Abstract
Mobilization of transposable elements (TEs) in plants has been recognized as a driving force of evolution and adaptation, in particular by providing genes with regulatory modules that impact their transcription. In this study, we employed an ATCOPIA93 Long terminal repeats (LTR) promoter-GUS fusion to show that this retrotransposon behaves like an immune-responsive gene during plant defense in Arabidopsis. We also showed that the reactivation of the endogenous ATCOPIA93 copy EVD, in the presence of bacterial stress, is not only negatively regulated by DNA methylation but also by Polycomb-mediated silencing — a mode of repression typically found at protein-coding and microRNA genes. Interestingly, one of the ATCOPIA93-derived soloLTRs is located upstream of the disease resistance gene RPP4 and is devoid of either DNA methylation or H3K27m3 marks. Through loss-of-function experiments, we demonstrated that this soloLTR is required for proper expression of RPP4 during plant defense, thus linking the responsiveness of ATCOPIA93 to biotic stress and the co-option of its LTR for plant immunity.
Introduction
TEs are repeated sequences that can potentially move and multiply in the genome. Their mobilization has been recognized as a driving force of evolution and adaptation in various organisms, in particular by providing genes with regulatory modules that can create or impact transcriptional programs (Chuong et al, 2016). The study of TE regulation is thus important in order to understand both the conditions for their transposition but also their influence, as full-length or truncated elements, on nearby gene regulation. This role in cis has been demonstrated in plants by artificially inducing insertions that confer gene regulation, e.g., the rice TE mPing (Naito et al, 2009) or the Arabidopsis TE ONSEN (Ito et al, 2011). However, the causal link between cis-regulatory properties of TEs and established expression patterns of nearby genes, requires loss-of-function experiments and has rarely been demonstrated (Chuong et al, 2016).
TE cis-regulatory effects can either be genetic in nature, such as when the TE contains regulatory motifs, or epigenetic through recruitment of dimethylation of histone 3 lysine 9 (H3K9m2) and cytosine DNA methylation, which are hallmarks of transposon control. DNA methylation in Arabidopsis is carried out by four pathways. While METHYLTRANSFERASE1 (MET1) (Kankel et al, 2003) maintains CG methylation, CHROMOMETHYLASE2 and 3 (CMT2 and CMT3) maintain CHG methylation (where H is any base pair but not a G) (Zemach et al, 2013; Stroud et al, 2014). The maintenance of CHH methylation requires either CMT2 for long heterochromatic repeats or RNA-directed DNA methylation (RdDM) mediated by DOMAINS REARRANGED METHYLASE 2 (DRM2) and accompanying small RNA machinery (Cao & Jacobsen, 2002; Chan, 2004; Deleris et al, 2016). In addition, the SNF2 family chromatin remodeler DECREASED DNA METHYLATION 1 (DDM1) is necessary for heterochromatic DNA methylation in all cytosine sequence contexts (Jeddeloh et al, 1999; Zemach et al, 2013). Furthermore, DNA methylation and H3K9m2 are mechanistically interconnected and, as a result, are largely co-localized throughout the genome (Du et al, 2015). Importantly, this histone and cytosine marking can also impact the nearby genes which then become epigenetically controlled because of the inhibitory effect of DNA and H3K9 methylation on promoter activity (Lippman et al, 2004; Liu et al, 2004; Huettel et al, 2006; Gehring et al, 2009). In addition, in both plants (Mathieu et al, 2005; Deleris et al, 2012; Weinhofer et al, 2010) and animals (Reddington et al, 2013; Saksouk et al, 2014; Basenko et al, 2015; Walter et al, 2016), the removal of DNA methylation at some TEs leads to H3K27 trimethylation (H3K27m3), an epigenetic mark deposited and interpreted by Polycomb Group (PcG) proteins, which normally target and silence protein coding genes that are often developmentally important (Förderer et al, 2016). Thus, there exists a potential for this alternative repression system to mediate silencing of TEs, but it has not been fully explored in plants, except in the endosperm, a nutritive and terminal seed tissue that is naturally DNA hypomethylated (Weinhofer et al, 2010; Moreno-Romero et al, 2016).
In accordance with the epigenetic control mediated by DNA and histone H3K9 methylation, mutations in DNA methylation pathway genes lead to reactivation of various subsets of TEs; however, these defects in chromatin regulation are not always sufficient for TE expression, and the activation of specific signaling pathways is sometimes needed. This has been exemplified by the Arabidopsis Long-Terminal Repeats (LTR)-retrotransposon ONSEN, which was shown to be reactivated after heat-stress, in wild-type plants and independently from a loss of DNA methylation in this context (Ito et al, 2011; Cavrak et al, 2014). In addition, ONSEN was not expressed in unstressed RdDM-defective mutants (nor in ddm1), but its induction was enhanced in RdDM-defective mutants subjected to heat stress (Ito et al, 2011). To our knowledge, ONSEN is the only described example of a TE which expression is modulated by DNA methylation during stress response. Thus, it is important to characterize other TEs that exploit plant signaling, in order to gain a deeper understanding of the connection between biotic/abiotic stresses and transposon activation, and how epigenetic silencing pathways exert their influence on this relationship.
