RNA interference: traveling in the cell and gaining functions?

40Review TRENDS in Genetics Vol.19 No.1 January 2003ssRNA annealingto antisense RNA?DNAMethylationInvertedrepeattransgeneSensetransgene?Transposableelement'aberrant'ssRNA?PrimerindependentRdRP?RNA synthesis??RepressivechromatinstructureShortor longdsRNA?PPPPdsRNA CAF? siRNA RISC-like?ActivatedRISC-like??NucleusCytoplasm?PRNA degradationExogenousdsRNAViraldsRNADicerPPPPdsRNA siRNA RISCActivatedPrimerRISCdependentRdRP?An A nPPRNA synthesisRNA degradationTRENDS in GeneticsFig. 1. Models of molecular pathways involved in double-stranded RNA (dsRNA)-mediated silencing. The basic mechanism, which is probably present in most eukaryotesundergoing dsRNA-mediated silencing, is indicated by the gray and blue boxes. The remaining steps seem to occur in at least some organisms, but their generality is currentlyunknown and their subcellular location is mainly hypothetical. The ‘triggers’ of silencing, either direct sources of dsRNA or transcription units producing singlestrandedRNAs that can be presumably converted to dsRNA, are colored red. Green RNA, endogenously transcribed single-stranded RNA; purple RNA, RNA synthesized bya putative RNA-directed RNA polymerase; blue and red RNA, double-stranded RNA introduced exogenously or resulting from viral replication, annealing of complementaryssRNAs and/or hairpin RNA. Proteins or protein complexes are indicated by yellow boxes: CAF, an Arabidopsis homolog of Dicer; Dicer, an RNase-III-like dsRNA-specificribonuclease; RdRP, an RNA-directed RNA polymerase; and RISC, RNA-induced silencing complex. Although dsRNA is depicted in single nuclear and cytoplasmic pools,depending on the source these molecules might be delivered differently to the processing Dicer enzymes. Similarly, the RISC and RISC-like complexes might have differentcomponents and associated effector proteins depending on their functions. Although a role for dsRNA in directing methylation of homologous DNA sequences has beendemonstrated in plants, the molecular machinery involved in this process and the actual nature of the ‘guide’ RNA have not been resolved. Recent evidence suggests thatthe RISC complex is equivalent to the miRNP complex (Fig. 2) in human cells [17].Argonaute family members, AGO1 and Aubergine (Aub),have been implicated in RNAi [26–28] (Table 1). In humancells, RNAi seems to occur predominantly in the cytosol[10,17], because a nucleoplasm confined transcript iscompletely resistant to degradation induced by a homologoussiRNA [29]. In addition, the only known human homolog ofDicer is located in the cytoplasm of HeLa cells [30] (Table 1).Notably, the rat homolog of eIF2C2 (GERp95), an Argonauteprotein and a component of human RISC [10,17],isaperipheral membrane protein that is associated primarilywith the Golgi complex or the endoplasmic reticulum [31].Exogenous (injected or fed) dsRNA and viral dsRNAprobably enter the cytoplasmic RNAi pathway directly(Fig. 1). In several organisms, however, RNAi can also beinduced by inverted repeat transgenes that are transcribedinto hairpin dsRNA in the nucleus [1,2,16]. ThisdsRNAprobably needs to be exported to the cytosol to targethomologous mRNAs effectively (Fig. 1). In both plants [32]and Drosophila [33], for example, inverted repeat transgenessilence more efficiently when the hairpin dsRNAcontains an intron and polyadenylation signals thatpresumably facilitate entry of the dsRNA into the mRNAexport pathway [34]. By contrast, a transgene designed toproduce non-polyadenylated, intronless hairpin dsRNA –which is presumably retained in nuclei – targets thedegradation of a homologous mRNA very poorly [1,35].Potential amplification processes in RNAi and PTGSGenetic analyses suggest that an amplification step mightbe required for efficient RNA-mediated silencing in severalhttp://tigs.trends.com

Review TRENDS in Genetics Vol.19 No.1 January 2003 41Table 1. Genetic and/or biochemical components of the RNAi/PTGS machinery discussed in the text a,bGene function A. thaliana C. elegans C. reinhardtii D. discoideum D. melanogaster H. sapiens N. crassa S. pombeRNase III þ RNA helicase CAF DCR-1 Dicer Dicer [C] DCR1Unknown (Argonaute) AGO1 RDE-1,AGO1 [C], eIF2C1, QDE2 AGO1ALG-1,ALG-2AGO2,AUB [C],PIWI [N]eIF2C2Putative RdRP SDE1/SGS2 RRF-1,RRPA QDE1 RDP1 [N]RRF-3RNA helicaseDRH-1,DRH2MUT6 [N]Gemin3[N þ C]dsRNA-binding proteinRDE-4RNase D likeMUT-7RecQ DNA helicaseQDE3UnknownGemin4[N þ C]a This list is not comprehensive and only some proteins are indicated for each of the following organisms: Arabidopsis thaliana, Caenorhabditis elegans, Chlamydomonasreinhardtii, Dictyostelium discoideum, Drosophila melanogaster, Homo sapiens, Neurospora crassa, Schizosaccharomyces pombe.b Where known, the subcellular localization of the proteins is shown in parentheses: C, cytoplasmic; N, nuclear.systems. Recent studies in C. elegans have led to a model inwhich primary siRNAs (derived from the trigger dsRNA)might prime the synthesis of additional dsRNA, using thetarget mRNA as a template, in a reaction catalyzed