Genetic Interactions between DNA Demethylation

and Methylation in Arabidopsis[OA]

Jon Penterman1, Rie Uzawa, and Robert L. Fischer*
Department of Plant and Microbial Biology, University of California, Berkeley, California 94720

DNA demethylation in Arabidopsis (Arabidopsis thaliana) is mediated by DNA glycosylases of the DEMETER family. Three
DEMETER-LIKE (DML) proteins, REPRESSOR OF SILENCING1 (ROS1), DML2, and DML3, function to protect genes from
potentially deleterious methylation. In Arabidopsis, much of the DNA methylation is directed by RNA interference (RNAi)
pathways and used to defend the genome from transposable elements and their remnants, repetitive sequences. Here, we
investigated the relationship between DML demethylation and RNAi-mediated DNA methylation. We found that genic
regions demethylated by DML enzymes are enriched for small interfering RNAs and generally contain sequence repeats,
transposons, or both. The most common class of small interfering RNAs was 24 nucleotides long, suggesting a role for an RNAi
pathway that depends on RNA-DEPENDENT RNA POLYMERASE2 (RDR2). We show that ROS1 removes methylation that
has multiple, independent origins, including de novo methylation directed by RDR2-dependent and -independent RNAi
pathways. Interestingly, in rdr2 and drm2 mutant plants, we found that genes demethylated by ROS1 accumulate CG
methylation, and we propose that this hypermethylation is due to the ROS1 down-regulation that occurs in these mutant
backgrounds. Our observations support the hypothesis that DNA demethylation by DML enzymes is one mechanism by which
Arabidopsis genes are protected from genome defense pathways.

