Small RNAs prevent transcription-coupled loss of histone H3 lysine 9 methylation in Arabidopsis thaliana.
ABSTRACT In eukaryotes, histone H3 lysine 9 methylation (H3K9me) mediates silencing of invasive sequences to prevent deleterious consequences including the expression of aberrant gene products and mobilization of transposons. In Arabidopsis thaliana, H3K9me maintained by SUVH histone methyltransferases (MTases) is associated with cytosine methylation (5meC) maintained by the CMT3 cytosine MTase. The SUVHs contain a 5meC binding domain and CMT3 contains an H3K9me binding domain, suggesting that the SUVH/CMT3 pathway involves an amplification loop between H3K9me and 5meC. However, at loci subject to read-through transcription, the stability of the H3K9me/5meC loop requires a mechanism to counteract transcription-coupled loss of H3K9me. Here we use the duplicated PAI genes, which stably maintain SUVH-dependent H3K9me and CMT3-dependent 5meC despite read-through transcription, to show that when PAI sRNAs are depleted by dicer ribonuclease mutations, PAI H3K9me and 5meC levels are reduced and remaining PAI 5meC is destabilized upon inbreeding. The dicer mutations confer weaker reductions in PAI 5meC levels but similar or stronger reductions in PAI H3K9me levels compared to a cmt3 mutation. This comparison indicates a connection between sRNAs and maintenance of H3K9me independent of CMT3 function. The dicer mutations reduce PAI H3K9me and 5meC levels through a distinct mechanism from the known role of dicer-dependent sRNAs in guiding the DRM2 cytosine MTase because the PAI genes maintain H3K9me and 5meC at levels similar to wild type in a drm2 mutant. Our results support a new role for sRNAs in plants to prevent transcription-coupled loss of H3K9me.
[show abstract] [hide abstract]
ABSTRACT: Gene silencing in eukaryotes is associated with the formation of heterochromatin, a complex of proteins and DNA that block transcription. Heterochromatin is characterized by the methylation of cytosine nucleotides of the DNA, the methylation of histone H3 at lysine 9 (H3 Lys 9), and the specific binding of heterochromatin protein 1 (HP1) to methylated H3 Lys 9 (refs 1-7). Although the relationship between these chromatin modifications is generally unknown, in the fungus Neurospora crassa, DNA methylation acts genetically downstream of H3 Lys 9 methylation. Here we report the isolation of KRYPTONITE, a methyltransferase gene specific to H3 Lys 9, identified in a mutant screen for suppressors of gene silencing at the Arabidopsis thaliana SUPERMAN (SUP) locus. Loss-of-function kryptonite alleles resemble mutants in the DNA methyltransferase gene CHROMOMETHYLASE3 (CMT3), showing loss of cytosine methylation at sites of CpNpG trinucleotides (where N is A, C, G or T) and reactivation of endogenous retrotransposon sequences. We show that CMT3 interacts with an Arabidopsis homologue of HP1, which in turn interacts with methylated histones. These data suggest that CpNpG DNA methylation is controlled by histone H3 Lys 9 methylation, through interaction of CMT3 with methylated chromatin.Nature 05/2002; 416(6880):556-60. · 36.28 Impact Factor
Article: An Arabidopsis SET domain protein required for maintenance but not establishment of DNA methylation.[show abstract] [hide abstract]
ABSTRACT: Cytosine methylation is critical for correct development and genome stability in mammals and plants. In order to elucidate the factors that control genomic DNA methylation patterning, a genetic screen for mutations that disrupt methylation-correlated silencing of the endogenous gene PAI2 was conducted in Arabidopsis: This screen yielded seven loss-of-function alleles in a SET domain protein with histone H3 Lys9 methyltransferase activity, SUVH4. The mutations conferred reduced cytosine methylation on PAI2, especially in non-CG sequence contexts, but did not affect methylation on another PAI locus carrying two genes arranged as an inverted repeat. Moreover, an unmethylated PAI2 gene could be methylated de novo in the suvh4 mutant background. These results suggest that SUVH4 is involved in maintenance but not establishment of methylation at particular genomic regions. In contrast, a heterochromatin protein 1 homolog, LHP1, had no effect on PAI methylation.The EMBO Journal 01/2003; 21(24):6842-52. · 9.20 Impact Factor
Article: H3 lysine 9 methylation is maintained on a transcribed inverted repeat by combined action of SUVH6 and SUVH4 methyltransferases.[show abstract] [hide abstract]
ABSTRACT: Transcribed inverted repeats are potent triggers for RNA interference and RNA-directed DNA methylation in plants through the production of double-stranded RNA (dsRNA). For example, a transcribed inverted repeat of endogenous genes in Arabidopsis thaliana, PAI1-PAI4, guides methylation of itself as well as two unlinked duplicated PAI genes, PAI2 and PAI3. In previous work, we found that mutations in the SUVH4/KYP histone H3 lysine 9 (H3 K9) methyltransferase cause a loss of DNA methylation on PAI2 and PAI3, but not on the inverted repeat. Here we use chromatin immunoprecipitation analysis to show that the transcribed inverted repeat carries H3 K9 methylation, which is maintained even in an suvh4 mutant. PAI1-PAI4 H3 K9 methylation and DNA methylation are also maintained in an suvh6 mutant, which is defective for a gene closely related to SUVH4. However, both epigenetic modifications are reduced at this locus in an suvh4 suvh6 double mutant. In contrast, SUVH6 does not play a significant role in maintenance of H3 K9 or DNA methylation on PAI2, transposon sequences, or centromere repeat sequences. Thus, SUVH6 is preferentially active at a dsRNA source locus versus targets for RNA-directed chromatin modifications.Molecular and Cellular Biology 01/2006; 25(23):10507-15. · 5.53 Impact Factor
Small RNAs Prevent Transcription-Coupled Loss of
Histone H3 Lysine 9 Methylation in Arabidopsis thaliana
Raymond A. Enke1, Zhicheng Dong2, Judith Bender2*
1Wilmer Eye Institute, Department of Ophthalmology, Johns Hopkins University School of Medicine, Baltimore, Maryland, United States of America, 2Department of
Molecular Biology, Cell Biology, and Biochemistry, Brown University, Providence, Rhode Island, United States of America
In eukaryotes, histone H3 lysine 9 methylation (H3K9me) mediates silencing of invasive sequences to prevent deleterious
consequences including the expression of aberrant gene products and mobilization of transposons. In Arabidopsis thaliana,
H3K9me maintained by SUVH histone methyltransferases (MTases) is associated with cytosine methylation (5meC)
maintained by the CMT3 cytosine MTase. The SUVHs contain a 5meC binding domain and CMT3 contains an H3K9me
binding domain, suggesting that the SUVH/CMT3 pathway involves an amplification loop between H3K9me and 5meC.
However, at loci subject to read-through transcription, the stability of the H3K9me/5meC loop requires a mechanism to
counteract transcription-coupled loss of H3K9me. Here we use the duplicated PAI genes, which stably maintain SUVH-
dependent H3K9me and CMT3-dependent 5meC despite read-through transcription, to show that when PAI sRNAs are
depleted by dicer ribonuclease mutations, PAI H3K9me and 5meC levels are reduced and remaining PAI 5meC is
destabilized upon inbreeding. The dicer mutations confer weaker reductions in PAI 5meC levels but similar or stronger
reductions in PAI H3K9me levels compared to a cmt3 mutation. This comparison indicates a connection between sRNAs and
maintenance of H3K9me independent of CMT3 function. The dicer mutations reduce PAI H3K9me and 5meC levels through
a distinct mechanism from the known role of dicer-dependent sRNAs in guiding the DRM2 cytosine MTase because the PAI
genes maintain H3K9me and 5meC at levels similar to wild type in a drm2 mutant. Our results support a new role for sRNAs
in plants to prevent transcription-coupled loss of H3K9me.
Citation: Enke RA, Dong Z, Bender J (2011) Small RNAs Prevent Transcription-Coupled Loss of Histone H3 Lysine 9 Methylation in Arabidopsis thaliana. PLoS
Genet 7(10): e1002350. doi:10.1371/journal.pgen.1002350
Editor: Gregory P. Copenhaver, The University of North Carolina at Chapel Hill, United States of America
Received March 25, 2011; Accepted August 30, 2011; Published October 27, 2011
Copyright: ? 2011 Enke et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by National Institutes of Health Grant GM-61148 to JB and by training grant T32 CA-09110 to RAE (http://www.nih.gov/). The
funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: email@example.com
The eukaryotic cell is under constant threat from transposons
and other invasive sequences. Transposons can drain cellular
resources for RNA and protein synthesis and can damage the cell
through expression of aberrant gene products or activation of
transposon movement. A major mechanism to protect against these
deleterious effects is to target transposons and other repetitive
sequences for silencing mediated through chromatin modifications.
