Mapping and Functional Analysis of
DNA Methylation in Arabidopsis
Xiaoyu Zhang,1,5Junshi Yazaki,3,5Ambika Sundaresan,3,5Shawn Cokus,1,5Simon W.-L. Chan,1,6
Huaming Chen,4Ian R. Henderson,1Paul Shinn,4Matteo Pellegrini,1Steve E. Jacobsen,1,2,*
and Joseph R. Ecker3,4,*
1Department of Molecular, Cell and Developmental Biology
2Howard Hughes Medical Institute
University of California, Los Angeles, Los Angeles, CA 90095, USA
3Plant Biology Laboratory
4Genomic Analysis Laboratory
The Salk Institute for Biological Studies, La Jolla, CA 92037, USA
5These authors contributed equally to this work.
6Present address: Section of Plant Biology, University of California, Davis, Davis, CA 95616, USA.
*Contact: firstname.lastname@example.org (S.E.J.); email@example.com (J.R.E.)
Cytosine methylation is important for transpo-
son silencing and epigenetic regulation of en-
dogenous genes, although the extent to which
this DNA modification functions to regulate
the genome is still unknown. Here we report
the first comprehensive DNA methylation map
of an entire genome, at 35 base pair resolution,
using the flowering plant Arabidopsis thaliana
as a model. We find that pericentromeric het-
erochromatin, repetitive sequences, and re-
gions producing small interfering RNAs are
heavily methylated. Unexpectedly, over one-
third of expressed genes contain methylation
within transcribed regions, whereas only ?5%
of genes show methylation within promoter
scribed regions are highly expressed and
constitutively active, whereas promoter-meth-
ylated genes show a greater degree of tissue-
specific expression. Whole-genome tiling-array
transcriptional profiling of DNA methyltransfer-
ase null mutants identified hundreds of genes
and intergenic noncoding RNAs with altered
expression levels, many ofwhich may beepige-
netically controlled by DNA methylation.
Cytosine DNA methylation is a conserved epigenetic si-
lencing mechanism involved in many important biological
processes, including defense against transposon prolifer-
ation, control of genomic imprinting, and regulation of
gene expression (Bird, 2002; Goll and Bestor, 2005). In
mammals, the de novo methyltransferases DNMT3a/
b and the maintenance methyltransferase DNMT1 are re-
sponsible for DNA methylation, which occurs primarily at
CG dinucleotides (Goll and Bestor, 2005). In plants, the
DNMT1 homolog MET1 maintains CG methylation, while
the DNMT3a/b homologs DRM1/2 and the plant-specific
methyltransferase CMT3 are responsible for methylation
at non-CG sites (Chan et al., 2005).
DNA methylation is critically important for normal devel-
opment in both animals and plants; null mutations in
mouse DNMT1 or DNMT3a/b result in embryonic lethality,
and both met1 and drm1 drm2 cmt3 triple mutants exhibit
developmental abnormalities in Arabidopsis (Chan et al.,
2005, 2006; Goll and Bestor, 2005). An unbiased ge-
nome-wide identification and functional analysis of sites
of DNA methylation should greatly broaden our under-
standing of how DNA methylation and gene expression
are correlated on a global scale. However, although the
detection of DNA methylation at individual loci (e.g., by bi-
a high-throughput manner to explore the methylated frac-
tion of a complete eukaryotic genome remains technically
A number of strategies for high-throughput detection of
DNA methylation have been recently described and have
led to important discoveries. They can be broadly divided
into two types: sequencing-based and microarray-based.
Sequencing-based analyses have included the direct se-
purification (hundreds of clones) (Selker et al., 2003), PCR
amplification products from bisulfite-treated DNA (hun-
dreds of sequences) (Rakyan et al., 2004), and genomic
DNA fractionated by methylation-sensitive restriction-
enzyme digestion (thousands of clones) (Rollins et al.,
Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc. 1189
2006). Recent advances in sequencing technology have
made it possible to assay over 1,000 preselected CG sites
in hundreds of genes (Bibikova et al., 2006). However, the
cost and labor required make it difficult for sequencing-
based methods to achieve whole-genome coverage. On
the other hand, microarray-based approaches rely on
the separation of methylated and unmethylated DNA by
methods such as the differential digestion by methyla-
tion-sensitive restriction enzymes (Ching et al., 2005;
Hatada et al., 2006; Lippman et al., 2004; Tran et al.,
2005a, 2005b). As just one example, digestion by the en-
zyme McrBC (which cuts near methylcytosines) followed
by hybridization to a PCR amplicon tiling array was per-
formed to detect DNA methylation at 1 kb resolution in
the repeat-rich heterochromatic knob region of Arabidop-
sis (1.5 Mb, or ?1.2% of the genome) (Lippman et al.,
2004). However, these methods are limited by either se-
quence context (i.e., specific restriction-enzyme recogni-
tion sites) or the imprecise cutting of McrBC relative to the
location of methylcytosine. Alternatively, methylated DNA
can be enriched using immunoprecipitation and hybrid-
ized to microarrays to achieve a more unbiased detection
(Keshet et al., 2006; Weber et al., 2005). In this case, the
resolution and coverage are mostly determined by the mi-
croarray platform. For example, a BAC array has been re-
cently used to determine the DNA methylation profile of
the human genome at 80 kb resolution (Weber et al.,
Previous methylation studies in Arabidopsis have iden-
tified a number of methylated transposons and indicated
that the transposon-rich heterochromatic regions are
heavily methylated (Hirochika et al., 2000; Lippman
et al., 2004; Miura et al., 2001; Singer et al., 2001; The Ara-
ray study utilizing methylation-sensitive restriction en-
zymes identified CG methylation clusters (CG but not
non-CG methylation) found predominantly within genes
(Tran et al., 2005a). A similar approach was used to deter-
mine restriction sites that are differentially methylated in
several Arabidopsis DNA methylation mutants (Tran
et al., 2005b). However, the comprehensive identification
of methylated regions in Arabidopsis has not been per-
formed, and the extent and distribution of DNA methyla-
tion is unknown. Furthermore, despite the importance of
DNA methylation in regulating gene expression, very few
endogenous genes have been identified as being directly
controlled by DNA methylation (Chan et al., 2005).
