Main

To investigate H2A.Z deposition in plant chromatin, we generated a high-resolution genome-wide map of H2A.Z in Arabidopsis by adapting the in vivo biotinylation system that we used to affinity purify Drosophila chromatin21. We tagged Arabidopsis H2A.Z with a peptide specifically recognized by the Escherichia coli biotin ligase BirA (biotin ligase recognition peptide, BLRP), and created transgenic plants coexpressing BLRP–H2A.Z with BirA. Cytological localization revealed that BLRP–H2A.Z has a diffuse nuclear distribution, but is excluded from heterochromatic chromocentres (Supplementary Fig. 1), the same pattern as that of endogenous H2A.Z17. After digestion with micrococcal nuclease to mostly mononucleosomes (Supplementary Fig. 1), we purified biotinylated chromatin from root tissue and co-hybridized the associated DNA with control DNA on tiling microarrays representing the entire Arabidopsis genome22. To ensure that our results were not influenced by potential tagging artefacts, we repeated the experiment with antibodies against endogenous H2A.Z17. We also mapped DNA methylation in roots (we have previously published a data set from aerial tissues22).

The maps generated by streptavidin pull-down and immunoprecipitation were virtually the same (Fig. 1 and Supplementary Fig. 2). The most notable feature was a strong, quantitative anticorrelation with DNA methylation (Pearson’s r = -0.81; Supplementary Tables 1 and 2). Distinct peaks of H2A.Z around the 5′ ends of genes were also evident (Fig. 1b). To visualize better the H2A.Z distribution, we aligned all Arabidopsis annotated sequences, which include genes, pseudogenes and transposable elements, at their 5′ ends, and stacked them from the top of chromosome 1 to the bottom of chromosome 5 (Fig. 2a and Supplementary Fig. 2). An obvious feature of this alignment is a vertical strip of high H2A.Z that roughly corresponds to the first nucleosome after the start of transcription. This pattern of H2A.Z deposition is consistent with those in yeast and humans10,11,12,13,14,15, indicating that this is a general feature of eukaryotic genes. There were also five conspicuous horizontal stripes of low H2A.Z incorporation. These correspond to transposon-rich, heavily methylated heterochromatin surrounding the five Arabidopsis centromeres. This pattern of incorporation is precisely the opposite to that of DNA methylation (Fig. 2b and Supplementary Fig. 2).

Figure 1: High-resolution maps of Arabidopsis H2A.Z and DNA methylation.
figure 1

a, H2A.Z (green) and DNA methylation (blue) profiles of Arabidopsis chromosome 2. Each vertical bar represents the log2 signal ratio of the test sample signal divided by the input control signal. The black circles denote the position of the centromeric sequence gap. b, c, More detailed views of a euchromatic (positions 547,000–587,000, b) and a heterochromatic (4,407,000–4,463,000, c) genomic region. DNA methylation from aerial tissues and roots is shown in blue; H2A.Z profiles obtained from two independent BLRP–H2A.Z transgenic lines and by means of immunoprecipitation of endogenous H2A.Z (Ab-1) are shown in green. Genes and transposons on the top and the bottom strands are shown above and below the line, respectively. 5′ peaks of H2A.Z in genes are emphasized by boxes in b. Unmethylated transposons with relatively high levels of H2A.Z are emphasized by boxes in c.

PowerPoint slide

Full size image
Figure 2: H2A.Z and DNA methylation are mutually exclusive.
figure 2

a, b, All annotated sequences from the Arabidopsis Information Resource release 7 (TAIR7, 31,762 sequences) were aligned at the 5′ end and stacked from the top of chromosome 1 to the bottom of chromosome 5. BLRP–H2A.Z is displayed as a heat map in a; root DNA methylation is displayed in b. Centromeric gap positions are indicated (cen1 to cen5). Note the high degree of anticorrelation between H2A.Z and methylation. c, d, Unmethylated transposable elements (listed in Supplementary Table 3). BLRP–H2A.Z is displayed as a heat map in c; root DNA methylation is displayed in d. e, All TAIR7 annotated sequences were k-means clustered (k = 3) on the basis of BLRP–H2A.Z patterns, and displayed as a heat map. For comparison, root DNA methylation of the same sequences is shown as a heat map in f.

