Histone acetylation recruits the SWR1 complex to regulate active DNA demethylation in Arabidopsis

Significance

Transgenes and some endogenous genes are protected from silencing by active DNA demethylation. How the demethylase ROS1 targets these specific genomic regions is poorly understood. Here, we show that H2A.Z and components of the conserved SWR1 complex are required for antisilencing. We found that ROS1 is recruited to some of its target regions by H2A.Z deposited by the SWR1 complex. Two bromodomain-containing proteins, AtMBD9 and NPX1, are critical for the recruitment of SWR1 complex to chromatin through recognition of histone acetylation marks created by the increased DNA methylation (IDM) complex, which is known to have a regulatory role in active DNA demethylation. Thus, we have identified a complete regulatory pathway initiated by the IDM complex in active DNA demethylation in Arabidopsis.

Abstract

Active DNA demethylation is critical for controlling the DNA methylomes in plants and mammals. However, little is known about how DNA demethylases are recruited to target loci, and the involvement of chromatin marks in this process. Here, we identify 2 components of the SWR1 chromatin-remodeling complex, PIE1 and ARP6, as required for ROS1-mediated DNA demethylation, and discover 2 SWR1-associated bromodomain-containing proteins, AtMBD9 and nuclear protein X1 (NPX1). AtMBD9 and NPX1 recognize histone acetylation marks established by increased DNA methylation 1 (IDM1), a known regulator of DNA demethylation, redundantly facilitating H2A.Z deposition at IDM1 target loci. We show that at some genomic regions, H2A.Z and DNA methylation marks coexist, and H2A.Z physically interacts with ROS1 to regulate DNA demethylation and antisilencing. Our results unveil a mechanism through which DNA demethylases can be recruited to specific target loci exhibiting particular histone marks, providing a conceptual framework to understand how chromatin marks regulate DNA demethylation.

As a conserved epigenetic mark, DNA methylation at the fifth position of cytosine plays an important role in plants and many other eukaryotic organisms (1, 2). A variety of biological processes, including the control of transposons and other repetitive DNA elements, generation of epialleles, genomic imprinting, genome interactions, and stress responses, are regulated by DNA methylation (2–5). In mammals, DNA methylation primarily occurs in the symmetric CG context, although non-CG methylation has been observed in brain tissues and embryonic stem cells (6–8). In plants, DNA methylation occurs in all 3 DNA sequence contexts: CG, CHG, and CHH (where H is C, A, or T) (9, 10). In Arabidopsis, the RNA-directed DNA methylation (RdDM) pathway is responsible for de novo methylation in all sequence contexts (2). CG methylation is maintained by DNA METHYLTRANSFERASE 1 (MET1) (11), while CHG methylation is maintained by the plant-specific DNA methyltransferase CHROMOMETHYLASE 3 (CMT3) (12). Asymmetric CHH methylation is maintained by DOMAIN REARRANGED METHYLTRANSFERASE 2 (DRM2) and CHROMOMETHYLASE 2 (CMT2) (13, 14). DRM2 is directed to specific genomic loci by the RdDM pathway, involving both 24-nucleotide small-interfering RNAs and long noncoding RNAs (4). DECREASE IN DNA METHYLATION1 (DDM1), a chromatin remodeling ATPase, facilitates CMT2 to catalyze RdDM-independent CHH methylation (14).

Genome-wide DNA methylation patterns are tightly controlled by the opposing actions of DNA methylation and demethylation pathways (2, 4, 15–18). How active DNA demethylation is regulated and how DNA demethylases are recruited to their target loci is poorly understood. In plants, active DNA demethylation is mediated by 5-methylcytosine DNA glycosylases through a DNA base-excision repair pathway (18–20). There are 4 known 5-methylcytosine DNA glycosylases in Arabidopsis: REPRESSOR OF SILENCING 1 (ROS1), DEMETER, and DEMETER-LIKE 2 and 3 (DML2 and DML3). DME is required for global DNA demethylation and gene imprinting in the endosperm (16, 21). ROS1 is necessary for active DNA demethylation in vegetative tissues at thousands of genomic regions and prevents transcriptional silencing of endogenous and transgenic loci (17, 20, 22, 23). DML2 and DML3 can excise 5-methylcytosine in vitro and mainly function redundantly with ROS1 in vivo (17). ROS1 is targeted to specific genomic loci through the increased DNA methylation (IDM) complex, which includes the methyl CpG-binding protein 7 (MBD7), IDM1, IDM2, IDM3, HDP1, and HDP2 proteins (23–29). The IDM complex catalyzes histone acetylation at a subset of DNA demethylation target loci. However, how histone acetylation in turn helps recruit DNA demethylases is unknown, since ROS1 does not physically interact with the IDM complex.

Histone modifications and histone variants play important roles in the regulation of gene expression, DNA repair, and a variety of other chromosomal processes (30), but the connection between histone modifications, histone variants, and active DNA demethylation remains elusive. The histone H2A variant H2A.Z is deposited into chromatin by the SWR1 chromatin-remodeling complex, which is conserved in eukaryotes (31–35). H2A.Z specifically localizes at euchromatin and has a genome-wide anticorrelation with DNA methylation (30, 36, 37).

In this study, we performed a forward genetic screen to identify cellular factors required for ROS1-mediated DNA demethylation and antisilencing. We isolated 2 SWR1 components, namely ARP6 and PIE1, and characterized the roles of the SWR1 complex and H2A.Z in regulating DNA demethylation. We discovered that the methyl-DNA-binding protein AtMBD9 and the yeast Bromodomain factor 1 (ScBDF1)-related protein nuclear protein X1 (NPX1) are associated with the plant SWR1 complex, and show that AtMBD9 and NPX1 function redundantly in active DNA demethylation by recognizing specific chromatin marks generated by IDM1. We have thus uncovered a complete regulatory pathway, initiated by the IDM complex, where the SWR1 complex, directed by AtMBD9/NPX1, functionally links the IDM-deposited histone marks to active DNA demethylation at specific loci in Arabidopsis through the physical recruitment of ROS1 by the SWR1-deposited H2A.Z.

Results

The SWR1 Subunits ARP6 and PIE1 Prevent DNA Hypermethylation and Gene Silencing.