Here we describe the first genome-wide analysis of
DNA methylation in the reference plant Arabidopsis thali-
ana. Using two biochemical methods in combination
with whole-genome tiling microarrays, we mapped the
methylated component of the Arabidopsis genome in
wild-type as wellas in themet1 and drm1drm2cmt3 triple
mutant plants at 35 bp resolution. In addition, RNA ex-
pression profiles of wild-type and mutant plants were de-
termined on both strands of the genome using the same
tiling-microarray platform. We found that genes methyl-
ated in their promoters tended to be expressed in a tis-
sue-specific manner, whereas genes methylated in their
coding regions were constitutively expressed at higher
levels. Analysis of the changes in DNA methylation and
the genome-wide identification of hundreds of genes as
well as many novel intergenic noncoding RNAs that ap-
pear to be regulated by DNA methylation. These results
represent the first genome-wide high-resolution mapping
of DNA methylation and the first systematic analysis of the
role of DNA methylation in regulating gene expression for
RESULTS AND DISCUSSION
Identification of Genomic Regions Containing
Methylated and unmethylated DNA was fractionated
by methylcytosine immunoprecipitation (mCIP) using
a monoclonal antibody that specifically recognizes meth-
ylcytosine (Keshet et al., 2006). The optimal antibody-to-
DNA ratio was determined by applying varying amounts
of antibody and assaying the amount of methylated DNA
recovered in the elution fraction by real-time PCR. We
found that 20 mg of antibody yielded maximal recovery
of methylated DNA from 2 mg of input genomic DNA; all
mCIP experiments described here were performed with
this ratio. Under these conditions, effective recovery of
methylated DNA was achieved for the FWA promoter re-
gion that contains extensive CG methylation (Figure 1A)
(Soppe et al., 2000). Unlike vertebrate genomes, plants
contain substantial amounts of non-CG DNA methylation.
Importantly, the elution fraction was also enriched for the
small transposon AtSN1 (primarily methylated at non-CG
sites), as well as for a region free of CG dinucleotides in
Ta3 with exclusively non-CG methylation (Figure 1A).
DNA from the unbound and elution fractions (depleted
and enriched for methylated DNA, respectively) was am-
plified and hybridized to whole-genome tiling microarrays.
We utilized a single-chip tiling microarray that covers
?97% of one strand of the ?120 Mb Arabidopsis genomic
sequence, with each of the 3.2 million 25 nt oligonucleo-
tide probes spaced within a 35 bp window (see the Sup-
plemental Data available with this article online).
Three independent statistical methods were used to
identify sites of DNA methylation from oligonucleotide
model (HMM) based on probe-level t statistics (Ji and
Wong, 2005), a Wilcoxon signed-rank test (Hollander and
Wolfe, 1999), and a nonparametric Kolmogorov-Smirnov
test developed in this study (see Supplemental Data). De-
spite utilizing different statistical approaches, all three
methods yielded similar results (Figure S1). The HMM
method was used for all subsequent analyses, as it pro-
videdaneasily interpretable bimodal distribution ofposte-
rior probability values and more defined boundaries be-
tween methylated and unmethylated regions (Figure S2).
To test the sensitivity of the mCIP-chip method, we
used an independent biochemical approach to separate
1190 Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc.
methylated and unmethylated DNA. DNA methylated at
CG dinucleotides was affinity purified using the methylcy-
tosine binding domain (MBD) from the human protein
MeCP2 (Cross et al., 1994) and hybridized to the same til-
ing array used for mCIP-chip (Figure 1B). The mCIP-chip
method identified ?95% of all regions detected by the
MBD-chip method and ?20% more genomic regions
(Figure 1C). This difference likely reflects the fact that
MBD binding requires a relatively high density of methyl-
cytosine in CG dinucleotides (Figure S3) as well as the
fact that only a subset of fractions eluted from the MBD
column were hybridized to microarrays.
Figure 1. Biochemical Methods to Fractionate Methylated and Unmethylated DNA and Comparison of Results
(A) The mCIP method. The relative amount of the unmethylated ACTIN-7 and the methylated FWA, AtSN1, and Ta3 in the unbound, wash, and bound
fractions as percentages of input was assayed by real-time PCR. Three biological replicates are shown.
(B) The MBD method. y axis = fold enrichment (upper half) or depletion (lower half) of FWA compared to ACTIN-7 in wild-type and in the met1 mutant
that lacks CG methylation.
(C) Overlap of methylated regions detected by the mCIP (gray circle) and MBD methods (ellipse).
(D) Methylation and RNA expression patterns of a typical euchromatic region (bp 439,000–468,500; left), a repeat-rich region from pericentromeric
heterochromatin (bp 4,827,000–4,849,500; middle), and the FWA gene (bp 13,038,000–13,043,000; right) on chromosome 4. White arrows indicate
the detection of two methylated CG sites near the 30end of FWA. A schematic representation of chromosome 4 is shown on top (open circle = cen-
Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc. 1191
Using the mCIP-chip data, we constructed and anno-
tated a set of methylation maps for each of the five chro-
mosomes. Methylated genomic regions identified with
this method were highly consistent with previous studies
that have characterized individual methylated genes
(Figure S4). Figure 1D shows results from three represen-
tative regions on chromosome 4, including a euchromatic
region, a repeat-rich pericentromeric heterochromatic re-
gion, and the euchromatic methylated gene FWA (Chan
of regions identified by mCIP-chip using genomic bisulfite
sequencing (Jacobsen et al., 2000). All mCIP-chip-posi-
tive regions were confirmed to contain methylcytosines
(in CG, CNG, or CNN contexts) (Figure S5). Regions previ-
ously known to be devoid of DNA methylation were also
categorized as unmethylated by the mCIP-chip method
(Figure S6). In addition, inclusion of the unmethylated
chloroplast genome (Ngernprasirtsiri et al., 1988) on the
microarray provided a large set of true negative probes
to systematically estimate the level of false discovery for
the mCIP-chip method and analysis processes. Within
the 154,478 bp chloroplast genome, only two regions rep-
resenting a mere 630 bp (0.4%) were detected as methyl-
ated. Thus, the mCIP-chip method and the analysis pro-
cedure yielded a low false-positive rate.
In the analyses of genome-wide DNA methylation pat-
terns presented in this study, a ‘‘methylated region’’ was
defined by combining adjacent probes with posterior
probabilities over 0.5, allowing a maximal gap of 200 bp,
and requiring a minimal run of 50 bp (see Supplemental
Data). We purposely chose this relatively stringent cutoff
to reduce falsely identified regions. Therefore, regions
where methylated cytosines are very sparsely distributed
may yield scores that are above background noise but do
not meet this stringent threshold. As one example, we dis-
covered that the last exon of the FWA gene contains two
methylated CG sites, which was detected as a peak with
a posterior probability of ?0.2 (Figure 1D). Thus, the anal-
ysis described below is likely a conservative estimate of
the amount of genomic methylation. However, the full dy-
namic range of methylation patterning can be visualized in
genome browsers (publicly available as described below)
for any particular genomic region.