In this study, we unravel the responsiveness of another family of Arabidopsis retroelements, ATCOPIA93 (Mirouze et al, 2009; Marí-Ordóñez et al, 2013), during PAMP-triggered immunity (PTI). PTI is defined as the first layer of active defense against pathogens and relies on the perception of evolutionary conserved Microbe- or Pathogen-associated molecular patterns (MAMPs or PAMPs) by surface receptors (Boutrot & Zipfel, 2017). ATCOPIA93 is a low-copy, evolutionary young family of LTR-retroelements, which is tightly controlled by DNA methylation, in particular CG methylation (Mirouze et al, 2009). The family representative EVD (AT5G17125) was found to transpose in ddm1 after eight generations of inbreeding (Tsukahara et al, 2009) as well as in genetically wild-type epigenetic recombinant lines (epiRILs) derived from crosses between wild-type and met1 (Mirouze et al, 2009, Marí-Ordóñez et al, 2013) or ddm1 (Marí-Ordóñez et al, 2013). EVD is 99.5% identical in sequence to the pericentromeric ATR (AT1G34967), which is predicted to encode a polyprotein but does not seem to be active in the latter conditions (Mirouze et al, 2009). Here, we first took advantage of an unmethylated ATCOPIA93 LTR-GUS fusion that we used as a reporter of promoter activity, since LTRs of retroelements contain cis-regulatory sequences that can recruit RNA Pol II (Chuong et al, 2016). We showed that this LTR exhibits the hallmarks of a promoter of an immune-responsive gene in the absence of epigenetic control. Accordingly, the corresponding methylated endogenous ATCOPIA93 retroelements, EVD and ATR, were only significantly reactivated after PAMP-elicitation in a DNA hypomethylated background, in met1 or ddm1 mutants. Interestingly, we demonstrated for the first time in wild-type plant vegetative tissues, a second layer of control of TE expression mediated by Polycomb silencing. Importantly, we showed that H3K27m3 co-exists with DNA methylation at EVD sequences but not at ATR, leading to a differential negative control between these two copies during immunity. Furthermore, we were able to test the implications of these findings for the regulation of the immune response. We identified an ATCOPIA93-derived soloLTR, unmethylated and not marked by H3K27m3, upstream of the RPP4 disease resistance gene. By measuring the impact of the genetic loss of this soloLTR, we could show that it has been co-opted for the proper expression of RPP4 during PAMP-triggered plant basal defense and thus that it plays a role as regulatory “enhancer” element. Thus, we established a link between the responsiveness of a TE to biotic stresses and the co-option of its derived soloLTR for plant immunity, where the repressive epigenetic modifications controlling the full-length active elements are absent on the derived regulatory sequence.
Results
ATCOPIA93-LTR::GUS transcriptional fusion behaves as a canonical immune-responsive gene
An ATCOPIA93-LTR::GUS construct –comprising the full EVD/ATR LTR upstream of a sequence encoding a GUS protein (schema Fig 1A, EV1A) – was transformed into the Arabidopsis wild type reference accession Columbia (Col-0), initially to serve as a reporter of DNA methylation levels. Unexpectedly, the LTR::GUS transgenes were not methylated in any of the transgenic lines obtained (Fig 1A, Fig EV1B). Instead, we generated a reporter of ATCOPIA93 promoter activity that we could exploit to assess ATCOPIA93 responsiveness during PTI in the absence of DNA methylation-mediated control. Plants containing the LTR::GUS transcriptional fusion were further elicited with either flg22 (a synthetic peptide corresponding to the conserved N-terminal region of bacterial flagellin that is often used as a PAMP surrogate) (Zipfel et al, 2004; Boutrot & Zipfel, 2017), or PtoΔ28E (a non-pathogenic Pseudomonas syringae pv. tomato DC3000 (Pto DC3000) in which 28 out of 36 effectors are deleted (Cunnac et al, 2011)), and the accumulation of GUS mRNA and protein was monitored over a 24 hour time course. In water-treated plants, at 24 hours post-infiltration (hpi), there was barely any GUS staining, indicating that the activity of the ATCOPIA93 LTR promoter is weak in this condition. By contrast, in both flg22 and PtoΔ28E treatments, an intense GUS staining was observed at 24 hpi (Fig 1B, top panel). This was associated with progressive GUS protein accumulation over the time-course, until it reaches a plateau (Fig 1B, bottom panel). At all the time points analyzed, the GUS expression was generally stronger in response to PtoΔ28E than flg22 and thus we focused on the former bacterial elicitor for the rest of the study. By analyzing GUS mRNA levels, we could then show that the GUS induction was transient, similarly as an immune-responsive gene induced rapidly during PTI such as WRKY29 (Asai et al., 2002) (Fig 1C, Fig EV1C). In addition, treatment with the virulent wild-type Pto DC3000 strain, which can inject type III effectors into the host cell, resulted in a partially compromised induction of GUS protein accumulation compared to a similar inoculum of PtoΔ28E (Fig 1D), suggesting that some bacterial effectors suppress the LTR responsiveness during PTI. This transcriptional behavior is reminiscent of typical PTI-induced genes whose induction is impaired by bacterial effectors that have evolved to suppress different steps of PTI to enable disease (Asai & Shirasu, 2015). Together, these data show that the ATCOPIA93 LTR::GUS transgene behaves like a canonical immune-responsive gene, which is transcriptionally reactivated during PTI and whose induction is suppressed by bacterial effectors. In support of this, we noticed in the LTR sequence the presence of two putative W-box elements, i.e. DNA sequences with the C/TTGACC/T (A/GGTCAA/G) motif, which are the cognate binding sites for WRKY transcription factors that are known to orchestrate transcriptional reprogramming during PTI (Rushton et al, 2010) (Fig EV1A). Importantly, we found that these two putative W-box elements are functional as the PtoΔ28E-mediated transcriptional induction of GUS was lost in part or entirely when W-box 1 and W-box 2 were mutated, respectively (Fig 1E).
