Molecular Cell, Vol. 6, 1077–1087, November, 2000, Copyright ©2000 by Cell Press
`
`Functional Anatomy of a dsRNA Trigger:
`Differential Requirement for the Two Trigger
`Strands in RNA Interference
`
`Susan Parrish,*† Jamie Fleenor,* SiQun Xu,*
`Craig Mello,‡ and Andrew Fire*§
`* Carnegie Institution of Washington
`Baltimore, Maryland 21210
`†Biology Graduate Program
`Johns Hopkins University
`Baltimore, Maryland 21218
`‡Program in Molecular Medicine
`Department of Cell Biology
`University of Massachusetts Cancer Center
`Worcester, Massachusetts 01605
`
`Summary
`
`In RNA-mediated interference (RNAi), externally pro-
`vided mixtures of sense and antisense RNA trigger
`concerted degradation of homologous cellular RNAs.
`We show that RNAi requires duplex formation between
`the two trigger strands, that the duplex must include
`a region of identity between trigger and target RNAs,
`and that duplexes as short as 26 bp can trigger RNAi.
`Consistent with in vitro observations, a fraction of in-
`put dsRNA is converted in vivo to short segments of
`ⵑ25 nt. Interference assays with modified dsRNAs in-
`dicate precise chemical requirements for both bases
`and backbone of the RNA trigger. Strikingly, certain
`modifications are well tolerated on the sense, but not
`the antisense, strand, indicating that the two trigger
`strands have distinct roles in the interference process.
`
`Introduction
`
`Double-stranded RNA (dsRNA) induces potent cellular
`responses in diverse biological systems (Fire, 1999;
`Sharp, 1999; Williams, 1999). Models to explain dsRNA
`responses have centered on possible defense against
`deleterious RNAs, including viral transcripts and replica-
`tion intermediates (Ratcliff et al., 1997; Williams, 1999)
`and transposons (Ketting et al., 1999; Tabara et al.,
`1999). In mammalian cells, dsRNA is associated with a
`sequence-nonspecific response that includes induction
`of interferon, phosphorylation of translation initiation
`factor eIF2 (which leads to a general block in translation),
`and induction of a 2⬘-5⬘ oligoadenylate synthetase
`(which can stimulate the RNA degrading enzyme RNase
`L) (Williams, 1999).
`A distinct type of response (termed RNAi) has been
`associated with dsRNA in numerous species of inverte-
`brates, plants, and protozoa (for reviews, Fire, 1999;
`Sharp, 1999). In these cases, the response to dsRNA
`includes a dramatic and sequence-specific destabiliza-
`tion of cellular RNA transcripts that correspond to the
`RNA trigger.
`A subset of the transgene-triggered processes re-
`ferred to in plants as posttranscriptional gene silencing
`
`§ To whom correspondence should be addressed (e-mail: fire@
`ciwemb.edu).
`
`(PTGS) appear to be related or identical to dsRNA-asso-
`ciated RNA destabilization, although it is not clear
`whether dsRNA is always the trigger (de Carvalho et al.,
`1992; Baulcombe, 1996; Vaucheret et al., 1998; Wasse-
`negger and Pellissier, 1998; Jorgensen et al., 1999). As
`with the global (non-gene-specific) responses to dsRNA
`in mammalian cells, PTGS in plants has been identified
`as an antiviral mechanism (Baulcombe, 1999).
`The dsRNA-associated gene-specific responses ob-
`served in plants and invertebrates are likely to involve
`(at some stage) the pairing of antisense RNA sequences
`derived from the trigger with the endogenous sense
`RNA. (This feature is held in common by all models
`proposed to date for PTGS.) Antisense nucleic acids
`have long been known to be involved in specific cases of
`physiological regulation and to be applicable in certain
`cases as tools for selective genetic disruption (Taka-
`yama and Inouye, 1990). The key (as yet unresolved)
`questions in analysis of dsRNA-associated PTGS are
`(1) Why are both strands required in the trigger RNA?
`and (2) How can dsRNA exert an effect at concentrations
`that are substantially lower than those of the endoge-
`nous target RNA? Several models have been proposed
`to explain the second observation, including the possi-
`bilities of multiround catalytic degradation of target
`RNAs using a denatured region of the double-stranded
`trigger RNA (Montgomery et al., 1998) or production of
`numerous short antisense RNA copies of the incoming
`trigger RNA (Wassenegger and Pellissier, 1998).
`Genetic screens for components necessary for PTGS/
`RNAi have been one approach toward identifying molec-
`ular components of the mechanism (e.g., Cogoni and
`Macino, 1999; Tabara et al., 1999; Dalmay et al., 2000;
`Mourrain et al., 2000). While the cloned genes and mu-
`tant strains provide hints and useful tools for future
`studies, direct biochemical roles for the genetically iden-
`tified PTGS factors have yet to be assigned.
