insight review articles
`
`Gene silencing as an adaptive
`defence against viruses
`
`Peter M. Waterhouse*, Ming-Bo Wang* & Tony Lough†
`
`*CSIRO Plant Industry, Canberra, ACT 2601, Australia (e-mail: peterw@pi.csiro.au)
`†HortResearch, Tennent Drive, Palmerston North, New Zealand (Present address: Genesis Research and Development Corporation, PO Box 50,
`Auckland, New Zealand)
`
`Gene silencing was perceived initially as an unpredictable and inconvenient side effect of introducing
`transgenes into plants. It now seems that it is the consequence of accidentally triggering the plant’s adaptive
`defence mechanism against viruses and transposable elements. This recently discovered mechanism,
`although mechanistically different, has a number of parallels with the immune system of mammals.
`
`Biology students are taught that the concept of
`
`vaccination came from Edward Jenner’s
`discovery that milkmaids and dairymen
`infected with the mild cowpox virus were
`protected against smallpox. It is less widely
`appreciated that plants can also be protected from a severe
`virus by prior infection with a mild strain of a closely
`related virus. This cross protection in plants was
`recognized as early as the 1920s, but its mechanism has
`been a mystery — plants do not possess an antibody-
`based immune system analogous to that found in animals.
`This was probably the first observation of a plant’s
`intrinsic defence mechanism against viruses which, 75
`years later, is just beginning to be understood. In the past
`decade, there has been considerable research into
`transgene-mediated virus resistance, co-suppression,
`virus-induced
`gene
`silencing
`(VIGS),
`antisense
`suppression and transcriptional gene silencing (TGS) in
`plants. There has also been intense research into RNA
`interference in Drosophila and nematodes, and quelling in
`fungi. These seemingly disparate endeavours have
`produced pieces of a jigsaw puzzle which, when put
`together, begin to reveal the existence and characteristics
`of a natural defence system in plants against viruses and
`transposable DNA elements. Many of the details
`and ramifications have yet to be determined, but the
`current picture is that of a wonderfully elegant system that
`can generically
`recognize
`invading viruses and
`transposable elements (TEs) and marshal the plant’s
`defences against them.
`
`Plant viruses and transposable DNA elements
`There are currently 72 different defined genera of plant
`viruses1, containing over 500 species, and there is scarcely a
`plant species — mono- or dicotyledon — that is not host to
`at least one virus. Plant viruses have a whole array of
`different particle morphologies, host ranges, vectors (for
`example, insects, nematodes, fungi, pollen, seeds or
`humans), genome organizations and gene expression
`strategies. They cause symptoms which at their least severe
`are unnoticeable, but range upwards through ringspots or
`mosaic leaf patterns, to widespread necrosis. The genomes
`of some plant viruses are encoded using single-stranded
`(ss) or double-stranded (ds) DNA; others have dsRNA
`genomes. However, over 90% of plant viruses have ssRNA
`genomes
`that are replicated by a virus-encoded
`RNA-dependent RNA polymerase (RDRP). Plants defend
`
`themselves by exploiting this requirement of most plant
`viruses to replicate using a double-stranded replicative
`intermediate.
`TEs are DNA sequences that have the capacity to move
`from place to place within a genome. They have been
`divided into two classes. Class I TEs are retroelements that
`amplify their copy number through reverse transcription of
`an RNA intermediate. They are particularly abundant in
`eukaryotes, and in plants comprise the greatest mass of TEs
`(in maize, this class of TE makes up over 70% of the nuclear
`DNA). Of the four types of retroelements in plants, the
`main class contains retrotransposons with direct long
`terminal repeats (LTRs). Class II TEs occur in all organisms,
`particularly prokaryotes; they have terminal inverted
`repeats (TIRs) ranging in size from 11 to several hundred
`base pairs. Within a class II TE family, one or more elements
`encode a transposase that has the potential to interact with
`TIRs to excise the elements and integrate them into other
`regions of the genome (for recent reviews of plant TEs, see
`refs 2, 3). Both classes of TEs can move around plant
`genomes, altering the function and structure of genes, and
`so accelerating genomic evolution. However, they are also
`parasitic mutagenic agents that have the potential to
`lacerate a genome4. To ensure survival, a plant needs to keep
`TE activity in check.
