Differential requirement of MED14 and UVH6 for heterochromatin transcription upon destabilization of silencing

Constitutive heterochromatin is commonly associated with high levels of repressive epigenetic marks and is stably maintained transcriptionally silent by the concerted action of different, yet convergent, silencing pathways. Reactivation of heterochromatin transcription is generally associated with alterations in levels of these epigenetic marks. However, in mutants for particular epigenetic regulators, or upon particular environmental changes such as heat stress, heterochromatin-associated silencing is destabilized without noticeable changes in epigenetic marks. This suggests that transcription can occur in a non-permissive chromatin context, yet the factors involved remain poorly known. Here, we show that heat stress-induced transcription of heterochromatin depends on the TFIIH component UVH6 and the Mediator subunit MED14. Mutants for these two factors exhibit hypersensitivity to heat stress, and under these conditions, UVH6 and MED14 are required for transcription of a high number of loci. We further show that MED14, but not UVH6, is required for transcription when heterochromatin silencing is destabilized in the absence of stress. In this case, MED14 requires proper chromatin patterns of repressive epigenetic marks for its function. We also uncover that MED14 regulates non-CG DNA methylation at a subset of RNA-directed DNA methylation target loci. These findings provide insight into the control of heterochromatin transcription upon silencing destabilization and identify MED14 as a regulator of DNA methylation.


Introduction
identified candidate mutations in zen1 and zen2 using mapping-by-sequencing from outcross F2 populations (supplementary figure 3B, C). zen1 plants contained a G to A transition in the MED14 (AT3G04740) gene, changing tryptophan for a stop codon at amino acid position 1090 (figure 2G). We identified a C to T mutation in the UVH6 (AT1G03190) gene in zen2 plants, causing a proline to leucine substitution at amino acid 320 (figure 2G). Complementation of zen1 and zen2 phenotypes with transgenes encoding WT versions of MED14 and UVH6 confirmed that MED14 and UVH6 mutations were responsible for the phenotypes observed in zen1 and zen2, respectively (figure 2F). Hence, zen1 and zen2 were renamed med14-3 and uvh6-3, respectively.
MED14 is the central subunit of the MEDIATOR complex, a large protein complex required for early steps of transcription initiation (Cevher et al., 2014;Soutourina, 2018). In Arabidopsis, MED14 function has been involved in cell proliferation and expression regulation of some cold-regulated or biotic stress-induced genes (Autran et al., 2002;Gonzalez et al., 2007;Hemsley et al., 2014;Wang et al., 2016;Zhang et al., 2013). UVH6 is the Arabidopsis ortholog of the human XPD and yeast RAD3 proteins (Liu et al., 2003), which are part of the transcription factor IIH (TFIIH) complex involved in transcription initiation and nucleotide excision repair (Compe and Egly, 2012). XPD is an ATP-dependent 5'->3' helicase and all amino acids required for XPD functions in yeast and human show remarkable conservation in UVH6 (Kunz et al., 2005).
Interestingly, all the mutations identified in UVH6 disrupt conserved residues (supplementary figure 4). In Arabidopsis, the UVH6 function was first described as necessary for tolerance to UV damage and heat stress (Jenkins et al., 1995(Jenkins et al., , 1997. Failure to isolate homozygous mutants for uvh6-2, a transfer-DNA (T-DNA) insertion line, suggested UVH6 to be an essential gene (Liu et al., 2003). Supporting this conclusion, we also failed to obtain homozygous plants for another uvh6 T-DNA insertion line (uvh6-5) (figure 2G).

