Dual control of ROS1-mediated active DNA demethylation by DNA damage-binding protein 2 (DDB2)

By controlling gene expression, DNA methylation contributes to key regulatory processes during plant development. Genomic methylation patterns are dynamic and must be properly maintained and/or re-estab-lished upon DNA replication and active removal, and therefore require sophisticated control mechanisms. Here we identify direct interplay between the DNA repair factor DNA damage-binding protein 2 (DDB2) and the ROS1-mediated active DNA demethylation pathway in Arabidopsis thaliana . We show that DDB2 forms a complex with ROS1 and AGO4 and that they act at the ROS1 locus to modulate levels of DNA methylation and therefore ROS1 expression. We found that DDB2 represses enzymatic activity of ROS1. DNA demethylation intermediates generated by ROS1 are processed by the DNA 3 0 -phosphatase ZDP and the apurinic/ apyrimidinic endonuclease APE1L, and we also show that DDB2 interacts with both enzymes and stimulates their activities. Taken together, our results indicate that DDB2 acts as a critical regulator of ROS1-mediated active DNA demethylation.


INTRODUCTION
As sessile organisms, plants are particularly exposed to the deleterious effects of environmental stress, which can affect integrity of the genome and epigenome by inducing DNA damage and chromatin alterations (Pecinka and Mittelsten Scheid, 2012). Therefore it would be of evolutionary benefit to plants to combine genome and epigenome surveillance processes, to efficiently deal with the chromosome instability that could be transmitted to the progeny. In Arabidopsis, interplays between genome and epigenome stability have been reported. Loss of function of DNA damage-binding protein 2 (DDB2), which is involved in the recognition of UV-induced DNA lesions during global genome repair (GGR; Chu and Chang, 1988;Molinier et al., 2008), and of the MutS homologue 1 (MSH1), involved in mismatch repair (MMR), lead to genome-wide alterations in DNA methylation (Virdi et al., 2015;. However, the underlying mechanisms remain to be determined. DNA methylation (5-methyl cytosine, 5-meC), a key component of the plant epigenome, contributes to the stable silencing of transposable elements (TEs) as well as to the regulation of gene expression (Law and Jacobsen, 2010). In Arabidopsis, DNA methylation is maintained by four main DNA methyltransferases: methyltransferase 1 (MET1), chromomethylase 3 (CMT3), chromomethylase 2 (CMT2) and domains rearranged methyltransferase 2 (DRM2) acting in the CG, CHG and CHH sequence contexts (where H = A, T or C), respectively (Law and Jacobsen, 2010;Zemach et al., 2013). Additionally, cytosines can be methylated de novo through RNA-directed DNA methylation (RdDM) involving different classes of small RNA (Cuerda-Gil and Slotkin, 2016). RdDM transcriptionally represses TEs and is involved in heterochromatin formation (Matzke and Mosher, 2014).
In order to prevent abnormal alterations of DNA methylation patterns, expression of DNA demethylases is under the complex control of different pathways (Penterman et al., 2007). Expression of DML3, for example, is regulated post-transcriptionally by the microRNA miR402 (Kim et al., 2010). Conversely, expression of ROS1 is regulated transcriptionally by a complex balance between DNA methylation and active DNA demethylation involving a particular sequence in the ROS1 promoter region called the DNA methylation monitoring sequence (MEMS), which serves as a methyl-rheostat (Lei et al., 2015;Williams et al., 2015). DNA methylation of MEMS activates expression of ROS1 whilst active DNA demethylation of MEMS represses it (Lei et al., 2015;Williams et al., 2015). Active DNA demethylation can also be controlled via regulation of DNA glycosylase/AP lyase activity. ROS1 contains an iron-sulfur cluster-binding motif that is essential for its enzymatic activity (Duan et al., 2015;Wang et al., 2016). Moreover, the iron-sulfur cluster assembly protein MET18 interacts with ROS1 and stimulates its activity (Duan et al., 2015;Wang et al., 2016).
Given that active DNA demethylation is related to the BER pathway, it is likely that BER factors play an important role. Indeed, the BER protein XRCC1 was shown to interact with ROS1 and ZDP and stimulate their enzymatic activities (Mart ınez-Mac ıas et al., 2013), thus strengthening the idea that specific DNA repair factors may act as key regulatory elements of DNA demethylation.
We have recently reported that DDB2 loss of function induces alterations in DNA methylation arising partially from ROS1 upregulation . Interestingly DDB2 also controls de novo DNA methylation by interacting with the RdDM factor AGO4 . Given that ROS1 expression is under the control of RdDM and active DNA demethylation we hypothesized that DDB2 may play a role in such antagonistic regulatory processes.
Here, we decipher an interplay between DDB2, AGO4 and the ROS1-mediated active DNA demethylation pathway at the ROS1 locus. We unveil that DDB2 together with ROS1 and AGO4 form a complex in planta. This, at least, tripartite complex acts predominantly at MEMS of the ROS1 locus, modulating its DNA methylation levels. We found that DDB2 interacts directly with ROS1 and inhibits its enzymatic activity. Moreover, we show that DDB2 also directly interacts with two downstream DNA demethylation factors, ZDP and APE1L, and stimulates their activities. Collectively, these results suggest that the GGR factor DDB2 has an important function during ROS1-mediated active DNA demethylation via the dual control of ROS1 expression and activity.

