Cell, Vol. 107, 465–476, November 16, 2001, Copyright 2001 by Cell Press
`
`On the Role of RNA Amplification
`in dsRNA-Triggered Gene Silencing
`
`Titia Sijen,1,4 Jamie Fleenor,3,4 Femke Simmer,1,4
`Karen L. Thijssen,1 Susan Parrish,2,3
`Lisa Timmons,3 Ronald H.A. Plasterk,1,5
`and Andrew Fire3,5
`1 Hubrecht Laboratory
`Center for Biomedical Genetics
`Uppsalalaan 8, 3584 CT
`Utrecht
`The Netherlands
`2 Biology Graduate Program
`Johns Hopkins University
`Baltimore, Maryland 21218
`3 Department of Embryology
`Carnegie Institution of Washington
`Baltimore, Maryland 21210
`
`Summary
`
`We have investigated the role of trigger RNA amplifica-
`tion during RNA interference (RNAi) in Caenorhabditis
`elegans. Analysis of small interfering RNAs (siRNAs)
`produced during RNAi in C. elegans revealed a sub-
`stantial fraction that cannot derive directly from input
`dsRNA. Instead, a population of siRNAs (termed sec-
`ondary siRNAs) appeared to derive from the action of
`a cellular RNA-directed RNA polymerase (RdRP) on
`mRNAs that are being targeted by the RNAi mecha-
`nism. The distribution of secondary siRNAs exhibited
`a distinct polarity (5ⴕ→3ⴕ on the antisense strand), sug-
`gesting a cyclic amplification process in which RdRP
`is primed by existing siRNAs. This amplification mech-
`anism substantially augments the potency of RNAi-
`based surveillance, while ensuring that the RNAi ma-
`chinery will focus on expressed mRNAs.
`
`Introduction
`
`RNA-mediated interference (RNAi) is a conserved gene
`silencing mechanism that recognizes double-stranded
`RNA (dsRNA) as a signal to trigger the sequence-spe-
`cific degradation of homologous mRNA (see Sharp, 2001
`for a recent review). Analyses of RNAi and related pro-
`cesses in diverse systems have uncovered several sur-
`prising properties, including the double-stranded char-
`acter of the trigger RNA and a catalytic aspect of the
`interference reaction. Indeed, a few molecules of dsRNA
`are sufficient in C. elegans or Drosophila cells to trigger
`the decay of a much larger population of target mRNAs
`(Fire et al., 1998; Kennerdell and Carthew, 1998).
`Several features of the RNAi mechanism have been
`proposed to contribute to the remarkable potency of
`the reaction. Some degree of amplification is likely to
`derive from cleavage of the dsRNA trigger into short
`pieces of 21–25 nt (called siRNAs) by the RNaseIII-like
`
`4 These authors made equal contributions to this work.
`5 Correspondence: plasterk@niob.knaw.nl; fire@ciwemb.edu
`
`nuclease DICER (e.g., Zamore et al., 2000; Bernstein et
`al., 2001). For the most commonly used dsRNA triggers
`(500–1000 bp), this would result in a 20- to 40-fold in-
`crease in the molar ratio of trigger to target. A simple
`(single-use) utilization of the siRNAs would be sufficient
`to explain the molar efficiency of RNAi in extracts of
`Drosophila, but would be insufficient to account for in
`vivo potency in C. elegans. A multiround mechanism
`(use of a single siRNA for hundreds or thousands of
`rounds of target degradation) would be much more effi-
`cient.
`An additional contribution to the potency of RNA-
`triggered gene silencing has been proposed to involve
`physical amplification of an aberrant RNA population
`through an RNA-directed RNA polymerase (RdRP) activ-
`ity (Dougherty and Parks, 1995). By producing a large
`number of copies of a triggering RNA, an RdRP activity
`might dramatically increase the effectiveness of RNAi.
`The possibility of RdRP involvement in posttranscrip-
`tional gene silencing has been supported by the isola-
`tion of an endogenous RdRP activity from tomato
`(Schiebel et al., 1993a, 1993b, 1998), followed by subse-
`quent demonstrations that factors with protein se-
`quence homology to this RdRP were required for effi-
`cient silencing in fungal, nematode, and plant systems
`(Cogoni and Macino, 1999; Smardon et al., 2000; Dalmay
`et al., 2000; Mourrain et al., 2000).