Active DNA demethylation is critical for preventing DNA hypermethylation and transcriptional silencing of certain transgenes and endogenous genes (22–25, 38). To identify antisilencing mutants in Arabidopsis, we performed a forward genetic screen with transgenic 35S:SUC2 plants, which exhibit short roots when grown on medium containing sucrose compared with wild-type Col-0 plants (38) (SI Appendix, Fig. S1A). The 35S:SUC2 plants also express the hygromycin phosphotransferase II (HPTII) and neomycin phosphotransferase II (NPTII) transgenes, which confer resistance to hygromycin and kanamycin, respectively. We isolated 4 mutant alleles with long roots on sucrose-containing media: arp6-4, arp6-5, pie1-6, and pie1-7 (Fig. 1A and SI Appendix, Fig. S1 B and C). Compared with 35S:SUC2 plants, the mutants were sensitive to kanamycin and displayed reduced transgene expression (SI Appendix, Fig. S1 D and E). In addition, AT1G26380 and AT1G62760, 2 endogenous loci, which are hypermethylated and silenced in idm1-1 and ros1-4 mutant plants (23, 25), also displayed reduced expression in arp6-4, arp6-5, pie1-6, and pie1-7 mutants (Fig. 1B). The reduced expression of AT1G26380 and AT1G62760 in arp6-4 and pie1-6 was released by the DNA methylation inhibitor 5-Aza-2′-deoxycytidine (5-Aza) treatment (SI Appendix, Fig. S1F). The expression of ARP6-3xFLAG-3xHA or PIE1-3xFLAG-3xHA fusions driven by their native promoters restored expression of the 35S:SUC2 transgene and the short-root phenotype in arp6-4 and pie1-6, respectively (SI Appendix, Fig. S2 A–D). Furthermore, ARP6-3xFLAG-3xHA and PIE1-3xFLAG-3xHA were enriched at the CaMV35S transgene promoter by chromatin immunoprecipitation (ChIP) (SI Appendix, Fig. S2 E–G). Thus, ARP6 and PIE1 are associated with chromatin at the transgene promoters, and protect transgenes and endogenous genes from silencing.

Fig. 1.

ARP6, PIE1, and H2A.Z prevent transcriptional silencing and DNA hypermethylation. (A) arp6-4, arp6-5, pie1-6, and pie1-7 mutants display long roots on sucrose-containing medium, similar to idm1-9 and ros1-14 (Upper). Seedlings grown on glucose-containing medium show normal root development (Lower). (B) Relative expression of 2 endogenous ROS1-targeted genes, AT1G26380 and AT1G62760, in mutants compared with control 35S:SUC2 plants. (C) Screenshots of DNA cytosine methylation status, H2A.Z deposition, and H3K18Ac at the CaMV35S transgene promoter in 35S:SUC2 control plants or the indicated mutants. At the top is a schematic representation of the 35S:SUC2 transgene, indicating the transcription start site (TSS) and some regulatory elements (putative transcription factor binding sites and TATA box). The CaMV35S transgene promoter is divided into region A and region B according to Lang et al. (25). Scales on the tracks of DNA methylation level are from −1 to +1. Two independent biological replicates of ChIP-seq data are displayed in each case. Data for region A are the average of 5 copies (1 in 35S:SUC2, 2 in 2x35S:HPTII, and 2 in 2x35S:NPTII); region B is present in 35S:SUC2 only. (D) The H2A.Z triple mutants, h2a.z-2 and h2a.z-3, show a long-root phenotype on sucrose-containing medium, similar to pie1-6 mutant plants. (E) Relative expression of transgenes in the 35S:SUC2 control plants and the indicated mutants. Values are means ± SD of 3 biological replicates relative to transcript levels in 35S:SUC2 plants in B and E. **P < 0.01, compared with 35S:SUC2 samples (2-tailed t test).

To determine if transgene silencing in the arp6-4 and pie1-6 mutants is associated with DNA methylation, we examined the DNA methylation levels of the CaMV35S transgene promoter in 35S:SUC2, arp6-4 and pie1-6 mutants by bisulfite sequencing. DNA methylation levels increased in arp6-4 and pie1-6 mutants relative to 35S:SUC2 plants, similar to what is observed in the antisilencing mutants idm1-9 and ros1-14 (Fig. 1C and SI Appendix, Fig. S3). Furthermore, the DNA methylation inhibitor 5-Aza promoted the expression of SUC2, HPTII, and NPTII in 35S:SUC2 plants and restored kanamycin resistance and transgene expression in arp6 and pie1 mutants (SI Appendix, Fig. S1 D and E). Taken together, these findings suggest that PIE1 and ARP6 prevent transgene silencing by limiting DNA methylation.

H2A.Z Prevents DNA Hypermethylation and Gene Silencing.

ARP6 and PIE1 are components of the SWR1 complex, which deposits H2A.Z into chromatin (31–35). To investigate the role of H2A.Z in antisilencing, we knocked out the 3 expressed isoforms of H2A.Z (HTA8, HTA9, and HTA11) in 35S:SUC2 transgenic plants (39–41). We obtained 2 triple mutant lines, h2a.z-2 and h2a.z-3 (SI Appendix, Fig. S4 A and B), which showed developmental aberrations (SI Appendix, Fig. S4C). Interestingly, the h2a.z mutants developed long roots on sucrose-containing media (Fig. 1D), and the expression of the SUC2, NPTII, and HPTII transgenes and of the endogenous genes AT1G26380 and AT1G62760 was reduced in these mutants compared with 35S:SUC2 plants (Fig. 1 B and E). Moreover, DNA methylation levels at the 35S promoter were increased in h2a.z-2 compared with 35S:SUC2 plants (Fig. 1C and SI Appendix, Fig. S3), and treatment with 5-Aza restored the expression of SUC2 and HPTII transgenes in the mutants (SI Appendix, Fig. S4D). H2A.Z was enriched at the CaMV35S promoter in 35S:SUC2 plants, and this enrichment was decreased in pie1-6, pie1-7, arp6-4, and arp6-5 mutants (Fig. 1C and SI Appendix, Fig. S5). Thus, H2A.Z is required to protect both transgenes and endogenous loci from DNA hypermethylation and silencing.

