Cooperation of Multiple Chromatin Modiﬁcations Can Generate Unanticipated Stability of Epigenetic States in Arabidopsis

Epigenetic changes of gene expression can potentially be reversed by developmental programs, genetic manipulation, or pharmacological interference. However, a case of transcriptional gene silencing, originally observed in tetraploid Arabidopsis thaliana plants, created an epiallele resistant to many mutations or inhibitor treatments that activate many other suppressed genes. This raised the question about the molecular basis of this extreme stability. A combination of forward and reverse genetics and drug application provides evidence for an epigenetic double lock that is only alleviated upon the simultaneous removal of both DNA methylation and histone methylation. Therefore, the cooperation of multiple chromatin modiﬁcations can generate unanticipated stability of epigenetic states and contributes to heritable diversity of gene expression patterns.


INTRODUCTION
Genetically determined loss of gene expression by mutation, insertion of transposons, or chromosomal rearrangements is usually irreversible, since the chance of precisely reconstituting the original DNA sequence is low. On the other hand, epigenetic loss of gene activity is defined as not affecting the DNA sequence but rather as chemically modifying DNA and associated proteins, thus altering the packaging of chromatin and its accessibility for the transcription machinery. Affected sequences are kept transcriptionally inactive by well-characterized pathways that establish DNA methylation and/or histone modifications. For several of these modifications, antagonistic enzymes have been described (Chen and Tian, 2007;Pfluger and Wagner, 2007;Ooi and Bestor, 2008), and many epigenetically regulated sequences undergo a cycle of silencing and activation in the life cycle of the organism. Familiar examples in developmental programs are imprinted genes, dosage-compensated chromosomes, or master regulatory genes under the control of the Polycomb/Trithorax system. Even genetic templates that can produce potentially deleterious transcripts and are usually under tight epigenetic control can become activated under stress conditions (for review, see Madlung and Comai, 2004;Chinnusamy and Zhu, 2009) or in the germ line in order to reinforce silencing via small RNA during transmission of genetic material to the next generation (Brennecke et al., 2008;Slotkin et al., 2009). However, some cases of genes with very durable epigenetic marks are also known (Chong and Whitelaw, 2004), and the stable transmission of their epigenetic state to subsequent generations has led to their denotation as epialleles (Finnegan, 2002). Examples in plants are a methylated transcription factor gene changing flower morphology in Linaria (Cubas et al., 1999) and the pigmentationcontrolling transcription factor genes in maize (Zea mays) that are downregulated by paramutation (for review, see Chandler et al., 2000). These famous cases were identified because of the striking phenotypes. It is likely that many more epialleles exist with less drastic morphological consequences but which nevertheless make a significant contribution to natural evolution and plant breeding (Kalisz and Purugganan, 2004).
Epialleles with remarkable stability have been observed in various tetraploid lines of Arabidopsis thaliana derived from a common diploid progenitor . The transgenic resistance marker gene, hygromycin phosphotransferase (HPT), under the control of the strong, constitutively active promoter of the Cauliflower mosaic virus (P35S) was present in these genetically identical lines either in fully active or completely silenced state. Both states were maintained during backcrosses to diploid lines homozygous for the HPT, giving rise to diploid lines C2R (resistant to hygromycin, active HPT) and C2S1 (diploid, sensitive to hygromycin, silent HPT). In crossing experiments with the tetraploid lines, the epialleles exerted a paramutation-like interaction in which the silent epiallele led to inactivation of the previously active counterpart (Mittelsten . The epialleles differ in the degree of DNA methylation and histone modification patterns (Hetzl et al., 2007;Foerster, 2009), as do many other active and inactive sequences. The epialleles show an extremely tight silencing (as described in the following): they were originally found in the tetraploid lines, and the epiallelic interaction occurred only in tetraploid intercrosses. Therefore, we refer to this phenomenon as polyploidyassociated transcriptional gene silencing (paTGS) even in the diploid lines. Most higher plants are polyploid (Masterson, 1994), and polyploidy is assumed to be a very important driving force in plant evolution and breeding (Stebbins, 1966). Furthermore, epigenetic changes are frequent in freshly formed polyploids (for review, see Osborn et al., 2003;Adams and Wendel, 2005). Paramutation-like epiallelic interaction can lead to significant shifts in the distribution of traits within populations of polyploid plants and drive their evolution more rapidly than anticipated by classical Mendelian genetics. Therefore, it is important to understand the characteristics of the epialleles that underwent paTGS. The described silent HPT epiallele offered an excellent model for this analysis, since its stability also in the diploid derivative line and the encoded protein allowed a selectionbased genetic screen for trans-and cis-acting factors involved in the maintenance of the silencing. Here, we demonstrate that the silent epiallele derived from the tetraploid line is under a double safeguard mechanism, which requires the concomitant loss of methylation of both DNA and histones for restoration of transcription. This is in contrast with many other transcriptionally silent sequences in the Arabidopsis genome that can be activated by removing only one of several inactive chromatin marks by mutation or specific inhibitors. Thus, epialleles in polyploid plants cannot easily revert and represent particularly stable states that are under tight control. For this reason, they might be highly relevant for long-term adaptation of gene expression patterns, breeding, and natural evolution.

