Fast co‐evolution of anti‐silencing systems shapes the invasiveness of Mu‐like DNA transposons in eudicots

Abstract

Transposable elements (TEs) constitute a major threat to genome stability and are therefore typically silenced by epigenetic mechanisms. In response, some TEs have evolved counteracting systems to suppress epigenetic silencing. In the model plant Arabidopsis thaliana, two such anti‐silencing systems have been identified and found to be mediated by the VANC DNA‐binding proteins encoded by VANDAL transposons. Here, we show that anti‐silencing systems have rapidly diversified since their origin in eudicots by gaining and losing VANC‐containing domains, such as DUF1985, DUF287, and Ulp1, as well as target sequence motifs. We further demonstrate that these motifs determine anti‐silencing specificity by sequence, density, and helical periodicity. Moreover, such rapid diversification yielded at least 10 distinct VANC‐induced anti‐silencing systems in Arabidopsis. Strikingly, anti‐silencing of non‐autonomous VANDALs, which can act as reservoirs of 24‐nt small RNAs, is critical to prevent the demise of cognate autonomous TEs and to ensure their propagation. Our findings illustrate how complex co‐evolutionary dynamics between TEs and host suppression pathways have shaped the emergence of new epigenetic control mechanisms.

Targeting of VANC anti‐silencing proteins is determined by sequence, density, and helical periodicity of sequence motifs within non‐coding regions of VANDAL elements.

Introduction

Transposable elements (TEs) are ubiquitous DNA sequences that move and self‐propagate across genomes. Because of their potential to create large effect mutations upon transposition or by facilitating chromosomal rearrangements through recombination, TEs constitute a major threat to genome function and integrity. However, TEs are usually under tight epigenetic control, notably by DNA methylation in plants and mammals (Slotkin & Martienssen, 2007), thus limiting their mutational impact. In the model plant Arabidopsis thaliana, mutants deficient in the chromatin remodeler DDM1 lose DNA methylation over most TE sequences and reactivate transcriptionally several hundreds of these, which results in increased transposition activity in few cases (Miura et al, 2001; Singer et al, 2001; Tsukahara et al, 2009; Quadrana et al, 2019). Also, many TEs transpose frequently in nature (Baduel et al, 2021), implying that they do occasionally evade repressive mechanisms, thus ensuring their continuous propagation.

How TEs escape epigenetic silencing remains largely unknown, except for the notable example of silencing suppression deployed by the VANDAL21 and VANDAL6 Mutator‐like DNA transposons (Fu et al, 2013; Hosaka et al, 2017), which are abundant in the A. thaliana genome (Kapitonov & Jurka, 1999). Specifically, these two TEs encode each a distinct VANC anti‐silencing protein, VANC21 or VANC6, which binds to distinctive short DNA motifs accumulated in non‐coding regions of cognate VANDAL copies, where they induce strong hypomethylation and transcriptional derepression (Hosaka et al, 2017). Furthermore, a VANDAL21 copy, called Hiun (Hi), mobilized in ddm1 (Tsukahara et al, 2009) and can also be mobilized in wild‐type background by transgenic expression of VANC21 protein (Fu et al, 2013). Unlike viral suppressors, which neutralize host defense responses broadly, VANC‐induced anti‐silencing is highly specific, as only related TE sequences are epigenetically reactivated. Thus, co‐evolution of VANC proteins and target DNA motifs may allow specific VANDALs to escape epigenetic silencing and propagate through the genome while minimizing host damage (Hosaka et al, 2017). Nonetheless, once activated, VANC‐induced anti‐silencing should perpetuate the epigenetically active state of target TEs, potentially leading to runaway transposition. However, wild‐type genomes typically contain very few full‐length VANDAL copies, implying that VANC‐mediated anti‐silencing must be interrupted at some point through a still unknown mechanism.

The A. thaliana genome contains 28 distinct VANDAL families in total. Previous analysis of F1 hybrids derived from a cross between wild type and a mutant defective in the DNA methyltransferase MET1, homolog of the mammalian DNMT1, identified lower methylation levels than the expected mid‐parental values for some VANDALs sequences, suggesting that they might be subjected to trans‐hypomethylation (Rigal et al, 2016).

Using a population of ddm1‐derived epigenetic recombinant inbred lines (epiRILs), we have previously found that transposed copies of the active VANDAL21 copy Hi induce efficient trans‐hypomethylation of homologous copies (Fu et al, 2013), indicating that this population provides a powerful tool for the systematic study of the anti‐silencing factors encoded by the VANDAL superfamily of TEs. Here, by combining methylome data for the epiRILs, quantitative epigenetics approaches, and ectopic expression of TE‐derived sequences, we have identified the complete set of active VANDAL‐encoded anti‐silencing systems in A. thaliana. Our results indicate that since their likely origin in the common ancestor of eudicots, VANCs and their target sequences diversified extensively. Furthermore, the A. thaliana VANC1‐encoded anti‐silencing system produced by a VANDAL1 copy has conserved features of the most likely ancestral system, including a Ulp1 protein domain and targeting of palindromic DNA sequences. We also show that target specificity of the distinct VANCs is determined by the sequence, density and spatial arrangement of ~10‐bp‐long motifs, which exhibit helical periodicity and hint to a cooperative DNA binding and homodimerization of VANCs. Last, we demonstrate that non‐autonomous VANDALs, which can serve as a reservoir of trans‐matching small RNAs, are also major targets of VANC‐induced anti‐silencing and that impairing this targeting by removal of the short‐sequence motifs triggers strong and concerted epigenetic re‐silencing of cognate autonomous copies. Together, our findings revealed the complex interplay between host silencing and TEs, as well as their interactions between autonomous and non‐autonomous copies that shaped the co‐evolution of VANDALs and have potentially contributed to the emergence of novel gene control mechanisms.