H2A.Z Promotes ROS1- and IDM1-Dependent DNA Demethylation.

To understand the effect of H2A.Z on DNA methylation, we examined the genome-wide DNA methylation profiles in 35S:SUC2 plants and in the arp6-4, pie1-6, h2a.z-2, idm1-9, and ros1-14 mutants. The overall DNA methylation patterns in genes and transposons were not altered in arp6-4, pie1-6, or h2a.z-2 mutants, nor in idm1-9 or ros1-14 (SI Appendix, Fig. S6), consistent with observations that lacking H2A.Z does not substantially perturb genic DNA methylation (41). However, we identified some specific genomic regions that were hypermethylated in the mutants relative to 35S:SUC2 plants (Fig. 2A). These hypermethylated regions are distributed in genes, transposable elements (TEs), and intergenic regions (SI Appendix, Fig. S7 A and B). The overlap in hyper-differentially methylated regions (DMRs) in different mutants was limited, due to the strict statistical criteria used for hyper-DMR calling. There were 135 hypermethylated regions shared by h2a.z-2, idm1-9, and ros1-14 mutants (Fig. 2B), 140 hypermethylated regions shared by h2a.z-2, arp6-4, and pie1-6 (Fig. 2C), and 76 hypermethylated regions shared by ros1-14, idm1-9, h2a.z-2, arp6-4, and pie1-6 (Dataset S2).

Fig. 2.

ARP6, PIE1 and H2A.Z function in active DNA demethylation. (A) Number of DMRs identified in arp6-4, pie1-6, h2a.z-2, idm1-9, and ros1-14, and overlap of hyper-DMRs between them. The P values of tests at these overlaps are <1 − 10⁻⁴. Further information about mapping, coverage, depth, and conversion rate can be found in Dataset S1. (B) Overlap of hyper-DMRs among h2a.z-2, idm1-9, and ros1-14. (C) Overlap of hyper-DMRs among h2a.z-2, arp6-4, and pie1-6. (D–F) Heatmap representation of the methylation level at h2a.z-2 hyper-DMRs (D), idm1-9 hyper-DMRs (E), and ros1-14 hyper-DMRs (F) in 35S:SUC2, arp6-4, pie1-6, h2a.z-2, idm1-9, and ros1-14 mutants. Results from 6 (35S:SUC2 control lines) or 2 (mutants) independent biological replicates are shown.

The effects of ROS1, IDM1, H2A.Z, ARP6, and PIE1 in preventing DNA hypermethylation are not limited to the 76 genomic regions, since a shared increase in methylation could be observed in all of the mutants in regions that had not been classified as hyper-DMRs (SI Appendix, Fig. S7C). We calculated the average DNA methylation levels in all sequence contexts at h2a.z-2–, idm1-9–, and ros1-14–dependent hypermethylated regions, and found that DNA methylation levels in all contexts were also increased at these regions in all of the antisilencing mutants (Fig. 2 D–F and SI Appendix, Fig. S8 A–C). The DNA methylation levels that were increased at h2a.z-2 hyper-DMRs were also increased in the previously published h2a.z triple mutant (41) (SI Appendix, Fig. S8D). At idm1-9–, h2a.z-2–, and ros1-14–dependent hyper-DMRs that correspond to genes or TEs, methylation increases also occurred in all sequence contexts (SI Appendix, Fig. S9). These results suggest that H2A.Z deposition by the SWR1 complex promotes ROS1- and IDM1-dependent DNA demethylation at some endogenous genomic regions.

H2A.Z Physically Interacts with ROS1.

To investigate the mechanism of H2A.Z function in DNA demethylation, we isolated H2A.Z-associated proteins through IP of H2A.Z from 35S:SUC2 plants. In the immunoprecipitate, we identified components of the SWR1 complex, including PIE1, ARP6, and SWC6 (Fig. 3A). Interestingly, peptides corresponding to ROS1 were also detected as associated with H2A.Z, but not with H2A.W, another histone variant in the H2A family (SI Appendix, Fig. S10A). We then immunoprecipitated ROS1 from transgenic plants expressing a ROS1-3xFLAG-3xHA fusion driven by the ROS1 native promoter, and identified HTA9 and HTA8, 2 H2A.Z isoforms (Fig. 3B and SI Appendix, Fig. S10B). MET18, a protein known to interact with ROS1 (42), was also found in these experiments (Fig. 3B). We confirmed the interaction between ROS1 and H2A.Z isoforms using a yeast 2-hybrid (Y2H) assay (minus-3 media) (Fig. 3C) and a luciferase complementation imaging assay in Nicotiana benthamiana leaves (Fig. 3D). Thus, H2A.Z physically interacts with ROS1.

Fig. 3.

H2A.Z interacts with ROS1. (A) Selected proteins detected by LC-MS/MS following IP using an anti-HTA9 antibody, which immunopurifies H2A.Z from Col-0 plants. Proteins included in this table have been previously reported (PIE1 and ARP6) or confirmed as interactors of H2A.Z in Y2H or split luciferase assays in this work (ROS1). (B) Unique identified peptides detected by LC-MS/MS following IP using anti-FLAG antibody in ros1-14/pROS1:ROS1-3xFLAG-3xHA transgenic plants. Proteins included in this table have been confirmed as interactors of ROS1 by independent methods or those previously reported (42). Results obtained in 2 independent biological replicates are shown (Rep1-Rep2) in A and B. (C) ROS1 and H2A.Z interact in a Y2H assay. BD, GAL4 binding domain; AD, GAL4 activation domain. (D) ROS1 and H2A.Z interact in a split luciferase complementation assay. The indicated proteins were transiently expressed fused to the N-terminal or the C-terminal part of the luciferase protein (nLuc or cLuc, respectively) in N. benthamiana leaves.

IDM1 Is Required for H2A.Z Deposition at Some Active DNA Demethylation Target Regions.

