Mum's the Word

As complete genomic sequences of model organisms continue to flow from the sequencing centers to an eager scientific community, there is a growing appreciation that full understanding of any genome must consider inherited information that the sequencing machines cannot detect. The rich history of plant genetics provides numerous examples of non-Mendelian phenomena demonstrating that identical DNA sequences can adopt alternative information states. These states can persist through a number of cell divisions, and in some cases, they may be inherited through many generations of organisms. Perhaps the most familiar example of such epigenetic modification is the cycling activity of transposons, which can shift between active and inactive states. Another classic example is paramutation: one allele alters the activity of its allelic partner on the homologous chromosome through a chro-matin level change that leaves the underlying DNA sequence unaltered (Hollick et al., 1997).

With the advent of genetic transformation of plants in the 1980s, a renewed interest in epigenetics grew from the fascinating and frustrating behavior of transgenes whose expression often become extinguished by silencing mechanisms operating above the level of the gene sequence (Matzke et al., 1989; Napoli et al., 1990; Vaucheret et al., 1998). Such silencing can take two forms, which at the surface appear to be mechanistically distinct, although cross-talk between the pathways is sug-gested by several studies (Ingelbrecht et al., 1994; Wassenegger et al., 1994; Jones et al., 1999; Mette et al., 1999). The first mechanism, post-transcriptional gene silencing (PTGS), is caused by turnover of specific transcripts in the cytoplasm and is mechanistically related to RNA interference (RNAi) in animals. In both PTGS and RNAi, RNA turnover is stimulated by double-stranded RNA and the silencing takes the form of a cytoplasmic metabolic state that can be transmitted systemically from cell to cell. The second mechanism, transcriptional gene silencing (TGS), reduces gene expression by suppressing transcriptional initiation and is often correlated with cytosine methylation of promoter regions. In the case of TGS, the epigenetic information is transmitted as a metastable chromatin state embedded within the chromosome. There is particular interest in transgene TGS mechanisms because of their importance in plant biotechnology and their possible contribution toward the management of complex genomes rich in repeated DNA.

Some of the molecular players participating in TGS have been identified through genetic screens for Arabidopsis suppressors of transcriptionally silenced transgenes. To date, five different trans-acting modifiers of transgene TGS have been reported in Arabidopsis: HOG1, SIL1, SIL2, SOM/DDM1, and MOM1 (Vongs et al., 1993; Furner et al., 1998; Mittelsten Scheid et al., 1998; Amedeo et al., 2000). Mutations in HOG1, SIL1 and SIL2 were identified by Furner et al. (1998) using a transcriptionally silenced multicopy chalcone synthase (CHS) locus as a screenable reporter. In addition to its silencing effect, the hog1 mutation causes DNA hypomethylation of genomic sequences, including the CHS transgene. At the same time, Furner and colleagues also demonstrated that ddm1, a mutation originally identified by its reduction of genomic methylation, released gene silencing of the CHS transgene. Paszkowski, Mittelsten Scheid, and colleagues performed an independent TGS modifier selection using an Arabidopsis strain (line A) containing a silenced multicopy (5× to 10×) transgene encoding hygromycin phosphotransferase (HPT) (Mittelsten Scheid et al., 1998). Their initial report described the identification of eight som mutations that reactivated and hypomethylated the line A HPT transgene locus and also reduced methylation of centromeric repeats. Of the eight som mutations, at least five are ddm1 alleles (Jeddeloh et al., 1999). These genetic studies establish a connection between DNA methylation and maintenance of TGS.

Other observations, however, question the role of DNA methylation in TGS. First, antisense expression of MET1, encoding the major maintenance cytosine methyltransferase in Arabidopsis, reduces methylation of the line A HPT reporter locus without releasing silencing (Mittelsten Scheid et al., 1998). Second, a novel class of mutations relieves transgene TGS while maintaining a normal level of cytosine methylation. Two members of this class, sil1 and sil2, do not alter methylation of the reactivated CHS transgene, nor do they affect ribosomal DNA methylation (Furner et al., 1998). Amedeo et al. (2000) recently reported the isolation of another member of this class, mom1, in a T-DNA tagging screen for reactivation of the line A HPT locus. Southern analysis and genomic sequencing did not detect significant methylation changes in the viral-based promoter of the HPT transgene in the mom1 mutant. These results suggest that cytosine methylation is not sufficient to enforce TGS, although minor methylation changes are difficult to rule out because of the complex repetitive nature of the reporter used.