DNA Methylation Landscape of the Arabidopsis
methylation, covering 22,554,840 bp and representing
?18.9% of the entire sequenced nuclear genome (The
Arabidopsis Genome Initiative, 2000). The chromosomal
distribution of methylated DNA is shown in Figure 2A as
the total length of methylated regions per 100 kb. As
expected, we found extensive DNA methylation in the het-
erochromatic regions of each of the five chromosomes,
including centromeres, pericentromeres, and the hetero-
chromatic knob on chromosome 4 (Figure 2A) (Hirochika
et al., 2001; The Arabidopsis Genome Initiative, 2000).
This distribution largely reflects the dense methylation of
transposons and other repetitive sequences that are clus-
tered in heterochromatin (Figures 2A and 2B).
A considerable amount of DNA methylation was also
found in euchromatic regions, including nonrepetitive in-
tergenic regions (?8% of total length) and genes. To per-
form a more detailed analysis of the level of DNA methyl-
ation in Arabidopsis protein-coding genes, we first
classified the 30,334 annotated genes into four cate-
gories: 14,948 ‘‘known expressed’’ genes (genes with ev-
idence of RNA expression and protein sequences with
known or predicted function), 10,475 ‘‘unknown ex-
pressed’’ genes (expressed genes with sequences show-
ing no homology to proteins of known or predicted func-
tion), 1,116 ‘‘nonexpressed genes’’ (computationally
predicted genes with no experimental evidence for ex-
pression), and 3,811 ‘‘pseudogenes’’ (annotated as pseu-
dogenes and very often containing homology to coding
sequences of transposons). While numerous methylated
genes were found in all four categories, the pseudogenes
and nonexpressed genes showed a much higher level of
methylation than known or unknown expressed genes
(Figure 2C), which likely reflects an enrichment for trans-
posons and other repeats within these two classes.
A High Level of DNA Methylation in Arabidopsis
Mammalian genes frequently contain small transposons
and repetitive DNA within their transcribed regions and
are commonly DNA methylated. Although the transcribed
regions of Arabidopsis genes are almost always transpo-
son-free, we found an unexpectedly high level of DNA
methylation in expressed genes (Figure 2C). Of the
25,423 expressed genes (excluding pseudogenes or non-
expressed genes), 15,627 (?61.5%) were entirely unme-
thylated (‘‘unmethylated’’), 1,331 (?5.2%) were methyl-
ated within their promoters (defined as the 200 bp region
ated’’). Interestingly, a surprisingly large number of genes
(8,465, ?33.3%) were methylated within transcribed re-
ated’’). Refined mapping of the position of DNA methyla-
tion within genes revealed a biased distribution toward
the 30half, whereas promoters and the immediate 30flank-
ing sequences of genes were hypomethylated (Figure 3A;
see Figure S9 for similar results from an alternative map-
ping method) (Tran et al., 2005a). This bias against pro-
moter and 30-end methylation was not found in pseudo-
genes and nonexpressed genes (Figure 3B), which
suggests that methylation at the 50and 30ends of ex-
pressed genes might be selected against, as methylation
may interfere with important activities such as transcrip-
tional initiation and termination.
DNA Methylation and Small RNAs
Small interfering RNAs (siRNAs) are known to cause RNA-
directed DNA methylation (RdDM) (Chan et al., 2004,
2005; Dalmay et al., 2000; Mette et al., 2000). We carried
1192 Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc.
out global comparisons between methylated genomic re-
gions and a large collection of Arabidopsis small RNA se-
quences determined by massively parallel signature se-
quencing (MPSS) (Lu et al., 2005). The majority of siRNA
clusters (i.e., endogenous loci corresponding to high local
concentrations of siRNAs) were heavily DNA methylated
(Figure 2D). However, ?63% of methylated regions were
amount of DNA methylation is maintained without persis-
tent targeting by siRNAs (Figure S7).
associated with siRNAs (?9%, compared to the genome
nance of genic methylation is largely independent of tar-
geting by siRNAs. This is consistent with the results from
gions contain MET1-dependent CG DNA methylation but
not siRNA-targeted non-CG methylation (Figure S5) and
also with the previous finding of CG methylation within
many Arabidopsis genes (Tran et al., 2005a).
Previous evidence suggested that microRNAs might
recruit DNA methylation enzymes to their target genes
(Bao et al., 2004). However, we found that annealing
sites in microRNA target genes were methylated at a level
slightly below the genome average (22 of 136, ?16.2%)
(see Figure S4 for PHB as an example). In addition, we
found that only one (MIR416a) of the 103 microRNA pre-
cursor genes was methylated. For trans-acting siRNAs
(tasiRNAs) (Allen et al., 2005; Peragine et al., 2004; Vaz-
quez et al., 2004), we found that 2 of the 5 tasiRNA-gen-
erating loci (TAS1b and TAS3) and 7 of 9 tasiRNA target
sites contained methylation. However, bisulfite sequenc-
ing of the methylated tasiRNA target sites in ARF3 re-
vealed CG methylation but an absence of non-CG meth-
ylation, which is a hallmark of RNA-directed DNA
methylation (Figure S8) (Chan et al., 2005). Furthermore,
DNA methylation persisted in the dcl2 dcl3 dcl4 triple
mutant that lacks detectable tasiRNAs (Henderson
et al., 2006). Overall, these results do not support a gen-
eral role for miRNAs or tasiRNAs in the active targeting of
Figure 2. DNA Methylation Landscape in the Arabidopsis Genome
(A) Chromosomal distribution of cytosine-methylated DNA. Top panels show the total length of repeats (y axis, left-side scale) and number of genes
(y axis, right-side scale) in a sliding 100 kb window. Bottom panels show the total length of methylated DNA in wild-type and in the met1 mutant in
100 kb windows. Arrows indicate the heterochromatic knob on chromosome 4.
(B) Percentage of interspersed, tandem, and inverted repeats that are methylated (green).
text for definition). y axis = length of methylated sequences as a percentage of the total length of DNA in each category.
(D) Methylation of siRNA clusters (defined in Lu et al., 2005).
Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc. 1193
Correlation between Genic DNA Methylation,
Expression Level, and Tissue Specificity
In order to examine the relationship between DNA methyl-
ation and gene expression patterns, we compared the
sites of DNA methylation with microarray expression
data from 79 different tissues or conditions (Schmid
et al., 2005). This analysis revealed a correlation between
the position of DNA methylation within a gene (i.e., pro-
moter or body), the level of gene expression, and its tissue
specificity. As shown in Figure 3D, the expression level of
body-methylated genes was significantly higher than that
of unmethylated genes, whereas that of promoter-methyl-
ated genes was generally lower. In addition, a distinct
higher tissue specificity (p < 10?37by a Kolmogorov-
Smirnov test) (Figure 3E). These results suggest that, in
general, body-methylated genes are constitutively ex-
pressed at a higher level, whereas promoter-methylated
genes tend to be expressed in a tissue-specific manner.