`A number of apparent constraints on the roles of RdRP
`activity in RNAi are suggested by experimental observa-
`tions. Embryonic extracts from Drosophila with no mea-
`surable RdRP activity can carry out a complete RNAi
`reaction (Zamore et al., 2000; P. Zamore, personal com-
`munication). This, combined with the absence in avail-
`able Drosophila or mammalian genomic sequences of
`a clear homolog of the RdRP-like genes implicated in
`other systems, argues that an RNAi reaction can pro-
`ceed without RdRP. It should be noted, however, that
`formation of unstable (transient) copy RNAs during the
`in vitro reaction might be difficult to detect, and that
`additional enzymes (such as RNA polymerase II and
`retroviral type reverse transcriptases) are capable of
`polymerizing RNA in response to certain RNA templates
`(e.g., Diener, 1991; Filipovska and Konarska, 2000; Mo-
`dahl et al., 2000). A more limited constraint on possible
`roles for RdRP in RNAi comes from experiments in which
`the two trigger strands have been modified differentially
`prior to injection into C. elegans or Drosophila (Parrish
`et al., 2000; Yang et al., 2000). These experiments
`showed a more stringent requirement for structure and
`sequence of the antisense strand of the original trigger,
`as compared to the sense strand. These “strand-prefer-
`ence” experiments do not rule out a role for RdRP in
`the interference reaction, but do severely limit models
`in which the RdRP carries out a multiround replication
`of a double-stranded trigger (e.g., Waterhouse et al.,
`1998) to produce exponential amplification: this type of
`exponential amplification would result in loss of memory
`of the difference between the original two strands and
`would thus be incompatible with the observed effects
`of strand-specific modification.
`
`Benitec - Exhibit 1011 - page 1
`
`

`
`Cell
`466
`
`Figure 1. Could siRNA-Primed Copying of Target RNAs by an RNA-
`Directed RNA Polymerase Contribute to RNAi?
`(A) A current model of the nucleic acid alterations during RNA inter-
`ference based primarily on in vitro studies of RNAi in Drosophila
`extracts (e.g., Zamore et al., 2000; Hammond et al., 2001; Bernstein
`et al., 2001; Elbashir et al., 2001). After cleavage of the dsRNA trigger
`into short siRNA segments, the individual antisense siRNAs pair with
`complementary mRNAs, with degradation of mRNA and (eventual)
`recycling of siRNAs.
`(B) shows that at the heart of the working model is an intermediate
`with the antisense strand of an siRNA hybridized to an mRNA target.
`Since the siRNAs possess a 3⬘-terminal hydroxyl group, the resulting
`intermediate might function as a template for elongation by an RdRP
`activity.
`(C) shows a possible consequence of the reactions proposed in (A)
`and (B), with the sequential activity of RdRP and a dsRNA-specific
`nuclease (e.g., DICER) leading to a target-dependent amplification
`of the siRNA population.
`
`Of the numerous roles proposed for RdRP during gene
`silencing, we were most intrigued by the possibility (Fig-
`ure 1) that antisense siRNAs that have annealed to a
`ssRNA target might be elongated by RdRP to produce
`longer stretches of dsRNA (Sijen and Kooter, 2000). This
`model is particularly attractive in that (1) siRNAs are
`known to have a 3⬘ hydroxyl group (Elbashir et al., 2001),
`which would be poised for elongation by an RNA poly-
`merase, (2) cleavage of the RdRP-elongated regions of
`dsRNA to produce short siRNAs would result in a net
`
`amplification of the initial population of siRNAs at the
`expense of target transcripts, and (3) this mode of ampli-
`fication utilizes the two input strands of the RNA trigger
`differentially; thus, there is no inconsistency with earlier
`results which had shown more stringent chemical re-
`quirements for the antisense strand of the initial trigger
`RNA (Parrish et al., 2000; Yang et al., 2000).
`The model in Figure 1C leads to a number of testable
`predictions; in particular, we would expect to observe
`a population of secondary siRNAs after RdRP-mediated
`synthesis of duplex RNAs followed by cleavage by
`RNaseIII/DICER activity. These secondary triggers
`would be derived primarily from sequences upstream
`of the initial trigger region on the target mRNA and would
`be expected to induce a secondary RNA interference
`reaction directed to any homologous target RNA.