In most eukaryotes, transposonchromatinis markedbymethylation
ofhistone H3atthelysine9 position(H3K9me).Insomeeukaryotes
includingmammals andplantstransposon chromatinis alsomarked
by cytosine methylation (5meC). An important question is how
H3K9me and 5meC are accurately maintained on transposons but
not on host genes.
A conserved strategy to maintain H3K9me and 5meC is to use
the modifications as methyltransferase (MTase) binding recognition
motifs. For example, in Arabidopsis thaliana, dimethylation of H3K9
(H3K9me2) maintained by three partially redundant histone
MTases—SUVH4 (also known as KYP, At5g13960), SUVH5
(At2g35160), and SUVH6 (At2g22740)—is associated with 5meC
maintained by the CMT3 cytosine MTase (At1g69770) [1–4]. The
SUVH MTases contain a 5meC binding domain and CMT3
contains an H3K9me binding domain, suggesting that the SUVH/
CMT3 pathway involves an amplification loop that can perpetuate
both H3K9me and 5meC [5,6]. Consistent with this model,
mutations in the CMT3 or MET1 (At5g49160) cytosine MTases,
which act to maintain 5meC in non-CG and CG sequence contexts
respectively, result in reduced H3K9me2 levels on transposons and
repetitive sequences [3,7–10]. In addition, a suvh4 suvh5 suvh6 triple
H3 K9 MTase mutant displays similar reduced non-CG methyl-
ation patterns to a cmt3 mutant . Although the H3K9me/5meC
amplification loop provides a mechanism to stably maintain both
modifications in untranscribed regions of the genome, at junctions
where modified sequences are transcribed through from nearby
unmodified promoters, H3K9me can be removed by transcription-
associated histone replacement or histone demethylation [7,11].
What prevents transcriptional destabilization of H3K9me patterns?
Duplicated Arabidopsis genes encoding the tryptophan synthe-
sis enzyme phosphoribosylanthranilate isomerase (PAI) provide an
ideal system to understand the balance between transcription and
SUVH-mediated H3K9me2/CMT3-mediated 5meC. In most
Arabidopsis strains there are three unlinked PAI gene duplications
that lack 5meC . However, in the Wassilewskija (Ws) strain
one of the PAI loci is rearranged as a tail-to-tail inverted repeat
(IR) of two genes PAI1–PAI4 (At1g07780), which triggers the
recognition of PAI sequences as invaders. The PAI1–PAI4 IR as
well as two unlinked singlet genes PAI2 (At5g05590) and PAI3
(At1g29410) are modified by H3K9me2 and 5meC, coextensive
with their regions of shared sequence identity [3,13] (see Figure S1
PLoS Genetics | www.plosgenetics.org1October 2011 | Volume 7 | Issue 10 | e1002350
for PAI gene maps). The PAI1–PAI4 IR is fused to a heterologous
promoter with a transcription start site approximately 500 base
pairs (bp) upstream of the PAI1 5meC boundary, which drives
constitutive expression of PAI1 transcripts . The polyadenylated
transcripts that accumulate from this locus consist of a majority class
that terminates normally in the PAI1 39 untranslated region at the
center of the IR and a minority class that extends through PAI1 into
palindromic PAI4 sequences to provide a source of fold-back
double-stranded RNA (dsRNA). Therefore the PAI1–PAI4 locus is
able to stably maintain H3K9me2 and 5meC on the IR sequences
even in the face of substantial read-through transcription. The PAI2
and PAI3 singlet genes also stably maintain H3K9me2 and 5meC
even though they are likely to be only partially silenced by limited
upstream modifications: at PAI2 5meC extends only 250 bp
upstream of the predicted transcription start site, and at PAI3
5meC extends only as far as the predicted transcription start site
Arabidopsis uses three cytosine MTase pathways to control
5meC: the CMT3 pathway maintains 5meC mainly in non-CG
contexts in conjunction with the SUVH H3K9 MTases, the
MET1 pathway maintains 5meC mainly in CG contexts, and the
DRM2 (At5g14620) pathway initiates 5meC on new invasive
sequences under the guidance of small RNAs (sRNAs), as well as
contributing to maintenance of non-CG methylation at some loci
. In a cmt3 or a suvh4 suvh5 suvh6 mutant, the Ws PAI genes are
depleted for 5meC in non-CG contexts [4,16]. In addition, in a
cmt3 met1 double mutant the PAI genes are depleted for 5meC in
all contexts . Therefore, the DRM2 pathway plays a minimal
role in the maintenance of PAI 5meC patterns. However, genetic
or epigenetic changes that impair the production of transcripts
that read through from PAI1 into palindromic PAI4 sequences at
the PAI1–PAI4 IR cause reduced levels of PAI 5meC in non-CG
contexts [13,14,17,18]. In light of these results, we hypothesized
that sRNAs processed from dsRNAs might underlie a mechanism
to prevent the loss of SUVH/CMT3-mediated modifications due
to read-through transcription, independently of the role for sRNAs
in guiding DRM2.
To test the hypothesis that sRNAs control the SUVH/CMT3
pathway, we used mutations in Arabidopsis dicer-like (DCL)
ribonucleases to block processing of sRNAs from dsRNAs, and
monitored the effects on Ws PAI gene H3K9me2 and 5meC levels.
Arabidopsis encodes four DCLs (reviewed in ). DCL1
(At1g01040) is specialized for processing 21 nucleotide (nt)
microRNAs (miRNAs) needed for developmental gene regulation,
whereas DCL2 (At3g03300), DCL3 (At3g43920), and DCL4
(At5g20320) have partially redundant roles in processing sRNAs
used in other silencing pathways. For example, DCL3 processes
24 nt sRNAs used to guide DRM2 to matching target sequences
such as transgene insertions and transposons [20,21]. In a dcl3
mutant DCL2 and DCL4 can partially compensate by processing
22 nt and 21 nt sRNAs respectively corresponding to the same
genomic target sequences [22,23].
Here we show that the dcl2 dcl3 dcl4 mutant has reduced levels of
H3K9me2 and non-CG methylation on PAI sequences relative to
wild type, corresponding to loss of PAI sRNAs. We also show that a
drm2 mutant maintains similar levels of PAI H3K9me2 and 5meC
relative to wild type. Therefore the PAI genes illustrate that DCL-
dependent sRNAs help maintain SUVH/CMT3-mediated modi-
fications through a distinct mechanism from their role in guiding
DRM2. In the dcl mutant there is a weaker reduction in PAI 5meC
levels but a similar or stronger reduction in PAI H3K9me2 levels
compared to a cmt3 mutant, indicating a connection between
sRNAs and maintenance of H3K9me2 patterns independent of
CMT3 function. We also show that upon inbreeding in the absence
of DCL function, remaining PAI 5meC is destabilized. Our results
reveal a new pathway for sRNA control of H3K9me2 and asso-
mechanism to use a product of read-through transcription—
sRNAs—as a means to counteract transcription-coupled loss of
H3K9me2 on transposons and repeats.
Dicer mutations cause reduced levels of PAI non-CG
To test whether dicer-dependent sRNAs contribute to mainte-
nance of PAI 5meC patterns, we generated strains where dicer
mutations were combined with the three methylated PAI loci from
Ws and assayed PAI 5meC patterns using both DNA gel blot and
bisulfite sequencing assays. For PAI DNA gel blot analysis we
cleaved genomic DNA with each of three 5meC-sensitive
restriction enzymes that have cleavage sites within methylated
PAI sequences: HincII (sensitive to methylation of the outermost
non-CG cytosines in 59 atGTCAACag 39, where the recognition
sequence is shown in uppercase), MspI (sensitive to methylation of
the outer non-CG cytosines in 59 CCGG 39, and HpaII (sensitive
to methylation of either the inner CG or outer non-CG cytosines
in 59 CCGG 39). HincII cleaves at the translational start codons of
PAI1, PAI2 and PAI4 but not PAI3, and the MspI/HpaII
isoschizomers cleave in the second introns of PAI2, PAI3, and
PAI4 but not PAI1 (Figure S1).
We found that genomic DNA prepared from dcl2, dcl3, and dcl4
single insertional null mutants and the dcl2 dcl4 and dcl3 dcl4 double
mutants had similar PAI cleavage patterns to wild type Ws genomic
DNA when assessed by HincII, MspI, or HpaII DNA gel blot assays
(Figure 1). In contrast, genomic DNA prepared from the dcl2 dcl3
and dcl2 dcl3 dcl4 mutants displayed increased cleavage with HincII
at PAI1–PAI4 and PAI2, and with MspI at PAI1–PAI4, PAI2, and
PAI3 relative to wild type Ws, diagnostic of partially reduced non-
CG methylation levels at all three PAI loci. Bisulfite sequencing of
PAI1 and PAI2 proximal promoter/first exon regions in the dcl2 dcl3
dcl4 mutant compared to wild type Ws and Ws cmt3 showed that
there was a partial loss of 5meC in CHG and CHH contexts.