Comparison of the location of genic methylation with
gene functional categories showed that the promoter-
methylated group was highly enriched for genes involved
in proteolysis, whereas the body-methylated group was
enriched for catalytic enzymes (Table S1). Transcription
factors were the most enriched category within the group
of unmethylated genes (Table S1).
Alteration of DNA Methylation in the met1 and drm1
drm2 cmt3 Mutants
Using the mCIP-chip method, we analyzed genome-wide
DNA methylation patterns in two DNA methylation-defi-
cient mutant backgrounds (met1 and a drm1 drm2 cmt3
tion alleles (Chan et al., 2006). It has been previously
shown by bisulfite genomic sequencing that, at all tested
loci, the drm1 drm2 cmt3 triple mutant eliminates the
vast majority of non-CG methylation but not CG methyla-
tion, whereas the met1 mutant lacks virtually all CG
Figure 3. DNA Methylation of Arabidopsis Genes
(A) Distribution of DNA methylation within known genes and expressed genes with unknown functions. One kilobase regions upstream and down-
stream of each gene were divided into 50 bp intervals, each gene was divided into 20 intervals (5% each interval), and the percentage of genes
with methylation in each interval was graphed. A schematic representation of a gene is shown as a thick horizontal bar (scaled to 2.5 kb, the average
length of Arabidopsis genes).
(B) Distribution of DNA methylation within pseudogenes and nonexpressed genes.
(C) siRNA-independent methylation in different regions of genes. ‘‘Promoter, distal’’ and ‘‘promoter, proximal’’ refer to 200 bp–1 kb and 0–200 bp
upstream of the transcription initiation site, respectively. The known and unknown expressed gene categories are included in this analysis.
averaged over 79 tissues and conditions (Schmid et al., 2005); vertical bars indicate the bins used. y axis = fraction of genes with given intensity level.
(E) Tissue specificity of promoter-methylated, body-methylated, and unmethylated genes measured by entropy level (Schug et al., 2005). Low en-
tropy values indicate high tissue specificity. Vertical bars on x axis indicate the bins used. y axis = fraction of genes with the given entropy value.
1194 Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc.
methylation as well as a substantial amount of non-CG
methylation (Cao and Jacobsen, 2002; Tariq et al.,
2003). Genome-wide DNA methylation profiles detected
by mCIP-chip in drm1 drm2 cmt3 were roughly similar to
that of wild-type; ?93% of the regions methylated in
wild-type remained methylated in drm1 drm2 cmt3 (see
Figure 1D for examples). This suggests that the majority
of non-CG methylation colocalizes with CG methylation
and is consistent with previous findings that loss of non-
CG methylation often does not disturb the remaining CG
methylation (Cao and Jacobsen, 2002). Interestingly, the
?7% of regions that did lose methylation in drm1 drm2
cmt3 showed a lower CG dinucleotide content relative to
the genome average (Figure S3) and thus may represent
regions that rely on non-CG methylation for the overall
maintenance of methylation patterning.
In contrast, ?64% of the regions methylated in wild-
type were no longer detected in met1. The greatest reduc-
that genic methylation often occurs only at CG sites and is
less likely to be associated with siRNAs (Figure 3C and
Figure S5) (Tran et al., 2005a). The residual methylated re-
gions in met1 were highly clustered in heterochromatin,
consisted of mostly repetitive
showed a much higher level of association with siRNA
clusters, and had a significantly higher CNG content (Fig-
ure S10). It is therefore likely that, in met1, the residual
methylation is maintained by CMT3 at CNG sites as well
as by the siRNA-directed activity of DRM1/2 (Chan
et al., 2005).
Loss of DNA Methylation in met1 Results in Massive
Transcriptional Reactivation of Pseudogenes
To infer the function of DNA methylation in regulating gene
type as well as the met1 and drm1 drm2 cmt3 DNA meth-
yltransferase mutants using genome-wide tiling-array ex-
of sense and antisense transcription were determined
separately by generating strand-specific probes and us-
nome (the same array used for methylation analysis) and
a second similar array covering the reverse genomic
strands. Massive reactivation of previouslysilenced trans-
in drastically elevated transcriptional activity at the
(Figure 4A) (Lippman et al., 2004; Miuraet al., 2001;Singer
et al., 2001). In contrast, relatively few transposons or
pseudogenes were activated in drm1 drm2 cmt3, and
the transcriptional activity remained grossly similar to
that of wild-type at the chromosomal level (Figure S11).
These findings are consistent with our observation that
CG methylation was largely unchanged in the absence
of non-CG methylation and suggest that, in most cases,
lencing of transposons and pseudogenes.
Promoter-Methylated Genes Are Preferentially
Overexpressed in met1
The relative expression level of each gene in met1 com-
pared to wild-type was determined from the hybridization
intensity of probes located within exons and UTRs (?35
probes per gene on average; see Supplemental Data for
details). We then compared gene-level changes for pro-
moter-methylated, body-methylated, and unmethylated
genes within each category (Figure 4B). For the known,
unknown, and nonexpressed categories, a significantly
larger fraction of promoter-methylated genes were in-
creased in steady-state RNA levels in met1 compared to
body-methylated or unmethylated genes (p < 0.01 by
a Kolmogorov-Smirnov test). In contrast, expression of
body-methylated genes did not appear to be systemati-
cally increased when compared to unmethylated genes
on a global scale (Figure S12). For pseudogenes, the ex-
pression of both promoter- and body-methylated genes
was significantly increased (p < 10?5). The preferential in-
crease in expression of promoter-methylated genes in
met1 was more obvious when only the most significantly
upregulated genes were examined (Figure 4C). These
findings suggest that methylation in promoter regions
plays a more profound role in downregulating gene ex-
pression for expressed genes and that both promoter
and body methylation are important for silencing transpo-
sons and pseudogenes. Because of its more subtle ef-
fects on methylation patterning, we did not see bulk
changes in the expression of promoter-methylated genes
in drm1 drm2 cmt3 (Figure S11); however, the expression
of many specific genes was altered (see below).
Antisense Gene Expression in DNA Methylation
Previous studies have revealed the presence of antisense
expression corresponding to a large fraction of Arabidop-
sis genes (Yamada et al., 2003). Methylation in the body of
genes has been proposed to suppress antisense tran-
scription from cryptic promoters to ensure normal sense
tisense overexpression were found in met1 (e.g., Fig-
ure 5A), such examples were relatively rare. Furthermore,
we did not observe a systematic increase of antisense
transcription from body-methylated genes in met1, and
the change in antisense transcription in body-methylated
genes was not significantly different from that in unmethy-
lated genes (Figure 5B). Similarly, there was no significant
difference between methylated genic regions or unmethy-
lated control genic regions with respect to changes in the
abundance of antisense transcripts (Figure 5B).