Results

Extensive trans‐hypomethylation of VANDALs in A. thaliana

We set out to determine whether other VANDAL copies in addition to Hi are also subjected to trans‐hypomethylation in the ddm1‐derived epiRILs. These lines have almost identical DNA sequences but segregate many differences in DNA methylation (Colomé‐Tatché et al, 2012) as well as few transpositionally active TEs (Quadrana et al, 2019). However, with the exception of Hi, no VANDAL mobilized in the epiRILs (Quadrana et al, 2019), enabling us to test DNA methylation levels over these TEs in the absence of confounding effects due to ongoing transposition. EpiRILs were derived from an initial cross between two isogenic individuals, one carrying a mutant allele of DDM1 and one WT. A single F1 was then backcrossed to the WT parental line and F2 DDM1/DDM1 progenies were propagated for six generations to generate a population of plants with mosaic epigenomes (Johannes et al, 2009); (Fig 1A). We obtained whole‐genome bisulfite sequencing (WGBS) data for a core collection of 16 epiRILs together with siblings of the two founder plants. Overall, single‐cytosine resolution methylomes confirmed the epihaplotype maps previously obtained using MeDIP and microarray‐based methylomes (Appendix Fig S1). Given the crossing scheme used to derive the epiRILs (Fig 1A), around 75% of their genome on average is of WT origin and exhibit indeed WT‐like methylation levels (Appendix Fig S2). Based on this property, we reasoned that putative trans‐hypomethylation of VANDAL sequences should be readily detected in the epiRILs as local DNA methylation losses over wt‐derived copies. We therefore analyzed the DNA methylation levels of wt‐derived VANDALs (Fig 1B) and detected 244 hypomethylated copies belonging to 20 VANDAL families, including several VANDAL21 sequences (Fig 1C and D). Only a small number of copies were affected per family (between 3 and 33 copies). Hypomethylation occurs both at CG and non‐CG sites, but to different degrees (Fig 1D). While non‐CG hypomethylation affects entire VANDAL sequences, CG hypomethylation is limited to short regions, resembling the sequence‐specific DNA methylation loss induced by VANC21 and VANC6 (Fu et al, 2013; Hosaka et al, 2017). Importantly, hypomethylated VANDALs were present only in ~25% of the epiRILs that carry the corresponding wt‐derived interval (Fig 1B and Appendix Fig S3). This last observation suggests that other loci, which should segregate independently of the wt‐derived VANDALs in most cases, induce hypomethylation in trans and hypomethylated VANDALs inherited from the ddm1 parent are obvious candidates. Altogether, our results indicate that most VANDAL families may be targets of anti‐silencing systems in A. thaliana.

Figure 1. Most VANDAL families are trans‐hypomethylated in epiRILs.

1. Cartoon depicting the crossing scheme used to generate the epiRIL population.
2. Genome browser tracks showing local hypomethylation of a WT‐derived VANDAL2 copy in Col‐0, ddm1 mutant, and several epRILs that carry the WT‐ or ddm1‐derived epihaplotype at this locus (indicated on the left).
3. Number of VANDAL copies per family as well as the number of WT‐derived VANDAL copies hypomethylated in at least one epiRIL.
4. Metaplot of DNA methylation on wt‐derived VANDAL21 or VANDAL1 copies within Col‐0, ddm1, or epiRIL60.

Sequence and syntax of motifs determine anti‐silencing specificity

Previous in vitro and in vivo studies revealed that VANC21 and VANC6, respectively, bind the short‐sequence motif “YAGTATTAY” and “AGTTGTMC” (where Y can be T or C; and M can be A or C). These motifs are located within non‐coding regions of VANDAL21 and VANDAL6 sequences, where they induce local CG hypomethylation (Fu et al, 2013; Hosaka et al, 2017). Thus, we searched in the epiRILs for short‐sequence motifs overrepresented within VANDAL21 sequences that lose CG methylation in the epiRILs and identified in this way a strong overrepresentation of the motif “YAGTATTAC” (Fig 2A). This result confirms the pattern described for VANC21‐binding sites (Hosaka et al, 2017) with hypomethylation around short‐sequence motifs extending much further at non‐CG than CG sites.

Figure 2. Sequence and syntax of motifs dictate VANC anti‐silencing specificity.

1. Metaplot of DNA methylation and relative motif density (arbitrary units: 1 = max; 0 = min) on wt‐derived VANDAL21 or VANDAL1 copies as well as around short‐sequence motifs within Col‐0, ddm1, or epiRIL60. Metaplots of DNA methylation for VANDAL1 and VANDAL21 are from Fig 1D.
2. Motif logo, number of hypomethylated and non‐hypomethylated WT‐derived copies, and size of VANDALs (in Kbp) carrying or not short‐sequence motifs. For each boxplot, the lower and upper bounds of the box indicate the first (Q1) and third (Q3) quartiles, respectively, the whiskers represent data range, bounded to 1.5 * (Q3–Q1). Sample sizes are indicated in gray.
3. Proportion of isolated or increasingly clustered motifs (medium: 2–3 motifs/1 Kb; high +4 motifs/1 Kb) across sequences belonging to the distinct VANDAL families. VANDAL families showing no hypomethylation in epiRILs are indicated in gray.
4. Spacing between consecutive short‐sequence motifs in the three possible relative orientations within specific VANDAL families. The proportion of palindromic sequence within each motif (i.e., Palindromic Index) is also shown.

Following this first confirmation of our approach, we set out to use the epiRILs to characterize the hypomethylation of all VANDALs. We searched for short‐sequence motifs overrepresented at CG‐hypomethylated regions within wt‐derived VANDAL sequences. We detected statistically overrepresented motifs for all TE families analyzed and in all cases local CG and broad non‐CG hypomethylation is observed around detected motifs (Fig 2A and B, Appendix Figs S4 and S5), reminiscent to the VANC21‐ and VANC6‐induced loss of DNA methylation (Fu et al, 2013; Hosaka et al, 2017). Consistent with only few copies per family being trans‐hypomethylated (Fig 1C), only a small fraction of VANDALs carries DNA‐sequence motifs, and these copies typically correspond to full‐length elements (Fig 2B). In fact, hypomethylated VANDAL sequences are much longer than non‐hypomethylated ones (Appendix Fig S6A).

Despite the strong association between the presence of motifs and hypomethylation, a sizable proportion of non‐hypomethylated copies do carry motifs (Fig 2B), which nonetheless accumulate at much lower density in these compared to hypomethylated VANDALs (Appendix Fig S6B), suggesting that the sole presence of motifs is not sufficient for hypomethylation targeting. Furthermore, motifs enriched in specific VANDAL families are also detected in other families (Fig 2C), which is of course expected given the high probability (P > 0.09) of finding a 8‐ to 10‐bp‐long motif in any 6,000‐bp‐long sequence. Given that each hypomethylated copy typically contains many short‐sequence motifs (Appendix Fig S6B), we reasoned that the specificity of VANDAL anti‐silencing systems may be determined by their local density, as has been proposed for VANC21 and VANC6 (Hosaka et al, 2017). To test this hypothesis, we investigated the clustering of short‐sequence motifs within VANDAL sequences. Compared to isolated motifs, which are ubiquitous among all VANDALs, clusters containing four or more motifs per 1,000 bp are almost exclusively overrepresented within cognate TE families (Fig 2C), with the notable exception of VANDAL1/1N1/2 and 2N1 families that share the same motif TGTACGTACA. In addition, motifs detected in VANDAL5 and VANDAL15 are also found in VANDAL6 and VANDAL16 copies, respectively. Notwithstanding this last result, which may be an indication that these families belong to the same anti‐silencing system, local accumulation of hypomethylation motifs seems to provide an additional layer of anti‐silencing specificity (Fig 2C).

