The Arabidopsis thaliana, Col-gl (Columbia-glabrous) ecotype was used for embryo isolation. Plants were cultivated in a controlled environment room with a long photoperiod of 16 hours of light and 8 hours of darkness, at a temperature of 22°C with cool white fluorescent light (100 mmole/m²/s).

After 24 hours of emasculation, fully matured WT ovules were pollinated with WT pollen. The pollinated plants were then incubated in the growth room until they reached a suitable stage for sampling. Arabidopsis embryos at DAP4, 5, 7, 9, and 12, corresponding to the globular, heart, torpedo, bending torpedo, and mature green stage embryos, respectively, were dissected from seeds using various methods specific to each stage. Globular embryos were obtained by gentle grinding the seeds with a pestle, while heart embryos were excised by making incision of the seeds with tweezers. For later-stage embryos, from torpedo to mature green, embryos were carefully extracted by puncturing the seed coat and endosperm with tweezers, followed by pushing embryos out. Each released embryo was manually collected using microcapillaries, one by one. All procedures used in this study, including pollination for embryo preparation, sampling of the embryos, genomic DNA extraction, bisulfite library construction, and the processing of whole-genome bisulfite sequencing (WGBS) data, were previously described in detail (Yoo et al., 2021).

Embryo DNA methylation data was divided into 50 bp fractional windows that contained at least three cytosines with at least five aligned reads in all stages for each cytosine context. For the circular global map, we used 5kb windows that contained more than ten cytosines with at least five reads for each methylation context.

The small RNA-Seq dataset was obtained from the NCBI GEO (GSE98553, GSE132066, GSE152971) (Lutzmayer et al., 2017; Plotnikova et al., 2019; Papareddy et al., 2020). We performed adapter trimming, quality control, and 18-to-26 nt-length read selection using Trimmomatic (0.39) and Cutadapt (v3.4). The pre-processed reads were mapped to rRNA/tRNA/snRNA sequences (RNACentral, v17) using bowtie (v1.3.0, with parameters -v 1 -m 0 -a) to filter out structural non-coding RNA reads. The remaining reads were first aligned using bowtie (1.3.0, -v 0 -m 0 -a), and then the aligned reads were re-aligned with ShortStack (v3.8.5, with parameters –align_only –mismatches 0 –mmap u –bowtie_m 1000 –ranmax 3) using the TAIR10 genome sequence as a reference. Reads aligned to known microRNA precursors were removed using Bedtools intersect (v2.30.0). The microRNA-free reads were then used as input for downstream analysis. Small RNA read clustering was conducted in a sample-by-sample manner using ShortStack (with parameters –dicermin 20 –dicermax 24 –foldsize 300 –pad 1(or 75), –mincov 1.0rpmm –nohp) (Johnson et al., 2016).

To calculate the average length of the regions where consecutive methylated probes were present at each stage of embryo development, we used 10 bp windows with at least one methylation site and at least five aligned reads. We merged windows with methylation levels above 5%, 10%, or 15% in all stages for each cytosine context if they were closer than 40 bp apart. The resulting merged windows were used to determine the average length of the methylated regions at each stage.

To identify regions with consecutive CHH methylation and sRNA clustering simultaneously in all stages of embryo development, we selected regions with consecutive CHH methylated probes in at least three out of five stages and a methylation level cutoff of 15%, along with sRNA clusters common in at least five out of eight stages. We divided the consecutive CHH methylated regions into two groups based on whether they overlapped with sRNA clustering or not. The average length of the regions with or without sRNA clustering was then calculated.