We also investigated whether changes in sense and an-
tisense transcripts in met1 were correlated (e.g., whether
elevated levels of antisense RNA were generally corre-
lated with decreased accumulation of sense RNAs). As
shown in Figure 5C and Figure S13, changes in sense
Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc. 1195
and antisense expression were found to be largely inde-
pendent for both body-methylated and unmethylated
with several recent studies of antisense transcription (Fa-
ghihi and Wahlestedt, 2006; Lu et al., 2005; Wang et al.,
2005) and do not support a general role for DNA methyla-
tion within genes in suppressing antisense transcription or
for antisense transcription generally interfering with sense
Profile of Genes Affected by DNA Methylation
Comparison of the genes that were most significantly in-
creased in expression in met1 and drm1 drm2 cmt3 (listed
in Tables S2 and S3) revealed an interesting distinction.
Most genes that showed an increase in RNA abundance
in met1 were pseudogenes clustered in pericentromeric
heterochromatic regions. In contrast, known genes dis-
fraction (?69%) of upregulated genes in drm1 drm2 cmt3
(Figures 6A and 6B and Figure S14).
While it is difficult to determine the fraction of gene ex-
pression changes in drm1 drm2 cmt3 that can be attrib-
uted to secondary effects, in many cases, the loss of
DNA methylation of specific genes correlated with their
overexpression, suggesting a direct role of non-CG meth-
ylation in regulating these genes. One such example is the
F box protein encoded by the gene At2g17690. As shown
in Figure 6C, the promoter region of At2g17690 contains
a tandem repeat (seven copies of a 32-mer) that is heavily
methylated and also associated with siRNAs, and the
gene is normally not expressed (see RT-PCR results in
Figure 6B). In met1, an incomplete reduction of promoter
methylation was accompanied by a mild increase in
At2g17690 gene expression. In contrast, promoter meth-
ylation was virtually eliminated in drm1 drm2 cmt3, result-
gene (Figure 6B).
These results provide evidence that non-CG methyla-
tion is important in regulating gene expression on a ge-
nome-wide scale and identify a host of candidate genes
that may be directly regulated by non-CG methylation. A
tentially functional genes may residewith the mechanisms
by which these two types of methylation are inherited. In
Figure 4. Global Changes of Gene Expression in met1 Com-
pared to Wild-Type
(A) Chromosome-level RNA expression levels in wild-type and met1.
Chromosome 4 is shown as an example. y axis = median probe-level
hybridization intensity over 100 kb windows. A schematic representa-
tion of chromosome 4 is shown (bottom) and labeled as in Figure 1. Ar-
row indicates the increased expression level in the heterochromatic
knob on chromosome 4 in met1.
(B) Global changes of gene expression in met1 compared to wild-type,
shown as cumulative distributions. x axis = gene-level expression fold
change in met1 compared to wild-type (log10scale).
(C) Percentage of genes with increased expression in met1 (listed in
Table S2) within different gene categories.
1196 Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc.
particular, the replication-coupled-maintenance methyla-
tion activity of MET1 at CG sites might mainly function to
maintain the stable silencing of transposons and pseudo-
genes, whereas the siRNA-directed DRM activity and the
histone methylation-directed CMT3 activity (Chan et al.,
2005) are likely to be dynamic and may play more impor-
tant roles in regulating the expression of endogenous
genes. Consistent with this idea, profound differences ex-
ist in the siRNA populations accumulated in different tis-
sues (Lu et al., 2005), which could account for differential
targeting of genes by DNA methylation.
Intergenic Noncoding RNAs Controlled by DNA
Methylated regions account for ?8% of the total length
of nonrepetitive intergenic regions, and we found that
widespread changes in the accumulation of intergenic
noncoding RNA (ncRNA) occurred in the DNA methyla-
tion mutants (Table S4). The most dramatic changes
were observed in met1, where we detected 264 overex-
pressed ncRNAs compared to wild-type (Figure 7A). Of
these, 214 (?80%) were hypomethylated in met1 relative
to wild-type (Figure 7B). Although many of these tran-
scripts (?67%) were repetitive and could represent un-
annotated transposons, 34 were found to be single-
copy in the genome (30 hypomethylated in met1), and
an additional 54 had 2–10 copies (36 hypomethylated
in met1) (Figure 7C). Strikingly, 87 of the 88 single- or
low-copy ncRNAs did not exhibit significant homology
with sequences from other organisms in GenBank (the
NR, EST, or GSS databases). These results suggest
that a large number of fast-evolving ncRNAs exist in
Figure 5. Global Changes of Antisense RNA Expression in met1
(A) An example (At2g05915) where loss of DNA methylation and increased antisense RNA accumulation coincide in met1.
(B) Loss of methylation within genes does not systematically cause overexpression of antisense RNA in met1. Graph shows the distribution of ex-
within these genes. Unmethylated genes and unmethylated segments within genes are shown as controls. x axis = fold change in antisense RNA
expression (log10 scale). y axis = percentage of regions or genes with a given fold change.
(C) Lack of correlation between changes in sense and antisense RNA expression for both unmethylated and body-methylated genes in met1. x axis =
fold change in antisense RNA expression (log10scale). y axis = fold change in sense RNA expression (log10scale).
Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc. 1197
the Arabidopsis genome whose expression is controlled
epigenetically by DNA methylation.
60 intergenic regions that were expressed in wild-type
plants but not in met1 (Table S5). Consistent with this re-
sult, 59 of the 60 suppressed ncRNAs were also found in
the GenBank database of expressed sequence tags
(dbEST); none of these were annotated with known func-
single-copy sequences in the genome, and many (?17%)
have homologous sequences in other plant genomes,
suggesting that they might have a conserved biological
function (or functions). Thus, it appears that in Arabidop-
sis, the normal expression of many potentially important
ncRNAs may be regulated by DNA methylation.
In summary, we have combined optimized biochemical
methods and high-density whole-genome tiling microar-
rays to enable the first high-resolution genome-wide char-
acterization of DNA methylation for any organism. Consis-
tent with previous studies, we found a significant
enrichment of methylation in heterochromatin and siRNA
clusters as well as an important role for methylation in si-
lencing transposons and pseudogenes (Lippman et al.,
2004; Miura et al., 2001; Singer et al., 2001). In addition,
we discovered an unexpectedly large amount of genic
methylation and uncovered intriguing distinctions among
differentially methylated genes with regard to their func-
tional classification, expression level, and tissue specific-
ity. In parallel, genome-wide transcription profiling al-
lowed the identification of hundreds of genes and
ncRNAs whose expression is affected by changes in
DNA methylation, and these analyses have uncovered
an unexpectedly important role for non-CG methylation
in the regulation of functional genes.