Therefore, the bisulfite sequencing data are consistent with the
DNA gel blot assays. The results indicate that DCL2 and DCL3 act
redundantly to maintain PAI non-CG methylation patterns.
Methylation of histone H3 at the lysine 9 position
(H3K9me) is a fundamental chromatin modification that
suppresses expression from invasive and repetitive se-
quences such as transposons. In plant genomes, regions
modified by H3K9me are maintained with precise bound-
aries. However, at junctions where H3K9me target regions
are subject to read-through transcription from outside
promoters, the stability of H3K9me patterns is jeopardized
by transcription-coupled processes that remove this
modification. We show that maintenance of H3K9me
patterns at such vulnerable sites requires small RNAs
corresponding to the H3K9me target region. We use a
sensitive reporter system to show that, in the absence of
small RNAs, target regions subject to read-through
H3K9me levels, followed by further losses in progeny
plants upon inbreeding. Our results support a new
function for small RNAs in maintaining accurate H3K9me
patterns in the plant genome.
immediate reduction in
Small RNAs Prevent Loss of H3K9me
PLoS Genetics | www.plosgenetics.org2 October 2011 | Volume 7 | Issue 10 | e1002350
To determine whether the miRNA processing dicer DCL1
contributes to the remaining PAI non-CG methylation in the dcl2
dcl3 dcl4 triple mutant relative to cmt3, we included a dcl1 dcl2 dcl3
dcl4 quadruple mutant strain in the DNA gel blot analysis
(Figure 1). Because dcl1 null alleles are embryo-lethal we used a
partial-function dcl1-9 allele that is viable but female-sterile
[24,25]. The dcl1 dcl2 dcl3 dcl4 mutant displayed similar cleavage
patterns to the dcl2 dcl3 dcl4 mutant, indicating that the dcl1-9
mutation does not enhance the partial loss of 5meC conferred by
mutation of the other three DCL genes. In subsequent studies we
focused on the dcl2 dcl3 dcl4 mutant, which has global depletion of
sRNAs other than miRNAs .
For comparison to the dcl mutants we included genomic DNA
prepared from cytosine MTase mutants in the Ws background
(Figure 1). DRM2 is the major cytosine MTase controlling
initiation of 5meC, but the related DRM1 MTase (At5g15380)
could also contribute to this pathway . Therefore we used a
drm1 drm2 double null insertional mutant. DNA from the Ws drm1
drm2 mutant displayed similar PAI cleavage patterns to wild type
Ws in all three DNA gel blot assays, and similar 5meC patterns to
wild type Ws in bisulfite sequencing analysis of PAI1 and PAI2
proximal promoter regions. DNA from a Ws met1 mutant
displayed increased cleavage at all three PAI loci with HpaII,
and partially increased cleavage at PAI1–PAI4 and PAI2 with
MspI, but no difference from wild type PAI cleavage patterns with
HincII in DNA gel blot assays, diagnostic of a partial loss of 5meC
in CG and CCG contexts. DNA from the cmt3 mutant displayed
nearly complete cleavage with HincII and MspI, and partially
increased cleavage with HpaII in DNA gel blot assays, diagnostic
of strong loss of 5meC mainly in CHG and CHH contexts.
Compared to the cytosine MTase mutants across the three DNA
gel blot assays, the dcl mutant PAI demethylation phenotypes are
consistent with a partial defect in the SUVH/CMT3 pathway
rather than a defect in the DRM or MET1 pathways.
DCL3 and DRM2 are key factors in establishing new 5meC
imprints. We used a previously developed genetic assay combining
the Ws PAI IR dsRNA source locus with an unmethylated PAI2
target gene from another strain background to show that dcl3 and
drm1 drm2 mutations impair the acquisition of new 5meC on PAI2
(Text S1, Figure S2). Therefore the PAI genes use the same
DCL3/DRM pathway for establishing 5meC imprints as other
characterized loci. However, once PAI 5meC patterns are
established, the DCL3/DRM pathway plays a minimal role in
long-term maintenance (Figure 1).
The dcl2 dcl3 dcl4 mutant has reduced levels of PAI
We used chromatin immunoprecipitation (ChIP) analysis with
H3K9me2-specific antibodies on chromatin prepared from the
dcl2 dcl3 dcl4 mutant compared to chromatin prepared from wild
type, suvh4 suvh5 suvh6, cmt3, drm1 drm2, or cmt3 drm1 drm2 strains to
determine whether the dcl mutations affect levels of H3K9me2 as
well as non-CG methylation on PAI sequences. Chromatin was
analyzed by quantitative PCR with primer pairs specific for the
PAI1 arm of the PAI1–PAI4 IR locus or the PAI2 singlet gene.
At both PAI1–PAI4 and PAI2 the dcl2 dcl3 dcl4 mutant had
reduced levels of H3K9me2 relative to wild type, although not as
strongly as in the suvh4 suvh5 suvh6 H3K9 MTase mutant (Figure 2).
Comparing the ChIP results to the assays for 5meC (Figure 1), the
reduced PAI H3K9me2 levels in the dcl2 dcl3 dcl4 mutant are still
sufficient to support substantial CMT3 activity. Therefore CMT3
might be able to use even sparsely distributed H3K9me2 as a
Figure 1. PAI non-CG methylation levels are reduced in dcl
mutants. (A), (B), and (C) DNA gel blot assays for PAI 5meC. Genomic
DNA from the indicated strains was cleaved with (A) HincII, (B) MspI, or
(C) HpaII (isoschizomer of MspI) and used in DNA gel blot analysis with a
PAI cDNA probe . PAI restriction maps are shown in Figure S1. P1–P4
indicates PAI1–PAI4, P2 indicates PAI2, and P3 indicates PAI3, with bands
diagnostic of 5meC on PAI-internal sites marked with asterisks. Lengths
of cleaved fragments in kilobases (kb) are indicated along the right
margins. Multiple mutant drm or dcl combinations are indicated with a
+ sign separating each mutation; for example, drm1+2 indicates a drm1
drm2 mutant and dcl2+3 indicates a dcl2 dcl3 mutant. WT indicates wild
type Ws. (D) Bisulfite sequencing of PAI1 and PAI2. Genomic DNA from
the indicated strains was treated with sodium bisulfite and used as a
template for PCR amplification of the top strands of the PAI1 or PAI2
proximal promoter/first exon regions. Eight independent clones were
sequenced for each locus. The percentage of 5meC out of total
cytosines sequenced within the region of PAI sequence identity (344 bp
for PAI1 or 338 bp for PAI2) is shown, divided into the contexts CG
(black), CHG (white), or CHH (gray). For comparison, previously
determined patterns for wild type Ws  and Ws cmt3  are shown.
drm indicates the drm1 drm2 mutant, and dcl indicates the dcl2 dcl3 dcl4
Small RNAs Prevent Loss of H3K9me
PLoS Genetics | www.plosgenetics.org3 October 2011 | Volume 7 | Issue 10 | e1002350
At both PAI loci the cmt3 mutant also had partially reduced
levels of H3K9me2, presumably because reduced 5meC levels
impair SUVH localization to PAI sequences. At PAI2 increased
transcription due to proximal promoter demethylation in a cmt3
mutant could also contribute to reduced H3K9me2 levels, perhaps
accounting for a stronger relative reduction at PAI2 than at PAI1–
PAI4 [3,16]. The dcl2 dcl3 dcl4 mutant had similar reduction in
H3K9me2 levelstothe cmt3mutantat PAI2,but astrongerreduction
at PAI1–PAI4. In contrast, the dcl2 dcl3 dcl4 mutant had weaker
reductions in PAI2 and PAI1–PAI4 non-CG methylation levels
compared to cmt3 (Figure 1). This comparison indicates that the dcl2
dcl3 dcl4 mutations impair maintenance of H3K9me2 independently
of effects on CMT3 function. If the dcl mutations acted by impairing
consequence, then the resulting reduction in H3K9me2 levels would
be expected to be less than in the cmt3 mutant.
At both PAI1–PAI4 and PAI2, the drm1 drm2 mutant displayed
similar levels of H3K9me2 to wild type, and the drm1 drm2 cmt3
mutant displayed similar levels of H3K9me2 to cmt3 (Figure 2).
Therefore, the DRM cytosine MTases do not contribute to
maintenance of PAI H3K9me2 patterns.