Transcriptomic analysis of uhv6 and med14 mutants in the absence of stress
To investigate the impact of med14-3 and uvh6-3 mutations on transcription genome-wide, we determined mRNA profiles of mutant leaves following incubation at either 23°C (med14-3_23, uvh6-3_23) or 37°C (med14-3_37, uvh6-3_37) by mRNA-seq. In this analysis, we also profiled the transcriptome at 23°C of another mutant allele of UVH6 (uvh6-4), which we isolated later while pursuing screening our L5 mutant population (supplementary figure 5A, B). The uvh6-4 mutation replaces a proline for a leucine at amino acid position 532 (figure 2G). Unlike uvh6-3, uvh6-4 mutants showed yellow-green leaves and reduced stature, a phenotype similar to the one previously described for the uvh6-1 mutant (supplementary figure 5C) (Jenkins et al., 1997). Suppression of heat stress-induced release of silencing was stronger in uvh6-4 than in uvh6-3 (supplementary figure 5B), and survival assays showed that uvh6-4 and uvh6-1 plants were more sensitive to heat stress than uvh6-3 plants ( figure 2E). This indicates that uvh6-4 is a stronger mutant allele of UVH6 than uvh6-3.
We first compared the mutant transcriptomes with that of the WT in the absence of heat stress. By applying stringent thresholds (fold change ≥ 4, false discovery rate < 0.01), we identified 628 differentially expressed genes (DEGs) in med14-3_23 (figure 3A), predominantly PCGs (597). As expected for a mutation of a protein required for transcription, the majority of med14-3 DEGs (385), including 23 TEs, showed decreased transcript accumulation. Only 7 DEGs were detected in uvh6-3_23, while 218 loci show differential transcript accumulation in uvh6-4_23, in agreement with uvh6-4 being a stronger mutant allele of UVH6. Unexpectedly, out of the 218 uvh6-4_23 DEGs, 156 were upregulated, suggesting that UVH6 mainly represses transcription at a subset of genomic loci at 23°C (figure 3A).  also show reduced transcript accumulation in uvh6-3 (supplementary figure 6A). The med14 and uvh6 mutations affect transcript accumulation at largely independent sets of loci (figure 3B).
Gene ontology analysis indicated that genes upregulated in med14-3_23 were enriched for biotic stress response genes (supplementary table 1). A similar enrichment was observed in uvh6-4_23 upregulated genes and in genes commonly upregulated in med14-3_23 and uvh6-4_23, indicating that MED14 and UVH6 repress genes involved in pathogen response. PCGs downregulated in med14-3_23 were enriched for genes associated with "positive regulation of transcription from RNA polymerase II promoter in response to heat stress". These included HsfB2A, HsfA4A, HsfA6b and HsfA3. HsfA6b, HsfA3 and another med14-3_23 downregulated gene, DREB2A, are partially required for thermotolerance (Huang et al., 2016;Sakuma et al., 2006;Schramm et al., 2007), suggesting that downregulation of these genes might be responsible for med14-3 hypersensitivity to heat stress (figure 2E). PCGs downregulated in uvh6-4_23 were enriched for genes associated with "response to UV" as well as genes involved in processes such as "anthocyanin biosynthesis", "regulation of flavonoids", "phenylpropanoid metabolism", which protect plants against UV radiation (Jansen et al., 1998). Therefore, downregulation of these genes likely plays a role in uvh6 mutant UV hypersensitivity (supplementary figure 6B) (Jenkins et al., 1995).

