The botanist Emil Heitz first defined heterochromatin as nuclear material that remains condensed throughout the cell cycle, unlike the rest of the chromosome, which unravels between cell divisions (Fig. 1)1. Biology textbooks now portray heterochromatin as a 'junkyard' of silent noncoding DNA and defective transposons. This picture has emerged because heterochromatin is composed of DNA sequences with little or no coding potential, repeated thousands of times, and silenced by the covalent modification of the DNA itself and of the histones around which the DNA is wound2.
Figure 1: Heterochromatin, transposable elements and PEV.
a, PEV in the Drosophila eye results in red-white mottling (left). It is caused by the juxtaposition of heterochromatin with the white locus following inversion (mottled-w4). PEV is suppressed by the RNA-helicase mutant spindle-E/homeless (right) and other mutants in RNAi (reproduced with permission from ref. 62). b, Heterochromatic knobs on maize chromosome 7 stained at pachytene. Variability in the size and location of knobs was used by B. McClintock to trace the origin of maize landraces in central America (reproduced with permission from ref. 83). c, Transposon-mediated variegation in maize. Defective derivatives of the CACTA transposable element, dSpm (or inhibitor, I), can epigenetically control the anthocyanin gene A1 when they are integrated into the A1 promoter. In the absence of the autonomous controlling element Suppressor-mutator (Spm), or Enhancer (En), gene expression is lost during development (right) leading to variegation (reproduced with permission from Cold Spring Harbor Archives).
High resolution image and legend (78K)But heterochromatin has a controversial past because of its ability to influence the regulation of nearby genes. In Drosophila, juxtaposition of eye-colour genes with heterochromatin results in eyes that are mottled red and white (Fig. 1). This observation led Muller to coin the term 'position effect variegation' (PEV)2, 3. Indeed, Goldschmidt working on PEV in Drosophila, and McClintock working with 'controlling elements' or transposable elements in maize (Fig. 1), sought to elevate the status of heterochromatic gene silencing to that of a regulatory mechanism underlying development — an idea that received little support at the time4. Recently, however, as the genetic basis for PEV, transgene silencing, viral resistance and transposon regulation has emerged, along with the sequence of heterochromatic regions in plant and animal genomes, this idea has gained credence. One such regulatory mechanism relies on post-transcriptional regulation mediated by RNA, or RNA interference (RNAi). Unexpectedly, RNAi has been found to have a central role in heterochromatic gene silencing, despite the classical view that 'silent' heterochromatin is not transcribed into RNA.
Here, we review how heterochromatic silencing depends on the processing of repeat RNA transcripts into short interfering RNAs (siRNAs), which then direct chromatin modification. This mechanism explains how different repeats found in various eukaryotic genomes can be similarly incorporated into heterochromatin. Although many questions remain, these studies have resurrected the controversial suggestion that heterochromatic silencing is important in evolution and development.
Repeated sequences and their modification
Most heterochromatin is found near centromeres and telomeres, and consists of tandem (satellite) repeats, which are sometimes interrupted by transposable elements (Box 1). Pericentromeric repeats range from a few kilobases (kb) in fission yeast to 100–400-base pair (bp) short repeats that are arranged in Megabase-pair (Mb) arrays in mammals, plants and Drosophila (Box 1). Heterochromatic repeats share little similarity between species, but in all cases, heterochromatin is silenced by conserved epigenetic modifications of histones and DNA. This epigenetic silencing, as well as the higher-order packaging of repeats into heterochromatin, is believed to prevent illegitimate recombination, which can lead to chromosomal rearrangements. Furthermore, chromosomes are protected from active transposons, which can cause mutations when they are integrated into genes.
Methylation, acetylation, phosphorylation and ubiquitination of the core histones H2A, H2B, H3 and H4, and histone variants such as H2A.Z and H3.3, are implicated in gene regulation. These modifications are collectively referred to as the histone code5. Many of these modifications are specific for heterochromatin or euchromatin, such as methylation of histone H3 lysine 9 and lysine 4, respectively. In Drosophila, the suppressor of PEV Su(var)3-9 (Suppressor of variegation 3-9) encodes a SET (Su(var)3-9, Enhancer-of-zeste, Trithorax)-domain protein that is conserved in plants, animals and yeast. This protein is responsible for histone H3 lysine 9 methylation (H3mK9)6. Methylated lysine residues on histone H3 are recognized by chromo-domain proteins such as the highly conserved hetero-chromatin protein 1 (HP1)6, which is also a suppressor of PEV. The crystal structures of SET domains and chromo-domains indicate specific residues that determine which methylated lysines are recognized by each protein7.