IDM1 acetylates histone H3 at K14, K18, and K23, facilitating ROS1-dependent DNA demethylation through an as yet unknown mechanism (23). To shed light on the functional interaction between IDM1-dependent histone acetylation, H2A.Z deposition, and ROS1-mediated DNA demethylation, we analyzed the presence of these epigenetic marks both at the CaMV35S promoter and genome-wide in the 35S:SUC2 control plants and the idm1-9, arp6-4, pie1-6, and ros1-14 mutants. H2A.Z enrichment at the CaMV35S promoter was decreased in arp6-4, pie1-6, and idm1-9 mutants, but not in ros1-14 mutants compared with 35S:SUC2 plants (Fig. 1C and SI Appendix, Fig. S5A), suggesting that histone acetylation by IDM1 may be required for H2A.Z deposition, and that H2A.Z deposition may be upstream of ROS1-mediated DNA demethylation. As expected, the H3K18 acetylation level at the CaMV35S promoter was lower in idm1-9 than in the 35S:SUC2 control plants (Fig. 1C). H3K18 acetylation levels were also reduced in arp6-4 (Fig. 1C). In Arabidopsis, H2A.Z is highly enriched at +1 nucleosomes, while its levels are low at promoters of genes and TEs (Fig. 4 A and B) (36, 37, 41). Similar patterns were observed in the deposition of H3K18ac at genes and TEs genome-wide (SI Appendix, Fig. S11A). As observed for the CaMV35S promoter region, H2A.Z enrichment was decreased at both genes and TEs in the arp6-4 and pie1-6 mutants as well as in idm1-9, but not in ros1-14 (Fig. 4 A and B). The idm1-9 mutant displayed significantly decreased H2A.Z and H3K18ac deposition at regions identified as hypermethylated in idm1-9, h2a.z-2, and ros1-14 (Fig. 4C and SI Appendix, Fig. S11B). As expected, deposition of H3K18Ac was substantially reduced in the idm1-9 mutant at the 76 shared hyper-DMRs (Fig. 4D). H2A.Z deposition at the idm1-9–dependent hypermethylated regions was higher in ros1-14 than in idm1-9 (Fig. 4 C, Left). It has been previously shown that the idm1-dependent hypermethylated regions are a subset of ros1-dependent hypermethylated regions (23). Taken together, these results suggest that H2A.Z functions downstream of IDM1 and upstream of ROS1 at idm1-9 hyper-DMRs to prevent DNA hypermethylation and silencing. However, at some of the hyper-DMRs, particularly IDM1-independent genomic regions, the H2A.Z levels appeared lower also in ros1 mutant plants (Fig. 4C). The expression of ARP6 or PIE1 was not reduced in ros1-4 mutant plants (SI Appendix, Fig. S11C). These results indicate that ROS1-mediated DNA demethylation may somehow feedback-regulate the deposition of H2A.Z. The decrease in H3K18 acetylation levels in arp6-4 at the hyper-DMRs (SI Appendix, Fig. S11B) indicates a possible mutual reinforcement between SWR1-dependent H2A.Z deposition and IDM1-dependent histone acetylation. The above data led us to propose that the histone acetylation marks created by IDM1 may promote the deposition of H2A.Z at certain genomic regions, and that H2A.Z may help recruit ROS1 for active DNA demethylation at these loci. We found that 3xFLAG-3xHA-tagged ROS1 enrichment was reduced at several tested loci in h2a.z compared with 35S:SUC2 plants (SI Appendix, Fig. S12), supporting the notion that H2A.Z is required for ROS1 recruitment to specific DNA demethylation target loci.

Fig. 4.

H2A.Z is differentially deposited at hyper-DMRs in antisilencing mutants. (A and B) Average enrichment of H2A.Z in genes (A) or TEs (B) delimited by a TSS and a transcription termination site (TTS). Information about the raw pairs, concordantly mapped pairs and uniquely mapped pairs can be found in Dataset S3. (C) Metaplot and heatmap representations of the H2A.Z ChIP signal at the hyper-DMRs of idm1-9, h2a.z-2, and ros1-14 in 35S:SUC2 control plants and idm1-9, ros1-14, arp6-4, and pie1-6 mutants. (D) Metaplot and heatmap representations of the H3K18Ac ChIP signals at the hyper-DMRs shared by idm1-9, ros1-14, arp6-4, pie1-6, and h2a.z-2 mutants.

H2A.Z and DNA Methylation Marks Coexist at the Shared Hyper-DMRs.

In the 76 hyper-DMRs shared by ros1-14, idm1-9, arp6-4, pie1-6, and h2a.z-2, DNA methylation levels were substantially increased in all 3 sequence contexts (CG, CHG, and CHH) compared with the 35S:SUC2 control, while H2A.Z deposition was markedly decreased (Fig. 5 A and B). A strong genome-wide anticorrelation between DNA methylation and H2A.Z deposition has been previously described (37). Interestingly, as shown in Fig. 5 A and B, the 76 shared hyper-DMRs are an exception to this rule, simultaneously displaying moderate levels of DNA methylation and H2A.Z deposition in the 35S:SUC2 control plants. With the aim of determining if the coexistence of DNA methylation and H2A.Z marks is a distinctive feature of this subset of genomic regions, we performed a simulation analysis in which we replaced each of the 76 hyper-DMRs with another genomic region of the same length and with the same level of methylation (Fig. 5C), and examined the level of H2A.Z deposition at this group of surrogate loci in 35:SUC2 plants. Notably, H2A.Z deposition was much lower at the simulation regions selected by their methylation level (Fig. 5D). In another simulation analysis, we replaced each of the 76 hyper-DMRs with another genomic region of the same length and with the same level of H2A.Z deposition (Fig. 5D), and found that the DNA methylation level at the simulation regions was close to zero and there was no increase in any of the antisilencing mutants (Fig. 5E). Therefore, DNA methylation and H2A.Z deposition were anticorrelated at the simulation regions. These results support that the coexistence of DNA methylation and H2A.Z marks is a distinct feature of the shared hyper-DMRs, consistent with our model that the IDM complex is recruited to genomic regions with DNA methylation (23, 24), and that the histone acetylation marks generated by the IDM complex are important for the SWR1 complex to deposit H2A.Z at these regions.

Fig. 5.

