Gene Silencing and Homology-Dependent Gene Silencing in Arabidopsis: Genetic Modifiers and DNA Methylation

Corresponding author: Ian J. Furner, Department of Genetics, University of Cambridge, Cambridge CB2 3EH, UK, ijf{at}mole.bio.cam.ac.uk (E-mail).

Communicating editor: D. PREUSS

	ABSTRACT

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Transgenes inserted into the plant genome can become inactive (gene silencing) or result in silencing of homologous cellular genes [homology-dependent gene silencing (HDG silencing)]. In an earlier study we reported HDG silencing of chalcone synthase (CHS) in Arabidopsis. This study concerns genetic revertants of one of the CHS HDG-silencing transgenic homozygotes. Two monogenic recessive trans-acting mutations (hog1 and ddm1) that impair gene silencing and HDG silencing were identified. These mutations reduce genomic DNA methylation and affect the quantity and size of CHS mRNA. These results imply that DNA methylation is necessary for both gene silencing and HDG silencing. Two further monogenic, trans-acting, recessive mutations (sil1 and sil2) reduce gene silencing but not HDG silencing. The existence of this mutant class shows that gene silencing involves genes that are not necessary for HDG silencing. A further mutant (C^att) was isolated and has an attenuated HDG-silencing T-DNA.

THE genes of higher plants occasionally undergo silencing, and this results in comparatively stable epimutant alleles. Such epimutants have been found to be correlated with increased DNA methylation in maize (DAS and MESSING 1994 ) and very recently in Arabidopsis (JACOBSEN and MEYEROWITZ 1997 ). Trivial variation in allelic methylation can be inherited in a Mendelian fashion (MESSEGUER et al. 1991 ). Transgenes introduced into higher plants frequently undergo silencing (reviewed in FINNEGAN and MCELROY 1994 ), and in some cases silencing has been correlated with increased DNA methylation (e.g., KILBY et al. 1992 ; reviewed in STAM et al. 1997 ).

More surprisingly, introduced transgenes can result in silencing of homologous endogenous genes. This phenomenon is called cosuppression (NAPOLI et al. 1990 ; reviewed in JORGENSEN 1995 ). Similarly, newly introduced transgenes can silence previously introduced transgenes by a process called transinactivation (reviewed in MATZKE and MATZKE 1993 , MATZKE and MATZKE 1995 ). A variety of silencing events between alleles (paramutation) or between duplicated members of gene families have also been reported (reviewed in MATZKE et al. 1996 ). The overall term to describe this huge variety of events is homology-dependent gene silencing (HDG silencing; reviewed in MEYER and SAEDLER 1996 ). Cosuppression, transinactivation, paramutation, and other examples of HDG silencing have, in some cases, been shown to correlate with DNA methylation of the silenced sequences, but the biological role of such methylation is uncertain. If the role of DNA methylation in silencing and HDG silencing is causal, removing the hypermethylation should also relieve silencing. Conversely, mutants impaired in silencing and/or HDG silencing might also be impaired in DNA methylation. If methylation merely accompanies changes brought about by other mechanisms, altering methylation should not relieve either silencing or HDG silencing.

Arabidopsis is a good model system to analyze gene silencing and HDG silencing. Some years ago we showed that transgenes introduced into Arabidopsis could undergo gene silencing and that this was correlated with methylation of a restriction site in the promoter (KILBY et al. 1992 ). Two inactive homozygous inbred lines (E and G) were generated. Each contained the nopaline synthase promoter fused to neomycin phosphotransferase (nosNPT). Tissue homozygous for the E and G inserts becomes resistant to kanamycin when grown in the presence of the demethylating agent 5 azacytidine, suggesting that the observed DNA methylation has a causal role in the nosNPT silencing.

In order to examine HDG silencing in Arabidopsis we produced transgenic plants containing extra copies of the chalcone synthase (CHS) gene (DAVIES et al. 1997 ). CHS is coded by a single gene at the TT4 locus. Mutants of TT4 have no obvious phenotype in the plant but do have a transparent testa (TT4) and yellow seeds (wild-type seeds are brown). CHS mediates an early step in the biosynthesis of the purple pigment anthocyanin. CHS can be induced by a variety of treatments, including high-intensity white light and growth on sucrose (DAVIES et al. 1997 ). Under such conditions, TT4 plants are purple and tt4 plants are green. Three independent HDG-silencing transgenic lines were identified by screening for transformed plants that were less purple than wild-type plants in the presence of high-intensity light. After backcrossing, three lines (A, B, and C) were produced, each of which was homozygous for a single HDG-silencing insert (A insert, etc). The A-insert homozygotes showed the least HDG silencing, the B-insert homozygotes were intermediate, and it was difficult to reproducibly detect anthocyanin in C-insert homozygotes.

Silencing by the C insert showed as a dominant epistatic effect on the TT4 locus, and plants either homozygous or heterozygous for the C insert had a green phenotype with yellow seeds (like tt4 mutants). Heterozygous plants generated by outcrossing the C-insert homozygotes were frequently variegated (striped), and this appeared to be related to the time taken for HDG silencing to act on the naive wild-type TT4 allele (DAVIES et al. 1997 ). The A, B, and C inserts were found to be highly methylated, and derivatives of the C-insert homozygotes that had lost their T-DNA retained partial silencing and methylation of the CHS copy at the TT4 locus. The partial TT4 silencing and methylation in the epimutant alleles is both stable and heritable (I. J. FURNER, unpublished results). These results suggest a role for methylation in HDG silencing of TT4, but they do not establish a causal role for it in the process.