PowerPoint slide

Full size image

Methylation is not distributed evenly within the genome. Transposons are heavily and uniformly methylated, some genes have short stretches of methylation, and most genes are unmethylated22,23,24,25,26. These three groups of sequences display a corresponding triphasic distribution of H2A.Z signal: low H2A.Z levels are found in transposons, intermediate levels in methylated genes, and high levels in unmethylated genes (Supplementary Fig. 3). One possibility is that the low levels of H2A.Z in transposons are caused by intrinsic sequence preferences, rather than DNA methylation. To test this, we examined the small number (49) of Arabidopsis transposons that are not methylated (Supplementary Table 3). Tellingly, all such transposons had high H2A.Z levels, indicating that low H2A.Z incorporation is not a feature of transposons per se (Figs 1c and 2c, d). Unmethylated transposons also lacked any discernible H2A.Z peaks, suggesting that these are unique features of endogenous genes. Unsupervised k-means clustering of annotated Arabidopsis sequences on the basis of H2A.Z patterns produced three groups that closely correspond to unmethylated genes, body-methylated genes and transposons (Fig. 2e, Supplementary Fig. 4 and Supplementary Table 4). Again, H2A.Z and DNA methylation levels showed a notable anticorrelation (Fig. 2f). DNA methylation and H2A.Z are thus mutually exclusive chromatin features, and our analyses show that this relationship is independent of sequence context, transcription or transcription potential (Supplementary Information and Supplementary Figs 5–11).

So far, our results indicate a strong anticorrelation between methylation and H2A.Z deposition, but we cannot distinguish which is causal. To address this issue, we took advantage of a line bearing a null mutation in the DNA methyltransferase MET1, met1-6 (refs 4 and 27). Mutations in MET1 cause major reductions in overall DNA methylation, and also significant hypermethylation mediated by other methyltransferases26. We reasoned that if DNA methylation influences H2A.Z deposition, changes in DNA methylation should be mirrored by changes in H2A.Z distribution. Notably, because met1 causes both losses and gains of DNA methylation, we should see both gains and losses of H2A.Z. To test our hypothesis, we mapped H2A.Z, as well as DNA methylation and transcription, in met1-6 plants.

Changes in DNA methylation indeed engendered changes in H2A.Z distribution (Fig. 3 and Supplementary Figs 12 and 13). To visualize these changes, we subtracted the wild-type H2A.Z data set from the met1 H2A.Z data set, so that high values represent increased H2A.Z incorporation in met1 (Supplementary Fig. 12). Examples of informative loci are shown in Fig. 3a–c. The FWA gene, which normally has 5′ methylation and lacks an H2A.Z peak, loses promoter methylation and gains 5′ H2A.Z in met1 (Fig. 3a). The retrotransposon At5g13205 is heavily methylated in wild type, but loses methylation and gains H2A.Z in met1 (Fig. 3b). Gene At1g22000, which encodes an F-box protein, is hypermethylated in met1, leading to loss of its 5′ H2A.Z peak (Fig. 3c).