The role of DNA methylation in heterochromatic gene silencing was recognized before that of histone modification8 (Box 2), even though it is less well conserved. DNA methylation is absent, or nearly absent in yeast, flies and nematodes, but a link between DNA methylation and histone methylation is well established in fungi apart from yeast, and in animals and plants8. Therefore, it is likely that DNA and histone modifications have a common role in gene silencing: they may even have a common origin.
In Neurospora crassa, DNA methylation depends on the Su(var)3-9 lysine 9 methyltransferase Dim-5 (ref. 9), whereas in Arabidopsis, plant-specific CNG and CNN methylation (where N is A, C, G or T) depends on the Su(var)3-9 homologue KRYPTONITE (KYP; also known as SUVH4), and on the CNG and CNN DNA methyltransferase CHROMOMETHYLASE3 (CMT3)10, 11, 12, 13. In Arabidopsis, methylation at histone H3 lysine 9 is lost from many transposons in the CG DNA methyltransferase mutant met1 (ref. 14), but it is retained by others15, and by centromeric satellite repeats16. In mammals, the DNA-methyltransferases Dnmt1 and Dnmt3b interact with the Suv39h (Suppressor of variegation 3-9 homologue) H3K9 methyltransferase and HP1, respectively, providing biochemical support for a functional link between DNA and histone methylation17, 18. Loss of H3mK9 in suvh39-/- knockout embryonic stem cells reduces DNA methylation at major centromeric satellites, but has little impact elsewhere19. Although dnmt1-/- or dnmt3-/- cells have normal amounts of heterochromatic H3mK9 (ref. 19), H3mK9 is lost when cells are treated with DNA methylation inhibitors20. So, histone H3 lysine 9 methylation and DNA methylation are interdependent, perhaps as a result of interactions between their respective methyltransferases. However, the inter-relationship between histone and DNA modification, although evolutionarily conserved, must depend, at least in part, on sequence specificity because not all DNA sequences are targeted equally for these modifications.
In Arabidopsis, the chromatin remodelling ATPase DDM1 (decrease in DNA methylation)21 is required for both DNA methyl-ation and H3mK9 (refs 16, 22): ddm1 mutants have been isolated along with met1 mutants in screens for loss of centromeric-repeat methylation23, 24. Nearly all the targets of DDM1 are transposons and tandem repeats, which are the focus of both DNA and histone methylation in plant genomes22, 25, 26. Both CG and CNG DNA methyltransferases contribute to transposon silencing in Arabidopsis27, as does histone deacetylation15, 28. H3mK9 seems to have a relatively minor role in transposon silencing, although double mutants between kyp and other histone methyltransferases have not yet been examined. As almost all DNA methylation and H3mK9 is confined to transposons and repeats, these elements must somehow be distinguished from genes. RNAi is one mechanism by which sequence-specific targeting might be achieved.
RNAi-mediated heterochromatic modifications in yeast
Heterochromatic silencing in the fission yeast Schizosaccharomyces pombe occurs at the centromeres, telomeres and the mating-type loci. Pericentromeric heterochromatin is composed of complex repeats, known as dg and dh. A single copy of each repeat is also found at the mating-type locus (Box 1). Silencing depends on the histone deacetylases Clr3 and Clr6, the histone H3K9 methyltransferase Clr4, and on the HP1 homologue Swi6 (ref. 29).