`In this paper, we demonstrate the production and
`biological activity of RdRP-dependent secondary trig-
`gers during RNA interference in C. elegans.
`
`Results
`
`Biochemical Evidence for Secondary siRNAs
`We first sought to demonstrate the existence of second-
`ary siRNAs through direct analysis of RNA populations.
`Although the appearance of short RNAs in the 21–25 nt
`range has universally been observed in studies of RNA-
`triggered gene silencing, the abundance of such RNAs
`varies considerably between systems.
`In particular,
`siRNAs observed during RNAi are apparently much less
`abundant in C. elegans than in plants and Drosophila
`(e.g., Hamilton and Baulcombe, 1999; Parrish et al.,
`2000; Yang et al., 2000). In order to characterize popula-
`tions of siRNA from C. elegans in detail, we used RNase
`protection assays. 32P labeled ssRNA molecules (used
`as probes) were hybridized to denatured cellular RNA,
`and the resulting material treated with ssRNA-specific
`ribonucleases to degrade any unhybridized probe. We
`used single-stranded probes from the sense strand in
`order to detect the siRNA signal while avoiding a back-
`ground due to breakdown products of the cellular mRNA
`target. To generate a large mass of C. elegans actively
`performing RNAi, we used a procedure in which animals
`are grown on bacteria engineered to express high levels
`of a specific dsRNA (Timmons and Fire, 1998; Fraser et
`al., 2000).
`Each RNase-protection experiment involves two seg-
`ments: a dsRNA trigger produced in bacteria and a
`probe RNA used to detect siRNA molecules. Figure 2
`shows results for two target genes: the muscle-specific
`gene unc-22 and the germline-specific gene pos-1. In
`each case, the strongest siRNA signals were obtained
`when the trigger and probe sequences corresponded.
`This population of siRNAs would be expected from mod-
`els in which a dsRNA-specific nuclease cleaves the orig-
`inal dsRNA trigger to produce siRNA segments. In addi-
`tion to the trigger-coincident siRNAs, we also detected
`populations of small antisense RNAs that correspond
`to regions of the target gene outside the original trigger.
`We tentatively refer to these as secondary siRNAs. The
`secondary siRNAs were generally detected at levels
`substantially below those of
`the trigger-coincident
`siRNAs, but were reproducibly observed using several
`
`Benitec - Exhibit 1011 - page 2
`
`

`
`RNA Amplification during RNAi
`467
`
`Figure 2. Biochemical Detection of Secondary siRNAs
`Analysis of small RNAs from wild-type animals grown on E. coli expressing dsRNA segments of unc-22 or pos-1. Total RNA was isolated and
`RNase protection assays were performed using various unc-22 or pos-1 specific probes (all of sense polarity).
`(A) Products of RNase protection assay (right: protected fragments of probe resolved on polyacrylamide-urea gel; left: detail of 16–30 nt
`portion of gel). Feeding on unc-22 dsRNA yielded siRNAs from the dsRNA segment comprising the food, but also produced siRNAs mapping
`upstream of this region. Lanes designated “⫹”: RNA from animals fed unc-22 dsRNA. To determine levels of probe-derived background,
`negative controls (“⫺”) were carried out by performing RNase protections with yeast tRNA as input RNA. A similar background in the siRNA
`size range was observed in RNase protection assays on RNA from animals grown on induced bacteria containing the feeding vector L4440
`with no insert (data not shown). RNase protection assays have also been carried out using RNA from IPTG-induced E. coli producing unc-22
`dsRNA; these showed some level of probe protection but no protected fragments in the siRNA size range (data not shown). Labels above
`the lanes indicate probes. “M”: 32P-labeled 25 nt RNA oligonucleotide marker.
`(B) Map of unc-22 mRNA with positions of probes and bacterially produced dsRNA.
`(C) Secondary siRNAs are also produced upon feeding with E. coli producing pos-1 dsRNA. Since pos-1 is a germline-specific gene, RNA
`was isolated from egg preparations. “⫹”: C. elegans populations fed with E. coli producing pos-1 dsRNA; “⫺”: equivalent RNA preparations
`from animals grown on E. coli containing the empty L4440 vector.