Correlation beteween DNA demethylation and H2A.Z enrichement at hyper-DMRs shared by idm1-9, ros1-14, arp6-4, pie1-6, and h2a.z-2 mutants. (A) DNA methylation levels at the hyper-DMRs shared by idm1-9, ros1-14, arp6-4, pie1-6, and h2a.z-2 mutants in 35S:SUC2 and indicated mutants. (B) H2A.Z level at the hyper-DMRs in A in 35S:SUC2 and indicated mutants. (C) DNA methylation levels at simulated regions with same length and DNA methylation in 35S:SUC2 control plants (corresponding to the hyper-DMRs in A) in 35S:SUC2 and indicated mutants. (D) H2A.Z level at the shared hyper-DMRs in A, simulations with same H2A.Z and simulations with same DNA methylation in 35S:SUC2 control plants. Compared with the simulated regions with the same DNA methylation, H2A.Z was largely enriched at the common hyper-DMRs shared by idm1-9, ros1-14, arp6-4, pie1-6, and h2a.z-2 mutants in 35S:SUC2 control plants. (E) DNA methylation levels at simulated regions with same length and H2A.Z enrichment in 35S:SUC2 control plants (corresponding to the hyper-DMRs in A) in 35S:SUC2 and indicated mutants.

The Bromodomain Proteins AtMBD9 and NPX1 Are Associated with the SWR1 Complex and Function in Active Demethylation and Antisilencing.

Although our results suggest that IDM1 promotes H2A.Z deposition at the hypermethylated regions in antisilencing mutants (Fig. 4C), we did not detect a direct interaction between IDM1—or its acetylation products—and the SWR1 complex subunits ARP6 or PIE1. To further characterize the connection between IDM1 and the SWR1 complex, we immunoprecipitated PIE1 from plants expressing a PIE1-3xFLAG-3xHA fusion driven by the PIE1 native promoter and subjected it to LC-MS/MS, with the aim of identifying potential unidentified components that may mediate an interaction with IDM1. Using this approach, we isolated a number of proteins putatively associated with the Arabidopsis SWR1 complex (SI Appendix, Fig. S13A). While most of these proteins are homologs of the yeast SWR1 complex components (43), several had not been previously reported in Arabidopsis, such as NPX1 and AtMBD9. NPX1, a bromodomain-containing protein, is conserved in plants and has sequence similarity to ScBDF1 (SI Appendix, Fig. S13 B and C). AtMBD9 is a member of the MBD family and also contains a bromodomain, which can recognize acetylated lysines in histones (44, 45). Probing a histone peptide array with the N-terminal part of NPX1, which contains the bromodomain (amino acids 1 to 509), revealed that the NPX1 bromodomain binds histone H3 peptides acetylated at K14 and K18, as well as other histone acetylation marks, but not H3 peptides with the repressive H3K9me2 mark (SI Appendix, Fig. S14 A–C). Mutating a conserved residue in the acetylation binding pocket of the bromodomain (V185A) abolished its binding to acetylated histone H3 peptides (SI Appendix, Fig. S14 C and D).

To investigate whether NPX1 and AtMBD9 are involved in DNA demethylation, we generated mutant alleles in the 35S:SUC2 background (SI Appendix, Fig. S15A). Both the npx1-4 and atmbd9-4 mutants showed weak root phenotypes on sucrose-containing media, compared with 35S:SUC2 plants (Fig. 6A). The expression of SUC2, NPTII, and HPTII transgenes was also reduced in npx1-4 and atmbd9-4, but the reduction was less pronounced than in idm1-9 or ros1-14 (Fig. 6B). The atmbd9 npx1 double-mutant, however, developed long roots on sucrose-containing media, which were longer than those of npx1-4 or atmbd9-4 single-mutant plants and comparable to those of idm1-9 or ros1-14 (Fig. 6A), indicating that functional redundancy exists between AtMBD9 and NPX1. The expression of SUC2, HPTII, and NPTII was reduced in atmbd9 npx1 double-mutant plants, similar to idm1-9 and ros1-14 mutants (Fig. 6B). The expression of the IDM1- and ROS1-targeted endogenous genes AT1G26380 and AT1G62760 was also reduced in atmbd9 npx1 compared with 35S:SUC2 plants, as in arp6-4, pie1-6, h2a.z-2, idm1-9, and ros1-14 (Fig. 1B). The expression of NPX1-3xFLAG-3xHA and AtMBD9-3xFLAG fusions driven by their native promoters restored the short-root phenotype in atmbd9 npx1, while mutation of the conserved amino acid in the bromodomain of NPX1 abolished this restoration (SI Appendix, Fig. S15 B and C). DNA methylation levels increased at the CaMV35S transgene promoter in atmbd9 npx1 plants relative to 35S:SUC2 plants, similar to the antisilencing mutants arp6-4, pie1-6, h2a.z-2, idm1-9, and ros1-14 (Fig. 1C and SI Appendix, Fig. S3). Furthermore, 5-Aza restored kanamycin resistance in the atmbd9 npx1 double-mutant (SI Appendix, Fig. S15D). DNA methylation levels in all sequence contexts at h2a.z-2 and idm1-9 hyper-DMRs were increased also in atmbd9 npx1 (SI Appendix, Fig. S15E). Thus, NPX1 and AtMBD9 have a redundant role in antagonizing transcriptional silencing, and are necessary for protecting specific genomic regions from hypermethylation.

Fig. 6.

AtMBD9 and NPX1 antagonize transcriptional silencing redundantly by recruiting the SWR1 complex to chromatin through the recognition of acetylated histone marks. (A) Root phenotype of npx1-4, atmbd9-4, and atmbd9 npx1 mutants on sucrose-containing medium (Upper) or glucose-containing medium (Lower). (B) Relative expression of transgenes in the 35S:SUC2 control plants and the indicated mutants. Values are means ± SD of 3 biological replicates. *P < 0.05, **P < 0.01, compared with atmbd9 npx1 plants. NS, not significantly different compared with atmbd npx1 plants (2-tailed t test). (C) Biotinylated peptide pull-down assays showing in vitro binding of the recombinant AtMBD9 bromodomain to acetylated histone H3 peptides. (D) Examples of DNA methylation status, H2A.Z deposition, H3K18Ac signal, and nucleosome density at several selected shared hyper-DMRs in antisilencing mutants. Scales on the tracks of DNA methylation level are from −1 to +1. The highlighted region (in the square) indicates regions with differential DNA methylation, H2A.Z signal and H3K18Ac signal. For DNA methylation, 2 biological replicates in each mutant were performed and 1 representative result is shown.

