are more variable and not strictly confined to the JmjC domain. DNA methylation is observed in regions both upstream and downstream of the JmjC domain (Figs 3A, 3D and S5B), and this variability in DNA methylation location does not appear to be conserved in species from the same family. For instance, in the Fabaceae family, the IBM1 orthologs in L. japonicus and G. max exhibit intronic methylation after the JmjC domain, whereas M. truncatula from the same family shows DNA methylation within the JmjC domain (S5C Fig). A similar pattern is noted in the Poaceae family: an IBM1 ortholog in P. virgatum has intronic DNA methylation within the JmjC domain, whereas in O. sativa, it is located before the JmjC domain. The mCHG introns are typically more enriched in longer introns (Fig 3E). For instance, a gene copy in M. esculenta, with three successive large introns (7.3kb, 3.6kb, 9.6kb), and another in T. cacao, with a 27.2 kb intron, both exhibit significant DNA methylation enrichment within these intron regions (S5A and S5B Fig). Interestingly, the IBM1 orthologs with mCHG introns also exhibit higher expression levels. (Fig 3F). These data support that many of these species likely use a similar mechanism to A. thaliana to sense heterochromatin content given the persistence of intron methylation within the JmjC domain of IBM1.

The DNA sequence underlying the heterochromatin sensor of IBM1 is evolving

In A. thaliana, DNA methylation in the 7th intron is present within a 150-bp repeat sequence that shows high similarity to YCF1, a gene encoded by the chloroplast genome [41]. Organellar sequences are often silenced by DNA methylation upon integration into the nuclear genome. In the IBM1 ortholog from A. lyrata, the DNA methylated region in the 7th intron is larger than that in A. thaliana, as it arises from two insertions of simple repeat elements approximately 200 bp upstream and 50 bp downstream of the sequence aligned with the DNA methylated intron of A. thaliana (Fig 4A). Intriguingly, this YCF1-like fragment is identified in A. lyrata, but not other Brassicaceae species, suggesting significant evolutionary divergence (S9 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. The sequence underlying the methylated intron of IBM1 is highly divergent between species.

(A) The plot illustrates the sequence comparison between A. lyrata and A. thaliana, which reveals an insertion of approximately 500 bp in the intron of A. lyrata. This inserted fragment comprises repeat sequences that are methylated. (B) The pie chart shows the proportion of DNA repeat types found in the introns covered by mCHG across the 65 IBM1 orthologs. (C) The bar plot illustrates the distribution of primary DNA repeat types across the introns of IBM1 orthologs in 34 species. (D) Gene structure diagrams show the different intron methylation patterns of duplicated IBM1 genes in B. stricta and C. rubella. (E) The bar plot displays the expression levels of duplicated IBM1 genes, categorized by the presence or absence of intronic methylation in B. stricta and C. rubella, respectively. (F) The bar plot displays the distribution of intronic methylation variation among 19 species that have duplicated IBM1 gene copies. The categories are defined as follows: ’All’ indicates that every IBM1 gene copy in a species has a methylated intron, ‘absent’ signifies that no copies of the IBM1 genes in a species have methylated introns, and ’variable’ denotes species where the methylation status differs among the IBM1 gene copies.

https://doi.org/10.1371/journal.pgen.1011358.g004

It is likely the heterochromatin sensing ability of IBM1 is conserved throughout many flowering plants given the presence of mCHG in introns within the JmjC domain of many species (Fig 3A and 3D). However, the sequence underlying the methylated introns is not conserved within species (Fig 4B, 4C and S9 Table). We found numerous alignments of simple and transposon repeats showing that sequence evolution within the intronic heterochromatin sensing region of IBM1 is quite dynamic between species (Fig 4B and 4C). For species that have a candidate heterochromatin sensor in an IBM1 intron, the actual sequence that is methylated isn’t as important as the methylation event itself. The fact that the sequence is continually changing between species, but remains methylated in many flowering plants suggests that there could be constant gains and loss of IBM’s ability to sense H3K9me2 levels over evolutionary time.

In fact, DNA methylation patterns in the intronic region of IBM1 orthologs not only varied among different species, but also among duplicated gene copies within the same species. For instance, a tandem duplication event happened just before the speciation of B. stricta and C. rubella, resulting in each species possessing two IBM1 gene copies, and DNA methylation and gene expression profiles also varied between the two gene copies (Fig 4D and 4E). Approximately 20% of species in this clade possess duplicated IBM1 genes (S4G Fig) and out of the 34 species for which there is DNA methylation data, 19 species contain duplicated IBM1 genes, and over half exhibit DNA methylation pattern divergence between the duplicates (Fig 4F). This indicates that upon duplication the heterochromatin sensor activity is not retained in duplicate copies.