It is likely that endogenous targets are under control of the same TGS mechanisms that suppress transgene expression. In this issue of THE PLANT CELL, on pages 1165–1178, Steimer et al. identify a silent endogenous target (TSI, for transcriptionally silent information), which is misregulated in Arabidopsis TGS mutants. Subtractive hybridization between the original silenced HPT line A and mom1 mutants yielded two cDNA clones, TSI-A and TSI-B, corresponding to transcripts reactivated in mom1 mutants. TSI is a member of a family of degenerate Athila retroelements located in the pericentromeric region of all five Arabidopsis chromosomes. Interestingly, transcriptional initiation begins in the middle of the element leading to the production of three major species of RNA. TSI transcription is also activated in hog1, sil1, som/ddm1 and met1 backgrounds. The fact that all transgene TGS modifiers identified to date (with the exception of sil2) affect TSI in a similar manner suggests that heterochromatic retroelement remnants are important endogenous silencing targets.

The results of Steimer et al. (2000) are consistent with the view that epigenetic phenomena are a consequence of chro-matin level regulation directed at transposable elements to reign in their potentially damaging effects (Martienssen, 1996, 1998a, 1998b; Yoder et al., 1997). Further support comes from another recent study identifying retrotransposons as targets for TGS in Arabidopsis. Hirochika et al. (2000) showed that the tobacco retrotransposon Tto1 becomes transcriptionally silenced and inactive when introduced into Arabidopsis at high copy number. Disruption of DDM1 function led to transcriptional activation and transposition of the Tto1 elements. In addition, one silent endogenous Arabidopsis retroelement (Tar17) was transcriptionally reactivated in ddm1 mutants, although no new Tar17 transposition events were detected (Hirochika et al., 2000). At least one Arabidopsis transposable element family is activated in a TGS mutant. Martienssen and colleagues (Singer et al., manuscript submitted) showed that Arabidopsis mutator-like transposable elements are demethylated, transcriptionally activated and transpose into novel locations within the genome of ddm1 mutants. Interestingly, Steimer et al.. (2000) did not detect a general release of retroelement silencing in the mom1 mutant despite the mutation's reactivation of transcription from the degenerate Athila elements.

While epigenetic regulation is commonly targeted to transposons and repetitive sequences, silencing can also affect low copy sequences. One example involves the small Arabidopsis gene family encoding phosphoribosyl anthranilate isomerase (PAI), a tryptophan biosynthetic enzyme. One functional family member (PAI2) is methylated and transcriptionally silenced in certain wild type strains (Bender and Fink, 1995), but the gene is reactivated in a ddm1 mutant background (Jeddeloh et al., 1998). Similarly, DDM1 is necessary to enforce silencing of the parentally imprinted gene MEDEA (MEA), which is inherited in a silenced state on the paternally transmitted chromosome (Vielle-Calzada et al., 1999). It is currently unknown whether PAI and MEA silencing are affected in other mutants altered in TGS, including mom1.

All the above cases involve the release of silencing, but paradoxically, certain TGS modifiers can lead to ectopic silencing. Hypermethylated silenced alleles of the floral developmental genes SUPERMAN (SUP) and AGAMOUS (AG) were identified in MET1 antisense, met1 and ddm1 lines (Jacobsen and Meyerowitz, 1997; Jacobsen et al., 2000). Although the majority of the genome is hypomethylated in these mutants, the novel sup and ag alleles are formed when methylation is targeted to these previously unmethylated low-copy genes. This redistribution of methylation parallels the silencing and ectopic methylation of tumor suppressor genes that can occur during the early stages of human carcinogenosis against a background of global DNA hypomethylation (Jones and Laird, 1999). The mechanisms involved in generating these abnormal methylation patterns are unknown, but they may involve induction of compensatory methylation activities with altered specificity or generation of regional chromatin states more accessible to methylation. Regardless of the mechanism, the potential evolutionary importance of stably silenced genes is demonstrated by the discovery of hypermethylated, silenced Lcyc alleles in toadflax, controlling natural variation in floral symmetry (Cubas et al., 1999).