Key components of the RNAi pathway include a small RNA-binding protein called Argonaute (Ago), a dsRNA ribonuclease termed Dicer (DCR), and an RNA-dependent RNA polymerase (RdRP; also known as RDR) (see reviews in this issue by Meister and Tuschl, page 343, and Baulcombe, page 356). In S. pombe, unlike higher eukaryotes, each of these genes is unique. ago-, dcr- and rdp- mutants are viable but defective in silencing of reporter genes integrated into centromeric (but not mating-type) repeats30. Transcripts from both strands of the centromeric repeats are the targets of RNAi30, and siRNAs corresponding to these repeats can be detected31. In all three RNAi mutants, repeat-associated H3mK9 is reduced, and both H3mK9 and Swi6 are lost from centromeric reporter genes, demonstrating that RNAi is responsible for the heterochromatic modifications30. H3mK9 recruits Swi6 by means of its highly conserved chromodomain7, which in turn recruits cohesin — a highly conserved protein required for sister-chromatid cohesion and chromosome segregation during mitosis32. Loss of cohesin results in lagging chromosomes at anaphase in each of the RNAi mutants, as it does in clr4- and swi6- mutants33, 34.
Biochemical purification of chromodomain complexes in fission yeast has yielded the RITS (RNAi-induced transcriptional gene silencing) complex, which includes siRNA from centromeric repeats as well as Ago, a chromodomain protein Chp1, and a novel protein Tas3 (ref. 35) (Fig. 2a). Chp1 and Tas3 are bound to the centromeres in an RNAi-dependent manner, and in this respect they resemble RdRP30, which may be recruited to the centromere by means of siRNAs and nascent repeat transcripts36. It is possible that siRNA also recruits the RITS complex by means of Ago35, whose PAZ domain (named after piwi–argonaute–zwille) binds siRNA (see reviews in this issue by Meister and Tuschl, page 343, and Ambros, page 350). The RITS complex may also be recruited to the centromere by means of the Chp1 chromodomain, which binds H3mK9 (refs 35, 37). This would explain the reliance of RNAi on the H3K9 methyltransferase Clr4 (ref. 38). In either case, the mechanism by which H3mK9 arrives at centromeric repeats in the first place is not clear (Fig. 2a).
Figure 2: RNAi and heterochromatic silencing.
a, Silencing in S. pombe. The reverse strand of the centromeric repeats (Box 1) is transcribed in wild-type cells, but rapidly processed into siRNA by the RNAi pathway (1). The resulting siRNAs are amplified by RdRP (2), which is recruited to the repeats. siRNAs are also found in the RITS complex (3). These siRNAs are not required for the assembly of RITS but they are required for targeting. Targeting may be mediated either directly or indirectly by means of H3mK9 (red circles), which in turn probably interacts with Chp1. The role of RITS, if any, in modification of H3K9 by Clr4 (4) is not yet understood, but this leads to recruitment of another chromodomain protein Swi6, which silences the forward strand by TGS (5). b, Silencing in Arabidopsis. Transposons and tandem repeats account for 95% of siRNA in Arabidopsis, at least some of which is processed by DCL3 and RdR2 (RNA-dependent RNA polymerase 2), although other enzymes may be involved. At least three groups of genes, whose products may form complexes, have been proposed to account for heterochromatic modifications that are guided by siRNA. The first includes DRM1 and/or DRM2, which encode redundant Dnmt3-like methyltransferases. These are required for de novo DNA methylation including non-CG methylation. The second group includes CMT3, which is required for CNG methylation and is recruited by histone H3mK9 by means of KYP and possibly other histone methyltransferases. The third group includes the Dnmt1-like methyltransferase MET1, which is required to maintain CG methylation, but may also establish methylation in the presence of siRNA. Also included in this group is the histone deacetylase HDA6, which is probably required indirectly for H3mK9. Each putative complex is thought to bind siRNA. The first two groups include at least one AGO gene, including AGO4 and AGO1. The mechanism by which siRNA accumulation depends on MET1 and DDM1 is unknown.