Gene body DNA methylation is frequently lost in Brassicaceae species due to reduction of CMT3 or IBM1 activity

Previous studies have shown that loss of CMT3 or reduced CMT3 activity is associated with a loss or a reduction of gbM [28,31]. Importantly, the loss of CMT3 has occurred multiple independent times suggesting that this is an evolving process [28,31]. One possible reason for the loss of CMT3 or a reduction in its activity could be due to reduced function of IBM1. Loss of IBM1 activity in A. thaliana leads to reduced expression of gbM genes and eventually lethality [36]. However, this is rescued by reduced activity or a complete loss of CMT3 [36,38]. To explore this possibility further, we analyzed transcriptomes and DNA methylome data from various Brassicaceae species, specifically focusing on CMT3, IBM1, and KYP expression and the prevalence of gbM (Fig 5A). In E. salsugineum, a noticeable decrease in IBM1 expression accompanies the loss of CMT3, coinciding with a substantial reduction in gbM. A similar pattern emerges in T. arvense, closely related to E. salsugineum. Although T. arvense retains CMT3, both IBM1 and CMT3 exhibit low expression levels, correlating with an absence of gbM. This trend persists within the Brassicaceae subclade; species like B. rapa and B. oleracea, showing diminished CMT3 and IBM1 expression, also display a marked decrease in gbM genes (Fig 5A and S10 Table). This is consistent with our previous results showing CMT3 in some Brassicaceae is under relaxed selective constraints [30].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Frequent expression and genetic variation of IBM1 and CMT3 in Brassicaceae.

(A) The heatmap on the left panel displays the expression levels of CMT3, IBM1 and KYP in Brassicaceae species. On the right panel, the scatter plot shows the gbM gene ratio for each species, which is calculated as the number of gbM genes divided by the total number of genes where DNA methylation could be measured in that species. The size of the points in the scatter plot is proportional to the gbM gene ratio. (B) A genome browser view of a region of synteny between I. lusitanica and A. thaliana depicting the absence of IBM1. (C) The mCG metaplots display the methylation patterns across all genes in newly analyzed Brassicaceae species I. lusitanica and R. raphanistrum. (D) Box plots showing per-site mCHG level across the genome, which reflects the enzyme activity of CMT3.

https://doi.org/10.1371/journal.pgen.1011358.g005

By mining publicly available reference Brassicaceae genomes, we identified two additional species (Raphanus raphanistrum and Isatis lusitanica) with potential loss or truncations of IBM1 or CMT3 (S6A Fig). However, the quality of these reference genomes was quite variable, requiring additional experiments to validate the loss or truncations of IBM1 or CMT3. We used transcriptome assembly as well as PCR of genomic DNA using universal primers to extract full-length IBM1 or CMT3 loci (S11 Table). We were unable to identify IBM1 in I. lusitanica suggesting it might be lost in this species. There is clear evidence that it is deleted within a region of synteny with A. thaliana (Fig 5B). We successfully identified CMT3 sequences across both species, although I. lusitanica exhibited a small N-terminal deletion in CMT3 (S6B Fig and S12 Table). Next, we estimated the expression level of IBM1, CMT3 and KYP across these species and evaluated the presence of gbM genes (Fig 5A). The number of potential gbM detected in I. lusitanica and R. raphanistrum was as low as E. salsugineum (Fig 5A and 5B) indicating it has been lost in these species. In R. raphanistrum, CMT3 expression was absent, coupled with a significant reduction in IBM1 expression. In I. lusitanica, in which we couldn’t identify IBM1, there were relatively lower CMT3 expression levels compared to Brassicaceae species with high numbers of gbM genes (Fig 5A). We also assessed the expression levels of CMT1 and CMT2, genes orthologous to CMT3, in species exhibiting reduced or absent IBM1 expression, finding no significant changes in their expression levels (S7 Fig).