`A complementary approach toward understanding
`PTGS comes from attempts to understand the chemical
`character of the trigger RNA that is critical for inducing
`interference. The nematode system offers a several ad-
`vantages for this analysis, since small quantities of syn-
`thetic trigger RNAs can be assayed for interference ac-
`tivity in vivo by a straightforward injection assay. This
`provides a sensitive means to test model substrates and
`to analyze the effects of various perturbations to the
`RNA trigger. In this paper, we investigate the chemical
`and sequence requirements for RNA-triggered gene si-
`lencing in C. elegans.
`
`Results
`
`Requirements for Length and Sequence
`Composition of the RNAi Trigger
`From first principles, it was conceivable that RNA-medi-
`ated interference required a specific sequence in the
`interfering RNA or target RNA. The ability to target large
`numbers of different genes in C. elegans for interference
`suggests that any such sequence must be common in
`
`Alnylam Exh. 1034
`
`

`

`Molecular Cell
`1078
`
`A
`
`S' UTR gfp coding r egion (717bp) 3' UTR
`Ila/I
`I'm//
`/l<tl 1071 l![cl llfllll
`325
`450
`,62 62J
`169
`
`1.5051
`IXlh
`
`2 ~
`242h
`
`GFP Interference
`
`- -,
`l.505X 1.5050
`I.S_l01!
`lo4b """'iE'i,7"iiii,"l_,~, ~
`6-h 94h
`1.172~
`717b
`dsRNA Size /hp}
`None
`742
`ds-unc22A
`93
`L5304
`717
`Ll729
`181
`L505 1
`154
`L5108
`L5058
`125
`L5050
`110
`L5059
`65
`L5052
`94
`L5301
`242
`
`+ +
`+++
`+ +Ill
`++
`++1/2
`++
`+ +
`+ +
`+1/2
`
`B
`dsRNA/bp Start
`
`Sequence of sense strand
`
`Fraction Ft twitching
`2s·
`16°
`nl 26 17890 uccucugucucugcuccucucggcgg (No A)
`26/124 0/200
`zr9 27 14287 aacccaucuccaccaaauaacuacaau (No G)
`0/600
`0/80
`,rll 32 4974 aaaaagcaagccaacgaaaccaaagggaccaa (No U)
`16/174 0/80
`zr7 37 9331 aaguugaagguuaagaaugaguugggagaggaugaag (No C) 32/100 27/100
`zr 3 81* 4096 gugcucggaaaaccaucuagcccauugggaccuuuggaagugucg 96/100 94/100
`aaugucuacgaagaucgcgcagauuuggaguggaaa
`
`C
`
`-=0.1
`OD
`,:
`:g
`! 05
`./;(J .. E
`.. .. 0
`
`·ao.2
`
`o.:u
`
`05
`0.2
`0
`0 ~
`c
`
`~
`c e
`
`dsZR3
`(81-mer)
`-e--
`
`dsZRI
`
`(26mer) --
`
`Figure 1. Sequence and Size Requirements
`for the RNA Trigger
`(A) A series of nonoverlapping fragments of
`the gfp coding region. dsRNAs correspond-
`ing to each fragment were prepared enzymat-
`ically and injected at concentrations in the
`range of 20–50 ␮g/ml into transgenic animals
`carrying a myo-3::gfp transgene (PD4251, see
`Experimental Procedures). Interference lev-
`els (loss of GFP) were quantitated as de-
`scribed in Experimental Procedures. Slight
`differences in potency between the short
`RNAs could reflect some variation in concen-
`tration or purity between RNA preparations.
`(B) Five short dsRNA sequences from the
`unc-22 coding region that were prepared syn-
`thetically. Fractions shown are numbers of F1
`progeny following injection that twitched in
`0.3 mM levamisole. Synthetic sense and anti-
`sense strands were annealed and injected
`into wild-type C. elegans (concentrations 9.5
`mg/ml [zr7, zr9, zr11]; 4.8 mg/ml [zr1]; 2.3 mg/
`ml [zr3]. Additional injections of zr11 at 19
`mg/ml showed no appearance of twitching
`progeny. As controls, single-stranded (sense
`and antisense) molecules of zr1 and zr3 were
`injected separately (4.8 mg/ml), with no
`twitching phenotypes seen in ⬎100 progeny
`each. Additional injections of dsRNA for zr1
`and zr3 into the RNAi-resistant strains rde-
`1(ne219),
`rde-2(ne221),
`rde-3(ne298), and
`rde-4(ne299) likewise produced no twitching
`progeny. All duplexes were blunt-ended ex-
`cept for zr3, for which the antisense strand
`had a 4 base 5⬘ extension.