AtCOPIA93 reactivation is negatively controlled by DNA methylation during PAMP-triggered immunity
We next analyzed, over the same period of time, the reactivation of the almost identical endogenous copies of AtCOPIA93: EVD and ATR. The induction of their expression upon bacterial challenge was generally weak in wild-type leaves (Fig 2A and 2B, Fig EV2). By contrast, we observed a consistent and transient induction of EVD/ATR at 3, 6 and 9 hpi in a ddm1 hypomethylated background (Stroud et al, 2013, Fig 2A and 2B left panel, Fig EV2), specifically in response to the PtoΔ28E strain. EVD/ATR transcript levels were also enhanced at 6 hpi in a bacteria-challenged met1 mutant (Fig 2B, right panel), which is impaired in CG methylation (Ordonnez et al., 2013). Together, these data indicate a tight negative control of ATCOPIA93 induction which is exerted by DNA methylation and is particularly relevant during bacterial challenge when the LTR is activated. Notably, at this developmental stage, mutations in the components of the DNA methylation pathways were not sufficient to enhance ATCOPIA93 expression in the absence of bacterial stress, in accordance with the transcriptional behavior of the unmethylated LTR::GUS fusion which displays weak promoter activity in water-treated plants (Fig 1).
ATCOPIA93-EVD is marked by H3K27m3 chromatin modification in addition to DNA methylation
Given previous observations in plants and mammals that some loci gain H3K27m3 marks upon their loss of DNA methylation (Mathieu et al, 2005; Deleris et al, 2012; Weinhofer et al, 2010; Reddington et al, 2013; Saksouk et al, 2014; Basenko et al, 2015), presumably mediating “back-up” transcriptional silencing of hypomethylated sequences, we thought that there could be an increase in H3K27m3 marks at ATCOPIA93-LTR in ddm1 plants. To test this possibility, we inspected publically available ChIP-chip datasets and found that ATCOPIA93 is marked by H3K27m3, not only in an hypomethylated mutant met1, but also unexpectedly in wild-type plants (Fig 3A). However, one limit of ChIP-chip and ChiP-seq datasets is that they do not allow precise determination of the genomic localization of immunoprecipitated repeated sequences, either because of cross-hybridization (ChIP-chip) or the impossibility to accurately map multiple repeated reads (ChIP-seq). To circumvent this problem, we designed specific qPCR primers to discriminate EVD-LTR from ATR-LTR sequences after ChIP by using upstream genomic sequences. In addition, we took advantage of the rare SNPs between EVD and ATR and used pyrosequencing to analyze the immunoprecipitated fragments from the ATCOPIA93 coding sequence (CDS), where specific primer design is impossible. Interestingly, with both these approaches, we found that there was a strong bias towards EVD molecules in the H3K27m3-IPs (Fig 3B, Fig EV3A and Fig 3C, Fig EV3B). This could be possibly due to a positional effect, as the EVD sequence is embedded in a larger domain of H3K27m3 that comprises seven adjacent genes (Fig 3A) (At5g17080 to At5g17140, coding for either cysteine-proteinases or cystatin-domain proteins). Nevertheless, there was less H3K27m3 in the CDS than in the LTR region (Fig 3B, Fig EV3A); this likely reflects the previously observed antagonism of DNA methylation/H3K9m2 and H3K27m3 (Mathieu et al, 2005; Deleris et al, 2012; Weinhofer et al, 2010) since DNA/H3K9 methylation levels are higher in the coding sequence of EVD than in the LTR (Fig EV3C)(Marí-Ordóñez et al, 2013). Finally, we found that H3K27m3 levels were strongly reduced in clf plants mutated for the Polycomb-Repressive Complex 2 (PRC2) H3K27 methylase CURLY LEAF (Förderer et al, 2016)(Fig 3B).
Next, we assessed whether H3K27m3 and DNA methylation could co-exist on the same molecules or whether the detection of both marks in wild type rosette leaves was only reflecting the contribution of different cell types, some marked by H3K27m3 at EVD and some by DNA methylation. To distinguish between these two possibilities, we analyzed the DNA methylation status of one representative CG site at the LTR region of EVD using a methylation sensitive enzyme assay on H3K27m3-IPed DNA, followed by qPCR (Fig EV1B, bottom panel). We observed amplification of the enzyme-treated DNA, comparable to the total input genomic DNA (Fig 3D). Based on this result we can conclude that the EVD DNA associated with H3K27m3 is methylated and that DNA methylation does not inhibit H3K27m3 deposition in this region.