Genome-wide suppression of heat-stress-induced transcriptional activation in uvh6 and med14
To assess the impact of med14 and uvh6 on transcript levels following heat stress, we compared med14-3_37 and uvh6-3_37 with WT-37 mRNA-seq datasets. Overall, heat stress-induced transcriptional activation of pericentromeric sequences was diminished in med14 and uvh6 mutant backgrounds, and transcripts from loci located on chromosome arms tended to accumulate at a lower level than in stressed WT plants ( figure   4A). Compared with med14, the impact of the uvh6 mutation on stress-induced transcriptional changes appeared more global ( figure 4A, supplementary figure 7). Accordingly, the number of DEGs was higher in uvh6-3_37 than in med14-3_37. We defined 1631 DEGs in med14-3_37, with the vast majority (1239) showing downregulation (figure 4B). Downregulated loci included 1124 PCGs and 115 TEs. While we detected only 7 DEGs in uvh6-3_23 (figure 3A), more than 6200 loci were differentially expressed in uvh6-3_37, with 80% of these (4949) being downregulated. A total of 4711 PCGs and 238 TEs displayed less transcript accumulation in uvh6-3_37 relative to WT-37 (figure 4B). The higher number of DEGs at 37°C relative to 23°C in the mutants indicates that MED14 and UVH6 functions are required for efficient transcription of a higher number of loci under heat stress.
PCGs upregulated by heat stress showed overall reduced transcript levels in uvh6-3 and med14-3, while transcript accumulation of PCGs downregulated by heat stress showed limited changes in med14-3 compared with uvh6-3 (supplementary figure 8A, B), suggesting again a more global impact of the uvh6 mutation on stress-induced transcriptional changes.
In the absence of stress, the med14 and uvh6 mutations affect transcript accumulation at rather few, largely independent set of loci ( figure 3B). Under heat stress, many loci downregulated in med14-3 were similarly affected in uvh6-3 (figure 4C). Even though this could be expected given the large number of genes downregulated in uvh6-3, we also observed that loci upregulated in one mutant also showed a similar tendency in the other ( figure 4C, supplementary figure 8C). This is remarkable as in both mutants, upregulation events are rare relative to downregulation events. These data suggest that MED14 and UVH6 have converging functions at many overlapping loci under heat-stress conditions. TEs transcriptionally upregulated by heat stress showed overall reduced transcriptional activation in the mutant backgrounds (figure 4D). Heat stress predominantly destabilized silencing at TEs of the DNA/En-Spm, DNA/MuDR, LTR/Copia and LTR/Gypsy superfamilies (supplementary figure 8D). Among these stressinduced TEs, TEs downregulated in uvh6-3_37 showed comparable proportions. Noticeably, TEs downregulated in med14-3_37 were enriched in LTR/Copia and LTR/Gypsy elements, suggesting that MED14 is preferentially required for heat-induced release of silencing at LTR retrotransposons.
We generated med14-3 uvh6-3 double mutants and assessed transcript accumulation from L5-GUS and selected TEs using RT-qPCR (supplementary figure 9). We found no synergy between the two mutations; at a given locus, the transcript levels in med14-3 uvh6-3 were similar to the ones detected in the mutant showing the strongest downregulation. These results suggest that, at least at these TEs, MED14 and UVH6 function in the same molecular pathway to promote transcription.
Together, our results indicate that MED14 and UVH6 are required for proper heat stress-induced transcriptional activation of heterochromatic TEs, and more generally play an important role in controlling transcription at a high number of genomic loci under stress conditions.

Transcription of methylated TEs requires MED14 but not UVH6
Given that UVH6 and MED14 are involved in transcriptional activation induced by heat-stress, we questioned whether their functions are also required for heterochromatin transcription occurring in mutants for epigenetic regulators. To address this question, we introduced uvh6-4 and med14-3 in the mom1-2 and ddm1-2 mutant backgrounds, which display constitutive release of transcriptional silencing at heterochromatic loci, and performed mRNA-seq. In ddm1, loss of silencing is associated with a strong reduction in DNA, H3K9me2 and H3K27me1 methylation levels (Ikeda et al., 2017;Vongs et al., 1993;Zemach et al., 2013), whereas silencing defects in mom1 mutants occur without major changes in these epigenetic marks (Amedeo et al., 2000;Habu et al., 2006;Han et al., 2016;Moissiard et al., 2014;Vaillant et al., 2006).
Because DNA and H3K9me2/K27me1 methylation levels are largely reduced in ddm1-2, while being mostly unaltered in mom1-2 (supplementary figure 10C) (Amedeo et al., 2000;Habu et al., 2006;Han et al., 2016;Moissiard et al., 2014;Vaillant et al., 2006), our data suggest that, upon silencing destabilization, MED14 is involved in transcription at a subset of heterochromatic TEs and requires DDM1-mediated epigenetic marks for its function. Supporting a role for DNA methylation in MED14 function, RT-qPCR assays showed that silencing release of MULE and TSI in the DNA hypomethylated met1-3 background was not suppressed by the med14-3 mutation (figure 5C). Remarkably, when considering TEs upregulated by heat stress, TEs depending on MED14 for transcriptional upregulation showed higher DNA methylation levels at all cytosine contexts compared to those independent of the med14-3 mutation (figure 5D, supplementary figure 10D).
Such strong bias for highly methylated elements was not observed at TEs that depended on UVH6 for heat stress-induced transcriptional upregulation (supplementary figure 10E). Furthermore, TEs transcribed in the WT in the absence of stress and downregulated by med14-3 were more methylated than those unaffected by the med14-3 mutation (figure 5E).
Therefore, we conclude that MED14 promotes transcript accumulation at a set of highly methylated TEs and requires proper DNA methylation patterns for this function. On the other hand, UVH6 is required for transcription in a heat stress-specific manner and appears to show a less pronounced preference than MED14 for highly methylated TEs.