To determine the dependency of TE methylation on CMT2 and RdDM pathways, we utilized previously identified lists of differentially methylated regions (DMRs) (Stroud et al., 2013), and recently derived CHH methylation pathway DMR classifications (Hsieh et al., 2023): DMRs that are regulated by the CMT2 or RdDM pathways in both leaf and embryos are classified as cCMT2 DMRs or cRdDM DMRs, respectively. DMRs that are regulated by CMT2 or unmethylated in the leaf and become newly targeted by the RdDM pathway in embryo are classified as eRdDM DMRs. Here, TEs were allocated into the three DMR classes using bedtools intersect -wa -a TE list -b DMR list. TEs allocated to more than one DMR class were removed, and only TEs associated with a single type of DMR class chosen for further analyses.

We utilized our recently reported optimized method to profile the genome-wide Arabidopsis embryo methylome at five developmental stages: DAP4, 5, 7, 9, and 12, which correspond to the globular, heart, torpedo, bending torpedo, and mature green stage embryos, respectively ( Figure 1A ) (Yoo et al., 2021). All libraries were made up of at least two biological replicates with robust CT conversion (>99%) and genome coverage (>40X), complying with the recommended standard for methylation data analysis (Ziller et al., 2015) ( Figure 1B ).

As previously reported (Cokus et al., 2008; Lister et al., 2008), Arabidopsis chromosomes exhibited high levels of pericentromeric methylation in all cytosine contexts, due to the accumulation of repetitive sequences and heterochromatic TEs ( Figure 1B ). Consistent with previous reports (Bouyer et al., 2017; Kawakatsu et al., 2017; Lin et al., 2017; Papareddy et al., 2020), CG methylation levels remained relatively constant throughout embryo development, whereas non-CG methylation was more dynamic ( Figure 1C ). In particular, as others have reported, there was a gradual increase in global CHH methylation levels in pericentromeric regions during embryo development ( Figure 1B ), which culminated as the seeds mature ( Figure 1C ). The increase in CHH methylation levels primarily resulted from increases at TEs, whereas gene body methylation levels in all cytosine contexts remained relatively stable ( Figure 1C ). CHH methylation levels in gene bodies increased only slightly ( Figure 1C ). CHG methylation levels in TEs showed a slight decrease, particularly from DAP7 to DAP9 ( Figure 1C ), but otherwise remained relatively stable as the seeds mature.

We next analyzed the methylation dynamics of Arabidopsis embryo development within genomic features, focusing on genes and TEs ( Figure 1 ). As observed genome-wide, we found that the methylation change from the beginning to end of embryogenesis was small in genic regions in all contexts (<0.7%; Figures 1D–F ). However, we observed distinct gene body methylation levels for both CHG and CHH contexts at each developmental stage ( Figures 1E, F ), indicating a coordinated change in methylome status during development. For CHH methylation at the 5’ and 3’ ends of genes, there was a slightly larger increase (1%) that was distinguishable from the globular-torpedo to the bending torpedo-mature green stages, likely due to the presence of overlapping TE sequences in the flanking regions of genes ( Figure 1F ). As noted above, the torpedo embryo stage at DAP7 showed the lowest CHG methylation levels ( Figures 1C–E ).

At TEs, CG methylation increased marginally throughout embryogenesis ( Figures 1G–I ). CHH methylation levels at TEs, as previously noted, demonstrated a profound and continuous increase during development ( Figure 1I ). Interestingly, a striking increase was shown in CHH methylation at TEs from DAP5 to DAP7, which coincided with a decrease in CHG methylation levels ( Figures 1C–H ). This observation may indicate that CHG and CHH methylation work together in a coordinated manner to repress TEs during heart (H) to torpedo (T) transition phase, which marks the beginning of maturation after pattern formation.

In mammalian embryogenesis, a unique form of de novo CG methylation occurs through the spreading of methylation from cis-regulatory regions to distal CpG sites (Turker, 2002). To investigate whether this phenomenon of methylation spreading also occurs in plants during embryo development, we calculated the average length of the region where CHH methylated probes exist consecutively at each embryo developmental stage. We merged windows with methylation levels above 5%, 10%, or 15% in all stages for each cytosine context if they were closer than 40 bp apart. As the embryos developed, the average length of consecutively methylated regions gradually increased, especially when we used higher methylation cutoffs (15%; Figure 2A ). This result suggests that genome-wide CHH methylation spreads as the embryo matures.