DDB2 represses ROS1 expression
Expression of ROS1 was shown to be promoted by DNA methylation and antagonized by active DNA demethylation pathways (Lei et al., 2015;Williams et al., 2015). Interestingly upregulation of ROS1 was observed in plants defective for the expression of the GGR factor DDB2 , suggesting that this DNA repair protein acts, directly or indirectly, as a repressor of ROS1 expression. In addition, DDB2 interconnects with the RdDM pathway via its interaction with AGO4 and therefore represents a material of choice for deciphering the underlying mechanisms of control of ROS1 expression. Examination of the DNA methylation profile obtained by whole-genome bisulfite sequencing (WGBS;  of the ROS1 locus revealed that, in ddb2 mutant plants, two regions exhibit alteration of DNA methylation profiles compared with wild-type (WT) plants ( Figure 1a). One hypomethylated region (proximal, P) overlaps with the MEMS (Lei et al., 2015;Williams et al., 2015;Figure 1a). Another region (distal, D) is located towards the 3 0 end of the coding region, between exons 15 and 17, and shows an enhanced DNA methylation level in ddb2 mutant plants (Figure 1a). Similar alterations of DNA methylation profiles are also identified in another ddb2 allelic mutant (Nossen ecotype), confirming the above-described trends ( Figure S1a). These observations allow us to postulate that DDB2 controls DNA methylation levels at these two particular regions, and therefore ROS1 expression. In order to characterize the underlying molecular mechanisms, we analyzed small RNA deep sequencing data  and identified that 24-nucleotide (nt) siRNAs overlap with the P and D regions (Figure 1a and Figure S1a in the Supporting Information). We found that abundance of 24-nt siRNA is reduced in the P region but a significant increase is detected in the D region in the ddb2 mutant plants compared with WT plants (Figure 1a). These data are in agreement with the well-characterized direct correlation between 24-nt siRNA abundance and DNA methylation level (Matzke and Mosher, 2014). In addition, using DDB2-FLAG immunoprecipitation followed by small RNA-seq (RIP; Schalk et al., 2017) and publicly available AGO4 RIP data (Mi et al., 2008), we identified that the 24-nt siRNAs matching with the P region are loaded into a DDB2-and/or an AGO4-containing complex whereas those matching with the D region are exclusively loaded into AGO4 (Figure 1a). These results are consistent with the differential 24-nt siRNA abundance measured in WT and ddb2 plants at the P and D regions ( Figure 1a). These data support the idea that DNA methylation levels of these two regions are under the control of DDB2 and/or AGO4 siRNAs and may play an important role in the regulation of ROS1 expression.
To further examine the contributions of RdDM, active DNA demethylation and DDB2 to the control of ROS1 expression we tested the DNA methylation levels of P and D regions in double ddb2-nrpd1 and ddb2-ros1 plants.
McrBC-quantitative PCR (qPCR) analyses showed that Exons and introns are shown as boxes and lines, respectively. The REP5 transposon is shown in grey. Regions exhibiting significant changes in DNA methylation levels (regions P and D) are highlighted in red. Bottom panels: schematic representation of the DNA methylation level obtained by whole-genome bisulfite sequencing for the ROS1 locus in ddb2-3 plants. The histograms represent hyper/hypomethylation (at least 10% difference) in ddb2-3 plants compared with wild-type (WT; Col) plants for a 100-bp window for the P and D regions, the abundance of 24-nt siRNAs at the ROS1 locus in WT (Col), ddb2-3 and nrpd1 plants and relative 24-nt siRNA enrichments in DDB2 and AGO4 RIP. (b) Immunodetection of the ROS1 protein in WT (Col), ddb2 and cen2 plants. Coomassie blue staining of the blot is shown. (c) Immunoblot analysis of DDB2 protein content associated with the insoluble chromatin fraction (P, pellet) and the soluble fraction (S, supernatant) from WT, cen2 and ros1 plants. Anti-histone H3 and anti-UGPase antibodies were used as controls for insoluble (P) and soluble fractions (S), respectively. Signal intensity relative to H3 or Coomassie is indicated below each lane. Coomassie blue staining of the blot is shown. (d) Percentage of DNA methylation in WT (Col), ddb2 and cen2 plants for P and D regions of the ROS1 locus. Data are presented as percentage of methylation (AESD) determined by McrBC-qPCR and are representative of three biological replicates. t-test *P < 0.01; ns, non-significant compared with WT (Col).
ddb2-induced DNA hypomethylation in the P region and DNA hypermethylation in the D region reverted to WT levels in both ddb2-nrpd1 and ddb2-ros1 double-mutant plants ( Figure S1b, c). The strong reduction in ROS1 expression levels in ddb2-nrpd1 plants ( Figure S1b) despite the WT level of DNA methylation indicates that DNA methylation may not be the only factor influencing the steady-state level of ROS1 mRNA. In addition, our results highlight that DDB2-and ROS1-mediated active DNA demethylation act antagonistically at the P region whereas both factors may negatively control de novo DNA methylation at the D region.