Considering the consequences of a breakdown in transcriptional silencing discussed above, one might expect TGS mutants to be hobbled by new mutations caused by reactivated transposons and stably inherited epigenetic alterations. Consistent with this prediction MET1 antisense lines and som/ddm1 mutants exhibit developmental defects due to inherited alterations in the genome (Finnegan et al., 1996; Kakutani et al., 1996; Ronemus et al., 1996; Amedeo et al., 2000). Specific genomic sites appear to be primary targets for alterations in these backgrounds, including sup, ag, and late flowering alterations mapping to FWA (Kakutani, 1998; Jacobsen et al., 2000). Curiously, the mom1 mutant does not show any developmental defects even after nine generations of inbreeding to allow mutations or epigenetic changes to accumulate. The discrepancy in the phenotypic effects of the different TGS mutations suggests that mom1 affects fewer loci than those under control of DNA methylation. This hypothesis is consistent with the differential expression screen reported by Steimer et al. (2000) that yielded two independent TSI clones from their secondary pool of 12 clones. One possibility is that MOM1 mediates silencing of sequences found only in particular chromosomal/subnuclear compartments, or possessing particular genomic organizations (e.g., multicopy, inverted repeats). In a compatible explanation, MOM1 may be restricted to one of several independent silencing pathways operating downstream of DNA methylation. Alternatively, MOM1 and DNA methylation may act on independent silencing pathways with overlapping target specificity. In this scenario, targets such as TSI, under the influence of both MOM1 and methylation-dependent silencing, may be reactivated when a single pathway is compromised. As a next step, it will be important to identify additional transcripts misregulated in mom1 mutants to define the spectrum of MOM1 action.

The deduced protein sequence of MOM1 has a number of interesting similarities, including a region related to SWI2/SNF2 chromatin remodeling engines (25 to 28% identity over a 340 to 390 amino acid window encompassing three of seven signature motifs) (Amedeo et al., 2000). The SWI2/ SNF2 family members most closely related to MOM1 fall in the CHD subfamily implicated in chromatin mediated repression in animals. Recently, a candidate Arabidopsis CHD orthologue, PICKLE/GYMNOS, was isolated through two independent genetic screens (Ogas et al., 1997; Eshed et al., 1999). Loss of function mutations in PICKLE were shown to derepress LEC1 expression, which is involved in specifying embryo identity (Ogas et al., 1999). It is not known if TGS targets are generally affected in pkl/gym mutants. The potential connection between TGS and chromatin remodelers is strengthened by the finding that DDM1 also encodes a SWI2/SNF2–like protein (Jeddeloh et al., 1999). It is important to stress, however, that despite the suggestive protein similarities, ATP-dependent chromatin remodeling has not yet been demonstrated for any of these proteins.

In addition to SWI2/SNF2–like proteins, at least one other chromatin component has been shown to be important in TGS in Arabidopsis: CURLY LEAF (CLF), which encodes a Polycomb-group (PcG) protein (Goodrich et al., 1997). PcG represents a class of proteins that forms a repressive chromatin structure mediating gene silencing of homeotic genes in Drosophila. Loss of CLF function in Arabidopsis results in ectopic expression of the AG and AP3 floral genes in vegetative tissues (Goodrich et al., 1997). Mutations in CLF, like pkl/gym alleles, were identified in screens for developmental mutants and not directly from searches for TGS modifiers. We anticipate that many more chromatin level modifiers of gene silencing will result from continued forward genetic screens, as well as functional genomics approaches.