A JUMONJI Protein with E3 Ligase and Histone H3 Binding Activities Affects Transposon Silencing in Arabidopsis

A conserved JUMONJI protein with E3 ubiquitin ligase activity affects transposon silencing.

Abstract

Transposable elements (TEs) make up a large proportion of eukaryotic genomes. As their mobilization creates genetic variation that threatens genome integrity, TEs are epigenetically silenced through several pathways, and this may spread to neighboring sequences. JUMONJI (JMJ) proteins can function as antisilencing factors and prevent silencing of genes next to TEs. Whether TE silencing is counterbalanced by the activity of antisilencing factors is still unclear. Here, we characterize JMJ24 as a regulator of TE silencing. We show that loss of JMJ24 results in increased silencing of the DNA transposon AtMu1c, while overexpression of JMJ24 reduces silencing. JMJ24 has a JumonjiC (JmjC) domain and two RING domains. JMJ24 autoubiquitinates in vitro, demonstrating E3 ligase activity of the RING domain(s). JMJ24-JmjC binds the N-terminal tail of histone H3, and full-length JMJ24 binds histone H3 in vivo. JMJ24 activity is anticorrelated with histone H3 Lys 9 dimethylation (H3K9me2) levels at AtMu1c. Double mutant analyses with epigenetic silencing mutants suggest that JMJ24 antagonizes histone H3K9me2 and requires H3K9 methyltransferases for its activity on AtMu1c. Genome-wide transcriptome analysis indicates that JMJ24 affects silencing at additional TEs. Our results suggest that the JmjC domain of JMJ24 has lost demethylase activity but has been retained as a binding domain for histone H3. This is in line with phylogenetic analyses indicating that JMJ24 (with the mutated JmjC domain) is widely conserved in angiosperms. Taken together, this study assigns a role in TE silencing to a conserved JmjC-domain protein with E3 ligase activity, but no demethylase activity.

Eukaryotic genomes contain a large proportion of transposable elements (TEs) and repetitive sequences (Tenaillon et al., 2010; Fedoroff, 2012). Through their mobilization and transposition, these sequences have a high mutagenic potential. This threatens genome integrity but also provides an important source of genetic variation for selection to act upon (Levin and Moran, 2011; Bennetzen and Wang, 2014). Thus, organisms have evolved mechanisms to limit TE activity. These mechanisms mostly involve DNA methylation, small RNAs, and histone modifications (Law and Jacobsen, 2010; Castel and Martienssen, 2013; Matzke and Mosher, 2014). TEs can be silenced to different degrees depending on which pathways dominate at individual loci. Many TEs are silenced through CG methylation that requires METHYLTRANSFERASE1 and DECREASED DNA METHYLATION1 for maintenance in Arabidopsis (Arabidopsis thaliana; Saze et al., 2003; Zemach et al., 2013). In RNA-dependent DNA methylation (RdDM), the generation of small-interfering RNA triggers the deposition of chromatin-associated silencing marks such as DNA methylation in CHG and CHH contexts (where H is A, C, or T) and histone H3 Lys 9 dimethylation (H3K9me2) at targeted loci (Castel and Martienssen, 2013; Matzke and Mosher, 2014). This involves the DNA methyltransferase DOMAINS REARRANGED METHYLTRANSFERASE (DRM1 and DRM2) and CHROMOMETHYLASE (CMT2 and CMT3) proteins (Cao and Jacobsen, 2002; Stroud et al., 2013, 2014). H3K9me2 is a repressive chromatin mark associated with the silencing of repeats and TEs (Jackson et al., 2002; Ebbs et al., 2005; Ebbs and Bender, 2006). The H3K9 methyltransferases KRYPTONITE (KYP/SUVH4), SUVH5, and SUVH6 regulate CHH methylation in a small-interfering RNA-independent manner and have been shown to be required for CMT3-dependent CHG methylation (Jackson et al., 2002; Ebbs et al., 2005; Ebbs and Bender, 2006; Stroud et al., 2013). CHG methylation and H3K9me2 reinforce each other through a positive feedback loop (Du et al., 2012).

The class II transposon Robertson’s Mutator element has originally been isolated from maize (Zea mays), where it transposes frequently (Lisch, 2012). Mutator elements from Arabidopsis have been characterized (Singer et al., 2001). For AtMu1, it was shown that it is targeted by several epigenetic silencing pathways, including RdDM (Lippman et al., 2003; Bäurle et al., 2007), and AtMu1 transposition was found in DNA methylation-deficient backgrounds and in the vegetative nucleus of pollen, where global reactivation of TEs occurs (Singer et al., 2001; Slotkin et al., 2009). AtMu1c is the most active of the three AtMu1 copies (Kabelitz et al., 2014). At the phylogenetic level, AtMu1c transposition was found in the Arabidopsis lineage (Kabelitz et al., 2014). AtMu1c contains two highly homologous terminal inverted repeats (TIRs) and a conserved transposase gene.

Chromatin structure is an important regulator of gene expression in all organisms. It is regulated to a large extent by the composition, localization, and posttranslational modification of nucleosomes (Struhl and Segal, 2013; Zentner and Henikoff, 2013). Nucleosomes consist of octamers of H2A, H2B, H3, and H4 proteins. All histones, but especially histone H3, can be modified at several residues with various modifications, including methylation, acetylation, and ubiquitination. Many of these modifications are reversible, thus contributing to the dynamic regulation of gene expression. For example, in the case of histone Lys methylation, histone methyltransferases perform mono-, di-, or trimethylation of Lys (Black et al., 2012). Conversely, the recently discovered histone demethylases reverse methylation. Two histone Lys demethylase classes are present in animals, plants, and yeast (Lu et al., 2008; Hong et al., 2009; Liu et al., 2010a; Mosammaparast and Shi, 2010; Chen et al., 2011). The JUMONJI (JMJ) gene class is characterized by the catalytic JumonjiC (JmjC) domain and contains Fe²⁺- and α-ketoglutarate-dependent dioxygenases (Klose et al., 2006a; Mosammaparast and Shi, 2010). Animal JMJ proteins have been implicated in many processes regulating development and disease (Klose et al., 2006a; Landeira and Fisher, 2011). In Arabidopsis, there are 21 JMJ genes, which can be categorized into five groups, namely, the KDM3/JHDM2, KDM4/JHDM3/JMJD2, KDM5/JARID1, JMJD6, and JmjC domain-only groups, based on sequence analysis and domain architecture (Lu et al., 2008; Hong et al., 2009). Members of the same clade tend to have similar target specificities (Klose et al., 2006a). About half of these genes have been functionally characterized so far, and they act in development and responses to endogenous and exogenous cues. For example, several characterized JMJ genes regulate flowering time through repression or activation of different target genes and different modifications (Noh et al., 2004; Yang et al., 2012; Crevillén et al., 2014; Gan et al., 2014). The histone H3 K4 demethylase JMJ14 is required for RNA-mediated DNA methylation (Deleris et al., 2010; Searle et al., 2010; Le Masson et al., 2012; Greenberg et al., 2013). A H3K4 demethylase from rice (Oryza sativa) has been implicated in the repression of TE sequences (Cui et al., 2013). The H3 K9 me2/me1 demethylase IBM1 prevents genes from being silenced through invasion of H3K9 methylation from neighboring TEs and repetitive elements (Saze et al., 2008; Inagaki et al., 2010). IBM1/JMJ25 does not generally target TEs (Inagaki et al., 2010; Rigal et al., 2012) and indirectly represses TEs through the activation of DCL3 and RDR2 (Fan et al., 2012). Very recently, JMJ24 has been proposed to function in the base transcription of silenced loci (Deng et al., 2015). However, its mode of action remains unclear.