The dcl2 dcl3 dcl4 mutant is depleted for PAI sRNAs
To determine whether reduced PAI non-CG methylation and
H3K9me2 levels in dcl2 dcl3 dcl4 correlate with loss of PAI sRNAs,
we used RNA gel blot analysis to detect PAI sRNAs (Figure 3). As
a negative control we used a mutant derivative of Ws, Dpai1–pai4,
where the PAI1–PAI4 IR source of dsRNA DCL substrates has
been deleted by homologous recombination between flanking
direct repeat sequences . As a positive control we used the
Dpai1–pai4 strain transformed with a PAIIR transgene consisting of
an IR of approximately 700 bp of PAI cDNA sequences
transcribed by the strong constitutive Cauliflower Mosaic Virus
35S promoter . We previously determined that PAI sRNAs
could be detected in the Dpai1–pai4(PAIIR) transgenic strain but
not in wild type Ws using RNA gel blot analysis with a PAI cDNA
riboprobe. Furthermore, high-throughput sRNA sequencing in the
C24 strain that has a similar PAI1–PAI4 IR to Ws detected PAI
sRNAs at low levels . We therefore optimized detection of rare
PAI sRNAs by designing a high-affinity locked nucleic acid (LNA)
probe corresponding to the sense strand of a 35 nt sequence in the
PAI fifth exon.
In wild type Ws the LNA probe detected low levels of PAI sRNA
species of both shorter and longer sizes, between 21 and 24 nt
relative to size markers, above the background signal in the Dpai1–
pai4 negative control strain (Figure 3). This pattern is consistent
with processing of PAI1–PAI4 palindromic transcripts into sRNAs
by more than one dicer. Correspondingly, the C24 high-
throughput sequencing analysis detected PAI sRNAs between 21
and 24 nt long covering the entire IR region . The wild type
Ws levels of endogenous PAI sRNAs were comparable to levels
detected in a hundred-fold dilution of RNA prepared from the
Dpai1–pai4(PAIIR) positive control transgenic strain (Figure 3). The
transgenic strain produced mostly smaller PAI sRNAs, as
Figure 2. PAI1 and PAI2 H3K9me2 levels are reduced in the dcl2
dcl3 dcl4 mutant. Primers specific for PAI1 or PAI2 were used for
quantitative PCR amplification from total input chromatin or chromatin
immunoprecipitated with antibodies specific for H3K9me2 from the
indicated strains. WT indicates wild type Ws, dcl indicates the dcl2 dcl3
dcl4 mutant, suvh indicates the suvh4 suvh5 suvh6 mutant, drm
indicates the drm1 drm2 mutant, and ddc indicates the drm1 drm2
cmt3 mutant. PCR results are displayed as fold change relative to WT.
Results were reproduced in three biological replicates with error bars
representing standard error among biological replicates.
Figure 3. The dcl2 dcl3 dcl4 mutant is depleted for PAI sRNAs.
sRNA fractions from the indicated strains were used in RNA gel blot
analysis with a 35 nt LNA oligonucleotide probe antisense to the PAI1
fifth exon sequence (PAI, top panel). A replicate blot was probed with a
21 nt oligonucleotide antisense to miR167 (middle panel). The low
molecular weight RNA portion of the ethidium bromide (EtBr) stained
gel, representing 5S RNA and tRNA, is shown prior to transfer of RNA to
a membrane (bottom panel). WT indicates wild type Ws, dcl indicates
the dcl2 dcl3 dcl4 mutant, and D(PAIIR) indicates Dpai1–pai4
transformed with the PAIIR transgene . The D(PAIIR) RNA was
loaded at ten-fold (1:10) and a hundred-fold (1:100) dilutions relative to
the other samples.
Small RNAs Prevent Loss of H3K9me
PLoS Genetics | www.plosgenetics.org4October 2011 | Volume 7 | Issue 10 | e1002350
previously observed for this strain using a PAI cDNA riboprobe
, presumably due to differences in PAI IR expression patterns
PAI sRNA species were depleted in the dcl2 dcl3 dcl4 strain to
similar background levels as detected in Dpai1–pai4 (Figure 3). This
result supports the hypothesis that loss of PAI sRNAs underlies the
reduction in SUVH-dependent H3K9me2 and CMT3-dependent
5meC on PAI sequences. The lack of residual PAI sRNAs in dcl2
dcl3 dcl4 indicates a minimal contribution of DCL1 to generating
these species, although DCL1 is functional in processing micro-
RNAs such as miR167.
The dcl2 dcl3 dcl4 mutant has destabilized PAI2 silencing
To determine whether loss of sRNAs causes destabilization of
remaining PAI silencing modifications upon inbreeding by self-
pollination, we introduced the dcl2 dcl3 dcl4 mutations into a Ws
pai1 reporter background where silencing of the PAI2 singlet gene
can be monitored by visual inspection. In wild type Ws, PAI1
expressed from the heterologous upstream promoter is the major
source of PAI enzyme; expression of PAI2 is impaired by
H3K9me2/5meC on proximal promoter sequences and PAI3 and
PAI4 do not encode functional enzyme due to polymorphisms .
In the Ws pai1 missense mutant, the impairment of PAI2 expression
is revealed through tryptophan deficiency phenotypes including
reduced size and blue fluorescence under ultraviolet (UV) light
caused by accumulation of the tryptophan precursor anthranilate
. The stable maintenance of PAI2 H3K9me2/5meC in pai1 is
reflected in stable maintenance of blue fluorescence across gen-
erations of inbreeding. Mutations that decrease PAI2 H3K9me2
and/or 5meC levels in the Ws pai1 background, including cmt3,
met1, and suvh4, result in reduced fluorescence [2,16,28].
The initial pai1 dcl2 dcl3 dcl4 strain displayed partially reduced
PAI 5meC patterns similar to the PAI1 dcl2 dcl3 dcl4 strain
(Figure 1, Figure 4). The pai1 dcl2 dcl3 dcl4 plants were larger and
less fluorescent than pai1 plants, reflecting the partial reduction of
non-CG methylation levels on PAI2 (Figure 4).
Examination of pai1 dcl2 dcl3 dcl4 inbred populations revealed
that blue fluorescence diagnostic of PAI2 silencing was not stably
maintained. In a population of 191 pai1 dcl2 dcl3 dcl4 plants we
found two non-fluorescent segregants (1.0%). In contrast, no non-
fluorescent individuals were found in control populations of
thousands of pai1 plants, consistent with our previous results. Each
of the non-fluorescent pai1 dcl2 dcl3 dcl4 plants yielded approxi-
mately 75% non-fluorescent and 25% fluorescent second-gener-
ation progeny (54 non-fluorescent out of 70 total progeny plants
[77%] for one line and 31 non-fluorescent out of 42 total progeny
plants [74%] for another line). Approximately one third of the
second-generation non-fluorescent plants lacked remaining PAI2
5meC in a HincII DNA gel blot assay, and these individuals
yielded 100% non-fluorescent third-generation progeny, whereas
the remaining second-generation non-fluorescent plants had
partial levels of PAI2 5meC and yielded approximately 75%
non-fluorescent third-generation progeny. For example, 12 out of
28 [43%] non-fluorescent second-generation progeny from one
line had fully demethylated PAI2 phenotypes in the HincII assay
and each of these individuals yielded 100% non-fluorescent
progeny; the pai1 dcl NF line shown in Figure 4 is derived from
one of these individuals. The segregation patterns are consistent
with reduced PAI2 5meC levels and increased expression
Figure 4. PAI2 5meC is destabilized in pai1 dcl2 dcl3 dcl4. (A) Blue fluorescence diagnostic of PAI2 silencing in the pai1 background.
Representative 2.5-week-old plants of the indicated strains are shown under visible (top) or UV (bottom) light. (B) DNA gel blot assays for PAI 5meC.
Genomic DNA from the indicated strains was cleaved with MspI or HpaII isoschizomers and used in DNA gel blot analysis with a PAI cDNA probe. P1–
P4 indicates PAI1–PAI4, P2 indicates PAI2, and P3 indicates PAI3, with bands diagnostic of 5meC on PAI-internal sites marked with asterisks. Bands
corresponding to PAI2, which controls the fluorescence phenotype, are highlighted in red. Molecular weights of fragments are as shown in Figure 1.
dcl indicates the dcl2 dcl3 dcl4 mutant, F indicates fluorescent, and NF indicates non-fluorescent.
Small RNAs Prevent Loss of H3K9me
PLoS Genetics | www.plosgenetics.org5 October 2011 | Volume 7 | Issue 10 | e1002350
occurring on just one of the two chromosomes in the parental non-
fluorescent plant and being inherited in a Mendelian fashion.