`(C) Concentration dependence for interfer-
`ence by dsRNA from zr1 and zr3.
`
`Concentration (mg/ml)
`
`11'1
`
`the genome. Sequences commonly used for interfer-
`ence are sufficiently long (400–1000 bp) that a require-
`ment for a short motif in the interfering segment might
`have been missed. As a more stringent test for the gener-
`ality of RNAi, eight nonoverlapping dsRNA segments of
`62–242 bp that span the 717 base gfp coding region (as
`well as the 5⬘ and 3⬘ nontranslated regions) were as-
`sayed for ability to interfere with GFP expression. Each
`of these segments produced an interference effect
`(albeit of slightly different magnitudes),
`indicating a
`lack of strong sequence specificity in interference (Fig-
`ure 1A).
`The unc-22 gene provides a somewhat more sensitive
`assay for gene function than gfp (see Experimental Pro-
`cedures). Chemical synthesis was used to prepare
`sense and antisense oligoribonucleotides correspond-
`ing to five nonoverlapping segments of 26, 27, 32, 37,
`and 81 nt from unc-22 (Figure 1B). The four shorter
`segments were distinguished in nucleotide composi-
`tion: these correspond to the longest segments of the
`unc-22 mRNA that lack A, G, U, and C residues, respec-
`tively. When injected as dsRNA, four of the segments
`
`(all except for the 27-mer lacking G) caused an Unc-22
`phenotype in a fraction of progeny. The short dsRNAs
`were not equal in their interference activities (Figure 1B);
`in particular, titrations of the 81 bp and 26 bp dsRNAs
`showed at least a 250-fold higher concentration require-
`ment for the 26 bp RNA (Figure 1C). The high concentra-
`tion of the short dsRNA required for interference raised
`the possibility that this sequence might be interfering
`with gene expression by an alternative mechanism. In
`particular, high doses of the 26-mer might conceivably
`obviate the need for an amplification or catalytic compo-
`nent of the interference. To address the relationship
`between interference induced by the 26-mer and inter-
`ference induced by longer dsRNAs, we assessed the
`requirements for the two strands and for the products
`of four genes that are required for full susceptibility to
`RNAi (rde-1, rde-2, rde-3, and rde-4; Tabara et al., 1999).
`Interference by the 26-mer required the products of the
`four rde genes and required both strands (data not
`shown), indicating common (if not identical) elements
`in the action of this dsRNA and that of previously charac-
`terized larger dsRNAs.
`
`

`

`Trigger Requirements for RNAi in C. elegans
`1079
`
`Injected RNA
`gfp
`IIIIC-22
`
`t
`t - - -
`t
`t
`t
`~ ,,,,,·-22
`t
`+ strand
`+++
`10011
`
`Interference
`gfp
`/IIIC-22
`+++
`+++
`+++
`+++
`
`+++
`
`+++
`
`+ ~
`•
`
`1111c-22 +++
`- strand
`10011
`
`+
`+ ==<=)+
`+ + ! +
`-
`+
`-
`-
`-
`
`+ ~ -
`
`+++
`
`Figure 2. Requirements for Double-Stranded Character in the RNAi
`Trigger
`The RNA molecules described in this figure are composed of seg-
`ments corresponding to the entire 717 base gfp coding region
`(green) and a 1034 base coding segment of unc-22 (unc22B [Fire
`et al., 1998]; red). RNAs were synthesized by standard enzymatic
`methods, annealed, and injected at approximately 40␮g/ml.
`
`Although no interference activity was observed for the
`27-mer dsRNA lacking G, it is possible that a higher
`dose or lower growth temperature would have shown
`such an effect. Alternatively, this sequence might corre-
`spond to an inaccessible region of the target transcript
`(due to secondary structure or protein coat).
`Taken together with the gfp data, the ability of four
`short double-stranded oligoribonucleotides with dra-
`matically different sequence compositions to produce
`a specific Unc-22 phenotype argues strongly against
`any sequence requirement for RNAi.
`
`Requirements for Double-Stranded Character
`in the Interfering RNA
`In earlier studies for which RNA was prepared enzymati-
`cally with T3 or T7 RNA polymerase, we had observed
`residual interference even with the most highly purified
`single-stranded RNA preparations (Fire et al., 1998). This
`left open the question of whether ssRNAs were capable
`of inducing interference or whether interference seen
`with ssRNA preparations was due to contaminating
`dsRNA. With the ability to make highly active interfering
`RNAs using an oligonucleotide synthesizer, it became
`possible to test this in a more critical manner. For the
`81 bp duplex (the most active of the dsRNA oligonucleo-
`tides), we observed no interference with sense or anti-
`sense alone, even when injected at concentrations (10
`mg/ml) that were 1000-fold above the effective concen-
`tration for the equivalent dsRNA (Figure 1C and data
`not shown).