Clustering of short‐sequence motifs may imply that VANCs interact with DNA as homomultimers, similarly to the mode of action described for some transcription factors (Avsec et al, 2021). To test this possibility, we explored the spatial arrangement of short‐sequence motifs (i.e., the distance between consecutive motifs organized as direct (DR), everted (ER), or inverted repeats (IR)). This analysis revealed that most hypomethylation motifs cluster as direct repeats spaced by 10 bp, which corresponds to one turn of the DNA helix. Notably, short‐sequence motifs within TEs belonging to VANDAL1/1N1/2 and 2N1 families appear to cluster indistinctly as DR, ER, or IR (Fig 2D), which is consistent with these motifs being highly palindromic (Palindromic Index 0.6–0.8). Taken together, these results establish that specificity of distinct anti‐silencing systems is likely determined by the identity, syntax, and spatial organization of short‐sequence motifs and that DNA‐bound VANCs potentially form arrays of homopolymers with helical periodicity.

Diversification of VANC‐dependent anti‐silencing systems within and across species

In order to maintain the functionality of the recognition system, diversification of short motifs within VANDALs needs to be accompanied by parallel variations in their cognate VANC proteins. However, investigating the evolution of TE‐encoding proteins is challenging because of the lack of reliable transcript annotations over TE sequences in reference genomes. To circumvent this limitation and determine the whole repertoire of VANDAL‐encoded VANC proteins in A. thaliana, we carried out deep long‐read Nanopore sequencing of cDNAs from ddm1 mutant plants (see Materials and Methods). By combining this high‐quality dataset with a previous “gene‐like” annotation of TEs (Panda & Slotkin, 2020), we could identify 160 VANDAL‐encoded transcripts together with their orientation, precise transcription initiation, and termination sites as well as their exon–exon boundaries (Fig 3A). In silico translation predicted 42 VANC‐encoding genes, encompassing 16 of the 28 VANDAL families annotated in the reference genome and including the previously characterized VANC21 and VANC6 (Hosaka et al, 2017) (Fig 3B). The number of VANC‐encoding genes varies greatly between TE families, likely reflecting differences in their coding potential. On the one hand, all the non‐autonomous VANDAL families (VANDAL1N1, VANDAL2N1, VANDAL5NA, VANDAL18NA/B, and VANDALNX1/2) lack any detectable VANC‐encoding transcripts. On the other hand, VANDAL6, VANDAL3, and VANDAL21 encompass the largest number of VANC‐encoding copies, supporting the notion that these families were subjected to recent amplification (Fu et al, 2013).

Figure 3. Diversification of VANC‐dependent anti‐silencing systems within and across species.

• A
  Genome browser view of full‐length cDNA nanopore reads and illumina short reads from ddm1‐mutant plants as well as de novo functional annotation of TE‐encoding transcripts together with TAIR10 gene and TE annotations. VANDAL4 and VANDAL5 copies are indicated as V4 and V5, respectively.
• B
  Number of VANC‐like encoding transcripts.
• C
  Phylogenetic relationship among predicted VANC proteins encoded by A. thaliana VANDALs and C. melo CUMULE. Structure and presence of conserved protein domains are indicated for each VANC.
• D
  Domain organization of VANC‐like proteins in the Pfam database.
• E
  VANC domain organizations across flowering plant species. The likely origin of Ulp1‐ and DUF287‐containing VANCs is indicated. Representative species from different groups of eudicots are shown.
• F
  Structure, predicted coding sequences, and localization of short‐sequence motifs of CUMULE.
• G
  Logo of the short‐sequence motif enriched within non‐coding regions of CUMULE. Statistical overrepresentation and Palindromic Index are also shown.
• H
  Spacing between consecutive short‐sequence motifs within CUMULE.

Sequence comparison of the 42 VANC proteins uncovered two main clusters, which are themselves made up of sub‐groups reflecting the different VANDAL families (Fig 3C). Detection of conserved domains in the Pfam database using Hidden Markov Models show that most VANC‐like proteins contain the domain DUF1985, either alone or in association with the Ulp1 or DUF287 domains (Fig 3C), and these combinations broadly explain the phylogenetic clustering. Remarkably, while Ulp1 domain is conserved across the tree of life (Appendix Fig S7A) and is associated with de‐sumoylation activities (Johnson, 2004), DUF1985 and DUF287 have uncharacterized functions and are almost exclusively encoded by Mu‐like elements (MULEs), named after the Mutator (Mu) DNA transposon of maize (Robertson, 1978). Placing VANC protein architectures (Fig 3D) into the plant phylogenetic tree indicated that DUF1985 alone or fused to Ulp1 preceded the radiation between rosids and asterids (Fig 3E; Appendix Fig S7B), tying the emergence of VANC proteins with that of eudicots. In addition, DUF1985 in combination with the Ulp1 domain is present across eudicot species (Appendix Fig S7A and B), suggesting that this domain organization is ancestral. Conversely, DUF287 is only detected in VANC‐like proteins encoded by Brassicaceae’s MULEs (Fig 3E; Appendix Fig S7A) and has no similarity with any other known protein, suggesting a recent de novo origin of this domain.

As the distribution, local density, and syntax of short motifs within VANDAL sequences is a reliable indicator of functional VANC‐mediated anti‐silencing in A. thaliana (Fig 2), we assessed whether similar motif organizations are present in the distantly related VANDAL‐like CUMULE from melon (van Leeuwen et al, 2007). Indeed, we found that this TE encodes a VANC‐like protein containing both DUF1985 and Ulp1 domains (Fig 3F) and carries outside of its coding sequences a high density of the quasi‐palindromic 11 bp motif TAAACGATCGT, arranged in a one‐helix‐turn periodicity (Fig 3G and H). Thus, CUMULE likely possesses the two components of a functional VANC‐induced anti‐silencing system, which suggests in turn that such systems are relatively ancestral. Taken together, these results illustrate the extensive diversification of MULE‐encoded VANC‐like factors and of their targeting sequences across eudicot species.