DNA gel blot analysis of a non-fluorescent pai1 dcl2 dcl3 dcl4 line
indicated a nearly complete loss of CCG methylation monitored
by MspI cleavage and partially reduced CG methylation
monitored by HpaII cleavage at PAI2 relative to the fluorescent
parental line, consistent with the reversion of tryptophan
deficiency phenotypes (Figure 4). However, PAI1–PAI4 and PAI3
maintained similar 5meC patterns to the fluorescent parental line.
Therefore, in pai1 dcl2 dcl3 dcl4 the loss of PAI2 5meC and
silencing detected by the blue fluorescence screen is not coupled to
destabilization of 5meC at the other PAI loci.
Both the initial loss of PAI non-CG methylation and the
stochastic further loss of PAI2 silencing and 5meC in dcl2 dcl3 dcl4
are similar to patterns we previously observed in the Dpai1–pai4
mutant . This comparison indicates that regardless of whether
PAI sRNAs are depleted by loss of DCL function or by loss of the
source of PAI dsRNA substrates for DCL cleavage (Figure 3), PAI2
silencing is similarly destabilized. The destabilization could be due
to a combination of effects at PAI2 including impairment of
H3K9me2 maintenance, increased transcription, and impairment
of the DRM2/DCL3 pathway in resetting 5meC imprints (Text
S1, Figure S2).
The dcl2 dcl3 dcl4 mutant has reduced non-CG
methylation levels at a subset of other SUVH/CMT3
To determine whether other SUVH/CMT3 target loci besides
the PAI genes have reduced 5meC levels in the dcl2 dcl3 dcl4
mutant, we used a survey approach with MspI DNA gel blot assays
(Figure 5). We monitored representative sequences of three types:
a degenerate (86% identical) inverted repeat locus IR1074 ,
highly repetitive 5S rDNA and 180 bp centromeric sequences
(CEN), or low-copy transposons Ta3  and Mu1 . At all of
these sequences, there was no difference in MspI cleavage between
wild type and the drm1 drm2 mutant, but greatly increased cleavage
in the cmt3 mutant, indicating that CCG methylation at the
monitored MspI sites is dependent on CMT3 with a minimal
contribution from the DRM MTases. For each sequence, we
tested both the pai1 dcl2 dcl3 dcl4 parental fluorescent strain and an
isogenic non-fluorescent progeny line to determine whether these
two strains had differences in 5meC patterns at loci other than
PAI2 (Figure 4).
The IR1074, 5S rDNA, and CEN sequences showed partially
increased MspI cleavage in both of the dcl mutant strains relative to
wild type and cmt3 (Figure 5). For the highly repetitive sequences
the increased cleavage was evident as a slight shift downwards in
the peak intensity of the ladder of cleaved bands (Figure 5B).
Therefore, similarly to the PAI genes, these sequences require
DCL function for maintenance of CMT3-dependent 5meC. The
sequences had similar 5meC patterns in the fluorescent versus
non-fluorescent pai1 dcl2 dlc3 dcl4 lines, indicating that the loss of
PAI2 5meC in the non-florescent line is not coupled to more
general destabilization of 5meC.
In contrast, Ta3 and Mu1 showed no differences in MspI
cleavage between the dcl mutant lines and wild type controls
(Figure 5). The Mu1 transposon has a polymorphic arrangement
between Ws and the Columbia (Col) strain in which the dcl alleles
were originally isolated, and the dcl2 dcl3 dcl4 strain carries both
Mu1 arrangements. Despite this complication, comparisons to wild
type versus cmt3 controls in each strain background showed no
evidence of demethylation in the dcl mutant lines. Therefore, not
all SUVH/CMT3 targets display DCL-dependent maintenance of
5meC. In light of our hypothesis that sRNAs prevent loss of
H3K9me2 due to read-through transcription, the different effects
of the dcl mutations at different SUVH/CMT3 target loci could
reflect the extent to which read-through transcription occurs
across the modified sequences. For example, the dcl-sensitive locus
IR1074 is likely to be transcribed across to make fold-back dsRNA
because this locus produces sRNAs even in an RNA-dependent
RNA polymerase rdr2 mutant background .
We also monitored H3K9me2 levels at the single-locus targets
IR1074 and Ta3 by ChIP in the dcl2 dcl3 dcl4 mutant and the same
control strains used for analysis of the PAI genes (Figure 6). At
IR1074 H3K9me2 levels were reduced in the dcl2 dcl3 dcl4 mutant
relative to wild type, although not as strongly as in the suvh4 suvh5
suvh6 mutant. In contrast, at Ta3 H3K9me2 was maintained at
similar levels between the dcl2 dcl3 dcl4 mutant and wild type. The
H3K9me2 ChIP results agree with the 5meC results indicating
that full modification of IR1074 but not Ta3 depends on DCL
function (Figure 5A, Figure 5C).
The cmt3 mutant had reduced levels of H3K9me2 at both
IR1074 and Ta3, presumably due to impaired SUVH localization
and/or increased transcription caused by loss of non-CG
methylation (Figure 5, Figure 6). However, at both loci the drm1
drm2 mutant maintained H3K9me2 at similar levels to wild type,
and the drm1 drm2 cmt3 mutant maintained H3K9me2 at similar
levels to cmt3 (Figure 6). Therefore, CMT3 but not the DRM
cytosine MTases contributes to maintenance of H3K9me2
patterns at IR1074 and Ta3.
At IR1074, the dcl2 dcl3 dcl4 mutant and the cmt3 mutant
displayed similar reductions in H3K9me2 levels relative to wild
type (Figure 6). However, dcl2 dcl3 dcl4 had a weaker reduction in
IR1074 non-CG methylation levels than cmt3 (Figure 5). This
relationship is similar to that observed for PAI1–PAI4 and PAI2
(Figure 1, Figure 2), and supports the view that loss of sRNAs in
the dcl2 dcl3 dcl4 mutant impairs maintenance of H3K9me2
patterns independently of CMT3 function at loci subject to read-
In Arabidopsis, H3K9me2 maintained by the SUVH4,
SUVH5, and SUVH6 histone MTases is used to guide 5meC in
non-CG contexts maintained by the CMT3 cytosine MTase [1–
4]. The SUVH MTases contain 5meC-binding domains, and
CMT3 contains an H3K9me-binding domain, leading to the
model that the SUVH/CMT3 pathway involves an amplification
loop between 5meC and H3K9me2 [5,6]. However, this
amplification loop is not sufficient to maintain full levels of
5meC and H3K9me2 on the Ws PAI gene duplications, including
a constitutively transcribed IR locus PAI1–PAI4 and partially
silenced singlet genes PAI2 and PAI3. In previous work we showed
that production of palindromic transcripts from the PAI1–PAI4 IR
is also required for maintenance of PAI non-CG methylation
[13,14,17,18]. For example, in a Dpai1–pai4 mutant the PAI2 and
PAI3 genes have reduced non-CG methylation, and the remaining
5meC on PAI2 is destabilized upon inbreeding . Here we use
mutations in the DCL dicer ribonucleases to show that PAI sRNAs
processed from PAI dsRNAs are the key species that reinforce the
SUVH/CMT3 amplification loop between H3K9me2 and non-
Arabidopsis uses DCL-dependent sRNAs incorporated into
argonaute (AGO) effector proteins as nucleic acid sequence-
specificity guides in a variety of pathways including miRNA
control of development, RNA interference, and guidance of 5meC
mediated by the DRM2 cytosine MTase [19,23]. The DRM2
pathway contributes together with the SUVH/CMT3 pathway to
Small RNAs Prevent Loss of H3K9me
PLoS Genetics | www.plosgenetics.org6October 2011 | Volume 7 | Issue 10 | e1002350
maintenance of non-CG methylation at many 5meC target loci
[15,31]. This overlap has obscured whether sRNAs have an
independent role in the SUVH/CMT3 pathway. However, the
Ws PAI genes maintain 5meC in non-CG contexts almost entirely
through the SUVH/CMT3 pathway once initial 5meC is
established (Figure 1, Text S1, Figure S2). The reduction in PAI
non-CG methylation and H3K9me2 levels in dcl mutant
backgrounds therefore indicates a direct connection between
Figure 5. 5meC levels are reduced at a subset of CMT3 target
loci in pai1 dcl2 dcl3 dcl4. Genomic DNA from the indicated strains
was cleaved with MspI, or with MspI plus HindIII for Mu1, and used in
DNA gel blot analysis with the indicated probes. Arrowheads indicate
positions of fully cleaved species. WT indicates wild type Ws. drm
indicates the drm1 drm2 double mutant. dcl F indicates a pai1 dcl2 dcl3
dcl4 fluorescent strain and dcl NF indicates a pai1 dcl2 dcl3 dcl4 non-
fluorescent strain with PAI2 demethylation, as shown in Figure 4. (A)
IR1074 is partially demethylated in dcl mutant strains (see Figure S1 for
MspI restriction map and probe information). (B) 5S rDNA and 180 bp
centromere repeat sequences (CEN) are partially demethylated in dcl
mutant strains. (C) Ta3 and Mu1 low-copy transposons are not
demethylated in dcl mutant strains. For the Mu1 blot, black and gray
arrowheads indicate bands diagnostic of complete cleavage for the Ws
or Col Mu1 arrangements respectively.