RESULTS paTGS Is Resistant to Treatments with DNA Methylation and Histone Deacetylase Inhibitors
Transcriptional inactivation in plants and mammals is frequently associated with methylation of cytosine residues in the DNA, an exchange of specific methylation of histone tails from active to inactive marks, and general deacetylation of histone tails (Chen and Tian, 2007;Vaillant and Paszkowski, 2007). Inhibitors specific for DNA methyltransferases and histone deacetylases exist, and they have been widely used as potentially activating agents for epigenetically silenced endogenes and transgenes (Chang and Pikaard, 2005). The DNA methylation inhibitor zebularine (ZEB) (Zhou et al., 2002) and the histone deacetylase inhibitor trichostatin A (TSA) (Yoshida et al., 1995) were therefore applied to test whether they would reactivate the silent HPT transgene. Seeds from the diploid line C2S1 with the inactive HPT and seeds from the HPT-expressing, hygromycin-resistant line C2R were germinated and plantlets grown for 3 weeks on plates containing 10 mg/mL of hygromycin in combination with 40 mM ZEB and/or 1.6 mM TSA, concentrations that were previously described to be effective in reactivating silenced targets and reducing methylation in all possible sequence contexts (Baubec et al., 2009) or were even higher than effective concentrations (Chang and Pikaard, 2005). ZEB causes growth retardation but allows the HPT-expressing line C2R to grow under selection upon all treatments. By contrast, no growth was observed in line C2S1 ( Figure 1A), even upon sequential application of the drugs prior to selection. The applied drug treatments could not, therefore, reactivate the HPT gene and restore the resistant phenotype.
Stringent hygromycin selection requires a certain amount of HPT RNA and protein to be produced. To determine whether the inhibitors would release subthreshold levels of gene expression, we performed RNA gel blot analysis using HPT-specific probes on total RNA extracted from C2S1 seedlings treated with 0, 20, 40, and 80 mM ZEB. These showed a minimal increase in HPT transcript but substantially less hybridization signal than in C2R ( Figure 1B). In addition, known epigenetic mutations, such as cmt3, drm1,2, and kyp that could not restore hygromycin resistance after introgression of the silent C2S1 epiallele (Milos, 2006), did also not further enhance the effect of zebularine treatments (see Supplemental Figure 1 online). Surprisingly, RNA gel blot analysis with a probe for a noncoding RNA transcribed from another copy of the P35S promoter, downstream of and in close proximity to the HPT gene (see Supplemental Figure 2 online), revealed strong reactivation of this second transcript after ZEB treatment of C2S1 ( Figure 1B). The pharmacological demethylation was effective, as demonstrated by methylation-sensitive restriction digest and subsequent DNA gel blotting ( Figure 1C), but was not sufficient to reactivate the HPT-driving promoter.
paTGS Can Be Released by Novel DDM1 and HOG1