The entire set of whole-genome DNA methylation
and gene expression data can be downloaded from the
gov/projects/geo/, accession numbers GSE5094 and
GSE5074). The data can also be viewed along with addi-
at http://signal.salk.edu/cgi-bin/methylome. These new
tools and genome-wide resources should serve as the raw
material for future studies to elucidate the global control of
DNA methylation patterning by mechanisms that include
Figure 6. Genes Overexpressed in met1 and drm1 drm2 cmt3
DNA Methyltransferase Mutants
(A) Composition of overexpressed genes within different gene cate-
(B) Examples of RT-PCR validation of overexpressed genes identified
by tiling-array analysis in met1 and drm1 drm2 cmt3.
(C) The F box-containing gene At2g17690 as anexample where lossof
promoter methylation in drm1 drm2 cmt3 is associated with ectopic
overexpression. Tracks are labeled as in Figure 1. Arrows at the top
represent the tandem repeats located in the promoter region. RT-
PCR validation result for this gene is shown in (B).
1198 Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc.
ing (Chan et al., 2004, 2005; Dalmay et al., 2000; Jackson
and Selker, 2001) and for investigations into the biological
functions of genes and noncoding transcripts whose ex-
pression is controlled by DNA methylation. Finally, whole-
studies of larger and more complex genomes, including
(Mockler et al., 2005).
Plant Materials and DNA Extraction
All plants used in this study were of the Arabidopsis thaliana Columbia
(Col-0) accession. The met1 and drm1 drm2 cmt3 mutants were pre-
viously published (Chan et al., 2006; Saze et al., 2003). The mutant al-
leles were met1-3 and drm1-2, drm2-2, cmt3-11. To avoid variability
caused by inbreeding, all homozygous mutant plant material used
for DNA methylation and RNA expression experiments was prepared
using F1 plants from segregating populations. Plants were grown un-
der continuous light. DNA was extracted from 5-week-old plants using
the Plant DNeasy Maxi Kit (QIAGEN) and sonicated to ?350 bp. The
entire aerial partsoftwoorthree plantswerepooled foreachbiological
The methylcytosine immunoprecipitation (mCIP) method was adapted
from a previous study (Keshet et al., 2006). Immunoprecipitation was
performed by incubating 2 mg of sonicated genomic DNA with 20 mg
mouse anti-methylcytosine monoclonal antibody (Calbiochem) in
600 ml of buffer FB (10 mM Tris-HCl [pH 7.5], 50 mM NaCl, 1 mM
EDTA) at 4?C for 12 hr. One hundred microliters each of Dynabeads
Protein G and Protein A (Dynal) were added to the mix and incubated
at 4?C for 6 hr. Dynabeads were washed six times by gentle mixing at
4?C for 10 min with 600 ml of buffer FB. Elution was performed three
times by vortexing in 200 ml TE containing 1.5%, 0.5%, and 0.1%
SDS, respectively. DNA was recovered from the unbound fraction,
each of the six washes, and the elution fraction by phenol-chloroform
extraction and ethanol precipitation. A fraction of the recovered DNA
was used for real-time PCR to determine the amount of the methylated
FWA promoter, AtSN1, and Ta3 as well as the unmethylated ACTIN
promoter in each fraction (Figure 1A). The remaining DNA from the un-
bound and elution fractionswas amplified (see below) and used for mi-
croarray hybridization. Three biological replicates were performed for
each genotype and yielded consistent results (see Supplemental
Affinity Purification of Methylated DNA Using the
Methylcytosine Binding Domain of the Human MeCP2 Protein
The expression and purification of the methylcytosine binding domain
(MBD) were performed as previously described (Cross et al., 1994;
Selker et al., 2003). Briefly, 10 mg of purified His-tagged MBD was af-
was subjected to binding by MBD in 10 ml loading buffer (20 mM
HEPES [pH 7.9], 100 mM NaCl, 10% glycerol, 0.1% Triton X-100,
0.5 mM PMSF) at a flow rate of 1 ml/min. After an initial wash with
taining a linear gradient of NaCl from 0.4 M to 1.0 M was used to elute
DNA with flow-through collected in 2 ml fractions, followed by a final
wash with 10 ml loading buffer containing 1.0M NaCl. DNA was recov-
ered from each elution. A fraction of the DNA was used to determine
the amount of the methylated FWA promoter and the unmethylated
ACTIN promoter by real-time PCR, which showed that the 0.5 M
NaCl fraction had the highest ACTIN/FWA ratio and that the 0.8 M
NaCl fraction had the highest FWA/ACTIN ratio (Figure 1B). The re-
maining DNA from these two fractions was amplified and used for
Figure 7. Expression of Intergenic Noncoding RNAs (ncRNAs)
Regulated by DNA Methylation
(A) RT-PCR validation of overexpressed ncRNAs in met1.
(B) Examples of hypomethylation correlated with ncRNA overexpres-
sion. Expression levels are shown for the forward (+) and reverse (?)
strands. RT-PCR validation results are shown below. The structures of
the overexpressed ncRNAs deduced from cloned sequences are dia-
grammed as open rectangles. DNA methylation was determined by
the mCIP-chip method.
(C) Copy number of ncRNAs that are overexpressed or suppressed in
Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc. 1199
microarray hybridization. Three biological replicates were performed
for each genotype.
Microarray Design, Experimental Procedures,
and Data Analyses
Detailed descriptions are included in the Supplemental Data.
Reverse Transcription-Polymerase Chain Reaction and Cloning
One microgram of total RNA was denatured at 65?C for 10 min, fol-
lowed by first-strand cDNA synthesis in a 50 ml reaction mix using
the Transcript First-Strand cDNA Synthesis Kit (Roche), and incubated
at 85?C for 5 min. One microliter of the mix was used with gene-spe-
cific primers for reverse transcription-polymerase chain reaction
(RT-PCR). The cycling parameters for ncRNAs were 95?C for 10 min,
21 cycles of 1 min at 95?C, 1 min at 58?C, 1 min at 72?C, and a final
elongation step at 72?C for 10 min. The cycling parameters for overex-
pressed genes were 95?C for 30 s, 30 cycles of 10 s at 95?C, 30 s at
60?C, 30 s at 72?C, and a final elongation step at 72?C for 1 min.
PCR primers are listed in Tables S6 and S7, respectively. For ncRNAs,
PCR products were purified with the QIAquick PCR Purification Kit
(QIAGEN) and cloned using the TOPO TA Cloning Kit (Invitrogen).
Genomic bisulfite sequencing was performed as previously described
(Cao and Jacobsen, 2002). The regions sequenced and the primers
used are listed in Table S8.