Depletion of the cognate GGR factor, CEN2, also leads to enhanced ROS1 expression (Schalk and Molinier, 2016; Figure 1b). Interestingly CEN2 influences DDB2 homeostasis (Matsumoto et al., 2015). Indeed, a higher chromatin DDB2 content could be detected in cen2 plants compared with WT plants (Figure 1c). Therefore, cen2 mutant plants represent a material of choice to better assess the role of DDB2 in the control of ROS1 expression. Thus we compared the DNA methylation levels of the P and D regions in WT, ddb2 and cen2 plants using McrBC-qPCR. Unlike ddb2 plants, cen2 plants, exhibit a higher DNA methylation level exclusively at the P region compared with WT plants and not at the D region (Figure 1d), consistent with the direct correlation between DNA methylation of the MEMS, overlapping the P region, and ROS1 expression (Lei et al., 2015). In addition, this strongly suggests that a prospective direct correlation may exist between DDB2 chromatin content and DNA methylation levels.
Taken together, these results strengthen the idea that control of ROS1 expression is regulated by a complex DNA methylation-based process in which GGR factors, and probably DDB2, play a crucial role. These results also suggest that the D region may serve as back-up rheostat to control ROS1 expression.
DDB2, ROS1 and AGO4 act directly at the ROS1 locus The above results indicate that GGR factors influence DNA methylation levels predominantly at the P region (MEMS), and highlight that the D region, in the absence of DDB2, may play a secondary role in the control of ROS1 expression. To further explore the relationship between DNA methylation levels at these particular regions and local enrichment of DDB2 and of key effectors of the DNA methylation/demethylation processes, we performed chromatin immunoprecipitation (ChIP) experiments in a set of Arabidopsis mutant plants.
Using DDB2-FLAG-expressing plants , ChIP-qPCR analyses revealed a significant enrichment of DDB2 in the P region whereas no difference was detectable in the D region compared with WT control plants ( Figure 2a). Interestingly, in rdr2 DDB2-FLAG-expressing plants, in which AGO4 and ROS1 are not expressed (Penterman et al., 2007), DDB2 enrichment in the P region was enhanced compared with DDB2-FLAG plants ( Figure 2a). These data are correlated with the higher DDB2 chromatin contents found in ago4 plants . Conversely, no significant difference in DDB2-FLAG enrichment was measured for the D region ( Figure 2a) suggesting an indirect effect on the DNA methylation level.
Similarly to cen2 and ago4 plants, DDB2 chromatin content is higher in ros1 plants than in WT plants (Figure 1c), suggesting that ROS1 may also influence DDB2 chromatin enrichment. Thus we investigated DDB2 content at the ROS1 locus in cen2 and ros1 DDB2-FLAG-expressing plants. For the P and D regions DDB2 enrichment was around the background level in both cen2 and ros1 plants ( Figure 2a). This indicates that DDB2 binding at the P region requires CEN2 and ROS1. Moreover, these data highlight that ROS1 and CEN2 similarly influence local (ROS1 locus) and global DDB2 chromatin content.
Given that interconnection exists between RdDM and DDB2  and that ROS1 expression is promoted by DNA methylation we performed ChIP experiments using anti-AGO4 antibody in WT and ddb2 plants.
In WT plants AGO4 enrichment was exclusively observed in the D region, whereas in ddb2 plants it was enhanced for both regions (Figure 2b). This observation is compatible with a possible direct correlation between the specific AGO4 content at the D region and the enhanced 24-nt siRNA abundance identified in ddb2 plants ( Figure 1a). Moreover, these findings strengthen the idea that DDB2 may act as a chaperone for AGO4, possibly controlling its availability on chromatin for DNA methylation effectors.
ROS1 expression was shown to be antagonized by itself at MEMS (Lei et al., 2015). Therefore, we tested the chromatin association of ROS1 with the P and D regions in ros1, ddb2 or cen2 plants expressing C-terminal FLAGtagged ROS1 protein. The ChIP-qPCR analyses revealed a significant enrichment of ROS1-FLAG in the P and D regions (Figure 2c) confirming the results of Lei et al. (2015). Interestingly, in ddb2 mutant plants, ROS1-FLAG enrichment was enhanced in the P region whereas no significant change was detectable for the D region ( Figure 2c). These data are correlated with the putative direct relationship between local enhancement of ROS1 content in the P region ( Figure 2c) and its reduced DNA methylation level measured in ddb2 plants ( Figure 1d). Conversely, cen2 plants exhibit reduced ROS1 enrichment in the P and D regions compared with ros1 ROS1-FLAG and ddb2 ROS1-FLAG plants (Figure 2c) that directly correlates with the DNA methylation levels measured in cen2 plants (Figure 1d).
Taken together, these ChIP experiments allow DDB2 to be added as a new local player in the control of ROS1 expression together with ROS1 and AGO4. Interestingly, our data shed light on a complex co-regulation of DDB2, ROS1 and AGO4 contents predominantly at the P region and more moderately at the D region. We propose that there is subtle regulation of the availability and stability of DDB2, ROS1 and AGO4 which modulates ROS1 expression, probably through fine tuning of DNA methylation profiles.