Polycomb-group proteins and DNA methylation exert differential negative control on ATCOPIA93 induction during PAMP-triggered immunity
The co-existence of DNA methylation and H3K27m3 at EVD-LTR suggests that there is dual control by both PcG- and DNA methylation- mediated silencing on the same molecule, in the same cell type. To test for the functional relevance of PcG silencing at ATCOPIA93-EVD, we challenged wild-type and clf mutant plants with PtoΔ28E and monitored ATCOPIA93 transcript levels by RT-qPCR analyses, using ddm1 mutant plants as a positive control. Results from these analyses revealed a variable but consistent reactivation of ATCOPIA93 expression in bacteria-elicited clf plants, though generally weaker than in elicited ddm1 plants (Fig 4A and 4B). These data indicate that both DNA methylation and PcG negatively regulate the transcriptional activation of ATCOPIA93 during plant defense. Furthermore, by pyrosequencing ATCOPIA93 cDNA in bacteria-elicited plants, we observed that while both EVD and ATR are induced in ddm1, EVD almost exclusively is reactivated in clf-elicited mutant background (Fig 4C, Fig EV4A). This is consistent with EVD exhibiting comparatively stronger H3K27m3 enrichment than ATR in wild-type (Fig 3). The pyrosequencing result also presumably explains, at least partly, the weaker ATCOPIA93 induction observed in clf compared to ddm1 after bacterial challenge (Fig 4A, 4B), as the mRNA is almost only contributed by EVD in clf (Fig 4C). In addition, as anticipated, we observed that in clf mutants, where H3K27m3 is reduced, H3K9m2 marks were retained to levels comparable to WT at EVD (Fig EV4B, top panel); this likely contributes to explain lower accumulation of EVD transcripts in clf than ddm1 (Fig 4D). In ddm1 mutants, where H3K9m2 is reduced, H3K27m3 marks were also retained to levels comparable to WT at EVD (Fig EV4B, lower panel) suggesting that H3K9m2/DNA methylation may exert a stronger repressive effect on EVD than polycomb group proteins.
Cis-regulation of the RPP4 disease resistance gene by a ATCOPIA93-derived, unmethylated soloLTR
The corollary of our findings on ATCOPIA93 regulation is that the presence of a ATCOPIA93 LTR in the genome, if deprived of DNA methylation and H3K27m3, can potentially lead to the transcription of downstream sequences, thus potentially affecting the transcription of adjacent genes. We found through a BLAST search, three new ATCOPIA93-derived sequences in the genome on the chromosome 3, in addition to the ones that were previously annotated on chromosomes 1, 3 and 4. Interestingly, apart from two copies on chromosomes 1 and 4, all other five sequences are present in the form of a soloLTR, which is the product of unequal recombination between the LTRs at the ends of a single retroelement (Fig EV5A). The functional W-box 1 was conserved in all of them making them potentially regulatory units responsive to PAMP-triggered immunity (Fig EV5). Interestingly, detection of transcription was detected in response to various bacterial challenges downstream of soloLTR-1, soloLTR-2 and soloLTR-5 (Fig EV5A). While soloLTR-1 and -2 are located upstream of a pseudogene and an intergenic region respectively, both of unknown function, the soloLTR-5 is embedded in the predicted promoter of the RECOGNITION OF PERONOSPORA PARASITICA 4 (RPP4) gene, less than 500 bp upstream of the annotated transcriptional start site. RPP4 is a canonical and functional disease resistance gene that belongs to the RPP5 cluster on chromosome 4, which is composed of 7 other Toll Interleukin-1 Receptor (TIR) domain-Nucleotide binding site (NBS) and Leucine-rich repeat (LRR) domain (TIR-NBS-LRR genes) (Noel, 1999). Among these genes, RPP4 was previously shown to confer race-specific resistance against the oomycete Hyaloperonospora arabidopsidis isolates Emwa1 and Emoy2 (Van Der Biezen et al, 2002). In addition, we observed that RPP4 was generally induced during PTI, either triggered by PtoΔ28E (Fig EV5B) or by various bacterial and oomycete PAMPs tested at 4 hpi (https://bar.utoronto.ca/eplant/AT4G16860, Tissue and Experiment eFP viewers, Biotic Stress Elicitors, Waese et al., 2017). Interestingly, the soloLTR-5 is completely DNA unmethylated and not marked by H3K27m3 nor H3K9m2 (Fig 5A, Ordonnez et al., 2013).