Several members of the KDM3/JHDM2 clade from both kingdoms contain RING-finger domains that may function in target protein ubiquitination (Lu et al., 2008; Zhou and Ma, 2008; Aiese Cigliano et al., 2013). However, in none of these proteins has the RING domain been functionally characterized so far. Here, we functionally characterize the RING-finger- and JmjC-domain-containing protein JMJ24. Like IBM1, JMJ24 belongs to the KDM3 clade of putative histone H3 K9 demethylases. Interestingly, although the cofactor binding residues of the JmjC domain indicate loss of catalytic activity, JMJ24 is conserved across the angiosperm lineage and binds to histone H3 through this domain. JMJ24 also has ubiquitination activity. JMJ24 has a role in transposon silencing by antagonizing H3K9me2 through locus-specific interactions. Taken together, JMJ24 likely functions in TE silencing through ubiquitination of histones or associated target proteins.

RESULTS

JMJ24 Counteracts Transcriptional Silencing of AtMu1c

In a reverse genetics screen for potential regulators of AtMu1c silencing, we found that a mutant in the JMJ24 gene had a moderate but consistent decrease in AtMu1c transcript levels (Fig. 1; Supplemental Fig. S1). At the morphological level, the jmj24-2 mutant, which carries a T-DNA insertion in the second exon (Supplemental Fig. S1), did not show any apparent defects. To corroborate this observation, we generated lines expressing an artificial microRNA against JMJ24 in the Columbia (Col) background. In these lines, JMJ24 transcript levels were decreased to 76 to 20% of the transcript levels in Col (Fig. 1). Correspondingly, AtMu1c transcript levels were also decreased and the degree of reduction was correlated with the strength of JMJ24 down-regulation. Thus, JMJ24 antagonizes AtMu1c silencing. We next asked whether overexpression of JMJ24 was able to reduce AtMu1c silencing. To this end, we generated transgenic plants overexpressing JMJ24 under the control of the 35S promoter in the Col wild-type background. We did not observe any obvious morphological alterations in the 35S::JMJ24 plants. Indeed, we observed reduced silencing of AtMu1c as evidenced by enhanced transcript levels (Fig. 1). The degree of reactivation correlated with the magnitude of JMJ24 overexpression. The analysis of unspliced AtMu1c transcripts (Kabelitz et al., 2014) in jmj24-2 and 35S::JMJ24 plants (Supplemental Fig. S2) suggests that the release of silencing occurs at the level of transcription (i.e. before splicing). Together, our results indicate that JMJ24 is both necessary and sufficient to antagonize AtMu1c silencing.

Figure 1.

JMJ24 negatively regulates AtMu1 silencing. Relative JMJ24 (gray) and AtMu1c (black) transcript levels were determined by qRT-PCR. Expression values were normalized to TUB6 and Col. jmj24, jmj24-2 mutant; JMJ24 amiR, lines carrying an artificial microRNA against JMJ24; 35S::JMJ24, JMJ24 overexpression construct. For each construct, three independent homozygous T3 lines were analyzed. Data shown are averages over six biological replicates with error bars representing se.

JMJ24 Is Conserved across Angiosperms But Has Lost Fe²⁺-Binding Activity

We next sought to determine the molecular function of JMJ24. JMJ24 contains a JmjC domain and two RING finger domains as well as a nuclear localization signal (Fig. 2A). JMJ24 is a single-copy gene in Arabidopsis and is most closely related to the KDM3/JHDM2 clade of JmjC proteins, which also contains IBM1/JMJ25 (Hong et al., 2009). JMJ24 and three additional members of this clade (JMJ26, 27, and 29) contain two RING domains of the RING-C2 type, which are potential E3 ubiquitin ligase enzymes (Lorick et al., 1999; Stone et al., 2005; Hong et al., 2009; Aiese Cigliano et al., 2013). Based on sequence similarity, the KDM3/JHDM2 clade is predicted to have histone H3 K9 demethylase activity. In line with this prediction, histone H3 K9 demethylase activity was demonstrated for IBM1 (Inagaki et al., 2010). To function as a demethylase, the JmjC domain requires Fe²⁺ and α-ketoglutarate as cofactors (Klose et al., 2006a). In JMJ24, the two residues that are required for α-ketoglutarate binding are present (Fig. 2A; Supplemental Fig. S3A). However, two of the three highly conserved residues that bind Fe²⁺ are not present, suggesting that JMJ24-JmjC may not be able to bind Fe²⁺ and thus may not be an active demethylase. Given this potential inactivity, it was interesting to study whether there are JMJ24-related proteins with a similarly mutated binding site in other species. We found potential JMJ24 orthologs in all angiosperm families investigated, including the basal angiosperm Amborella trichopoda and a representative selection of monocot and dicot families, but not in the moss Physcomitrella patens and not outside the plant kingdom (Fig. 2B). These results confirm and extend a recent phylogenetic study, which identified a putative JMJ24 ortholog in A. trichopoda (Qian et al., 2015). Comparing the conservation of the five amino acid residues that are essential for cofactor binding (F/T/YHD/EKH), we found that all JMJ24-like proteins had a degenerated motif (e.g. THNKF in JMJ24; Fig. 2A). Moreover, in the JMJ24-like proteins, the second His of the Fe²⁺-binding motif was always converted into a Phe and the Asp at position three mostly into an Asn or Lys. Thus, although JMJ24 has lost the THDKH motif of the canonical KDM3-JmjC domain, the high degree of conservation of the modified JMJ24 motif suggests that this motif continues to be under positive selection and that the JMJ24-JmjC domain is functionally active, albeit probably not as a demethylase. Together, our phylogenetic analysis of protein sequences revealed that JMJ24 has evolved before the separation of monocotyledonous and dicotyledonous plants during basal angiosperm evolution and may have a functional activity differing from that of the canonical JmjC domain.

Figure 2.