MED14 regulates non-CG DNA methylation
We sought to determine whether med14 mutation affect DNA methylation by profiling genome-wide DNA methylation levels in WT and med14-3 seedlings by BS-seq. Overall, DNA methylation levels were mostly unaltered at CG sites, and showed a moderate reduction at non-CG sites in med14 compared with the WT (figure 6A). Calculating average methylation levels along all genomic PCGs, euchromatic TEs and pericentromeric TEs revealed that non-CG methylation was specifically decreased at pericentromeric TEs in med14-3 (supplementary figure 11A, B). Because low variations on average methylation levels could mask strong changes at a limited number of loci, we divided the genome in 100-bp bins and determined differentially methylation regions (DMRs) in med14-3 relative to the WT. This analysis confirmed that the med14-3 mutation predominantly induced a decrease in DNA methylation at non-CG sites, and preferentially alters methylation of pericentromeric regions of the chromosomes (figure 6B, C). CHG and CHH hypomethylation occurred concurrently (supplementary figure 11C) indicating that MED14 regulates non-CG methylation at these loci.
The Mediator complex is involved in initiation of Pol II transcription and Pol II has been reported to be involved in a pathway that regulates DNA methylation (Stroud et al., 2013). Furthermore, at several heterochromatic loci, Mediator promotes Pol II-mediated production of long noncoding scaffold RNAs, which serve to recruit Pol V to these loci (Kim et al., 2011). To assess whether MED14 and Pol II regulate DNA methylation at the same loci, we determined DNA methylation levels of med14 hypomethylated DMRs in the nrpb2-3 Pol II mutant allele using previously published data (Zhai et al., 2015). For the vast majority of these genomic regions, DNA methylation levels were unaltered in nrpb2-3 (supplementary figure 12A), indicating that MED14 regulates DNA methylation largely independently of Pol II.
In the Arabidopsis genome, CHG methylation is mostly mediated by the H3K9me2-directed CMT3 chromomethylase, while CHH methylation is maintained by CMT2 and the RdDM pathway at largely distinct genomic regions (Stroud et al., 2014;Zemach et al., 2013). RdDM requires the production of noncoding RNAs by Pol IV and Pol V, which are eventually required to target and recruit the RdDM effector complex containing the DRM2 de novo methyltransferase to its genomic targets (Matzke and Mosher, 2014). We used published data (Stroud et al., 2013) to determine non-CG methylation levels at med14 non-CG hypomethylated DMRs in mutants for CMT3, CMT2, Pol IV (NRPD1), Pol V (NRPE1) and DRM1/2. med14 CHG hypomethylated DMRs showed nearly WT methylation levels in cmt2, whereas they were largely hypomethylated in cmt3 (supplementary figure 12B), in agreement with the prominent role of CMT3 over CMT2 in controlling CHG methylation (Stroud et al., 2014). Interestingly, many med14 CHG hypo DMRs showed reduced DNA methylation level in the nrpd1, nrpe1 and drm1/2 RdDM mutants (supplementary figure 12B). Strikingly, med14 CHH hypomethylated DMRs showed strongly reduced DNA methylation level in these RdDM mutants ( figure 6D). This was not merely due to a genome-wide impact of RdDM deficiency on CHH methylation since the same number of randomly selected genomic regions showed much less reduction in CHH methylation in the RdDM mutants ( figure 6E). Conversely, loci with reduced CHH methylation in drm1/2, nrpd1 or nrpe1 all showed lower CHH methylation in med14-3 (supplementary figure   12C). Together, these results indicate that MED14 regulates non-CG methylation at a subset of loci, likely through RdDM.