ROS1 interacts with DDB2 and AGO4
Given that ROS1 and DDB2 bind DNA directly, and that both proteins are enriched at the P region of ROS1, we wondered whether DDB2 could physically interact with ROS1. Using in vitro pull-down experiments, we demonstrated that ROS1 interacts directly with DDB2 ( Figure 3a). In addition, the ROS1-FLAG protein co-immunoprecipitated with endogenous DDB2 protein (Figure 3b), confirming the aforementioned in vitro data. ROS1-dependent active DNA demethylation antagonizes RdDM (Zhu, 2009;Tang et al., 2016). Since DDB2 interacts directly with ROS1 and also assembles with AGO4 in a high-molecular-weight complex , it is possible that these three proteins are part of the same protein complex. Indeed, endogenous DDB2 and AGO4 co-immunoprecipitated with ROS1-FLAG ( Figure 3c). Collectively, these data provide compelling evidence that ROS1, DDB2 and AGO4 form a complex that may trigger either silencing or anti-silencing at particular loci.

DDB2 inhibits ROS1 5-meC DNA glycosylase activity
Since DDB2 and ROS1 interact both in vitro and in vivo, we asked whether DDB2 exerts any effect on the enzymatic activity of ROS1. ROS1 was incubated, in either the absence or the presence of DDB2, with a 51-mer duplex oligo substrate that contained a 5-meC:G at position 29 ( Figure 4a). ROS1 is a bifunctional DNA glycosylase/AP lyase that catalyzes both the release of 5-meC and the subsequent cleavage of DNA at the resulting abasic site by b,d-elimination (Morales-Ruiz et al., 2006). In this assay, the 5-meC DNA glycosylase/lyase activity is detected by the appearance of 28-nt incision products with either 3 0 -PUA or 3 0 -P ends. We observed that increasing amounts of DDB2 strongly inhibit the processing of 5-meC by ROS1 (Figure 4b).
Although the base excision and strand incision steps catalyzed by ROS1 are highly coordinated (Ponferrada-Mar ın et al., 2009), they are the result of separate DNA glycosylase and AP lyase activities that can be uncoupled by Purified DDB2 fused to a His tag was fixed to a Ni-Sepharose column and incubated with either MBP-ROS1 or MBP. After washes, the proteins associated with the resin were separated by SDS-PAGE, transferred to a membrane and immunoblotted with antibodies against MBP. (b) In vivo pull-down of ROS1 with DDB2 protein. WT (Col) and ros1 ROS1-FLAG plants were used for immunoprecipitation assays using anti-FLAG antibody. Coomassie blue staining of the blot is shown. *Cross-reacting signal. (c) In vivo pull-down of ROS1 with DDB2 and AGO4 proteins. WT (Col) and ros1 ROS1-FLAG expressing plants were used for immunoprecipitation assays using anti-FLAG antibody. Coomassie blue staining of the blot is shown. *Cross-reacting signal. (a) Schematic diagram of molecules used as DNA substrates. Double-stranded oligonucleotides containing either a single 5-meC or an AP site (indicated as X) on the upper strand. The fluor-labeled 5 0 -end of the upper strand is indicated by an asterisk. The size of the 5 0 -end labeled fragment generated after DNA glycosylase/lyase activity is indicated. (b) DDB2 inhibits the DNA glycosylase activity of ROS1. Purified ROS1 was incubated with a doublestranded oligonucleotide substrate containing a single 5-meC at the 5 0 -end-labeled upper strand, in the absence or presence of purified wild-type (WT) or mutant versions of DDB2. Graph data on the left are mean AE SE from three independent experiments. A representative gel is shown on the right. (c) DDB2 has no effect on the AP lyase activity of ROS1. Purified ROS1 was incubated with a doublestranded oligonucleotide substrate containing an AP site at the 5 0 -end-labeled upper strand, in the absence or the presence of purified WT and mutant versions of DDB2. Graph data on the left are mean AE SE from three independent experiments. A representative gel is shown on the right. We therefore asked whether DDB2 inhibits either the DNA glycosylase activity or the AP lyase activity of ROS1. To specifically detect ROS1 AP lyase activity, we performed enzymatic reactions with an analogous DNA substrate containing an abasic site instead of the target 5-meC at position 29 (Figure 4c). The results show that DDB2 does not exert any detectable effect on ROS1 AP lyase activity. Taken together, these results indicate that DDB2 specifically inhibits the 5-meC excision step catalyzed by ROS1.
Since both DDB2 and ROS1 have DNA-binding capacity, we next asked whether the inhibitory effect of DDB2 on the 5-meC excision activity of ROS1 could be a consequence of competition for the DNA substrate. To test this hypothesis, the 5-meC-containing DNA was either pre-incubated or not with DDB2 before adding ROS1 to the reaction mixture ( Figure S2). We found that pre-incubating the DNA substrate with DDB2 does not increase ROS1 inhibition, which argues against a simple competition effect. This result is in agreement with the fact that a K314E-mutated form of DDB2, which is impaired in DNA-binding , retains most of the inhibitory capacity exerted by WT DDB2 on ROS1 (Figure 4b, c). Taken together, these results suggest that DDB2 directly inhibits excision of 5-meC catalyzed by ROS1.