Mutant Alleles
Since the silent HPT transgene allowed for a reactivation assay based on hygromycin selection, we performed a forward genetic screen to identify factors involved in this robust epigenetic regulation of the HPT promoter. Diploid C2S1 plants carrying the silent HPT transgenic locus were mutagenized by random T-DNA insertion, and M2 progeny of 20,000 independent transformants was screened for hygromycin resistance. We identified three novel alleles of DECREASE IN DNA METHYLATION1 (DDM1), a member of the ATP-dependent SWI2/SNF2 chromatin remodeling gene family (Vongs et al., 1993;Mittelsten Scheid et al., 1998;Jeddeloh et al., 1999) and one novel allele of the HOMOLOGOUS GENE SILENCING1 (HOG1) gene, coding for an S-adenosyl-L-homocysteine (SAH) hydrolase (SAHH) (Furner et al., 1998;Rocha et al., 2005). Mutations in DDM1 (At5g66750) have been previously shown to interfere with maintenance of transcriptional gene silencing at numerous endogenous and transgenic inserts by decreasing DNA and H3K9 methylation (Mittelsten Scheid et al., 1998;Jeddeloh et al., 1999;Soppe et al., 2002;Mathieu et al., 2003). HOG1 (or SAHH1, At4g13940) is required to convert SAH into homocysteine. This degradation is essential for recycling of the methyl-group donor S-adenosyl-L-methionine (SAM) and prevents inhibition of trans-methylation reactions through increased levels of SAH (Weretilnyk et al., 2001). HOG1 is involved in maintaining transcriptional gene silencing at numerous targets (Furner et al., 1998;Rocha et al., 2005;Mull et al., 2006;Jordan et al., 2007), while another SAHH-related gene (SAHH2, At3g23810) has no role in silencing or DNA methylation (Rocha et al., 2005). The DDM1 alleles were named ddm1-11 to ddm1-13, in continuation of the already available mutant alleles (Jeddeloh et al., 1999;Jordan et al., 2007): ddm1-11 has a 38-bp deletion in exon V, ddm1-12 has a 30-bp deletion in exon XIV, and ddm1-13 has a T-DNA insertion in exon VII ( Figure 2A). In contrast with the widely used alleles ddm1-2, with a point mutation generating a G-to-A transition in the splice donor site of intron XI (Jeddeloh et al., 1999), and ddm1-5, with an 82-bp insert in exon II (Mittelsten Scheid et al., 1998;Jeddeloh et al., 1999), the new mutations are all in conserved signature motifs that are characteristic of SWI2/SNF2 family proteins (Bork and Koonin, 1993) and affect the domains that are important for ATP-dependent chromatin remodeling, namely, SNF2_N and DEAD/DEAH (Figure 2A). This may explain why plants with the new alleles survived the stringent hygromycin selection in the M2 generation during the screen, while plants with the ddm1-5 allele showed partial reactivation and survived only in F4 after introgression Milos, 2006). A direct comparison of the ddm1-5 F4 seedlings with the corresponding M4 generation seedlings obtained from the novel alleles further illustrates the differences in resistance ( Figure 2B), confirmed by HPT expression analysis (see below). ddm1-12 was used as a representative ddm1 allele in the following experiments. The new HOG1 allele, named hog1-7 in continuation of previously identified alleles (Rocha et al., 2005), has a rearranged T-DNA insertion in the 39 UTR ( Figure 2C). Although this mutation is not likely to cause a complete loss of function, it affects HOG1 mRNA levels and stability, as revealed by quantitative RT-PCR ( Figure 2D).
We analyzed the degree of HPT reactivation in 3-week-old M4 mutant seedlings. Quantification of HPT transcripts with real-time PCR using cDNA obtained from reverse-transcribed total RNA from the ddm1-12 and hog1-7 mutants indicated a similar abundance as in the active line C2R (1-fold 6 0.35 and 0.96-fold 6 0.23 in hog1-7 and ddm1-12 mutants, respectively; Figure 3A). This is in agreement with a similar loss of DNA methylation at the P35S promoter, as shown by DNA gel blot analysis ( Figure  3B). To quantify the degree of DNA demethylation specifically at the promoter upstream of HPT, we applied bisulfite sequencing  (A) DDM1 gene region with indicated UTRs (white boxes), exons (filled boxes), and introns (lines). Functional domains are indicated by colored boxes, while mutations are indicated by insertions or deletions (D). Below: reading frame analysis in the ddm1 alleles. Coding sequence is indicated by the gray bar, and conserved SWI2/SNF2 signatures (Bork and Koonin, 1993) are shown below. White glyphs indicate potential translation initiation sites in the 59 region (aa(A/G)(A/C)aAUGCcg; Rangan et al., 2008). Coding reading frames (in different colors) and encoded protein size are predicted in wild-type and mutant alleles. Light-gray bars indicate nonplant DNA insertions. (B) Allele comparison by hygromycin selection in analogous generations: F4 from crosses between C2S1 3 ddm1-5 and M4 in the novel alleles. C2S1 and C2R are used as controls. to DNA of the hog1-7 and ddm1-12 mutants ( Figure 3C). Total DNA methylation was reduced from 29% in C2S1 to 0% in hog1-7 and to 17% in ddm1-12, while CG-specific methylation was reduced from 87 to 1% and 22% in hog1-7 and ddm1-12, respectively. We observed a similar decrease in methylation at CHG sites, where the hog1-7 mutation resulted in 0% residual methylated CHG sites, while the ddm1-12 mutation maintained 32% of the methylated CHG sites compared with 57% in C2S1.
CHH-specific methylation, with 13% of all available sites in C2S1, was significantly decreased in hog1-7 with 0%, while it remained unaltered in the ddm1-12 mutant, indicating that DDM1 is not required to maintain methylation at these largely nonsymmetrical sites ( Figure 3C).
To complement the analysis of chromatin changes in the mutants, we further analyzed ddm1-12-specific and hog1-7-specific changes in histone modifications at the HPT transgene promoter (A) Quantification of HPT mRNA levels in wild-type, C2S1, C2R hog1-7,C2S1, and ddm1-12,C2S1 seedlings normalized to EIF4A2. Error bars represent SD from triplicate analyses. (B) p35S DNA methylation analysis in C2S1, C2R, and mutant plants by DNA gel blotting of DNA digested by methylation-sensitive restriction enzymes. (C) Promoter DNA methylation analysis by bisulfite sequencing representing total (black) and sequence context-specific ( m CG, red; m CHG, blue; m CHH, green) methylation. (D) Analysis of histone modifications and histone H3 abundance normalized to H3 or input at the HPT promoter by ChIP in C2S1 and mutant lines. Gray columns (right of the colored columns and very small) represent samples precipitated without antibodies. by chromatin immunoprecipitation (ChIP) using specific antibodies or antisera against the heterochromatic mark histone H3 Lys-9 dimethylation (H3K9me2) and the euchromatic marks histone H3 Lys-4 trimethylation (H3K4me3) and histone H4 panacetylation (H4panAc) (for review, see Fuchs et al., 2006). General nucleosome occupancy in the examined regions was analyzed by ChIP with antibodies recognizing histone H3 independent of modifications. Enrichment of the DNA fragments in the modification-specific precipitates was measured by quantitative PCR in triplicate and was related to their loading with histone H3. The prevalence of heterochromatic H3K9me2 in C2S1 was drastically reduced in the hog1-7 and ddm1-12 mutants ( Figure 3D, red columns). H3K4me3 increased in both mutants compared with C2S1, although only slightly in hog1-7 and much more pronounced in ddm1-12 ( Figure  3D, green columns). An increase of H4 acetylation was observed in both mutants, again with a stronger increase in ddm1-12 ( Figure  3D, violet columns). Remarkably, nucleosome occupancy measured as histone H3 abundance relative to input DNA was comparable between C2S1 and hog1-7 but almost totally lost in ddm1-12. This should be considered when interpreting the relative enrichment or depletion of histone marks in the mutants ( Figure 3D, blue columns).