Putative JMJ24 orthologs are present across the angiosperms. A, Schematic overview of the JMJ24 protein domains and cofactor binding sites. Numbers indicate amino acid positions of domain start and end. Amino acid residues required for Fe²⁺ binding and α-ketoglutarate binding are shown above the protein schematic. Corresponding residues in the JMJ24 sequence are shown below. NLS, nuclear localization signal; RING, RING finger domain. B, Phylogenetic tree with the KDM3 clade from Arabidopsis (ARATH, JMJ24, JMJ25/IBM1, JMJ26, JMJ27, and JMJ29), Homo sapiens KDM3A, KDM3B, and putative orthologs of JMJ24 from the basal angiosperm A. trichopoda (AMBTR), monocotyledonous species Sorghum bicolor (SORBI), Oryza sativa ssp. japonica (ORYSJ), Brachypodium distachyon (BRADI), Hordeum vulgare (HORVD), Zea mays (ZEAMA), and the dicotyledonous species Arabidopsis lyrata (ARALL), Solanum lycopersicum (SOLLC), Glycine max (SOYBN), Vitis vinifera (VITVI), Populus trichocarpa (POPTR), and Physcomitrella patens (PHYPA) using full-length protein sequences. Putative orthologs were identified using Delta BLAST and Inparanoid. The Arabidopsis JMJ18 sequence was used as outgroup. The phylogenetic tree was prepared using Phylogeny.fr and the default settings. Right panel, conservation of the residues required for Fe²⁺ and α-ketoglutarate binding in the aligned protein sequences. Red, amino acids conserved in canonical JmjC sequence; cyan, amino acids conserved in JMJ24-related proteins but deviating from consensus sequence. C, Iron binding analysis of JMJ24-JmjC. The amount of protein-bound iron was measured by ICP-OES and normalized to protein input. The analysis was performed in triplicate. GST served as a negative and JMJ18-JmjC as a positive control (for protein inputs, see Supplemental Fig. S3B). The Fe concentration was normalized to the input protein concentration to obtain the number of Fe molecules per protein molecule.

Next, we tested the ability of the JMJ24-JmjC domain to bind Fe directly. JMJ24-JmjC was expressed in Escherichia coli and purified (Supplemental Fig. S3B). As a control, we also expressed and purified JMJ18-JmjC, in which all important residues are conserved. JMJ18-JmjC was previously shown to be a functional histone H3K4me3/me2 demethylase (Yang et al., 2012). Using inductively coupled plasma optical emission spectrometry (ICP-OES), we analyzed the presence of Fe in GST, GST-JMJ24-JmjC, and GST-JMJ18-JmjC (Fig. 2C; Supplemental Table S1). GST-JMJ18-JmjC had a near equimolar Fe content (70%) consistent with the binding of one Fe²⁺ per JmjC molecule. For GST-JMJ24-JmjC and GST alone, only background levels were observed (17–18%). This is consistent with the notion that JMJ24 is unable to bind Fe²⁺ and is therefore not an active histone demethylase. It is also consistent with a previous report that did not find demethylase activity for JMJ24 using an in vitro assay (Deng et al., 2015).

Nuclear JMJ24 Shows E3 Ubiquitin Ligase Activity

We next sought to identify alternative molecular functions of JMJ24 by investigating the potential E3 ubiquitin ligase activity of the two RING finger domains. As E3 ubiquitin ligases are known to bind to E2 enzymes promiscuously (Kraft et al., 2005), we first tested whether JMJ24 was able to interact with the AtUBC10 E2 enzyme in the yeast two-hybrid system. Using β-galactosidase activity as a readout, we observed interaction of AtUBC10 with the full-length JMJ24 protein (amino acids 6–945) or with truncated versions of JMJ24 containing an N-terminal fragment (amino acids 6–432), both RING domains (amino acids 205–432), or only RING1 (amino acids 205–281; Supplemental Fig. S4). No interaction was found when JMJ24-RING2 (amino acids 337–441) was used as bait. Thus, JMJ24 interacts with AtUBC10 through the N-terminal RING1 domain. To test whether JMJ24 functions as an E3 ubiquitin ligase, we next performed an in vitro ubiquitination assay (Stone et al., 2005). The BB protein (Disch et al., 2006), which was used as a positive control, underwent polyubiquitination under our assay conditions (Fig. 3A). For GST-JMJ24-RING1, we observed a size shift of about 10 kD, consistent with monoubiquitination of the protein (Fig. 3A, lane 4, bottom panel). A band of the same size appeared in the corresponding lane in an immunoblot against the FLAG-tagged ubiquitin (Fig. 3A, top panel). A similar but weaker band shift was found for GST-JMJ24-RING2 (lane 8). Thus, JMJ24 is a functional E3 ubiquitin ligase with RING1 being the primarily active RING domain.

Figure 3.

JMJ24 is a nuclear protein with monoubiquitination activity. A, In vitro ubiquitin ligase assay of JMJ24 RING domains. GST-tagged JMJ24-RING1 (lane 4), JMJ24-RING2 (lane 8), and His₆-tagged BB (lane 13) undergo autoubiquitination as evidenced by the shifts in electrophoretic mobility (arrowheads) in the presence of E1 and E2. Immunoblots were processed with anti-FLAG antibodies (upper panel) to visualize FLAG-ubiquitin, or anti-GST and anti-His₆ (lower panel) antibodies, respectively. His₆-BB (Disch et al., 2006) served as a positive control and catalyzes the formation of high molecular weight ubiquitin chains. GST served as negative control. GST-tagged JMJ24-RING1 and RING2 catalyze the attachment of monoubiquitin (arrowheads). The intense band at 25 kD in lanes 4, 8, 10, 11, and 13 (arrow) likely represents an E2-ubiquitin adduct formed independently of E3 activity. B, Confocal microscopy images of JMJ24 subcellular localization. JMJ24::JMJ24-vYFP was stably transformed into jmj24-2 mutants and 3-d-old root tips were imaged. a, DAPI-stained nuclei (blue), b, JMJ24-vYFP signal (yellow), c, bright-field signal, and d, merged image of a, b, and c. Bar = 10 μm.

To test whether the predicted nuclear localization signal in form of a Trp-Arg-Cys motif (WRC-motif) near the N terminus is functional, we determined the subcellular localization of JMJ24-YFP. In Arabidopsis plants stably transformed with JMJ24::JMJ24-YFP, we observed nuclear fluorescence (Fig. 3B), as indicated by the overlap of YFP and 4′,6-diamino-phenylindole (DAPI) signals. This transgenic line complemented the jmj24-2 mutant phenotype, as evidenced by restoration of AtMu1c transcript levels (Supplemental Fig. S5). Taken together, our results indicate that JMJ24 is an E3 ubiquitin ligase localized to the nucleus.

JMJ24-JmjC Binds Histone H3

Even if JMJ24-JmjC is not an active demethylase, the domain may still function as a histone binding module. We therefore tested whether differently modified histone H3 peptides were able to interact with GST-JMJ24-JmjC purified from E. coli (Shi et al., 2006). As controls, we used GST-PHD-AtING1, for which a very high affinity for histone H3K4me3 was reported (Lee et al., 2009), and GST. GST-JMJ24-JmjC was precipitated by unmodified H3 amino acids 1 to 20 or H3 amino acids 1 to 20 that was mono-, di-, or trimethylated at either K4 or K9 (Fig. 4A). In contrast, GST-JMJ24-JmjC was not precipitated by unmodified H3 amino acids 21 to 44. GST alone was not precipitated by the tested H3 peptides (amino acids 21 to 44 or amino acids 1 to 20 K4me3). Thus, JMJ24-JmjC binds to the very N terminus of histone H3 irrespective of the methylation states of K4 and K9 in vitro. However, it cannot be excluded that in vivo the substrate specificity is further determined by other parts of JMJ24, by modifications of the protein that are missing in recombinant JMJ24-JmjC, or by interacting molecules, as was reported for other JMJ proteins (Hou and Yu, 2010).