Figure 6. Maintenance of IR1074 but not Ta3 H3K9me2 patterns
is impaired in dcl2 dcl3 dcl4. Primers specific for IR1074 or Ta3 were
used for quantitative PCR amplification from total input chromatin or
chromatin immunoprecipitated with antibodies specific for H3K9me2
from the indicated strains. WT indicates wild type Ws, dcl indicates the
dcl2 dcl3 dcl4 mutant, suvh indicates the suvh4 suvh5 suvh6 mutant, drm
indicates the drm1 drm2 mutant, and ddc indicates the drm1 drm2 cmt3
mutant. PCR results are displayed as fold change relative to WT. Results
were reproduced in three biological replicates with error bars
representing standard error among biological replicates.
Small RNAs Prevent Loss of H3K9me
PLoS Genetics | www.plosgenetics.org7 October 2011 | Volume 7 | Issue 10 | e1002350
(Figure 1, Figure 2). DCL2 and DCL3 are the key dicers required
for maintaining H3K9me2 and 5meC patterns on the PAI genes,
suggesting a preference for longer 22 and 24 nt sRNAs in this
ChIP analysis shows that the dcl2 dcl3 dcl4 mutant has reduced
H3K9me2 levels at PAI loci similar to or stronger than in the cmt3
mutant (Figure 2). However, the dcl2 dcl3 dcl4 mutant has weaker
reductions in PAI 5meC levels than the cmt3 mutant (Figure 1).
Therefore reduced H3K9me2 levels in the dcl mutant cannot be
accounted for as a secondary effect of impaired CMT3 function.
Instead, the ChIP results support the view that the dcl mutations
directly impair maintenance of H3K9me2 patterns. In this view,
the partial loss of PAI H3K9me2 in dcl2 dcl3 dcl4 reduces CMT3
localization, resulting in reduced PAI non-CG methylation as a
Similarly to the PAI genes, a subset of other SUVH/CMT3
target loci including highly repetitive 5S rDNA and CEN
sequences have partially reduced non-CG methylation levels in
the dcl2 dcl3 dcl4 mutant (Figure 5). However, some loci such as the
low copy transposons Ta3 and Mu1 can maintain full 5meC levels
relative to wild type despite the loss of DCL function. This
variation could reflect the degree to which different SUVH/
CMT3 target loci are transcribed across. This variation could also
reflect which RNA polymerases are most active at different loci.
Arabidopsis encodes five RNA polymerases: the conserved
eukaryotic RNA polymerases POLI, POLII, and POLIII, and
plant-specific POLIV and POLV implicated in targeting DRM2-
dependent 5meC . In particular, POLV is proposed to
transcribe across target loci to make ‘‘scaffold’’ transcripts that
recruit sRNA/AGO complexes and components of the DRM2
pathway [33,34]. Because of its specialized role in making
silencing-associated transcripts, POLV might be less disruptive of
H3K9me2 than other RNA polymerases designed to express host
genes. In this case, protein-encoding loci like the PAI genes that
are transcribed by RNA POLII, and 5meC targets that depend on
RNA POLII for scaffold transcript synthesis , might have a
stronger dependence on an sRNA-based mechanism to maintain
H3K9me2 than POLV-transcribed regions of the genome.
Our previous studies with allelic variants of the PAI1–PAI4 IR
locus support the hypothesis that the level of transcription across
the locus determines the extent to which sRNAs are needed for
maintenance of PAI 5meC levels. In one study we used transgene-
expressed sRNAs to direct 5meC and transcriptional silencing to
the upstream promoter that drives transcription through PAI1–
PAI4, thereby impairing production of PAI dsRNAs and sRNAs
. In this transgenic strain the PAI1–PAI4 locus was able to
maintain full 5meC levels, whereas the PAI2 and PAI3 singlet
genes had partially reduced non-CG methylation levels. These
patterns are consistent with a model where the decreased
transcription of PAI1–PAI4 specifically reduces its dependence
on PAI sRNAs. In a second study we characterized a mutant
derivative of Ws where a rearrangement in the center of the PAI1–
PAI4 IR introduces a new polyadenylation site and reduces the
levels of transcripts that extend into palindromic PAI4 sequences,
without altering promoter sequences or the level of transcription
across the locus . In the rearrangement mutant the PAI1–PAI4
IR locus as well as the PAI2 and PAI3 singlet genes had partially
reduced non-CG methylation levels. These patterns are consistent
with a model where read-through transcription at all three loci
together with reduced PAI sRNAs results in loss of H3K9me2 and
5meC at all three loci, similarly to the situation in the dcl2 dcl3 dcl4
mutant. Taken together, our results support a homeostatic
mechanism where sRNAs produced from heterochromatic regions
by read-through transcription feed back to counteract depletion of
H3K9me2 and associated 5meC levels caused by read-through
The mechanistic relationship between sRNAs and maintenance
of H3K9me2 patterns remains to be determined. In the fission
yeast Schizosaccharomyces pombe, an sRNA-loaded AGO protein in
the RITS effector complex interacts with nascent transcripts at
centromeric repeats to recruit the Clr4 H3K9 MTase (reviewed in
). Plants could use an analogous effector complex interaction
mechanism to target SUVH H3K9 MTases to specific regions of
the genome. Consistent with this possibility, in a suvh4 suvh5
mutant background, the remaining SUVH6 MTase maintains
levels of H3K9me2 and associated 5meC similar to wild type at
the PAI1–PAI4 IR but not at the PAI2 singlet gene ; this locus-
specific activity could reflect preferential interactions between
SUVH6 and effector complexes that assemble near a site of
dsRNA synthesis. Alternatively, sRNA-AGO complexes could
recruit intermediate factors that then promote SUVH activity at
specific targets. A third possibility is that sRNAs could guide a
pathway that protects heterochromatic sequences from H3K9
demethylation. For example, the IBM1 JumonjiC domain H3K9
demethylase acts to prevent H3K9me2 and non-CG methylation
from accumulating in transcribed genes [11,37,38]. IBM1 could
be excluded from also acting at heterochromatic sequences
through a mechanism that involves sRNA-AGO complexes.
Furthermore, sRNA-dependent mechanisms that promote addi-
tion of H3K9me2 or prevent removal of H3K9me2 could operate
Pathways where sRNA-AGO complexes guide H3K9me to
appropriate regions of the genome have been identified in
organisms ranging from fission yeast to the protozoan Tetrahymena
thermophila to the insect Drosophila melanogaster, even though these
organisms lack 5meC [39–41]. Our discovery that Arabidopsis
also uses sRNAs to maintain H3K9me could represent a plant-
specific variation on this fundamentally conserved strategy. In this
case, the sRNA/SUVH/CMT3 pathway and the sRNA/DRM2
pathway could have both evolved from a basal mechanism
involving sRNA-AGO guidance of H3K9 MTases. Consistent
with this possibility, SUVH variants that lack catalytic activity but
maintain methyl-DNA binding are required for DRM2-dependent
The plant sRNA/H3K9me maintenance mechanism is inter-
woven with the SUVH/CMT3 chromatin binding amplification
loop and partially redundant functions of the MET1 and DRM
pathways to create a reinforced silencing network. However, loss
of the sRNA/H3K9me maintenance mechanism cannot be
completely buffered by the other pathways, and results in both
immediate reductionsand longer-term
H3K9me2 and 5meC. The unique properties of the PAI genes
make them ideal reporters to further understand how sRNAs are
harnessed to control maintenance of H3K9me2 on appropriate
target sequences in plant genomes.