by aputative RNA-directed RNA polymerase (RdRP) [4,36](Fig. 1, cytoplasm). The newly synthesized dsRNA wouldthen be cleaved by Dicer to generate secondary siRNAs at asufficient concentration to achieve efficient target mRNAdegradation by RISC [4,36]. In support of this model, theinjection of short antisense RNA oligomers into C. eleganscan trigger silencing of endogenous genes, and this effect isdependent on a functional Dicer (DCR-1) [37].Putative homologs of a tomato RdRP are required forPTGS triggered by sense transgenes in Arabidopsis, forquelling in Neurospora, and for RNAi in C. elegans andDictyostelium discoideum [1–4,36,38–40]; however, biochemicalactivity remains to be demonstrated definitivelyfor any of the RdRP homologs implicated in RNAi andPTGS [5]. In addition, in both Drosophila and humans, theresults of many experiments argue against an obligatoryrole for an RdRP in dsRNA-induced RNAi [5,9,12],although an RdRP-like activity has been reported inextracts from Drosophila embryos [41].Because siRNA production and target mRNA cleavagemost probably occur in the cytosol, it is tempting tospeculate that the RdRP-dependent amplification stepdescribed in C. elegans also occurs in the cytoplasm (Fig. 1).Notably, RDE-1 and RDE-4 are required to initiate RNAiinduced by injected or transgene-produced long dsRNAs[20], but not for silencing triggered by short antisenseRNAs [37]. Thus, the RdRP-generated dsRNA might bedelivered to Dicer differently to the way in whichexogenous or transgenic dsRNA is delivered [19] (Fig. 1,cytoplasm). In Dictyostelium, RRPA, a putative RdRPrequired for RNAi, contains an N-terminal RNA helicasedomain that has not been found in other RdRPs [40].Intriguingly, the closest homolog of this helicase domain isthat of C. elegans Dicer, suggesting that domain swappingmight have occurred between Dictyostelium Dicer andRRPA [40]. If this is true, then perhaps Dicer and an RdRPinteract in a complex such that the dsRNA generated byRdRP activity is immediately accessible and rapidlyprocessed by Dicer.In C. elegans, RDE-1, RDE-4 and RRF-1 (a putativeRdRP) are required for dsRNA-induced silencing in thesoma, even when the trigger dsRNA is expressed from atransgene in the nuclei of target tissues [16,19,36]. SDE1/SGS2 (a putative RdRP) and AGO1 (the first describedmember of the Argonaute family [15]) are also requiredfor PTGS induced by sense transgenes in Arabidopsis[38,39,42]. They are, however, completely dispensable forPTGS induced by the expression of inverted repeat transgenesthat produce hairpin dsRNA [42]. This suggests thatthe putative amplification step catalyzed by ArabidopsisSDE1/SGS2 might be different to that mediated byC. elegans RRF-1 [42].In plants, there is also an SDE1/SGS2-dependentspreading of silencing from the region homologous to thetrigger dsRNA into the adjacent non-homologous 5 0 and 3 0regions of a target transgene [43]. This is not consistentwith the simple notion that primary siRNAs prime 5 0 ! 3 0dsRNA synthesis on sense mRNA. A possible explanationis that trigger dsRNAs or siRNAs interact with thetransgene DNA and induce changes in chromatin structure[38,43] (Fig. 1, nucleus, repressive chromatin structure).This would lead to the production of ‘aberrant’, butnearly full-length transgenic RNAs that could be used bySDE1/SGS2 for primer-independent amplification [38,43](Fig. 1, nucleus, ‘aberrant ssRNA?’).In several plant species, PTGS of transgenes usuallycorrelates with DNA methylation in transcribed regions[1–3,44]. Chromatin modifications associated with (orpreceding) DNA methylation might lead to the productionof truncated aberrant RNAs, as has been reported infilamentous fungi [45]. A role for DNA methylation and/orchromatin modification in PTGS of sense transgenes isalso supported by the observation that mutations inMET1/DDM2 (a DNA methyltransferase) and in DDM1(a member of the SWI2/SNF2 family of chromatinremodelingproteins) can affect both the degree andpersistence of post-transcriptional silencing in Arabidopsis[44], although indirect effects, such as saturation of thePTGS machinery by the unregulated expression ofendogenous RNAs, cannot be ruled out.The nature of the postulated aberrant RNA remainsunknown [1,2,5,38], but PTGS can be induced in soybeanhttp://tigs.trends.com

42Review TRENDS in Genetics Vol.19 No.1 January 2003and tobacco by transgenes expressing ribozyme-truncatedtranscripts that are not polyadenylated and are preferentiallyretained in the nucleus [46].InDrosophila, thepost-transcriptional silencing of alcohol dehydrogenasetransgenes – a phenomenon that is similar to PTGSinduced by sense transgenes in plants – is dependent onthe Argonaute family member Piwi [47]. Notably, Piwi islocalized predominantly in the nucleus of interphase cells[48] (Table 1). In addition, in Schizosaccharomyces pombe,RDP1 (a putative RdRP homolog) is associated withcentromeric DNA [49] (Table 1).It is therefore tempting to speculate that in Arabidopsisthe mechanism dependent on SDE1/SGS2 and AGO1might function to produce dsRNA from