Discussion
Previous studies have shown that heat stress or mutations in certain silencing factors can trigger heterochromatin transcription without modifying levels of repressive epigenetic marks (Amedeo et al., 2000;Lang-Mladek et al., 2010;Moissiard et al., 2012;Pecinka et al., 2010;Tittel-Elmer et al., 2010). That transcription could occur in an otherwise repressive environment suggested that specific mechanisms were involved (Tittel-Elmer et al., 2010). Here, we identified MED14 and UVH6 as critical factors for heterochromatin transcription during heat stress. We showed that UVH6 is dispensable for heterochromatin transcription in silencing mutants such as mom1 and ddm1, while MED14 is solely required when heterochromatic marks are not altered. Additionally, we showed that MED14 participates in maintenance of DNA methylation at a subset of RdDM-dependent loci.
XPD, the human UVH6 ortholog, is the central subunit of the TFIIH complex, which is crucial for nucleotide exchange repair and is considered a global transcription factor (Compe and Egly, 2016). Our data show that uvh6 mutations impair transcription of many genes and TEs specifically at elevated temperature. This suggests that UVH6 is not generally required for transcription initiation in Arabidopsis, but is rather involved in a stress-specific transcription mechanism. Previous studies showed that UVH6 belongs to the most essential factors regarding thermotolerance (Jenkins et al., 1997;Larkindale et al., 2005), although the molecular pathway involved is not known. Interestingly, heat-induced accumulation of the canonical heat-responsive factors HSFs and HSPs is independent of UVH6 (Hu et al., 2015;Larkindale et al., 2005), reinforcing the notion that UVH6 is not required for transcription of all genes during heat stress. Human TFIIH has been shown to be involved in selective transcriptional responses to various stimuli through posttranslational modifications or recruitment of transcription factors (Chen et al., 2000;Chymkowitch et al., 2011;Compe et al., 2007;Keriel et al., 2002;Sano et al., 2007;Traboulsi et al., 2014). Therefore, UVH6 may cooperate with HSFs or other transcription factors during heat stress. In human, XPD is involved in many functions on top of its well-established roles in transcription and repair, sometimes in other complex than TFIIH (Compe and Egly, 2016). To get a better understanding of UVH6-dependent transcription in heat stress, futures efforts should try to determine if UVH6 acts as a component of the TFIIH complex or separately.
Mediator is a large protein complex organized in a head, middle and tail modules, with a transiently associated CDK8 kinase module (Soutourina, 2018). MED14 connects the three main modules and is critical for Mediator architecture and its function as a co-activator of Pol II transcription (Cevher et al., 2014). We found that MED14 preferentially stimulates transcription of highly methylated TEs in control and stressed conditions. TEs derepressed in mom1 mutants require MED14 for transcription, and importantly, the same Although MED14 makes multiple contacts with the different Mediator modules, the C-terminal part of yeast and human MED14 has been mapped to the tail module (Nozawa et al., 2017;Tsai et al., 2014).
Accordingly, C-terminal truncations of MED14 led to dissociation of the tail module in yeast (Li et al., 1995;Liu and Myers, 2012). The med14-3 mutation isolated in our study induces a stop codon at amino acid 1090 of MED14, truncating 614 amino acids at the C-terminal end. The Mediator subunits are relatively well conserved between yeast, human and Arabidopsis (Bäckström et al., 2007). By analogy, the med14-3 mutation reported here may be expected to lead to tail module dissociation. Interestingly, the tail module seems important for recruiting the Mediator complex to chromatin (Jeronimo and Robert, 2017;Soutourina, 2018). Thus, the Mediator tail module may mediate the preference of MED14 for DNA methylated loci.
In fission yeast, mutations of some subunits from the Mediator head and middle modules induce defects in heterochromatin silencing at pericentromeres and concomitant loss of the heterochromatic mark H3K9me2