Mutations in DDM1 or HOG1 Affect Methylation of DNA and Histones Globally
Transcriptional silencing associated with DNA methylation and heterochromatic marks can be released by different means, including specific inhibitors or loss of function of epigenetic regulators. As shown above, the silent HPT transgene that was found in the polyploid lines did not respond to inhibitors. It also remained suppressed in the background of many mutations representing the known epigenetic regulatory pathways (Milos, 2006;Baubec et al., 2009;Foerster, 2009). This raised the question of why and how the new mutations in DDM1 and HOG1 proved to be exceptions and whether this could hint at an underlying mechanism. Both mutants have been reported to interfere with transcriptional gene silencing at many other targets in the Arabidopsis genome (Lippman et al., 2004;Jordan et al., 2007), but many of these were also expressed in those other mutants that did not reactivate the HPT gene. However, mutations in DDM1 and HOG1 have in common that they reduce DNA methylation and heterochromatic histone modifications at the HPT transgene. This effect of DDM1 loss has also been described for other targets (Gendrel et al., 2002;Johnson et al., 2002;Soppe et al., 2002;Probst et al., 2003). Mutations in HOG1 cause DNA hypomethylation at transgenic and endogenous repeats (Furner et al., 1998;Rocha et al., 2005;Mull et al., 2006), and the function of the HOG1 gene regulating the level of the methyl group donor SAM suggested that its loss would also affect histone methylation (Rocha et al., 2005). To challenge the hypothesis that removal of both marks is a common feature of ddm1 and hog1, we characterized DNA methylation and histone methylation in the novel mutant alleles in general and also at other sequences to allow for a direct comparison of the extent and specificity of the effects.
In agreement with results published for other alleles (Vongs et al., 1993;Furner et al., 1998;Jeddeloh et al., 1999;Rocha et al., 2005), global DNA methylation in hog1-7 and ddm1-12 was reduced to 2.7% (60.47) and 1.7% (60.13), respectively, in comparison to 5-methyldeoxycytosine (5-mdC) levels of 5.9% (60.5) in the parental line C2S1, which is similar to wild-type levels (Rozhon et al., 2008) (Figure 4A). A significant proportion of the DNA methylation in wild-type Arabidopsis is found at repetitive sequences (Martinez-Zapater et al., 1986) and disappears in ddm1 or hog1 mutants (Vongs et al., 1993;Furner et al., 1998). This is also true for the new alleles: DNA gel blot analysis of DNA methylation at centromeric 180-bp repeats ( Figure 4B) showed drastic hypomethylation in both mutants. However, the demethylation was more pronounced in the ddm1-12 mutant, especially for the CG sites ( Figure 4B). A certain difference was also evident after cytological analysis of the usually compact heterochromatic chromocentres by immunofluorescence, revealing dispersed 5-mdC localization in ddm1-12, where just 14% (n = 104) of nuclei retained chromocentric 5-mdC signals ( Figure 4C). This is in agreement with other reports (Soppe et al., 2002). Nuclei of hog1-7, however, maintained most 5-mdC (89%, n = 80) at the chromocenters (CCs), close to wild-type nuclei (91%, n = 109), in accordance with the DNA gel blot methylation analysis of the centromeric repeats. This suggests that loss of DNA methylation in the hog1-7 mutant occurs primarily at other parts of the genome. H3K9me2, as revealed by immunostaining, also colocalizes with CCs in wild-type nuclei (71%, n = 129) but is reduced in both mutants to 8 and 10% of nuclei having wild-type morphology (n = 114 and 107, respectively; Figure 4D), as also previously reported for ddm1 .
The loss of chromocentric H3K9me2 signals in hog1-7 nuclei, independent of the remaining DNA methylation, suggests a direct effect of SAM depletion on histone methylation. The cytological evidence was further substantiated by loss of silencing accompanied by reduced DNA and histone methylation at the retrotransposon without long terminal repeats LINE1-4 (At2g01840) (Lippman et al., 2003) in the ddm1-12 and hog1-7 mutations (see Supplemental Figures 3A to 3C online), as well as by decreased levels of H3K9me2 and H3K4me3 in hog1-7 analyzed by immunoblot (see Supplemental Figure 3D online). This provides further evidence of globally reduced histone methylation in hog1 mutants, independent of the Lys residue analyzed.