non-polyadenylated,prematurely terminated or misprocessed RNAs inthe nuclear compartment (Fig. 1, nucleus, ‘primer independentRdRP?’). Perhaps the RdRP itself operates as a sensorfor the accumulation of aberrant RNAs above a certainthreshold [5]. The resulting dsRNA could be either processedto siRNAs by a nuclear Dicer (see below) or exported to thecytoplasm as part of a ribonucleoprotein complex thatdelivers the dsRNA to a cytosolic Dicer. But only the lattermight be able to induce PTGS because, as discussed above,dsRNA from inverted repeat transgenes apparently needsto be exported for efficient silencing [32,33].A similar model could explain co-suppression triggeredby single-copy dispersed transgenes as well as certaintransposon and repetitive DNA-silencing phenomena[3,4,13,19,38]. But it should be noted that some aberrantRNAs could exit the nucleus and the SDE1/SGS2-dependent step could occur in the cytosol. Furthermore,this RdRP might have dual roles in both the nuclear andthe cytoplasmic compartments.The apparent differences in the requirement or thecharacteristics of an amplification step, when comparingArabidopsis, C. elegans, Drosophila and human, mightreflect true mechanistic differences among organisms.Alternatively, at least in species in which RdRPs homologsare encoded by multigene families, such as Arabidopsis,Dictyostelium, Neurospora and C. elegans [3,4,36,38–40],different family members might participate in differentsteps of the pathway. Thus, the discrepancies mightultimately reflect our current lack of understanding ofthe role of several components. In addition, as mentionedabove, the enzymatic activity of the putative RdRPsinvolved in RNAi and PTGS remains to be verified. Thisis particularly relevant because loss of function of RRF-3(a possible RdRP) in C. elegans enhances the sensitivityto RNAi in several tissues [50]. Thus, RRF-3 behaves asa negative modulator of the RNAi response, which isdifficult to reconcile with its postulated activity (but seeRef. [50]).Nuclear phenomena associated with RNAi and PTGSIn C. elegans, RNAi mediated by injected long dsRNAvirtually abolished the accumulation of a homologouscytoplasmic transcript, but it also produced a substantialreduction of nascent RNA in the nucleus [51]. RNAi is alsotriggered by dsRNA homologous to some introns, suggestingthat at least certain precursor mRNAs (premRNAs)can be targeted for degradation [52].http://tigs.trends.comIn the green alga Chlamydomonas reinhardtii, MUT6(a DEAH-box RNA helicase required for PTGS [53]) islocalized in the nucleus (K. van Dijk, B. Jeong andH. Cerutti, unpublished; Table 1). This helicase associateswith a single-stranded RNA-binding protein and with aribonuclease containing a staphylococcal nuclease-likedomain. Because MUT6 is required for the degradationof non-polyadenylated transposon transcripts [53], whichare presumably retained in nuclei, these observationssuggest the existence of a nuclear RISC-like complex(Fig. 1, nucleus). In organisms with a single, cytoplasmicDicer, RISC or a similar ribonucleoprotein (RNP) complexmight be imported into the nucleus. But evidence fromC. elegans suggests that most pre-mRNAs do not seem tobe targeted by RNAi [51,52], possibly because they areprotected by RNP complexes and/or rapidly processed tomature mRNAs and exported to the cytosol [34]. Thus, anuclear RISC-like complex might be capable of degradingonly certain transcripts, such as non-polyadenylatedtransposon RNAs or accessible pre-mRNAs.In human cells, RNAi seems to be limited mostly to thecytoplasm [10,17,29]. A nucleoplasm-sequestered transcriptwas found to be resistant to degradation induced bya homologous siRNA. But the same RNA, in the biochemicallydefined nuclear fraction, was partiallydegraded when undergoing export from the nucleus [29].A possible explanation is that transcripts become accessibleto the cytoplasmic RNAi machinery before completetransit of the mRNAs through the nuclear pores [29].Alternatively, nuclear RISC-like complexes might exist inhuman cells but only target accessible transcripts, such asthose that undergo remodeling of associated proteinsconcomitant with nuclear export [34]. As suggestedpreviously [54], a fraction of the RISC complexes mightbe located at the nuclear pores and act as gatekeepers,scanning the RNAs as they are being exported.The human RISC complex is proposed to containeIF2C1 and/or eIF2C2, Gemin3 (a putative DEAD-boxRNA helicase) and Gemin4 [10,17]. Gemin3 and Gemin4are also components of another multiprotein complex,containing the SMN (survival of motor neurons) protein,that functions in the assembly and restructuring of diverseRNP particles involved in transcription, mRNA splicingand rRNA processing [55,56]. Gemin3 and Gemin4 arelocalized both in the nucleus and in the cytosol [56],whereas the rat homolog of eIF2C2 is associated mainlywith Golgi or endoplasmic reticulum membranes [31](Table 1). The subcellular location of the whole RISCcomplex remains to be determined.The Arabidopsis genome encodes four putative Dicerhomologs [57,58]. One of these, Carpel Factory (CAF), hasbeen implicated recently in the generation of microRNAs[59], and it is predicted to be a nuclear protein [60]. Thus,at least in Arabidopsis, dsRNA might also be processed tosiRNAs in the nucleus (Fig. 1). Indeed, non-polyadenylatedtransgenic hairpin dsRNA, which is designed toinhibit the transcription of homologous promoters and ispresumably retained in the nucleus, is processed to smallRNAs [1,35]. Putative siRNAs corresponding to a viroid,which is replicated in the nucleus by the host RNApolymerase II, have also been detected in tomato [61]. In