Figure 4.

JMJ24 interacts with histones. A, JMJ24-JmjC binds to N-terminal histone H3 peptides (amino acids 1–20) in vitro. GST-tagged JMJ24-JmjC was immunoprecipitated with biotinylated histone peptides with different modifications. GST served as a negative and GST-tagged AtING1 (Lee et al., 2009) as a positive control for histone binding. no hist., control with no added histone peptides. B, In vivo JMJ24 protein interaction partners as isolated by immunoprecipitation and mass spectrometry (LC-MS/MS). Native JMJ24-vYFP protein complexes were immunoprecipitated from transgenic 35S::JMJ24-vYFP plants, and coprecipitated proteins were subsequently identified by LC-MS/MS (see also Supplemental Table S4). Two biological experiments with two technical replicates each were performed and the number of individual peptides and the cumulative score is shown for each experiment. C, JMJ24 specifically binds Histone H3 in vivo. JMJ24-YFP was purified from cross-linked nuclear protein extracts of transgenic 35S:JMJ24-YFP and nontransgenic Col-0 seedlings. Copurification of histone H2B and H3 was assessed by immunoblotting with specific antibodies.

To identify JMJ24-interacting proteins in vivo, we performed purification of native protein complexes followed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) using 7-d-old seedlings of a complementing 35S::JMJ24-vYFP transgenic Arabidopsis line in the jmj24-2 background (Supplemental Fig. S5). In two separate experiments with two technical replicates each, we identified JMJ24 peptides with a very high score. We also identified histone peptides of H2A, H2B, H3.1/H3.3, H3.3, and H4 with high scores (Fig. 4B). Neither of these peptides was found in a control experiment using Col-0. Among the H3 peptides, we found one peptide specific for H3.3 and several others that were redundant between H3.1 and H3.3 isoforms (Stroud et al., 2012; Shu et al., 2014).

We next investigated whether the binding of JMJ24 to histone H2B was likely direct or via H3. To this end, we repeated the coimmunoprecipitation of JMJ24-YFP with cross-linking under mild conditions to preserve protein-protein interactions. The coprecipitated proteins were analyzed by immunoblotting for the presence of H3 and H2B. We were able to precipitate H3 but not H2B with JMJ24-YFP, indicating that JMJ24 binds H3, but not H2B (Fig. 4C). In summary, these results suggest that full-length JMJ24 interacts with nucleosomes in vivo via binding to H3.

Genetic Interaction of JMJ24 with Histone Methyltransferases at AtMu1c

TEs are silenced through a number of silencing pathways involving RdDM and histone H3 K9 methylation. In order to begin to place JMJ24 within the existing silencing pathways, we analyzed the interaction of JMJ24 with the histone H3K9 methyltransferases KYP, SUVH5, and SUVH6 (Jackson et al., 2002; Ebbs et al., 2005; Ebbs and Bender, 2006). To this end, we crossed T-DNA insertion mutants of these genes with jmj24-2 and with 35S::JMJ24, respectively. In kyp single mutants, we observed a loss of AtMu1c silencing using the kyp-7 and kyp-4 alleles (Fig. 5). In kyp-7 jmj24-2 and kyp-4 jmj24-2, the silencing of AtMu1c was partially restored. Conversely, in kyp-7 35S::JMJ24 and kyp-4 35S::JMJ24, AtMu1c reactivation was stronger than in kyp single mutants or JMJ24 overexpressor lines. Thus, JMJ24 still functions in a kyp mutant background. This is consistent with JMJ24 acting at least in part independently of KYP. It was reported previously that SUVH5 and SUVH6 act partially redundantly with KYP (Ebbs and Bender, 2006). In suvh5 suvh6 double mutants, AtMu1c silencing was not affected (Fig. 5B); however, in the kyp-4 suvh5 suvh6 triple mutant, AtMu1c transcript levels were massively induced, indicating that H3K9me2 is an important component of AtMu1c silencing. Interestingly, modulation of JMJ24 activity by mutation or overexpression in the kyp-4 suvh5 suvh6 background did not have an effect. Surprisingly, the same was true for the suvh5 suvh6 double mutant background, although AtMu1c levels in this background were overall low. These findings suggest that JMJ24 requires SUVH5 and SUVH6 for its action. Together, these results are consistent with the idea that JMJ24 opposes H3K9me2 by antagonizing the H3K9 histone methyltransferases SUVH5 and SUVH6 (and possibly KYP).

Figure 5.

Genetic interaction of JMJ24 with KYP, SUVH5, and SUVH6. Relative AtMu1c transcript levels as determined by qRT-PCR in seedlings with modified JMJ24 activity and loss of KYP activity (kyp-7; A) or loss of KYP (kyp-4), SUVH5, and SUVH6 activities (B), respectively. Expression values of AtMu1c were normalized to TUB6 and Col. A, Results of one representative experiment are shown. Error bars indicate se of three technical replicates. A second experiment is shown in Supplemental Figure S6. B, Average of three biological replicates is shown. Error bars represent se. Statistical analysis was performed using ANOVA with log₂-transformed data, followed by Tukey’s test. Different letters indicate significant differences.

JMJ24 Represses H3K9me2 and CHG Methylation at AtMu1c

To further test this idea, we analyzed H3K9me2 levels at the AtMu1c locus by chromatin immunoprecipitation (ChIP). In jmj24-2, there was a slight increase in H3K9me2 that tested not significant (Fig. 6A). Conversely, in the JMJ24 overexpressor, we observed a significant reduction in H3K9me2 levels. In kyp-7, the overall levels were much reduced, but a similar effect of JMJ24 activity was observed, with increased levels in jmj24-2 kyp-7 and reduced levels in 35S::JMJ24 kyp-7 compared to kyp-7 that tested significant for the comparison between jmj24-2 kyp-7 and 35S::JMJ24 kyp-7 (Fig. 6A). Thus, JMJ24 antagonizes H3K9me2 levels at AtMu1c.

Figure 6.

Effect of loss of JMJ24 on H3K9me2 and DNA methylation. A, Histone H3K9me2 levels relative to global H3 at the AtMu1c transposase region as determined by ChIP in seedlings with modified JMJ24 and KYP activities. Overexpression of JMJ24 reduces H3K9me2, while loss of H3K9me2 enhances it. Data from three biological replicates were normalized to input and H3, and averaged. Error bars represent se. *P < 0.05 and **P < 0.005 (Student’s t test). Data for control amplicons with previously described (Mathieu et al., 2005; Kabelitz et al., 2014) high (CACTA) or low (ACTIN) H3K9me2 accumulation are shown in Supplemental Figure S6. B, DNA methylation analysis of the AtMu1c TIR by bisulfite sequencing in seedlings with modified JMJ24 and KYP activities. Percentages of DNA methylation were calculated from 20 independent clones each (compare with Supplemental Table S2).