Inhibition of SAHH Interferes with Maintenance of paTGS
The similar but not identical consequences of mutations in DDM1 and HOG1 on general DNA and histone methylation let us postulate that their comparable and exclusive role among TGS mutants in the maintenance of paTGS would occur through directly and simultaneously affecting DNA and histone methylation at the HPT promoter. A genetic approach to simultaneously reduce histone methylation and DNA methylation in all sequence contexts would require combination of at least six mutations and renders plants with severe developmental aberrations (Chan et al., 2006;Johnson et al., 2008). Therefore, we tried to mimic the hog1 mutation by applying the specific SAHH inhibitor dihydroxypropyladenine (DHPA). The adenosine homolog DHPA was shown to induce hypomethylation and release of posttranscriptionally silenced transgenes in tobacco (Nicotiana tabacum; Kovarik et al., 1994Kovarik et al., , 2000a. We first established the applicable dose range in Arabidopsis and analyzed the effectiveness of DHPA by germinating and growing seeds of a line with a transcriptionally silent, highly repetitive b-glucuronidase (GUS) transgene insertion on chromosome III (L5) (Morel et al., 2000) that is reactivated in the background of numerous epigenetic mutations (Elmayan et al., 2005), including hog1-7 and ddm1-12, or by treatment with DNA methylation inhibitors (Baubec et al., 2009). DHPA treatments had only mild growth effects at the applied concentrations of 50 to 200 mM but successfully induced transcriptional reactivation of the GUS transgene (see Supplemental Figures 4A and 4B online). We subsequently compared DHPA inhibitors with drugs that change either DNA or histone modification. There is no inhibitor that specifically reduces histone methylation while leaving DNA methylation undisturbed. Thus, we applied the histone deacetylation inhibitor TSA, which has repeatedly been shown to convert silent into transcriptionally active genes (Chen and Pikaard, 1997;Xu et al., 2005). ZEB interferes specifically with DNA methylation (Zhou et al., 2002). We performed a side-by-side comparison of wild-type seedlings grown for 3 weeks on media containing either TSA, ZEB, or DHPA in the previously established dose ranges (Chang and Pikaard, 2005;Baubec et al., 2009; this article). We first analyzed transcriptional activation of endogenous repeats by quantitative real-time PCR. As observed in the mutant background (see Supplemental Figures 3A and 3B online), the retrotransposon without long terminal repeats LINE1-4 (At2g01840) showed significant and dose-dependent transcript abundance ( Figure  5A) and DNA hypomethylation ( Figure 5B) after ZEB or DHPA, but not TSA treatments. Corresponding to the degree of transcriptional activation, we observed a significant, though not complete, reduction of H3K9me2 ( Figure 5C). The active mark H3K4me3 increased but did not reach the levels seen in ddm1 mutants (see Supplemental Figure 3C online). This is plausible since both histone modifications are likely to require SAM, which is a limiting factor in hog1 and upon DHPA but not in ddm1. Data describing expression, DNA methylation, and histone modification for two other genomic sequences and cytological analysis of treated nuclei support the findings (see Supplemental Figures 5 and 6 online).
Although the retroelement LINE1-4, other repetitive sequences, and the second promoter of the transgene were transcriptionally activated by ZEB alone, silencing at the HPT promoter itself was not released by this drug ( Figure 1B). Therefore, we evaluated the effects of DHPA treatments on expression, DNA methylation, and histone modification of the silenced HPT gene, asking whether the drugs would mimic the effects of the hog1 mutation and release paTGS. The answer was affirmative, and high concentrations (200 mM) of SAHH inhibitor resulted in HPT expression up to 60% of the level in the hygromycinresistant line C2R ( Figure 6A). RNA gel blots with specific probes (C) Chromocentric 5-mdC localization measured by immunofluorescence is lost only in ddm1-12 but not in hog1-7. DAPI, 4',6-diamidino-2phenylindole.
(D) H3K9me2 compaction measured by immunofluorescence is disrupted in both mutants. The pie charts represent the percentage of nuclei with corresponding morphology. Gray, compact signals; white, dispersed signals. revealed that both promoters were activated concordantly, with the primary promoter producing nearly as much HPT transcript as in the active state of the control line C2R ( Figure 6B). DNA gel blot analysis of DNA methylation indicated dose-dependent hypomethylation at both promoters upon DHPA treatment, with CHG methylation more affected than CG ( Figure 6C). This is in accordance with gradual demethylation at the HpaII/MspI recognition sequence m C m CGG sites after DHPA treatments (Kovarik et al., 2000b). Quantification of histone modifications at the HPT promoter after DHPA treatments revealed loss of H3K9me2 and a slight gain of H3K4me3 already after 50 mM DHPA treatments ( Figure 6D), as in hog1-7. In summary, the chemical interference produced by DHPA application has a similar effect as the genetically determined decrease of functional SAHH by the hog1-7 mutation. Both cause a reduction of methylation of DNA and the associated histones of several genomic sequences, including the HPT transgene that underwent polyploidy-associated gene silencing. The lack of HPT reactivation upon depletion of only one type of methylation, in contrast with its restored transcription upon interference with both modifications simultaneously, suggests that this epiallele, and probably similar ones, are under a double-safeguard control that renders gene suppression extremely stable against epigenetic perturbation (Figure 7).