Multiple sequence‐specific anti‐silencing systems coexist in A. thaliana

We next set out to identify the VANC‐encoding copies responsible for the family‐specific anti‐silencing in the epiRILs. One possibility would be that hypomethylation of VANDALs is due to the activity of ddm1‐derived, VANC‐encoding TEs that segregate in the epiRILs. To test this hypothesis, we considered hypomethylation of wt‐derived VANDALs as traits and performed (epi)QTL mapping using the hundreds of parental DMRs that segregate in this population (Colomé‐Tatché et al, 2012; Cortijo et al, 2014). Starting with the epiQTL mapping of VANDAL21 and VANDAL6 hypomethylation, our approach accurately identified the full‐length reference copies encoding the active VANC21 and VANC6 previously characterized (Hosaka et al, 2017) (Appendix Fig S8A and B), demonstrating that trans‐hypomethylation in the epiRILs is determined by ddm1‐derived, VANC‐encoding VANDALs. Following this confirmation, we performed epiQTL mapping on the remaining VANDAL families with hypomethylated copies. In most cases, one or two (epi)QTL intervals were detected per family (Appendix Fig S8A), thus revealing a simple genetic architecture of anti‐silencing activities overall. However, several (epi)QTL intervals were shared by various TE families, suggesting that some VANCs have broad activity. Most noticeably, one (epi)QTL interval in chromosome 1 is shared by the four families VANDAL1/1N1/2 and 2N1 (Fig 4A), which also have similar short‐sequence motifs (Fig 2B), pointing to a common anti‐silencing system. Consistent with this interpretation and the simple genetic architecture of anti‐silencing, we typically identified a single VANDAL copy expressed in ddm1 and that encodes a full‐length VANC (Fig 3) within each epiQTL interval (Appendix Fig S8A). In total, there are at least 10 independent anti‐silencing systems associated with up to 10 VANC‐encoding VANDAL copies (Appendix Fig S8A). Importantly, five of such systems involved the uncharacterized Ulp1‐containing VANCs, providing a unique opportunity to investigate their anti‐silencing activity experimentally.

Figure 4. Ulp1‐containing VANC1 induces sequence‐specific hypomethylation.

1. (epi)QTL mapping of VANDAL1, 1N1, 2, and 2N1 trans‐hypomethylation in 105 epiRILs. The VANC‐encoding VANDAL1 and VANDAL2 located within the single (epi)QTL interval are indicated in each case.
2. Schematic diagram of structures of candidate VANDAL1 copy and the modified transgene spanning VANC1 used. Boxes indicate exons.
3. Genome browser tracks showing hypomethylation effect of VANC1 on a VANDAL1 and VANDAL2 copy.
4. VANC1‐induced DNA hypomethylation is shown for each TE at CHG sites and CHH sites. VANDAL1, 1N1, 2, and 2N1 copies are indicated by colors.
5. Metaplot of DNA methylation around short‐sequence motifs within wild‐type (grey) or VANC1‐expressing plants (red).

Ulp1‐containing VANC1 induces sequence‐specific hypomethylation

To assess the function of Ulp1‐containing VANC factors, we transformed wild‐type A. thaliana plants with VANC‐containing sequences from the VANDAL1 (AT1TE56425) and VANDAL2 (AT1TE31190) candidate copies identified on our epiQTL analysis (Fig 4A and B; Appendix Fig S9A). WGBS of transformed plants revealed that the ectopic expression of VANC1, but not VANC2, is sufficient to induce strong and specific non‐CG hypomethylation of both VANDAL1 and VANDAL2 sequences (Fig 4C and D; Appendix Fig S9B and C). Furthermore, in VANC1‐transgenic plants (VANC1‐TG), the loss of non‐CG DNA methylation tends to affect the entire VANDAL sequences, while CG hypomethylation is constrained to short regions enriched in the motif “TGTACGTMY” (Fig 4C and E), reproducing the pattern of hypomethylation observed in the epiRILs (Figs 1B and 2A). Together, these findings demonstrate that the ectopic expression of Ulp1‐containing VANC1 induces strong and sequence‐specific hypomethylation of related VANDALs.

We next tested by RT–PCR experiments whether VANC1 can induce expression of VANDAL1/2‐encoded genes and found that VANB and VANC, but not the putative transposase VANA, which carries disabling mutations including a premature stop, are transcriptionally reactivated in VANC1‐expressing lines (Appendix Fig S10A–C). We then assessed whether expression of VANC1 can induce transposition of target sequences, as was previously found for VANC21 (Fu et al, 2013). We estimated VANDAL1 and VANDAL2 copy numbers by comparing WGBS coverage between VANC1‐TG and control samples, but did not observe significant differences in coverage (Appendix Fig S11), consistent with the lack of expression of the putative transposase VANA. This last result demonstrates that VANC1‐induced hypomethylation does not require and is not sufficient to trigger transposition.

Impaired VANC targeting of non‐autonomous copies induces family‐wide epigenetic re‐silencing

Beyond the demethylation of cognate VANDAL1 copies, the ectopic expression of VANC1 also induces efficient demethylation of TEs belonging to the non‐autonomous VANDAL1N1 and VANDAL2N1 families (Figs 4D and 5A). Copies belonging to these two families are short, do not produce mRNAs in ddm1 (Fig 3B), and lack any predicted open reading frames (ORFs). Therefore, VANDAL1N1 and VANDAL2N1 copies must rely on factors encoded by other TEs for their amplification. To hijack the transposition machinery, such non‐autonomous TEs have terminal sequences that are recognized by transposases encoded by autonomous TEs. Accordingly, terminal sequences of VANDAL1N1 and VANDAL2N1 copies are almost identical to the ones of VANDAL1 and VANDAL2, respectively (Appendix Fig S12A and B). Notwithstanding, internal sequences of VANDAL1N1 and VANDAL2N1 have no sequence homology with their cognate autonomous TEs (Fig 5A), nor with any other sequence in the A. thaliana genome, suggesting that these non‐autonomous families originated from complex sequence rearrangements. Despite their chimeric origin and the lack of sequence homology, internal regions of VANDAL1N1 and VANDAL2N1 copies have high densities of the VANC1 short motif “TGTACGTMY” (Fig 5A and B). However, the exact spatial organization of these motifs differs from that of VANDAL1 and VANDAL2 (Appendix Fig S12C), indicating that non‐autonomous copies have accumulated VANC‐targeting motifs anew.

Figure 5. VANC‐induced hypomethylation of non‐autonomous copies prevents family‐wide epigenetic resilencing.