Review TRENDS in Genetics Vol.19 No.1 January 2003 43addition, HC-Pro, a cytoplasmic viral suppressor of PTGS,inhibits the production of siRNAs associated with the posttranscriptionalsilencing of transgenes [2,62] but does notreduce the small RNAs corresponding to promoterdirecteddsRNA in tobacco [1,35].Recent work involving viral suppressors of RNAmediatedsilencing in Nicotiana benthamiana andsilencing-defective Arabidopsis mutants has demonstratedthe existence of two size classes of siRNAs [58].Long (24–26 nt) siRNAs are correlated with systemicsilencing in N. benthamiana and methylation of retroelementsequences in Arabidopsis [58]. RetrotransposonsiRNAs in both plant species correspond exclusively to thislong class. By contrast, short (21–22 nt) siRNAs arecorrelated with sequence-specific degradation of a transgenicmRNA [58]. All of these observations support theexistence of at least two populations of small RNAs inplants (in addition to microRNAs). Because subtle alterationsin Dicer structure have been postulated to alter thespacing between the catalytic centers [3], long and shortsiRNAs might be processed by different Dicer homologs,which are presumably located in the nuclear (for longsiRNAs?) or in the cytoplasmic (for short siRNAs?)compartments.In several plant species, dsRNA can direct methylationof homologous DNA sequences [1,2,35]. Methylation ofgenomic DNA occurs even when silencing is induced byRNA viruses, with sequences homologous to nuclear DNA,that replicate exclusively in the cytoplasm [1,2,54]. Thisagain suggests that there is communication between thecytoplasm and the nucleus. When the dsRNA hashomology to a promoter, it induces transcriptional silencingin association with DNA methylation [1,35]; however,it is not known whether long dsRNA or processed smallRNAs are involved in this process (Fig. 1, nucleus).Connections between the RNAi and the PTGS machineryand chromatin and/or genomic DNA modifications arealso starting to emerge in other organisms. In C. elegans,mutations in the putative RNA exonuclease MUT-7reactivate transgenic arrays that are silenced by apolycomb-dependent, presumably transcriptional, mechanism[16]. Some polycomb group homologs, which arenormally involved in chromatin repression, are alsorequired for RNAi under certain experimental conditions[63]. Likewise, QDE3, a putative RecQ DNA helicase, isnecessary for quelling in Neurospora [64]. InDrosophila,mutations in Piwi block PTGS and one aspect oftranscriptional silencing [47].Several recent reports have directly implicated theRNAi and PTGS machinery in heterochromatin formationand genome rearrangements [49,65–68] (Fig. 1, nucleus).In many eukaryotes, heterochromatin is characterizedby a high density of histone H3 methylated at lysine 9(H3-Lys9) [69]. This modification results in the binding tohistone H3 of heterochromatin protein 1 (HP1), andpresumably other factors, and formation of a transcriptionallyrepressive chromatin structure [69]. H3-Lys9 methylationalso leads to DNA methylation in Neurospora andArabidopsis [69]. But DNA methylation might be a secondarymodification that contributes, in certain organisms, tothe stability and inheritability of the silent chromatinstate.In S. pombe, deletion of DCR1 (a Dicer homolog), RDP1(a putative RdRP homolog) or AGO1 (an Argonautehomolog) results in the transcriptional derepression oftransgenes integrated in centromeric regions, the loss ofH3-Lys9 methylation and the dissociation of SWI6 (theS. pombe homolog of HP1) from this chromatin domain[49]. In addition, small RNAs complementary to bothstrands of the centromeric repeats are detected in wildtypecells [65]. These observations have led to a model inwhich dsRNA originating from the pericentric repeats bybi-directional transcription is processed to siRNAs, whichin turn induce H3-Lys9 methylation and heterochromatinformation [49]. Analysis of repetitive DNA integrated intoa euchromatic region has shown that the RNAi machineryis required for initiating but not maintaining the heterochromaticstate [66].Studies in the ciliated protozoan Tetrahymena thermophilasupport the model of heterochromatin formationpostulated for S. pombe [67,68]. Tetrahymena contains atranscriptionally inactive micronucleus (with an intactgenome), which gives rise to a transcriptionally activemacronucleus with extensive deletions corresponding toabout 15% of the micronuclear genome. This programmedDNA rearrangement is impaired in a mutant with defectsin a