DDB2 stimulates post-incision events in the DNA demethylation pathway
The plant DNA demethylation pathway initiated by ROS1 is a multistep BER process in which additional proteins are required to catalyze downstream repair stages that culminate in the replacement of 5-meC with unmethylated C. As indicated above, the product generated by ROS1 is a single-nucleotide gap with either 3 0 -P or 3 0 -PUA ends. These non-canonical 3 0 termini must be converted to 3 0 -OH ends before DNA polymerase and ligase activities can complete DNA demethylation. We have previously reported that the DNA 3 0 -phosphatase ZDP removes the blocking 3 0 -P, allowing subsequent DNA polymerization and ligation steps needed to complete demethylation. Furthermore, ZDP and ROS1 interact in vitro and co-localize in vivo (Mart ınez-Mac ıas et al., 2012). We therefore asked whether DDB2 exerts any effect on the DNA-3 0 phosphatase activity of ZDP. Purified ZDP was incubated with a 5 0 -end-labeled single-nucleotide gapped substrate (Table S1), either in the absence or presence of DDB2, and measured the conversion from 3 0 -P to 3 0 -OH ( Figure 5a). Interestingly, and in stark contrast to the effect on ROS1, we found that DDB2 stimulates ZDP activity. In the absence of ZDP, the DDB2 protein showed no detectable activity on the gapped (a) DDB2 stimulates the DNA 3 0 -phosphatase activity of ZDP. Purified ZDP was incubated with a double-stranded oligonucleotide substrate containing a single-nucleotide gap flanked by 3 0 -phosphate and 5 0 -phosphate ends at the 5 0 -end-labeled upper strand, in the absence or presence of purified WT or mutant versions of DDB2. Graph data on the left are mean AE SE from three independent experiments. A representative gel is shown on the right. (b) DDB2 stimulates the 3 0 -phosphodiesterase activity of APE1L. Purified APE1L was incubated with a double-stranded oligonucleotide substrate containing a single-nucleotide gap flanked by 3 0 -PUA and 5 0 -phosphate ends at the 5 0 -end-labeled upper strand, in the absence or presence of purified WT or mutant versions of DDB2. Graph data on the left are mean AE SE from three independent experiments. A representative gel is shown on the right. (c) DDB2 interacts with ZDP and APE1L. Purified ZDP and APE1L fused to MBP, or MBP alone, were fixed to an amylose column. The proteins bound to the column were incubated in the presence of His-DDB2. After washes, the proteins associated with the resin were separated by SDS-PAGE, transferred to a membrane and immunoblotted with antibodies against the His-tag. The other types of non-canonical 3 0 -termini generated by ROS1 are 3 0 -PUA ends. We have previously shown that APE1L, one of three Arabidopsis AP endonucleases, has a potent 3 0 -phosphodiesterase activity that processes the 3 0 -PUA blocking groups to generate 3 0 -OH. Furthermore, APE1L and ROS1 interact in vitro and in vivo (Li et al., 2015a). To test the effect of DDB2 on the 3 0 -phosphodiesterase activity of APE1L we incubated both proteins, either singly or in combination, with a 5 0 -end-labeled DNA gapped substrate containing a 3 0 -end and measured its conversion to 3 0 -OH (Figure 5b). We found that DDB2 enhances APE1L activity, although at higher concentrations than those required to stimulate ZDP. In the absence of APE1L, the DDB2 protein exhibited no detectable activity on 3 0 -PUA ends (Figure 5b, lane 1), suggesting that the enhanced activity indicates stimulation of APE1L. APE1L is also endowed with a weak AP-endonuclease activity that was not affected by DDB2 ( Figure S3a).
The K314E mutated form of DDB2 retains the stimulatory capacity exerted by WT DDB2 either on ZDP or APE1 (Figure 5), which suggests that stimulation is independent of DNA-binding by DDB2. We performed in vitro pull-down assays to prove a direct interaction between DDB2 and either ZDP or APE1L. As shown in Figure 5(c), MBP-ZDP and MBP-APE1L, but not MBP alone, bound to His-DDB2. Taken together, these results suggest that DDB2 interacts directly with ZDP and APE1L and stimulates their 3 0 -phosphatase and 3 0 -phosphodiesterase activity, respectively.
We also examined the possible effect of DDB2 on additional stages of the DNA demethylation pathway. The DNA polymerase(s) responsible for gap filling have not yet been identified, but it has been reported that the final ligation step is catalyzed by LIG1 (C ordoba-Cañero et al., 2011;Li et al., 2015b). However, we could not detect any effect of DDB2 on LIG1 DNA ligase activity ( Figure S3b). Collectively, these results suggest that DDB2 stimulates post-incision events during the DNA demethylation pathway, which may help to avoid accumulation of repair intermediates with blocked 3 0 -ends.