In this study we set out to disturb gene silencing, HDG silencing, and methylation of the C-insert homozygotes (C/C plants). As we used a previously described mutant (ddm1) in this work, it is necessary to first review the biology of this line. The ddm1 (decreased DNA methylation) mutation is a monogenic recessive trait, and in the homozygous state it results in a genome-wide demethylation of both single-copy and multicopy sequences (VONGS et al. 1993 ). The hypomethylation induced by the mutation tends to persist in outcross heterozygotes. Hypomethylation induced by 5-azacytidine also persists for at least two generations in the absence of the drug (KILBY et al. 1992 ). The precise biochemical lesion of the ddm1 mutant is uncertain (KAKUTANI et al. 1995 ). Inbreeding the ddm1 line results in the generation of stable epigenetic variants (RICHARDS 1997 ). The effects of ddm1 (VONGS et al. 1993 ) on C-insert-induced HDG silencing and gene silencing were examined. Mutants affecting gene silencing and HDG silencing were isolated and characterized.

	MATERIALS AND METHODS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

General methods:
Culture of plants, media, and growth conditions were as previously described (KILBY et al. 1992 ). Anthocyanin assays, sucrose induction, and Northern and Southern blots were as in our earlier article (DAVIES et al. 1997 ). In these experiments, light induction was for 9 days with an 8-hr photoperiod. Segregations of the reactivated transgenes were performed on plates supplemented with 300 µg/ml kanamycin. The sole exception was the segregations of the inactive nosNPT homozygotes E and G, which were plated on 50 µg/ml kanamycin (KILBY et al. 1992 ). The level of drug resistance in the various homozygotes was quantified by germinating seeds on a range of drug concentrations. After 10 days, five seedlings were harvested per plate, and the chlorophyll content of each was measured spectrophotometrically (LICHTENTHALER et al. 1983). The results were plotted against the drug concentration, and the concentration resulting in a 50% reduction in chlorophyll content was estimated by interpolation. Crosses between plants were performed using emasculated flowers. Typically immature (closed) flowers were emasculated and pollinated 3 days later. Unless they were crossed plants were maintained by self-fertilization. The term "family" is used to describe the progeny of a single self-fertilized plant. The C-insert homozygous line (C/C) was also made homozygous for a second mutation (clv1-2) as a control for stray phenotypically wild-type seeds or pollen (LEYSER and FURNER 1992 ). The ddm1 allele used in this study was ddm1-2 (KAKUTANI et al. 1995 ).

Mutant isolation:
M₂ lines derived from clv1-2 C/C plants were screened for anthocyanin-producing plants under inducing conditions. The M₂ lines were made either by 16 kilorads (Kr) X-irradiation (FURNER and PUMFREY 1992 ) or by 100 mM ethylmethane-sulphonate (EMS) treatment (LEYSER and FURNER 1992 ). Approximately 40,000 plants were screened in a variety of experiments. Thirteen putative HDG-silencing modifier lines were found. Unfortunately, most of these had lost their T-DNA inserts, presumably by deletion (DAVIES et al. 1997 ). Only two lines had both the T-DNA and reproducibly higher anthocyanin levels (b1 and 15B). It became apparent that some of the HDG-silencing modifier lines also relieved the gene silencing of the NPT and hygromycin phosphotransferase (HPT) genes on the C insert. In order to isolate more gene-silencing modifiers, ~30,000 M₂ seeds (as generated above) were plated on 100 µg/ml kanamycin to screen for resistant lines. Two resistant plants, CX2 and CX3, were successfully recovered. At the end of the screens there were four new, stable, true-breeding lines. Each was shown to contain the C-insert T-DNA by Southern blots (data not shown). The 15B line was isolated from EMS-derived material, and the other mutations were X-ray induced. Each line was from a different batch of mutagenized seeds, and the mutations are presumably independent in origin. The lines used in this study and their origins are described in Table 1.

Generation of T-DNA-free derivatives of hog1 and sil1:
In order to isolate the hog1 and sil1 mutations without the C insert, homozygous plants were crossed to wild type. F₃ families (from individual F₂ plants) segregating 3/4 plants resistant to kanamycin (300 µg/ml) were identified in the expectation that they would be homozygous for the modifier and segregating for the T-DNA (e.g., hog1/hog1, C/O). In the F₄ generation, 100%-sensitive families were identified (e.g., hog1/hog1 O/O; nontransgenic) from the segregating families. DNA from the sil1-no-T-DNA (sil1NOT) and hog1NOT homozygotes was examined using Southern blots to confirm the absence of the T-DNA (data not shown). The two NOT lines were crossed to the two inactive nosNPT homozygotes (E and G) from the earlier gene-silencing study (KILBY et al. 1992 ). The segregation of kanamycin resistant:sensitive plants was examined in the F₂ generation.

	RESULTS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

The C insert contains multiple copies of the T-DNA:
The transforming construct is shown in Figure 1. The T-DNA (and the transgenic C/C line) contains the CHS, HPT, and NPT genes. The structure of the C insert was examined using Southern blots of genomic DNA from C/C plants and probes to various regions of the T-DNA. The fragment patterns were complex (data not shown), presumably because of extensive rearrangement during transformation or integration. It was not possible to produce a map of the insert. However, it was possible to estimate approximate copy numbers for the various regions of the insert. These were NPT and HPT, five to nine copies; pBR322, at least three copies; and CHS, at least three copies (in addition to the TT4 copy). Given that the T-DNA is 16 kb in length (Figure 1) and that there appear to be three or more copies of most regions, the C insert is a complex, scrambled structure greater than 50 kb in length.