homolog of Piwi [67]. Wild-type but not mutant cellscontain small RNAs that hybridize preferentially tomicronucleus-specific sequences [67]. In addition, H3-Lys9methylation is required for the targeted DNA eliminationin Tetrahymena [68]. It is thought that the smallRNAs, which are processed from bi-directionally transcribedmicronucleus-specific sequences, scan the macronucleargenome, thereby directing H3-Lys9 methylationand subsequent genomic deletions [67,68]. Although smallRNAs presumably target chromatin modification througha pairing mechanism, the recognition step and thecomponents that link small RNAs to histone modificationare currently unclear. Nevertheless, these observationsextend the range of dsRNA-mediated processes andencourage the examination of transcriptional regulationby dsRNA.MicroRNAsMicroRNAs constitute a class of noncoding small RNAsthat are phylogenetically widespread in invertebrates,vertebrates and plants [55,57,59,70–72]. The smalltemporal RNAs (stRNAs) lin-4 and let-7, which belong toa subclass of miRNAs, were initially identified in C. elegansas essential regulators of the timing of development [70].Mature stRNAs are about 21 nt and can form imperfectduplexes with sequences in the 3 0 untranslated regions oftarget mRNAs, leading to translational repression [70].Although the biological role of most miRNAs is unknown,they were thought originally to function in translationalcontrol or other undefined mechanisms of genetic regulationthat are clearly distinct from the RNA cleavagedirected by siRNAs. But several new findings are blurringthe differences between miRNAs and siRNAs [17,71,72].MicroRNAs are usually recovered as single-strandedRNAs of 20–25 nt that have been processed from longerhttp://tigs.trends.com

44Review TRENDS in Genetics Vol.19 No.1 January 2003stem-loop precursors. In animals, precursor miRNAs (premiRNAs)are apparently encoded as imperfectly pairinginverted repeats of 60–70 nt [55,70]. In plants, however,some predicted pre-miRNAs are more than three timeslonger than those in animals and involve more extensive orcomplex stem-loop structures [57,59]. Most pre-miRNAsare encoded in intergenic regions and are probablytranscribed from autonomous promoters [57,59,70](Fig. 2); however, several pre-miRNAs are clustered ingenomic regions and are apparently synthesized as asingle polycistronic RNA [70,73]. In addition, a fewmiRNAs might also be processed from introns, asbyproducts of pre-mRNA splicing, or even from putativeprotein-coding RNAs [57,59,70].In mammalian cells, maturation of miRNAs involves atleast two steps [73]. Both single and clustered miRNAsseem to be expressed as longer transcripts that areprocessed into pre-miRNAs of 60–70 nt in the nucleus[73] (Fig. 2, nucleus, ‘pre-processing’). The pre-miRNAsare then exported to the cytoplasm and processed by Dicer,and possibly by other factors, into mature miRNAs ofabout 21 nt [73] (Fig. 2, cytoplasm). Recently, numeroushuman miRNAs have been found to be associated in a 15SRNP complex that includes Gemin3, Gemin4 and eIF2C2[55]. The human homolog of let-7 is a component of thisMonocistronicunitTRENDS in GeneticsPre-processingCAFPre-miRNARNP?Pre-miRNAPolycistronicunitDicerPmiRNANucleusCytoplasmPmiRNAA nPRNAdegradationIntron containedunitFunction(s)?PmiRNP?miRNP PA nPTranslationrepressionFig. 2. Model of the origin and potential functions of microRNAs. The indicatedpathways are mainly supported by experimental evidence in Arabidopsis, Caenorhabditiselegans, Drosophila and human. Green RNA, endogenously transcribedsingle-stranded RNA; red RNA, microRNAs and microRNA precursors. Protein orprotein complexes are indicated by yellow boxes: miRNP, ribonucleoprotein complexcontaining human Gemin3, Gemin4 and eIF2C2. Several steps related tomiRNA biogenesis and function are hypothetical (see text for details). Recent evidencesuggests that the miRNP complex is equivalent to the RISC complex (Fig. 1)in human cells [17].?microRNP (miRNP) and can direct the cleavage of aperfectly complementary target RNA [17]. These observationshave led to the proposal that the miRNP is thehuman RISC and can carry out both target cleavage in theRNAi pathway and translational control in the miRNApathway [17].The biogenesis of at least some plant miRNAs seems tobe different to that in animals. Although CAF, a homolog ofDicer, has been implicated in the generation of miRNAs inArabidopsis [59], pre-miRNAs are detected poorly or not atall in Arabidopsis, whereas mature miRNAs are observedreadily [57,59]. In addition, whereas in metazoans premiRNAsincrease when Dicer activity is reduced [74],mutants of CAF show significantly lower levels of miRNAswithout any concomitant accumulation of precursor [59].This suggests that because CAF is predicted to be anuclear protein [60], at least some plant miRNAs might beprocessed either co-transcriptionally or shortly aftertranscription from transient primary transcripts (Fig. 2,nucleus). Mature miRNAs might then be transported tothe cytosol as part of a RNP complex and/or they mighthave a role in the