`Observations that an effective RNAi trigger requires
`both sense and antisense strands leave several possibil-
`ities as to the contributions of the two strands to the
`triggering mechanism. Figure 2A shows experiments de-
`
`signed to test whether interference by ssRNA could be
`stimulated by covalent linkage to an unrelated segment
`of dsRNA. Each molecule has an ssRNA region corre-
`sponding to one target gene linked to a dsRNA region
`corresponding to a second target. When injected into C.
`elegans, these molecules produce interference effects
`corresponding only to the double-stranded portion of
`the trigger. These experiments indicate that triggering
`of RNAi requires both strands in the region of identity
`to the target.
`A second means to address the nature of the two-
`strand requirement comes from experiments in which
`both strands of RNA are provided to the animal under
`conditions that either favor or prevent formation of du-
`plex. Although the two strands of unc-22 RNA can pro-
`duce an interference effect when injected separately
`into the same animal, a stimulation of activity is seen if
`the strands are first annealed before injection (Fire et
`al., 1998). To further investigate the requirements for
`formation of dsRNA structure, we produced a pair of
`stem-loop molecules in which the two strands of unc-
`22 were both present but would be prevented from rapid
`hybridization by topological constraints. No interference
`with unc-22 function was observed when these two
`stem–loop molecules were coinjected (Figure 2B). These
`results indicate a requirement for sense/antisense du-
`plex formation in the RNAi trigger.
`
`Requirements for Homology between Trigger
`and Target RNAs
`To assess homology requirements for RNAi, we used a
`series of altered gfp coding regions with different de-
`grees of identity to the transgene target (Figure 3A;
`Crameri et al., 1996; Cormack et al., 1997; Fukumura et
`al., 1998; Cohen and Fox, unpublished data). We found
`effective interference with dsRNAs that were 96% identi-
`cal to the target sequence (193 nt maximum uninter-
`rupted identity), less effective interference with a trigger
`that was 88% identical (41 nt maximum uninterrupted
`identity), and no interference with dsRNA triggers that
`were 78% or 72% identical to the target (23 and 14 nt
`maximum uninterrupted identity; Figure 3B). By compar-
`ison with dilutions of the wild-type gfp dsRNA, we found
`that the 78% identity and 72% identity RNAs reduced
`the effectiveness of interference by at least 100-fold
`(data not shown).
`
`Requirements for Quality of the Interfering
`RNA Duplex
`The availability of altered gfp isoforms allowed us to
`test the triggering capacity of heteroduplexes in which
`mismatches between the two interfering strands were
`present. As shown in Figure 3C, such heteroduplexes
`can be triggers for RNAi, but with reduced effectiveness
`when compared to perfectly matched duplexes. No sig-
`nificant difference was observed as a function of which
`strand was more closely related to the RNAi target.
`
`Saturation of the RNAi Machinery in C. elegans
`To measure the saturability of the RNAi machinery in C.
`elegans, we carried out experiments in which a defined
`concentration of a single dsRNA species (an 81 bp seg-
`ment of unc-22) was mixed with increasing concentra-
`
`

`

`Molecular Cell
`1080
`
`A
`- - - - - •
`W ild type
`- - - - -(cid:127) ,a-••--•-•••--•-• Crameri
`u11•• (cid:127)
`t 11 ••• ••n••••••• • •••-• • • (cid:127)
`11 • • ,1111 1 C ormack
`111 111 111 (cid:127)
`1111(cid:127) 111111(cid:127) • • •11 11111111111 1 1(cid:127)
`11 1Cohen/Fox
`(cid:127) 1 1 1
`11
`11111(cid:127) 1111 1 11 1111111111 11
`11 u 111 1111 1 Seed
`1 1111 1 1
`- 20b
`
`111
`
`% Id e ntity
`with target
`99%
`96%
`88%
`78 %
`72%
`
`G F P
`Longest Identity
`with target
`interference
`52 I
`+ + +
`19 3
`+ + +
`+ +
`41