1. Schematic diagram of structures of full‐length VANDAL1 and VANDAL2 copies and their derived non‐autonomous VANDAL1N1 and VANDAL2N1, respectively. Regions with high sequence homology as well as the location of motifs associated with hypomethylation are indicated in grey and pink, respectively.
2. Genome browser tracks showing hypomethylation effect of VANC1 on a VANDAL1N1 and VANDAL2N1 copy.
3. Density (#siRNAs/1,000 bp) of perfectly multiple‐matching 24‐nt‐long small RNAs and sequence divergence (% global dissimilarity) from functional VANC‐encoding VANDAL1 is shown for each VANDAL1 and VANDAL1N1 TE (red and pink dots, respectively).
4. Schematic diagram of transgene structures of original VANDAL1N1 (1N1) and modified version lacking all VANC1 short‐sequence motifs (1N1Δmotif).
5. Comparison of CHG hypomethylation between replicates of plants containing VANC1, VANC1 + 1N1, or VANC1 + 1N1Δmotif transgenes. VANDAL1, 1N1, 2, and 2N1 copies are indicated by colors.
6. Genome browser tracks showing the effect of 1N1 and 1N1Δmotif transgenes on VANC1‐induced hypomethylation over a VANDAL1 and VANDAL1N1 copy. Methylation levels over the VANDAL1N1 copy (AT5TE61035) in 1N1 samples reflect the average methylation of the endogenous and introduced copy.

Given that VANC activity is not required for transposition (Fu et al, 2013), and that DNA methylation of TE sequences primarily acts to repress their transcription, it is very intriguing that VANDAL1N1 copies with no transcriptional potential are nonetheless efficiently targeted by VANC1‐induced demethylation. One possibility is that the terminal sequences of non‐autonomous copies could act as reservoirs of small RNAs that can then trigger the epigenetic silencing of VANC‐encoding TEs in trans via the RNA‐directed DNA methylation (RdDM) pathway. Indeed, when active VANCs are expressed, VANDAL‐matching 24‐nt small RNAs are strongly reduced (Rigal et al, 2016). Under this scenario, deficient hypomethylation of related VANDAL sequences would lead to the continuous accumulation of trans‐matching small RNAs, which in turn could counteract VANC activity. Consistently, re‐analysis of small RNA sequencing data obtained from wild‐type inflorescences (Creasey et al, 2014) shows that the single VANC1‐encoding copy (AT1TE56425) and several VANDAL1N1 copies generate abundant perfectly multiple‐matching 24‐nt‐long small RNAs, the density of which decreases with the DNA‐sequence divergence between these TEs (Fig 5C). We thus hypothesized that VANC‐induced hypomethylation of non‐autonomous TEs may prevent the silencing of related autonomous VANDALs through identity‐based RdDM. Under this scenario, impairing VANC targeting of a related VANDAL copy should trigger epigenetic silencing of the whole TE family. To test directly this hypothesis, we introduced in the genome of VANC1‐TG plants a copy of the non‐autonomous VANDAL1N1 (AT5TE61035), or a modified version of it that is devoid of the short‐sequence motifs required for VANC1 targeting (1N1 and 1N1Δmotif copies, respectively; Fig 5D). Supporting our hypothesis, introduction of the 1N1Δmotif sequence was sufficient to fully abolish VANC1‐induced non‐CG hypomethylation of the 5′ terminal region of the endogenous VANC1‐encoding copy, whereas introduction of the 1N1 copy had no effect (Appendix Fig S13). To confirm this result genome wide, we obtained WGBS for two independent 1N1 and 1N1Δmotif transgenic lines and found that all VANDAL1/1N1/2 and 2N1 copies become systematically re‐methylated following the introduction of the 1N1Δmotif sequence (Fig 5E and F; Appendix Fig S14). Furthermore, 1N1Δmotif‐induced DNA remethylation goes beyond the terminal regions with high sequence homology between copies (Fig 5F), implying the involvement of heterochromatin spreading and/or production of secondary small RNAs. The capacity of 1N1Δmotif to induce strong silencing of related VANDALs is reminiscent to the silencing activity of the natural Mu killer locus from maize (Slotkin et al, 2005), which is a non‐autonomous mutated derivative of an active Mu transposon. Together, these results demonstrate that VANC‐induced hypomethylation of non‐autonomous copies is critical to avoid small RNA‐mediated epigenetic re‐silencing of related VANDAL sequences and that homologous copies lacking VANC motifs can act as VANDAL killers. Thus, coordinated acquisition and diversification of VANC‐specific short‐sequence motifs across related copies is key for the evolution of efficient anti‐silencing systems and propagation of VANDALs.

Discussion

Epigenetic control of TEs imposes strong selective constraints, engaging hosts and TEs in intimate co‐evolutionary dynamics (Hurst & Werren, 2001; Cosby et al, 2019). One possible outcome of these dynamics is the evolution of TEs that can escape host repression and propagate across the genome. Here, we have exploited the ddm1 epiRIL population in combination with long‐read transcriptome sequencing, quantitative genetic approaches, ectopic expression of TEs, and evolutionary analysis to identify and characterize a remarkably diverse family of VANC anti‐silencing factors encoded by VANDAL DNA transposons specifically in eudicots.

The epiRILs were designed to investigate the epigenetic basis of phenotypic traits (Johannes et al, 2009; Cortijo et al, 2014). Our work demonstrates that this experimental population also provides a powerful system to study VANDALs anti‐silencing, which in turn calls for its use to identify other types of TE‐encoded anti‐silencing systems in A. thaliana. Indeed, the McClintock’s Suppressor–mutator (Spm) element from Maize encodes the anti‐silencing factor TnpA, which can bind to, and induce hypomethylation of, cognate Spm copies (Gierl et al, 1988; Schläppi et al, 1994). Given our observations that the A. thaliana genome contains numerous Spm families showing evidence of trans‐hypomethylation in the epiRILs (Appendix Fig S15), including the highly active Spm3 (Miura et al, 2001; Kato et al, 2004; Quadrana et al, 2019), a next step would be to characterize Spm anti‐silencing systems in this species. Furthermore, ddm1 mutants have also been obtained in other plants, including tomato (Corem et al, 2018), rice (Tan et al, 2018), and maize (Li et al, 2014), thus providing useful systems to study systematically the existence of TE‐encoded anti‐silencing factors across plants. As a matter of fact, the putative transposase of Mu, MURA, was shown to demethylate the terminal inverted repeats (TIRs) of cognate transposons (Burgess et al, 2020), indicating that TE‐encoded anti‐silencing systems could be much more common than previously thought.