View larger version (15K): In this window In a new window Download PPT slide   Figure 1. —The construct used to generate the C-insert plants (upper part of the figure) and a detailed restriction map of the CHS region of the construct (lower part of the figure). The plasmid PGJD-4 is ~22 kb long and the T-DNA ~16 kb long. The ampicillin-resistance gene (AMP) and the tetracyclin-resistance gene (TET) are expressed in bacteria. The other three genes (NPT, HPT, and CHS) have plant promoters; the left (LB) and right borders (RB) are indicated. The restriction sites are EcoRI (E), HindIII (H), and MspI (M). The MspI map of CHS is redrawn from DAVIES et al. 1997 ; the construction of PGJD-4 is described in the same article. The HPT and NPT genes have no MspI fragments longer than 300 bp. The NPT gene has the nos promoter, the HPT gene has the CAMV 35S promoter, and the CHS gene is a full-length genomic clone driven by its own promoter.

The C/C plants are sensitive to both hygromycin and kanamycin, implying that both the HPT and NPT genes are inactive (Table 1). The gene inactivity (silencing) of the HPT and NPT genes and the HDG silencing of the CHS gene are presumably a function of the C-insert location, structure, and/or methylation. In the earlier study we demonstrated high levels of methylation of CHS in C/C plants (DAVIES et al. 1997 ). The inactivity of the C-insert-encoded drug markers suggests that the methylation (and silencing) extends to HPT and NPT transgenes. This point is examined in a subsequent section.

The stably silenced phenotypically green C-insert homozygote was mutagenized and screened for purple (phenotypically wild-type) revertants (see MATERIALS AND METHODS). Two true-breeding purple lines with brown seeds were isolated (b1 and 15B).

The C^att mutation is an attenuated C insert:
The b1 line was crossed to C/C plants and wild-type plants, and the phenotypes of the F₁ and F₂ plants noted (Table 2). The F₁ plants of the cross of b1 to wild type were purple, and the F₂ generation segregated 3:1, pale purple:purple. Because no green plants were seen in the F₂ generation, the modifier did not segregate away from the T-DNA, and the line therefore contains an attenuated insert (C^att) with reduced HDG silencing. As the F₁ plants of the cross between C^att homozygotes and C homozygotes are green, C^att is recessive to C. This was confirmed in the F₂ generation, in which the plants were segregating 3:1 green:pale purple. The anthocyanin level in C^att homozygotes is comparatively low, less than 10% that found in wild-type plants (Figure 2), and the seeds are brown (Table 7). Southern blots of C^att/C^att and C/C genomic DNA cut with a variety of enzymes were made and probed with the CHS gene; the restriction fragment patterns were indistinguishable (data not shown). This result implies that there was no rearrangement of the C-insert-encoded CHS gene copies in the C^att mutant.

View larger version (14K): In this window In a new window Download PPT slide   Figure 2. —Anthocyanin levels in a variety of homozygous lines induced using white light. Results are expressed as OD530/plant as a measure of anthocyanin content. The genotypes are as indicated, and all lines were homozygous for the C insert (except wild type). The mean standard errors are shown (n = 10). The growth condition and assays were as previously described (DAVIES et al. 1997 ).

The hog mutation is a trans-acting modifier of HDG silencing and gene silencing:
The 15B line was isolated on the basis of its purple phenotype under inducing conditions (Table 7). The 15B plants are small and per plant contain about 1/4 as much anthocyanin as wild-type plants (Figure 2). In other experiments in which anthocyanin was expressed on the basis of fresh weight, the 15B line was found to contain more anthocyanin than wild type (data not shown). Genetic analysis suggested that the 15B line contained a single simple recessive mutation. The trait was called hog1 (homology-dependent gene silencing 1). Data on dominance of the purple phenotype in the F₁ generation are presented in Table 3A, and the F₂-generation segregations are presented in Table 4. The F₁-generation plants of crosses to wild type are variegated (showing silencing), and the F₁-generation plants of the crosses to C-insert homozygotes are green (silenced), implying that the hog1 mutation is recessive. In the F₂ generation the hog1 trait segregated 3:1 (green:purple) in crosses to C/C plants and 7:9 (purple:green) in crosses to wild-type plants (Table 4). The genetics of this line is a good fit to an ideal second site trans-acting recessive modifier of HDG silencing.

During routine work on the hog1 homozygotes, it became apparent that they were more kanamycin resistant than the parental line. Similar results were obtained with hygromycin (Table 7). The hog1 C/C line shows at least a fivefold increase in both kanamycin and hygromycin resistance compared with the parental C/C plants. The result implies that hog1 relieves the gene silencing of the T-DNA-encoded HPT and NPT genes. In order to examine the inheritance of the reactivation, high-level kanamycin resistance was analyzed in the F₁ (Table 3B) and F₂ (Table 5) generations of hog1 crosses to wild type and the C/C line. The results were anomalous. The reactivated NPT was dominant in the F₁ generation of crosses to wild type and recessive in crosses to C/C plants. In the F₂ generation, the reactivated NPT gene behaved like a broadly recessive trait, but there was an unexpected excess of resistant seedlings in crosses to C/C plants and a deficit in crosses to wild type (Table 5). In general, plant color (purple:green) appears to give more straightforward genetics with the hog1 trait than does the associated NPT reactivation. Several crosses involving 15B (hog1) showed deficits of seeding classes, and the purple plants were unhealthy. This appears to be due to other unlinked EMS-induced mutations, because T-DNA-free derivatives of hog1 are wild type (see T-DNA-free lines, below).

The ddm1 mutation also relieves HDG silencing and gene silencing:
We suspected that the C-insert-induced HDG silencing was related to its methylation and perhaps also to methylation of the CHS copy at TT4 (DAVIES et al. 1997 ). In order to examine these propositions, we crossed C-insert homozygotes and ddm1 homozygotes, produced F₁ and F₂ generations, and examined the segregation of phenotypes (purple:green) of the latter. If ddm1 was acting as a trans-acting modifier of HDG silencing of CHS, the expected ratio should be 7:9 (purple:green). This is what was observed (Table 4).