In Arabidopsis (Arabidopsis thaliana), DNA demeth-
ylation is carried out by bifunctional helix-hairpin-
helix DNA glycosylases of the DEMETER (DME)
family (Agius et al., 2006; Gehring et al., 2006; Morales-
Ruiz et al., 2006; Penterman et al., 2007). The DME
family consists of DME, DEMETER-LIKE 2 (DML2),
DML3, and REPRESSOR OF SILENCING1 (ROS1).
DNA demethylation by DME occurs during reproduc-
tive development and is required for genomic imprint-
ing and seed viability (Choi et al., 2002; Gehring et al.,
2006). DNA demethylation by ROS1, DML2, and
DML3 protects genes from potentially deleterious
methylation (Penterman et al., 2007; Zhu et al., 2007).
DML enzymes demethylate approximately 179 loci, of
which nearly 80% are genes (Penterman et al., 2007).
DML demethylation at genic loci primarily occurs at
the 5# end and the 3# end and, as a set, the genes
demethylated by DMLs are representative of the ge-
nome at large (Penterman et al., 2007). The origin and
underlying causes of methylation and demethylation
at these genes are not known.
1
Present address: Department of Microbiology, University of
Washington, Seattle, WA 98195.
* Corresponding author; e-mail rfischer@berkeley.edu.
The author responsible for distribution of materials integral to the
findings presented in this article in accordance with the policy
described in the Instructions for Authors (www.plantphysiol.org) is:
Robert L. Fischer (rfischer@berkeley.edu).
[OA]
Open Access articles can be viewed online without a sub-
scription.
www.plantphysiol.org/cgi/doi/10.1104/pp.107.107730
Plant Physiology, December 2007, Vol. 145, pp. 1549–1557, www.plantphysiol.org Ó 2007 American Society of Plant Biologists
ROS1 is necessary for expression of an RD29AT
LUCIFERASE (RD29ATLUC) transgene and the en-
dogenous RD29A gene (Gong et al., 2002). In
RD29ATLUC plants, ROS1 removes RNA-directed
DNA methylation (RdDM) from the promoters of
RD29ATLUC and RD29A, as well as DNA methyl-
ation at other loci (Gong et al., 2002; Penterman et al.,
2007; Zhu et al., 2007). RdDM is mediated by RNA
interference (RNAi; Bender, 2004; Chan et al., 2005).
Aberrant transcripts produced from the RD29ATLUC
reporter gene are processed into small interfering
RNA (siRNA), which are likely loaded into ARGO-
NAUTE4 (AGO4) and AGO6 effector complexes and
used to direct methyltransferases to the promoters of
RD29ATLUC and the endogenous RD29A gene (Gong
et al., 2002; Zheng et al., 2007). It is not known if the
antagonism between ROS1 and RdDM extends to
endogenous loci demethylated by ROS1.
Current evidence suggests that the primary function
of RdDM and RNAi is to defend the genome from
transposable elements (Bender, 2004; Chan et al., 2005).
At least two independent RNAi pathways can lead to
RdDM. One pathway includes RNA-DEPENDENT
RNA POLYMERASE6 (RDR6), SUPPRESSOR OF GENE
SILENCING3/SILENCING DEFECTIVE2 (SDE2), SDE3,
and AGO1, and the other RNAi pathway is composed of
RNA polymerase IV (PolIV), RNA-DEPENDENT RNA
POLYMERASE2 (RDR2), DICER-LIKE3 (DCL3), and
AGO4 (Chan et al., 2005). Evidence suggests that the
PolIV/RDR2/DCL3/AGO4 pathway catalyzes the
majority of RdDM in Arabidopsis (Chan et al., 2005;
Li et al., 2006; Lu et al., 2006; Pontes et al., 2006;
Kasschau et al., 2007; Zhang et al., 2007). The PolIV/
RDR2/DCL3/AGO4 pathway is required for de novo
1549
Penterman et al.
methylation by the methyltransferase DOMAINS RE-
ARRANGED METHYLTRANSFERASE2 (DRM2) and
for the maintenance of non-CG methylation genome-
wide by DRM2 and CHROMOMETHYLTRANSFER-
ASE3 (Bender, 2004; Chan et al., 2005). The PolIV/
RDR2/DCL3/AGO4 pathway maintains non-CG
methylation by using a positive feedback loop (Pontes
et al., 2006; Zhang et al., 2007). At heterochromatic loci,
PolIV is hypothesized to generate single-stranded
RNA molecules (Herr et al., 2005; Kanno et al., 2005;
Pontes et al., 2006). RDR2 then synthesizes double-
stranded RNAs with which DCL3 endonuclease gen-
erates 24-nucleotide siRNAs, a hallmark of the PolIV/
RDR2/DCL3/AGO4 pathway (Xie et al., 2004; Herr
et al., 2005; Pontes et al., 2006; Kasschau et al., 2007).
The siRNAs then direct DRM2 to the loci from which
they originated, maintaining or establishing the level
of DNA methylation. PolIV then initiates the cycle
again to maintain non-CG methylation (Pontes et al.,
2006; Zhang et al., 2007), and METHYLTRANSFER-
ASE1 (MET1) maintains CG methylation, which may
not require additional cycles of the siRNA pathway
(Bender, 2004; Chan et al., 2005). Mutations that affect
any step of this pathway block de novo DNA methyl-
ation, significantly reduce genomic non-CG methylation,
and cause the derepression of several transposable
elements (Kanno et al., 2004, 2005; Xie et al., 2004; Chan
et al., 2005, 2006; Herr et al., 2005). The PolIV/RDR2/
DCL3/AGO4 pathway might have a role in establish-
ing methylation at endogenous loci demethylated by
ROS1 because ROS1 and AGO4 are antagonists in
transgenic RD29ATLUC plants (Zheng et al., 2007).
Mutations in RDR2, DEFECTIVE IN RNA-
DIRECTED DNA METHYLATION1 (DRD1), and genes
that encode PolIV reduce the level of ROS1 RNA
(Huettel et al., 2006, 2007). It is not known how
reduced ROS1 expression due to mutations in the
RDR2 and PolIV genes affects DNA demethylation
activity at endogenous ROS1 gene targets.
Here, we investigated the genetic relationship be-
tween DNA demethylation and DNA methylation
in Arabidopsis. We show that genic regions demeth-
ylated by DML enzymes are enriched for siRNAs and
generally contain sequence repeats, transposons, or
both. The most common class of siRNA at DML target
genes was 24 nucleotides long, suggesting a role for
the PolIV/RDR2/DCL3/AGO4 pathway at these loci.
By using mutations in RDR2 and DRM2, we show that
ROS1 removes de novo methylation directed by the
PolIV/RDR2/DCL3/AGO4 pathway and an RDR2-
independent RNAi pathway. Interestingly, in rdr2 and
drm2 mutant plants, we found that genes demeth-
ylated by ROS1 in wild-type plants accumulate meth-
ylation at CG sites, and we propose that this
hypermethylation is due to the ROS1 down-regulation
that occurs in these mutant backgrounds. Our obser-
vations support the hypothesis that DML enzymes
protect genes from the genome defense pathway by
demethylating nearby sequence repeats and transpos-
able elements.
1550
RESULTS
Loci Demethylated by DML Enzymes Correspond to
Regions with Small RNAs, Repeats, and Transposons
DNA methylation is generally found at transposons,
repetitive sequences, and regions targeted by siRNAs.
We searched for these features at intergenic and genic
loci (n 5 33 and n 5 146, respectively) previously
identified as demethylated by DMLs (Penterman et al.,
2007) by using databases that contain sequenced
siRNAs and computationally identified repeats and
transposons (Gustafson et al., 2005; Lu et al., 2005;
Rajagopalan et al., 2006; Kasschau et al., 2007; Zhang
et al., 2007). Sequenced siRNAs were found at 60% (n 5
88) of genes and 64% (n 5 21) of intergenic loci that are
demethylated by DML enzymes (Fig. 1A). The gene set
demethylated by DMLs is significantly enriched for
siRNAs relative to the number of Arabidopsis genes
targeted by siRNAs (approximately 20%; Kasschau
et al., 2007). Transposons and various sequence repeats
were also common to genic (48%; n 5 70) and intergenic
(55%; n 5 18) loci demethylated by DML enzymes (Fig.
1A). When combined, 75% of genes (n 5 110) and 76%
of intergenic loci (n 5 25) demethylated by DMLs had
siRNAs, transposable elements and repeats, or both
(Fig. 1A). These data show that loci demethylated by
DMLs are enriched for siRNAs and commonly overlap
with sequence repeats and transposable elements.
Multiple RNAi pathways function in Arabidopsis,
producing siRNAs that generally are 20 to 24 nucleo-
tides long. The size of siRNAs at a given locus can
provide clues as to which RNAi pathway targets that
locus. Data about the size of siRNAs were available for
105/109 loci targeted by DML demethylation and
siRNAs. At these loci, siRNAs ranged from 20 to 24
nucleotides in length, and multiple classes of siRNAs
were often observed at a single locus (Fig. 1B; data not
shown). The siRNA profile of DML-target loci was
skewed toward the longer siRNA classes (Fig. 1B).
Nearly 80% of all loci demethylated by DMLs were
targeted by a 24-nucleotide-long siRNA (Fig. 1B), which
is a hallmark of the PolIV/RDR2/DCL3/AGO4 path-
way (Xie et al., 2004; Herr et al., 2005; Kasschau et al.,
2007). These observations suggest that the PolIV/
RDR2/DCL3/AGO4 RNAi pathway targets DNA meth-
ylation to most loci demethylated by DML enzymes.
ROS1 Demethylation Antagonizes de Novo
Methylation Directed by the PolIV/RDR2/DCL3/AGO4
RNAi Pathway
The PolIV/RDR2/DCL3/AGO4 pathway protects
the genome from transposable elements and repetitive
DNAs by establishing de novo methylation and main-
taining non-CG methylation (Chan et al., 2004, 2005,
2006). We used the rdr2-1 loss-of-function mutation to
test the role of this pathway in methylation of ROS1-
target loci because RDR2 is specific to and essential for
the PolIV/RDR2/DCL3/AGO4 pathway (Xie et al.,
2004; Lu et al., 2006; Pontes et al., 2006; Kasschau et al.,
Plant Physiol. Vol. 145, 2007