Figure 3: H2A.Z incorporation changes in met1-6 mutant plants.
figure 3

a–c, Wild type (WT) root DNA methylation (dark blue), met1-6 root DNA methylation (purple), WT H2A.Z (antibody, green), WT H2A.Z profile subtracted from the met1-6 H2A.Z profile (two sets of independent paired experiments, light blue), and met1-6/WT transcription (red) for FWA in a, copia-like transposable element At5g13205 that loses methylation and gains H2A.Z in met1-6 in b, and F-box gene At1g22000 that is hypermethylated and loses H2A.Z in met1-6 in c. The 5′ region of FWA methylated in WT is emphasized by boxes in a. d, All TAIR7 annotated sequences were aligned at the 5′ end and stacked from the top of chromosome 1 to the bottom of chromosome 5. The WT H2A.Z pattern subtracted from the met1-6 H2A.Z pattern is displayed as a heat map. The same data after k-means clustering (k = 3) are shown in e. For comparison, root DNA methylation of sequences arranged as in e is shown as a heat map in f. g, WT methylation levels (left) and met1-6 methylation levels (right) for probes representing a significant decrease of H2A.Z in met1-6 (Supplementary Fig. 12). The histogram is cumulative for three independent methylation data sets. Grey histograms in the background show the signal distribution for all probes. h, Kernel density plot, which has the effect of tracing the frequency distribution, of all probes in the data set displayed in d (black trace), transposable elements upregulated in met1-6 (red trace), and transposable elements not upregulated in met1-6 (blue trace).

PowerPoint slide

Full size image

To get a comprehensive view of H2A.Z dynamics in met1-6, we aligned and arranged all annotated Arabidopsis sequences as in Fig. 2a. The same conspicuous pericentric stripes were evident in this profile (Fig. 3d and Supplementary Fig. 13)—H2A.Z levels are increased in transposable elements, which lose most of their methylation and become reactivated in met1 (refs 22 and 23). Unbiased sorting of the data produced three clusters that roughly encompass unmethylated genes, methylated genes and transposons, respectively (Fig. 3e, Supplementary Fig. 13 and Supplementary Table 4; sequences are categorized as in ref. 22). The changes in H2A.Z closely correspond to DNA methylation—sequences that gain H2A.Z in met1 are methylated in wild type (Fig. 3f and Supplementary Fig. 13). Conversely, loci with decreased H2A.Z incorporation are unmethylated in wild type, but methylated in met1-6 (Fig. 3g). Overall, changes in DNA methylation were mirrored by changes in H2A.Z in a manner that strongly argues that methylation inhibits H2A.Z incorporation.

Because some transposons and genes undergo transcriptional upregulation in met1 plants22, we had an opportunity to test whether H2A.Z incorporation is negatively influenced by methylation or positively influenced by transcription. Within genes, there is a robust correlation between DNA methylation in wild type and H2A.Z changes in met1-6 (average Pearson’s r = 0.51, Supplementary Table 2), but there is no correlation between transcriptional and H2A.Z changes (average Pearson’s r = 0.05). FWA, which is strongly overexpressed in met1, has reduced levels of H2A.Z in the body of the gene, where it has no methylation in wild type (Fig. 3a). Similarly, of the handful of transposons that are not methylated in wild type, two (At4g10690 and At5g35205) are nevertheless upregulated in met1 (Supplementary Fig. 14). Both also have less H2A.Z in met1 than in wild type, the opposite of other transposons.

Because only about half of all transposable elements are upregulated in met1, we could ask whether those elements preferentially gain H2A.Z, as would be expected if H2A.Z incorporation was associated with transcriptional activity. To ensure that the size of the data sets and methylation are not an issue, we compared 12,500 probes that represent activated transposons and silent transposons, respectively, and have identical methylation profiles. We found that both transposon classes are equally enriched in H2A.Z (Fig. 3h and Supplementary Fig. 15). Thus, changes in DNA methylation, rather than transcription, cause the redistribution of H2A.Z we observe in met1.

Our results show that DNA methylation excludes H2A.Z. An intriguing question is whether H2A.Z can also exclude methylation. Some of our data suggest that this is indeed the case. The most notable feature of H2A.Z incorporation, the 5′ genic peak, is independent of DNA methylation (Fig. 2e, f and Supplementary Figs 4 and 6), yet methylation is strongly excluded from this area22,23. Likewise, the higher H2A.Z levels in the bodies of less-transcribed genes (Supplementary Information and Supplementary Figs 6 and 7) might explain the puzzling observation that the chances of a gene becoming methylated increase with transcription (up to about the 70th percentile)22,23.