The purple plants from this cross were harvested individually and screened for families showing 100% purple and 100% strongly kanamycin-resistant plants. Such plants were assumed to be homozygous for both ddm1 and the C insert. Such ddm1 C/C plants have ~40% wild-type levels of anthocyanin (Figure 2), and the plants are strongly kanamycin and hygromycin resistant. This suggests that, like hog1, the ddm1 mutation relieved gene silencing as well as HDG silencing (Table 7). Genetic studies of the ddm1 (C/C) isolate included complementation in the F₁ generation (Table 3), segregation in the F₂ generation of the purple phenotype (Table 4), and segregation in the kanamycin-resistant phenotype (Table 5). In the F₁ generation the ddm1-related phenotypes both behave like dominant traits (F₁ plants are purple and kanamycin resistant). In many of the F₂ segregations, purple and kanamycin-resistant seedlings were in excess of the recessive expectation.

The sil1 and sil2 mutants impair gene silencing but not HDG silencing:
As the hog1 and ddm1 mutations relieved both gene silencing and HDG silencing, resulting in purple kanamycin-resistant plants, it seemed possible that selecting for reactivation of NPT (by kanamycin resistance) would generate similar lines. Two such lines were isolated (CX2 and CX3; MATERIALS AND METHODS and Table 1). The plants were green under inducing conditions, and the anthocyanin levels were only slightly higher than those of C-insert homozygotes (Figure 2), so the plants showed relief of gene silencing of NPT but little—if any—relief of HDG silencing of CHS. CX3 seeds are brown (Table 7), perhaps suggesting a slight relief of HDG silencing in the testa. CX2 seeds are yellow (Table 7). The two mutations appear to be independent monogenic recessive modifiers of silencing and were named sil1(CX3) and sil2(CX2). In studies of dominance in F₁, both lines gave ambiguous results in crosses to C/C plants, and sil1 was clearly recessive to wild type (Table 3). In the F₂ generation both sil1 and sil2 behaved like simple monogenic recessive modifiers of NPT silencing, segregating 3:1 (sensitive:resistant) in crosses to C-insert homozygotes and 3:13 in crosses to wild type (Table 5). The sil2 mutant shows weak and variable kanamycin resistance in the C/C background. The penetrance of sil2 is low, and the trait proved difficult to score. For this reason, only limited work was done on this mutant. The F₂ segregation data are incomplete (Table 5).

If sil1 and sil2 were alleles at a single locus, the F₂ generation of the cross between them should be entirely kanamycin resistant. Conversely, if the traits were at different loci, the F₂ generation should segregate 7:9 resistant:sensitive. The result was intermediate (Table 5). It may be that the genes affect one another, resulting in an excess of kanamycin-resistant seedlings over the expected 7:9 ratio. It is hard to see how sil1 and sil2 could be alleles at a single locus and have 1/3 sensitive seedlings in the F₂ generation. The sil2 mutant does not reactivate the HPT gene of the C insert, whereas sil1 does (Table 7)—which is further evidence that the two mutants are in different genes. Finally, the sil2 mutant codes comparatively weak and impenetrant kanamycin resistance (in the C/C background), and it may be that the sil1 and sil2 mutants are allelic and the resistant class was underscored. Further work will be needed to resolve this point.

The F₁ complementation data on crosses between ddm1, hog1, sil1, and sil2 (Table 3) are difficult to interpret. The F₂ generation of the cross between hog1 and ddm1 segregated 7:9 purple:green (Table 4) and showed an unexpected excess of kanamycin-sensitive seeds (Table 5). The two mutants are independent, noncomplementing, and unlinked. A similar case can be made for crosses of ddm1 and hog1 to sil1, respectively. These crosses gave a good fit to a 7:9 ratio for kanamycin resistant:sensitive (Table 5). Insufficient F₂ segregation data were obtained with the sil2 mutant to draw any meaningful conclusion about possible allelism to the other mutants.

Generation and analysis of T-DNA-free lines of hog1 and sil1:
In order to isolate a T-DNA-free line from hog1 c/c homozygotes, such plants were crossed to wild type. In the F₃ generation, families homozygous for hog1 and segregating the C insert were identified. Families homozygous for hog1 and lacking an insert were identified in the F₄ generation (MATERIALS AND METHODS). These lines, hog1NOT (no T-DNA) and sil1NOT, were purple (under inducing conditions) and kanamycin sensitive. They have been maintained by self-fertilization to the F₆ generation—and, unlike ddm1 inbreds, these plants have no obvious phenotypic abnormalities. As hog1NOT is of normal stature, the small stature and slow growth of the original isolate (15B) was caused by other mutations that were lost in the crosses to produce the NOT derivative.

Crosses were made to two 5-azacytidine-reactivable methylated nosNPT homozygotes (E and G; Table 1 and KILBY et al. 1992 ). The resulting F₂-generation plants and control crosses to wild type were scored for kanamycin resistance. The experiment was partially successful (Table 6). Crosses of the inactive homozygotes to wild type gave no reactivation. The sil1 mutation was effective in reactivating the G insert and had little effect on the E insert. In contrast, the hog1 mutation was highly effective in reactivating the E insert but had a relatively weak effect on the G insert. In the two most effective reactivations, the ratio of resistant:sensitive seedlings approached the 3:13 ratio expected of an unlinked recessive modifier (Table 6).

RNA level in sucrose-induced lines:
It is difficult to produce RNA from light-induced plants (DAVIES et al. 1997 ). In contrast, intact RNA can be easily obtained from sucrose-induced seedlings. A Northern blot was made of total RNA from sucrose-induced seedlings of the homozygotes listed in Table 7 (Figure 3, upper panel). Sucrose induction resulted in a different pattern of phenotypes from light induction, so the levels of anthocyanin were measured in parallel samples under sucrose induction (Figure 3, lower panel).