`(D) Map of pos-1 mRNA with positions of probes and bacterially produced dsRNA.
`
`different combinations of trigger and probe sequences.
`Although the detection limits of the system preclude a
`definitive measurement of siRNA levels for each trigger/
`probe combination, two points emerge rather clearly
`from the analysis. First, occurrence of a detectable sec-
`ondary antisense population was limited to cases in
`which the probe sequence was upstream (closer to the
`5⬘ end of the target mRNA) as compared with the trigger
`sequence. Second, the abundance of secondary siRNA
`molecules appeared to decrease as a function of dis-
`tance from the primary trigger.
`
`Transitive RNAi
`Secondary siRNAs might be expected to act as func-
`tional RNAi triggers, targeting any homologous mRNA
`
`sequences for degradation. To test this hypothesis, it
`is necessary to distinguish between targeting by the
`initial dsRNA trigger and by the secondary siRNAs. This
`is most conveniently carried out by means of a “transi-
`tive RNAi” assay. Essentially, such an assay entails a cell
`with two populations of target RNA: the first population
`(primary target) has a segment which matches the
`dsRNA trigger; the second population has no homology
`to the initial dsRNA trigger, but has a segment which is
`identical to the primary target.
`Figure 3 shows an example of transitive RNAi in which
`both primary and secondary target RNAs are transgene-
`derived transcripts carrying gfp. The primary target in
`this experiment encodes a nuclear-targeted GFP-LACZ
`fusion protein (NLS-GFP-LACZ), while the secondary
`
`Benitec - Exhibit 1011 - page 3
`
`

`
`Cell
`468
`
`Figure 3. Assays for Transitive RNAi Using
`Distinct gfp Transgenes
`this assay
`The transgenic line used for
`(PD4251) carries two different gfp reporter
`constructs (A). pSAK2 produces nuclear-
`localized GFP fused at the C terminus to
`additional sequences encoding E. coli ␤-galac-
`tosidase (lacZ). pSAK4 produces mitochon-
`drially localized GFP with no additional se-
`quences at the C terminus. PD4251 animals
`express both nuclear and mitochondrial GFP
`forms in all cells of the body musculature (Fire
`et al., 1998). Young adult progeny of adult
`animals injected with specific dsRNA seg-
`ments (B) were examined to determine the
`level of interference with nuclear- and mito-
`chondrial-targeted gfps.
`(C and D) Mock injected control animals
`with both GFP isoforms expressed in each
`muscle cell.
`(E and F) Progeny of animals injected with
`ds-lacZU. This injection produced a strong
`transitive RNAi effect, interfering in a majority
`of cells not only with the nuclear targeted
`gfp::lacZ transgene, but also with the mito-
`chondrial-targeted gfp. (A bright “X” shape in [F] shows vulval muscles fortuitously included in the photo; these cells are generally nonresponsive
`to parentally injected dsRNA; Fire et al., 1998)
`(G and H) Progeny of animals injected with ds-lacZL. This segment had only a modest effect on the expression of mitochondrially targeted
`gfp, so that the majority of cells continue to produce GFP in mitochondria but not nuclei. (F) and (H) are representative of the strongest
`transitive RNAi response in each population, while (E) and (G) are representative of the weakest effect. As negative controls, PD4251
`animals injected with a variety of unrelated dsRNA segments (unc-22A, unc-22B, lin-26IVS3) showed no evident decrease in either nuclear
`or mitochondrial GFP. Animals injected with gfp dsRNA show near-complete (98%) loss of both nuclear and mitochondrial GFP (Fire et al.,
`1998).