`23
`14
`
`B
`dsRNA
`Trigger
`Wild Type (99%)
`Crameri (96%)
`Cormack (88%)
`Coh en/Fox (78%)
`Seed (72%)
`C
`
`Heteroduplex RNAi triggers
`Incoming Sense
`I ll]jii jjjj\j jjjjlll@Pfi!iili!iHI@ Piil1ncoming Anlisense
`- - - - - - • f a rget mRNA
`
`Sense Strand
`of Trigger
`Wild Type (99 %)
`Wild Type (99%)
`Wild Type (99%)
`Wild Type (99 %)
`Wild Type (99 %)
`Cramcri (96%)
`Cormack (88 %)
`Cohcn/Fo< (78 %)
`Seed ( 72 '7<)
`None
`
`, nt iscnsc Strand
`of Trigger
`Crameri (96 %)
`Cormack (88 %)
`Cohen/Fox (78 %)
`Seed (72 %)
`None
`Wild Type (99%)
`Wild Type (99%)
`Wild T)•pc (99%)
`Wild Type (99%)
`Wild Type (99%)
`
`GFP
`interference
`+++
`+112
`+I•
`+/-
`+/-
`+++
`+1/2
`+/(cid:173)
`+/(cid:173)
`+/-
`
`Figure 3. Requirements for Homology between Trigger and Target
`RNAs
`Wild-type gfp and four multiply mutant versions were inserted into
`standard vectors for synthesis of RNA using T3 and T7 RNA polymer-
`ases. “Crameri” (GFP-uv; Crameri et al., 1996) is a result of combina-
`torial mutagenesis, “Cormack” (Cormack et al., 1997), “Cohen/Fox”
`(Cohen and Fox, unpublished data), and “Seed” (Fukumura et al.,
`1998) are codon-optimized synthetic forms of the gene that were
`originally produced for use in yeast nuclei, yeast mitochondria, and
`human cells, respectively.
`(A) Regions of identity between the five versions of GFP and the
`target gene over the 717 bp GFP coding sequence. (Note that the
`gfp variant used in the transgene target differs at five positions
`from wild-type gfp.) Darkened bars represent regions of identity of
`ⱖ5 bp.
`(B and C) Interference activities of homoduplex and heteroduplex
`dsRNAs. Single-stranded RNAs were prepared from each gene and
`annealed as described in Experimental Procedures and injected at
`concentrations of approximately 40␮g/ml into GFP reporter strain
`PD4251. GFP levels in progeny animals were scored by counting
`GFP-positive cells in L4 larvae and young adults (see Experimental
`Procedures). Although the wild-type gfp sense and antisense RNAs
`in these experiments were purified through two rounds of electro-
`phoresis, we were unable to completely remove low level interfer-
`ence activity of each ssRNA preparation alone (“⫹/⫺” signals, pre-
`sumably due to trace levels of dsRNA contamination). For this
`reason, a low level of interference by Cohen/Fox::WildType or
`Seed::WildType heteroduplexes would have been missed.
`
`tions of an unrelated competitor dsRNA. The competitor
`dsRNA (codon-humanized gfp) was unrelated to any
`gene in C. elegans. We found in these experiments that
`the RNAi effect was indeed saturable (Figure 4A). Satura-
`tion of the response required double-stranded character
`in that the individual single strands of the competitor
`RNAs had less than 2% of the competitor activity seen
`with dsRNAs (data not shown). Competition by the long
`gfp dsRNA was surprisingly effective, with complete
`competition observed even with a 5-fold lower mass of
`
`A
`
`B
`
`75
`
`100
`'Q'
`0
`.!!l
`E
`"' >
`.2
`.E
`o.c
`C
`:E
`5 ~
`!
`~
`8
`" 25
`.l:! .. .,
`l:!
`!
`
`1.2 mg/ml dsZR3
`-
`- -0.12 mg/ml dsZR3
`
`0 =
`
`=
`"'
`"'
`~
`~
`Q
`Unrelated Competitor dsRNA (mg/ml)
`
`29bp
`6Sbp
`Jbp
`::::::==== ::::: LS0S9 (97bp)
`
`778bp
`6Sbp
`3bp
`::::::=:::::::::::::::==:::::::::::::::= u953 (846bp)
`
`RNA Concentration
`GFP Interference
`+ +
`L5059 2.3 µM ( ISO µg/ml)
`+ +
`0.47 µM (30 µg/m l)
`0.09 µM (6 µg/ml)
`+1/2
`0.02 µM (l.2 µg/ml)
`L4953 0.89 µM (500 µ g/ml)
`
`Figure 4. cis and trans Saturability of the RNAi Machinery in C.
`elegans
`(A) Response following injection of wild-type animals with mixtures
`of a synthetic dsRNA oligonucleotide (zr3, 81 bp) with increasing
`concentrations of an unrelated dsRNA (enzymatically prepared
`dsRNA corresponding to the humanized “Seed” version of gfp (724
`bp). Percentage of twitching animals (in 0.3 mM levamisole) amongst
`F1 progeny is shown.