To test whether the presence of this presumably PAMP-responsive ATCOPIA93-soloLTR has a bona fide impact on the expression of RPP4 during PTI and could be co-opted for regulatory functions, we took a loss-of-function approach. We transformed rrp4 knock-out (KO) mutants with transgenes consisting of the entire RPP4 genomic region under the control of its native promoter (~3kb upstream of the TSS, comprising the whole upstream gene unit) or under the control of the same promoter sequence with a deletion for soloLTR-5 (Fig 5B, “WT” and “ΔLTR” constructs). We further analyzed the primary transformants for RPP4 expression in response to either water or PtoΔ28E treatments. In the absence of the soloLTR-5, RPP4 expression was reduced in both mock-inoculated (Fig 5C, significant decrease between the blue boxplots) and PtoΔ28E-elicited plants (Fig 5C, significant decrease between the red boxplots); in addition, the induction of RPP4 expression upon bacterial elicitation was no longer significant in the absence of the soloLTR-5 (Fig 5C). These results provide evidence that the ATCOPIA93-derived LTR is required for both the basal expression and bacterial-stress induction of RPP4. In addition, to assess whether soloLTR-5 contributes to RPP4 induction during PtoΔ28E elicitation through the conserved W-box1 (Fig EV5C), which had a partial effect on the induction of the LTR::GUS fusion (Fig 1E), we transformed the rpp4 KO mutant plants with a construct comprised of the RPP4 gene under the control of its promoter mutated in the W-box 1 element (Fig 5B, “w1” construct). We found that RPP4 expression levels in the absence of PtoΔ28E-challenge were globally unchanged when the W-box element was mutated (Fig 5D, not significant between the blue boxplots). However, the induction of RPP4 expression was significantly reduced in transgenic plants containing the “w1” mutant version versus plants containing the “WT” transgene upon bacterial challenge (Fig 5D, significant decrease between the red boxplots) and induction no longer occurred in these “w1” transgenic plants (not significant between mock-treated and bacteria-treated “w1” plants). These results demonstrate that the LTR responsiveness to bacterial PAMPs contributes to proper induction of RPP4 and is mediated at least in part by a functional W-box element.
Finally, to assess the relevance of this layer of regulation of RPP4 during immune responses against H. arabidopsidis (to which RPP4 confers race-specific resistance), we treated LTR::GUS plants with NLP20, the active peptide of the oomycete PAMP NPP1 that was previously shown to be non-cytotoxic in planta (Oome et al, 2014). We found that the LTR::GUS fusion was similarly responsive to this oomycete PAMP (Fig 5E), showing that NLP20 induces the same ATCOPIA93-LTR regulation as PtoΔ28E, in accordance with the fact that the PTI responses induced by unrelated PAMPs largely overlap (Katagiri, 2004; Schwessinger & Zipfel, 2008; Zipfel et al, 2006).
Together, these results show that a ATCOPIA93-derived soloLTR has been co-opted during evolution to cis-regulate RPP4 expression during basal immunity.
Discussion
ATCOPIA93 has been a widely used model to study plant transposon biology and epigenetics over the last years (Tsukahara et al, 2009; Mirouze et al, 2009; Tsukahara et al, 2012; Marí-Ordóñez et al, 2013; Reinders et al, 2013; Rigal et al, 2016; Oberlin et al, 2017). However, with the exception of DNA methylation-defective mutants, the conditions required for the activation of this family, have not been fully explored. Here, we show that the ATCOPIA93 LTR, in the absence of negative epigenetic control, has the hallmarks of an immune-responsive gene promoter: i) responsiveness to unrelated PAMPs, ii) transient activation upon PAMP elicitation (like early PAMPs-induced genes), iii) suppression of transcriptional activation by bacterial effectors, iv) full dependence on biotic stress-response elements for activation (W-box cis-regulatory elements). While EVD transcripts had been so far only observed in discrete cell types in the absence of DNA methylation (Marí-Ordóñez et al, 2013), we show here that in the presence of the adequate signaling and transcription factors, it can be expressed in other tissues, such as adult leaves.
The connection between TE activation and stress response was particularly well-addressed in studies of the Tnt1 family of transposons in tobacco, which was found to be responsive to various biotic and abiotic stresses (Pouteau et al, 1991, 1994; Moreau-Mhiri et al, 1996; Mhiri et al, 1997; Grandbastien et al, 1997). Additionally, different Tnt families were induced by distinct biotic challenges and functional analyses further proved that the structural motifs present in the LTR sequences of Tnt1 families provided specific transcriptional reactivation to specific stresses (Beguiristain et al, 2001). Here, we show that one single element can be induced by PAMPs from bacterial and oomycetes pathogens; thus, in the future it will be important to determine whether the two functional W-boxes in EVD/ATR LTR are differentially involved in the induction by different pathogens, which would indicate the binding of different transcription factors to the same ATCOPIA93 LTR. Our findings in Arabidopsis, the primary model plant species for epigenetic analyses, should allow for the investigation of the poorly-understood phenomenon of permissiveness of transposon expression to certain stresses but not to others, and test whether this differential permissiveness could be epigenetically regulated, as was previously proposed for Tnt1 (Grandbastien et al, 2005).