`
`target encodes a mitochondrially targeted GFP (MtGFP)
`which has no sequences from lacZ (both transgene
`mRNAs are driven by the myo-3 promoter). As a control,
`animals carrying only one of the two transgene con-
`structs show the expected effects: both GFPs are dra-
`matically reduced in progeny of animals injected with
`dsRNA corresponding to GFP, while only the NLS-GFP-
`LACZ construct is affected by dsRNAs corresponding
`to lacZ (data not shown). A line carrying both transgene
`constructs produces both nuclear LACZ-GFP and mito-
`chondrial GFP (PD4251; Figures 3C and 3D). Injection
`of dsRNA segments from lacZ into the line carrying both
`transgenes produces a transitive effect: reduction of
`both nuclear GFP-lacZ and mitochondrial GFP. Of two
`different lacZ segments tested, a trigger that was lo-
`cated just 3⬘ to the gfp::lacZ junction (ds-lacZU) was
`most potent in the transitive RNAi assay, producing re-
`duction of mitochondrial GFP to background in 60% of
`targeted cells, while a dsRNA trigger located further
`downstream (ds-lacZL) produced a more modest effect
`(reduction of GFP in 28% of cells) (Figure 3 and data
`not shown).
`A second example of transitive RNAi is presented in
`Figure 4. In this case, the primary target is an unc-22::gfp
`fusion transgene (Figure 4C), while the secondary target
`is an endogenous gene (unc-22; Brenner, 1974; Moer-
`man et al., 1988). Injection of dsRNA corresponding to
`gfp into wild-type animals (no transgene) produced no
`phenotype; injection of dsgfp RNA into animals carrying
`a transgene expressing GFP alone produced a decrease
`in GFP but no unc-22 phenotype. Injection of dsgfp
`RNA into animals expressing the unc-22::gfp transgene
`produced the twitching phenotype that is characteristic
`of loss of unc-22 expression (e.g., ds-gfpA; Figure 4C).
`
`To test whether transitive RNAi could proceed with
`endogenous genes as targets, we carried out the two
`experiments shown in Figure 5. In-frame deletion alleles
`of unc-22 and unc-52 provide a useful genetic tool: these
`alleles each produce proteins that lose a fraction of the
`coding region (658 amino acids for unc-22(st528); 150
`amino acids for unc-52(ra511)) but retain full wild-type
`function (Kiff et al., 1988; Fire et al., 1991; Rogalski et
`al., 1993; Mullen et al., 1999). As expected, dsRNAs
`corresponding to the deleted regions produced strong
`gene-specific RNAi effects in wild-type animals, but no
`effect in animals homozygous for the corresponding
`deletion alleles. The test for transitive RNAi in each case
`consists of introducing these trigger RNAs into hetero-
`zygous animals carrying both wild-type and mutant al-
`leles. In each case, we found a strong transitive RNAi
`effect: heterozygotes exhibited interference with both
`deletion and wild-type alleles. These experiments dem-
`onstrate that transitive RNAi is not limited to transgene
`targets, but can also target physiological expression of
`cellular genes.
`
`Structural Requirements for Triggering
`of Transitive RNAi
`Certain features of transitive RNAi are illuminated by
`the requirements for structure and dose of the primary
`trigger. A prediction of the model in Figure 1C is that the
`effect should exhibit a defined polarity, with interference
`depending on the order of the two segments in the
`primary target mRNA. This was the case, as shown by
`the lack of sensitivity to transitive RNA when the order
`of segments in the transgene construct was reversed
`(Figure 4E).
`
`Benitec - Exhibit 1011 - page 4
`
`

`
`RNA Amplification during RNAi
`469
`
`Figure 4. Assays for Transitive RNAi Using a Chimeric unc-22::gfp
`Transgene
`Transgenic lines used for these assays carry the C. elegans myo-3
`promoter driving the indicated combinations of the gfp coding re-
`gion (717 nt) and a segment within the unc-22 gene (unc-22z; 486
`nt). Following propagation of clonal transgenic lines for several gen-
`erations, transitive RNAi was assayed by injecting adults with a
`variety of dsRNAs. After ⵑ3.5 days, injected animals and postinjec-
`tion progeny (⬎50 animals derived from 5–20 injected parents) were
`scored for twitching in levamisole. Assays marked with an “*”
`showed twitching predominantly in the injected adults; the re-
`maining positive assays showed twitching in both injected adults
`and progeny, while negative assays showed twitching in neither
`injected adults nor progeny.
`(A and B) Segments used in this analysis. mRNA structures are
`shown; the gfp coding region is interrupted in each DNA construct
`by three 51 nt introns. The gfp-derived dsRNAs (Parrish et al., 2000)
`were each functional in primary RNAi, as assayed by reduction of
`GFP in injected adults and progeny.