View larger version (30K): In this window In a new window Download PPT slide   Figure 3. —Northern blots of total RNA from sucrose-induced seedlings probed with CHS (upper panel) and rDNA (center panel). The lower panel shows anthocyanin determinations of parallel samples. All lines were homozygous for the C insert except wild type. The rDNA probe was that used by VONGS et al. 1993 , and the CHS probe was as previously described (DAVIES et al. 1997 ). The samples used in the anthocyanin assays were the seedlings derived from 10 mg (~500) seeds grown in 2% sucrose for 10 days. The mean standard errors are shown (n = 5).

The wild type has a high level of intact CHS mRNA and high levels of anthocyanin. The C-insert homozygotes had only trace amounts of low-molecular-weight CHS mRNA and no detectable anthocyanin. In the hog1 C-insert double homozygote, CHS mRNA was found to be both abundant and intact, and the seedlings had levels of anthocyanin comparable to wild type. In the ddm1 C-insert double homozygote there was a considerable increase in CHS mRNA size and abundance compared with the C-insert homozygotes, but little intact CHS mRNA and only low levels of anthocyanin were found. The other homozygotes contained no detectable CHS mRNA and little or no anthocyanin. Overall, the anthocyanin level shows a strong correlation with the level of intact CHS message. The hog1 and ddm1 mutations affect both the size and abundance of the CHS mRNA, presumably by affecting its transcription and/or turnover. The blot was reprobed with an rRNA probe (Figure 3, center panel), and the loadings were comparable.

The ddm1 and hog1 mutations result in demethylation of CHS, NPT, HPT, and the rDNA:
A series of Southern blots were made from DNA of the lines listed in Table 7 using the methylation-sensitive isoschitzomers MspI and HindIII or HpaII and HindIII (Figure 4). In the wild type, CHS was unmethylated and the rDNA highly methylated. In the C-insert homozygotes, sil1(C/C), sil2(C/C), and C^att/C^att, the CHS, NPT, HPT, and the rDNA genes were all highly methylated (poorly cut). In contrast, the ddm1(C/C) and the hog1(C/C) lines showed demethylation of the three T-DNA genes and the rDNA. The results show the expected global demethylation of ddm1. The results also imply that hog1 is a similar methylation-deficient mutant to ddm1, and sil1 and sil2 must act in a manner that is methylation independent, as these mutants have wild-type levels of methylation. The sil2 track on the four blots is underloaded, and the hybridization is less intense than in the adja-cent tracks. This gives a misleading impression of sil2-induced demethylation of the HPT and NPT genes. Longer exposure of these blots revealed that the sil2 (C/C) methylation pattern is indistinguishable from the C/C methylation pattern (data not shown). The hog1 homozygote showed less cutting (more methylation) than the ddm1 homozygote with each of the four probes.

View larger version (67K): In this window In a new window Download PPT slide   Figure 4. —Southern blots of DNA of a variety of homozygous lines cut with the methylation-sensitive enzymes MspI (M) and HindIII or HpaII (H) and HindIII. The probes, plant lines, and molecular weights are indicated. All lines were homozygous for the C insert except wild type.

	DISCUSSION

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

The evidence for the involvement of methylation in epimutation, cosuppression, transinactivation, paramutation, and other silencing and HDG silencing events is strong but largely correlative (STAM et al. 1997 ). The molecular mechanisms involved in these phenomena have been the subject of a fairly lively debate with no clear consensus emerging, even on the most basic points (MATZKE et al. 1996 ; MEYER and SAEDLER 1996 ; STAM et al. 1997 ). At this point it is unclear whether silencing and HDG-silencing phenomena are aspects of a single process or a collection of diverse processes with little or nothing in common. The study of mutants impaired in silencing has the potential to reveal the molecular mechanisms involved. The only published mutants that specifically enhance gene silencing in Arabidopsis affect a meiotically reversible transgene-silencing phenomenon (DEHIO and SCHELL 1994 ). Two loci were identified (egs1 and egs2), and the data suggested post-transcriptional control. The molecular basis of the mutant phenotype is uncertain.

While this manuscript was in revision, some novel Arabidopsis mutants that relieve the gene silencing of an inactive HPT transgene were reported (MITTELSTEN SCHEID et al. 1998 ). A total of eight putative mutants (som 1–8) were described, three mutants were found to be alleles of the DDM1 gene (VONGS et al. 1993 ), one mutant appeared to be a recessive mutant of another locus, and the rest could not be classified because of slow F₁ resilencing of the transgene. All the lines showed global reductions in methylation. The authors also showed that the ddm1 mutant could relieve the HPT gene silencing. Overall, this study implies a causal role for methylation in gene silencing.

Silencing and homology in fungi:
HDG-silencing-like processes have been found in Neurospora (COGONI and MACINO 1997 and references therein), Ascobolous (MALAGNAC et al. 1997 and references therein), Schitzophyllum (SCHUURS et al. 1997 ), and Coprinus (FREEDMAN and PUKKILA 1993 ). These processes have common elements, including presence of duplicated transgenes and methylation of both the transgenes and endogenous genes. In Neurospora the gene silencing in the vegetative phase is called "quelling," and deficient (qde) mutants have been found (COGONI and MACINO 1997 ). The 15 mutants recovered allowed the identification of three complementation groups and a nonrecessive class. The mutants accumulate transgene-sense RNA, but the molecular defect in the lines is unclear.