Figure 1. DML-target loci have features associated with DNA methyl-
ation. A, Percentage of DML-target genes and intergenic loci with
siRNAs, transposons and repetitive elements, or both. B, Percentage of
DML-target loci with 20-, 21-, 22-, 23-, and 24-nucleotide siRNAs.
2007; Zhang et al., 2007). An F1 plant heterozygous for
ros1-3 and rdr2-1 was self-pollinated to generate a
homozygous ros1-3; rdr2-1 F2 line, and the methylation
levels of three ROS1-target genes (At1g26400,
At1g34245, and At5g38550) in ros1-3; rdr2-1 F3 plants
were measured and compared to wild-type and ros1-3
control plants. Hypermethylation of these loci was
significantly less in ros1-3; rdr2-1 F3 relative to ros1-3
plants (Fig. 2, A–C). For the At1g26400 (Fig. 2A) and
At5g38550 (Fig. 2C) loci, CNG and CNN hypermethyl-
ation was reduced to wild-type levels in ros1-3; rdr2-1
F3 plants, but CG sites remained hypermethylated
relative to wild-type plants. For the At1g34245 locus
(Fig. 2B), both CG and non-CG methylation levels
Plant Physiol. Vol. 145, 2007
DNA Demethylation and Methylation
were reduced in ros1-3; rdr2-1 F3 plants, although the
levels of CG and CNG methylation were intermediate
between the levels observed in wild-type and ros1-3
plants. Thus, full CNG and CNN hypermethylation of
loci in ros1-3 mutants requires a PolIV/RDR2/DCL3/
AGO4 pathway. These results are consistent with the
high proportion of 24-nucleotide siRNAs at DML-
target loci and indicate that ROS1 demethylation pro-
tects endogenous genes from methylation directed by
the PolIV/RDR2/DCL3/AGO4 pathway.
The PolIV/RDR2/DCL3/AGO4 pathway both es-
tablishes and maintains non-CG methylation. In the
experiments reported above, ROS1 and RDR2 activities
were simultaneously lost when the ros1-3; rdr2-1 F2
double mutant plants were formed (ros1; rdr2 in Fig.
2, A–C). Thus, the reduction of non-CG methylation
observed in the ros1-3; rdr2-1 genetic background
might reflect the role the PolIV/RDR2/DCL3/AGO4
pathway has in both establishing and maintaining
hypermethylation of ROS1-target loci. Alternatively,
these results might reflect only the maintenance func-
tion of the PolIV/RDR2/DCL3/AGO4 pathway. To
clarify the role of the PolIV/RDR2/DCL3/AGO4 path-
way at these loci, we defined this pathway’s role in
maintaining DNA hypermethylation in the ros1-3 mu-
tant background. This was accomplished by allowing
hypermethylation to be established and maintained by
the PolIV/RDR2/DCL3/AGO4 pathway in a ros1-3;
RDR2/rdr2-1 F2 mutant plant (see methylation data in
Fig. 5, B–D), which was subsequently self-pollinated,
and the effect of the loss of RDR2 activity upon
maintaining hypermethylation was assessed by com-
paring the methylation levels of loci in ros1-3; rdr2-1 F3
progeny to the levels observed in ros1-3 F3 plants (Fig.
2, A–E, compare ros1 in A–C to ros1; rdr2 in D–F).
CNG and CNN hypermethylation at At1g26400 and
At1g34245 was significantly reduced in ros1-3; rdr2-1 F3
progeny relative to ros1-3 plants (compare ros1; rdr2 in
A and B to D and E, respectively), indicating that the
PolIV/RDR2/DCL3/AGO4 pathway maintains CNG
and CNN hypermethylation at these loci in ros1-3
plants. At At5g38550, CNN hypermethylation was
significantly reduced in ros1-3; rdr2-1 F3 progeny rela-
tive to ros1-3 plants (compare ros1; rdr2 in C and E),
indicating that the PolIV/RDR2/DCL3/AGO4 path-
way maintains CNN hypermethylation at At5g38550
in ros1-3 plants.
We next determined the role of the PolIV/RDR2/
DCL3/AGO4 pathway in establishing hypermethyl-
ation at ROS1-target loci by comparing the results
obtained when ROS1 and RDR2 activity were simul-
taneously lost (ros1; rdr2 in Fig. 2, A–C) to those when
the maintenance function of RDR2 was lost in ros1-3;
rdr2-1 F3 progeny (ros1; rdr2 in Fig. 2, D–F). We found
an equivalent reduction in CNG and CNN hyper-
methylation at At1g26400 (compare ros1; rdr2 in A and
D) and Ag1g34245 (compare ros1; rdr2 in B and E),
making it impossible to differentiate the maintenance
and establishment functions of the PolIV/RDR2/
DCL3/AGO4 pathway at these loci in ros1 mutant
1551
Penterman et al.
Figure 2. ROS1 removes DNA methylation established and maintained
by the PolIV/RDR2/DCL3/AGO4 pathway and DRM2. A to F, Percent-
age of methylation at At1g26400 (A and D), At1g34245 (B and E), and
At5g38550 (C and F) in wild-type, ros1-3, ros1-3; rdr2-1, and ros1-3;
drm2-1 plants. The ros1-3; rdr2-1 genotype in A to C was generated by
selfing a ros1-3; rdr2-1 F2 line, and the ros1-3; rdr2-1 genotype in D to F
was generated by selfing a ros1-3; RDR2/rdr2-1 plant. The ros1-3;
drm2-1 F3 plant was generated by selfing a ros1-3; drm2-1 F2 line.
1552
plants. At At5g38550, it was not possible to distinguish
the establishment and maintenance function of the
PolIV/RDR2/DCL3/AGO4 pathway in hypermethyl-
ation of CNN sites because an equivalent reduction
was observed in both plants (compare ros1; rdr2 in C
and F). However, the establishment and maintenance
function of the PolIV/RDR2/DCL3/AGO4 pathway
in hypermethylation of CNG was distinguishable; the
PolIV/RDR2/DCL3/AGO4 pathway was required for
establishment, but not maintenance, of CNG hyper-
methylation (compare ros1; rdr2 in C and F), indicating
that the PolIV/RDR2/DCL3/AGO4 pathway estab-
lishes de novo CNG methylation at At5g38550 in ros1
mutant plants. After establishment, a different path-
way maintains CNG hypermethylation at At5g38550
in ros1-3 mutants. Collectively, these data show that
DNA demethylation by ROS1 in wild-type plants
removes methylation that in most, but not all, cases
is established and maintained by the PolIV/RDR2/
DCL3/AGO4 pathway.
ROS1 Removes Methylation That Has Multiple,
Independent Origins
Mutations that disrupt the PolIV/RDR2/DCL3/
AGO4 pathway do not affect the establishment or
maintenance of CG hypermethylation at At1g26400,
At5g38550, and At1g34245 in ros1 mutant plants (ros1;
rdr2 in Fig. 2, A–C). The lack of a role by the PolIV/
RDR2/DCL3/AGO4 pathway in CG hypermethyl-
ation might be a reflection of the fact that CG methylation
is symmetric and is mediated by MET1 methyltrans-
ferase without input from an RNAi pathway (Chan
et al., 2005). Alternatively, a redundant RNAi pathway,
distinct from the PolIV/RDR2/DCL3/AGO4 pathway,
could establish this CG hypermethylation. To test the
latter idea, we investigated the role of the de novo
methyltransferase DRM2 in establishing hypermethyl-
ation at At1g26400, At1g34245, and At5g38550 in the
ros1-3 mutant background. If an RNAi pathway other
than the PolIV/RDR2/DCL3/AGO4 pathway estab-
lishes hypermethylation in ros1 plants, then drm2 mu-
tations should lower CG hypermethylation levels more
than rdr2 mutations. This is because DRM2 is the only
known de novo methyltransferase in Arabidopsis
(Chan et al., 2005), so that both the PolIV/RDR2/
DCL3/AGO4 pathway and the hypothetical redun-
dant pathway would use DRM2 for de novo DNA
methylation. A plant heterozygous for the ros1-3 and
drm2-1 mutations was self-pollinated to generate a ros1-3;
drm2-1 F2 line, and the methylation of these loci was
measured in F3 plants (ros1; drm2 in Fig. 2, A–C). As
expected, ros1-3; drm2-1 mutant plants displayed little
or no CNG or CNN hypermethylation at At1g26400,
At1g34245, and At5g38850. Thus, mutations in DRM2
prevented the establishment and/or maintenance of
nearly all non-CG hypermethylation at loci in ros1-3;
drm2-1 F3 plants (Fig. 2, A–C). By contrast, CG hyper-
methylation at At1g26400 and At5g38550 was not af-
fected in ros1-3; drm2-1 F3 double mutants, suggesting
Plant Physiol. Vol. 145, 2007
 DNA Demethylation and Methylation