The bromodomain of AtMBD9 binds H3K14Ac, H3K18Ac, and H3K23Ac peptides in a pull-down assay, and mutating a conserved residue (D1209A) within the bromodomain abolished this binding (Fig. 6C). IP-MS experiments using tagged AtMBD9 confirmed that AtMBD9 is associated with PIE1 and SWR1 components (SI Appendix, Fig. S16). In Y2H assays and luciferase complementation imaging in N. benthamiana leaves, we found that both NPX1 and AtMBD9 interact with PIE1 (SI Appendix, Fig. S17). These results indicate that AtMBD9 and NPX1 associate with the plant SWR1 complex and may function together as readers of histone acetylation marks.

Several examples of the shared hyper-DMRs in atmbd9 npx1 as well as in the ros1-14, idm1-9, arp6-4, pie1-6, and h2a.z-2 mutants are shown using integrative genomics viewer screenshots, which also display the ChIP-seq results for H2A.Z deposition, H3K18ac enrichment, and nucleosome density in these genomic regions targeted for active DNA demethylation (Fig. 6D and SI Appendix, Fig. S18). ChIP-qPCR assays confirmed that H2A.Z deposition at several tested DNA hypermethylated loci was reduced in atmbd9 npx1, as well as in idm1-9 (SI Appendix, Fig. S19).

Discussion

In this study, we showed that ARP6 and PIE1 are 2 cellular factors that inhibit DNA hypermethylation at specific genomic regions and prevent transcriptional gene silencing. In Arabidopsis, mutations in the SWR1 subunits PIE1 or ARP6 are known to cause similar developmental phenotypes as those seen in h2a.z mutants, including early flowering, reduction in plant size, curly leaves, shortened siliques, and reduced fertility (39, 41, 46–49), consistent with the role of the SWR1 complex in depositing H2A.Z. Our results suggest that the SWR1 complex, in addition to regulating plant development, also functions in active DNA demethylation. H2A.Z interacts with ROS1, and h2a.z mutant plants show DNA hypermethylation and gene-silencing phenotypes. Thus, the histone variant H2A.Z is important for active DNA demethylation at specific genomic regions.

Our results show that the Arabidopsis SWR1 complex associates with 2 bromodomain-containing proteins, NPX1 and AtMBD9, which recognize acetylated histone marks deposited by the IDM complex. The presence of double bromodomains in the yeast BDF1 highlights the significance of histone acetylation in H2A.Z deposition. In addition, the recruitment of the SWR1 complex to chromatin in yeast is regulated by the sequence-specific transcription factor Swc2/YL-1 (43). The MBD domain of AtMBD9 may be important for targeting the SWR1 complex to certain methylated genomic regions with histone acetylation marks to deposit H2A.Z. Chromatin modifiers are often associated with MBD proteins. In mammals, MBD2 is part of the histone deacetylation complex MeCP1 (50). In Arabidopsis, the chromatin remodeling protein DDM1 binds in vitro and colocalizes in vivo with AtMBD5, -6, and -7 (51). Our study on AtMBD9 and NPX1 provides insight into the mechanism of recruitment of the SWR1 complex, through recognition of both euchromatic (histone acetylation) and heterochromatic (DNA methylation) marks.

Previous work showed that the histone acetyltransferase IDM1 exists in a protein complex that specifically recognizes certain methylated genomic regions through the MBD domains of its subunits MBD7 and IDM1, and through the DNA binding domain of HDP2 (24). Our results here suggest that the bromodomain-containing proteins NPX1 and AtMBD9 recognize acetylated histone marks established by IDM1 and help attract the SWR1 complex to chromatin for H2A.Z deposition at the 35S:SUC2 transgene and specific endogenous genomic loci. In these particular regions, we propose that H2A.Z interacts with ROS1 and mediates its recruitment for active DNA demethylation to prevent hypermethylation and the spread of DNA methylation (Fig. 7). These specific genomic regions are characterized by a coexistence of DNA methylation and H2A.Z marks, and thus are an exception to the general rule of anticorrelation between H2A.Z deposition and DNA methylation (37). IDM1-dependent acetylation and SWR1-directed H2A.Z deposition could possibly act through a mutual reinforcement mechanism during active DNA demethylation since the H3K18 acetylation mark at the 35S:SUC2 transgene and at some endogenous ROS1 targets were reduced in the arp6-5 and atmbd9 npx1 mutants (Fig. 1C and SI Appendix, Fig. S11B).

Fig. 7.

Working model for the role of SWR1 and H2A.Z in active DNA demethylation. The methylated genomic target regions of active DNA demethylation are recognized by the MBD-containing proteins in the IDM complex (i). The histone H3K14Ac, H3K18Ac and H3K23Ac marks generated by IDM1 are recognized by the bromodomain proteins AtMBD9 and NPX1 in the SWR1 complex (ii). The SWR1 complex deposits H2A.Z to the targeted methylated regions. H2A.Z may also reinforce histone H3 acetylation (iii). H2A.Z interacts with and recruits ROS1 to initiate active DNA demethylation (iv). The model explains the behavior of particular genomic regions where moderate levels of DNA methylation and H2A.Z deposition coexist.

Since our model only applies to a subset of genomic regions demethylated by ROS1, additional mechanisms for directing ROS1 for locus-specific demethylation must exist. H2A.Z has broad genomic distribution and a wide range of functions (36, 37, 39–41, 52–54), many of which are clearly unrelated to active DNA demethylation, since the antisilencing/DNA demethylation mutants do not show any hypermethylation at the +1 nucleosomes, where H2A.Z is highly enriched (Fig. 4A and SI Appendix, Fig. S20), and incorporation or removal of H2A.Z can be carried out by specific histone chaperones in yeasts and metazoans (55, 56). ROS1 has domains that are important in the recognition of substrate DNA (57); this and perhaps other as yet unknown factors may function together with H2A.Z in recruiting ROS1 to the aforementioned particular genomic regions that break the rule of anticorrelation between DNA methylation and H2A.Z deposition.