`
`Targeted RNA degradation
`Although not recognized at the time, evidence of a plant’s
`intrinsic defence mechanism to counter viruses and
`transposons came from the initially mystifying results of co-
`suppression and transgene-mediated virus resistance.
`Transformation with antisense gene constructs has been
`used in plant research since 1987 (ref. 5). From previous
`work on natural antisense in bacteria6, it was thought that
`hybridization of antisense RNA to the target messenger
`RNA interfered with its transport or translation7. So it was
`surprising to find subsequently that transformation of
`plants with transgene constructs encoding sense mRNA
`homologous to endogenous genes could also suppress the
`activities of these genes8–10. It was similarly perplexing when
`plants
`transformed with virus-derived
`transgenes,
`designed to provide protection through a protein-mediated
`mechanism11, gave protection against viruses even
`when little or no transgene protein (transprotein) was
`produced12(Fig. 1).
`Further analysis of the co-suppressed and virus-
`resistant transprotein-free plants revealed that in both cases
`
`834
`
`© 2001 Macmillan Magazines Ltd
`
`NATURE | VOL 411 | 14 JUNE 2001 | www.nature.com
`
`Alnylam Exh. 1054
`
`

`

`insight review articles
`
`consistent with the phenomenon of VIGS; here, plant viruses that
`contain sequences homologous to nuclear-expressed genes act to
`induce silencing of the targeted genes26.
`It therefore seems likely that one of the roles of the dsRNA-
`induced RNA degradation system of plants is to protect them against
`virus infection. This is a way to generically detect the replication of an
`invading ssRNA virus and destroy it, by specifically degrading both
`replicating and translatable forms of its genome.
`
`Nuclease specificity and location
`Specific fragments from mRNAs or viral genomes have been
`identified in gene-silenced or virus-resistant tissues, indicating
`that the targeted RNA degradation starts with endonucleolytic
`cleavage at one or more sites and is followed by exonucleolytic
`degradation14,27,28. Further investigation has found that sense and
`antisense ~25-nucleotide RNAs, with homology to the target RNA,
`are found consistently in plants showing co-suppression, antisense
`suppression, VIGS and virus resistance, but not in the appropriate
`control plants29–33. This is a critical finding. It supplies further
`evidence that these different forms of silencing are all acting by the
`same mechanism; from here on we refer to them generically as post-
`transcriptional gene silencing (PTGS). It also provides a strong link
`between PTGS and a phenomenon called RNA interference (RNAi),
`which is the targeted inhibition of gene activity by introduction,
`usually by injection, of dsRNA into a number of lower eukaryotes,
`including nematodes and Drosophila34,35.
`Looking at the biochemistry of the process of RNAi provides a
`good indication of what is probably happening in plants (Fig. 2).
`The processes of RNAi have been examined in Drosophila embryos,
`and embryo extracts, using radiolabelled dsRNA and target
`ssRNA36–40. Target ssRNA is not significantly degraded when sense
`or antisense RNAs are also introduced. However, the target RNA is
`degraded within minutes of adding homologous dsRNA. The degra-
`dation rapidly produces short sense and antisense ~21-nucleotide
`RNAs from both the dsRNA and the target ssRNA as a two-step
`process39,40. The dsRNA is degraded in ~21-nucleotide steps from
`both ends by an enzyme called Dicer-1 (CG64792, DCR1). The
`cleavage process, which is similar to that of Escherichia coli RNase III,
`produces ~21-nucleotide dsRNA fragments with 3፱ overhangs of
`2–3 nucleotides, and 5፱-phosphate and 3፱-hydoxyl termini40. Each
`fragment is associated with, and cleaved by, a separate Dicer-
`containing complex. The current model for the second step of the
`degradation
`is that the Dicer-containing, small
`interfering
`ribonucleoprotein (siRNP) complex alters in such a way that the
`strands of short dsRNA become unpaired and guide the complex to
`complementary target RNAs. This probably requires recruitment of
`
`Figure 1 Potato plants
`challenged with potato virus Y.