Our analyses indicate that VANCs likely originated in eudicots soon after their divergence from monocots and contain a characteristic N‐terminal DUF1985, which in its ancestral form is typically combined with a C‐terminal Ulp1 domain. This observation, together with the finding that Ulp1‐containing VANC1 induces strong sequence‐specific anti‐silencing in A. thaliana, suggest that these domains were at the origin of VANDALs anti‐silencing systems. Incidentally, these results also reveal that VANC proteins derived from the fusion of a de novo originated and a captured cellular domain, DUF1985 and Ulp1, respectively. In yeast, the N‐terminal and C‐terminal domains of Ulp1‐containing proteins are, respectively, involved in protein targeting and removal of small ubiquitin‐related modifier (SUMO) (Johnson, 2004), which has been shown to repress the retrotransposon Ty1 in this species (Bonnet et al, 2021) as well as to contribute to heterochromatin formation in multiple organisms (Maison et al, 2016; Ninova et al, 2020; Andreev et al, 2022; Sheban et al, 2022). Thus, it is reasonable to speculate that the N‐terminal domain DUF1985 guides VANCs to target sequences while the C‐terminal domain Ulp1 participates in their hypomethylation, possibly by affecting chromatin‐associated SUMO. Notably, TEs encoding Ulp1‐containing proteins are pervasive across the tree of life (Marín, 2010; Böhne et al, 2011; Lisch, 2015), suggesting that acquisition of desumoylation activities could be a recurrent evolutionary response of TEs to escape epigenetic silencing.

The deep conservation of Ulp1‐containing VANCs contrasts with the many VANCs in Brassicaceae species that lack this domain and that have instead the uncharacterized DUF287, including VANC21 and VANC6. DUF287 has no sequence similarity with any other type of protein described so far, indicating that it has originated de novo. How these, as well as other TE‐encoded orphan proteins, such as the accessory factor MURB encoded by Mu (Lisch, 2002), originated remains elusive. Strikingly, DUF287‐containing families already account for almost half of the VANDAL copies in the A. thaliana genome, suggesting that this derived anti‐silencing factor may have contributed to the invasiveness of VANDALs in this group of species (Dupeyron et al, 2019). Because the origin of VANC‐like proteins predated the radiation of eudicot species, during a time of intensive diversification and rapid evolution, it is tempting to speculate that the conflicts between host suppression pathways and TEs may have contributed to the evolution of epigenetic control systems in this remarkably diverse clade of plants. In this sense, the characteristic VANC domain DUF1985 has been recurrently captured and fused to diverse host proteins containing distinct DNA‐ or chromatin‐binding domains (Appendix Fig S7B), likely leading to the creation of new cellular functions. Determining the precise function of VANC‐containing DUF1985, Ulp1 and DUF287 domains will be key to understanding how the emergence of new TE‐encoded functions shapes the evolution of TEs and host silencing mechanisms.

VANC proteins bind DNA in vitro and in vivo (Hosaka et al, 2017) and we provided evidence that highly specific self‐recognition is determined by the sequence, density, and spatial arrangement of short motifs, which are typically spaced by 10bp within non‐coding regions of VANDALs. The finding that this motif syntax is conserved across distantly related VANDALs points to a functional role. For instance, helical periodicity may enhance binding and polymerization of VANCs on the DNA sequence. Such homopolymerization of VANCs may shield VANDAL DNA sequences from DNA methyltransferases while inducing active DNA demethylation. Based on our findings, a key priority for the future will be to investigate the cooperative DNA‐binding, polymerization and demethylation activity of VANC proteins.

Our study revealed that VANC‐induced anti‐silencing systems target autonomous as well as non‐autonomous VANDAL families, which appear to have accumulated VANC‐targeting sequence motifs anew. The latter type of TEs derives from full‐length copies by truncation as well as accumulation of random mutations. To ensure propagation, these elements must hijack the transposition machinery from other TEs, leading some authors to consider non‐autonomous TEs as analogous to hyperparasites (Robillard et al, 2016). Our findings now establish that non‐autonomous copies also hijack the anti‐silencing mechanisms of related VANDALs to promote their own hypomethylation. This observation was initially puzzling as VANC seems to be not essential for transposition (Fu et al, 2013). However, we found that non‐autonomous VANDALs may serve as important reservoirs of multiple‐matching 24‐nt‐long small RNAs targeting autonomous copies in a homology‐dependent manner. Indeed, the density of these siRNAs decreases with the divergence between non‐autonomous and autonomous VANDALs. In contrast, a handful of well‐spaced motifs within non‐autonomous copies are sufficient to trigger VANC‐induced hypomethylation and hence reduction in small RNA accumulation. Such persistent VANC targeting beyond the recognition of small RNAs would protect related VANDALs (i.e., belonging to the same family) from host silencing mechanisms.

The remarkable capacity of the 1N1Δmotif sequence to induce strong and concerted epigenetic re‐silencing of related VANDALs is reminiscent of the silencing activity of the Mu killer locus, which is a naturally occurring non‐autonomous derivative of the maize Mu transposon that express a long hairpin transcript (Slotkin et al, 2005). TE sequences are frequently mutated and rearranged, particularly during transposition, and it has been proposed that non‐autonomous copies may be a common source of transposon silencing triggers (Slotkin et al, 2005; Burgess et al, 2020; Wang et al, 2020). However, unlike VANDAL killer, which does not produce long transcripts and is associated with 24‐nt‐long siRNAs, the hairpin transcript of Mu killer is processed into 22‐nt‐long small RNAs, which trigger de novo DNA methylation of full‐length Mu copies (Burgess et al, 2020). Therefore, different transposon silencing triggers may rely on distinct molecular mechanisms. TE transgenes can be methylated through the identity‐based silencing mechanism, which is mostly dependent on 24‐nt‐long small RNAs produced from endogenous TEs by the plant‐specific RNA Polymerase IV (Fultz & Slotkin, 2017). De novo establishment of RdDM targeting is still enigmatic, although recent research showed the importance of transcription by Pol II (Sigman et al, 2021). Conversely, reinforcement of DNA methylation can be mediated by RdDM‐dependent and ‐independent mechanisms, which both rely on the presence of remaining epigenetic mark(s) at target loci (To et al, 2020). In this sense, VANCs do not erase all epigenetic marks over target TEs, as CG methylation outside motifs remains unaffected, providing the epigenetic memory that is required for resilencing. Indeed, DNA methylation of VANDAL21 is rapidly restored when VANC21‐TG is segregated apart (Fu et al, 2013), indicating that VANC‐induced hypomethylated VANDALs are in a labile epigenetic state that can be readily resilenced. Under this scenario, non‐autonomous VANDAL copies that are no longer targeted by VANCs can induce remethylation of related VANDAL sequences through identity‐based Pol IV‐RdDM (Fultz & Slotkin, 2017). Gain and loss of VANC‐targeting sequence motifs, or even the perturbation of their helical periodicity by accumulation of short indels, may happen remarkably fast due to imprecise transposition, replication slippage, unequal crossing over, and/or small‐scale mutations. Therefore, the accumulation of mutations during sustained VANDAL proliferation is expected to eventually transform non‐autonomous VANDAL copies from hyperparasites to killers. Such spontaneous formation of VANDAL killers may in turn provide an efficient self‐control mechanism to limit runaway VANDAL proliferation, protecting genome persistence and of the TEs it contains.