that DRM2 is not required for CG hypermethylation at

these loci (ros1; drm2 in Fig. 2, A and C). CG sites at
At1g34245 were also hypermethylated in ros1-3; drm2
plants relative to wild-type plants (ros1; drms in Fig. 2B).
However, the CG methylation level at At1g34245 in
ros1-3; drm2 was approximately 50% lower than in
ros1-3 plants (ros1 in Fig. 2B), yet was still 3-fold higher
than the wild type (WT in Fig. 2), indicating that DRM2
was required for a certain level of CG hypermethylation
at At1g34245 in ros1 mutant plants. Taken together,
the data indicate that ROS1 actively removes CG meth-
ylation that is mostly independent of DRM2.
The CNG and CNN methylation profiles at At1g26400
and At5g38550 in ros1-3; drm2 F3 plants were similar to
the profiles observed in ros1-3; rdr2 F3 plants, indicating
that DRM2 and RDR2 are required to establish the same
patterns of CNG and CNN hypermethylation at these
loci (Fig. 2, A and C). The CNG methylation level at
At1g34245 in ros1-3; drm2 F3 plants was lower than the
level measured in ros1-3; rdr2 F3 plants (Fig. 2B), sug-
gesting that DRM2 establishes a certain level of CNG
methylation at At1g34245 in ros1-3 plants in an RDR2-
independent manner. De novo methylation by DRM2
requires RNAi (Chan et al., 2005). Therefore, these
results suggest that ROS1 demethylation at At1g34245
removes de novo methylation that is directed by RDR2-
dependent and -independent RNAi pathways.