`The three plants on the right
`are non-transgenic and are
`susceptible to the virus. The
`three plants on the left contain
`an untranslatable virus-
`derived transgene yet are
`immune to the virus18.
`
`the transgenes were being highly transcribed in the nucleus, but the
`steady-state levels of their mRNAs in the cytoplasm were very low.
`This led to the proposal12,13 that the transgene mRNA was somehow
`perceived by the cell as unwanted and induced sequence-specific
`degradation, by a targeted nuclease, of itself and other homologous
`or complementary RNA sequences in the cytoplasm. Thus, in the co-
`suppressed and virus-resistant lines, not only the transgene mRNAs
`but also the mRNA from the homologous endogenous gene and the
`invading virus RNA (with homology to the transgene) were being
`degraded. The concept of transgene RNA-directed RNA degradation
`was supported by the results of an experiment in which plants, with a
`co-suppressed ȋ-glucuronidase (GUS) reporter gene, were inoculat-
`ed with a wild-type plant virus or the same virus engineered to
`contain GUS sequences in its genome. The plants were susceptible to
`the wild-type virus, but resistant to the virus containing the
`GUS-encoding sequence. The virus has an RNA genome and
`replicates exclusively in the cytoplasm, so the simple explanation is
`that the GUS sequence within the virus genome was specifically
`degraded in the cytoplasm by the same mechanism that was causing
`co-suppression of the nuclear-expressed genes14.
`This conclusion raised a number of questions. How do the
`nucleases in the cell know which RNAs to degrade and which to leave
`alone, or more specifically, how do they distinguish transgene RNA
`from endogenous gene mRNA? Why does this not happen to the
`mRNAs from all transgenes? And why would a plant want to
`specifically degrade these RNAs? A critical observation was that in
`both the co-suppression and virus-transgene transformation experi-
`ments, only a proportion of the initial transformants showed co-
`suppression or virus resistance, and these plants generally contained
`multiple, methylated copies of the transgenes.
`
`How is the degradation system triggered?
`Several theories have been advanced to explain how this sequence-
`specific degradation system might be activated. It was proposed
`initially that the high copy number of the transgenes produced
`excessively high levels of transgene mRNA and that this level induced
`the degradation system12,13. Other researchers suggested that the
`methylation of the transgenes made them produce aberrant (for
`example, prematurely terminated) RNA and that this aberrance
`induced the system14–17. A compelling proposal was that the system is
`induced and directed by dsRNA and that multiple transgenes
`favoured the likelihood of their integration as inverted repeats which,
`by transcriptional readthrough from one transgene into the other,
`would produce duplex-forming, self-complementary RNA18. This
`was supported by the demonstration that transgenes deliberately
`designed to produce self-complementary (hairpin or hp) RNA or
`dsRNA were highly efficient at inducing targeted virus resistance and
`gene silencing18–20. Furthermore, an investigation of simple co-
`suppression and antisense constructs found a perfect correlation
`between the integration of these constructs as inverted repeats and
`the induction of silencing19, and analysis of similar loci detected the
`presence of hpRNAs transcribed from them21.
`Why would the plant want to degrade dsRNA and ssRNAs of simi-
`lar sequence? Healthy plants do not contain dsRNA or extensively
`self-complementary ssRNA. In fact, for many years, plant virologists
`have used the presence of dsRNA in plant extracts to diagnose viral
`infection22. This seems to be the key. Most plant viruses have ssRNA
`genomes and replicate in the cytoplasm using their own RDRP to
`produce both sense and antisense (termed plus-strand and minus-
`strand) RNA. Evidence from research on the RNA bacteriophage
`Qȋ23 suggests that the plus and minus strands of a ssRNA virus form
`full-length dsRNA only as an artefact of extraction24. However, when
`the mammalian 2፱,5፱-oligoadenylate system (which is activated
`specifically by dsRNA) was transformed into plants, it was activated
`by infection with either of the two ssRNA plant viruses tested25.