Supplemental Data include Supplemental Experimental Procedures,
Supplemental References, 14 figures, and 9 tables and can be found
with this article online at http://www.cell.com/cgi/content/full/126/6/
We thank Howard Cedar for the initial mCIP protocol, Jim Carrington
for miRNA and tasiRNA target gene information, Zhirong Bao and
Sean R. Eddy for providing the RECON repeat library, and T. Gingeras
and colleagues at Affymetrix Laboratories for assistance in tiling-array
development.S.W.-L.C., X.Z.,J.Y., S.E.J., andJ.R.E.designedthe ex-
periments. X.Z. generated the mCIP and MBD methylation data. J.Y.
and A.S. generated the expression data. X.Z., M.P., S.C., and S.W.-
L.C. analyzed the methylation data. S.C., M.P., and H.C. developed
analysis methods for the expression data. J.Y., S.C., M.P., X.Z., and
H.C. analyzed the expression data. H.C. and J.R.E. designed the tiling
arrays. S.W.-L.C., J.Y., I.R.H., A.S., and P.S. generated the validation
data. X.Z. wrote the manuscript. S.E.J. and J.R.E. edited the manu-
script. This work was supported by grants from the NIH ENCODE
Program HG003523 (S.E.J. and J.R.E.), NIH grant GM60398 (S.E.J.),
and NSF 2010 Project DBI0520253 (J.R.E.). A.S. was supported by
the NIH/NIGMS-funded UCSD Genetics Training Program (T32
GM08666). S.W.-L.C. is a DOE Energy Biosciences Fellow of the Life
Science Research Foundation. I.R.H. is supported by an EMBO
Long-Term Postdoctoral Fellowship. S.E.J. is an investigator of the
Howard Hughes Medical Institute.
Received: June 30, 2006
Revised: August 1, 2006
Accepted: August 7, 2006
Published online: August 31, 2006
Allen, E., Xie, Z., Gustafson, A.M., and Carrington, J.C. (2005). micro-
RNA-directed phasing during trans-acting siRNA biogenesis in plants.
Cell 121, 207–221.
Bao, N., Lye, K.W., and Barton, M.K. (2004). MicroRNA binding sites in
template chromosome. Dev. Cell 7, 653–662.
Bibikova, M., Lin, Z., Zhou, L., Chudin, E., Garcia, E.W., Wu, B.,
Doucet, D., Thomas, N.J., Wang, Y., Vollmer, E., et al. (2006). High-
throughput DNA methylation profiling using universal bead arrays.
Genome Res. 16, 383–393.
Bird, A. (2002). DNA methylation patterns and epigenetic memory.
Genes Dev. 16, 6–21.
Cao, X., and Jacobsen, S.E. (2002). Locus-specific control of asym-
metric and CpNpG methylation by the DRM and CMT3 methyltransfer-
ase genes. Proc. Natl. Acad. Sci. USA 99 (Suppl 4), 16491–16498.
Chan, S.W., Zilberman, D., Xie, Z., Johansen, L.K., Carrington, J.C.,
and Jacobsen, S.E. (2004). RNA silencing genes control de novo
DNA methylation. Science 303, 1336.
Chan, S.W., Henderson, I.R., and Jacobsen, S.E. (2005). Gardening
the genome: DNA methylation in Arabidopsis thaliana. Nat. Rev.
Genet. 6, 351–360.
Chan, S.W., Henderson, I.R., Zhang, X., Shah, G., Chien, J.S.-C., and
Jacobsen, S.E. (2006). RNAi, DRD1, and Histone Methylation Actively
Target Developmentally Important Non-CG DNA Methylation in Arabi-
dopsis. PLoS Genet. 2, e83.
Ching, T.T., Maunakea, A.K., Jun, P., Hong, C., Zardo, G., Pinkel, D.,
Albertson, D.G., Fridlyand, J., Mao, J.H., Shchors, K., et al. (2005).
Epigenome analyses using BAC microarrays identify evolutionary con-
servation of tissue-specific methylation of SHANK3. Nat. Genet. 37,
Cross, S.H., Charlton, J.A., Nan, X., and Bird, A.P. (1994). Purification
of CpG islands using a methylated DNA binding column. Nat. Genet. 6,
Dalmay, T., Hamilton, A., Mueller, E., and Baulcombe, D.C. (2000). Po-
tato virus X amplicons in arabidopsis mediate genetic and epigenetic
gene silencing. Plant Cell 12, 369–379.
Faghihi, M.A., and Wahlestedt, C. (2006). RNA interference is not in-
volved in natural antisense mediated regulation of gene expression
in mammals. Genome Biol. 7, R38.
Goll, M.G., and Bestor, T.H. (2005). Eukaryotic cytosine methyltrans-
ferases. Annu. Rev. Biochem. 74, 481–514.
Hatada, I., Fukasawa, M., Kimura, M., Morita, S., Yamada, K., Yoshi-
kawa, T., Yamanaka, S., Endo, C., Sakurada, A., Sato, M., et al.
(2006). Genome-wide profiling of promoter methylation in human. On-
cogene 25, 3059–3064.
Henderson, I.R., Zhang, X., Lu, C., Johnson, L., Meyers, B.C., Green,
P.J., and Jacobsen, S.E. (2006). Dissecting Arabidopsis thaliana
DICER function in small RNA processing, gene silencing and DNA
methylation patterning. Nat. Genet. 38, 721–725.
Hirochika, H., Okamoto, H., and Kakutani, T. (2000). Silencing of retro-
transposons in arabidopsis and reactivation by the ddm1 mutation.
Plant Cell 12, 357–369.
Hollander, M., and Wolfe, D.A. (1999). Nonparametric Statistical
Methods, Second Edition (New York: John Wiley and Sons, Inc.).
Jackson, J.P., Lindroth, A.M., Cao, X., and Jacobsen, S.E. (2002).
Control of CpNpG DNA methylation by the KRYPTONITE histone H3
methyltransferase. Nature 416, 556–560.
Jacobsen, S.E., Sakai, H., Finnegan, E.J., Cao, X., and Meyerowitz,
E.M. (2000). Ectopic hypermethylation of flower-specific genes in Ara-
bidopsis. Curr. Biol. 10, 179–186.
Ji, H., and Wong, W.H. (2005). TileMap: create chromosomal map of
tiling array hybridizations. Bioinformatics 21, 3629–3636.
Keshet, I., Schlesinger, Y., Farkash, S., Rand, E., Hecht, M., Segal, E.,
Pikarski, E., Young, R.A., Niveleau, A., Cedar, H., and Simon, I. (2006).
Evidence for an instructive mechanism of de novo methylation in can-
cer cells. Nat. Genet. 38, 149–153.