Mutations in DRM2 and RDR2 Lead to CG

Hypermethylation at ROS1-Target Loci
At1g26400, At1g34245, and At5g38550 all contain low
levels of non-CG methylation in wild-type plants (Fig. 2,
A–C; Penterman et al., 2007). Given the previous results
showing non-CG DNA methylation by the PolIV/
RDR2/DCL3/AGO4 pathway, single rdr2 and drm2
mutations in a wild-type ROS1 background would be
expected to reduce non-CG methylation levels of these
loci because ROS1 would remove methylation and
no new non-CG methylation would be established by
pathways using RDR2 and DRM2. The overall non-
CG methylation level at At1g26400, At1g34245, and
At5g38550 in rdr2 and drm2 single mutants was less or
similar to the level observed in wild-type plants (Fig. 3,
A–C).
As for CG methylation, the rdr2 and drm2 mutations
should have no effect or reduce it if they are needed to
re-establish CG methylation after ROS1 demethylation
in wild-type plants. However, contrary to expecta-
tions, At1g26400, At1g34245, and At5g38550 were hy-
permethylated at CG sites in rdr2-1 and drm2-1 single
mutants relative to wild-type plants (Fig. 3, A–C). The
degree of hypermethylation was similar to that ob-
served in ros1 mutant plants (Fig. 3, A–C).
Is the CG hypermethylation of ROS1-target loci
general to mutants defective in RNAi and RdDM, or
is it specific to the PolIV/RDR2/DCL3/AGO4 RNAi
pathway? To address this question, At1g26400, At1g34245, Figure 3. ROS1-target loci are hypermethylated in rdr2-1 and drm2-1 mutant
and At5g38550 were bisulfite sequenced from plants plants. A to C, Percentage of methylation at At1g26400 (A), At1g34245 (B),
bearing mutations in RDR6, which is required for and At5g38550 (C) in wild-type, ros1-3, rdr2-1, drm2-1, and rdr6-11 plants.

Plant Physiol. Vol. 145, 2007 1553

Penterman et al.
RdDM and transcriptional silencing of highly tran-
scribed sense transgenes (Chan et al., 2005). No hyper-
methylation was observed at ROS1 target loci (Fig. 3,
A–C). These observations suggest that ROS1-target loci
are specifically affected by mutations in RDR2 and
DRM2.
ROS1 Is Down-Regulated in drm2 Mutant Plants
rdr2-1 and drm2-1 mutations cause CG hypermethyl-
ation at ROS1 target loci (Fig. 3). The level of CG
hypermethylation is similar to the level observed in
ros1-3; rdr2-1 and ros1-3; drm2-1 double mutant plants
(Fig. 2, A–C). Moreover, plants with mutations in RDR2
and PolIV genes have significantly reduced levels of
ROS1 RNA (Huettel et al., 2006). Down-regulation of
ROS1 might explain why ROS1 target loci become
hypermethylated at CG sites in rdr2 and drm2 mutants.
To test this idea, semiquantitative reverse transcription
(RT)-PCR was used to compare ROS1 expression levels
between rdr2, drm2, rdr6, ROS1/ros1-3; rdr2-1, and ROS1/
ros1-3; RDR2/rdr2-1 plants. As previously shown, the
expression of ROS1 was significantly down-regulated
in rdr2-1 plants (Fig. 4; Huettel et al., 2006). In drm2
mutant plants, ROS1 expression was further reduced
and could barely be detected by RT-PCR (Fig. 4). By
contrast, ROS1 expression was not reduced in rdr6-11
mutant plants (Fig. 4). These results show that proper
ROS1 expression depends on DRM2 and RDR2, but not
on RDR6. Thus, there is a positive correlation between
the expression level of ROS1 and the CG methylation
status of its target genes. In drm2 and rdr2 mutants,
ROS1 expression is down-regulated and its target loci
are hypermethylated at CG sites. In the wild type and
rdr6-11, there is a high level of ROS1 expression and
ROS1-target loci are hypomethylated. ROS1-target loci
might be hypermethylated at CG sites in drm2 and rdr2
plants because ROS1 expression is down-regulated.
RDR2 Is Required for DNA Demethylation by ROS1
To demonstrate that the down-regulation of ROS1 in
rdr2-1 mutant plants affects ROS1 demethylation, we
tested the ability of ROS1 to demethylate and restore
CG hypermethylated loci to wild-type levels in sib-
lings that differ in the presence and absence of RDR2
(Fig. 5A). We crossed ros1-3; RDR2/rdr2-1 plants to
rdr2-1 single mutants to generate ros1/ROS1; RDR2/
rdr2-1 and ros1/ROS1; rdr2-1 progeny (Fig. 5A). We
compared the methylation levels of these progeny to
the levels of the parents to see if reintroduction of
ROS1 alleles in progeny could restore hypermethylated
alleles to a more hypomethylated state. The level of CG
methylation at At1g26400, At1g34245, and At5g38550
in heterozygous ros1/ROS1; RDR2/rdr2-1 progeny was
much less than in both parents, indicating that these
loci were demethylated by ROS1 (Fig. 5, B–D). How-
ever, in ros1/ROS1; rdr2-1 siblings, CG methylation
levels at At1g26400, At1g34245, and At5g38550 were
similar to or exceeded the methylation levels of the
1554
Figure 4. ROS1 expression is down-regulated in rdr2-1 and drm2-1
mutants. Shown is the expression of ROS1 and ACTIN11 (Actin) in
ros1-3, rdr2-1, ros1-3/ROS1; rdr2-1/RDR2, ros1-3/ROS1; rdr2-1, drm2-1,
and rdr6-11 plants (lanes 1–6, respectively).
parents, indicating that the rdr2 mutation inhibited
ROS1 demethylation (Fig. 5, B–D). These results show
that the CG hypermethylation observed at ROS1 target
loci in rdr2 and drm2 plants (Fig. 3, A–C) is because
ROS1 demethylation requires a wild-type RDR2 allele.
DISCUSSION
In this study, we showed that the loci demethylated
by DML enzymes contain features commonly targeted
by the plant’s genome defense pathway. These features
include transposable elements and their remnants,
repetitive sequences, and siRNAs (Fig. 1A). A signif-
icant fraction of DML-target loci matched 24 nucleotide
siRNAs (Fig. 1B), which suggested that DML demeth-
ylation removes methylation directed by the PolIV/
RDR2/DCL3/AGO4 RNAi pathway (Xie et al., 2004;
Kasschau et al., 2007). We demonstrated an antagonis-
tic relationship between ROS1 and the PolIV/RDR2/
DCL3/AGO4 pathway and uncovered evidence that
suggests that ROS1 also removes DNA methylation
directed by an RDR2-independent RNAi pathway (Fig.
2B). We found that DRM2 is also required for ROS1
expression and demonstrate that wild-type RDR2 and
DRM2 alleles are required for DNA demethylation by
ROS1 (Figs. 3, A–C, 4, and 5, B–D). Our data support
the hypothesis that DNA demethylation by DML DNA
glycosylases protects genes from the genome defense
pathways and show one mechanism by which ROS1
demethylation is potentially regulated by DNA meth-
ylation pathways in Arabidopsis.
DML demethylation protects genes from potentially
deleterious methylation (Penterman et al., 2007). At
genes, DML enzymes primarily demethylate the 5# and
Plant Physiol. Vol. 145, 2007
 DNA Demethylation and Methylation