Materials and Methods
T-DNA insertional dcl alleles were obtained from the Arabi-
dopsis Biological Resource Center (ABRC) or from the laboratory
of James Carrington at Oregon State University. The dcl2-1, dcl3-
1, and dcl4-2 mutations are likely null alleles originally isolated in
the Col strain [21,23]. The dcl1-9 mutation is a partial function
allele originally isolated in Ws, but then crossed five times to the
Landsberg erecta (Ler) strain [24,25]. Each dcl mutant was crossed
to Ws. PCR-based genotype markers were used to identify dcl
mutant progeny homozygous for the three PAI loci from Ws
Small RNAs Prevent Loss of H3K9me
PLoS Genetics | www.plosgenetics.org8October 2011 | Volume 7 | Issue 10 | e1002350
(Table S1). Each dcl allele was crossed a second time with Ws to
increase the proportion of the genome contributed by the Ws
parent. The resulting dcl single mutant strains were then crossed
with each other to generate double, triple, and quadruple mutant
combinations. The dcl mutations were also crossed into the Ws
pai1 reporter strain . The Ws drm1 drm2 double T-DNA
insertional null strain was obtained from the laboratory of Steven
Jacobsen at UCLA . The Col cmt3-11T T-DNA insertional
null strain was obtained from the ABRC . The Ws pai1, Ws
Dpai1–pai4, Ws Dpai1–pai4(PAIIR), Ws cmt3i11a, Ws x met1-1, and
pai1 suvh4R302* suvh5-1 suvh6-1 strains were previously described
[4,14,16,17,28]. Ws cmt3illa and Ws drm1 drm2 mutants were
crossed to make the Ws drm1 drm2 cmt3 strain.
Analysis of 5meC patterns
Plant genomic DNA preparation and DNA gel blot assays for
5meC were performed as previously described . Bisulfite
sequencing of the top strands of PAI1 and PAI2 proximal promoter
regions was performed as previously described . PAI bisulfite
sequencing primers are listed in Table S1.
Total RNA was extracted with TRIzol reagent (Invitrogen)
using the manufacturer’s protocol. Low molecular weight (LMW)
RNA was enriched by precipitating high molecular weight RNA
out of solution with 0.5 M NaCl, 10% polyethylene glycol (MW
8000). The remaining LMW RNA was precipitated with 100%
ethanol and resuspended in water treated with diethyl pyrocarbo-
nate. LMW RNA was fractionated on a 17% acrylamide 7 M urea
gel and transferred to a Hybond-N membrane (GE Healthcare).
sRNA 59 ends were chemically crosslinked to the membrane as
previously described . Membranes were hybridized in
OligoHyb buffer (Ambion) overnight at 42uC with
labeled oligonucleotide probes. Probes were either an LNA
modified PAI1 exon 5 sense 35-mer (Exiqon) or an miR167
antisense 21-mer (Table S1). Probed membranes were washed
three times with a 26SSC, 0.1% SDS solution. sRNA sizes were
estimated from an ethidium bromide-stained low molecular weight
DNA ladder (USB), and by comparison to the PAI sRNA species
observed in the Dpai1–pai4(PAIIR) control strain .
32P 59 end-
Formaldehyde crosslinking and chromatin preparations were
performed as previously described  starting with two grams of
aerial tissue from three-week-old plants grown in soilless potting
medium (Fafard mix 2) under continuous illumination. Chromatin
was immunoprecipitated with anti-H3K9me2 monoclonal anti-
body  or carried through the protocol with no antibody added
as a control (mock precipitation). Immunoprecipitations were
performed as previously described . Each ChIP assay was
performed in at least three independent biological replicates.
Quantitative PCR amplification of immunoprecipitated DNA was
performed using the 7300 Real-Time PCR System (ABI), with
three replicate reactions for each sample. ChIP primer sequences
are listed in Table S1.
(A) PAI gene restriction maps. Arrows represent the regions of
shared sequence identity among the duplicated PAI genes. Black
arrows indicate duplications that share at least 98% identity and
the gray arrow indicates that the PAI3 duplication is more
divergent, with only 90% identity to the other genes. The number
Restriction maps of loci analyzed by DNA gel blot.
of each PAI gene is listed on the arrowhead. The Ws PAI genes
carry 5meC across the regions represented by the arrows.
Restriction patterns unique to specific strains are indicated with
Ws, Col, or Col/Ler. HincII sites are indicated by red vertical bars.
MspI/HpaII sites are indicated by black vertical bars. Only the
HincII and MspI/HpaII relevant to the 5meC DNA gel blot assays
are shown. PAI-internal sites that are methylated in wild type Ws
are indicated with asterisks. The regions covered by the probe used
in PAI DNA gel blots are indicated by blue lines. The sizes of
restriction products detected in DNA gel blot analysis are given in
kb. The transcription start site (TSS) position for each PAI gene is
indicated. (B) IR1074 restriction map. Arrows represent the
duplicated segments of the IR, with the degenerate nature of the
duplication indicated with black versus gray arrows. MspI sites are
indicated by black vertical bars. Internal sites that are methylated
in wild type Ws are indicated with asterisks. The region covered by
the probe used in the IR1074 DNA gel blot shown in Figure 5A is
indicated by the blue line over the sequences flanking the right
duplication. The sizes of restriction products detected in DNA gel
blot analysis are given in kb.
of new PAI2 5meC. (A) Diagram of crossing scheme to test for
initiation of PAI2 5meC and silencing. Col/Ler PAI genes are
indicated by red arrows. Ws PAI genes are indicated by black or
grey arrows, representing functional and non-functional genes
respectively. 5meC is indicated by boxes around the affected
genes. The question mark indicates that initiation of 5meC on
PAI2 depends on the genetic background. (B) DNA gel blot assay
for PAI 5meC. Genomic DNA from the indicated strains in the
indicated generations was cleaved with HincII and used in DNA
gel blot analysis with a PAI cDNA probe. P1–P4 indicates pai1–
PAI4, ColP1 indicates Col/Ler PAI1, and P3 indicates PAI3, with
bands diagnostic of 5meC on PAI-internal sites marked with
asterisks. P2 indicates unmethylated Ws, Col, or Ler PAI2, P2*
indicates methylated Ws PAI2, and Col/LerP2* indicates
methylated Col or Ler PAI2, which is a higher molecular weight
species than in Ws due to a flanking HincII polymorphism (see
Figure S1). Molecular weights of fragments are as shown in
Figure 1. (C) Blue fluorescence diagnostic of PAI2 silencing in the
pai1–PAI4 background. Representative 2.5-week-old plants of the
indicated strains in the indicated generations are shown under
visible (top) or UV (bottom) light. Ws pai1 x Col/Ler indicates
strains that are homozygous for Ws pai1–PAI4 and homozygous
for Col/Ler PAI2, as shown in (A), with the indicated dcl or drm
mutations present in both parents. drm indicates the drm1 drm2
The dcl3 and drm1 drm2 mutations impair acquisition
otides used in this study are shown.
Oligonucleotide sequences. Sequences for oligonucle-
new PAI2 5meC. Results, Materials and Methods, and References
supporting the experiments shown in Figure S2 are provided.
The dcl3 and drm1 drm2 mutations impair acquisition of
We thank Hiroshi Kimura, for the gift of H3K9me2 monoclonal
antibodies, and the laboratories of David Baulcombe and Blake Meyers,
for sharing data with us pre-publication. We also thank Jacqueline Gores,
for technical assistance, and Tricia Serio and Mark A. Johnson, for
comments on the manuscript.
Small RNAs Prevent Loss of H3K9me
PLoS Genetics | www.plosgenetics.org9 October 2011 | Volume 7 | Issue 10 | e1002350
Conceived and designed the experiments: RAE ZD JB. Performed the
experiments: RAE ZD JB. Analyzed the data: RAE ZD JB. Contributed
reagents/materials/analysis tools: RAE ZD JB. Wrote the paper: JB.
1. Jackson JP, Lindroth AM, Cao X, Jacobsen SE (2002) Control of CpNpG DNA
methylation by the KRYPTONITE histone H3 methyltransferase. Nature 416:
2. Malagnac F, Bartee L, Bender J (2002) An Arabidopsis SET domain protein
required for maintenance but not establishment of DNA methylation. EMBO J
3. Ebbs ML, Bartee L, Bender J (2005) H3 lysine 9 methylation is maintained on a
transcribed inverted repeat by combined action of SUVH6 and SUVH4
methyltransferases. Mol Cell Biol 25: 10507–10515.
4. Ebbs ML, Bender J (2006) Locus-specific control of DNA methylation by the
Arabidopsis SUVH5 histone methyltransferase. Plant Cell 18: 1166–1176.
5. Lindroth AM, Shultis D, Jasencakova Z, Fuchs J, Johnson L, et al. (2004) Dual
histone H3 methylation marks at lysines 9 and 27 required for interaction with
CHROMOMETHYLASE3. EMBO J 23: 4286–4296.
6. Johnson LM, Bostick M, Zhang X, Kraft E, Henderson I, et al. (2007) The SRA
methyl-cytosine-binding domain links DNA and histone methylation. Curr Biol
7. Johnson L, Cao X, Jacobsen S (2002) Interplay between two epigenetic marks.
DNA methylation and histone H3 lysine 9 methylation. Curr Biol 12:
8. Tariq M, Saze H, Probst AV, Lichota J, Habu Y, et al. (2003) Erasure of CpG
methylation in Arabidopsis alters patterns of histone H3 methylation in
heterochromatin. Proc Natl Acad Sci U S A 100: 8823–8827.