Neurospora also shows a silencing associated with the sexual cycle (repeat induced-point mutation; RIP). This is associated with GC to AC mutations and extensive methylation of the endogenous gene (ROUNTREE and SELKER 1997 ). The alleles produced by this process are both epimutants (because of the methylation) and true mutants (because of the point mutations). The dim-2 mutation, which abolishes methylation, results in expression of a RIP copy of the am gene. This implies that the methylation—and not the mutations—causes the silencing. In these experiments the transcriptional initiation was normal, but elongation of the transcripts from the methylated am gene was impaired. Demethylation induced by the dim-2 mutation relieves the blockage to elongation and allows the generation of full-length transcripts. Very recently, a novel DNA methyltransferase was cloned from Ascobolous (MALAGNAC et al. 1997 ). Disruption of the gene impairs de novo methylation and an HDG-silencing-like phenomenon called methylation induced premeiotically (MIP). The proven absolute requirement for methylation in HDG silencing in Neurospora and Ascobolous implies that a similar requirement might exist in higher plants.

The hog1 and ddm1 mutants are impaired in maintenance methylation, gene silencing, and HDG silencing:
In this study we identified a novel, second-site modifier of HDG silencing and gene silencing. The hog1 trait is a monogenic recessive mutation that is unlinked to the HDG silencing C insert. The C-insert homozygotes contain trace amounts of low-molecular-weight CHS message. In plants homozygous for hog1 and the C insert, the abundance and size of the CHS message are similar to those in wild-type plants. The hog1 mutation also reduces the level of methylation of the three C-insert transgenes and the rDNA. This reflects a failure in maintenance methylation, as the parental C-insert homozygotes were methylated in all four genes. The hog1 mutation also results in expression of the silenced NPT and HPT genes of the C insert. The previously described methylation defect (ddm1; VONGS et al. 1993 ) is nonallelic but shows most of the same properties as hog1 in the C-insert homozygous background, including loss of HDG silencing of CHS; demethylation of all four genes; increases in abundance and size of CHS transcripts (the effect was less complete than with hog1 under the conditions used); and reactivation of the C-insert-encoded HPT and NPT genes.

A model for gene silencing and HDG silencing:
Taken together, these results imply that methylation is absolutely necessary for both gene silencing and HDG silencing. The mutation-induced demethylation affects both the size and abundance of mRNA from the HDG-silenced CHS gene, restoring it toward wild-type levels. HDG silencing is more complex than gene silencing in that both the transgene and the endogenous gene have to be silenced. The model outlined in Figure 5 proposes that methylated transgenes produce a signal (X) that results in two things: silencing of the endogenous gene and methylation of the endogenous gene (CHS copies at TT4 become methylated in the presence of the C insert; DAVIES et al. 1997 ). In the absence of DDM1 or HOG1 function, maintenance methylation fails and methylation is lost from the CHS copies on the C insert and at TT4. At this point it is unclear whether demethylation of TT4, loss of production of X from the C insert, or both are important in the release from HDG silencing. The hog1 and ddm1 relief of gene silencing at NPT and HPT is directly comparable to the relief of silencing of methylated am alleles by dim-2, and the mechanism may be the same: that is an effect on mRNA elongation rather than initiation (ROUNTREE and SELKER 1997 ).

View larger version (15K): In this window In a new window Download PPT slide   Figure 5. —A model for the roles of the DM1, HOG1, SIL1, and SIL2 gene products in gene silencing and HDG silencing. The model proposes that the C insert produces a diffusable signal, X, that results in methylation and silencing of the CHS copy at TT4. Production of X is dependent on C-insert methylation, and the methylation-deficient mutant homozygotes (hog1 and ddm1) do not produce it (center panel). Transgene silencing is dependent on both methylation and SIL1 and SIL2. In the homozygous sil1 and sil2 mutants, NPT and HPT are methylated and expressed (lower panel). The model proposes that SIL1 and SIL2 code proteins that bind the methylated C insert and impose gene silencing on HPT and NPT (upper panel). The sil1 and sil2 mutant homozygotes still produce X, and this results in HDG silencing of TT4. The way the model is drawn implies that the effect of X is at the DNA level.

Mutants specifically impaired in gene silencing but not HDG silencing:
Selection for plants with high-level kanamycin resistance from the inactive NPT gene on the C insert allowed the isolation of two new mutants, sil1 and sil2. The sil mutants impair silencing of the C-insert-encoded NPT but have little or no effect on HDG silencing of CHS. Gene silencing appears to require both methylation and SIL functions, whereas HDG silencing requires only methylation and no SIL functions. One proposal that could potentially explain these observations is that the SIL1 and SIL2 gene products recognize methylated DNA and interfere with the production of functional mRNA (Figure 5, lower panel). In the model, the HDG-silencing signal, X, can be produced irrespective of the absence of the SIL gene products. In the sil mutants gene silencing would be relieved, but HDG silencing would remain, as X would still be produced. This is not a particularly novel model, and other authors have proposed roles for methyl-binding proteins in transcriptional repression (reviewed in KASS et al. 1977). The model conspicuously fails to explain the failure of sil2 to reactivate HPT. However, the model does fit in with the frequent finding of transcriptional defects associated with gene silencing (STAM et al. 1997 ).

Silencing deficient mutants reactivate other inserts:
The hog1 and sil1 mutants were segregated away from the C insert and crossed to plants homozygous for 5-aza-cytidine-reactivable nosNPT inserts (KILBY et al. 1992 ). The reactivation was imperfect, but in the two cases where it worked (hog1 and the E insert, sil1 and the G insert), the ratio of resistant:sensitive seedlings in the F₂ generation approached the expected 3:13 ratio of an unlinked, recessive, transgene-silencing modifier. The result implies that hog1 and sil1 are not specific for the C-insert silencing and have activating effects on other silent inserts. Both the NPT on the C insert and the same gene on the E and G inserts have the nos promoter (the C-insert HYG gene has the 35S promoter). Because hog1 and sil1 can work on both different inserts and promoters, their effects must be nonspecific. Perhaps they have the potential to act as modifiers of epimutants of endogenous genes, such as some alleles of sup1 (JACOBSEN and MEYEROWITZ 1997 ).