High resolution image and legend (67K)Mating-type silencing is unaffected in ago-, dcr- and rdp- mutants30, as well as in chp1- and tas3- mutants35, indicating that some other mechanism must be responsible for the maintenance of mating-type silencing in the absence of RNAi. This mechanism involves histone deacetylation because silencing is lost in histone deacetylase mutants, or when histone deacetylase is inhibited by trichostatin A (TSA)29. When TSA inhibition is removed, mating-type silencing only returns slowly in RNAi mutants39. Unstable transgene silencing conferred by ectopic mating-type repeats also depends on RNAi33, 39. These observations suggest a role for RNAi in establishment rather than in maintenance of silencing39. However, it has recently been reported that maintenance of mating-type silencing also requires RNAi, but only in the absence of either one of two transcription factors (presumably repressors) in the ATF/CREB (activating transcription factor/cAMP response-element binding protein) family40. So RNAi has a role in maintenance of mating-type silencing, which is similar to its role at the centromere30, but this role is obscured by ATF.
The recent excitement surrounding the role of RNAi and microRNAs (miRNAs) in development (see review in this issue by Ambros, page 350) has overshadowed the fact that the first biological function attributed to RNAi was defence against transposons and viruses (see review in this issue by Baulcombe, page 356). Many of the genes required for RNAi in Caenorhabditis elegans, Drosophila and Chlamydomonas are also required for transposon suppression41, 42, 43. Moreover, siRNAs cloned from Drosophila and trypanosomes match several classes of repetitive sequences44, 45. The ability of Drosophila heterochromatic tandem repeats, known as Suppressor of Stellate (Su(Ste)), to silence the homologous Stellate genes depends on the RNA helicase spindle-E(spn-E)43. Mutations in this gene lead to the activation of transposons and other genomic repeats. In a dramatic example from Tetrahymena, macronuclear-DNA-elimination of repeats is RNAi-dependent and requires H3mK9, the AGO homologue TWI1 (related to piwi) and a chromodomain protein Pdd1p (programmed DNA degradation) (refs 46, 47).
Heterochromatic silencing in plants
Heterochromatic silencing in plants also involves RNAi, but DNA methylation is an additional factor (Box 2). Most (90–95%) endogenous siRNAs in Arabidopsis correspond to transposons and repeats whose histones and DNA are heavily methylated16, 22, 26, 48. Readthrough transcription of inverted repeats, transcription after the insertion of repeats into another transposon, or other mechanisms could account for the occurrence of transposon dsRNA49. Moreover, bidirectional transcription has been detected in met1 and ddm1 mutants26. As long as they are transcribed from one strand, tandem repeats can theoretically maintain a population of siRNAs through multiple rounds of replication mediated by RdRP36.
Despite the prevalence of siRNAs in Arabidopsis, only a handful of Arabidopsis transposons are activated in RNAi mutants15, 48, 50. AGO and other RNAi components are encoded by partially redundant multigene families (see review in this issue by Baulcombe, page 356), and at least some double ago mutants are lethal, complicating this analysis51. However RNAi-mediated silencing cannot be the only mechanism by which transposons are silenced, because siRNAs for many transposons are lost in dicer-like3 (dcl3) and rna-dependent rna polymerase2 (rdr2) mutants, but silencing is hardly affected48.
Once transposon silencing has been established by RNAi, silencing may be maintained by other means. In met1 and ddm1 mutants, which lose DNA methylation, transposons are activated, and most remain active (preset) in backcrosses15, 26. This means that MET1 and DDM1 can maintain transposon silencing, but cannot re-establish it once it is lost. Transposon silencing is also lost in the histone deacetylase mutant hda6/sil1 (where sil1 is an allele of hda6) but, in this case, silencing is re-established in backcrosses to wild-type plants. Most transposon siRNAs are lost from met1 mutants, but are retained in sil1 mutants. Furthermore, MET1 can re-silence some transposons in backcrosses, and the siRNA corresponding to these transposons are retained in met1 mutants15. It is possible, therefore, that siRNA (in cis) may be required for silencing de novo by means of MET1 and SIL1. A silencing complex that includes SIL1, MET1, DDM1 and RNA has been proposed on the basis of parallel complexes found in mammalian cells (Fig. 2b)15.