3# ends, a pattern opposite to the overall distribution

of wild-type DNA methylation (Zhang et al., 2006;
Penterman et al., 2007; Vaughn et al., 2007; Zilberman
et al., 2007). This led us to hypothesize that DML
demethylation was a factor keeping gene ends meth-
ylation free. Our data show that nearby transposons
and repeats are common characteristics of DML-
demethylated genes (Fig. 1A). These features, which
are inherent targets of the RNAi pathways and RdDM
(Chan et al., 2005), are likely the reason methylation is
directed to some DML-demethylated genes (Penterman
et al., 2007). Thus, at some loci, DML demethylation
protects genes from genome defense pathways that are
targeting repeats and transposable elements.
Methylation within Arabidopsis genes is concen-
trated in the middle and distributed away from the 5#
and 3# ends, suggesting that 5# and 3# methylation
is detrimental to gene function (Zhang et al., 2006;
Gehring and Henikoff, 2007; Vaughn et al., 2007;
Zilberman et al., 2007). Our data and recent reports
suggest that gene ends are kept free of methylation by
two distinct mechanisms. The 5# and 3# ends of genes
generally lack siRNAs, which indicates that genic ends
are kept free of methylation by controlling where
methylation is directed (Kasschau et al., 2007). How-
ever, our data indicate that when gene ends are
targeted by siRNAs, DML demethylation can prevent
the methylation of some gene ends (Figs. 1, A and B,
and 2; Penterman et al., 2007). Thus, protecting gene
ends from methylation involves two distinct layers of
regulation: (1) siRNAs are generally not directed to the
ends of genes (Kasschau et al., 2007); and (2) DML
demethylases target gene ends that are targeted by
siRNAs.
Our data also suggest that the methylation removed
by ROS1 has multiple, distinct origins. In this study,
we examined the role of the PolIV/RDR2/DCL3/
AGO4 pathway and DRM2 in de novo methylation
of loci in ros1 mutant plants. At At1g26400 and
At5g38550 in ros1-3 plants, the PolIV/RDR2/DCL3/
AGO4 pathway and the de novo methyltransferase
DRM2 were required for non-CG hypermethylation,
whereas CG hypermethylation did not require the
PolIV/RDR2/DCL3/AGO4 pathway and DRM2 (Fig.
2, A and C). When and how CG hypermethylation is
established at these loci is not clear. However, as CG
methylation is efficiently maintained by MET1, estab-
lishment of CG methylation would only have to occur
infrequently or in progenitor cells to be maintained
indefinitely. We found that the PolIV/RDR2/DCL3/
AGO4 pathway was only partially responsible for di-
recting DRM2 de novo methylation at the At1g34245
locus, as ros1; rdr2 mutants had greater CG and CNG
methylation than ros1; drm2 mutants (Fig. 2B). This

Figure 5. RDR2 is required for ROS1 DNA demethylation. A, Inheri- (rdr2 and ros1-3; RDR2/rdr2-1) and their progeny (ROS1/ros1-3;
tance of hypermethylated loci by siblings that differ in the absence or RDR2/rdr2-1 and ROS1/ros1-3; rdr2-1). Eight to 10 clones from the
presence of an RDR2 allele. B to D, Percentage of methylation at parents were sequenced, and 17 to 20 clones from the progeny were
At1g26400, At1g34245, and At5g38550 in the parental genotypes sequenced.