To address this issue, we mapped DNA methylation in plants with a strong loss-of-function allele of PIE1 (the conserved catalytic component of Swr1) that disrupts proper deposition of H2A.Z17. The overall methylation pattern in pie1-5 plants remained similar to that in wild type (Supplementary Table 5), but there was a modest but consistent increase in DNA methylation (Supplementary Fig. 16). To visualize the methylation changes in pie1, we subtracted the methylation patterns of matched wild-type controls (F2 siblings) from pie1 and displayed the resulting data as a heat map (Fig. 4a and Supplementary Fig. 16). This analysis revealed genome-wide hypermethylation of gene bodies. Using the ChIPOTle algorithm28, we identified 1,201 hypermethylated regions (corresponding to 1,172 genes) for further analysis (threshold P < 10-7, Supplementary Table 6).

Figure 4: H2A.Z protects from DNA methylation.
figure 4

a, All TAIR7 annotated sequences were aligned at the 5′ end and stacked from the top of chromosome 1 to the bottom of chromosome 5. The WT methylation pattern subtracted from the pie1 methylation pattern is displayed as a heat map. Shown are results from data set 1; all three data sets are shown in Supplementary Fig. 16. b, Bisulphite sequencing results for five loci. We sequenced 12 clones from each genotype, except for At1g69850 (10 clones in pie1) and At4g38190 (11 clones in pie1). c, Polymerase chain reaction (PCR) products from bisulphite-converted genomic DNA were digested with TaqI, which recognizes TCGA and will cut only if the C is unconverted (and therefore methylated). L, 100 bp ladder; TaqI, PCR product digested with TaqI; Unc, uncut PCR product. Note the greater digestion, which represents greater methylation, in pie1 compared to that in WT. d, All genes were aligned at the 5′ end and average scores for each 100-bp interval are plotted from 2 kb away from the gene (negative numbers) to 3 kb into the gene (positive numbers). The data were smoothed with a 5-point sliding window. The dashed line represents the point of alignment. The black line traces methylated genes; the green line traces unmethylated genes and the red line traces genes hypermethylated in pie1. e, Genes were grouped into percentiles on the basis of transcription levels. The red line traces the number of genes hypermethylated in pie1 within each percentile (left y axis). The black line traces DNA methylation enrichment (all genes) and the green line traces H2A.Z enrichment in unmethylated genes (right y axis). The data were smoothed with a 10-point sliding window. The scale of the right y axis was set to start at zero to enable comparison between methylation and H2A.Z.

PowerPoint slide

Full size image

In plants, DNA methylation can occur at any cytosine5. Most methylation is found in symmetric CG sites, like it is in animals, and is mediated by MET1, but there is also a substantial amount of methylation in other sequence contexts catalysed by other methyltransferases (hence the hypermethylation observed in met1)25,26. To determine how the pie1 mutation affects DNA methylation in different contexts, we used bisulphite sequencing to analyse the methylation of individual cytosines in five loci scored as hypermethylated by ChIPOTle: At1g69850 (a nitrate transporter), At3g22340 (a COPIA-like retrotransposon), At4g03480 (an ankyrin-repeat-containing protein), At4g38190 (a cellulose synthase) and At5g37450 (a protein kinase). All five showed a modest but consistent gain of CG methylation (Fig. 4b, c), confirming the microarray analysis. There was very little non-CG methylation at any of the loci in either wild type or pie1 (data not shown). Interestingly, all of the loci had some methylation in wild type, so the overall genomic hypermethylation we observed in pie1 is likely to be primarily caused by increased methylation of normally lightly methylated loci rather than de novo methylation of previously unmethylated loci.