The RdDM pathway plays a critical role in de novo methylation in Arabidopsis (Matzke and Mosher, 2014; Chow et al., 2020; Erdmann and Picard, 2020), however, it is possible that sRNAs are only required for the initial CHH methylation setting during early embryogenesis. This initial CHH methylation could then act as a signal for its spreading to nearby sites at later stages of embryogenesis, similarly to CG methylation in mammals. In this scenario, CHH methylation spreading would occur even in the absence of matching sRNAs to nearby sites. Alternatively, sRNA clusters corresponding to neighboring regions may be continuously required for the spreading of CHH methylation. We measured the average length of 24nt sRNA clusters that are merged when they exist within 75 bp at each stage using publicly available sRNA expression data during embryogenesis (Papareddy et al., 2020). As the embryos developed, the average length of merged sRNA clusters for 24nt sRNAs gradually increased ( Figure 2B ). This suggests that expression of sRNAs corresponding to nearby sites is still required for the methylation of neighboring unmethylated CHH sites during embryo development.

To investigate the relationship between the expansion of sRNA expression and the spreading of CHH methylation during embryo development, we compared the average length of regions where consecutive CHH-methylated windows overlapped with merged 24nt sRNA clusters. Our results revealed a linear increase in the length of the sRNA clusters during development in CHH overlapping regions, demonstrating that there is a positive correlation between the spreading of CHH methylation and expansion of sRNA-producing regions ( Figure 2C , green bars). CHH methylated regions were also expanded in the absence of sRNA clusters as embryos matured ( Figure 2C , pink bars). However, these regions were formed of far fewer loci and reflected less extensive CHH spreading. In addition, when we mapped the position of these non-sRNA overlapping CHH methylated regions, we found that they were tightly centromeric, and as such likely to be conferred by CMT2 ( Supplementary Figure S1 ). To further investigate the positive correlation of CHH methylation spreading with sRNA clusters, we visualized individual overlapping loci, and observed that expanding sRNA expression was coincident with CHH methylation spreading and accumulation ( Figures 2D–F ), providing support for previous observations that increased embryo CHH methylation is conferred by the RdDM pathway (Hsieh et al., 2023).

Taken together, our data suggest that the RdDM pathway plays an important role in both initiating and spreading CHH methylation to nearby regions, as well as in the accumulation of CHH methylation levels. This can be achieved, at least in part, by the continuous expression of sRNA clusters followed by expression of sRNA clusters in neighboring regions, resulting in wider CHH methylation regions and higher CHH methylation levels at later stages of embryogenesis.

CHH methylation is known to be regulated by two pathways; the RNA-independent CMT2 pathway and the RdDM pathway (Zemach et al., 2013; Matzke and Mosher, 2014; Kenchanmane Raju et al., 2019). RdDM targets are methylated by de novo methyltransferases, DRM1 and 2. Recent papers described the delineation of embryo CHH loci into classes based on their mechanism of catalysis using drm1/2 and cmt2 mutants (Kawakatsu et al., 2017; Hsieh et al., 2023). Briefly, if an embryonic CHH DMR is targeted by CMT2 both in the vegetative (leaves) and reproductive embryo stages, that DMR is denoted as a conserved CMT2 (cCMT2) DMR. If a DMR is targeted by DRM1/2 both in leaves and embryos, that DMR is denoted as a conserved RdDM (cRdDM) DMR. If a DMR is targeted by CMT2 or unmethylated in leaves but becomes exclusively targeted by DRM1/2 in embryos, that DMR is denoted an embryonic RdDM (eRdDM) DMR (Hsieh et al., 2023)