Despite the growing body of knowledge concerning the nature and genome-wide landscape of diverse epigenetic modifications, the interplay and hierarchy between the various epigenetic modifications are not well understood. In particular, little is known about how active DNA demethylation may be regulated by chromatin marks. The results presented here offer an example of how chromatin marks can determine the DNA methylation status through the sequential actions of chromatin modifiers, shedding light on the coordinated control of gene expression, and providing a conceptual framework for the interplay between chromatin modifications and active DNA demethylation.

Experimental Procedures

Plant Materials, Mutant Screening, and Map-Based Cloning.

35S:SUC2 plants in this study refer to transgenic plants previously reported; the EMS-mutagenized library was generated and screened for mutants based on the long-root phenotype (38, 58). To map the mutant genes, mutants were crossed to wild-type plants of the Landsberg ecotype. Seedlings were grown vertically on 1/2 Murashige and Skoog (MS) plates with 1% sucrose with 20 μg/L Hygromycin and 1% agar and showed segregation of long and short root plants. We screened for mutants with a long-root phenotype among 7-d-old seedlings. Genetic mapping and gene cloning was performed as described previously (38). The T-DNA mutants arp6-1 (SAIL_599_G03), pie1-5 (SALK_096434), npx1-1 (SAIL_123_A10), npx1-2 (WiscDsLox422C12), atmbd9-2 (SALK_121881), and atmbd9-3 (SALK_039302C) were ordered from the Arabidopsis Biological Resource Center (https://www.arabidopsis.org) and genotyped by PCR. Plants were grown in long-day (16-h light/8-h dark) or short-day (8-h light/16-h dark) conditions at 22 °C.

DNA Methylation Inhibitor 5-Aza Treatment.

For tests of antibiotic resistance, plants were grown on glucose-containing 1/2 MS medium with 50 mg/L kanamycin. For 5-Aza treatments, both 50 mg/L kanamycin and 20 µM 5-Aza were added to 1% glucose-containing 1/2 MS medium. An equal volume of DMSO was added to the medium as control. The seedlings were photographed and harvested for RNA extraction after 14 d.

Mutant Plant Complementation.

For the complementation of mutants, ARP6, PIE1, or NPX1 genomic DNA with 2-kb upstream region (as the native promoter region) was amplified from genomic DNA of Col-0 with primers listed in SI Appendix, Table S1 and cloned into the pENTR/d-TOPO vector (Invitrogen). After sequencing, through LR recombination reaction, the genomic DNA was subcloned into the pEarly305 vector (with a 3xFLAG-3xHA tag at the C terminus) using LR clonase II (Invitrogen). AtMBD9 genomic DNA with 2-kb upstream region (as the native promoter region) was amplified from genomic DNA of Col-0 with primers listed in SI Appendix, Table S1. The full-length of genomic AtMBD9 DNA was divided into 2 fragments and the 2 fragments were digested with KpnI and SalI, and SalI and SbfI, respectively. The 2 fragments were cloned to pCambia 1305 vector (with a 3xFLAG tag at the C terminus) by T4 DNA ligase (New England Biolabs). Unsegregated T3 plants growing on 1% sucrose-containing medium and that were identified as homozygous complementation lines were then used for the additional experiments. The ros1-14/proROS1:ROS1-3xFLAG-3xHA transgenic plants used for IP and ChIP assays were generated in a previous publication (59). The ros1-14h2a.z/proROS1:ROS1-3xFLAG-3xHA transgenic plants were generated by crossing ros1-14/proROS1:ROS1-3xFLAG-3xHA with h2a.z mutant plants. The h2a.z mutant was described previously (41).

Generation of h2a.z Mutants in the 35S:SUC2 Transgene Background.

To generate the h2a.z mutant alleles using the CRISPR-Cas9 system, 3 20-bp sgRNA oligos targeting HTA8, HTA9, or HTA11, were inserted into the BbsI sites of 18T-AtU6-chim, 18T-AtU3b-chim, and 18T-At7SL-chim, respectively. To cascade the 3 single-guide RNA (sgRNAs) together, the pAtU6-sgRNA fragment was digested with HindIII and XhoI, the pAtU3b-sgRNA fragment with XhoI and XbaI and the pAt7SL-sgRNA fragment with XbaI and XmaI from their host vectors. These 3 fragments were inserted into the HindIII and XmaI sites of the psgRNA-Cas9-At to obtain the p3×sgRNA-Cas9-At and then p3×sgRNA-Cas9-At was subcloned into the pCAMBIA1300 binary vector (60). The construct was transformed into 35S:SUC2 plants using Agrobacterium tumefaciens GV3101 by the standard floral-dip method (61). Seeds from T4 plants contain homozygous h2a.z mutations and were genotyped by PCR or sequencing (JIELI BIOTEC).

Real-Time qPCR.

Total RNA was extracted from 0.1 g of 14-d-old Arabidopsis seedlings with the RNeasy plant kit (Qiagen). Two micrograms of mRNA was converted to cDNA with M-MuLV Reverse Transcriptase (New England Biolabs) and the cDNAs were used as templates for real-time PCR with iQ SYBR green supermix (Bio-Rad).

Whole-Genome Bisulfite Sequencing and Data Analysis.

Fourteen-day-old seedlings grown on 1/2 MS medium with 1% glucose and 0.8% agar were used for extraction of genomic DNA, according to the manufacturer’s instructions (Qiagen). Six replicates were performed in 35S:SUC2 plants. Two replicates were performed for h2a.z-2, idm1-9, ros1-14, arp6-4, pie1-6, and atmbd9npx1. Bisulfite conversion, library construction, and deep sequencing were performed by the Genomics Core Facility at the Shanghai Center for Plant Stress Biology, China.

For data analysis, data were trimmed using trimmomatic (62) with parameters “LEADING:20 TRAILING:20 SLIDINGWINDOW:4:15 MINLEN:50.” Clean reads were mapped to the Arabidopsis thaliana TAIR 10 genome using BSMAP with parameters “-m 0 -x 1000 -w 2”, and the python script of BSMAP was used to remove potential PCR duplicates and extract methylation ratios (63).

Identification of DMRs.