`Therefore, replication of a ssRNA plant virus can produce sufficient
`stretches of dsRNA to be recognized as such within the plant. This is
`
`NATURE | VOL 411 | 14 JUNE 2001 | www.nature.com
`
`© 2001 Macmillan Magazines Ltd
`
`835
`
`

`

`Table 1 Genetic components of post-transcriptional silencing pathways
`Gene function Plants*
`Worms*
`Flies*
`Fungi*
`Algae*
`TE act.† PAZ‡
`RDRP
`SGS2
`EGO1
`QDE1
`SDE1§
`AGO1
`
`RDE1
`
`QDE2
`
`No
`
`Yes
`
`insight review articles
`
`a
`
`b
`
`c
`
`d
`
`e
`
`Translation
`initiation factor
`RecQ, DEAH
`or Upf1p
`helicase
`RNaseIII+
`helicase¶
`Chromatin
`remodelling
`Methyl
`transferase
`?
`?
`?
`
`SDE3
`
`MUT7||
`SMG-2
`
`QDE3
`
`MUT6|| Yes||
`
`CAF1
`
`K12H4.8 DCR1
`
`Yes
`
`DDM1
`
`MET1
`
`SGS3
`
`Yes
`
`No
`Yes
`
`RDE 4
`RDE2&3
`MUT2
`No
`No
`No
`Yes?
`Methylation
`*Plants, Arabidopsis thaliana; worms, Caenorhabditis elegans; flies, Drosophila melanogaster;
`fungi, Neurospora crassa; algae, Chlamydomonas reinhardtii.
`†TE act., activation of transposon activity in organisms mutant for this gene function.
`‡PAZ, presence of the PAZ domain identified by Cerruti et al.54.
`§SGS2 and SDE1 are different descriptions of the same gene.
`||The effect on transposon activation was assessed only for the MUT7 and MUT6 helicases.
`MUT7 also has an RNAse D motif.
`¶CAF1 and K12H4.8 have been identified by sequence homology to DCR1 but they have not
`been shown formally to be involved in PTGS.
`
`Figure 2 Proposed mechanism, based on RNAi, for dsRNA-directed ssRNA cleavage in
`PTGS. Introduced dsRNA (a) attracts Dicer-1-like proteins to its termini (b). The
`heterodimer complex at each end cleaves a 21-nucleotide dsRNA fragment (c), and the
`exposed ends of the shortened dsRNA each attract a new Dicer complex, which cleaves
`a further 21-nucleotide dsRNA fragment. This progressive exonuclease-like shortening
`continues until the dsRNA is completely cleaved. Dicers, loaded with dsRNA, acquire
`further components (blue ellipse), melt their dsRNA fragments and use one strand to
`hybridize to homologous ssRNA and cleave it in the middle of the 21-nucleotide guide-
`recognized sequence. One half of the dimer (shown as gold) directs hybridization and
`endonuclease cleavage giving it sense specificity. Thus Dicer complexes loaded from
`one end of a dsRNA will cleave sense-strand mRNA (d), whereas complexes loaded
`from the other end will cleave mRNA of the opposite sense (e).
`
`additional proteins to the complex39. Once hybridized to a target
`RNA, the complex cleaves it at a position approximately in the
`middle of the recognized 21-nucleotide sequence. The whole two-
`step process results in dsRNA being cleaved with a ~21-nucleotide
`(two helical turns36) periodicity from their termini, and the
`appropriate target RNAs being cleaved with the same periodicity but
`with a frame shift of ~10 nucleotides.
`The second step of the degradation probably takes place exclu-
`sively in the cytoplasm, as silencing does not reduce the full-length
`transcript levels in the nucleus27. However, the first step could occur
`in both the cytoplasm and the nucleus. Many mRNA degradation
`mechanisms involve the association of RNA with ribosomes, so it
`might be assumed that this would be the site of siRNP-mediated
`degradation. But several studies using protein-synthesis inhibitors
`have shown that neither ongoing translation nor association of the
`target RNAs with the ribosome are required for this degradation15,41.