As H3K9me2 acts in a positive feedback loop with CHG methylation, we next analyzed DNA methylation levels at the TIR of AtMu1c in different mutant backgrounds by bisulfite sequencing. We observed reduced overall methylation in kyp-7 (Fig. 6B, Total). In kyp-7 35S::JMJ24, CHG methylation was reduced compared to kyp-7 (Fig. 6B), consistent with the observed reduction in H3K9me2 (Fig. 6A). Total methylation levels in jmj24-2 were slightly increased (Fig. 6B). Thus, modulation of JMJ24 activity affected the level of DNA methylation at AtMu1c, possibly through its role in antagonizing H3K9me2.

Genetic Interaction with DNA Methyltransferases

We next investigated the genetic interaction between jmj24-2 and genes that affect CHH and CHG DNA methylation. To this end, we crossed jmj24-2 into the cmt3-11 and drm1-2 drm2-2 cmt3-11 backgrounds (Cao et al., 2003; Zhang et al., 2006), which both caused a strong loss of AtMu1c silencing, suggesting that AtMu1c silencing is mainly controlled by CMT3 (Fig. 7, A and B). Loss of jmj24-2 suppressed this release in silencing markedly, consistent with what we observed in the wild-type background and suggesting that JMJ24 and CMT3 largely act in parallel (Fig. 7A). However, jmj24-2 suppressed the effect of the drm1-2 drm2-2 cmt3-11 mutant much more strongly than it did with any other mutant tested (10-fold repression; Fig. 7B), indicating that upon loss of asymmetric DNA methylation JMJ24 activity becomes critical. This may reflect an increased sensitivity to changes in H3K9me2 in this background. Taken together, our results are consistent with a model where JMJ24 suppresses epigenetic silencing at AtMu1c by antagonizing histone H3K9me2, possibly by binding to histones and monoubiquitinating an as yet unknown target protein.

Figure 7.

Genetic interaction of JMJ24 with CMT3 and DRM1 DRM2 CMT3. Relative AtMu1c transcript levels in jmj24-2 double mutants with cmt3-11 (A) and drm1 drm2 cmt3 (B) as determined by qRT-PCR. Expression values of AtMu1c were normalized to TUB6 and Col. Averages of three biological replicates are shown. Error bars represent se.

JMJ24 Widely Affects Silencing of Transposons and Repetitive Elements

We next asked whether JMJ24 affected the expression/silencing of other TEs apart from AtMu1c. To this end, we performed transcriptome analysis using ATH1 microarrays comparing seedlings of three genotypes with increasing doses of JMJ24 (jmj24-2, Col, and 35S::JMJ24). To efficiently analyze the TEs contained on the microarray, we used a refined probe set annotation (Slotkin et al., 2009). To maximize the effects of modulated JMJ24 activity and minimize potential background effects, we compared transcript levels in the jmj24-2 mutant with those in an overexpressor line that was created in the jmj24-2 mutant background. Several categories of DNA transposons and one class of LTR retrotransposons (Gypsy AtGP1) were significantly affected (Fig. 8A, Table I). Among the affected DNA transposons, there were both MuDR (ARNOLD and VANDAL) and non-MuDR elements (CACTA). Interestingly, at the global scale, most TEs showed a release of silencing in the jmj24-2 mutant compared to the overexpressor line 35S::JMJ24. Thus, JMJ24 can have opposite effects on different TEs. This may depend on the chromosomal environment of the element and on the type and level of silencing present at the locus.

Figure 8.

JMJ24 affects TE silencing positively and negatively. A, Empirical cumulative distribution function (CDF) plots for DNA TEs, MuDR DNA TEs, non-MuDR DNA TEs, and CACTA DNA TEs (Slotkin et al., 2009). The distribution shows the log₂ fold change of 35S::JMJ24 against jmj24-2 based on ATH1 microarray data. A shift of the log₂ fold change distribution of the TE category specified in the header (magenta) compared to the log₂ fold change distribution of all TEs (blue) to the left indicates lower expression. B and C, Relative transcript levels of several TEs behaving similar to AtMu1c (B) or opposite (C) in jmj24-2 and 35S::JMJ24 as determined by qRT-PCR. Expression values were normalized to TUB6 and Col. Error bars are se of n biological replicates. *P < 0.05 and **P < 0.005 (Student’s t test).

Table I. JMJ24 widely affects TE silencing TE categories with changed transcript level in 35S::JMJ24 compared to jmj24-2. Transcript levels of jmj24-2 and 35S:JMJ24 were determined by ATH1 microarrays. TE annotation and categorization were adopted from Slotkin et al. (2009). P values are Benjamini-Hochberg (BH) corrected. n, Number of probes in this category. Arrows indicate the direction of transcript level difference in 35S::JMJ24 compared to jmj24-2.

TE Category	n	P (BH Corrected)
Retrotransposon.LTR.Gypsy.ATGP1I	10	0.0319	↓
DNA-Transposon	575	0.0000	↓
DNA-Transposon.MuDR	306	0.0052	↓
DNA-Transposon.MuDR.nonTIR	44	0.0001	↓
DNA-Transposon.MuDR.nonTIR. ARNOLD1	11	0.0271	↓
DNA-Transposon.nonMuDR	269	0.0000	↓
DNA-Transposon.nonMuDR.CACTA	85	0.0001	↓
DNA-Transposon.nonMuDR.CACTA. ATENSPM1	29	0.0041	↓
DNA-Transposon.nonMuDR.CACTA. ATENSPM1A	6	0.0361	↓

To further characterize the effect of JMJ24 at the level of individual TEs, we selected six TEs from the microarray analysis (VANDAL2 and IS112A; Eisen et al., 1994; Slotkin et al., 2009) and from previous studies (Bäurle et al., 2007; Zheng et al., 2009; Deleris et al., 2010) for expression analysis by qRT-PCR. The selected TEs fell into two classes (Fig. 8, B and C). The first class, containing IG/LINE, COPIA2, and MEA/ISR, displayed reduced transcript levels in jmj24-2 and increased transcript levels in 35S::JMJ24, reminiscent of what we observed for AtMu1c (Fig. 8B). Conversely, the second class, containing IS112A, a CACTA TE, and VANDAL2, displayed increased transcript levels in jmj24-2 and reduced transcript levels in 35S::JMJ24 (Fig. 8C). These findings corroborate our observation that JMJ24 can have apparently opposing outcomes both at the level of TE classes and at the single-element level. The transcriptome analysis suggests that globally the negative function of JMJ24 may prevail as the affected TE categories indicate an increased expression in jmj24-2. The different outcome may be caused by the context of the elements and the contribution of individual silencing pathways to each locus. Notably, transcript levels of members of various silencing pathways were not affected in jmj24-2 mutants. In summary, JMJ24 acts as an antisilencing factor at some loci, including AtMu1c, but as a silencing enhancer at others.