Plant material
The ddm1 Vaucheret (Morel et al., 2000). Plants were grown in soil or in vitro in a growth cabinet at 23°C, 50% humidity, using long day conditions (16h light, 8h dark). For in vitro conditions, seeds were surface sterilized with calcium hypochlorite and sowed on solid Murashige and Skoog medium containing 1% sucrose (w/v).
For all other molecular data presented in this study, we used lines backcrossed six times for med14-3 and GUS assay Following heat or control treatment, rosette leaves were transferred to 3ml of a staining solution composed of 400 µg/ml 5-bromo-4-chloro-3-indolyl-β-D-glucuronic acid, 10 mM EDTA, 50 mM sodium phosphate buffer pH 7.2, 0.2% triton X-100. Leaves were placed in a desiccator, subjected to void for 5 minutes two times, and subsequently incubated 20h to 24h at 37°C. Chlorophyll was then repeatedly dissolved in ethanol to allow proper staining visualization.

Mutagenesis, screening and mapping
We used EMS-mutagenized seeds from a previously described study (Ikeda et al., 2017). To screen for mutants deficient in heat stress-induced release of silencing of the L5-GUS transgene, one leaf per M2 plant was dissected, and leaves from four plants were heat-stressed together with a 24h incubation at 37°C in dH2O. Leaves were subsequently subjected to GUS staining as described above. To isolate mutant candidates, a second round of screening was applied to each individual of M2 pools that contained leaves with reduced GUS signal relative to the non-mutagenized progenitor L5 line. Center (Seki et al., 1998(Seki et al., , 2002 and its stop codon was removed by PCR. The promoter and cDNA were cloned into a pBluescript SK plasmid supplemented with attP sites by BP recombination, and subsequently introduced into pB7FWG2 by LR recombination. For the p35S::MED14 construct, the MED14 cDNA was introduced by LR recombination into a pBINHygTX plasmid supplemented with attR sequences. For p35S::UVH6-GFP construct, the UVH6 cDNA without stop codon was amplified from Col-0 RNA and introduced by BP recombination into the pDONR/ZEO vector (Invitrogen). The fragment was introduced into pH7FWG2 by LR recombination. The med14-3 and uvh6-3 mutants were complemented by Agrobacterium-

Heat stress and UV-C irradiation
Rosette leaves were cut with forceps and transferred to 6-well tissue culture plates containing 3ml dH2O.
They were subsequently incubated for 24h in a 23°C or 37°C growth cabinet with otherwise standard conditions. For molecular analysis, nine to twelve rosette leaves from three to four seedlings were pooled for heat or control treatment. Rosette leaves were then dried on absorbent paper, flash-frozen in liquid nitrogen and stored at -80°C or directly processed.
For survival assays, seeds were sowed in vitro, stratified for 72h in the dark at 4°C and grown 7 days in standard conditions before heat or UV treatment. Heat stress was applied for 24h or 48h. UV-irradiation was performed in an Et-OH sterilized UV chamber (GS Gene Linker, Bio-Rad) equipped with 254 nm bulbs. Plate lids were removed before irradiation at 10 000 J / m2 and placed back immediately. Irradiated seedlings were transferred to a dark growth cabinet with standard conditions for 24h to block photoreactivation before recovering in light for five days.

RNA analysis
Total RNA was extracted in TRIzol reagent, precipitated with isopropanol and washed two times in ethanol 70%. Integrity was assessed by running 1ug of RNA through an agarose gel after RNA denaturation in 1X MOPS 4% formaldehyde for 15 minutes at 65°C. 2ug of RNA were then DNase treated using 2 unit of RQ1 DNAse (Promega) in 15ul, following manufacturer's instructions. DNase-treated RNAs were further diluted to 40ul in RNase-free H2O before subsequent analysis. 50ng of RNA was used as input for reverse transcription PCR (RT-PCR). End-point RT-PCR was performed with the one-step RT-PCR kit (Qiagen) following manufacturer's instructions in a final volume of 10ul. For 18S rRNA, MULE, 106B, TSI and 180bp, we respectively performed 20, 26, 35, 28 and 37 cycles. RT-qPCR were performed in a final volume of 10ul with the SensiFAST TM SYBR® No-ROX One-Step Kit (Bioline) in an Eco Real-time PCR system (Illumina). Quantification cycle (Cq or Ct) values were analyzed following the 2 -ΔΔCT method (Livak and Schmittgen, 2001). The mean of biological replicates from the control condition was subtracted to each ΔCq value to calculate ΔΔCq. Means and standard errors from biological replicates were calculated from 2 -ΔΔCq values.