DISCUSSION
An undisputed definition of epigenetic inheritance is still lacking, but most descriptions refer to its reversible nature to distinguish it from genetic alterations inscribed in the DNA sequence. As is often the case in biology, this sharp distinction does not hold upon closer inspection. While many epigenetically regulated genes undergo programmed, regular, or random cycles of activation and suppression in the course of development, others have proven to be extremely stably silenced. Among them are many transposable elements, for which redundant control by different DNA methyltransferases (Kato et al., 2003) or a special reinforcement by small RNA silencing in the germ line (Brennecke et al., 2008;Slotkin et al., 2009) have been described. However, even transposons exhibit a surprising diversity in response to epigenetic interference in Arabidopsis where the role of welldefined epigenetic pathways can be studied in numerous mutants. Loss-of-function of DNA methyltransferases, argonaute proteins, histone methyltransferases, or histone deacetylases causes transcriptional activation of overlapping but not identical subsets of elements (Lippman et al., 2003). Most of these elements can also be activated by drugs that reduce either DNA methylation or histone modifications (Chang and Pikaard, 2005). Here, we have described a case of epigenetic transcriptional silencing that is surprisingly resistant to genetic and chemical interference, since removing one chromatin modification alone does not restore transcriptional activity from the potentially strong viral P35S promoter. Based on results from forward and reverse mutational screens (Milos, 2006;Baubec, 2008;Foerster, 2009), complemented by inhibitor experiments, we have provided evidence that two epigenetic features, namely, symmetric DNA methylation and histone methylation, cooperate to generate a double safeguard system that controls transcriptional suppression. Hence, both modifications have to be unlocked to convert the silent epiallele into an active one. This could be achieved by a loss of functional DDM1, a member of the SWI2/SNF2 chromatin remodeling ATP-dependent helicase family. Mutations in DDM1 known to decrease DNA methylation (Vongs et al., 1993;Kakutani et al., 1996Kakutani et al., , 1999 also reduce the levels of histone H3 dimethylation at Lys-9 (Gendrel et al., 2002;Habu et al., 2006). A partial interdependence of DNA methylation and H3K9me2 in Arabidopsis was further described in mutants of other genes whose products were supposed to act primarily on either DNA or H3K9 methylation (Johnson et al., 2002;Soppe et al., 2002;Tariq et al., 2003). These studies reveal a complex and possibly mutual interplay of DNA and histone methylation at different targets that can also depend on transcriptional activity. However, this interdependence does not apply to the silencing described in this study, since DNA or histone methyltransferase mutations alone did not reactivate the silent epiallele in our study. Even the concomitant reduction of both modifications was effective only above a certain threshold: mutant allele ddm1-5, isolated based on its strong reactivation of a transcriptionally silent HPT repeat (Mittelsten Scheid et al., 1998;Jeddeloh et al., 1999) but probably not a complete loss-offunction mutation (this study), did not evoke significant hygromycin resistance before the third inbred homozygous generation. Significant activity of the primary P35S promoter in the first homozygous mutant generation was only obtained with the three new DDM1 mutations that disrupt the conserved regions of the protein and are likely more deleterious. While numerous previously mentioned studies describe the large-scale consequences of ddm1 mutations for gene expression, transposon activation, and diverse chromatin modifications, the mechanistic connection between these effects and the remodeling activity of the protein extrapolated from in vitro experiments (Brzeski and Jerzmanowski, 2003) still remains to be uncovered. In this context, it is interesting that we observed decreased nucleosome abundance in DDM1-deficient plants. This could link the nucleosome remodeling function of DDM1 to the maintenance of DNA and histone methylation by facilitating a permissive environment for DNA and histone methyltransferases. Since ddm1 is frequently investigated in the context of histone modifications (Gendrel et al., 2002;Lippman et al., 2004;Habu et al., 2006), lower nucleosome occupancy should be considered in quantitative comparisons.
By contrast, hog1 mutations have so far only been analyzed for their effects on specific targets (Rocha et al., 2005;Mull et al., 2006) and general gene expression (Jordan et al., 2007). Nevertheless, the precise functional annotation of the gene product and the biochemical evidence for its role in regulating SAH levels  (Rocha et al., 2005) make it easier to speculate about its mode of action. The methyl group donor SAM is a central hub for numerous methylation reactions modifying DNA, proteins, and metabolites (Roberts and Selker, 1995;Loenen, 2006;Roje, 2006). Therefore, substrate competition by even slightly increased SAH levels is expected to change many reactions simultaneously. The focus of HOG1 analysis has so far been on DNA as a methylation acceptor (Furner et al., 1998;Rocha et al., 2005), especially since changes in histone methylation levels were not detected in a weak hog1 allele termed sah1L459F (Mull et al., 2006). Nevertheless, the new hog1 allele brings about a substantial loss of H3K9me2 from CCs despite only slight decrease in DNA methylation, as well as a globally reduced methylation at several histones (though to different degrees). Furthermore, the mutant effects of transcriptional activation of the HPT transgene and endogenous transposable elements can be mimicked with a specific SAHH inhibitor. Together, these findings indicate that HOG1 is indeed a central factor in chromatin modification. This is further emphasized by the relatively small overlap of gene expression changes between hog1-1 and treatment with the inhibitor 5-azacytidine (Jordan et al., 2007) that reduces DNA methylation and probably also 5-mdC-dependent histone methylation. Changing expression of many more genes indicates that hog1 acts through interference with additional components. A central role of HOG1 for the plant as a whole is also evident from the severe phenotypic consequences of even subtle mutations and the embryonic lethality observed in loss-of-function mutants (Rocha et al., 2005). Due to the central role of SAM, there are probably many more, non-chromatin-related factors involved. It should be emphasized that the SAM:SAH ratio may also be modified by metabolic regulation or by sulfur availability (Nikiforova et al., 2006). Thus, the dependence of several chromatin components on the levels of SAM and SAH offers a path by which environmental or nutritional cues can inscribe a signature in the epigenetic outfit of the genome.
The forward genetic screen for reactivation of the HPT allele resulted in only 21 primary mutant candidates, surprisingly few for a mutant population derived from 20,000 independent T-DNA transformation events compared with other screens following insertional mutagenesis (Budziszewski et al., 2001). In addition, several candidates turned out to carry mutations within the marker gene itself (A. Foerster, unpublished data). This, together with finding three alleles of the DDM1 gene, indicates saturation of the screen for trans-acting mutations. The mechanism of epigenetic control depicted in the double lock model makes these results nevertheless plausible: the need to eliminate two different chromatin modifications simultaneously requires either rare double mutations in two independent pathways or single mutations affecting the two modifications equally, making the screen a very stringent quest for strong modifications. Although very different in their assumed mode of action, DDM1 and HOG1 fulfill the latter conditions. Therefore, the double lock model is not only supported by the molecular data, but also by the general outcome of the forward screen.
It could be asked whether the data presented here, based mainly but not exclusively on the HPT transgene, have relevance beyond this particular situation. The advantage of this experimental system is that it represents a gene whose activity is absolutely nonessential for the plants unless under selection and thereby does not bias the propagation or segregation of either the active or inactive states. It is inserted in an intergenic region  and is therefore unlikely to cause an insertional mutation. The random rearrangement producing a duplication of the P35S promoter during the initial transformation event even allowed these two regulatory elements to be compared, with the surprising result that the identical sequences, in the same genomic location and with a distance of only 2 kb between them, respond quite differently to mutations and inhibitor effects. As pointed out before (Rocha et al., 2005), the silencing system in plants was not invented to inactivate manmade transgenes. Along this line, we demonstrated a clear epigenetic effect of the SAHH inhibitors and the hog1 and ddm1 mutant alleles on individual endogenous targets. A significant overlap of genes differentially regulated in both mutants, mainly but not exclusively transposons (T. Baubec and O. Mittelsten Scheid, unpublished data) further indicates more sequences under double control and a significant relevance of tight silencing beyond the HPT transgene.
More important is thinking about the role of polyploidy in generating a stable epiallele. While a diploid progenitor line containing the very same transgene always maintained high expression, partial or complete silencing was found in several independent autotetraploid derivatives . However, these were generated by protoplast culture and regeneration, leaving other parameters, such as hormone effects, tissue culture conditions, or even propagation of preexisting epigenetic states in individual cells, as possible sources of silencing, rather than polyploidization. Nevertheless, an association with polyploidy is very likely based on the trans-acting silencing between inactive and active epialleles, which is limited to tetraploid hybrids , and with a specific set of genes that are differentially expressed in the tetraploid lines. Although gene expression changes in autotetraploids are less frequent compared with freshly formed allopolyploids (Wang et al., 2004), polyploidization is recognized as being a significant source of genetic as well as epigenetic changes in many different plant species (for review, see Osborn et al., 2003;Adams and Wendel, 2005). paTGS can apparently generate very tightly controlled epialleles with an extremely low frequency of reversion and with the potential to be propagated and even spread among plant populations. It should be considered to be an important source of epigenetic diversity with an evolutionary impact.