`Furthermore, for each ribosome to be associated with enough siRNP
`complexes to ensure effective degradation of target RNA, the siRNPs
`would have to be expressed at very high levels. Perhaps the degrada-
`tion complexes act as gatekeepers, located at the nuclear pores and
`plasmodesmata, scanning the RNAs that pass through. This would
`allow mRNAs to be efficiently screened and destroyed, if recognized,
`as they exit the nucleus, thus leading to gene silencing. Viral RNAs
`
`would be scanned and destroyed as they attempt to spread from cell
`to cell through the plasmodesmata.
`
`What genes are involved in PTGS and RNAi?
`The similarity of induction, degradation and associated short
`dsRNAs in RNAi, quelling and PTGS indicates an underlying evolu-
`tionarily conserved mechanism. Analysis of mutants defective in these
`processes in Caenorhabditis elegans, Neurospora and Arabidopsis
`confirm this closeness, showing that there are a number of common
`essential enzymes or factors (Table 1). In all three species, mutation of
`an RDRP or a protein with homology to eIF2C, a rabbit protein
`thought to be involved in translation initiation42, blocks silenc-
`ing32,43–48. Another class of essential silencing proteins, those with
`homology to one of three types (RecQ, DEAH or Upf1p) of helicase,
`has been found in C. elegans, Neurospora, Chlamydomonas reinhardtii
`and Arabidopsis49–53. The roles of these proteins remain to be elucidat-
`ed. They are probably not the equivalents (or parts thereof) of Dicer-1
`in Drosophila; comparisons of the Dicer-1 sequence with genome
`databases identify K12H4.8 in C. elegans and CAF1 in Arabidopsis as
`homologues. These two Dicer-like proteins each have an RNA helicase
`domain, RNase III motifs and a PAZ domain39,54 (Fig. 3).
`There are two further categories of silencing-deficient mutants in
`plants and nematodes. One contains mutations of proteins that affect
`the structure and/or transcriptional status of chromatin, including
`DDM1, which remodels chromatin structure, MET1, which is a
`methyltransferase, and RDE2, RDE3 and MUT2, which seem to be
`involved with repressing the activity of TEs. The other category
`contains SGS3 in Arabidopsis and RDE4 from C. elegans, whose
`functions are a complete mystery. SGS3 has been cloned and
`sequenced, but has no recognizable motifs or matches with other
`sequences in available databases.
`These mutation studies show that PTGS, RNAi and quelling are
`not just the result of Dicer complexes waiting to degrade dsRNA and
`homologous ssRNA that invades a cell. Other associated processes
`are clearly involved, including a possible link to the translation
`apparatus, an RDRP, and interactions with chromosomal DNA.
`
`The role of methylation and chromatin remodelling in PTGS
`DNA methylation and chromatin structure have an integral role in
`TGS. In this form of silencing, the promoter and sometimes the
`
`836
`
`© 2001 Macmillan Magazines Ltd
`
`NATURE | VOL 411 | 14 JUNE 2001 | www.nature.com
`
`

`

` DEAH helicase
`
` PAZ
`
`RIII
`
`RIII
`
`dsB
`
`
`
` PAZ PAZ
`
`
`
`PiwiPiwi
`
`Figure 3 Distribution of domains on DCR1-like and AGO1-like proteins. Top panel
`illustrates the domains on DCR1-like proteins of ~2,000 amino acids (including
`DCR1, CAF1 and K12H4.8); bottom panel represents AGO1-like proteins of ~900
`amino acids (including AGO1, RDE1 and QDE2). RIII, RNase III domain; dsB, double-
`stranded RNA-binding domain(s). For more detailed information on domains, see
`http://www.sanger.ac.uk/Software/Pfam.
`
`coding region of the silenced transgenes are densely methylated55.
`Methylation, or methylation-associated chromatin remodelling, of
`promoter sequences is thought to prevent binding of factors neces-
`sary for transcription55. The coding sequences of PTGS-inducing
`transgenes are also frequently found to be methylated. PTGS can be
`established in plants with a mutant methyltransferase (metI), but
`during growth, the silencing becomes impaired, reactivating the
`silenced gene in sectors of the plant56. Furthermore, PTGS can
`fail to establish in mutant plants lacking the chromatin remodelling
`protein DDM1. These results suggest a role for DNA methylation
`and/or chromatin structure in both establishment and maintenance
`of PTGS.