The conditional induction of EVD is reminiscent of the heat-stress responsive element ONSEN (Ito et al, 2011; Cavrak et al, 2014) and expands the repertoire of Arabidopsis model TEs that can potentially highjack the transcriptional host machinery during stress responses. Thus, ONSEN is not a unique case, and many Arabidopsis TEs could exhibit this restricted reactivation pattern, where lack of DNA methylation will result in transposition if compounded by specific stress signaling. This would imply that the “mobilome"–the fraction of TEs with transposition activity– observed in ddm1 unstressed mutants (Tsukahara et al, 2009) is likely to be underestimated. Here, we focused on the somatic regulation of ATCOPIA93 in leaf tissues–where transpositions would not be mitotically inherited thus are difficult to detect–with the aim to test the impact of this regulation on defense gene regulation. However, future studies should address the exciting question of enhanced germinal transposition when wild type and ddm1 flowers will be subjected to PAMP elicitation or infected with pathogens.
One major difference between ATCOPIA93 and ONSEN is that ATCOPIA93 stress-induced expression in the wild type is more tightly controlled by epigenetic regulation for ATCOPIA93 than for ONSEN. Importantly, we revealed that ATCOPIA93-EVD, in addition to be subjected to DNA/H3K9 methylation, is targeted by an additional layer of epigenetic control through PcG silencing, which is generally associated with negative regulation of protein-coding genes and miRNA genes in vegetative tissues (Förderer et al, 2016). While DNA methylation and H3K27m3 marks have been described to be largely mutually exclusive, with DNA methylation generally preventing K27m3 deposition (Mathieu et al, 2005; Deleris et al, 2012; Weinhofer et al, 2010; Reddington et al, 2013; Saksouk et al, 2014), we found co-existence of these two epigenetic marks at EVD-LTR at the molecular level. This had already been observed in the endosperm at the pericentromeric GYPSY elements (Moreno-Romero et al, 2016) as well as in mammals where lower densities of CG methylation were found to allow H3K27m3 deposition (Statham et al, 2012; Brinkman et al, 2012). Accordingly, EVD LTR has only five methylated CGs (Fig 1, Marí-Ordóñez et al, 2013), which is low relative to the size of the LTR (roughly 400 bp). Thus, in Arabidopsis vegetative tissues, the co-existence of CG methylation and H3K27m3 can occur, likely constrained by CG density. As for the differential marking between EVD and ATR, it is presumably due to, or at least favored by, spreading from the neighboring genes marked by H3K27m3. This idea is supported by the almost complete loss of H3K27m3 at EVD-LTR in clf, a mutant for the PRC2 component CLF, which was recently shown to be involved in the spreading phase of H3K27m3 at the Flowering Locus C (FLC) (Yang et al, 2017). Future studies should identify and delineate TEs that can potentially recruit PRC2 in cis from the ones that become H3K27m3-marked due to an insertional effect, and the ones that exhibit both characteristics. Importantly, we showed that PcG silencing is functional at EVD ATCOPIA93, the modest effect of H3K27m3 loss on total ATCOPIA93 expression presumably due to the specific EVD control by PcG and the functional redundancy of H3K27m3 with DNA methylation, which co-exists at this site. The role of PcG in TE silencing is actually supported by the recent evidence of negative regulators of PcG, such as ALP1, which is encoded by a domesticated transposase (Liang et al, 2015) and which is likely evolved to protect TEs from Polycomb-mediated silencing. Interestingly, this negative control of EVD by PcG adds an additional epigenetic layer of restriction specifically at the functional ATCOPIA93 member (which is also less DNA methylated than its pericentromeric counterpart ATR), and presumably limits its somatic transposition while the corresponding soloLTR in the RPP4 promoter is activated for proper regulation during immune response. Notably, this double and differential mC/H3K27m3 marking could allow for unique members of one TE family to be differentially regulated, in particular at discrete stages of development since DNA methylation and PcG are not equivalent in their lability; this might provide the family members with different and discrete windows of opportunities to be expressed and transpose in some restricted tissues, and account for a not-yet appreciated strategy of TEs to adapt to their host.
Interestingly, the ATCOPIA93-LTR::GUS fusion was consistently found to be unmethylated in various transgenic lines in the wild-type Col-0 background, concordant with LTR::GUS reactivation in response to PtoΔ28E. This lack of de novo methylation upon transformation may be explained by weak LTR transcriptional activity in untreated plants thus preventing expression-dependent RNA-directed DNA Methylation (Fultz & Slotkin, 2017) and/or by low levels of CHH methylation/siRNAs at ATCOPIA93 (Fig 1A) (Mirouze et al, 2009; Marí-Ordóñez et al, 2013)), thus preventing identity-based silencing in trans (Fultz & Slotkin, 2017). Similarly, in wild type plants, the ATCOPIA93-soloLTRs appear usually to be unmethylated (Fig EV5A), in particular the soloLTR-5 described in detail here. This absence of methylation allowed us to test for a role of the ATCOPIA93 LTR as a “fully competent” transcriptional module in immunity, i.e., not masked by DNA methylation. This is a different role from the one previously described as an epigenetic module, interfering negatively with downstream expression, when the LTR was artificially methylated in trans by siRNAs produced by EVD after a burst of transposition in specific epiRIL lines (Marí-Ordóñez et al, 2013). The latter results may provide an explanation for the peculiar epigenetic control of EVD, which is almost exclusively controlled by CG methylation (Fig1A, Mirouze et al, 2009), although it belongs to an evolutionary young family of TEs: the preferred targets of POLYMERASE V, siRNAs and the RNA-directed DNA methylation (RdDM) pathway (Zhong et al, 2012). We propose that the low levels of EVD LTR siRNAs, which could methylate the soloLTR-5 in trans if present in larger quantities, could be the result of evolutionarily selection, so that soloLTR-5 remains unmethylated and proper immune response can be properly activated.