Although most silencing targets are transposons, phenotypic change can result from the silencing of genes. For example, the imprinted gene FWA encodes a homeodomain protein that controls flowering, and it is normally expressed in the endosperm; however, FWA is silenced in vegetative tissues by methylation of tandem repeats in its promoter52, 53. FWA transgenes are silenced when they are integrated into wild-type plants by means of Agrobacterium transformation, but not when they are transformed into dcl3, rdr2 or ago4 mutants, in which case ectopic expression and late flowering results54. Double mutants of the redundant de novo non-CG DNA methyltransferases domains rearranged methylase 1 and domains rearranged methylase 2 (drm1 and drm2) (Box 2) also fail to silence introduced transgenes: thus DRM1 and DRM2 are likely to establish DNA methylation in this case54, 55, 56. RNAi mutants (and loss of non-CG methylation) have little impact on silencing of endogenous FWA54, but CG methylation and silencing is lost in met1 mutants, resulting in heritable late-flowering mutant epialleles52. This indicates a role for MET1 (Dnmt1-like) in maintenance of silencing, and a role for RNAi, DRM1 and DRM2 in establishment of silencing (Fig. 2b)54. However, the possibility that RNAi has a redundant role in silencing maintenance cannot be ruled out.
Interestingly the tandem repeats that make up the promoter of FWA are generated by the insertion of a short interspersed nucleotide element (SINE) related to A. thaliana SINE 2 (AtSN2) (ref. 26). These repeats accumulate siRNA and H3mK9, both of which are lost in met1 mutants (ref. 26). So loss of transposon siRNA may account for the inheritance of late-flowering FWA epialleles from met1 mutants, and for the failure to establish FWA-transgene silencing in dcl3, rdr2 and ago4 mutants54. siRNAs corresponding to AtSN1 are also lost in all four mutants15, 48, 50, indicating that SINE elements may be targets of this silencing mechanism.
The loss of siRNA from met1 and ddm1 mutants and the epigenetic inheritance of transposon activity in plants may be similar to the loss of siRNA in clr4- mutants and the epigenetic inheritance of mating-type de-repression in S. pombe38, 39. In both cases, 'maintenance' chromatin-modification enzymes may also function in establishment by guiding interactions with siRNAs. Despite differences in DNA methylation, the dependence of heterochromatin silencing on RNAi is similar in S. pombe and Arabidopsis, including the strand-specific regulation of centromeric repeats (B. May et al., unpublished data). Furthermore, fission yeast retrotransposon silencing is more dependent on histone deacetylation than on RNAi (K. R. Hansen et al., unpublished data), resembling Arabidopsis transposon silencing in this respect (ref. 15). Therefore, both DNA methyltransferases and histone deacetlyases on the one hand, and RNAi and histone methyltransferases on the other, participate in establishment and maintenance of heterochromatic silencing, although different targets may be involved in each mechanism15, 28, 30.
RNAi-dependent silencing in plants
The earliest examples of sequence-specific RNA-mediated gene silencing came from plants (Box 3). Recently, models for RNA-dependent silencing have allowed the isolation of silencing mutants in Arabidopsis. Transcriptional gene silencing (TGS) and methylation of three different genes depends, at least in part, on the transcription of a homologous inverted repeat, which ultimately results in RNA-dependent DNA methylation or RdDM. Silencing of PAI2 and SUPERMAN (SUP) is stabilized by inverted repeats of each entire gene, including the promoter10, 55, 57. In contrast, silencing of a bacterial transgene can be achieved using inverted repeats of the nopaline synthase promoter (NOSpro) alone58. In at least two of these three examples, transcription of the inverted repeat is required for silencing28, 59. Furthermore, for all these genes, histone and DNA methylation is disrupted in otherwise viable and fertile mutants. Silencing of PAI2 and SUP requires CMT3 and KYP10, 11, 13, whereas silencing of NOSpro depends on MET1 and on the histone deacetylase HDA6/SIL1 (ref. 28).