`The mechanisms of PTGS and TGS may have more in common
`than was previously thought. In PTGS, the short RNAs derived from
`the transcribed region of the transgene act as guides for siRNPs to
`degrade target ssRNA. In TGS plants, hpRNAs containing promot-
`
`I/R transgenes
`
`Methyltransferase
`
` CH3
`
`Hairpin RNA
`
`RNA-directed
`methylation
`
`insight review articles
`
`er-region sequences are processed into short dsRNAs, and seem to
`direct methylation30. Similarly, virus-replicated RNAs direct
`sequence-specific DNA methylation57,58 and are associated with
`short dsRNAs58. It is possible that the steps of PTGS and TGS are, in
`fact, the same and differ only in their target sequences: hpRNA or
`dsRNA is cleaved by the plant homologue of Dicer-1 into
`~21-nucleotide dsRNAs to guide specific ssRNA degradation in the
`cytoplasm, and a similar ribonucleoprotein complex passes into the
`nucleus to direct chromatin remodelling/methylation of homolo-
`gous DNA (Fig. 4). Thus, production of hpRNA/dsRNA that
`contains promoter sequences leads to the methylation/altered state
`of the promoter DNA, causing TGS, whereas hpRNA/dsRNA that
`contains coding-region sequences leads to the degradation of
`homologous mRNA, causing PTGS. The methylation of coding-
`region DNA in PTGS and the potential degradation of promoter-
`sequence transcripts in TGS would be irrelevant by-products, as
`methylated coding regions are readily transcribed58,59 and promoter
`sequences are not usually transcribed.
`It seems unlikely that the DNA methylation mechanism associat-
`ed with PTGS and TGS is involved directly in protecting plants
`against most RNA viruses. The vast majority of these viruses have
`exclusively cytoplasmic
`lifecycles and no homologous DNA
`sequences in plant genomes. It is possible that dsRNA-directed
`methylation is involved in inhibiting the handful of known plant
`retroviruses or pararetroviruses during their DNA phases within the
`nucleus. It is even more likely that the mechanism is primarily for
`defence against TEs.
`
`Defence against transposons in plants
`DNA methylation may have evolved as an epigenetic means of
`containing the spread of TEs in host genomes4,60. De novo DNA methy-
`lation was first detected in plants during the inactivation of class II
`TEs61 and has been associated with both transcriptional inactivation
`
`Replicating
`RNA Virus
`
`dsRNA
`
`RNA-directed
`Dicer complex
`
`Cleavage
`
`Spread for
`systemic
`silencing
`
`Figure 4 A model for the initiation and operation of PTGS. The hpRNA or dsRNA produced from either an inverted-repeat transgene or a replicating virus is cleaved into ~21-
`nucleotide fragments by the Dicer-containing complex and used as guides for cleavage of homologous ssRNA (described in more detail in Fig. 2). The ~21-nucleotide dsRNA, in
`some sort of complex, is transported into the nucleus to direct DNA methylation of homologous sequences or into neighbouring cells to act as a mobile PTGS signal.
`
`NATURE | VOL 411 | 14 JUNE 2001 | www.nature.com
`
`© 2001 Macmillan Magazines Ltd
`
`837
`
`

`

`insight review articles
`
`a
`
`b
`
`c
`
`d
`
`e
`
`Figure 5 Possible ways in which transposons may generate hpRNA or dsRNA. a, LTR
`transposon integrated as an inverted repeat. b, LTR transposon integrated in the
`opposite polarity into another copy of the same transposon. c, TIR transposon
`adjacent to an endogenous promoter. d, TIR transposon (Mu). e, LTR transposon
`adjacent to an endogenous promoter. Red blocks represent either the inverted or
`direct terminal-repeat sequences. Green blocks represent the coding regions of the
`TEs. Dark blue arrows below represent the transcripts generating dsRNAs or hpRNAs,
`and the drawing below each transcript represents the structure it may form (that is,
`either a hairpin or a dsRNA structure). Blue box represents the transcription terminator
`sequence and light blue arrows represent readthrough transcription. Large green
`arrows represent endogenous promoters.