To conclude, our findings reveal that the co‐evolution between host silencing and TEs, as well as their interactions with hyperparasitic non‐autonomous copies, shaped the diversification and invasive success of VANDAL TEs, with potential implications for the emergence of novel gene control mechanisms.

Materials and Methods

Plant materials

The A. thaliana Col‐0, ddm1‐2 mutant, and the 16 epiRILs (Johannes et al, 2009) lines used in this work were described before (Colomé‐Tatché et al, 2012; Quadrana et al, 2019). All plants were grown in long days (16 h:8 h light:dark) at 23°C. The VANC1 and VANC2 constructs were generated by amplifying genomic sequences of VANCs by PCR and cloned into pPLV01 vector double digested by HpaI and Eco53kI using NEBuilder (NEB). The 1N1 and 1N1Δmotif constructs were cloned into SmaI‐digested pGreenII‐0179. For 1N1Δmotif, “TGTACGTMY” motifs were converted to “TGTATATMY” by PCR‐based site‐directed mutagenesis. Constructs were transformed into wild‐type (VANC1 and VANC2) or VANC1‐TG (1N1 and 1N1Δmotif) plants of A. thaliana Col‐0 ecotype by floral dip (Clough & Bent, 1998).

Whole‐genome bisulfite sequencing and DMRs detection

DNA from epiRILs was extracted using a standard CTAB protocol. Bisulfite conversion, BS‐seq libraries, and sequencing (paired‐end 100 nt reads) were performed by BGI Tech Solutions (Hong Kong). For WGBS of transgenic plants, bisulfite treatment and library preparation were conducted as previously described (Fu et al, 2013). In all cases, paired‐end reads were trimmed using Trimmomatic program (version 0.33) with the following parameters “ILLUMINACLIP:TruSeq3‐PE.fa:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36” (Bolger et al, 2014). Mapping of trimmed sequences to Arabidopsis reference genome (TAIR10) with option “‐n 1 ‐l 20,” removal of identical reads, and counting of methylated and unmethylated cytosines were performed by Bismark ver. 0.15.0 (Krueger & Andrews, 2011). MethylKit package v0.9.4 (Akalin et al, 2012) was used to calculate differential CG methylation in 100 bp non‐overlapping windows (DMRs) between epiRILs and wild type. Significance of calculated differences was determined using Fisher’s exact test and Benjamin–Hochberg (BH) adjustment of P‐values (FDR < 0.05) and methylation difference cutoffs of 40%. Metaplots of mCG across wt‐derived VANDAL copies shown in Fig 1D and Appendix Fig S4 were performed using deeptools v3.4.0 and DNA methylation data from wt, ddm1, and the indicated epiRIL in each case. DNA methylation ratio in 120 bp bin for each cytosine context was calculated and compared between WT and VANC1‐TG. Bins whose change in CG methylation ratio was 0.5 or more were determined as CG‐hypoDMRs. Significance of decrease in DNA methylation for TEs in each cytosine context (Figs 4D and 5E) was accessed by value (Mn/Cn − Mt/Ct)/(1/√Cn + 1/√Ct), where Mn, Cn, Mt, and Ct are methylated cytosine (M) and total cytosine (C) counts mapped for each TEs in the non‐transgenic (n) and transgenic (t) plants, respectively (Fu et al, 2013). Overview of the bisulfite data for the epiRILs is provided in Appendix Table S1.

Targeted bisulfite analysis

Conventional bisulfite sequencing analysis for endogenous AT1TE56425 (VANDAL1) was performed as described previously (Saze & Kakutani, 2007), using primers listed in Appendix Table S2. For each sample, at least 15 clones were sequenced.

Short motif detection

DNA sequences in epiRILs CG‐hypoDMRs and within annotated VANDALs were extracted using fastaFromBed command (bedtools) and analyzed by meme (Bailey, 2011) with the following parameters: ‐dna ‐oc ‐nostatus ‐time 18000 ‐maxsize 60000 ‐mod anr ‐nmotifs 3 ‐minw 8 ‐maxw 12 ‐revcomp. For CUMULE (GenBank: AY524004), DNA sequences outside coding regions, predicted using GENSCAN (Burge & Karlin, 1997), were extracted and processed as described above. Presence of specific short‐sequence motifs across all VANDAL copies was evaluated using fimo script. Isolated, intermediate (2–3 motifs/1kbp) and high (+4 motifs/1 kbp) density of motifs were determined by counting the number of motifs in 1,000 bp windows. Distance between consecutive motifs was obtained using closestBed command (bedtools) and the frequency distribution of these distances was represented as a heatmap. Palindromic Index for motifs was calculated as the average fraction of palindromic DNA within motif instances using an in‐home script (accessible at https://github.com/LeanQ/palindromes). DNA sequences in VANC1 CG‐hypoDMRs overlapping VANDAL1/2/1N1/2N1 (N = 325) were used for prediction of VANC1‐targeted motifs by DREME script of MEME software version 4.11.0 (Bailey, 2011).

epiQTL mapping of VANDAL’s hypomethylation

Using methylation level based on MedIP data (Colomé‐Tatché et al, 2012) for each wt‐derived VANDAL copy containing CG‐hypoDMRs as a trait and a total of 126 parental differentially methylated regions (DMRs) that segregate in a Mendelian fashion in 105 epiRILs (i.e., stable DMRs) as physical markers (Colomé‐Tatché et al, 2012), we performed individuals epiQTL mappings based on the multiple QTL model (mqmsacn) from the R/qtl package. Genome‐wide significance was determined empirically for each trait using 1,000 permutations of the data. LOD significance thresholds were chosen to correspond to a genome‐wide false‐positive rate of 5%. To summarize epiQTL results obtained for the different copies of the same VANDAL family, LOD scores were first transformed to P‐values using the following function in R: pchisq(LOD*(2*log(10)), df = 1, lower.tail = FALSE)/2. Meta‐analysis was calculated as previously described (Sasaki et al, 2019). Statistical threshold was defined as the 1% (P = 0.01) lowest meta‐analysis P‐values genome wide.