`(B) Two dsRNAs that each have 65 bp of identity to a gfp reporter
`transgene (PD4251, see Experimental Procedures). The shorter RNA
`(L5059, see Figure 1A) has only short segments on each end (derived
`from the vector multiple cloning site), while the L4953 RNA is linked
`in cis to an extended region of dsRNA corresponding to unc-22.
`The progeny of animals injected with L4953 RNA show a strong
`twitching phenotype but no evident reduction in the number of cells
`with GFP fluorescence. Note that a longer region of gfp (717 base)
`dsRNA in cis to unc-22 dsRNA produced effective gfp interference
`(line 1 of Figure 2).
`
`the competitor (24 ␮g/ml of competitor dsRNA versus
`120 ␮g/ml of interfering RNA). This could reflect a gen-
`eral preference for longer dsRNA in triggering RNAi,
`greater stability of the longer dsRNA, or specific proper-
`ties of the individual sequences used in the competition
`assay.
`It was conceivable that the ability to saturate the RNAi
`response represented a limitation in the number of
`dsRNA molecules that could be recognized. This could
`be the case, for instance, if the limiting factor were re-
`sponsible for the initial recruitment of individual dsRNA
`molecules to a multicomponent complex. To address
`the nature of the saturation, we produced molecules in
`which competitor RNA was linked in cis to the interfering
`RNA. These experiments demonstrated that triggering
`activity of a dsRNA could be effectively competed by
`excess RNA provided in cis (Figure 4B).
`
`(cid:127)
`

`

`Trigger Requirements for RNAi in C. elegans
`1081
`
`A
`
`' o......._ /o-
`_.,,r,
`. o °'-c'Qn2
`
`a-th10 S
`
`BASE
`
`2'-fluoro F
`2'-amino , f{2
`2'-d eoxy H
`
`0
`•
`
`0
`
`B
`Sense Strand
`or Trieger
`Unmodified RNA
`cytidine->cx-thiocy tidine
`guanosine->et-thioguanosine
`adenoslne->a-thioadenosinc
`uracil->cx-thiouracil
`uracil->2'-nuo rouracil
`Unmodified RNA
`uracll->2'-ami nouracil
`Unmodified RNA
`eytidine- >2'-aminocytidine
`Unmodified RNA
`uracil->2'-deoxythymldine
`Unmodified RNA
`cytidl ne-> 2 • -deoxyeytid in e
`Unmodified RNA
`Unmodified RNA
`DNA
`
`Antisense Strand
`or Trieger
`Unmodified RNA
`cytidine->cx.-thiocytidine
`guanos inc->CX•thioguanos inc
`adenosine->n -thioadenos ine
`uracil- >a-thiouracil
`Unmodified RNA
`uracil->2'-fluorour acil
`Unmodified RNA
`uracil->2'-aminouracil
`Unmodified RNA
`cytidine->2'-aminocytidine
`Unmodified RNA
`uracil->2'-d eoxythymldlne
`Unmodified RNA
`cytidine->2'-deoxycytidine
`DNA
`Unmodified RNA
`
`11nc-22
`interCerence
`+++
`+++
`+++
`+++
`++
`+++
`+++
`++
`
`++
`
`++
`+
`++
`+
`
`Figure 5. Effects of Backbone Modifications on Activity of the RNAi
`Trigger
`(A) A schematic RNA backbone with a description of the modifica-
`tions used in this work. Note that the Uracil→deoxyThymidine sub-
`stitution also entails a change in the base moiety (see Figure 6A).
`(B) Activities of backbone modified RNAs. RNAs were prepared as
`described in Experimental Procedures, annealed, and injected into
`wild-type C. elegans. The degree of interference was assessed by
`examination of phenotypes of progeny animals with and without
`levamisole treatment. Although assays were in general qualitative,
`we have included an indicator of interference strength based on
`severity of phenotypes for the majority class of progeny animals
`scored. Unmodified RNA and modified RNAs described as “⫹⫹⫹”
`produced a strong twitching phenotype in the absence of levamisole
`(Fire et al., 1998). Modified RNAs described as “⫹⫹” produce a weak
`twitching phenotype that is evident without levamisole treatment as
`a twitch during movement; these animals twitch strongly following
`levamisole treatment. Interference described as “⫹” was observed
`only in levamisole and only in a fraction of progeny. Interference
`described as “⫹/⫺” was at a level that was indistinguishable from
`the preparations of unmodified sense or antisense RNA used to
`make the modified duplex. Titrations of unmodified RNA (Fire et al.,
`1998) indicate that levels of twitching described as ⫹⫹ correspond
`to activities in a range of 3%–10% of unmodified RNA, while those
`described as “⫹” correspond to 1%–3% of unmodified RNA.