Plant Physiol. Vol. 145, 2007 1555

Penterman et al.
observation implies that another distinct RNAi path-
way directs DRM2 de novo methylation to At1g34245
in ros1-3 mutant plants. Thus, ROS1 demethylation
might antagonize more than one RNAi pathway,
which is consistent with its housekeeping function
(Penterman et al., 2007).
Previously, ROS1 was shown to be transcriptionally
down-regulated in plants with mutations in RDR2,
DRD1, and the PolIV genes (Huettel et al., 2006). This
effect on ROS1 expression appears to be specific to
mutations that disrupt the PolIV/RDR2/DCL3/
AGO4 RNAi pathway, because mutations in RDR6,
which functions in another RNAi pathway, do not
affect ROS1 expression (Fig. 4). ROS1 was also down-
regulated in plants with mutations in the de novo
methyltransferase DRM2 (Fig. 4). Because RDR2, DRD1,
PolIV genes, and DRM2 encode enzymes that function
at different, critical steps of the PolIV/RDR2/DCL3/
AGO4 pathway (Li et al., 2006; Pontes et al., 2006),
these observations suggest that it is the function of the
pathway, rather than any one component, that is
important for ROS1 expression. Our data also show
that the PolIV/RDR2/DCL3/AGO4 pathway is neces-
sary for ROS1 demethylation. In drm2 and rdr2 mutants,
loci demethylated by ROS1 accumulate methylation
at CG sites (Fig. 3, A–C), and hypermethylated al-
leles inherited from the parents are demethylated by
ROS1 in progeny that have a wild-type RDR2 allele,
whereas these alleles remain hypermethylated in ho-
mozygous rdr2 siblings (Fig. 4, A–D). Thus, down-
regulation of ROS1 in rdr2 and drm2 mutants results in
minimal ROS1 demethylation and a unique methyl-
ation pattern at its targets. Identifying the mechanisms
of ROS1 down-regulation by mutations that disrupt
the PolIV/RDR2/DCL3/AGO4 RNAi pathway is
crucial to determining its functional significance in
Arabidopsis.
MATERIALS AND METHODS
Transposon, Repeat, and siRNA Analysis
The data used to search for transposable elements, repeats, and siRNAs
were described previously (Gustafson et al., 2005; Lu et al., 2005; Rajagopalan
et al., 2006; Kasschau et al., 2007; Zhang et al., 2007) and can be accessed at the
following databases: http://asrp.cgrb.oregonstate.edu, http://epigenomics.
mcdb.ucla.edu/smallRNAs, and http://mpss.udel.edu/at. The databases
were queried using the coordinates for each DML target locus (Penterman
et al., 2007). Size information for siRNAs was available for most loci except
ones whose siRNAs were only reported by Lu et al. (2005). Annotation of
transposable elements was based on RepBase v10.0.1 (Jurka, 2000).
Plant Materials
The descriptions and genotyping protocols for ros1-3, dml2-1, and dml3-1
were previously described (Penterman et al., 2007). The rdr2-1 and rdr6-11
were obtained from the Arabidopsis Biological Resource Center (Ohio State
University). The drm2 allele was a kind gift from Steve Jacobsen (University of
California, Los Angeles). The descriptions and genotyping protocols for rdr2-1,
rdr6-11, and drm2 were previously described (Cao and Jacobsen, 2002; Peragine
et al., 2004; Xie et al., 2004). The drm2 allele was introgressed into the
Columbia-0 background four times prior to use in the experiments. All genetic
1556
experiments involving ros1-3 with rdr2-1 or drm-1 were done in the ros1-3;
dml2-1; dml3-1 background.
Bisulfite Sequencing
Methods and primer sequences used in bisulfite sequencing experiments
were previously described (Penterman et al., 2007). Around eight to 10 clones
were sequenced for all experiments unless noted otherwise.
RT-PCR Analysis of Genes Demethylated ROS1
Twenty-day-old plants were used to isolate total RNA as described (Choi
et al., 2002). RT-PCR conditions and ACTIN11 primers were as previously
described (Penterman et al., 2007). Primers for amplification of ROS1 tran-
scripts were ROS13F (5#-TGGAAGGGATCCGTCGTGGATTCT-3#) and
ROS1R (5#-CCATCTTGCTGAGAGCAACA-3#).
ACKNOWLEDGMENT
We thank Christian Ibarra for help with bisulfite sequencing.
Received August 21, 2007; accepted September 24, 2007; published October 19,
2007.
LITERATURE CITED
Agius F, Kapoor A, Zhu JK (2006) Role of the Arabidopsis DNA glyco-
sylase/lyase ROS1 in active DNA demethylation. Proc Natl Acad Sci
USA 103: 11796–11801
Bender J (2004) DNA methylation and epigenetics. Annu Rev Plant Biol 55:
41–68
Cao X, Jacobsen SE (2002) Role of the arabidopsis DRM methyltransferases
in de novo DNA methylation and gene silencing. Curr Biol 12: 1138–1144
Chan SW, Henderson IR, Jacobsen SE (2005) Gardening the genome: DNA
methylation in Arabidopsis thaliana. Nat Rev Genet 6: 351–360
Chan SW, Henderson IR, Zhang X, Shah G, Chien JS, Jacobsen SE (2006)
RNAi, DRD1, and histone methylation actively target developmentally
important non-CG DNA methylation in arabidopsis. PLoS Genet 2: e83
Chan SW, Zhang X, Bernatavichute YV, Jacobsen SE (2006) Two-step
recruitment of RNA-directed DNA methylation to tandem repeats. PLoS
Biol 4: 1923–1933
Chan SW, Zilberman D, Xie Z, Johansen LK, Carrington JC, Jacobsen SE
(2004) RNA silencing genes control de novo DNA methylation. Science
303: 1336
Choi Y, Gehring M, Johnson L, Hannon M, Harada JJ, Goldberg RB,
Jacobsen SE, Fischer RL (2002) DEMETER, a DNA glycosylase domain
protein, is required for endosperm gene imprinting and seed viability in
arabidopsis. Cell 110: 33–42
Gehring M, Henikoff S (2007) DNA methylation dynamics in plant
genomes. Biochim Biophys Acta 1769: 276–286
Gehring M, Huh JH, Hsieh TF, Penterman J, Choi Y, Harada JJ, Goldberg
RB, Fischer RL (2006) DEMETER DNA glycosylase establishes MEDEA
polycomb gene self-imprinting by allele-specific demethylation. Cell
124: 495–506
Gong Z, Morales-Ruiz T, Ariza RR, Roldan-Arjona T, David L, Zhu JK
(2002) ROS1, a repressor of transcriptional gene silencing in Arabidop-
sis, encodes a DNA glycosylase/lyase. Cell 111: 803–814
Gustafson AM, Allen E, Givan S, Smith D, Carrington JC, Kasschau KD
(2005) ASRP: the Arabidopsis Small RNA Project Database. Nucleic
Acids Res 33: D637–D640
Herr AJ, Jensen MB, Dalmay T, Baulcombe DC (2005) RNA polymerase IV
directs silencing of endogenous DNA. Science 308: 118–120
Huettel B, Kanno T, Daxinger L, Aufsatz W, Matzke AJ, Matzke M (2006)
Endogenous targets of RNA-directed DNA methylation and Pol IV in
Arabidopsis. EMBO J 25: 2828–2836
Huettel B, Kanno T, Daxinger L, Bucher E, van der Winden J, Matzke AJ,
Matzke M (2007) RNA-directed DNA methylation mediated by DRD1
and Pol IVb: a versatile pathway for transcriptional gene silencing in
plants. Biochim Biophys Acta 1769: 358–374
Jurka J (2000) Repbase update: a database and an electronic journal of
repetitive elements. Trends Genet 16: 418–420
Plant Physiol. Vol. 145, 2007