mRNA-sequencing
Total RNA was extracted and treated as indicated above except that following DNase treatment, RNAs were further purified in phenol-chloroform. Sequencing libraries were generated and sequenced as 50bp singleend reads at Fasteris S.A. (Geneva, Switzerland). Read mapping and quantification of gene expression were performed as previously reported (Ikeda et al., 2017). To allow comparisons between uvh6-3, med14-3 and uvh6-4 ( figure 3 and supplementary figure 6), the uvh6-4 sample and its corresponding WT were artificially converted to non-stranded libraries by merging sense and antisense reads and re-calculating RPKM values at each locus. For comparisons of WT at 37°C versus (vs) WT at 23°C, ddm1-2 vs WT and mom1-2 vs WT, differentially expressed loci (PCGs and TEs) were defined by a log2 fold change >= 1 or <= -1, a false discovery rate (FDR) < 0.01 and only loci defined as differentially expressed in both replicates were retained.
When reads could be assigned to a specific strand (ddm1-2 and mom1-2 libraries), differential expression was tested in both orientations for each annotation, and only loci that were differentially expressed on the same orientation in both replicates were retained. For all other comparisons, since a single replicate was analyzed, the log2 fold change threshold was increased to >= 2 or <= -2. Gene ontology analysis was performed using Panther Overrepresentation Test (05/12/2017 release) using the 27/12/2017 Gene Ontology database (Ashburner et al., 2000).
To analyze TE transcription in WT and med14-3 in standard conditions (23°C) (figure 5E), we aligned reads from WT and med14-3 with STAR (Dobin et al., 2013) and retained multi-mapped reads randomly assigned.
We counted reads on TAIR10 transposon annotations and selected TEs with a minimum RPKM value of one in WT, a minimum length of 200 bp and that had at most 10% of their length intersecting a protein coding gene annotation, regardless of their orientation.

Whole-genome bisulfite sequencing
After 24h incubation at 23°C in dH2O of 16-day-old rosette leaves from L5 and med14-3, genomic DNA was extracted using the Wizard® Genomic DNA Purification Kit (Promega) following manufacturer's instructions.
C. Transcripts from TSI and MULE loci were analyzed by RT-qPCR in rosette leaves from indicated genotypes at control temperature (23°C). Data were normalized to the reference gene AT5G12240 and further normalized to the mean of L5 samples at 23°C. Error bars illustrate standard errors of the mean across three biological replicates. Statistically significant differences between means of mom1, ddm1, met1 and combinations of these mutations with med14-3 were tested by unpaired bilateral Student's ttest.
D. DNA methylation levels at CG, CHG and CHH contexts of TEs upregulated in heat-stressed WT samples, distinguishing TEs downregulated in med14-3 from TEs not downregulated in med14-3, were calculated in WT samples subjected to a control stress at 23°C. Statistical differences between data sets were tested by unpaired two-sided Mann-Whitney test.
E. RPKM values at TEs were calculated using multi-and uniquely-mapped reads in WT and med14-3 in control conditions (23°C) (see methods). TEs above one RPKM in WT were grouped according to their log2 fold change in med14-3 and DNA methylation levels at CG, CHG and CHH contexts in WT at 23°C were calculated for each group. Statistical differences between data sets were tested by unpaired twosided Mann-Whitney test. Transcript accumulation in rosette leaves from Col-0 wild type (WT), arp6-1 and atmorc6-3 mutants was quantified by RT-qPCR at five loci overexpressed in heat-stress. Samples had been subjected to a control stress at 23°C or a heat stress treatment at 37°C. Data were normalized to the geometric mean of the reference genes AT5G12240 and ACT2. Values were further normalized to the mean of L5 samples at 23°C. Error bars indicate standard errors of the mean across three biological replicates. For each locus, differences in mean at 37°C between the WT and the mutants was tested by unpaired twotailed Student's t-test, but did not reveal any significant difference (P > 0.05). C. Representative pictures of 16-day-old seedlings of the indicated genotypes grown in soil and in long day conditions. Scale bar: 1cm.