`2⬘-fluorouracil, 2⬘-aminouracil, 2⬘-deoxythymidine, and 2⬘-deoxy-
`cytidine were incorporated into individual strands of the 742 nt unc-
`22A segment using T3 and T7 RNA polymerases (Experimental Pro-
`cedures). Incorporation of ␣-thio nucleotides was carried out with
`an RNA hairpin carrying unc-22 sequences in the stem. This allows
`rapid formation in the polymerase reaction mix of duplex structure,
`which appeared to stabilize the resulting modified RNA. RNA con-
`centrations in injection mixes were in the range of 30–60 ␮g/ml.
`RNA:DNA hybrids were constructed with synthetic RNA and DNA
`corresponding to the 81 nt zr3 segment (Figure 2) and were injected
`at 5 mg/ml.
`
`Chemical Requirements in the RNA Trigger:
`Phosphate–Sugar Backbone
`We tested several modifications in the phosphate–sugar
`backbone for their effects on the ability of dsRNA to
`trigger interference (Figure 5). Modification of phosphate
`residues to thiophosphate could be effectively carried
`
`out by incorporating thiophosphate nucleotide analogs
`with T7 and T3 RNA polymerase. Although the ␣-thio-
`phosphate modifications caused some chemical insta-
`bility in the RNA, we were able to demonstrate interfer-
`ence activity following incorporation of any single
`modified residue. Modifications of A, C, or G residues
`were compatible with full
`interference activity, while
`modified U caused some decrease in interference activ-
`ity (Figure 5B). Interestingly, Zamore et al. (2000) have
`noted a preference for U residues in RNA-associated
`cleavage in vitro. RNAs with two modified bases also
`had substantial decreases in effectiveness as RNAi trig-
`gers (data not shown); modification of more than two
`residues greatly destabilized the RNAs in vitro and we
`were not able to assay interference activities.
`A second position at which modifications were tested
`was the 2⬘ position of the nucleotide sugar. Modification
`of cytidine to deoxycytidine (or uracil to thymidine) on
`either the sense or the antisense strand of the trigger
`was sufficient to produce a substantial decrease in inter-
`ference activity (Figure 5B). In the case of cytidine to
`deoxycytidine substitution, this effect must be a conse-
`quence of a change at the 2⬘ position, while the effects
`of uracil to thymidine substitution could reflect effects
`of the additional methyl group on the thymidine base.
`Modification of uracil with 2⬘-fluorouracil was com-
`patible with RNAi activity, while modification with
`2⬘-aminouracil or 2⬘-aminocytidine produced a de-
`crease in activity comparable to that seen with the de-
`oxynucleotide modification.
`A second means to assess requirements at the 2⬘
`position involves the question of whether RNA:DNA hy-
`brids can trigger RNAi. Such hybrids were prepared
`synthetically and enzymatically and found to lack inter-
`ference activity (Figure 5B and data not shown).
`Interestingly we observed a preferential effect on in-
`terference activity for several modifications (uracil→
`2⬘-aminouracil, cytidine→2⬘-aminocytidine, uracil→thy-
`midine, and cytidine→2⬘-deoxycytidine) depending on
`whether the sense or antisense strand was modified. In
`each case, trigger activity was more sensitive to modifi-
`cation of the antisense strand than of the sense strand.
`
`Chemical Requirements for the RNA
`Trigger: RNA Bases
`We tested a small number of RNA base modifications for
`their effects on the efficacy of RNA interference reaction.
`The modifications were chosen based on the commer-
`cial availability of nucleoside triphosphates that could
`be incorporated using T3 or T7 RNA polymerases. Five
`such base analogs (Figure 6A) were tested; the uracil
`analogs 4-thiouracil, 5-bromouracil, 5-iodouracil, and
`5-(3-aminoallyl)-uracil could be readily incorporated in
`place of uracil, while inosine was incorporated in place
`of guanosine.
`As with the backbone modifications, we were particu-
`larly interested to learn whether there were distinct base
`requirements for the two strands of the RNAi trigger.
`As shown in Figure 6, 4-thio-uracil and 5-bromo-uracil
`(which were compatible with interference) and inosine
`(which was compatible but produced a substantial de-
`crease in interference activity) showed no detectable
`difference in effect between the two strands. By con-
`
`

`

`Molecular Cell
`1082
`
`A
`
`0
`
`5-Metbyt CH.r
`S.Bromo n,..