Transposable elements have been proposed to contribute not only to the diversification of disease resistance genes, which are among the fastest evolving genes, but also, following their diversification, to the evolution of their cis-regulation, as part of their maturation process (Lai & Eulgem, 2017). Compelling evidence exists for the latter role (Hayashi & Yoshida, 2009; Tsuchiya & Eulgem, 2013; Deng et al, 2017; Lai & Eulgem, 2017). In the present study, we have brought another demonstration for the cis-regulatory role of TEs, and for the first time we have linked the co-option of a soloLTR for proper expression of a functional disease resistance gene (RPP4) and the responsiveness of the corresponding full-length retroelement (COPIA93 EVD/ATR), through its LTR, during basal immunity. TEs have been long thought to be a motor of adaptive genetic changes in response to stress (McClintock, 1984). The link we established between responsiveness of a retroelement to biotic stress and its co-option for regulation of immunity provides experimental support to McClintock’s early model that TEs play a role in the genome response to environmental cues. Future studies should address the extent of the TE repertoire with such restricted expression to adapt to their host, in the light of recent findings showing that not only do disease resistance gene clusters contain many TEs, but also they are among the most frequent transposition targets observed (Quadrana et al, 2016).
Materials&Methods
Plant material and growth condition
Plants were grown at 22°C with an 8h light/16h dark photoperiod (short days) and experiments were generally performed on 4.5 to 5-week-old rosette leaves. Apart from Figure 3, where plants were analyzed in the absence of treatment, plants were infiltrated with a syringe with either water (“mock”), synthetic flg22 (Genescript) at 1µM concentration or a suspension of bacteria as described below, always at the same time in the morning (between 10 and 11.30am, depending on the number of plants to infiltrate). Plants were then covered with a clear plastic dome until tissue harvest to allow high humidity (Xin et al, 2016). For Figure 4B, 3.5-week-old seedlings grown on MS plates were transferred to liquid MS for at least 24h then infiltrated with either water or a suspension of bacteria and put back to light for 2 hours.
Mutant lines
We used the met1-3 allele (Saze et al, 2003), the ddm1-2 allele (Vongs et al, 1993) and the clf-29 allele (Bouveret, 2006). For Figures 1 and 2, first generation homozygous met1-3 and ddm1-2 mutants were genotyped and used for analysis; for Figure 4, second generation ddm1-2 homozygous mutants were used.
Generation of transgenic lines
- LTR::GUS transgenic lines
EVD-LTR was cloned into a pENTR/D-TOPO vector, then in a pBGWFS7 binary vector, upstream of the GUS sequence. LTR::GUS constructs were transformed in the Col0 accession by standard Agrotransformation protocol (Clough & Bent, 1998). Primary transformants were selected with Basta herbicide. Four lines were selected on the basis of 3:1 segregation of the transgene (single insertion) and brought to T3 generation (#2, #12, #4, #6) where the transgene was in a homozygous state and all four lines behave similarly as for GUS expression. Most experiments were performed on stable T3 lines (#12) homozygotes for the LTR::GUS transgene, and some in their progeny (T4) after checking that the absence of DNA methylation persisted. The mutations in W-boxes 1 and 2 were introduced in pENTR vector by overlapping PCR and the mutated LTRs cloned in pBGWFS7. Experiments were performed on individual primary transformants for WT and mutated constructs. Transgenic plants were sequenced to verify the presence of the mutations at the LTR transgene.
- pRPP4::RPP4 transgenic lines
For the “WT” construct, a 3kb sequence upstream of RPP4 predicted TSS was cloned in pENTRD-TOPO; for the “ΔLTR” construct, the same 3kb sequence minus the soloLTR-5 was synthesized and cloned in the same vector; for the “w1” construct, site-directed mutagenesis was used on the “WT” pENTRD-TOPO. A fragment corresponding to the RPP4 gDNA (with introns and UTRs) was then amplified from wild-type plants and cloned after the RPP4 promoter in the three different “WT”, “ΔLTR” and “w1” pENTRD-TOPO vectors using restriction enzymes. The resulting vectors were recombined with pH7WG. Primary transformants were selected on hygromycin and analyzed individually.
Bacterial strains and preparation of inocula
The bacterial strains used are Pseudomonas syringae pv tomato Pto DC3000 (“Pto”) and a non-pathogenic derivative of Pseudomonas syringae pv tomato Pto DC3000 in which 28 out of 36 effectors are deleted (Cunnac et al, 2011), referred here as “PtoΔ28E”. Bacteria were first grown on standard NYGA solid medium at 28°C with appropriate selection, then overnight on standard NYGB liquid medium. Bacteria were pelleted and washed with water twice. Suspensions of 2.108cfu/ml were used for PtoΔ28E except for Figure 1C where suspensions of 1.107cfu/ml were used to compare to a same inoculum of Pto DC3000.