Relationships to other mutants:
The hog mutant is not an allele of the DDM1 locus (Table 4). In the original description of the ddm1 mutant a similar nonallelic mutant was mentioned but not described (mutant C IN VONGS et al. 1993 ). The hog1 mutant might be allelic to this mutant. The hog1 mutant does not show slow transgene resilencing, as do four of the som mutants (MITTELSTEN SCHEID et al. 1998 ), and hog1 is nonallelic to ddm1 (three of the som mutants are DDM1 alleles). It is possible that hog1 is allelic to som2, which shows rapid transgene silencing and is nonallelic to ddm1. Further work will be needed to define how many genes affect methylation in Arabidopsis. It seems likely that such an analysis will have to be pursued by examining segregations in the F₂ generation, because in the F₁ generation, complementation experiments with methylation mutants can give misleading results because of the persistence of the transgene expression. This was evident with the ddm1 mutant in this study (Table 3 and Table 4) and four of the eight som mutants (MITTELSTEN SCHEID et al. 1998 ). In contrast to ddm1, hog1, and the som mutants, the sil mutants are not methylation deficient and define at least one novel silencing-related gene that has not been previously described.

A cis-acting mutant affecting HDG gene silencing:
The identity of the signal produced by the C insert (X) is uncertain. However, the attenuated derivative of the C insert isolated in this study is a good candidate for a mutation reducing the level of X. The C^att trait is recessive to the C insert, and the C^att mutant shows reduced HDG silencing and higher levels of anthocyanin compared with the C insert. The silencing of the transgenes in the C^att homozygotes resembles that of the C-insert homozygotes (HPT and NPT both inactive). The ge-netic data suggest that the C^att T-DNA has a mutation resulting in reduced production of X (compared to C). The reduced level of X results in increased expression of CHS presumably from TT4. The results do not distinguish between the possibilities that the C^att mutation is either within the T-DNA or that it is merely linked to the C insert. There is no obvious reduction in CHS methylation in the C^att homozygote DNA, but it should be borne in mind that the homozygotes have multiple copies of the gene as well as the TT4 copy, and demethylation of single copies could easily be missed. It is possible that C^att is not a mutant at all and is a stable epimutant of the C insert.

The identity of the HDG-silencing signal (X):
The model (Figure 5) proposes that HDG-silencing inserts produce a signal, X, that acts in trans to the target gene and results in both HDG silencing and methylation of the target gene. Production or perception of X is DNA methylation dependent and homology dependent. Recently it has been reported that petunia plants showing CHS HDG silencing contain truncated transcripts polyadenylated at unusual positions (METZLAFF et al. 1997 ). These appear to result from a degradative process. It has also been shown that reproducing RNA viroids can provoke methylation of genomic DNA with homology to them (WASSENEGGER et al. 1994 ). The simplest proposition that could account for these observations is that X is a transcript derived from the methylated DNA of the C insert and that it differs in some structural fashion from the normal transcripts from TT4. For example, if methylation blocks elongation in Arabidopsis as it does in Neurospora, 3' truncated transcripts would presumably be produced that might interact with the normal transcripts, resulting in their degradation. Such molecules might also be involved in the methylation of the CHS copy at TT4 by a mechanism similar to the viroid example (WASSENEGGER et al. 1994 ).

Genetics of mutations affecting epimutation:
This study has identified four loci that affect gene silencing or HDG silencing. The mutants were assessed indirectly by their failure to show gene silencing and/or HDG silencing. In several cases, these mutants showed contradictory patterns of dominance in the F₁ and F₂ generations and often excesses of expressed types in the F₂ generation. The ddm1 mutant shows the strongest effects. It seems probable that this is a result of segregation of semistable epigenetic variation in gene and transgene expression levels superimposed on the underlying genetic segregation. The ddm1 mutation is recessive—but demethylation provoked by it can persist (VONGS et al. 1993 ), and plants homozygous for this trait give rise to stable epimutants (RICHARDS 1997 ). In general, sil1 and sil2 behave like simple monogenic recessive traits in F₂, whereas ddm1 and hog1 give more unusual ratios. An excess of purple kanamycin-resistant plants is found in many crosses in which the C insert has been maintained in the ddm1 background. In an earlier article, we pointed out that HDG silencing of the naive wild-type TT4 allele is slow, resulting in variegation (DAVIES et al. 1997 ). Similarly, once the C insert has been demethylated by ddm1, resilencing (by remethylation) is also slow. In both examples, de novo methylation must be involved. In this work, the genetics of the modifier genes was studied indirectly by observing their effects on silencing and HDG silencing of the C-insert transgenes and TT4. The modifiers, in effect, alter the epigenetic states of the transgenes and TT4 by relieving silencing. Reestablishment of a new equilibrium between silent and expressed genes takes generations.

	FOOTNOTES

¹ Present address: Department of Biology, University of York, Heslington, York YO1 5DD, UK.

	ACKNOWLEDGMENTS

We thank ERIC RICHARDS for providing the ddm1 mutant. We also thank CAROLYNE SHARON CHIPPS for typing the manuscript and GARETH DAVIES for part of Figure 1. The C-insert homozygote is being distributed by the Nottingham Arabidopsis Stock Centre, Nottingham, UK, and the modifiers will be made available in a similar fashion. The work was carried out in accordance with the Ministry of Agriculture Food and Fishery guidelines (license PHF174B/60L60).

Manuscript received December 20, 1997; Accepted for publication March 3, 1998.

	LITERATURE CITED

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

DAS, O. P. and J. MESSING, 1994  Variegated phenotype and developmental methylation changes of a maize allele originating from epimutation. Genetics 136:1121-1141[Abstract].