Only one in more than forty silencing mutants isolated in the three screens for disrupted silencing of PAI2, SUP or NOSpro, is defective in RNAi: a single mutant allele of the Ago homologue ago4 is defective in silencing SUP50. It might, therefore, be concluded that RNAi is of little importance in RNA-dependent DNA methylation. However, at least one of the inverted-repeat transgenes (NOSpro) gives rise to siRNA, suggesting that the RNAi machinery is involved28. Furthermore, inverted-repeat silencing systems in addition to SUP have defects in non-CG methylation in ago4 mutants, resembling drm1 and drm2 double mutants55, even when silencing of target loci is unaffected60. Moreover, the inverted repeats themselves are hypomethylated in ago4 (ref. 60), which may contribute to production of functional protein in the case of SUP. Finally, a mutant allele of another Ago homologue, ago1, has defects in DNA methylation that are associated with post-transcriptional gene silencing (PTGS)61. Thus RNAi clearly plays a role in RdDM in plants, but this role can be obscured by other silencing mechanisms.
Transposons are regulated by the same mechanisms as those required for RdDM15. Most retrotransposons are silenced by MET1 and HDAC6/SIL1; the retrotransposons AtCOPIA44/TA3 and AtCOPIA4 also require the H3K9 methyltransferase KYP, the chromo-methylase CMT3 and either AGO4 or AGO1 (refs 15, 50) (Fig. 2b). It is possible that SUP, PAI2 and the transposable element COPIA on the one hand, and NOSpro and the transposable element GYPSY on the other, have distinct classes of cis-acting regulatory sequences. Alternatively, SUP, PAI2 and COPIA may represent unstable (facultative) heterochromatin that requires H3mK9 for silencing. NOSpro and GYPSY may be stably silenced by further histone deacetylation (such as that of H2A, H2B and H4), as well as by DNA methylation. Further analysis of histone modifications will be required to identify which of these possibilities is the case.
RNAi and heterochromatin in animals
RNAi also has a role in heterochromatin formation in animals. In Drosophila, the Ago mutants aubergine and piwi disrupt TGS of transgene tandem repeats, which lose H3mK9 (ref. 62). In spindle-E mutants, H3mK9 is also lost, and two HP1 homologues are no longer associated with heterochromatic foci62. Like the histone H3K9 methyltransferase Su(var)3-9, all these mutants also suppress PEV62. These same genes regulate the post-transcriptional silencing of retrotransposons and tandem repeats, in addition to aspects of transcriptional silencing of transgenes63. Also in Drosophila, mutants of dcr-2, but not dcr-1, affect the accumulation of siRNA derived from an inverted repeat transgene64. Drosophila dcr-2 mutants are similar to dcl3 mutants in Arabidopsis in that both primarily affect siRNA, and are phenotypically wild type in all other respects. This suggests that in Drosophila, DCR-2 is only involved in the silencing of heterochromatin. aubergine and piwi mutants, on the other hand, resemble Arabidopsis ago1 mutants in having defects in transgene and heterochromatic silencing, and in having strong developmental phenotypes61, 63.
A role for RNAi in centromeric silencing in C. elegans is difficult to assess because canonical centromeres are lacking. However, RNAi regulates transposons in the germ line, although not in somatic cells in which Tc1 transposable elements are active49. Silencing of extra-chromosomal transgene arrays also occurs in the germ line and is associated with H3mK9 (ref. 65). So, germ line heterochromatin formation may depend on RNAi. It would not be surprising if transposons were also associated with H3mK9. Germline-specific components of the RNAi machinery have been identified in C. elegans49, which may be directly involved in targeting heterochromatin formation as occurs in S. pombe, Arabidopsis and Drosophila.
Linking RNAi with heterochromatic silencing in mammals has been more difficult, particularly because very few repeat-derived siRNAs have been cloned (see review in this issue by Ambros, page 350). Nonetheless, the localization of H3mK9 and HP1 to pericentromeric heterochromatin in mouse is abolished by RNase treatment66. Moreover, mice with null mutations of the Suv39h histone methyltransferases show reduced viability and chromosomal instabilities19, reminiscent of clr4- mutants in S. pombe, which are also defective in RNAi38. Dicer-null mice arrest very early in development, and this has been attributed to a defect in stem-cell maintenance67. Lethality could also result from the loss of centromeric silencing and chromosome segregation defects, which may be more problematic for multicellular organisms than for S. pombe. Centromeric satellite repeats are methylated in wild-type mouse cells, and some classes of repeats are expressed from both strands19, 68. The generation of conditional Dicer knockouts will be necessary to test for transcriptional activation of these repeats and for changes in methylation. Finally, two out of nine Ago homologues in mice have been knocked out, and both mutants show spermatogenesis defects and male sterility69, 70. One of these mutants is blocked completely at meiosis I (ref. 70), which could be a consequence of heterochromatin defects.