In order to understand the methylation dynamics of these three DMR classes during development, we divided our data according to DMR class and examined their average methylation level during embryogenesis ( Figure 3A , also see Methods). Average CG methylation levels of cCMT2 DMRs increased from DAP4 to DAP12 (p < 0.0001), whereas those of cRdDM DMRs decreased during the same period (p < 0.0001) ( Figure 3A ). Both the levels and pattern of CG methylation in eRdDM DMRs laid between those of cCMT2 DMRs and cRdDM DMRs ( Figure 3A ). Average CHG methylation level change was unique, whereby cCMT2 and eRdDM DMRs showed decreases from DAP5 to DAP7 (p< 0.0001 for cCMT2 DMRs and p < 0.0001 for eRdDM DMRs) whereas cRdDM DMRs showed large increases in CHG methylation from DAP7 to DAP12 (p<0.0001) ( Figure 3A ). CHH methylation levels were continuously increased during embryogenesis in all classes ( Figure 3A ). The most dynamic epigenetic change observed during embryogenesis remained the constant increase in CHH methylation, and among them, cRdDM-targets displayed the largest change. This was particularly striking from DAP9 to DAP12, reaching 60% CHH methylation levels at DAP12 ( Figure 3A ).

To explore the overall location of CMT2/RdDM-dependent DMRs, we mapped cCMT2-, cRdDM-, and eRdDM DMRs to chromosomes. All three classes were enriched in pericentric chromatin ( Figure 3B ). However, consistent with previous reports, cCMT2 DMRs were almost entirely located in pericentromeric regions (He et al., 2021) ( Figure 3B , green trace), whereas cRdDM DMRs were much more evenly spread throughout chromosome arms ( Figure 3B , pink trace). Interestingly, eRdDM DMRs were most similar in profile to cCMT2 DMRs, with clear pericentromeric enrichment, although they demonstrated comparative increases in chromosome arm enrichment ( Figure 3B , black trace). These data indicate that chromosomal location, which is highly associated with local chromatin state, may be key to determining CHH enzyme dependency, consistent with previous reports (Sigman and Slotkin, 2016).

Recently, it was reported that some regions regulated by CMT2 in wild-type Arabidopsis seedlings are accessed by the RdDM pathway in the absence of the DDM1 nucleosome remodeling factor, which in wild-type plants mediates heterochromatic DNA methylation in all cytosine contexts (Zemach et al., 2013; He et al., 2021). These regions were referred to as “CMT2-to-RdDM loci” by the authors. Some eRdDM DMRs exhibited similar traits to CMT2-to-RdDM loci, as they were initially methylated by CMT2 in somatic tissues and subsequently switch to the RdDM pathway after fertilization during embryonic development (Hsieh et al., 2023). This prompted us to investigate the extent of overlap between eRdDM DMRs and CMT2-to-RdDM loci by He et al. We conducted a parallel comparison of these regions and found that most of the DMRs were not shared between two datasets ( Figure 3C ), indicating that eRdDM DMRs are distinct from the CMT2-to-RdDM loci.

The purpose of CHH methylation in Arabidopsis is widely accepted to be TE silencing, to prevent insertional mutagenesis in key tissues, such as the developing embryo. To explore TE dynamics during embryo development, we categorized TEs based on their dependency on CHH methylation pathway. As for our whole-genome analyses, we annotated TEs according to the three DMR classes outlined above (cCMT2, cRdDM and eRdDM) (Hsieh et al., 2023), and calculated the proportion of each ( Figure 4A ). Then, only those TEs annotated exclusively with a single category of DMR were used for further analyses; 3,790 for cCMT2 TEs, 5,402 for cRdDM TEs and 2,646 for eRdDM TEs ( Figure 4A ).