Full‐length cDNA nanopore sequencing

Total RNA was extracted from 100 mg of rosette leaves from ddm1‐2 plants using the Nucleo‐spin RNA Plant mini kit (Macherey‐Nagel). Library preparation and Nanopore sequencing were performed as previously described by Domínguez et al (2020). Briefly, 10 ng of total RNA was amplified and converted into cDNA using SMART‐Seq v4 Ultra Low Input RNA kit (Clontech). About 17 fmol of cDNA was used for library preparation using the PCR Barcoding kit (SQK‐PBK004 kit, ONT) and cleaned up with 0.6× Agencourt Ampure XP beads. About 2 fmol of the purified product was amplified during 18 cycles, with a 17‐min elongation step, to introduce barcodes. Samples were multiplexed in equimolar quantities to obtain 20 fmol of cDNA, and the rapid adapter ligation step was performed. Multiplexed library was loaded on an R9.4.1 flowcell (ONT) according to the manufacturer’s instructions. A standard 72 h sequencing was performed on a MinION MkIB instrument. MinKNOW software (version 19.12.5) was used for sequence calling.

RT–PCR

Total RNA was extracted from seedlings of WT and VANC1‐TG using TRIzol (Thermo Fisher), and treated with DNase I (Invitrogen). cDNA was synthesized using 3 μg of total RNA by SuperScript III (invitrogen). Ten times diluted cDNA was used as a template for RT–PCR. Primers used for RT–PCR were listed in Appendix Table S2.

Functional annotation of TE‐encoding genes

Long reads from ddm1 plants were mapped on the Arabidopsis reference genome (TAIR10) using minimap v2.11‐r797 (H. Li, 2018) with the following options ‐ax splice ‐G 30k ‐t 12 and STAR v2.5.3a (Dobin et al, 2013) with the following options ‐‐outFilterMultimapNmax 50 ‐‐outFilterMatchNmin 30 ‐‐alignIntronMax 10000 ‐‐alignSJoverhangMin 3, respectively. Previously published short reads (Oberlin et al, 2017) were also mapped on the Arabidopsis reference genome (TAIR10) using STAR (Dobin et al, 2013). Transcript annotation was performed using the FLAIR pipeline (Tang et al, 2020) available at https://github.com/BrooksLabUCSC/FLAIR. First, splicing junctions based on short‐read sequencing data were extracted using “junctions_from_sam.py” script and used to correct ONT long reads using “flair.py correct” script. Transcript isoforms were then detected using the “flair.py collapse” script, and transcripts supported by at least five long reads were retained. Annotated transcripts overlapping VANDAL elements were extracted using intersectBed and translated in silico using getorf vEMBOSS:6.6.0.0. Putative VANCs proteins were identified by BLAST against the functionally characterized VANC21 and VANC6 (Hosaka et al, 2017). Conserved protein domains were detected using HHMscan (https://www.ebi.ac.uk/Tools/hmmer/search/hmmscan) against the Pfam database.

Phylogenetic analysis

Putative VANC proteins from A. thaliana as well as CUMULE were aligned using MAFFT v7.271 and trimmed with trimAl v1.4.rev15 (Capella‐Gutiérrez et al, 2009). Phylogenetic tree was generated with PhyML v20160207 (Guindon & Gascuel, 2003) using subtree pruning and regrafting (SPR) topological moves. Phylogenetic tree of VANDAL1/2/1N1 and 2N1 copies was generated with Clustal OMEGA (https://www.ebi.ac.uk/Tools/msa/clustalo/) using 250 bp sequences of terminal inverted repeats (TIRs). Termini of TEs were determined by target site duplications. Sequence divergence between AT1TE56425 and VANDAL1 and VANDAL1N1 copies was calculated using the command line version of Blast2seq (bl2seq), with the following parameters ‐p blastn ‐e 0.05 ‐D 1 ‐r 2 ‐G 5 ‐E 2.

Dot plot analysis

DNA sequences of non‐coding regions in VANDAL1/2/1N1/2N1 were compared by dot plot analysis. Dot plots were made with EMBOSS dotmatcher (https://www.bioinformatics.nl/cgi‐bin/emboss/dotmatcher) default setting (10 window size, 23 threshold).

Small RNA analysis

Small RNA data from Col‐0 wild‐type inflorescence were obtained from Creasey et al (2014). Reads were trimmed using the Trimmomatic program (version 0.33) and first mapped on AT1TE56425 sequence using Bowtie2 with the following parameters: ‐‐local ‐‐very‐sensitive. Mapped 24‐bp‐long reads were then extracted using Samtools and aligned on the collection of VANDAL1 and VANDAL1N1 sequences, excluding AT1TE56425, using Bowtie2 with the following parameters: ‐‐local ‐‐very‐sensitive ‐k 10.

Author contributions

Taku Sasaki: Conceptualization; Resources; Data curation; Formal analysis; Validation; Investigation; Visualization; Methodology; Writing—review & editing. Kyudo Ro: Methodology. Erwann Caillieux: Methodology. Riku Manabe: Methodology. Grégoire Bohl‐Viallefond: Data curation. Pierre Baduel: Formal analysis; Writing—review & editing. Vincent Colot: Conceptualization; Resources; Supervision; Writing—review & editing. Tetsuji Kakutani: Conceptualization; Resources; Supervision; Writing—review & editing. Leandro Quadrana: Conceptualization; Resources; Data curation; Formal analysis; Supervision; Validation; Investigation; Visualization; Methodology; Writing—original draft; Project administration.

In addition to the CRediT author contributions listed above, the contributions in detail are:

TS, VC, TK, and LQ conceived the project. TS and KR performed VANC1 transgenic experiments. TS and RM performed 1N1Δmotif experiments. EC extracted genomic DNA for WGBS of the epiRILs and GB‐V processed the WGBS data. LQ analyzed the WGBS data, ONT results, and performed evolutionary analyses. PB and LQ performed epiQTL mapping. TS and LQ interpreted the data. LQ drafted the manuscript with additional input from TS, PB, VC, and TK. All the authors read and approved the manuscript.

Disclosure and competing interests statement

The authors declare that they have no conflict of interest.

Supporting information

Appendix

The EMBO Journal (2022) 41: e110070.

Data availability

Original Scripts are available on GitHub (https://github.com/LeanQ/palindromes). WGBS of epiRILs is available at the European Nucleotide Archive (ENA) under project PRJEB47214 (https://www.ebi.ac.uk/ena/browser/view/PRJEB47214) and NCBI Gene Expression Omnibus (GEO) as GSE62206 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE62206). WGBS of VANC1 transgenic lines has been deposited in the DDBJ under project PRJDB12220 (https://ddbj.nig.ac.jp/resource/bioproject/PRJDB12220).