DISCUSSION

Here, we reported the characterization of the JmjC- and RING-domain-containing protein JMJ24 from Arabidopsis. We investigated the activity of the functional domains and the global role of JMJ24 in TE silencing during seedling development. The JmjC domain is conserved from yeast to humans and belongs to the superfamiliy of Fe²⁺-dependent dioxygenases (Hou and Yu, 2010). Using Fe²⁺ and α-ketoglutarate as cofactors and in the presence of oxygen, the methyl group of a methyl Lys is converted to hydroxymethyl, which is then released as formaldehyde. Both cofactors are absolutely essential for demethylase activity (Klose et al., 2006a). JMJ24 belongs to the KDM3/JHDM2 group, as does the closely related H3K9me2/me1 demethylase IBM1, indicating a specificity for H3K9me2/me1 (Inagaki et al., 2010). Five well characterized residues are required to bind the two cofactors. In JMJ24, the two amino acid residues required for α-ketoglutarate binding are conserved, but only one of three residues required for Fe²⁺ binding is present. Accordingly, no Fe binding was detected experimentally for JMJ24-JmjC. Thus, it is highly unlikely that JMJ24 is an active demethylase. Despite this apparent loss of functionality, JMJ24 is conserved during angiosperm evolution with clear orthologs in all dicotyledonous and monocotyledonous species examined. Interestingly, the two essential amino acids, which are required for Fe²⁺ binding and are substituted in JMJ24, are highly conserved among JMJ24 orthologs across monocotyledonous and dicotyledonous plants, suggesting that they are functionally important and under positive selection. This function may be to bind histones; its exact nature remains to be investigated. To our knowledge, JMJ24 is the first JmjC-domain protein from plants with an inactive JmjC domain that has been functionally characterized.

Interestingly, the founding member of the JMJ family, Jarid2, also lacks histone demethylase activity (Klose et al., 2006a; Landeira and Fisher, 2011). Jarid2 plays important roles in development and disease as a component of the PRC2 silencing complex. Jarid2 is required to target and assemble PRC2 onto target genes and may do so through association with noncoding RNA and methylation by PRC2 (Kaneko et al., 2014; Sanulli et al., 2015). Thus, there is precedent for the notion that the JmjC domain can degenerate, yet be maintained in functionally active proteins. An attractive hypothesis is that the JMJ24 JmjC domain has been retained as a histone H3 reader domain that binds modified or unmodified H3 in order to target H3 or associated proteins for ubiquitination. Our in vivo immunoprecipitation experiments confirmed that histone H3 is associated with JMJ24. It is a likely explanation that we recovered all four nucleosome-constituting histones in the more sensitive LC-MS/MS analysis because the conditions of purification were mild enough to leave nucleosomes intact. Our analyses do not support the notion that JMJ24 binds a particular H3 modification with high specificity. The in vitro binding studies indicated a broad binding of modified N-terminal H3 peptides. However, it is probable that other parts of the protein, unknown binding partners, combinatorial histone modifications, or modifications of JMJ24 itself affect the binding specificity in vivo. It is well established for other histone Lys demethylases, such as PHF8, KDM7A, and KDM4, that they have dual specificity depending on the chromatin context and the presence of other binding factors (Klose et al., 2006b; Horton et al., 2010; Lin et al., 2010; Liu et al., 2010b; Yang et al., 2010).

In contrast to the JmjC domain, the RING domains of JMJ24 have E3 ubiquitin ligase activity, as demonstrated in vitro using AtUBC10 as E2 enzyme. Interestingly, RING1 was much more active than RING2. Moreover, we only detected monoubiquitination in our experiments. So far, the target of JMJ24 E3 ligase activity remains elusive. One interesting possibility is that histone H3 or another histone protein is the target of this activity. Polyubiquitination is generally regarded as a signal for proteasomal degradation. In contrast, monoubiquitination is a stable protein modification. So far, H3 ubiquitination has only been reported for centromeric H3 variants in yeast and mammals (Hewawasam et al., 2010; Niikura et al., 2015). Monoubiquitination of H2A and H2B and their role in activating gene expression and PRC2 silencing, respectively, are well established (Weake and Workman, 2008). However, as we did not observe association with H2B (and H2A) and neither modification was associated with the regulation of TEs, they are unlikely target proteins. Alternatively, the target of ubiquitination may be a protein that is associated with histone H3 and histone H3 may serve to (negatively) regulate E3 ligase activity. RNA-DEPENDENT RNA POLYMERASE2 (RDR2) was recently reported to interact with JMJ24 (Deng et al., 2015). RDR2 was not found in our immunoprecipitation experiments; however, one possible scenario is that JMJ24 binds RDR2 transiently in order to ubiquitinate it. Future studies are required to test these possibilities. There is precedent for the combination of histone demethylation and E3 ligase activity. Recently, the animal histone demethylase LSD2 implicated in cancer cell growth has been found to have E3 ubiquitin ligase activity toward O-GlcNAc transferase (Yang et al., 2015). JMJ24 shares the RING domains with numerous KDM3 clade members (Zhou and Ma, 2008; Aiese Cigliano et al., 2013). Here, we demonstrated E3 ubiquitin ligase activity for the RING domain of a JMJ protein.

As described above, JMJ24 falls into the KDM3/JHDM2 clade of JmjC proteins, indicating specificity for H3K9me2/me1. In plants, H3K9me2 is associated with heterochromatin and silenced TEs. Although JMJ24 has lost demethylase activity, it may still recognize and antagonize H3K9me2. This hypothesis is confirmed by the H3K9me2 ChIP and the double mutant analyses with the H3K9 methyltransferases KYP, SUVH5, and SUVH6. Loss of JMJ24 activity correlated with enhanced H3K9me2 at AtMu1c, and overexpression of JMJ24 correlated with reduced H3K9me2. This was especially true in the kyp-7 background, where H3K9me2 levels were reduced but not abolished (because of redundancy with SUVH5 and SUVH6). When SUVH5 and SUVH6 were simultaneously deleted, JMJ24 overexpression did not have an effect. We found during this study that H3K9me2 strongly silences AtMu1c in the Col accession. The sensitivity of transcript analysis by qRT-PCR and accession-specific differences provide a plausible explanation for why previous studies did not observe such a strong reactivation in suvh mutants (Lippman et al., 2003; Ebbs et al., 2005). Our findings, together with the phylogenetic grouping of JMJ24, suggest that H3K9me2 may be the epigenetic mark that is targeted by JMJ24.

In summary, we propose a model where JMJ24 recognizes histone H3 through the inactivated JmjC domain and subsequently mono- or polyubiquitinates H3 or an associated protein (e.g. a histone H3 methyltransferase or reader protein), thereby marking it for degradation or inactivation (Fig. 9). Thus, JMJ24 antagonizes the silencing mark H3K9me2 independently of demethylation and promotes basal level transcription. JMJ24 represents an interesting new actor of the RdDM pathway. JMJ24 may play a role in TE reactivation during development or during the establishment of silencing of new TE insertions.

Figure 9.

Model for JMJ24 action during RdDM. JMJ24 encodes an inactive histone demethylase that associates with histone H3. At the AtMu1c locus, JMJ24 antagonizes repressive H3K9me2 marks, which are deposited by KYP and SUVH5/6 and promote DNA methylation. JMJ24 (genetically) antagonizes SUVH5 function, and the positive effect of JMJ24 on AtMu1c expression depends on the presence of SUVH5/6. An as yet unknown target protein is (mono)ubiquitinated by JMJ24 and thus inactivated. In the absence of JMJ24, this target protein promotes silencing of AtMu1c, likely through the methylation of H3K9 or cytosine. Thus, the JMJ24 pathway acts to sustain basal transcription of AtMu1c and other TEs during development or the establishment of silencing at novel TE insertions.