Plant Growth and Chemical Treatments
Stratified seeds were surface-sterilized with 5% sodium hypochlorite and 0.05% Tween 80 for 6 min and washed and air-dried overnight. Sterilized seeds were germinated and grown in Petri dishes containing agarsolidified germination medium in growth chambers under 16-h-light/8-hdark cycles at 218C. For treatments with hygromycin (Calbiochem), TSA (Sigma-Aldrich), ZEB (Sigma-Aldrich), and DHPA (donated by Ales Kovarik), seeds were sown and grown directly on drug-containing plates under the conditions described above. Hygromycin (10 mg/mL), zebularine (20,40,and 80 mM),and DHPA (50,100,and 200 mM) in aqueous solution or TSA (1.6 and 3.2 mM) dissolved in DMSO were added to the germination medium before solidifying.

Mutant Screen and Mapping
Diploid C2S1 plants (in the background of accession Zü rich) were mutagenized by random T-DNA insertion after Agrobacterium tumefaciens transformation with p1'barbi (Mengiste et al., 1997). M2 seeds from 20,000 mutant M1 plants were harvested in pools of 15 M1 plants and selected on hygromycin-containing medium. HPT-expressing and nonexpressing lines, C2R and C2S1, were used as positive and negative controls, respectively. Hygromycin-resistant plants were further propagated, and hygromycin resistance was followed in subsequent generations after selfing and outcrossing to the wild type. Sequences flanking the T-DNA insertion that were genetically linked with the mutations (ddm1-13 and hog1-7) were identified by thermal asymmetric interlaced PCR as described (Liu et al., 1995). Other mutations were identified by sequencing of candidate genes (as in the case of ddm1-11 and ddm1-12).