RNAi-dependent heterochromatin and chromosomal imprinting
At least in principle, RNAi-dependent histone and DNA modification can 'imprint' chromosomal sequences in such a way that these modifications are subsequently maintained throughout cell division — resembling heterochromatin in this respect1. This might, in part, account for epigenetic mechanisms of gene silencing in plants and animals, such as imprinting (whereby gene expression depends on the parent of origin for the chromosome). Parent-of-origin-specific imprints imposed in the germ line are 'remembered' after fertilization, and are associated with changes in DNA and histone H3 methyl-ation71, 72. In plants, both FWA and MEDEA are imprinted in the endosperm, and are associated with tandem repeats and DNA methyl-ation53, 55, 73, 74. The silencing of FWA depends on RNAi and hetero-chromatic transposons26, 54, although a role for RNAi in the gametophyte has not yet been demonstrated. Transposons and repeats have been found at many imprinted loci in mammals and noncoding transcripts have been implicated. However, a role for RNAi itself is far from clear71, 75, 76.
Paramutation in maize is an example of a chromosomal imprint where a silent state can be transferred from one allele to another in somatic cells77. This imprint is stable and heritable, and depends on upstream tandem repeats that are differentially methylated in paramutagenic and paramutable alleles77. It will be interesting to determine whether these repeats correspond to small RNAs.
RNA components direct several other forms of epigenetic silencing, and although a role for RNAi has not been demonstrated, aspects of this heterochromatic regulation share common features with that mediated by RNAi. X-chromosome inactivation in mammals is responsible for dosage compensation in XX females78, and is mediated by a noncoding RNA, Xist (X-inactive specific transcript). Transcripts from both strands of this internally repetitive region are required for X-chromosome silencing, which is characterized by histone and DNA methylation79. X-chromosome inactivation is imprinted in the mouse, in that only the paternal chromosome is silenced in the extraembryonic tissues78. Dosage compensation in Drosophila uses a different mechanism, but also uses noncoding roX RNA that interacts with chromodomains80.
Finally, RNAi mutants in animals and plants frequently have stem-cell defects (see reviews in this issue by Ambros, page 350 and Baulcombe, page 356). Stem-cell divisions can be thought of as chromosomal imprints if the parental chromosome set is retained in mother cells of asymmetric stem-cell lineages — as has been indicated in some studies of mammalian cells81. It is tempting to implicate centromeric transcripts in these imprints, given that RNAi is required for normal chromosome segregation33, 34. The transcribed strands are inherently asymmetric, at least in yeast30 and plants (B. May et al., unpublished data), providing a plausible mechanism.
Perspective
RNAi provides a mechanism for programmed, sequence-specific silencing by means of heterochromatic modifications, which is likely to be particularly important for silencing transposons and repeats. In 1951 McClintock proposed gene control by transposable elements as a potential mechanism for developmental regulation82, 83. Although imprinting and cellular memory may well be regulated in this way, heterochromatic silencing could prove to be important over a longer timeframe in population genetics and evolution. Heterochromatin is a dynamic component of the genome1 and varies between natural ecotypes of most species83, 84. In wide crosses between diverse strains, such as those responsible for generating natural allopolyploids (individuals containing multiple chromosome sets derived from different species), siRNAs from one parent might not match polymorphic repeats from the other. If siRNA pools are largely maternal, cytoplasmic siRNA could contribute to chromosome loss and transposon activation (hybrid dysgenesis) if they fail to match paternal chromosome sequences85, 86, 87. Conversely, loss of silencing in transposon-regulated genes could lead to increased gene expression, and contribute to hybrid vigour25. Aspects of the RNAi-mediated heterochromatic silencing mechanism are still unclear. In particular, we know little about the polymerase responsible for generating dsRNA, or how specific sequences are targeted. However, this RNAi-dependent gene silencing mechanism will surely have widespread implications for the biological role of heterochromatin.