JMJ24 antagonizes RdDM-mediated epigenetic silencing of the DNA transposon AtMu1c and other TEs and thus formally acts as an antisilencing factor at these loci. IBM1/JMJ25 also acts as an antisilencing and boundary factor in Arabidopsis, acting to protect genic regions from the invasion of nearby heterochromatic marks and silencing (Saze et al., 2008; Inagaki et al., 2010; Rigal et al., 2012). Another antisilencing factor is the JMJ-domain protein Epe1 from Schizosaccharomyces pombe, which is required for heterochromatin boundary formation (Trewick et al., 2007; Tamaru, 2010). Neither of them targets TEs. So far, we can only speculate why plants may require an antisilencing factor directed toward TEs. One possibility is that this ensures low levels of transcription of a TE, which may be required to sustain silencing by RNA-mediated recruitment of the silencing machinery (Fultz et al., 2015). Alternatively, it may act as a preadaptation to extreme environmental conditions and allow rapid adaptation to changing environmental conditions (McClintock, 1984; Biémont and Vieira, 2006; Lisch, 2013). It is thought that extreme environmental conditions cause reactivation and transposition of TEs, which increases genetic variability in the population and may yield individuals with increased stress tolerance. Low-level transcription may facilitate reactivation during such extreme conditions.

Notably, some TE loci found to be regulated by JMJ24 in our microarray analysis were affected in the opposite direction as AtMu1c. A recent study investigating the transcriptome of jmj24-1 pollen by tiling arrays found that among 200 misregulated TEs, 80% were downregulated and 20% upregulated (Deng et al., 2015), thus confirming our observations in seedlings. Notably, in pollen, TEs are globally hyperactivated compared to seedlings (Slotkin et al., 2009), facilitating their experimental detection. This global reactivation together with the different mode of transcript detection (tiling array versus microarray) may explain why slightly differing observations were made in the two studies. It is possible that the apparent outcome of JMJ24 activity at individual loci depends on the locus-specific interplay of various silencing pathways, as well as the developmental stage. For example, it is conceivable that at loci (formally) repressed by JMJ24, the reduced transcription upon loss of JMJ24 causes the overall activity of the locus to fall below a certain threshold so that the locus is no longer targeted by another silencing pathway, thereby causing a net increase of transcript levels in jmj24-2. Consequently, increased activity of the TE locus upon JMJ24 overexpression may reinforce the activity of the other silencing pathway, resulting in a net decrease of expression. JMJ24 was found to be associated with the AtSN1, IG/LINE, and SDC loci (Deng et al., 2015) and may be associated with additional TEs investigated in this study. Identifying the target(s) of ubiquitination remains an important question for future studies investigating the function of JMJ24 in the RdDM pathway.

MATERIALS AND METHODS

Plant Materials and Growth Conditions

Plants were grown in long-day conditions on soil in a greenhouse or on GM plates (1% [w/v] Glc) at 23°C day/21°C night cycles for 10 to 14 d before analysis. All analyses were performed in the Col-0 background. The jmj24-2 (SALK_021260), kyp-7 (SALK_069326; Mathieu et al., 2007), cmt3-11 (N16392), and drm1-2 drm2-2 cmt3-11 (N16384; Henderson and Jacobsen, 2008) mutants were obtained from the European Arabidopsis Stock Centre. kyp-4 (SALK_044606), suvh5 suvh6 (SALK_074957 and SAIL_864_E08), and kyp-4 suvh5 suvh6 were obtained from J. Bender, Brown University (Ebbs and Bender, 2006).

Construction of Transgenic Lines

JMJ24 artificial microRNA, 35S::JMJ24, and JMJ24::JMJ24-vYFP lines were generated by Agrobacterium tumefaciens-mediated transformation. The JMJ24 artificial microRNA construct was designed according to Ossowski et al. (2008). A PCR product containing the artificial JMJ24 microRNA (oligonucleotides 297–300; Supplemental Table S3) was subcloned into pGEM-T Easy (Promega), sequenced, and transferred to 35S::pBarM (ML595) using BamHI and XhoI. For 35S::JMJ24 and 35S::JMJ24-vYFP, the genomic sequence of JMJ24 encoding the full-length protein was amplified with oligonucleotides 482 to 484, introducing BclI restriction sites. After sequencing, the JMJ24 fragment was transferred via BclI into 35S::pBarM (ML595) and 35S::pBarM-vYFP (IB30), respectively. For JMJ24::JMJ24-vYFP, the genomic JMJ24 sequence was amplified in two fragments. The JMJ24 5′ region (containing promoter, 5′ untranslated region, and the first half of the coding region) was amplified with oligonucleotides 417/418, introducing SphI and PstI restriction sites. The JMJ24 3′ region (containing the second half of the coding region and 3′ untranslated region) was amplified with oligonucleotides 419/420, introducing KpnI and PstI restriction sites. After sequencing, both fragments were assembled in pUC-ML939. Venus YFP was amplified with oligonucleotides 477/ 478, introducing SalI restriction sites, and assembled with JMJ24::JMJ24 in pUC-ML939. The whole cassette was then transferred via AscI into pBarMAP (ML516) (Adamski et al., 2009).

Gene Expression Analysis

RNA was extracted from 7- to 14-d-old seedlings using a hot-phenol RNA extraction protocol (Kabelitz et al., 2014). Total RNA was treated with TURBO DNAfree (Ambion), and 10 µg was reverse transcribed with SuperScript III (Invitrogen) according to the manufacturers’ instructions. cDNA was diluted 1:100 into qPCR reactions with GoTaq qPCR Master Mix (Promega), and a Roche LightCycler 480 instrument was used for measurements. Expression was normalized to TUBULIN6 using the comparative CT method. Primer sequences are listed in Supplemental Table S3. For microarray analysis, RNA from 9-d-old seedlings of three biological replicates per genotype was purified over an RNeasy Plant RNA extraction column (Qiagen) and processed for hybridization of Affymetrix ATH1 GeneChips (Atlas Biolabs). Arrays were further processed using the R/Bioconductor packages affy (Gautier et al., 2004) and limma (Ritchie et al., 2015) for rma normalization and differential expression analysis. Significantly affected transposon categories based on ATH1 probe set annotations by Slotkin et al. (2009) were identified using a Wilcoxon Rank Sum test. P values were Benjamini-Hochberg corrected. Empirical cumulative distribution function plots were done using the R package Lattice.