Supplementary figure 6
A. Reads per million per kilobase (RPKM) values in the WT and uvh6-3 mutants of loci significantly upregulated and downregulated in uvh6-4 at 23°C. Statistical differences between distributions in WT and uvh6-3 were tested by unpaired two-sided Mann-Whitney test.
B. UV survival assays. Seven-day-old seedlings of the indicated genotypes were UV irradiated at 10kJ / m2, returned to standard conditions with 24h of dark followed by five days recovery in light.

Supplementary figure 7
Transcriptional changes in WT plants subjected to heat stress (top), in med14-3_37 (middle) and uvh6-3_37 (bottom) relative to heat-stressed WT, represented along chromosomes one to four by log2 ratios of mean RPKM values in 100kb windows. (left) and uvh6-3_37 (right) showed for both med14-3_37 and uvh6-3_37 relative to heat-stressed WT plants.
D. Relative frequency of TE superfamilies in the Arabidopsis genome (TAIR10, white) and the following datasets: TEs upregulated in WT plants subjected to heat stress (red), among these, TEs downregulated in med14-3_37 (dark blue) or in uvh6-3_37 (green) relative to a WT at 37°C.

Supplementary figure 9
Transcripts from six loci were analyzed by RT-qPCR in L5 control plants, med14-3, uvh6-3 mutants and med14-3 uvh6-3 double mutants at 23°C and 37°C. Data were normalized to the reference gene AT5G12240 and further normalized to the mean of L5 samples at 23°C. Error bars illustrate standard errors of the mean across three biological replicates. For each temperature treatment, statistical differences between means of mutant conditions were tested by ANOVA followed by post-hoc analysis using Tukey's Honest Significant Difference test (*: p-value < 0.05, **: p-value < 0.005, ***: p-value < 0.0001). See methods. Data for the L5-GUS transgene was already displayed in figure 2B.
C. TEs commonly upregulated in ddm1 and mom1 were aligned at their 5'-end or 3'-end and average cytosine methylation levels in the indicated nucleotide contexts were calculated from 3 kb upstream to 3 kb downstream in WT, ddm1 and mom1. Upstream and downstream regions were divided in 100bp bins, while annotations were divided in 40 bins of equal length. D. DNA methylation levels at CG, CHG and CHH contexts were calculated for TEs upregulated in heatstressed WT samples, distinguishing TEs downregulated in med14-3 from TEs not downregulated in med14-3. DNA methylation levels were calculated in WT at 23°C and 37°C.
E. DNA methylation levels at CG, CHG and CHH contexts were calculated for TEs upregulated in heatstressed WT samples, distinguishing TEs downregulated in uvh6-3 from TEs not downregulated in uvh6-3. DNA methylation levels were calculated in WT at 23°C and 37°C.

Supplementary figure 11
A. TEs localized in chromosome arms or pericentromeres were aligned at their 5'-end or 3'-end and average cytosine methylation levels in the indicated nucleotide contexts were calculated from 3 kb upstream to 3 kb downstream in WT and med14-3. Upstream and downstream regions were divided in 100bp bins, while annotations were divided in 40 bins of equal length.
B. Average DNA methylation levels were calculated at protein coding genes as in A.
B. DNA methylation levels in CHG context in the indicated genotypes at med14-3 hypo-CHG DMRs.
C. DNA methylation levels in CHH context in WT and med14-3 at hypo-CHH DMRs identified in drm1/2, nrpd1 and nrpe1. Statistically significant differences between distributions in WT and med14-3 were tested by unpaired two-sided Mann-Whitney test.