Identification of DMRs was performed as described previously (25) with some modifications. We had 3 batches of BS-Seq of 35S:SUC2 wild-type (each batch has 2 replicates). We combined 2 replicates of each sample. Only cytosines with a depth of at least 4 in the libraries were retained for further analysis. Differentially methylated cytosines (DMC) were identified if the P value from the 2-tailed Fisher’s Exact test was not higher than 0.05. We used a 200-bp sliding window with sliding step of 50 bp. DMC numbers were counted in each window. A region was selected as an anchor region if it had at least 2 DMCs. The actual boundary of each anchor region was then adjusted as the locations of first and last DMCs in the region. If the distance between 2 anchor regions was ≤100 bp, they were combined into a larger region. Only regions containing at least 10 DMCs were selected as candidate DMR. Each mutant was compared with each of the 3 35S:SUC2 wild-types and only the intersection of the 3 candidate DMRs set are reported as final DMRs. Further information about the identified hyper-DMRs can be found in Dataset S2.

ChIP Assay and Data Analysis.

The ChIP assay was performed as described previously (36) with few modifications. Five grams of seedlings were cross-linked with 1% formaldehyde in PBS buffer and ground in liquid nitrogen. Sonicated chromatin was incubated overnight with anti-H2A.Z (36), anti-H3K18ac (ab1191, Abcam), or anti-FLAG (F1804, Sigma). Immunoprecipitated DNA was treated with RNase A (Qiagen) and purified with the QIAquick purification kit (Qiagen). For individual locus detection, ChIP products were diluted with 80 μL of TE buffer, and inputs were diluted with TE buffer at the ratio of 1:100; 2 μL were used for each qPCR and 3 technical replicates were performed. Primers are listed in SI Appendix, Table S1.

For ChIP-seq, the DNA was sequenced at the Genomics Core Facility of the Shanghai Center for Plant Stress Biology, Shanghai, China. The quality of the sequencing data were first checked with FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). The first 8 bp of each read and reads after 86 bp (H2AZ ChIP)/90 bp (H3K18Ac) were trimmed. Clean paired-end reads were mapped to the TAIR10 A. thaliana genome using Bowtie2 (64) with parameters “–very-sensitive–no-unal–no-mixed–no-discordant -k 2” and only uniquely aligned reads were kept. Duplicated reads were removed using the SAMtools “rmdup” command (65).

The normalization method for H2A.Z ChIP-Seq was performed as previously described (66). Data were normalized so that the H2A.Z signal inside of background regions (defined as one-quarter of TEs with lowest ChIP signal) were the same. Briefly, TEs that do not overlap with genes were used as input for feature counts to count the ChIP fragment number in 35S:SUC2 wild-type. Then the TEs were sorted by value (=ChIP_fragment_number/TE_length) and the lowest 25% TEs were chosen as background regions. The fragments were pooled together in the background and the background signal in each genotype was calculated as signal = [(ChIP_background_rep1/ChIP_total_rep1+ChIP_background_rep2/ChIP_total_rep2)/2]/[Input_background/Input_total]. The normalization factors were calculated as norm_factor_mut = signal_mut/ signal_WT.

Histone Peptide Array and Pull-Down Assays.

The coding sequence of N-terminal NPX1 (1 to 509 aa), including the bromodomain, was amplified and cloned into the pET28a vector under the direction of the One Step Cloning Kit (Vazyme Biotech). The bromodomain mutation site was introduced by site-direct mutagenesis. The constructs were transformed into Escherichia coli BL21 for fusion protein expression. The His-tag fusion proteins were purified with Ni-NTA His Bind Resin (Novagen).

The coding sequence of the bromodomain of AtMBD9 was amplified using the primers in SI Appendix, Table S1 and cloned to pGEX-4T-1 vector. GST-tag fusion proteins were purified as indicated previously (38). The bromodomain mutation site was introduced by site-directed mutagenesis.

MODified Histone Peptide Array (Active Motif) was used to screen the specific histone marks for NXP1 binding. For pull-down assays, the biotinylated histone H3K9me2, H3K14Ac, H3K18Ac, or H3K23Ac peptides (EpiCypher) were incubated with 5 μg of His-Tag fusion or GST-Tag fusion proteins. The outputs of the incubation were detected by Western blot.

Y2H Assay.

In brief, the cDNA sequences were cloned into pGADT7-AD or pGBKT7-BD vectors and the pair of genes to be tested for interaction were cotransformed into the yeast strain AH109. Yeast cells expressed binding domain-ROS1 and activation domain-HTA2, -6, -8, -9, or -11. Y2H assays were performed as described previously (67).

IP and LC-MS/MS Analysis.

For IP, about 10 g of floral tissue for each epitope-tagged transgenic line were used. Dynabeads (10003D, Invitrogen) conjugated with FLAG antibody (F1804, Sigma), anti-H2A.W (36), and anti-H2A.Z (36), respectively, were applied for IP. Affinity purification was performed as described in a previous publication (68); the protein samples were subjected to LC-MS/MS analysis as detailed in a previous publication (25). For the IP samples of ROS1 or PIE1, to increase the number of detected peptides and the subsequent protein sequence coverage, the bound-bead proteins were first eluted by acid elution with 0.1 M glycine HCl at pH 3.5, and then the bound-bead proteins subsequently were digested on-bead by the addition of 100 µl of 100 mM NH₄HCO₃ containing 1 μg of Lys-C (Promega) overnight at 37 °C (69). All peptides were purified using StageTips before LC MS/MS analysis. We combined the 2 LC MS/MS results from products of glycine elution and the products of on-beads digestion together as the final IP result of PIE1 or ROS1.

Split Luciferase Complementation Assays.

Split luciferase complementation assays were performed using A. tumefaciens GV3101 carrying different constructs in 4-wk-old tobacco leaves. The coding sequences of the indicated proteins were cloned into pCAMBIA-cLUC and pCAMBIA-nLUC vectors. Luciferase activity was detected at 48 h postinfiltration. The detailed procedure is as described previously (25).

Data Access.

The MNase-Seq data are downloaded from a previously published paper (70); the Sequence Read Archive accession number is SRP120232. All high-throughput sequencing data generated in this study were submitted to the National Center for Biotechnology Information’s Gene Expression Omnibus (GSE115170).