Kanno T, Huettel B, Mette MF, Aufsatz W, Jaligot E, Daxinger L, Kreil DP,
Matzke M, Matzke AJ (2005) Atypical RNA polymerase subunits
required for RNA-directed DNA methylation. Nat Genet 37: 761–765
Kanno T, Mette MF, Kreil DP, Aufsatz W, Matzke M, Matzke AJ (2004)
Involvement of putative SNF2 chromatin remodeling protein DRD1 in
RNA-directed DNA methylation. Curr Biol 14: 801–805
Kasschau KD, Fahlgren N, Chapman EJ, Sullivan CM, Cumbie JS, Givan
SA, Carrington JC (2007) Genome-wide profiling and analysis of
Arabidopsis siRNAs. PLoS Biol 5: e57
Li CF, Pontes O, El-Shami M, Henderson IR, Bernatavichute YV, Chan
SW, Lagrange T, Pikaard CS, Jacobsen SE (2006) An ARGONAUTE4-
containing nuclear processing center colocalized with Cajal bodies in
Arabidopsis thaliana. Cell 126: 93–106
Lu C, Kulkarni K, Souret FF, MuthuValliappan R, Tej SS, Poethig RS,
Henderson IR, Jacobsen SE, Wang W, Green PJ, et al (2006) MicroRNAs
and other small RNAs enriched in the Arabidopsis RNA-dependent
RNA polymerase-2 mutant. Genome Res 16: 1276–1288
Lu C, Tej SS, Luo S, Haudenschild CD, Meyers BC, Green PJ (2005)
Elucidation of the small RNA component of the transcriptome. Science
309: 1567–1569
Morales-Ruiz T, Ortega-Galisteo AP, Ponferrada-Marin MI, Martinez-
Macias MI, Ariza RR, Roldan-Arjona T (2006) DEMETER and
REPRESSOR OF SILENCING 1 encode 5-methylcytosine DNA glyco-
sylases. Proc Natl Acad Sci USA 103: 6853–6858
Penterman J, Zilberman D, Huh JH, Ballinger T, Henikoff S, Fischer RL
(2007) DNA demethylation in the Arabidopsis genome. Proc Natl Acad
Sci USA 104: 6752–6757
Peragine A, Yoshikawa M, Wu G, Albrecht HL, Poethig RS (2004) SGS3
and SGS2/SDE1/RDR6 are required for juvenile development and the
production of trans-acting siRNAs in Arabidopsis. Genes Dev 18:
2368–2379
Plant Physiol. Vol. 145, 2007
DNA Demethylation and Methylation
Pontes O, Li CF, Nunes PC, Haag J, Ream T, Vitins A, Jacobsen SE,
Pikaard CS (2006) The Arabidopsis chromatin-modifying nuclear
siRNA pathway involves a nucleolar RNA processing center. Cell 126:
79–92
Rajagopalan R, Vaucheret H, Trejo J, Bartel DP (2006) A diverse and
evolutionarily fluid set of microRNAs in Arabidopsis thaliana. Genes
Dev 20: 3407–3425
Vaughn MW, Tanurd Ic M, Lippman Z, Jiang H, Carrasquillo R,
Rabinowicz PD, Dedhia N, McCombie WR, Agier N, Bulski A, et al
(2007) Epigenetic natural variation in Arabidopsis thaliana. PLoS Biol
5: e174
Xie Z, Johansen LK, Gustafson AM, Kasschau KD, Lellis AD, Zilberman
D, Jacobsen SE, Carrington JC (2004) Genetic and functional diversi-
fication of small RNA pathways in plants. PLoS Biol 2: E104
Zhang X, Henderson IR, Lu C, Green PJ, Jacobsen SE (2007) Role of RNA
polymerase IV in plant small RNA metabolism. Proc Natl Acad Sci USA
104: 4536–4541
Zhang X, Yazaki J, Sundaresan A, Cokus S, Chan SW, Chen H, Henderson
IR, Shinn P, Pellegrini M, Jacobsen SE, et al (2006) Genome-wide high-
resolution mapping and functional analysis of DNA methylation in
arabidopsis. Cell 126: 1189–1201
Zheng X, Zhu J, Kapoor A, Zhu JK (2007) Role of Arabidopsis AGO6 in
siRNA accumulation, DNA methylation and transcriptional gene si-
lencing. EMBO J 26: 1691–1701
Zhu J, Kapoor A, Sridhar VV, Agius F, Zhu JK (2007) The DNA
glycosylase/lyase ROS1 functions in pruning DNA methylation pat-
terns in Arabidopsis. Curr Biol 17: 54–59
Zilberman D, Gehring M, Tran RK, Ballinger T, Henikoff S (2007)
Genome-wide analysis of Arabidopsis thaliana DNA methylation un-
covers an interdependence between methylation and transcription. Nat
Genet 39: 61–69
1557