Nucleic Acid Isolation and Gel Blot Analysis
Pools of 50 to 100 seedlings (age as indicated for the individual experiments) were shock-frozen in liquid nitrogen, homogenized, and subsequently used for DNA or RNA extraction using Phytopure (Amersham) or RNAeasy (Qiagen) kits, respectively.
For DNA methylation analysis by DNA gel blot, 10 mg of genomic DNA were digested overnight with 1 to 2 units of HpaII (sensitive to m C m CGG) or MspI (sensitive to m CCGG) restriction enzymes. Subsequently, samples were electrophoretically separated on TAE agarose gels, depurinated for 10 min in 250 mM HCl, denatured for 30 min in 0.5 M NaOH and 1.5 M NaCl, and neutralized twice in 0.5 M Tris, 1.5 M NaCl, and 1 mM EDTA at pH 7.2 for 15 min. For RNA gel blot analysis, 10 mg of total RNA were denatured with 15% glyoxal and DMSO for 1 h at 508C and separated on 1.4% agarose gels in 10 mM sodium phosphate buffer, pH 7.0, in a Sea2000 circular flow electrophoresis chamber (Elchrom Scientific). DNA and RNA gels were blotted onto Hybond N + membranes (GE Healthcare) overnight with 203 SSC and washed, and the samples were UV cross-linked using a Stratalinker (Stratagene). Hybridization was performed as described (Church and Gilbert, 1984). Radioactively labeled sequence-specific probes were synthesized from 25 ng of template DNA in the presence of 50 mCi of [a-32 P]dCT-(Hartmann Analytic) using the Rediprime labeling kit (Amersham). Signals of exposures in the linear range were detected with phosphor imager screens (Bio-Rad) and scanned with a Molecular Imager FX (Bio-Rad).

Quantification of Global DNA Methylation
Total cytosine methylation was determined by cation exchange HPLC as described by Rozhon et al. (2008). All samples were analyzed in triplicate, and 5-mdC values were expressed as a percentage of total cytosine.

Reverse Transcription and Quantitative Real-Time PCR
RNA samples were treated with 5 units of DNase I (MBI Fermentas), 0.4 units of RNasin, and 4 mL of 103 DNase I buffer for 40 min at 378C to remove residual DNA contamination, extracted with phenol:chloroform (24:1), and subsequently ethanol-precipitated. Reverse transcription was performed on 1 mg of RNA with 0.2 mg of random hexamer primers (MBI Fermentas) using 1 unit of RevertAid H Minus M-MuLV-RTase (MBI Fermentas) at 428C for 11/2 h. Real-time PCR analysis was performed with the 23 SensiMix Plus SYBR and Fluorescein Kit (Quantace) protocol using an iQ5 real-time PCR system (Bio-Rad Laboratories). Ct values were analyzed using Excel (Microsoft). The primer sequences are listed in Supplemental Table 1 online.

In Situ GUS Detection
GUS activity was detected by staining plant tissue in 0.1 M sodium phosphate buffe, pH 7.0, 10 mM EDTA, 0.1% Triton X-100, 100 mg/mL chloramphenicol, 2 mM potassium ferrocyanide, 2 mM potassium ferricyanide, and 0.5 mg/mL X-glucuronide after 30 min vacuum infiltration and overnight incubation at 378C. Subsequent washes with 70% ethanol at 378C were performed to remove chlorophyll and enhance contrast. All samples were analyzed using a Leica MZ16FA binocular microscope with a Leica DFC300FX CCD camera. Images were acquired with the Leica Application Suite and processed with Adobe Photoshop (Adobe).

Immunofluorescence
For the preparation of nuclei, 21-d-old plantlets were rinsed in 10 mM Tris buffer, pH 7.5, fixed by vacuum infiltration in 4% formaldehyde/Tris buffer, rinsed in Tris buffer, chopped in 500 mL chromosome isolation buffer (15 mM Tris, 2 mM Na 2 EDTA, 0.5 mM spermin, 80 mM KCl, 20 mM NaCl, 15 mM b-mercaptoethanol, and 0.1% Triton X-100, pH 7.5), and filtered through a 50-mm nylon mesh. Fifty microliters of suspension was transferred onto microscope slides, and nuclei were attached to the slide using a cytospin centrifuge (MPW) at 2500 rpm for 10 min.

Bisulfite Conversion, Sequencing, and Evaluation
After treatment with RNase A and proteinase K, 1 to 2 mg of genomic DNA were digested overnight with BamHI (MBI Fermentas). Subsequent bisulfite conversion was performed using the Epitect conversion kit (Qiagen) and controlled for completion as described (Hetzl et al., 2007). Converted DNA was used for PCR amplification (see Supplemental Table  1 online). PCR-amplified DNA was cloned using CloneJet (MBI Fermentas) and ligation mixes transformed into DH5a cells (Invitrogen), sequenced by terminal labeling using BigDye Terminator v3.1 (Applied Biosystems), and read at vbcgenomics.com. The sequence information obtained was analyzed with CyMATE (www.gmi.oeaw.ac.at/cymate; Hetzl et al., 2007) and Excel (Microsoft).

ChIP
ChIP was performed as described (www.epigenome-noe.net/ researchtools/protocol.php) using 3-week-old seedlings. The chromatin was immunoprecipitated with antibodies to histone H3 (Abcam; ab1791), H3K4me3 (Upstate; 07-473), H4pentaAc (Millipore; P62805), H3K9me2 (T. Jenuwein, 4677; Abcam, ab1220). Immunoprecipitated DNA was purified using a Qiagen PCR purification kit and eluted in 50 mL of EB buffer. Quantitative real-time PCR was performed in a total reaction volume of 25 mL, and quantitative PCR conditions were according to the 23 SensiMix Plus SYBR and Fluorescein kit (Quantace) protocol using an iQ5 real-time-PCR system (Bio-Rad Laboratories). Quantitative PCR data were evaluated as a ratio either to input DNA or to H3 abundance (Haring et al., 2007), as indicated.

Supplemental Data
The following materials are available in the online version of this article.