1200 Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc.
Lippman, Z., Gendrel, A.V., Black, M., Vaughn, M.W., Dedhia, N., Download full-text
McCombie, W.R., Lavine, K., Mittal, V., May, B., Kasschau, K.D.,
et al. (2004). Role of transposable elements in heterochromatin and
epigenetic control. Nature 430, 471–476.
P.J. (2005). Elucidation of the small RNA component of the transcrip-
tome. Science 309, 1567–1569.
Malagnac, F., Bartee, L., and Bender, J. (2002). An Arabidopsis SET
domain protein required for maintenance but not establishment of
DNA methylation. EMBO J. 21, 6842–6852.
Mette, M.F., Aufsatz, W., van der Winden, J., Matzke, M.A., and
Matzke, A.J. (2000). Transcriptional silencing and promoter methyla-
tion triggered by double-stranded RNA. EMBO J. 19, 5194–5201.
Miura, A., Yonebayashi, S., Watanabe, K., Toyama, T., Shimada, H.,
and Kakutani, T. (2001). Mobilization of transposons by a mutation
abolishing full DNA methylation in Arabidopsis. Nature 411, 212–214.
Mockler, T.C., Chan, S., Sundaresan, A., Chen, H., Jacobsen, S.E.,
and Ecker, J.R. (2005). Applications of DNA tiling arrays for whole-
genome analysis. Genomics 85, 1–15.
Ngernprasirtsiri,J., Kobayashi, H.,and Akazawa, T.(1988). DNAmeth-
ylation as a mechanism of transcriptional regulation in nonphotosyn-
thetic plastids in plant cells. Proc. Natl. Acad. Sci. USA 85,4750–4754.
Peragine, A., Yoshikawa, M., Wu, G., Albrecht, H.L., and Poethig, R.S.
ment and the production of trans-acting siRNAs in Arabidopsis. Genes
Dev. 18, 2368–2379.
Rakyan, V.K., Hildmann, T., Novik, K.L., Lewin, J., Tost, J., Cox, A.V.,
Andrews, T.D., Howe, K.L., Otto, T., Olek, A., et al. (2004). DNA meth-
ylation profiling of the human major histocompatibility complex: a pilot
study for the human epigenome project. PLoS Biol. 2, e405.
Rollins, R.A., Haghighi, F., Edwards, J.R., Das, R., Zhang, M.Q., Ju, J.,
and Bestor, T.H. (2006). Large-scale structure of genomic methylation
patterns. Genome Res. 16, 157–163.
Saze, H., Scheid, O.M., and Paszkowski, J. (2003). Maintenance of
CpG methylation is essential for epigenetic inheritance during plant
gametogenesis. Nat. Genet. 34, 65–69.
Schmid, M.,Davison, T.S.,Henz,S.R.,Pape, U.J.,Demar,M.,Vingron,
M., Scholkopf, B., Weigel, D., and Lohmann, J.U. (2005). A gene ex-
pression map of Arabidopsis thaliana development. Nat. Genet. 37,
Schug, J., Schuller, W.P., Kappen, C., Salbaum, J.M., Bucan, M., and
Stoeckert, C.J., Jr.(2005). Promoter features related totissue specific-
ity as measured by Shannon entropy. Genome Biol. 6, R33.
Selker, E.U., Tountas, N.A., Cross, S.H., Margolin, B.S., Murphy, J.G.,
Bird, A.P., and Freitag, M. (2003). The methylated component of the
Neurospora crassa genome. Nature 422, 893–897.
Singer, T., Yordan, C., and Martienssen, R.A. (2001). Robertson’s
Mutator transposons in A. thaliana are regulated by the chromatin-re-
modeling gene Decrease in DNA Methylation (DDM1). Genes Dev. 15,
Soppe, W.J., Jacobsen, S.E., Alonso-Blanco, C., Jackson, J.P., Kaku-
tani, T., Koornneef, M., and Peeters, A.J. (2000). The late flowering
phenotype of fwa mutants is caused bygain-of- function epigenetic al-
leles of a homeodomain gene. Mol. Cell 6, 791–802.
Tamaru, H., and Selker, E.U. (2001). A histone H3 methyltransferase
controls DNA methylation in Neurospora crassa. Nature 414, 277–283.
Tariq, M., Saze, H., Probst, A.V., Lichota, J., Habu, Y., and Paszkow-
ski, J. (2003). Erasure of CpG methylation in Arabidopsis alters pat-
terns of histone H3 methylation in heterochromatin. Proc. Natl. Acad.
Sci. USA 100, 8823–8827.
The Arabidopsis Genome Initiative. (2000). Analysis of the genome
sequence of the flowering plant Arabidopsis thaliana. Nature 408,
Henikoff, S. (2005a). DNA methylation profiling identifies CG methyla-
tion clusters in Arabidopsis genes. Curr. Biol. 15, 154–159.
Tran, R.K., Zilberman, D., de Bustos, C., Ditt, R.F., Henikoff, J.G., Lin-
droth, A.M., Delrow, J., Boyle, T., Kwong, S., Bryson, T.D., et al.
(2005b). Chromatin and siRNA pathways cooperate to maintain DNA
methylation of small transposable elements in Arabidopsis. Genome
Biol. 6, R90.
Vazquez, F., Vaucheret, H., Rajagopalan, R., Lepers, C., Gasciolli, V.,
Mallory, A.C., Hilbert, J.L., Bartel, D.P., and Crete, P. (2004). Endoge-
nous trans-acting siRNAs regulate the accumulation of Arabidopsis
mRNAs. Mol. Cell 16, 69–79.
Wang, X.J., Gaasterland, T., and Chua, N.H. (2005). Genome-wide
prediction and identification of cis-natural antisense transcripts in Ara-
bidopsis thaliana. Genome Biol. 6, R30.
Weber, M., Davies, J.J., Wittig, D., Oakeley, E.J., Haase, M., Lam,
W.L., and Schubeler, D. (2005). Chromosome-wide and promoter-
specific analyses identify sites of differential DNA methylation in nor-
mal and transformed human cells. Nat. Genet. 37, 853–862.
Yamada, K., Lim, J., Dale, J.M., Chen, H., Shinn, P., Palm, C.J., South-
wick,A.M., Wu,H.C., Kim, C.,Nguyen, M.,etal. (2003). Empirical anal-
ysis of transcriptional activity inthe Arabidopsis genome. Science 302,
All raw microarray data (CEL files) for DNA methylation and expression
analyses have been deposited in GEO under the accession numbers
GSE5094 and GSE5074, respectively. The ncRNA sequences
have been deposited in GenBank under the accession numbers
Cell 126, 1189–1201, September 22, 2006 ª2006 Elsevier Inc. 1201