To investigate the overall location of TEs depending on their CHH mechanistic class, we mapped cCMT2, cRdDM, and eRdDM TEs to the chromosomes. Similar to the chromosomal location of each CHH DMR ( Figure 3B ), we found cCMT2 TEs were more highly concentrated in pericentromeric regions than other classes of TEs. Furthermore, we observed that longer cCMT2 TEs (≥ 3kb) were more highly concentrated in pericentromeric regions compared to shorter cCMT2 TEs (≤ 500 bp) ( Figure 4B ). In contrast, cRdDM TEs were located throughout the chromosomes and most of them were shorter than 500bp ( Figure 4B ). Interestingly, the majority of eRdDM TEs were also shorter than 500bp, similar to cRdDM TEs ( Figure 4B ), indicating that accessible length is a fundamental attribute of loci targeted by RdDM. Additionally, we observed that eRdDM TEs were predominantly located similarly to the shorter cCMT2 TEs, with a slight shift towards chromosomal arms ( Figure 4B ).

We compared the overall cytosine methylation levels for the three TE classes, mainly revealing differences in flanking regions. Consecutive high levels of methylation in all contexts surrounding TE bodies was observed in cCMT2 TEs ( Figure 4C ), but not in cRdDM TEs ( Figure 4D ). This is consistent with previous observations of comprehensive heterochromatin formation and DNA methylation in pericentromeric regions, since cCMT2 TEs were almost all pericentromeric, and cRdDM TEs were much more spread along the chromosome ( Figure 4B ) (Zemach et al., 2013; Yelagandula et al., 2014; Zhao et al., 2019; Bourguet et al., 2021). eRdDM TE methylation levels were intermediate between the other two classes, displaying the same clear delineation of CG and CHG methylation levels between TE bodies and flanking regions as cRdDM sites, but much less so for CHH methylation ( Figure 4E ).

We next examined the average length of TEs classified into the three DMR classes. cCMT2 TEs tended to be longer than cRdDM TEs, although with large variation ( Figure 4F , p < 0.0001). eRdDM TEs were even shorter in size than cRdDM TEs ( Figure 4F , p < 0.0001). Considering that 29% of eRdDM TEs were cCMT2 TEs in leaves ( Figure 4A , Supplementary Figure S2 ), this is quite intriguing. Based on their size and location, this result raises the idea that eRdDM TEs that originated as CMT2 TEs in leaves were shorter, less-heterochromatic CMT2 sites. We speculate that these sites transitioned to RdDM targets due to changes in local chromatin organization and the presence of highly robust RdDM activity during the developmental transition upon fertilization.

To test whether the composition of TE superfamilies within each TE class is correlated to CHH-conferring enzyme dependency, we investigated the proportion of representative TE superfamilies within each TE class. In Arabidopsis, retroelements represent the most common TE type, comprising ~10Mb of the genome, closely followed by DNA transposon orders Helitrons (8 Mb) and TIR elements (7 Mb) (Ahmed et al., 2011). TIR elements are so called due to their Terminal Inverted Repeats, and two of the most frequent TIR superfamilies in Arabidopsis are Mutator-like (MuDR) which is also one of the most mobile elements (Quadrana et al., 2016), and En-Spm. Retroelements can be further delineated into LTR (long terminal repeat) TEs such as Gypsy and Copia [another highly mobile elements, (Quadrana et al., 2016)], and non-LTRs such as LINEs and SINEs (Quesneville, 2020).