Given the wide-spread hypermethylation caused by the pie1 mutation, we asked whether the hypermethylated loci are representative of the genome as a whole. As might be expected, pie1 hypermethylated genes have high levels of H2A.Z in wild type (that is, those generally found in unmethylated genes; Fig. 4d). They are also generally enriched in genes transcribed at a low level, with greatest enrichment around the 30th transcription percentile (Fig. 4e). This pattern is very different from that of normally methylated genes, which are most prevalent around the 70th percentile (Fig. 4e), and is also unlike unmethylated genes, which are enriched in both low and highly expressed genes22. pie1 hypermethylated genes do, however, closely parallel the overall distribution of H2A.Z (Fig. 4e). These loci also include 17 of the 49 transposons that are enriched in H2A.Z and unmethylated in wild type (Supplementary Tables 3 and 6), a 10-fold overrepresentation (P = 10-4, Fisher’s exact test). Thus, sequences that are generally preferred targets of DNA methylation (gene bodies and transposons) are hypermethylated in pie1, consistent with the presence of low levels of DNA methylation in these sequences in wild type (Fig. 4b, c). The high levels of H2A.Z found at these loci apparently protect them from developing full-blown DNA methylation, probably explaining the observed relationship between gene transcription and DNA methylation22.

How methylation silences genes has been a vexing question for decades. A popular model is that proteins that bind to methylated DNA engender silencing by recruiting histone deacetylases6. However, careful gene disruption studies in mice have shown that these proteins are unlikely to fully account for methylation-induced repression29,30. Previous work has provided strong evidence that H2A.Z contributes to promoter competence16,17,18,19. Therefore, exclusion of H2A.Z would represent a new mechanism of gene silencing by DNA methylation. H2A.Z incorporation, in turn, is likely to protect gene promoters from DNA methylation, contributing to gene activity and preventing silencing. Given that DNA methylation and H2A.Z are both ancient chromatin components, their interaction probably has an important general role in regulating eukaryotic gene expression.

Methods Summary

We adapted the biotin-mediated affinity purification system we developed in Drosophila tissue culture cells21 to allow protein purification from Arabidopsis plants. Biotinylated H2A.Z was purified largely as described21. Endogenous H2A.Z was immunopurified (IP) as described17, except the IP was performed in TNE (10 mM Tris, pH 8.0, 100 mM NaCl, 1 mM EDTA).

Our methylated DNA IP protocol (MeDIP), microarray design and labelling protocol are described in ref. 22. All labelled samples were sent to NimbleGen Systems for hybridization, except the pie1 samples, which were hybridized at the Fred Hutchinson Cancer Research Center DNA array facility. For bisulphite sequencing, 2 μg genomic DNA for each sample was bisulphite-converted with the Qiagen EpiTect kit.

Online Methods

Transgenic lines

We adapted the biotin-mediated affinity purification system that we developed in Drosophila tissue culture cells21 to allow protein purification from Arabidopsis plants. We constructed a binary plasmid that contained the E. coli biotin ligase, BirA, driven by the Arabidopsis ACT2 promoter, and the Arabidopsis H2A.Z gene At1g52740 driven by its endogenous promoter and tagged at the amino terminus with the BLRP. BLRP is a high affinity substrate for BirA, which biotinylates a lysine residue within the peptide. We sent the plasmid to the UC Riverside Plant Transformation Research Center (http://www.ptrc.ucr.edu), where transgenic Arabidopsis lines were created by vacuum infiltration in ecotype Columbia.