Histochemical GUS staining
GUS staining was performed as in (Yu et al, 2013). Briefly, leaves were placed in microplates containing a GUS staining buffer, vacuum infiltrated three times during 15 min, and incubated overnight at 37 °C. Leaves were subsequently washed several times in 70% ethanol.
SDS-PAGE and Western blotting
Leaf total protein extracts were obtained by using the Tanaka method, quantified by standard BCA assay and 100 µg were resolved on SDS/PAGE. After electroblotting the proteins on a Polyvinylidene difluoride (PVDF) membrane, GUS protein analysis was performed using an antibody against the GFP since pBGWFS7 contains a GUS-GFP fusion and the anti-GFP antibody (Clontech #632380) was more specific, and stained with a standard coomassie solution to control for equal loading.
DNA extraction and bisulfite conversion
DNA extraction and bisulfite conversion was performed as in Yu et al, 2013 except that the DNA was not sonicated before bisulfite conversion and 17 to 22 clones were analyzed per experiment.
RNA extraction and qRT-PCR analyses
Total RNA was extracted using RNeasy Plant Mini kit (Qiagen or Macherey-Nagel). One µg of DNA-free RNA was reverse transcribed using qScript cDNA Supermix (Quanta Biosciences) and either oligo(dT) and random hexamers mix or a transcript-specific primer for GUS mRNA analysis. cDNA was then amplified in RT q-PCR reactions using Takyon SYBR Green Supermix (Eurogentec) and transcript-specific primers on a Roche Light Cycler 480 thermocycler. For each biological replicate, two or three technical replicates were averaged when the qPCR corresponding values were within 0.5 cycles. Expression was normalized to UBIQUITIN (At2g36060) expression. In addition, for Figure 4A, two reverse-transcription reactions were performed for each biological replicate – in particular, in order to obtain enough cDNA for pyrosequencing- and qPCRs technical replicates averaged. The PCR parameters are: 1 cycle of 10 minutes at 95°C, 45 cycles of 10 s at 95°C, 40 s at 60°C.
Chromatin immunoprecipitation and ChIP-qPCR analyses
ChIPs were performed as in (Bernatavichute et al, 2008), starting with 0.3 g to 1 g (per ChIP) of adult leaves that were previously crosslinked by vacuum-infiltration of a 1% formaldehyde solution. Antibodies against H3K27m3 and H3K9m2 are from MILLIPORE (07-449) and ABCAM (ab1220) respectively. 2 µl of a 1:10 dilution of the IP was used for qPCR. The PCR parameters are: 1 cycle of 10 minutes at 95°C, 45 cycles of 10 s at 95°C, 40 s at 60°C.
Methylation-sensitive enzyme assay (“Chop-assay”)
Two hundred ng of gDNA (Fig EV1) or 10 to 20 ng of ChIP-DNA (10 ng for Input DNA) (Fig 3) was digested overnight at 37C with 1 µL or 0,5 µl respectively of Sau96I enzyme (Thermoscientist FD0194). As Sau96I cannot be heat-inactivated, DNA was then purified with a clean-up column (Macherey Nagel nucleospin column) (Fig 1A) or, when the amount of material was limited (Fig 3) by standard phenol-chloroform extraction using glycogen to precipitate the DNA. DNA was eluted in 20 µl of water or pellets were resuspended in 20 µl of water; qPCR were performed using 0,3ul and primers that amplify an amplicon spanning the Sau96I site. The same amount of the corresponding non-digested DNA was used for qPCR as a control and to normalize the data.
Pyrosequencing
ATCOPIA93 DNA (ChIP-DNA, cDNA or gDNA as a control) was amplified with a biotinylated (forward) primer in the same region where RNA levels were analyzed and containing a SNP between EVD and ATR; the biotinylated PCR product (40 µl reaction) was pulled down with streptavidin beads (sigma GE17-5113-01) and the sense biotinylated strand sequenced with a Pyromark Q24 (Qiagen) on the sequencing mode. Input DNA was used as a control for equal contribution of each SNP. Analysis and quality check of the peaks were done with the Pyromark Q24 companion software which delivers pyrograms indicating the % of each nucleotide at the interrogated SNP. These percentages were directly plotted for each biological replicate in the Extended Views and averaged for clarity of presentation in the main figures.
Authors contributions
A.D., J.Z., A.Y., J.W. performed the experiments, A.D. and J.Z. analyzed the data; J.D. performed bioinformatic searches; A.D, J.Z. and L.N. designed the experiments; A.D. wrote the manuscript.
Acknowledgments
We thank the members of the Navarro Lab for their input and discussions as well as the Bourc’his and Felix Labs for their help with pyrosequencing, D. Bouyer and L. Quadrana for discussions, H. Keller for valuable comments, M. Greenberg and M. Boccara for critical reading of the manuscript and discussions.
Funding source: This work was funded by an ANR-retour post doc (ANR-11-PDOC-0007-01) granted to A.D. and a Human Frontier Scientific Program Career Development Award (HFSPCDA-00018/2014) granted to A.D.
All authors have seen and approved the manuscript.