We examined whether any TE superfamilies were enriched in a particular TE class. As shown in Figures 5A, B , cCMT2 TEs were highly enriched for Gypsy LTR retrotransposons, which is known to be the longest TE in mean length and concentrated in pericentromeric regions (Quesneville, 2020). So most Gypsy elements acquire CHH methylation via the CMT2 pathway. In contrast, both cRdDM and eRdDM TEs were depleted for LTR targets, instead enriched for the Helitron and MuDR superfamilies of TEs, of which the mean lengths were significantly shorter than LTR retrotransposons ( Figures 5A, B ). All TE superfamilies, regardless of their average length, exhibited the same trend: longer TEs were targeted by the CMT2 pathway, while shorter TEs were targeted by the RdDM pathway, within each family ( Figure 5B ). For instance, SINEs were generally the shortest TE among the seven different superfamilies we examined. However, there were SINE TEs targeted by CMT2 which were, although long relative to other SINEs, were still shorter than RdDM-targeted TEs in the Helitron superfamily ( Figure 5B ). In addition, the mean length of eRdDM TEs were shorter than cRdDM TEs even within each superfamily ( Figure 5B ). This suggests a sequence based mechanism influences CHH methylation targeting, whereby once an element reaches a certain size, the sequence composition is such that CHH targeting switches from RdDM to CMT2 based mechanisms. Shorter TEs that are unmethylated, or even CMT2-regulated in leaves, may become accessible to the RdDM pathway after the transition to reproductive development in the embryo. Taken together, and consistent with a previous report (Zemach et al., 2013), our data highlight that TE size, relative to the superfamily, is one of the most important factors for CHH methylation targeting mechanisms.

All three DMR classes of CHH-methylated loci in embryo were found across all TE superfamilies ( Figures 5A, B ). Next we examined methylation dynamics of each TE family during embryo development. In general, cCMT2 TEs showed uniform methylome dynamics throughout the superfamilies ( Supplementary Figure S3A ), whereas both cRdDM- and eRdDM TEs showed more variable methylome dynamics in each superfamily ( Supplementary Figures S3B, C ). Consistent with the result shown in Figure 3A , while all the TEs showed gradual increase of CHH methylation, cRdDM TEs showed the most striking increase in CHH methylation for all superfamilies ( Supplementary Figure S3B , cRdDM_CHH). CG methylation levels were decreased specifically at DAP12 ( Figure 3A ), an all seven superfamilies of cRdDM TEs followed a similar pattern ( Supplementary Figure S3B , cRdDM_CG). It is tempting to speculate that the decrease in CG methylation in cRdDM TEs located in chromosomal arms might induce an active demethylation mechanism, resulting in an increase in sRNA generation and CHH de novo methylation. The large increase in CHH methylation of TEs during late embryo development is supported by previous reports that genes involved in RdDM such as AGO4, DRM2 and DMS3, and concomitant sRNAs, showed highest expression during late embryogenesis (Belmonte et al, 2013; Bouyer et al., 2017; Papareddy et al., 2020). Interestingly, Helitron transposons, which were the most enriched in terms of copy number and genome coverage among the superfamilies we analyzed ( Figure 5A ), showed the lowest CG and CHG methylation levels in RdDM at all stages ( Supplementary Figures S3B, C ). Taken together, our data suggest that the CHH regulatory pathways mediating TE methylation dynamics are independent of TE superfamily.

We next investigated the impact of TE length on embryo methylation levels and dynamics in different CHH regulatory pathways at DAP4 ( Figures 6A–C ) and DAP12 ( Figures 6D–F ). Almost all longer (>3 kb) TEs belonged to the cCMT2 TE category in all cytosine contexts (green in Figures 6A–F ), which was consistent with the findings in Figures 4F , 5B . Moreover, methylation levels for cCMT2 TEs tended to converge at a specific level as TE length increased in all cytosine contexts. Methylation levels increased from DAP4 to DAP12, especially in CG and CHH contexts ( Figures 6A–F ). This suggests that cCMT2 TEs maintain specific ranges of methylation levels regardless of their length, consistent with their collective methylation pattern around the centromere, observed above ( Figure 4C ; Supplementary Figure S3A ). The CG methylation range for cCMT2 TEs was broader at DAP4, becoming narrower at DAP12 ( Figures 6A–D ). Interestingly, the opposite trend was observed for CHH methylation, as DAP12 exhibited a broader range of CHH methylation. These results indicate that as the embryo matures, cCMT2 TEs gain a higher and more uniform level of CG methylation (around 95%) whereas CHH increases are more variable, resulting in a wider range of levels DAP12. Intriguingly, although many cCMT2 TEs displayed low CHH methylation at DAP4 (around 5%), some short cCMT2 TEs exhibited very high CHH methylation ( Figure 6C ).