DAVIES, G. J., M. A. SHEIKH, O. J. RATCLIFFE, G. COUPLAND, and I. J. FURNER, 1997  Genetics of homology-dependent gene silencing in Arabidopsis: a role for methylation. Plant J. 12:791-804[Medline].

DEHIO, C. and J. SCHELL, 1994  Identification of plant genetic loci involved in a post-transcriptional mechanism for meiotically reversible transgene silencing. Proc. Natl. Acad. Sci. USA 91:5538-5542[Abstract/Free Full Text].

COGONI, C. and G. MACINO, 1997  Isolation of quelling-defective (qde) mutants impaired in post-transcriptional transgene-induced gene silencing in Neurospora crassa.. Proc. Natl. Acad. Sci. USA 94:10,233-10,238[Abstract/Free Full Text].

FINNEGAN, J. and D. MCELROY, 1994  Transgene inactivation: plants fight back!. Bio/Technol. 12:883-888.

FREEDMAN, F. and P. J. PUKKILA, 1993  De novo methylation of repeated sequences in Coprinus cinereus.. Genetics 135:357-366[Abstract].

FURNER, I. J. and J. E. PUMFREY, 1992  Cell fate in the shoot apical meristem of Arabidopsis thaliana.. Development 115:755-764[Abstract].

JACOBSEN, S. E. and E. M. MEYEROWITZ, 1997  Hypermethylated SUPERMAN epigenetic alleles in Arabidopsis.. Science 277:1100-1102[Abstract/Free Full Text].

JORGENSEN, R. A., 1995  Co-suppression, flower color patterns, and metastable gene expression states. Science 268:686-691[Abstract/Free Full Text].

KAKUTANI, T., J. A. JEDDELOH, and E. J. RICHARDS, 1995  Characterization of an Arabidopsis thaliana DNA hypomethylation mutant. Nucleic Acids Res. 23:130-137[Abstract/Free Full Text].

KASS, S. U., D. PRUSS, and A. P. WOLFE, 1997  How does DNA methylation repress transcription? Trends Genet. 13:444-449[Medline].

KILBY, N. J., H. M. O. LEYSER, and I. J. FURNER, 1992  Promoter methylation and progressive transgene inactivation in Arabidopsis.. Plant Mol. Biol. 20:103-112[Medline].

LEYSER, H. M. O. and I. J. FURNER, 1992  Characterization of three shoot apical meristem mutants of Arabidopsis thaliana.. Development 116:397-403.

LICHTENTHALER, H. K. and A. R. WELBURN, 1983  Determination of total carotenoids and chlorophylls a and b of leaf extracts in different solvents. Biochem. Soc. Trans. 603:591-592.

MALAGNAC, F., B. WENDEL, C. GOYON, G. FAUGERON, and D. ZICKLER et al., 1997  A gene essential for de novo methylation and development in ascobolus reveals a novel type of eukaryotic DNA methyltransferase structure. Cell 91:281-290[Medline].

MATZKE, M. A. and A. J. M. MATZKE, 1993  Genomic imprinting in plants: parental effects and trans-inactivation phenomena. Annu. Rev. Plant Physiol. Plant Mol. Biol. 44:53-76.

MATZKE, M. A. and A. J. M. MATZKE, 1995  How and why do plants inactivate homologous (trans)genes? Plant Physiol. 107:679-685[Medline].

MATZKE, M. A., A. J. M. MATZKE, and W. B. EGGLESTON, 1996  Paramutation and transgene silencing: a common response to invasive DNA? Trends Plant Sci. 1:382-388.

MESSEGUER, R., M. W. GANAL, J. C. STEFFENS, and S. D. TANKSLEY, 1991  Characterization of the level, target sites and inheritance of cytosine methylation in tomato nuclear DNA. Plant Mol. Biol. 16:753-770[Medline].

METZLAFF, M., M. O'DELL, P. D. CLUSTER, and R. B. FLAVELL, 1997  RNA-mediated RNA degradation and chalcone synthaseA silencing in petunia. Cell 88:845-854[Medline].

MEYER, P. and H. SAEDLER, 1996  Homology-dependent gene silencing in plants. Annu. Rev. Plant Physiol. Plant Mol. Biol. 47:23-48.

MITTELSTEN SCHEID, O., K. AFSAR, and J. PASZKOWSKI, 1998  Release of epigenetic gene silencing by trans-acting mutations in Arabidopsis.. Proc. Natl. Acad. Sci. USA 95:632-637[Abstract/Free Full Text].

NAPOLI, C., C. LEMIEUX, and R. JORGENSEN, 1990  Introduction of a chimeric chalcone synthase gene into petunia results in reversible co-suppression of homologous genes in trans.. Plant Cell 2:279-289[Abstract/Free Full Text].

RICHARDS, E. J., 1997  DNA methylation and plant development. Trends Genet. 13:319-323[Medline].

ROUNTREE, M. R. and E. U. SELKER, 1997  DNA methylation inhibits elongation but not initiation of transcription in Neurospora crassa.. Genes Dev. 11:2383-2395[Abstract/Free Full Text].

SCHUURS, T. A., E. A. M. SCHAEFFER, and J. G. H. WESSELS, 1997  Homology-dependent silencing of the SC3 gene in Schizophyllum commune.. Genetics 147:589-596[Abstract].

STAM, M., J. N. M. MOL, and J. M. KOOTER, 1997  The silence of genes in transgenic plants. Ann. Bot. 79:3-12[Abstract/Free Full Text].

VONGS, A., T. KAKUTANI, R. A. MARTIENSSEN, and E. J. RICHARDS, 1993  Arabidopsis thaliana DNA methylation mutants. Science 260:1926-1928[Abstract/Free Full Text].

WASSENEGGER, M., S. HEIMES, L. RIEDEL, and H. L. SÄNGER, 1994  RNA-directed de novo methylation of genomic sequences in plants. Cell 766:567-576.