DISCUSSION
In this study we demonstrate that DDB2 contributes to the control of ROS1 expression through fine-tuning of the DNA methylation level of the ROS1 locus. We reveal that ROS1, DDB2 and AGO4 form a complex in vivo, allowing DDB2 to mediate a sophisticated control of RdDM and of active DNA demethylation pathways predominantly at MEMS (Figure 6a). Moreover, we unveil that DDB2 directly interacts with three components of the active demethylation pathway, ROS1, ZDP and APE1L, inhibiting ROS1 glycosylase activity and stimulating ZDP and APE1L activities (Figure 6b). Taken together, our results strengthen the notion that a direct interplay exists between DNA repair and DNA methylation dynamics.

DDB2 controls active DNA demethylation
Our analyses show that DDB2 acts at several stages during active DNA demethylation. We demonstrated that DDB2 interacts in vitro and in vivo with ROS1 and inhibits its 5-meC DNA glycosylase activity. This differs from the stimulatory effect reported for other ROS1 interactors (Mart ınez-Mac ıas et al., 2013;Duan et al., 2015;Wang et al., 2016). Interestingly, we found that the inhibitory effect of DDB2 is not a consequence of direct competition between ROS1 and DDB2 for the DNA substrate. It would thus be interesting to map the DDB2-ROS1 interaction domains in order to better decipher how the enzymatic activity of ROS1 is inhibited. Upon removal of 5-meC, gaps flanked by 3 0 -P or 3 0 -PUA blocking ends are generated and must be processed to allow efficient gap filling and DNA ligation (Mart ınez-Mac ıas et al., 2012;Li et al., 2015a). We found that DDB2 interacts with the Arabidopsis DNA phosphatase ZDP and AP endonuclease APE1L, stimulating their activities. The intrinsic capacity of DDB2 to bind changes in DNA structure (Wittschieben et al., 2005) may explain why DDB2 acts at this step of the active DNA demethylation process. This result allows us to speculate that, by sensing DNA demethylation intermediates, DDB2 could be a general factor in active DNA demethylation including DME-, DML2-and DML3-mediated processes. Another DNA repair factor, XRCC1, which acts as a scaffolding protein during BER, interacts with ZDP and stimulates its enzymatic activity (Mart ınez-Mac ıas et al., 2013). Together these data stress the point that, as a DNA repair process, many factors related to short/long patch BER or NER may contribute to active DNA demethylation.