When measuring gene body mCG, A. thaliana exhibited a typical bell-shaped gbM pattern (Fig 5C). However, both I. lusitanica and R. raphanistrum lacked this pattern, similar to E. salsugineum, which correlates with a substantial reduction of gbM (Fig 5A and 5C). Notably, the basal mCG levels in I. lusitanica and R. raphanistrum were slightly higher than in E. salsugineum (Fig 5C). This could be the result of higher methylation levels in TE-like methylated genes, which increases basal mCG levels. This is further supported by the fact that mCHG levels in I. lusitanica and R. raphanistrum were also slightly higher than in E. salsugineum (S8A Fig). We also analyzed the methylation levels in TEs. Our analyses reveal that R. raphanistrum exhibits higher methylation levels of both CG and CHG in TEs compared to A. thaliana, while I. lusitanica shows higher CG but lower CHG methylation levels than A. thaliana (S8B Fig). These differences in TE methylation patterns might contribute to the observed variations in gene body methylation and overall methylation profiles among these species. We hypothesized that even though some of the Brassicaceae species have CMT3, its expression or enzymatic activity might be reduced or absent, consistent with the reduction of gbM (Fig 5A). We tested this by measuring mCHG levels and mCHG symmetry, both of which should be present if CMT3 activity is functional. Genome-wide per-site mCHG levels were measured across species, revealing R. raphanistrum and I. lusitanica with low mCHG levels, similar to gbM-absent species such as E. salsugineum (Fig 5D). In Arabidopsis, mCCG is underrepresented compared to other CHG sites, as the CCG methylation depends on MET1 as well as CMT3 [56]. We have conducted additional per site methylation analysis on CCG sites and confirmed that they show low mCCG level in Arabidopsis as expected (S8C Fig). We also assessed the methylation symmetry at CWG sites (W = A or T) preferentially targeted by CMT3. Furthermore, R. raphanistrum and I. lusitanica exhibited a predominantly asymmetric mCWG pattern, aligning with the pattern observed in E. salsugineum, which lacks CMT3 (S9 Fig).

WGBS and RNA-seq data acquisition

We obtained whole genome bisulfite sequencing (WGBS) and RNA-seq data for natural accessions of A. thaliana from the 1,001 Genomes Project [43–45,63]. The WGBS and RNA-seq data for A. thaliana mutants were obtained from NCBI [11,41,46–48], with specific details provided in S2 Table. Additionally, we acquired WGBS and RNA-seq data for 34 angiosperm species from NCBI [19,28,32,64–67], with the sources detailed in S7 Table. Furthermore, we generated WGBS and RNA-seq data for two Brassicaceae family species, and both the sequencing and processed data are available at NCBI GEO database (GSE252913).

Plant materials and transgenic line analysis

The IBM1 coding sequence from Col-0 was driven by the UBQ10 promoter. The construct was transformed into the A. thaliana Cnt-1 accession by Agrobacterium-mediated flower dipping method [68]. Three transgenic lines were selected and RNA was extracted to estimate IBM1 expression. The RT-qPCR analysis was conducted using the Luna Universal One-Step RT-qPCR Kit. The expression levels of IBM1 were quantified and subsequently normalized against the expression levels of UBQ10. The genomic DNA from those lines were subjected to WGBS analysis. For the DNA methylation and RNA-seq analyses of the two Brassicaceae species, namely Raphanus raphanistrum, and Isatis lusitanica, the plants were grown under identical conditions to those used for A. thaliana. Leaf tissues were collected from each species at the 8-week-old growth stage for subsequent analyses.

Whole genome bisulfite sequencing library preparation

Libraries were prepared following the MethylC-seq protocol [69]. Briefly, genomic DNA was isolated from leaf tissues using the DNeasy Plant Mini Kit (Qiagen). Subsequently, genomic DNA was sonicated to achieve 200 bp fragments, and then end-repair was performed by the End-It DNA End-Repair Kit (Epicentre). This end-repaired DNA was subjected to A-tailing using the Klenow 3′–5′ exo− enzyme (New England Biolabs). The subsequent step involved the ligation of methylated adapters to the A-tailed DNA, using T4 DNA Ligase (New England Biolabs). Following adapter ligation, the DNA was bisulfite converted with the EZ DNA Methylation-Gold Kit. Finally, the library was amplified using KAPA HiFi Uracil + Readymix Polymerase (Roche).