`
`and increased transitional mutation62–64. Many studies have shown the
`involvement of methylation with transposon inactivation61,65–67 and
`demethylation with transposon activation68,69. A recent demonstration
`of this comes from the study of the retrotransposon Tto1 in Arabidop-
`sis70. As this TE becomes transcriptionally silenced, it also becomes
`increasingly methylated, and demethylation of the element, in a
`hypomethylation mutant ddm1 background, reactivates its transcrip-
`tion. However, it is still not entirely clear whether the methylation
`itself inactivates the transposon or whether it is a secondary effect of
`inactivation caused by a change in chromatin structure.
`Epigenetic inactivation of TEs occurs in many, possibly most,
`cases by its insertion into or near an already heterochromatic block of
`genomic DNA and the radiation of this repressed state into the
`elements3. But TEs integrating into euchromatic areas may well be
`the target for dsRNA-induced silencing. There are a number of
`scenarios (Fig. 5) of how TEs could produce dsRNA or hpRNA to
`trigger this mechanism. The LTRs of class I TEs contain promoter
`sequences, so two TE copies integrating as an inverted repeat could
`produce hpRNA transcripts of these sequences. TEs often integrate
`within each other, potentially generating transcribable, complex
`inverted-repeat sequences. Some transposons, such as Robertson’s
`mutator (Mu), have convergently arranged genes which produce
`transcripts that, by failing to terminate or be polyadenylated at the
`appropriate sequences, have regions of complementarity with each
`other71. Insertion of a class II TE adjacent to an endogenous promoter
`directing transcription across the elements could produce RNA with
`self-complementarity from the TIRs. An adjacent endogenous
`promoter directing transcription from the reverse end of a TE could
`produce antisense RNA that might hybridize with the TE RNA to
`produce dsRNA. It is also possible that the reverse transcription of
`
`retrotransposon RNAs produces intermediates in the cytoplasm
`similar to replicating RNA viruses which, although RNA/DNA
`hybrids rather than RNA/RNA hybrids, act as triggers for dsRNA-
`induced silencing. Indeed, normal infection of plants with cauli-
`flower mosaic pararetrovirus, which will produce a similar
`RNA/DNA intermediate, triggers a PTGS-like response72. But the
`most convincing evidence that the dsRNA-induced silencing
`mechanism is suppressing TEs is that such silenced elements are
`reactivated in a number of PTGS- and RNAi-defective mutants
`(Table 1), and that some of the ~21-nucleotide dsRNAs from RNAi
`and PTGS extracts contain sequences of TEs (ref. 40, and A. J. Hamil-
`ton and D. C. Baulcombe, personal communication). The TEs are
`probably controlled by methylation of their DNA and degradation of
`their transposase mRNA.
`
`PTGS can spread systemically through a plant
`PTGS has three phases: initiation, maintenance and, remarkably,
`spread16,73. Transgenes and viruses can initiate PTGS, as can
`exogenous DNA delivered by bombardment or Agrobacterium
`infiltration16, and grafting of unsilenced scions onto silenced
`rootstock73. These last methods give localized delivery points for
`PTGS that spreads from these points into other tissues. It seems to
`spread by a non-metabolic, gene-specific diffusible signal that is
`capable of travelling both between cells through plasmodesmata, and
`long distances via the phloem16,73. For example, new tissue
`growing from a GUS-expressing scion, grafted onto a GUS-silenced
`rootstock, shows progressive silencing of its GUS transgene73. The
`signal seems to be sequence specific, to move uni-directionally
`from source to sink tissue, and can traverse at least 30 cm of wild-type
`stem grafted between the GUS-expressing scion and GUS-silenced
`rootstock73.