nuclear compartment (Fig. 2).Unlike animal miRNAs, whose targets are difficult toidentify by sequence complementarity, some Arabidopsisand rice miRNAs have been found to show perfect ornear-perfect complementarity to potential mRNA targets[71,72]. This suggested that they might function as siRNAs[71,72]. Indeed, an Arabidopsis miRNA (miR171/miRNA39)was found to direct specific cleavage of complementarytranscripts corresponding to a family of transcriptionalregulators [72]. Thus, despite the presence of miRNAs inboth plants and animals, current studies imply differencesin the structure of predicted precursor RNAs, in the timingand (presumed) subcellular localization of processing and,possibly, in the actual function of mature miRNAs.Biological functions of RNA-mediated silencingMutational inactivation of components of the RNAi andPTGS machinery affects at least three distinct eukaryoticprocesses: the defense response against viruses, transposonmobility and the development of multicellular organisms[1–5,54]. The RNAi and PTGS processes wereoriginally proposed to have evolved to counteract genomicparasites [1–4,54,75], but it is becoming apparent thatdsRNA-mediated mechanisms are also involved in thenormal regulation of endogenous genes. Drosophila malesuse an RNAi mechanism to degrade Stellate transcripts[75]; one miRNA has been implicated in the specificcleavage of target mRNAs corresponding to the SCARE-CROW-like transcription factors in Arabidopsis [72]; andseveral miRNAs that are complementary to protein-codingsequences have been identified in Arabidopsis and rice[57,59,71].Intriguingly, a high proportion of the predicted miRNAtargets function as developmental regulators in plants,suggesting that miRNAs might have a role in coordinatinggrowth and development [71]. Consistent with thisinterpretation, Arabidopsis CAF mutants, which aredefective in miRNA processing, show pronounced developmentalalterations [59,60].InC. elegans, developmentaldefects resulting from reduced function of Dicer and thehttp://tigs.trends.com

Review TRENDS in Genetics Vol.19 No.1 January 2003 45Argonaute-like proteins ALG-1 and ALG-2 have also beenattributed to the improper processing of miRNA precursorsand a reduction in mature stRNA expression [3,74].Itis therefore tempting to propose the existence of ancient,miRNA-mediated mechanisms that regulate endogenousgenes in eukaryotes. But most endogenous small RNAscloned from animals, and several from plants, do notmatch protein-coding or structural RNAs and theirmechanistic roles, with the exception of C. elegans stRNAs,remain unknown [55,70,74].Dicer processes precursor dsRNAs to make both siRNAsand miRNAs. In organisms encoding only one Dicer, asingle pathway might handle small RNAs. In other words,the miRNA and siRNA pathways might be interchangeablefrom biogenesis of the small RNA to interaction withits target. The final outcome, such as mRNA degradationor, for example, translational repression, would depend onthe degree of complementarity to the target RNA and,presumably, on associated effector proteins [3,17]. Thismodel is consistent with the findings that short hairpinRNAs, resembling miRNA precursors, can induce RNAi onperfectly homologous target mRNAs [3,17], and that thehuman RISC seems to be equivalent to the 15S miRNPthat is associated with many miRNAs [17,55]. Alternatively,two distinct pathways, intersecting at the Dicercatalyzed step, might be involved in the generation andfunction of at least some miRNAs and siRNAs [3]. This issupported by the requirement of different Argonauteproteins for the production of functional stRNAs or siRNAsin C. elegans [16,19,74]. In organisms where Dicer isencoded by a multigene family, such as Arabidopsis [57,58],cytoplasmic and nuclear processing pathways mightoperate.Recent findings have also implicated small RNAs inchromatin and/or DNA modifications and genomerearrangements, such as heterochromatin formationin S. pombe and DNA elimination in Tetrahymena[1,35,49,65–68]. This suggests that dsRNA-mediatedprocesses might have a role in genome organization andtranscriptional control. It is clear that despite muchprogress resulting from a combination of genetics andbiochemistry, we are only just beginning to understand themechanistic complexity of RNA-mediated silencing, itsbiological implications, and the differences and similaritiesamong different eukaryotes.AcknowledgementsI thank S. Jacobsen, T. Clemente and members of mylaboratory for helpful comments. My apologies to colleagueswhose work could not be cited due to spacelimitations. Supported by NIH grant GM62915.References1 Matzke, M. et al. (2001) RNA: guiding gene silencing. Science 293,1080–10832 Vance, V. and Vaucheret, H. (2001) RNA silencing in plants – defenseand counterdefense. Science 292, 2277–22803 Hannon, G.J. (2002) RNA interference. Nature 418, 244–2514 Plasterk, R.H.A. (2002) RNA silencing: the genome’s immune system.Science 296, 1263–12655 Zamore, P.D. (2002) Ancient pathways programmed by small RNAs.Science 296, 1265–12696 Zamore, P.D. et al. (2000) RNAi: double-stranded RNA directs theATP-dependent cleavage of mRNA at 21 to 23 nucleotide intervals. Cell101, 25–337 Bernstein, E. et al. (2001) Role for a bidentate ribonuclease in theinitiation step of RNA interference. Nature 409, 363–3668 Hammond, S.M. et al. (2000) An RNA-directed nuclease mediates posttranscriptionalgene silencing in Drosophila cells. Nature 404,293–2969 Nykänen, A. et al. (2001) ATP requirements and small interfering RNAstructure in the RNA interference pathway. Cell 107, 309–32110 Martinez, J. et al. (2002) Single-stranded antisense siRNAs guidetarget RNA cleavage in RNAi. Cell 110, 563–57411 Elbashir, S.M. et al. (2001) RNA interference is mediated by 21- and 22-nucleotide RNAs. Genes Dev. 15, 188–20012 Holen, T. et al. (2002) Positional effects of short interfering RNAstargeting the human coagulation trigger tissue factor. Nucleic AcidsRes. 30, 1757–176613 Han, Y. and Grierson, D. (2002) Relationship between small antisenseRNAs and aberrant RNAs associated with sense transgene mediatedgene silencing in tomato. Plant J. 29, 509–51914 Hammond, S.M. et al. (2001) Argonaute2, a link between genetic andbiochemical analyses of RNAi. Science 293, 1146–115015 Bohmert, K. et al. (1998) AGO1 defines a novel locus of Arabidopsiscontrolling leaf development. EMBO J. 17, 170–18016 Tabara, H. et al. (1999) The rde-1 gene, RNA interference, andtransposon silencing in C. elegans. Cell 99, 123–13217 Hutvágner, G. and Zamore, P.D. (2002) A microRNA in a multipleturnoverRNAi enzyme complex. Science 297, 2056–206018 Cerutti, L. et al. (2000) Domains in gene silencing and celldifferentiation proteins: the novel PAZ domain and redefinition ofthe Piwi domain. Trends Biochem. Sci. 25, 481–48219 Tabara, H. et al. (2002) The dsRNA binding protein RDE-4 interactswith RDE-1, DCR-1, and a DExH-box helicase to direct RNAi inC. elegans. Cell 109, 861–87120 Grishok, A. et al. (2000) Genetic requirements for inheritance of RNAiin C. elegans. Science 287, 2494–249721 Ketting, R.F. et al. (1999) Mut-7 of C. elegans, required for transposonsilencing and RNA interference, is a homolog of Werner syndromehelicase and RNaseD. Cell 99, 133–14122 Kennerdell, J.R. and Carthew, R.W. (1998) Use of dsRNA-mediatedgenetic interference to demonstrate that frizzled and frizzled 2 act inthe wingless pathway. Cell 95, 1017–102623 Svoboda, P. et al. (2000) Selective reduction of dormant maternalmRNAs in mouse oocytes by RNA interference. Development 127,4147–415624 Ratcliff, F. et al. (1997) A similarity between viral defense and genesilencing in plants. Science 276, 1558–156025 Bitko, V. and Barik, S. (2001) Phenotypic silencing of cytoplasmicgenes using sequence-specific double-stranded short interfering RNAand its application in the reverse genetics of wild type negative-strandRNA viruses. BMC Microbiol. 1, 34–4526 Williams, R.W. and Rubin, G.M. (2002) ARGONAUTE1 is required forefficient RNA interference in Drosophila embryos. Proc. Natl Acad.Sci. USA 99, 6889–689427 Kataoka, Y. et al. (2001) Developmental roles and molecularcharacterization of a Drosophila homologue of Arabidopsis Argonaute1,the founder of a novel gene superfamily. Genes Cells 6,313–32528 Harris, A.N. and Macdonald, P.M. (2001) aubergine encodes aDrosophila polar granule component required for pole cell formationand related to eIF2C. Development 128, 2823–283229 Zeng, Y. and Cullen, B.R. (2002) RNA interference in human cells isrestricted to the cytoplasm. RNA 8, 855–86030 Billy, E. et al. (2001) Specific interference with gene expression inducedby long, double-stranded RNA in mouse embryonal teratocarcinomacell lines. Proc. Natl Acad. Sci. USA 98, 14428–1443331 Cikaluk, D.E. et al. (1999) GERp95, a membrane-associated proteinthat belongs to a family of proteins involved in stem cell differentiation.Mol. Biol. Cell 10, 3357–337232 Wesley, S.V. et al. (2001) Construct design for efficient, effective andhigh-throughput gene silencing in plants. Plant J. 27, 581–59033 Kalidas, S. and Smith, D.P. (2002) Novel genomic cDNA hybridshttp://tigs.trends.com

46Review TRENDS in Genetics Vol.19 No.1 January 2003produce effective RNA interference in adult Drosophila. Neuron 33,177–18434 Reed, R. and Hurt, E. (2002) A conserved mRNA export machinerycoupled to pre-mRNA splicing. Cell 108, 523–53135 Mette, M.F. et al. (2001) Resistance of RNA-mediated TGS to HC-Pro, aviral suppressor of PTGS, suggests alternative pathways for dsRNAprocessing. Curr. Biol. 11, 1119 – 112336 Sijen, T. et al. (2001) On the role of RNA amplification in dsRNAtriggeredgene silencing. Cell 107, 465–47637 Tijsterman, M. et al. (2002) RNA helicase MUT-14-dependent genesilencing triggered in C. elegans by short antisense RNAs. Science 295,694–69738 Dalmay, T. et al. (2000) An RNA-dependent RNA polymerase gene inArabidopsis is required for posttranscriptional gene silencingmediated by a transgene but not by a virus. Cell 101, 543–55339 Mourrain, P. et al. (2000) Arabidopsis SGS2 and SGS3 genes arerequired for posttranscriptional gene silencing and natural virusresistance. Cell 101, 533–54240 Martens, H. et al. (2002) RNAi in Dictyostelium: the role of RNAdirectedRNA polymerases and double-stranded RNase. Mol. Biol. Cell13, 445–45341 Lipardi, C. et al. (2001) RNAi as random degradative PCR: siRNAprimers convert mRNA into dsRNAs that are degraded to generatenew siRNAs. Cell 107, 297–30742 Béclin, C. et al. (2002) A branched pathway for transgene-inducedRNA silencing in plants. Curr. Biol. 12, 684–68843 Vaistij, F.E. et al. (2002) Spreading of RNA targeting and DNAmethylation in RNA silencing requires transcription of the target geneand a putative RNA-dependent RNA polymerase. Plant Cell 14,857–86744 Morel, J.