Unlike cCMT2 TEs, the majority of both RdDM TEs were found to reside in shorter TEs, with eRdDM TEs being slightly shorter than cRdDM TEs ( Figures 6A–F ). The overall methylation patterns, particularly for CG and CHG methylation, were similar between cRdDM TEs and eRdDM TEs ( Figures 6A–E ). The eRdDM TEs exhibiting high CG methylation levels showed almost the same levels as those in cCMT2 TEs, suggesting that these eRdDM TEs in the reproduction stage likely originated from cCMT2 TEs in the vegetative stage. CHH methylation levels of eRdDM TEs were found to be lower than those in cRdDM TEs at DAP4, and the increasing rate of CHH methylation of during embryogenesis was slower. Therefore, the highest CHH methylation level observed in eRdDM TEs was around 45%, compared with cRdDM TEs, which exhibited levels around 95% at DAP12 ( Figures 6C–F ).

While it has been known that heterochromatic TEs have higher GC content than euchromatic TEs (Zemach et al., 2013), examining the frequency of all cytosine contexts may provide insights into the relationship between CHH regulatory dependency of TEs and their sequence composition. Our analysis revealed that cCMT2 TEs exhibited significantly higher cytosine frequency across all contexts compared to cRdDM TEs and eRdDM TEs ( Figure 7A . p < 0.0001). eRdDM TEs demonstrated the lowest CG and CHG frequency, but slightly higher CHH frequency compared to cRdDM TEs ( Figure 7A ). We measured the frequencies of CG, CHG and CHH in the three TE classes as well as those of all Arabidopsis TEs as a reference based on each TE superfamily. We also calculated the expected frequencies using arithmetic calculations ( Table 1 ). As a result, we observed that the cytosine frequencies of most cCMT2 TEs were higher than two RdDM classes, as well as all TEs across all cytosine contexts. However, the patterns of cytosine frequencies differed between cRdDM TEs and eRdDM TEs, depending on the context. Both cRdDM TEs and eRdDM TEs exhibited lower CHG frequency compared to all TEs except for the SINE superfamily in cRdDM TEs (0.052) ( Table 1 ; Figure 7A ). Between cRdDM TEs and eRdDM TEs, eRdDM TEs showed lower CHG frequency but higher CHH frequency, except for the En-Spm superfamily ( Table 1 , Figure 7A ). Notably, the CHH frequency of cRdDM TEs was the lowest among the three TE classes, suggesting that TEs with low CHH frequency are preferentially targeted by the RdDM pathway during vegetative growth. Since the region with the highest CHG frequency in cCMT2 TEs is likely to be more heterochromatic than the CHG region in RdDM TEs, our analysis suggests that the difference between cCMT2 TEs and eRdDM TEs may be attributed, at least in part, to the higher CHG frequency observed in cCMT2 TEs. This further implies that shorter CMT2 TEs with lower CHG frequency may contribute to epigenetic reprogramming during the transition from the vegetative to the reproductive lifecycle, and they may be newly targeted by the RdDM pathway, becoming eRdDM TEs.

Recent research has shown that although euchromatic TEs are usually regulated by RdDM, young and LTR transposons in euchromatin are typically regulated by CMTs. Furthermore, longer and younger LTR transposons tend to have higher CHG methylation levels (Wang and Baulcombe, 2020). We analyzed the cytosine frequencies of predicted de novo LTRs using publicly available data (Wang and Baulcombe, 2020) and found that intact LTRs with higher LTR sequence similarity, which are thought to be more recently introduced in the genome, displayed diverse cytosine frequencies. By contrast, older LTRs with low similarity due to mutations had similar cytosine frequencies, suggesting they have decayed over time during evolution ( Figure 7B ). Taken together, our findings highlight the correlation between length, location, cytosine ratio, DNA methylation, and regulatory pathways of TEs.