`5-lodo I-
`ll
`
`5-(3-AmlnoAUyt)
`Nl(zCH2CH2Cllr
`
`H
`
`B Sense Strand
`
`of Trii:i:er
`Unmodified RNA
`uracil->4-thiouracil
`Unmodified RNA
`uracil->5-bromouracil
`Unmodified RNA
`uracil->5-iodouracil
`Unmodified RNA
`uracil->5-(3-aminoallyl)uracil
`Unmodified RNA
`guanosi ne->inosine
`Unmodified RNA
`
`Antisense Strand
`unc-22
`interference
`of Trii:i:er
`Unmodified RNA
`+++
`+++
`Unmodified RNA
`+ + +
`uracil->4-thiouracil
`+++
`Unmodified RNA
`+ + +
`u racil->5-bromouracil
`+ +
`Unmodified RNA
`+
`uracil->5-iodour acil
`+ +
`Unmodified RNA
`uracil->5-(3-aminoallyl)uracil +/-
`Unmodified RNA
`+
`+
`guanosine->inosine
`
`gfpG Trigger
`
`lacZL Trigger
`
`unc22A Trigger
`
`C
`
`Figure 6. Effects of Base Modification on Ac-
`tivity of the RNAi Trigger
`(A) Formulas for uracil and guanosine with a
`description of the modifications used in this
`work.
`(B) Activities of backbone modified RNAs.
`RNAs were prepared as described in Experi-
`mental Procedures, annealed, and injected
`into wild-type C. elegans. The degree of inter-
`ference was assessed as in Figure 5.
`(C) Activities of 5-(3-aminoallyl)-uracil sub-
`stituted RNAs. The triggering segments
`unc22A, gfpG, lacZL, and unc54A were as
`described in Fire et al. (1998). RNA concentra-
`tions were 30 ␮g/ml for lacZL and gfpG, and
`40␮g/ml for unc-22A. gfpG and lacZL injec-
`tions were carried out in strain PD4251 (Fire
`et al., 1998). unc-22A and unc-54A injections
`were carried out in wild type (N2) animals. For
`unc-22A injections, twitching fractions shown
`are in the absence of levamisole (L4 larvae
`and young adults). Some animals with weak
`twitching in the presence of levamisole were
`observed in antisense-modified unc-22A in-
`jections (57/200 animals); this was only mar-
`ginally above the background level of weak
`levamisole-induced twitching following injec-
`tion of the preparation of sense RNA alone
`(30/200 animals); this limited signal could rep-
`resent residual contamination of dsRNA in the
`individual ssRNA preparations. For unc-54A
`injections, movement was scored in L1 larvae
`and again in young adults. Titrations of 5-(3-
`aminoallyl)uracil sense-modified RNAs for
`unc-54A indicate that this RNA is at least 15-
`fold more active than the equivalent anti-
`sense modified unc-54A trigger. Similar
`experiments with lacZL, gfpG, and unc-22A
`indicate in each case at least a 10-fold differ-
`ential strand effect for 5-(3-aminoallyl)uracil
`substitution.
`
`unc54A Trigger
`progeny as adults
`progeny as lanae
`RNA Trigger
`motile
`motile
`-
`no dsRNA
`_ modified antlsense. __ . _. 30µg/ml .•• _ ••.•••• motile •••.• _ •.•. __ .•• motUe ___ • _. _
`modified sense
`6()µg,'ml
`s low
`paralyzed
`12µgfml
`slow
`slow
`______ . ___ ••. __ ••.. __ . • . 2µg/ml •.••••••• ____ slow __ ••. ___ .•. __ . . • motile __ .• _ ••
`unmodified dsRNA
`SOµg/ml
`slow
`paralyzed
`. •. • • . ...... . ....... ... .. tllj.lg/ml .. ...•...... slow ....... ..... slow/paralyzed • ..
`
`trast, 5-iodouracil and 5-(3-aminoallyl)-uracil showed
`remarkably strand-specific effects, with substantially
`greater effects on the antisense strand.
`It was important to test whether greater sensitivity
`to modification for the antisense strand was a general
`phenomenon or a unique property of the unc-22A seg-
`ment used as a trigger. Although the sense and anti-
`sense strands of the unc-22 had comparable base com-
`position, it was conceivable that some specific feature
`of the sequence might account for the difference in
`susceptibility between the two strands. We therefore
`tested three other RNAi trigger segments for sensitivity
`to modification by 5-(3-aminoallyl)-uracil (Figure 6C). For
`each trigger, we observed (a) full or near full activity with
`a modified sense strand and (b) greatly reduced activity
`of the modified antisense strand. We obtained a rough
`quantification of the effect by carrying out titrations of
`the active sense-modified RNAs. These experiments in-
`dicated minimal differences of 10- to 25-fold between
`the antisense and sense