`(C) A twitching phenotype was observed when the injected dsRNA
`corresponded to sequences from gfp downstream of the unc-22::gfp
`junction. Note that ds-gfpA produced the most effective twitching
`response, presumably by producing the highest molar concentration
`of siRNAs immediately downstream of the unc-22::gfp junction.
`(D and E) Transitive RNA was specific to the structure and arrange-
`ment of the initial dsRNA trigger and transgene.
`
`Interference showed a dose response to the concen-
`tration of primary trigger, with a modest interference
`response observed at doses as low as 1.5 ⫻ 106 mole-
`cules per injected parent (data not shown). Given the
`expression levels of unc-22 (Fire et al., 1991), and as-
`suming equal dispersion of trigger RNA among the cells
`of the affected progeny, this corresponds to a stoichi-
`ometry on the order of ⵑ100 molecules of trigger RNA
`for ⵑ5000 molecules of target mRNA in each muscle
`cell of the affected animals. Triggering also appeared
`to be structure-specific: although some interference
`was observed with sense or antisense RNA preparations
`alone, there was a dramatic stimulation upon mixing the
`two preparations. As with previous studies (e.g., Fire
`et al., 1998), it was not straightforward to distinguish
`whether residual activity of our ssRNA preparations was
`due to low levels of dsRNA contamination even after
`purification. In any case, these data indicate that the
`
`Figure 5. Transitive RNAi Can Operate on Native Chromosomal
`Genes
`(A) Maps of wild-type unc-22 and an in-frame deletion (st528) that
`retains wild-type function (Moerman et al., 1988; Benian et al., 1993;
`Kiff et al., 1988; black, exons; white, introns). unc-22 null mutants
`exhibit a strong twitching behavior in the absence of levamisole (we
`used unc-22(e66) as a canonical null for this analysis; Brenner, 1974).
`The strong twitching phenotype is not seen with animals that have
`a single functional dose of the wild-type or st528 allele. Following
`injection of ds-unc22X RNA, twitching without levamisole was ob-
`served in 100% of unc-22(⫹) animals, 0% of unc-22(st528)/
`unc-22(st528) animals, and 60% of unc-22(st528)/⫹ animals.
`(B) Maps of wild-type unc-52 and a deletion allele that removes
`nonessential sequences (unc-52(ra511); Mullen et al., 1999; black,
`exons; white, introns; hatched, alternatively spliced exons). The null
`phenotype for unc-52 is a zygotic-effect embryonic lethality with
`paralysis (Williams and Waterston, 1994; Rogalski et al., 1993). A
`chromosomal deficiency (mnT11; Herman et al., 1982) was used to
`definitively determine unc-52(⫹)/unc-52(o) and unc-52(ra511)/
`unc-52(o) phenotypes. Animals that have a single functional dose
`of the wild-type or ra511 allele show no lethal or visible phenotype.
`Following injection of ds-unc52A RNA, embryonic lethality with pa-
`ralysis was observed in 100% of unc-52(⫹) animals, 0% of unc-
`52(ra511)/unc-52(ra511) animals, and 100% of unc-52(ra511)/⫹ an-
`imals.
`
`initial triggering reaction is either fully dependent on, or
`greatly stimulated by, delivery of a trigger RNA with
`double-stranded character.
`Not all potential trigger RNAs were capable of produc-
`ing transitive interference. For each target RNA, we ob-
`served a graded effect as a function of distance between
`primary and secondary target sequences. The precise
`relationship between distance and effectiveness ap-
`peared to depend on the details of the experiment (com-
`pare positional dependence in Figures 3E–3H with that
`
`Benitec - Exhibit 1011 - page 5
`
`

`
`Cell
`470
`
`in Figure 4C), but in each case, the effect decreased
`with increasing distance between the segments.