Affinity purification

About 100 seeds were sterilized in 20% bleach and 0.5% Tween-20 for 10 min. Seeds were germinated in 300 ml of Gamborg’s B-5 medium supplemented with 5 mM biotin, and roots harvested after four weeks. Four grams of roots were ground in liquid nitrogen to a fine powder, suspended in 20 ml of modified Honda buffer (25 mM Tris, pH 7.6, 0.44 M sucrose, 10 mM MgCl2, 2 mM spermine, 0.1% Triton X-100, 10 mM β-mercaptoethanol) and homogenized with a tissue homogenizer. The homogenate was filtered through Miracloth, transferred to a 30 ml round-bottom glass tube, and spun at 4,000 r.p.m. (2,000g) at 4 °C in an SS-34 rotor for 10 min. The pellet was resuspended in Honda buffer B (Honda buffer minus spermine), spun in a microcentrifuge at 1,500 r.p.m.(200g) at 4 °C for 2 min, and resuspended in 1 ml of TNE (10 mM Tris, pH 8.0, 100 mM NaCl, 1 mM EDTA). The suspension was warmed to 37 °C and digested with micrococcal nuclease in the presence of 4 mM CaCl2 (Supplementary Fig. 1) to liberate nucleosomes. The reaction was stopped with 25 mM EDTA and spun at high speed in a microcentrifuge for 5 min at 4 °C. Biotinylated proteins were purified from the supernatant as described21. Endogenous H2A.Z was immunopurified as described17, except the IP was performed in TNE. The antibodies are predicted to cross-react with all three Arabidopsis H2A.Z proteins17.

Microarray analysis

Our microarray design is described in ref. 22. We analysed DNA methylation in five independent samples from the Columbia ecotype: two from wild-type roots (WT root-1 and root-2), two from met1-6 roots (met1-6 root-1 and root-2), and one from met1-6 aerial tissues. We followed our protocol, as described in ref. 22, except we omitted the T7 RNA polymerase-mediated amplification step for all samples except aerial met1-6. Instead, sufficient amplification was achieved in the labelling step. We also used our wild-type aerial methylation data published in ref. 22.

For pie1 methylation analysis, we mapped methylation in three pie1 replicates and three matched wild-type controls (F2 siblings). DNA was extracted from tissue collected from >100 whole 12-day-old seedlings to eliminate the possibility of detecting random variations in DNA methylation24. The samples were amplified with the Sigma WGA2 kit before labelling. For ChIPOTle analysis28, outliers were removed from each data set by median smoothing (3-probe window), the three pie1 and wild-type data sets were averaged, wild type was subtracted from pie1, and the resulting data set was smoothed (triangular smoothing, y = 0.25(xn - 1) + 0.5(xn ) + 0.25(xn + 1)) and normalized to a mean of zero. We removed the 270-kb mitochondrial DNA insertion on chromosome 2 before analysis. 1,201 peaks were called with a conservative threshold of P < 10-7. As a control, we determined the number of ‘negative’ peaks that would represent hypomethylation: only 53 peaks were called. Even assuming the unlikely scenario that all the negative peaks are false positives, the false-positive rate would be 4%.

We assayed H2A.Z in eight samples: BLRP-1 and BLRP-2 were from one transgenic line, BLRP-3 and BLRP-4 from an independent transgenic line, wild-type Ab-1 and met1-6 Ab-1 were paired immunoprecipitation experiments from wild-type and met1-6 roots, respectively, and wild-type Ab-2 and met1-6 Ab-2 were a second set of paired experiments. All samples except BLRP-2, BLRP-3 and BLRP-4 were amplified by T7 RNA polymerase22; the rest were sufficiently amplified in the labelling step.

Expression analysis of two independent met1-6 RNA samples (paired with two independent wild-type samples) was carried out as described in ref. 22, except random hexamers were used for complementary DNA synthesis instead of an oligo d(T) primer. All labelled samples were sent to NimbleGen Systems for hybridization, except the pie1 samples, which were hybridized at the FHCRC DNA array facility.

Bisulphite sequencing

Two micrograms of genomic DNA for each sample were bisulphite-converted with the Qiagen EpiTect kit. PCR products were cloned with the Invitrogen PCR4 TOPO kit. Primer sequences are available on request.