`To account for the specificity of the signal, it must consist (at least
`in part) of the transgene product, probably in the form of RNA73. The
`concept of cell-to-cell and long-distance spread of endogenous RNAs
`within plants remains somewhat controversial, but is not unprece-
`dented. For instance, plant viruses have genomes composed of RNA
`and, when they infect their host, their RNA spreads throughout the
`plant. Viral-encoded movement proteins facilitate the movement of
`viral RNA between cells through plasmodesmata in the form of
`either a ribonucleoprotein complex or intact virions. To fulfil
`this role, movement proteins have the capacity to move between
`cells, bind viral RNA and dilate the size exclusion limit of plasmodes-
`mata74. Simpler still, viroids — plant pathogens with small
`(~350 nucleotide) naked RNA genomes encoding no proteins — also
`infect and spread though plants, presumably associated with
`host proteins75.
`There is an emerging picture of RNA mobility in plants that
`potentially impacts on other plant processes, including transport of
`the gene-specific silencing signal. Examples of host RNAs moving
`from cell to cell include the KNOTTED1 transcription factor and its
`corresponding mRNA76, and the transcript that encodes sucrose
`transporter 1, which has been localized to the enucleate sieve
`elements, presumably having been transported there from the asso-
`ciated companion cell77. Perhaps the most convincing demonstra-
`tion of intercellular movement of endogenous plant RNA, and
`potentially signalling, is the demonstration that mRNA is found in
`the phloem of rice and cucurbits78,79. Mobility of pumpkin phloem
`RNA was demonstrated using grafting experiments. In one instance,
`a transcript encoding a transcription factor, NACP, was detected in
`the meristem of cucumber scions that had been heterografted onto a
`pumpkin rootstock78. Thus RNA molecules, derived from the
`silenced transgene, might move from cells where this gene is silenced,
`possibly with cellular protein factors fulfilling a role similar to viral
`movement proteins, to induce silencing in other cells expressing the
`same transgene. This raises three questions — is the signal the
`~21-nucleotide dsRNA, how is the signal propagated, and what is the
`natural (non-transgenic) role of the signal?
`
`838
`
`© 2001 Macmillan Magazines Ltd
`
`NATURE | VOL 411 | 14 JUNE 2001 | www.nature.com
`
`

`

`Figure 6 Tobacco plants showing potato virus Y (PVY)
`susceptibility, immunity and resistant/recovery
`symptoms. a, Non-transgenic tobacco plant challenged
`with PVY, showing a uniform chlorotic mosaic on leaves
`throughout the plant. b, Virus-challenged transgenic
`plant containing a transgene that expresses a hpRNA
`derived from PVY sequences, showing immunity to the
`virus and no symptoms. c, A whole leaf, and d, close-
`up, from a PVY-challenged transgenic plant showing
`resistance/recovery symptoms. As the plant grows the
`leaves show fewer and fewer yellow patches; only the
`yellow patches contain detectable levels of virus. The
`pattern suggests that as the virus attempts to spread
`from the phloem into the surrounding cells, a signal is
`emitted that allows more distal cells to be forewarned
`and resist and restrict the virus.
`
`a
`
`c
`
`insight review articles
`
`b
`
`d
`
`Defence is a two-step process
`Most plant species are immune to most plant viruses, possibly due to
`compatibility requirements between the RDRP or movement
`protein of a specific virus and the host factors present in the plant.
`However, it is also possible that the only virus/host combinations
`leading to an infection are ones in which the virus can overcome or
`avoid the plant’s PTGS defence mechanism. For example, potyvirus-
`es encode a protein, HC-Pro, that potently disables PTGS31,80,81 by
`directly or indirectly preventing the dsRNA cleaving activity of the
`Dicer complex. Therefore, a plant challenged with a potyvirus
`becomes a battleground in the fight between a defence and a counter-
`defence strategy. It is a race between the cell’s recognition and
`degradation of viral RNA, and the virus expressing HC-Pro to
`inactivate the degradation machinery. Perhaps the mobile signal,
`seen in the artificial situation of transgenes and grafting experiments,
`is a reflection of a plant’s fallback strategy. If the first-challenged cells
`are not quick enough to recognize and destroy the virus, but can send
`a warning message to uninfected cells of their imminent invasion,
`these cells can have