`
`The Cellular RdRP Homolog rrf-1 Is Required
`in Somatic Cells for Production of Secondary
`siRNA Triggers and for Transitive RNAi
`Genetic screens for factors responsible for RNA-trig-
`gered silencing phenomena in diverse organisms have
`identified (among many other components) factors with
`substantial homology to a cellular RdRP isolated from
`viroid-infected tomato leaves (Schiebel et al., 1998; Co-
`goni and Macino, 1999; Smardon et al., 2000; Dalmay
`et al., 2000; Mourrain et al., 2000). C. elegans has four
`members of this gene family (ego-1, rrf-1, rrf-2, and rrf-3)
`(Smardon et al., 2000). Two of these genes, ego-1 and
`rrf-1, are closely linked (0.9 kb apart in tandem orienta-
`tion), while rrf-2 and rrf-3 map to distinct loci. ego-1
`is an essential gene required for fertility: adult ego-1
`homozygotes can only be derived as progeny of hetero-
`zygous mothers, thus it is not possible to carry out RNAi
`assays in the complete absence of maternal and zygotic
`ego-1 product (Smardon et al., 2000). Despite this limita-
`tion, Smardon et al. (2000) were able to demonstrate a
`requirement for ego-1 in producing an efficient RNAi
`response in the adult germline; no role for ego-1 during
`RNAi in somatic tissue has been detected.
`To extend our understanding of the RdRP gene family
`in C. elegans, we produced deletion alleles of the rrf-1,
`rrf-2, and rrf-3 genes through PCR-based screening of
`a chemical deletion library (Figure 6A; protocol from
`Jansen et al., 1997). We obtained single deletion alleles
`for each rrf- gene: rrf-1(pk1417) deletes 401 aa, including
`the majority of the residues conserved in the RdRP fam-
`ily; rrf-2(pk2040) deletes the presumed promoter region
`and the first five exons; rrf-3(pk1426) produces an out-
`of-frame truncation after the fourth exon, effectively re-
`moving most or all of the RdRP domain. These three
`deletions would be predicted to behave as genetic nulls.
`Each of the three rrf deletions was viable and fertile;
`none showed any obvious morphological or growth de-
`fects (the rrf-3(pk1426) strain produces a slightly higher
`incidence of male progeny than wild-type; the source
`of this effect has not been characterized). While this
`work was being carried out, an additional transposon
`(Tc1)-induced allele of rrf-3 was obtained (F.S. and R.P.,
`unpublished data; protocol from Zwaal et al., 1993). Al-
`though the majority of our analysis was carried out with
`the three deletion alleles, the transitive RNAi properties
`of rrf-3 (see below) were confirmed with the Tc1 allele.
`As shown in Figure 6B, the rrf-2 and rrf-3 deletion
`strains were sensitive to RNA interference in all tissues
`(soma and germline) and for all assays performed (both
`standard RNAi assays and transitive RNAi assays). For
`rrf-2(pk2040), we observed no differences from wild-
`type in any of the RNAi assays. These results indicate
`either a redundant role for RRF-2 in RNAi or (alterna-
`tively) a role in a distinct cellular process. Interestingly,
`the rrf-3 deletion and Tc1 insertion strains both showed
`reproducible increases in sensitivity to RNAi when com-
`pared to wild-type animals. This increase in sensitivity
`is evident for several different target genes and for both
`standard and transitive RNAi assays (Figure 6B and data
`not shown). While it is interesting to speculate on possi-
`ble negative roles for rrf-3 in the RNAi response (e.g.,
`
`loss of rrf-3 function might release specific RdRP cofac-
`tors for use in RNAi), the nature of the effect will require
`further experimental analysis; the major conclusion that
`we can draw at this point is that RRF-3 is nonessential
`for the RNAi responses tested.
`By contrast to the RNAi sensitivity observed in rrf-2
`and rrf-3 mutants, we observed complete resistance of
`the rrf-1 deletion allele to certain RNAi triggers. As
`shown in Figure 6B, there was a strong correlation be-
`tween site (tissue) of function for the target gene and
`the efficacy of interference: interference for genes ex-
`pressed in somatic tissue was lost in rrf-1 deletion mu-
`tants, while interference was retained for genes ex-
`pressed in the germline. Consistent with our analysis of
`rrf-1(pk1426), D. Conte and C. Mello (personal communi-
`cation) have observed loss of RNAi in soma but not
`germline tissue in an independently isolated set of rrf-1
`missense mutations.