DDB2-mediated control of ROS1 expression
Expression of ROS1 is under the complex influence of DNA methylation and active DNA demethylation pathways (Penterman et al., 2007;Lei et al., 2015;Williams et al., 2015). The DNA methylation level of the MEMS is positively correlated with ROS1 expression (Lei et al., 2015;Williams et al., 2015). In this system, ROS1 counteracts RdDM to remove methylation at its own promoter. Therefore, factors that simultaneously influence DNA methylation and active DNA demethylation may represent putative key sophisticated regulators of ROS1 expression. Indeed, the GGR factor DDB2 also carries such dual features. DDB2 was shown to influence de novo DNA methylation by controlling the local abundance of 24-nt siRNAs via interaction with AGO4  as well as to repress ROS1 activity (this study).
Given that DNA methylation at MEMS is targeted by ROS1-dependent active demethylation, the slight but significant loss of DNA methylation detected at the MEMS in ddb2 mutant plants may reflect a local increase in ROS1 activity. This hypomethylation correlates with the enhanced ROS1-binding identified in ddb2 plants and with the loss of the DDB2 inhibitory effect of ROS1. Apparently, this ddb2-induced MEMS hypomethylation associated with ROS1 overexpression challenges the model in which DNA methylation of the MEMS sequence restores ROS1 expression to WT levels (Lei et al., 2015;Williams et al., 2015). However, it should be noted that DDB2 deficiency increases ROS1 expression to well over WT levels (Figures 1b and S1c). Together, these observations suggest that the repressive effect of DDB2 on ROS1 expression is not predominantly mediated by changes in MEMS DNA methylation. We propose that DDB2 has a dual function. First, through direct interaction with AGO4 and ROS1, stimulating methylation and inhibiting demethylation and second as a transcriptional repressor of ROS1 expression. Interestingly, mammalian DDB2 has been reported as a transcriptional repressor of a constitutive gene (Minig et al., 2009). Moreover, it is tempting to speculate that the association of DDB2 with ROS1 may influence its own stability since DDB2 is subjected to ubiquitin-mediated proteolytic degradation (Molinier et al., 2008). An important question is why DDB2 is required to control the expression Figure 6. Model for DDB2 control of ROS1mediated active DNA demethylation. (a) DDB2, together with AGO4 and ROS1, act as regulators of ROS1 expression. Schematic representation of the ROS1 locus. The REP5 transposon is shown in grey. The P and D regulatory regions are highlighted in red. Top panel: at the P region in WT plants, DDB2 prevents active DNA demethylation by ROS1 (thin red arrow) allowing AGO4-mediated DNA methylation (thick green arrow). Conversely, ROS1 actively demethylates the D region. Bottom panel: in the absence of DDB2, active DNA demethylation and AGO4-mediated methylation are enhanced at the P and D loci, respectively, and ROS1 expression increases. of ROS1. One possibility is that DDB2 functions to fine-tune the MEMS methylstat in order to prevent genome-damaging effects. We propose a model (Figure 6a) in which DDB2 is recruited to the ROS1 promoter region to avoid the accumulation of DNA demethylation intermediates arising from the cycling process of DNA methylation/ demethylation at MEMS. Such recruitment, which might involve the capacity of DDB2 to detect altered conformations of DNA (Wittschieben et al., 2005), stimulates the processing of such intermediates (Figure 6b), inhibits ROS1 demethylation activity, represses ROS1 expression and stimulates RdDM-dependent DNA methylation. The net effect is a WT steady-state level of both DNA methylation and ROS1 expression. In a ddb2 mutant background, MEMS methylation decreases and ROS1 expression increases over WT levels via the release of ROS1 activity. Interestingly, we found that a region located between exons 15 and 17 of the ROS1 locus, called the D region, displays an enhanced DNA methylation level in ddb2 mutant plants. This hypermethylation could be directly correlated with the enhanced level of ROS1 expression and is possibly the result of transcription-coupled mechanisms involving MET1 (Teixeira and Colot, 2009;Miura et al., 2009). Moreover, in ddb2 mutant plants the methylation level of the D region is positively correlated with enhanced 24-nt siRNA abundance and AGO4 binding. We identified that this region is under the influence of both RdDM and ROS1. Thus, the decrease in MEMS DNA methylation might be counterbalanced by the RdDM-dependent hypermethylation of the D region leading to restoration and/or enhanced ROS1 expression. Consistent with this, plants impaired in RdDM exhibit both MEMS and D region hypomethylation (Williams et al., 2015). Loss of DNA methylation at the P region and hypermethylation at the D region were also observed in met1 mutant plants, but unlike in the ddb2 mutant plants ROS1 expression was strongly reduced (Rigal et al., 2012;Williams et al., 2015). This discrepancy between both mutants could be explained by the minimal level of DNA methylation maintained at MEMS, probably due to the influence of DDB2 on de novo DNA methylation . This may reflect that RdDM takes over DNA demethylation at this particular region and that DDB2 acts as a dual regulator. In addition, our results allow us to propose that the D region may serve as back-up regulatory MEMS to maintain, restore or enhance ROS1 expression, especially when DDB2 homeostasis varies. Elevated ROS1 expression was also detected in plants defective for the expression of the cognate GGR factors CEN2 and RAD10 (Schalk and Molinier, 2016). Contrary to the ddb2 mutant plants, cen2 plants exhibit hypermethylation of the MEMS in agreement with the already published model (Lei et al., 2015;Williams et al., 2015). CEN2 is part of the XPC complex that recognizes bulky DNA lesions during the NER pathway. CEN2 stabilizes XPC on chromatin and inhibits ubiquitin-dependent degradation of DDB2, enhancing its retention on chromatin (Matsumoto et al., 2015). Therefore, the enrichment of DDB2 on chromatin in cen2 mutant plants would entail that hypermethylation of the MEMS is be due to the inhibition of ROS1 chromatin loading/activity.
Given that ROS1 expression is under the complex control of different factors such as the DNA methyltransferases MET1, CMT3, DRM2 and the H3K9 histone methyl transferase/demethylase KYP (kryptonite), IBM1 (increase in BONSAI methylation 1), respectively (Rigal et al., 2012) it would be of great interest to establish how DDB2 interconnects with these factors. Additionally, it is likely that expression of ROS1 is not only regulated by DNA methylation. The fact that recovery of methylation at the P locus in ddb2 nrpd1 double mutants does not restore ROS1 expression to WT levels ( Figure S1b) supports this idea. It should also be considered that certain growth conditions (i.e. upon UV exposure) that influence DDB2 homeostasis/dynamics may also affect ROS1 expression. Indeed, it has been reported that upon UV exposure the steady-state level of DDB2 mRNA increases (Al Khateeb and Schroeder, 2009;Biedermann and Hellmann, 2010) while ROS1 decreases (Q€ uesta et al., 2013). On the other hand ros1 mutants show increased photorepair (Q€ uesta et al., 2013), which suggests that such ROS1 repression, mediated either by upregulation of DDB2 or by other factors, is important for a coordinated response to UV-induced DNA damage. This correlation emphasizes that contents of DDB2 and ROS1 are closely related and support the hypothesis that environmental stress may also control the DNA methylation landscape in a DDB2-ROS1-dependent manner.