-B. et al. (2000) DNA methylation and chromatin structureaffect transcriptional and post-transcriptional transgene silencing inArabidopsis. Curr. Biol. 10, 1591–159445 Rountree, M.R. and Selker, E.U. (1997) DNA methylation inhibitselongation but not initiation of transcription in Neurospora crassa.Genes Dev. 11, 2383–239546 Buhr, T. et al. (2002) Ribozyme termination of RNA transcripts downregulateseed fatty acid genes in transgenic soybean. Plant J. 30,155–16347 Pal-Bhadra, M. et al. (2002) RNAi related mechanisms affect bothtranscriptional and posttranscriptional transgene silencing in Drosophila.Mol. Cell 9, 315–32748 Cox, D.N. et al. (2000) piwi encodes a nucleoplasmic factor whoseactivity modulates the number and division rate of germline stem cells.Development 127, 503–51449 Volpe, T.A. et al. (2002) Regulation of heterochromatic silencing andhistone H3 lysine-9 methylation by RNAi. Science 297, 1833–183750 Simmer, F. et al. (2002) Loss of the putative RNA-directed RNApolymerase RRF-3 makes C. elegans hypersensitive to RNAi. Curr.Biol. 12, 1317–131951 Montgomery, M.K. et al. (1998) RNA as a target of double-strandedRNA-mediated genetic interference in Caenorhabditis elegans. Proc.Natl Acad. Sci. USA 95, 15502–1550752 Bosher, J.M. et al. (1999) RNA interference can target pre-mRNA:consequences for gene expression in a Caenorhabditis elegans operon.Genetics 153, 1245–125653 Wu-Scharf, D. et al. (2000) Transgene and transposon silencing inChlamydomonas reinhardtii by a DEAH-box RNA helicase. Science290, 1159–116254 Waterhouse, P.M. et al. (2001) Gene silencing as an adaptive defenceagainst viruses. Nature 411, 834–84255 Mourelatos, Z. et al. (2002) miRNPs: a novel class of ribonucleoproteinscontaining numerous microRNAs. Genes Dev. 16, 720–72856 Charroux, B. et al. (2000) Gemin4: a novel component of the SMNcomplex that is found in both gems and nucleoli. J. Cell Biol. 148,1177–118657 Llave, C. et al. (2002) Endogenous and silencing-associated smallRNAs in plants. Plant Cell 14, 1605–161958 Hamilton, A. et al. (2002) Two classes of short interfering RNA in RNAsilencing. EMBO J. 21, 4671–467959 Reinhart, B.J. et al. (2002) MicroRNAs in plants. Genes Dev. 16,1616–162660 Jacobsen, S.E. et al. (1999) Disruption of an RNA helicase/RNAse IIIgene in Arabidopsis causes unregulated cell division in floralmeristems. Development 126, 5231–524361 Papaefthimiou, I. et al. (2001) Replicating potato spindle tuber viroidRNA is accompanied by short RNA fragments that are characteristic ofpost-transcriptional gene silencing. Nucleic Acids Res. 29, 2395–240062 Mallory, A.C. et al. (2001) HC-Pro suppression of transgene silencingeliminates the small RNAs but not transgene methylation or themobile signal. Plant Cell 13, 571–58363 Dudley, N.R. et al. (2002) Using RNA interference to identify genesrequired for RNA interference. Proc. Natl Acad. Sci. USA 99,4191–419664 Cogoni, C. and Macino, G. (1999) Posttranscriptional gene silencing inNeurospora by a RecQ DNA helicase. Science 286, 2342–234565 Reinhardt, B.J. and Bartel, D.P. (2002) Small RNAs correspond tocentromere heterochromatic repeats. Science 297, 183166 Hall, I.M. et al. (2002) Establishment and maintenance of aheterochromatin domain. Science 297, 2232–223767 Mochizuki, K. et al. (2002) Analysis of a piwi-related gene implicatessmall RNAs in genome rearrangement in Tetrahymena. Cell 110,689–69968 Taverna, S.D. et al. (2002) Methylation of histone H3 at lysine 9 targetsprogrammed DNA elimination in Tetrahymena. Cell 110, 701–71169 Lachner, M. and Jenuwein, T. (2002) The many faces of histone lysinemethylation. Curr. Opin. Cell Biol. 14, 286–29870 Pasquinelli, A.E. (2002) MicroRNAs: deviants no longer. Trends Genet.18, 171–17371 Rhoades, M.W. et al. (2002) Prediction of plant microRNA targets. Cell110, 513–52072 Llave, C. et al. (2002) Cleavage of Scarecrow-like mRNA targetsdirected by a class of Arabidopsis miRNA. Science 297, 2053–205673 Lee, Y. et al. (2002) MicroRNA maturation: stepwise processing andsubcellular localization. EMBO J. 21, 4663–467074 Grishok, A. et al. (2001) Genes and mechanisms related to RNAinterference regulate expression of the small temporal RNAs thatcontrol C. elegans developmental timing. Cell 106, 23–3475 Aravin, A.A. et al. (2001) Double-stranded RNA-mediated silencingof genomic tandem repeats and transposable elements in theD. melanogaster germline. Curr. Biol. 11, 1017–1027http://tigs.trends.com