9. Mathieu O, Reinders J, Caikovski M, Smathajitt C, Paszkowski J (2007)
Transgenerational stability of the Arabidopsis epigenome is coordinated by CG
methylation. Cell 130: 851–862.
10. Pillot M, Baroux C, Vazquez MA, Autran D, Leblanc O, et al. (2010) Embryo
and endosperm inherit distinct chromatin and transcriptional states from the
female gametes in Arabidopsis. Plant Cell 22: 307–320.
11. Inagaki S, Miura-Kamio A, Nakamura Y, Lu F, Cui X, et al. (2010)
Autocatalytic differentiation of epigenetic modifications within the Arabidopsis
genome. EMBO J 29: 3496–3506.
12. Melquist S, Luff B, Bender J (1999) Arabidopsis PAI gene arrangements, cytosine
methylation and expression. Genetics 153: 401–413.
13. Luff B, Pawlowski L, Bender J (1999) An inverted repeat triggers cytosine
methylation of identical sequences in Arabidopsis. Mol Cell 3: 505–511.
14. Melquist S, Bender J (2003) Transcription from an upstream promoter controls
methylation signaling from an inverted repeat of endogenous genes in Arabidopsis.
Genes Dev 17: 2036–2047.
15. Lister R, O’Malley RC, Tonti-Filippini J, Gregory BD, Berry CC, et al. (2008)
Highly integrated single-base resolution maps of the epigenome in Arabidopsis.
Cell 133: 523–536.
16. Bartee L, Malagnac F, Bender J (2001) Arabidopsis cmt3 chromomethylase
mutations block non-CG methylation and silencing of an endogenous gene.
Genes Dev 15: 1753–1758.
17. Bender J, Fink GR (1995) Epigenetic control of an endogenous gene family is
revealed by a novel blue fluorescent mutant of Arabidopsis. Cell 83: 725–734.
18. Melquist S, Bender J (2004) An internal rearrangement in an Arabidopsis
inverted repeat locus impairs DNA methylation triggered by the locus. Genetics
19. Ramachandran V, Chen X (2008) Small RNA metabolism in Arabidopsis. Trends
Plant Sci 13: 368–374.
20. Chan SW, Zilberman D, Xie Z, Johansen LK, Carrington JC, et al. (2004) RNA
silencing genes control de novo DNA methylation. Science 303: 1336.
21. Xie Z, Johansen LK, Gustafson AM, Kasschau KD, Lellis AD, et al. (2004)
Genetic and functional diversification of small RNA pathways in plants. PLoS
Biol 2: E104.
22. Gasciolli V, Mallory AC, Bartel DP, Vaucheret H (2005) Partially redundant
functions of Arabidopsis DICER-like enzymes and a role for DCL4 in producing
trans-acting siRNAs. Curr Biol 15: 1494–1500.
23. Henderson IR, Zhang X, Lu C, Johnson L, Meyers BC, et al. (2006) Dissecting
Arabidopsis thaliana DICER function in small RNA processing, gene silencing and
DNA methylation patterning. Nat Genet 38: 721–725.
24. Jacobsen SE, Running MP, Meyerowitz EM (1999) Disruption of an RNA
helicase/RNAse III gene in Arabidopsis causes unregulated cell division in floral
meristems. Development 126: 5231–5243.
25. Schauer SE, Jacobsen SE, Meinke DW, Ray A (2002) DICER-LIKE1: blind
men and elephants in Arabidopsis development. Trends Plant Sci 7: 487–491.
26. Cao X, Jacobsen SE (2002) Role of the Arabidopsis DRM methyltransferases in de
novo DNA methylation and gene silencing. Curr Biol 12: 1138–1144.
27. Molnar A, Melnyk CW, Bassett A, Hardcastle TJ, Dunn R, et al. (2010) Small
silencing RNAs in plants are mobile and direct epigenetic modification in
recipient cells. Science 328: 872–875.
28. Bartee L, Bender J (2001) Two Arabidopsis methylation-deficiency mutations
confer only partial effects on a methylated endogenous gene family. Nucleic
Acids Res 29: 2127–2134.
29. Lu C, Kulkarni K, Souret FF, MuthuValliappan R, Tej SS, et al. (2006)
MicroRNAs and other small RNAs enriched in the Arabidopsis RNA-dependent
RNA polymerase-2 mutant. Genome Res 16: 1276–1288.
30. Lippman Z, May B, Yordan C, Singer T, Martienssen R (2003) Distinct
mechanisms determine transposon inheritance and methylation via small
interfering RNA and histone modification. PLoS Biol 1: E67.
31. Cao X, Jacobsen SE (2002) Locus-specific control of asymmetric and CpNpG
methylation by the DRM and CMT3 methyltransferase genes. Proc Natl Acad
Sci U S A 99 Suppl 4: 16491–16498.
32. Pikaard CS, Haag JR, Ream T, Wierzbicki AT (2008) Roles of RNA polymerase
IV in gene silencing. Trends Plant Sci 13: 390–397.
33. Wierzbicki AT, Haag JR, Pikaard CS (2008) Noncoding transcription by RNA
polymerase Pol IVb/Pol V mediates transcriptional silencing of overlapping and
adjacent genes. Cell 135: 635–648.
34. Wierzbicki AT, Ream TS, Haag JR, Pikaard CS (2009) RNA polymerase V
transcription guides ARGONAUTE4 to chromatin. Nat Genet 41: 630–634.
35. Zheng B, Wang Z, Li S, Yu B, Liu JY, et al. (2009) Intergenic transcription by
RNA polymerase II coordinates Pol IV and Pol V in siRNA-directed
transcriptional gene silencing in Arabidopsis. Genes Dev 23: 2850–2860.
36. Moazed D (2009) Small RNAs in transcriptional gene silencing and genome
defence. Nature 457: 413–420.
37. Miura A, Nakamura M, Inagaki S, Kobayashi A, Saze H, et al. (2009) An
Arabidopsis jmjC domain protein protects transcribed genes from DNA
methylation at CHG sites. EMBO J 28: 1078–1086.
38. Saze H, Shiraishi A, Miura A, Kakutani T (2008) Control of genic DNA
methylation by a jmjC domain-containing protein in Arabidopsis thaliana. Science
39. Volpe TA, Kidner C, Hall IM, Teng G, Grewal SI, et al. (2002) Regulation of
heterochromatic silencing and histone H3 lysine-9 methylation by RNAi.
Science 297: 1833–1837.
40. Liu Y, Mochizuki K, Gorovsky MA (2004) Histone H3 lysine 9 methylation is
required for DNA elimination in developing macronuclei in Tetrahymena. Proc
Natl Acad Sci U S A 101: 1679–1684.
41. Pal-Bhadra M, Leibovitch BA, Gandhi SG, Rao M, Bhadra U, et al. (2004)
Heterochromatic silencing and HP1 localization in Drosophila are dependent on
the RNAi machinery. Science 303: 669–672.
42. Johnson LM, Law JA, Khattar A, Henderson IR, Jacobsen SE (2008) SRA-
domain proteins required for DRM2-mediated de novo DNA methylation.
PLoS Genet 4: e1000280.
43. Cao X, Springer NM, Muszynski MG, Phillips RL, Kaeppler S, et al. (2000)
Conserved plant genes with similarity to mammalian de novo DNA
methyltransferases. Proc Natl Acad Sci U S A 97: 4979–4984.
44. Chan SW, Henderson IR, Zhang X, Shah G, Chien JS, et al. (2006) RNAi,
DRD1, and histone methylation actively target developmentally important non-
CG DNA methylation in Arabidopsis. PLoS Genet 2: e83.
45. Pall GS, Hamilton AJ (2008) Improved northern blot method for enhanced
detection of small RNA. Nat Protoc 3: 1077–1084.
46. Saleh A, Alvarez-Venegas R, Avramova Z (2008) An efficient chromatin
immunoprecipitation (ChIP) protocol for studying histone modifications in
Arabidopsis plants. Nat Protoc 3: 1018–1025.
47. Kimura H, Hayashi-Takanaka Y, Goto Y, Takizawa N, Nozaki N (2008) The
organization of histone H3 modifications as revealed by a panel of specific
monoclonal antibodies. Cell Struct Funct 33: 61–73.
Small RNAs Prevent Loss of H3K9me
PLoS Genetics | www.plosgenetics.org10October 2011 | Volume 7 | Issue 10 | e1002350