DNA methylation analysis

WGBS data were processed using Methylpy [70], following the methodology outlined in reference [39]. Initially, read quality filtering and adapter trimming were conducted using Cutadapt v1.9.dev1. The qualified reads were then aligned to the species-specific reference genome using Bowtie 2.2.4 [71], ensuring that only uniquely aligned and non-clonal reads were retained. The genome assembly version of 34 species used for mapping are provided in S7 Table. To calculate the non-conversion rate of unmodified cytosines in the sodium bisulfite reaction, a fully unmethylated sequence, chloroplast or lambda (see unmethylated sequence used for each species in S7 Table), was used as a control. A binomial test, requiring a minimum coverage of three reads, was used to determine the DNA methylation status of cytosines.

To determine gbM genes in each of the 34 species, we counted the number of methylated and total cytosines for each methylation context (CG, CHG, and CHH) within the coding regions of primary transcripts for each gene. We then calculated the percentage of methylated sites for each context across all coding regions in each species. This percentage served as the background probability of methylation at a single site within coding sequences (CDS). Using this background probability, along with the total counts of cytosines and methylated cytosines on CDS, we calculated p-values based on a binomial distribution. These p-values represent the cumulative probability of observing a greater number of methylated cytosines in a given gene than expected by chance. Subsequently, we adjusted the p-values using the Benjamini–Hochberg False Discovery Rate (FDR) method to compute q-values. A gene was classified as having gbM if it had reads mapping to at least 20 CG sites, a q-value less than 0.05 for mCG, and q-values greater than 0.95 for both mCHG and mCHH. To calculate the gbM ratio, the total number of gbM genes was divided by the overall count of genes with adequate coverage in each species. This process of calculating gene coverage required counting CG sites within each gene, selecting those genes where at least 40% of the CG sites had a minimum coverage of three reads or more.

In natural A. thaliana accessions, we identified ectopic non-CG genes characterized by high non-CG methylation (mCHG and/or mCHH with a q-value less than 0.05) in certain accessions, while typically existing as gbM genes in over 90% of all accessions. Specifically, in Cnt-1, which has the highest number of ectopic mCHG in genes, we compared the mCHG levels of these genes between the wild type Cnt-1 and an IBM1 overexpression transgenic line in Cnt-1 (IBM1-ox). The average DNA methylation ratio of each gene was calculated, and the number of the gbM genes that gained more than 2% mCHG (> 0.02) was plotted. The DNA methylation ratio of the mCHG-gain gbM genes in Cnt-1 (n = 636) were plotted as heatmaps.

Using binomial tests, we identified introns enriched with mCHG in IBM1 of different species. For this, we counted the methylated and total cytosines for each methylation context (CG, CHG, and CHH) specifically within the 7th intron of IBM1. We then calculated the percentage of methylated sites for each context in all introns for each species. This species-specific percentage served as the background probability. With this background probability and the total counts of cytosines and methylated cytosines in the IBM1 intron region, we computed p-values using a binomial distribution. These p-values were then adjusted using the Benjamini–Hochberg False Discovery Rate (FDR) method to derive q-values. An intron was classified as mCHG-enriched if its mCHG q-value was less than 0.05.

In the genic metaplots for A. thaliana, we divided the gene body into 20 equal windows. Similarly, the 1,000 base pairs upstream and downstream of the gene were each divided into 20 windows. Within each window, we calculated the weighted DNA methylation [72]. Subsequently, we computed the average weighted methylation for each window across all genes. These average values were then plotted using R to create the metaplots.

Per site methylation and symmetry analysis

The methylation ratios at individual CHG sites were calculated under the condition that each CHG site exhibited a minimum read coverage of three. Additionally, a CHG site was included in the analysis only if at least one CHG site was identified as being methylated. For the analysis of methylation symmetry, we selectively focused on CWG sites (W = A or T) to exclude the influence of MET1’s activity on CCG sites. Both strands of CWG sites were required to have a minimum coverage of at least three reads, and at least one CWG site was identified as being methylated.

Isoform quantification for RNA-seq data

Quality filtering and adapter trimming of the RNA-seq reads were conducted using Trimmomatic v0.33 [73], using default parameters. For each species, the processed reads were then aligned to their respective transcriptome fasta files using Kallisto v0.50.0 [74] for transcript quantification. Specifically in A. thaliana, to quantify the expression of short transcripts in genes with long introns, we generated truncated versions of these transcripts, comprising only the UTR and CDS regions preceding the long intron. These truncated transcripts were subsequently incorporated into the transcriptome files used for mapping. Similarly, for other Brassicaceae species with long introns in IBM1, we created and added shortened versions of IBM1 transcripts into their respective transcriptome files for precise quantification of the short isoform expression of IBM1.