Recombinant Protein Expression, Ubiquitin Ligase Assay, and Fe Binding Analysis

The JMJ24 RING domains were amplified from cDNA with oligonucleotides 552/667 (RING1) and 668/669 (RING2), respectively, subcloned into pGEM-T Easy, sequenced, and transferred to pGEX-4T-1 (Amersham Biosciences). The JMJ24-JmjC domain was amplified from cDNA with oligonucleotides 1000/1001, subcloned into pGEM-T Easy, sequenced, and transferred to pGEX-4T-1. The JMJ18-JmjC domain was amplified from cDNA with oligonucleotides 1185/1186, subcloned into pGEM-T Easy, sequenced, and transferred into pGEX-5X-3. All recombinant proteins were expressed in Escherichia coli strain BL21 (DE3). BB and AtUBC10 E2 were previously described and were expressed as His₆-fusion proteins via the pQE system and purified with Ni-NTA affinity resins (Qiagen; Disch et al., 2006). The JMJ24 RING domains, JMJ24 and JMJ18-JmjC domains, were expressed as GST fusions via pGEX-4T-1 (Amersham Biosciences) and purified with a glutathione affinity resin (Pierce). The in vitro ubiquitin ligase assay was performed as previously described (Stone et al., 2005). Twenty-microliter reactions containing 50 mm Tris-HCl, pH 7.5, 10 mm MgCl₂, 0.05 mm ZnCl₂, 1 mm ATP, 0.2 mm DTT, 10 mm phosphocreatine, 2 units of creatine kinase (Sigma-Aldrich), 200 ng of yeast E1 (Sigma-Aldrich), 500 ng of purified E2 HIS₆-AtUBC10, 500 ng of purified GST-JMJ24-RING, GST, or HIS₆-BB, and 10 µg FLAG-ubiquitin (Sigma-Aldrich) were incubated at 30°C for 2 h. Reactions were stopped by adding 6 µL of SDS-PAGE sample buffer (125 mm Tris-HCl, pH 6.8, 20% [v/v] glycerin, 4% [w/v] SDS, and 10% [v/v] β-mercaptoethanol) and boiled for 5 min at 95°C. Reactions were analyzed by SDS-PAGE followed by immunoblotting using anti-FLAG (Sigma-Aldrich), anti-GST or anti-His₆ (both Novagen), and secondary IR800 (Novagen) antibodies, followed by detection with the LI-COR Odyssey system. For Fe binding analysis, 10 µmol protein was wet-ashed with nitric acid at 100°C overnight and measured by ICP-OES at 239.562 nm. The Fe concentration was normalized to the input protein concentration to obtain the relative number of Fe molecules per protein molecule.

DNA Methylation and ChIP Analysis

Bisulfite sequencing and ChIP analysis were performed as described (Kabelitz et al., 2014). For ChIP, chromatin was extracted from 7-d-old seedlings and sheared with a Bioruptor (Diagenode). ChIP was performed using anti-H3 (Abcam ab1791) or anti-H3K9me2 (Wako 302-32369) antibodies. Precipitated DNA was quantified by qPCR and normalized to input and H3.

Histone Peptide Binding Assay

The histone peptide binding assay was performed as described (Shi et al., 2006). In brief, 1 µg of biotinylated histone peptides (Millipore) was incubated with 10 µg of GST-fused protein in binding buffer (50 mm Tris-HCl, pH 7.5, 300 mm NaCl, 0.1% NP-40, and 1 mm PMSF plus protease inhibitors) overnight at 4°C with rotation. After 1 h incubation with Streptavidin Dynabeads (Invitrogen) and extensive washing with TBST, bound proteins were analyzed by SDS-PAGE and immunoblotting with anti-GST antibodies (Novagen).

Pull-Down and Mass Spectrometry Analysis

Two grams of 7-d-old seedlings was collected and flash-frozen in liquid nitrogen. Protein immunoprecipitation was performed using the μMACS GFP isolation kit (Miltenyi Biotec) and a published protocol (Smaczniak et al., 2012). Immunoprecipitated proteins were digested with trypsin (Trypsin Gold, Mass Spectrometry Grade; Promega), purified, and desalted using C18 columns (Teknokroma). The spectra acquired from the Easy-nLC coupled to a Q Exactive Plus (Thermo Fisher Scientific) were analyzed using MaxQuant protein quantification software (Cox and Mann, 2008). Immunoprecipitations from Col and 35S::YFP were used as controls for unspecifically bound proteins.

For immunoblotting of JMJ24 pull-downs, 2 g of 7-d-old seedlings was cross-linked with 0.5% formaldehyde solution for 10 min, blocked with Gly, and flash-frozen in liquid nitrogen. Protein immunoprecipitation was performed as described above. Precipitated proteins were analyzed by immunoblotting as described above and detected with antibodies against GFP (Abcam ab290), H3 (ab1791), and H2B (ab1790). Col-0 was used as a control to identify nonspecific binding.

Yeast Two-Hybrid Analysis

Yeast two-hybrid assays were performed with the DupLEX-A System (Origene; Golemis et al., 2008). Bait genes were cloned into pEG202 (lexA-DBD) and prey genes into pJG4-5 (B42-AD) vector. Yeast strain EGY48 (pSH18-34) was cotransformed by using the Frozen-EZ Yeast Transformation II Kit (Zymo Research). Three independent transformants per combination were tested on appropriate dropout medium with Gal, raffinose, and X-gal.

DAPI Staining and Confocal Images

Three-day-old seedlings were fixed in PBST with 3% glutaraldehyde for 6 h. DAPI staining was done overnight in PBST with 5% DMSO. After washing three times, pictures were taken using a Zeiss LSM 710 confocal microscope.

Accession Numbers

Sequence data from this article can be found in the GenBank/EMBL data libraries under the following accession numbers: JMJ24 (At1g09060), AtMu1c (At5g27345), JMJ18 (At1g30810), KYP/SUVH4 (At5g13960), SUVH5 (At2g35160), SUVH6 (At2g22740), DRM1 (At5g15380), DRM2 (At5g14620), CMT3 (At1g69770), IG/LINE (At5g27845), COPIA2 (At1g18930), MEA/ISR (At1g02580), IS112A (At1g43590), CACTA TE (At4g03745), and VANDAL2 (At2g12170). Microarray data files are available from the Gene Expression Omnibus database under accession number GSE72954.

Supplemental Data

The following supplemental materials are available.

• Supplemental Figure S1. JMJ24 gene structure and expression level in jmj24-2.

• Supplemental Figure S2. JMJ24 affects AtMu1 silencing at the level of transcription.

• Supplemental Figure S3. JMJ24 JmjC-domain alignment and recombinant protein expression.

• Supplemental Figure S4. The JMJ24-RING1 domain interacts with AtUBC10.

• Supplemental Figure S5. Complementation of transgenic lines used in this study.

• Supplemental Figure S6. Biological replicate for jmj24-2 kyp-7 double mutant analysis.

• Supplemental Figure S7. Control amplicons for the anti-H3K9me2 chromatin immunoprecipitation experiment shown in Figure 6A.

• Supplemental Table S1. Quantification of Fe analysis by ICP-OES.

• Supplemental Table S2. Details of AtMu1c TIR bisulfite sequencing (compare with Fig. 6B).

• Supplemental Table S3. List of primers used in this study.

• Supplemental Table S4. List of peptides identified in JMJ24 immunoprecipitation.

Glossary

TE

transposable element

RdDM

RNA-dependent DNA methylation

TIR

terminal inverted repeat

ICP-OES

inductively coupled plasma optical emission spectrometry

DAPI

4′,6-diamino-phenylindole

LC-MS/MS

liquid chromatography-tandem mass spectrometry

ChIP

chromatin immunoprecipitation