DDB2 plays a key role in DNA methylation dynamics
Our study unveils DDB2 as an important factor controlling active DNA demethylation, adding a new level of knowledge about the complex regulation of DNA methylation dynamics. Actually, ddb2 mutant plants exhibit genomewide alterations in DNA methylation . These changes are explained by the misregulation of the local control of de novo DNA methylation by the DDB2-AGO4-siRNA complex and also by misregulation of ROS1 and DML3 expression . In the course of this study we showed that ROS1 forms, in vivo, a complex with AGO4 and DDB2. Therefore, DDB2-dependent control of the local abundance of 24-nt siRNA  and of the demethylation activities provides compelling evidence that DDB2 directly interconnects the RdDM and the active DNA demethylation pathways. More generally we could consider that DDB2 plays a regulatory role in the control of DNA methylation dynamics, ensuring efficient shaping of the methylome by interconnecting/coordinating different pathways.

Generation of transgenic plants
The genomic DNA (ATG to stop codon) of Arabidopsis ROS1 was amplified by PCR using primers described in . ROS1 genomic DNA was sequenced and cloned between the NcoI and AvrII sites into the pOEX2 vector (Molinier et al., 2004). The pOEX2 ROS1-FLAG plasmid was mobilized into Agrobacterium tumefaciens to transform ros1-3 and ddb2-3 Arabidopsis plants.
The DDB2-FLAG construct  was used to transform rdr2 plants.

Immunoprecipitation assays
One gram of 10-day-old ros1 ROS1-FLAG-expressing seedlings was used to extract total soluble proteins in 3 ml of IP buffer (Pazhouhandeh et al., 2011). Anti-FLAG gel affinity (Sigma, http:// www.sigmaaldrich.com/) was used for immunoprecipitation. The precipitate was washed four times in IP buffer, resuspended in 50 ll of SDS sample buffer and heated for 3 min at 100°C prior to immunoblotting. Anti-FLAG horseradish peroxidase (Sigma; 1:10 000 dilution) was used to detect the ROS1-FLAG protein.

McrBC treatment
Genomic DNA (1.5 lg) was digested with the McrBC enzyme (New England Biolabs, https://www.neb.com/) for 8 h at 37°C. Undigested genomic DNA (1.5 lg) was used as a control. DNA methylation levels at the P and D regions of ROS1 were determined by real-time PCR and percentage DNA methylation was calculated as described in . Experiments were done in triplicate using independent biological samples. Three technical replicates were performed for each independent biological sample. Primers are listed in Table S2.

Chromatin immunoprecipitation
The ChIP experiments were performed according Pazhouhandeh et al. (2011) using 10-day-old seedlings grown in vitro (Roudier et al., 2011). The immunoprecipitated DNA was analysed by realtime qPCR. Experiments were done in triplicate using independent biological samples. Three technical replicates were performed for each independent biological sample.

Quantitative PCR
Quantitative PCR was performed using a LightCycler 480 and LightCycler 480 SYBR green I Master mix (Roche, http://www.roc he.com/) following the manufacturer's instructions. Normalization was performed relative to the housekeeping genes Actin2, Ubiquino-cytochrome C Reductase and Hexokinase 1 as described in .

Enzyme assays
Double-stranded oligonucleotides (20 nM) were incubated at the specified temperature in a reaction mixture containing 50 mM 2-amino-2-(hydroxymethyl)-1,3-propanediol (TRIS)-HCl pH 8.0, 1 mM DTT, 0.1 mg ml À1 BSA and the corresponding amount of the indicated proteins in a total volume of 50 ll. Reactions mixtures containing APE1L or LIG1 also included 2 mM ATP and either 2 mM or 5 mM MgCl 2 , respectively. Reactions were stopped by adding 20 mM EDTA, 0.6% sodium dodecyl sulfate and 0.5 mg ml À1 proteinase K, and the mixtures were incubated at 37°C for 30 min. DNA was extracted with phenol/chloroform/ isoamyl alcohol (25:24:1) and ethanol precipitated at À20°C in the presence of 0.3 mM NaCl and 16 lg ml À1 glycogen. Samples were resuspended in 10 ll of 90% formamide, heated at 95°C for 5 min, and separated in a 12% denaturing polyacrylamide gel containing 7 M urea. Fluorescein or Alexa Fluor-labeled DNA was visualized using the fluorescence mode of an FLA-5100 imager and analyzed using Multigauge software (Fujifilm, http:// www.fujifilm.com/).

CONFLICT OF INTEREST
The authors have no conflict of interest to declare.

SUPPORTING INFORMATION
Additional Supporting Information may be found in the online version of this article. Figure S1. DNA methylation profiles at the ROS1 locus. Figure S2. Inhibition of ROS1 activity by DDB2 is not due to DNAbinding competition. Figure S3. DDB2 has no effect on the apurinic/apyrimidinic endonuclease activity of APE1L and the DNA ligase activity of LIG1 .  Table S1. The DNA sequences of oligonucleotides used as substrates. Table S2. List of primers used for quantitative PCR. Data S1. Experimental procedures.