`We used two assays to address the production of
`secondary siRNAs in the RdRP mutants. These assays
`were carried out for somatic targets, since infertility of
`ego-1 mutants (likely to affect germline RdRP; Smardon
`et al., 2000) would confound our biochemical and ge-
`netic assays. We first transformed each rrf deletion mu-
`tation with a DNA construct (myo-3::unc-22Z::gfp, as
`shown in Figure 4) that allows a functional test for transi-
`tive interference. In these assays, we observed no loss of
`transitive interference in rrf-2(pk2040) and rrf-3(pk1426),
`while rrf-1(pk1417) completely blocked the transitive in-
`terference. In parallel, we assayed directly for physical
`production of secondary trigger molecules (Figure 6C).
`By this assay, we failed to detect upstream (secondary)
`siRNAs in rrf-1(pk1417) animals. rrf-2(pk2040) and rrf-
`3(pk1426) retained the ability to produce the secondary
`triggers. Interestingly, rrf-1(pk1417) mutants retain the
`ability to produce a small population of siRNA molecules
`corresponding to the original trigger RNA. The siRNAs
`produced in rrf-1(pk1417) may represent the primary
`trigger RNAs. These results are consistent with an
`RdRP-independent cleavage of the initial dsRNA trigger,
`followed by RdRP- and target-dependent amplification
`of the trigger population.
`A variety of genes have been shown to play essential
`or contributory roles in RNAi in C. elegans. To identify
`additional genetic requirements for transitive RNAi, we
`first assayed two genes for which the most straightfor-
`ward genetic tools were available. rde-1 and rde-4 are
`the only C. elegans genes known to be essential for
`RNAi in all tissues. Since both genes are dispensable for
`organismal viability and fertility, the assays for transitive
`RNAi were straightforward. We found that both genes
`were required for the transitive RNAi assay (Figure 6B).
`We note an ambiguity that is inherent in both siRNA
`and transitive RNAi assays: since both assays depend
`on early steps in the RNAi pathway, the results with rrf-1,
`rde-1, and rde-4 mutants do not distinguish between (1)
`a specific loss of secondary siRNAs and (2) a decrease
`in secondary siRNAs as a result of inefficiency in earlier
`stages in the RNAi pathway (e.g., primary siRNA produc-
`tion). For rde-1, this ambiguity is addressed by previous
`results. Extracts of rde-1 mutant animals are compara-
`ble to wild-type extracts in cleavage of labeled dsRNA
`into short siRNA fragments (Ketting et al., 2001). This
`initial cleavage process also proceeds in vivo: after in-
`jection of a 32P-labeled dsRNA trigger into the syncytial
`
`Benitec - Exhibit 1011 - page 6
`
`

`
`RNA Amplification during RNAi
`471
`
`germline, rde-1(ne300) null mutants are comparable to
`wild-type in the production of 32P-labeled siRNAs (Par-
`rish and Fire, 2001). rde-4 mutants have also been ana-
`lyzed in the in vivo assay; rde-4 shows a decreased
`primary siRNA production, suggesting a possible defect
`in primary siRNA production (Parrish and Fire, 2001).
`For the RdRP products, the straightforward assay for
`cleavage of labeled dsRNA after germline injection (or
`extract preparation) is not available: since ego-1 is an
`essential gene, we have no source of healthy RdRP(⫺)
`animals for direct assays of siRNA production.
`A second test that has been used to address muta-
`tional effects on the role of siRNAs in the interference
`reaction involves injection of a large population of syn-
`thetic siRNAs directed at a specific target sequence.
`The siRNAs are prepared with the characteristic duplex
`structure and 2-base 3⬘ overhang (Elbashir et al., 2001).
`For C. elegans, synthetic siRNAs of 24–25 bp yield ro-
`bust interference in wild-type animals and partially by-
`pass the RNAi defect in rde-4 mutants (but not in rde-1
`mutants) (Caplen et al., 2001; Parrish and Fire, 2001).
`When tested in the rrf-1 mutant backgrounds (point or
`deletion), we observed no interference by preformed
`siRNAs, even at concentrations 10-fold above those re-
`quired for interference in a wild-type background (Figure
`6B and data not shown).
`
`An Essential Role for Secondary siRNAs and RdRP
`Activity in the RNAi Mechanism
`The insensitivity of rrf-1 mutants to phenotypic interfer-
`ence in the soma suggested that the initial siRNA:target
`interaction might be insufficient to produce a pheno-
`typically significant effect on gene expression. This
`was particularly surprising with an unc-22 target, since
`a