Identification of IBM1 and CMT3 in the Brassicaceae species

The orthologs of IBM1 and CMT3 across four Brassicaceae species were identified using OrthoFinder [75] and the extracted coding sequences were used for phylogenetic analysis using MEGA [76]. The de novo transcriptome assembly was performed using Trinity [77]. The presence/absence of the IBM1 and CMT3 transcripts were identified by BLASTing the A. thaliana gene sequence against the assembled transcriptome. To amplify the IBM1 and CMT3 gene from the genomic DNA, the degenerate primers were designed (S11 Table), and the genes were PCR-amplified using Q5 High-Fidelity 2X Master Mix (New England Biolabs).

Homologous genes identification and phylogeny analysis

We extracted the JmjC gene family from the One Thousand Plants (1KP) Consortium’s orthogroupings [78], using the A. thaliana IBM1 gene identifier (AT3G07610). The 1KP Consortium identified a single orthogroup that encompasses the IBM1 proteins, along with three other JmjC family genes (JMJ27, JMJ26, JMJ29) from A. thaliana, with a total of 9,258 protein sequences. The corresponding coding sequences (CDS) for these genes were also obtained from the 1KP Consortium. We expanded our dataset to include sequences from 69 species not covered in 1KP, including 45 from the Brassicaceae family. Their annotated CDS and protein sequences were obtained from Phytozome or EBI, with specific data sources detailed in S6 Table. For these species not included in iKP orthogroup, protein sequences showing reciprocal best BLAST hits with A. thaliana JMJ27 (AT4G00990), JMJ26 (AT1G11950), JMJ29 (AT1G62310), and IBM1 (AT3G07610) were added in the orthogroup. In total, the JmjC gene family included 9,521 sequences from 1,130 species. Then, according to Interproscan, sequences were retained if they included the same PFAM domain (JmjC domain PF02373) or ProSiteProfiles domain (JmjC domain profile PS51184) as A. thaliana. These filtered sequences included 1,528 sequences from 602 species, from which 10 sequences were excluded due to discrepancies between their protein and CDS sequences in terms of codon-to-amino acid conversion.

Then, to estimate the JmjC gene tree, a protein alignment was carried out using Pasta [79] with the default setting. The resulting alignment was back-translated using the coding sequence into an in-frame codon alignment. Then, Gblocks was used to retain only conserved codons with default settings, but allowing for a 50% gapped position. This conserved codon alignment then served as the input for phylogenetic estimation using RAxML [80], which included 500 rapid bootstrap replicates. The generated tree was rooted at the green algae clade and subsequently edited using the R package ggtree. From this comprehensive gene tree, a subtree comprising IBM1 orthologous genes from 34 species with available methylome data was extracted using ggtree.

Identification of introns where decreased mCHG correlates with full-length transcripts in A. thaliana

For each gene featuring a long intron (≥1kb) enriched with mCHG, we determined the average expression of both short and long isoforms across replicates, conducting these calculations independently for wild-type and suvh456 samples. Subsequently, we identified genes exhibiting a difference greater than 2 in the short-to-long isoform ratio between suvh456 mutants and wild type.

Calculation of SNP mutation rates

To determine the SNP density within the A. thaliana population, we divided the SNP genotype data of IBM1 genes (sourced from the 1,001 Genomes Project) into three groups: SNPs in the CDS region, SNPs within the methylated 7th intron, and SNPs in other intron regions not covered by methylation. We then calculated the SNP density and Tajima’s D values separately for each of these SNP matrices. Tajima’s D values were calculated by R package PopGenome while SNP density was calculated as the proportion of sites with SNP in each of three groups.

To calculate the nucleotide mutation rate between A. thaliana and A. lyrata, we first generated separate sequence alignments for the coding sequences (CDS) and intron regions of both species. The intron regions were further categorized into methylated regions and non-methylated regions. For each of these three groups—CDS, methylated introns, and non-methylated introns—we counted the number of identical nucleotides shared between the two species. The nucleotide difference for each group was then determined by calculating 1 minus the proportion of identical bases.

Annotation of repeats and motifs in IBM1 introns across various species

The DNA sequences of the introns for each of the 34 species were extracted from their respective genome assemblies using GFF annotations. RepeatMasker [81] was then used to identify various types of transposable elements, utilizing Repbase [82] as the reference library.